Premotor synaptic plasticity limited to the critical period for song learning

Abstract

Synaptic plasticity has been hypothesized to underlie learning and memory. Understanding of how such plasticity might produce motor learning is limited, in part because of the paucity of model systems with a tractable learned behavior under control of a discrete neural circuit. Songbirds possess both of these traits, thereby providing an excellent model for studying vertebrate motor learning. We report unique evidence of long-term depression (LTD) in the juvenile songbird premotor robust nucleus of the arcopallium (RA). LTD induction at RA recurrent collateral synapses requires NMDA receptors, postsynaptic depolarization, and postsynaptic calcium, and can be reversed by high-frequency stimulation. In adult birds, which have exited the critical period for sensorimotor learning and cannot modify their song, we were no longer able to induce LTD at RA collateral synapses. Furthermore, testosterone-induced premature maturation of song in juveniles abolishes LTD. LTD in nucleus RA therefore makes an excellent candidate mechanism to mediate song learning during development and is well-suited to provide insight into other forms of vertebrate motor learning.

Long-lasting changes in synaptic strength, such as long-term potentiation (LTP) and long-term depression (LTD), have emerged as leading cellular models for the modification of neural circuitry during learning (1). In particular, associative learning induces LTP in the hippocampus (2). Skill learning can increase extracellular field potentials and partially occlude LTP at cortico-cortical synapses in the motor cortex (3), suggesting a similar role for cortical synaptic plasticity in motor learning. However, whether cortical synaptic plasticity underlies motor learning remains unclear, partly because of the distributed nature of cortical motor control and because of the paucity of behavioral paradigms controlled by sufficiently discrete cortical regions.

Songbirds provide a highly advantageous model for studying the neural substrate of motor learning. Juvenile birds learn their songs during a critical period by first listening to an adult male tutor (sensory phase) and then practicing, and listening to themselves (sensorimotor phase), until they produce a stereotyped copy of the tutor's vocalization (4) (Fig. 1A). Early vocalizations are highly variable; after thousands of repetitions, however, the song is modified to match the tutor's song and becomes highly stereotyped, or “crystallized” (5).

Fig. 1.

Song learning in zebra finches is restricted to a critical period and is mediated by discrete nuclei. (A) Timeline depicting the three major phases of song learning in zebra finches with respect to days after hatching. (B) Diagram of the song system showing the direct motor pathway (HVC to RA to syrinx/respiration) and the AFP (HVC to Area X to DLM to LMAN to RA). (C) Simplified schematic of RA depicting cell types, connections, and recording setup.

A discrete set of brain nuclei mediates song learning and production (6) (Fig. 1B). This song system can be divided into two functional pathways: the direct motor pathway, which is necessary for song production, and the anterior forebrain pathway (AFP), a cortical-basal ganglia pathway necessary for song learning and some forms of adult song plasticity, but not for the production of learned song (7–11). The premotor robust nucleus of the arcopallium (RA) occupies a crucial position in the song system as the final telencephalic nucleus in the direct motor pathway and the only motor nucleus to receive input from the AFP (Fig. 1B). Because the AFP is needed for learning, but not for producing learned song, its target RA is the best candidate structure for storing circuit changes that underlie learning (12). Consequently, some form of synaptic plasticity in RA has been incorporated into essentially all theoretical models of song learning (13–15). However, there has been no experimental evidence for such plasticity.

Here we report synaptic plasticity at recurrent collateral synapses in nucleus RA. LTD at these synapses is present only during the critical period for song learning and is abolished by early crystallization of song. These collateral synapses are the most numerous in RA (16) and are well suited computationally to support stabilization of learned patterns (17) and coordination of multiple muscle groups (18). Our results provide a unique candidate cellular mechanism for mediating behavioral changes during the sensorimotor phase of song learning.

Results

RA Collateral Synapses in Juvenile Zebra Finches Show the Capacity for LTD.

We performed whole-cell voltage-clamp recordings in acute slices of juvenile (45–60 d posthatch, dph) zebra finch brain and antidromically activated collateral synapses between RA projection neurons. Low-frequency stimulation (LFS; 5-Hz presynaptic stimulation for 2 min with postsynaptic depolarization to 0 mV) caused a significant reduction in excitatory postsynaptic current (EPSC) amplitude at RA collateral synapses (58 ± 7% of baseline, n = 14, P = 0.0001) (Fig. 2), which lasted as long as the recordings could be maintained (maximum 50 min postconditioning). This LTD was not accompanied by a change in the paired-pulse ratio (119 ± 12% of baseline, n = 9, P = 0.20), suggesting that release probability was not altered. High-frequency stimulation (HFS; 100-Hz presynaptic stimulation for 1 s, delivered three times with postsynaptic depolarization to 0 mV), which induces LTP in other brain regions (1, 19), caused EPSC amplitude to return to baseline levels (Fig. 2E) (147 ± 39% of baseline, n = 5, P = 0.275). Conversely, HFS did not alter EPSC amplitude when applied alone (92 ± 8% of baseline, n = 6, P = 0.39). This finding was significantly different from the response to LTD induction (P = 0.021). Thus, LFS induces LTD but HFS induces dedepression, but not potentiation, at RA collateral synapses.

Fig. 2.

Pairing LFS with postsynaptic depolarization induces LTD at collateral synapses in RA. (A) Sample traces taken from the experiment in B showing EPSCs elicited by collateral synapse stimulation under baseline conditions (black) and after application of the conditioning protocol (gray). (B) EPSC amplitude plotted across one experiment. Time 0 represents the first EPSC recorded after application of the conditioning stimulus. (C) Summary data showing mean and SEM of 18 such experiments in 1-min time bins. (D) LTD induction is blocked by 50 μM APV, hyperpolarization of the postsynaptic cell to −90 mV, or inclusion of 20 mM BAPTA in the recording pipette. Each point represents the average EPSC amplitude from 13 to 17 min postconditioning for one recording. (E) HFS (3× 1 s 100-Hz trains) reverses the LTD induced by LFS (n = 5).

Transmission at collateral synapses is mediated by both AMPA and NMDA receptors (20), and LTD in mammals usually requires NMDA receptor activation. We found that induction of LTD at RA collateral synapses was blocked by the application of the NMDA receptor antagonist APV (2-amino-5-phosphonovaleric acid; 50 μM) (98 ± 6% of baseline, n = 5, P = 0.75) (Fig. 2D and Fig. S1 A and B). Including the the calcium chelator BAPTA (20 mM) in the recording electrode also blocked LTD induction (95 ± 3% of baseline, n = 7, P = 0.10) (Fig. 2D and Fig. S1C). Additionally, postsynaptic hyperpolarization to −90 mV during conditioning blocked LTD (89 ± 5% of baseline, n = 11, P = 0.06) (Fig. 2D). The resulting change in EPSC amplitude was significantly different from the LTD induced with depolarization to 0 mV (P = 0.018). Thus, LFS coupled with postsynaptic depolarization induces NMDAR- and postsynaptic calcium-dependent LTD at RA collateral synapses in juvenile zebra finches.

LTD in RA Is Restricted to the Critical Period for Sensorimotor Learning.

Zebra finches are closed-ended learners; their song crystallizes and does not change once they reach adulthood. We hypothesized that if LTD at RA collateral synapses was involved in the sensorimotor phase of song learning, then it would not occur in adulthood when intact birds no longer modify their songs. To test this theory we attempted to induce LTD at RA collateral synapses in slices from adult zebra finches (>92 dph). RA collateral synapses in adults did not show the capacity for LTD (97 ± 6% of baseline, n = 7, P = 0.61) (Fig. 3). Young adults (92–95 dph) and older adults (>120 dph) showed similar degrees of LTD (P = 0.40). Normalized EPSC amplitude after conditioning in juveniles was significantly less than in adult birds (P = 0.01). Thus, the capacity for LTD at RA collateral synapses that exists in juveniles is absent in adults.

Fig. 3.

RA collateral synapses in adult brains do not show the capacity for LTD. (A) Sample traces taken from the experiment in B showing two EPSCs elicited by collateral synapse stimulation under baseline conditions (black) and after application of the conditioning protocol (gray). (B) EPSC amplitude plotted across one experiment. Time 0 represents the first EPSC recorded after application of the conditioning stimulus. (C) Summary data showing the mean and SEM of 11 such experiments.

Premature Song Crystallization Abolishes LTD in RA.

Our data suggest that the bird's ability to modify its song may relate directly to the capacity for RA collateral synapses to undergo LTD. Alternatively, the loss of both the ability to modify song and LTD in adults may also be because of two coincident developmental processes with no direct relationship. To distinguish between these two possibilities we used testosterone to prematurely crystallize juvenile song. Testosterone has been implicated in natural song crystallization (21), and artificially increasing testosterone levels in juvenile birds causes their songs to be impoverished when they reach adulthood (22). Furthermore, juvenile canaries treated with testosterone produce adult-like respiratory motor gestures (23). We therefore hypothesized that testosterone administration in juvenile birds would cause a premature crystallization of song, providing us with the opportunity to dissociate the time-course of song development from general development and determine how closely LTD in RA is related to song plasticity. Juvenile birds (38–49 dph) were implanted subcutaneously with testosterone pellets (see Methods) and housed individually in sound attenuation chambers. Song was continuously recorded until birds were removed for electrophysiology experiments (10–14 d later). A control group of juveniles (43–53 dph) were implanted with blank pellets and subjected to the same experimental protocol as testosterone-implanted birds. Testosterone-implantation caused a significant increase in levels of circulating testosterone compared with blank implants (25.5 ± 8.9 ng/mL, n = 5, and 0.3 ± 0.003 ng/mL, n = 6, respectively; P = 0.012). Fig. 4 shows representative motifs from age-matched blank- and testosterone-implanted birds; the spectral features of the blank-implanted bird's song are much more variable from rendition to rendition than those of the testosterone-implanted bird's song. One of the hallmarks of song crystallization is a decrease in spectral variability across renditions of song. Therefore, to determine whether song had crystallized we measured the trial-to-trial variability in spectral frequency of the bird's song over the course of one day using the “pitch” measurement in Sound Analysis Pro (24) (Fig. 5). Briefly, pitch was measured for each syllable sung on the last day before electrophysiological recording. The coefficient of variation (CV) of these pitch measurements was then calculated for each syllable and those values were averaged for each bird (see Methods). Testosterone treatment caused a significant decrease in the CV of pitch compared with treatment with blank implants (0.17 ± 0.01, n = 7, and 0.27 ± 0.02, n = 7, respectively, P = 0.0012) (Fig. 5C). The CV of pitch for testosterone-implanted juveniles was not significantly different from that of five randomly selected adult birds (0.19 ± 0.07, n = 5, P = 0.63) (Fig. 5C). Despite the difference in pitch variability, testosterone- and blank-implanted birds matched their tutor to a similar degree, and there was no difference in similarity to tutor song between the two groups (55 ± 9% and 62 ± 4%, respectively; P = 0.53; Sound Analysis Pro similarity score). Thus, testosterone induces premature song crystallization, prematurely reducing the pitch variability of juvenile song to the level of adults.

Fig. 4.

Testosterone reduces juvenile song variability. (A) Three spectrograms showing representative motifs from a 58-dph juvenile 14 d after receiving a blank implant. Each motif was taken from a different song bout. Note the variability in spectral features from rendition to rendition. (B) Three spectrograms showing representative motifs recorded from a testosterone-implanted 51-dph juvenile 10 d after receiving the implant. Each motif was taken from a different song bout. Note that each rendition is very similar to the others.

Fig. 5.

Testosterone reduces juvenile song variability. (A) Scatterplot of the SAP pitch measurement for each of the first 1,000 syllables sung on one day by a 58-dph juvenile 14 d after receiving a blank implant. (B) Same as A for the first 1,000 syllables sung on 1 d by a 51-dph juvenile 10 d after receiving a testosterone implant. (C) Average CV of pitch from the final day of song recording for adults and blank- and testosterone-implanted juveniles. Each point represents the mean CV of all syllables for one bird and the horizontal bar shows the mean for each group.

We prepared acute brain slices from implanted birds and measured the capacity for LTD at RA collateral synapses. In slices prepared from testosterone-treated birds, as with adult birds, application of the LFS conditioning protocol caused no change in EPSC amplitude (104 ± 9% of baseline, n = 6, P = 0.68) (Fig. 6C). Blank-implanted birds, however, showed LTD (52 ± 5% of baseline, n = 6, P = 0.0003) (Fig. 6C) that was not significantly different from normal juvenile birds (P = 0.54) and the two groups were therefore combined (Fig. 2C). There was no correlation between magnitude of LTD and the CV of pitch for either blank or testosterone-implanted birds (P = 0.56). There was also no difference in baseline transmission between untreated, blank-, and testosterone-implanted juveniles, as measured by paired-pulse ratio and CV of EPSC amplitude, (P = 0.95 and P = 0.796, respectively). Furthermore, acute application of 0.1 nM testosterone to slices prepared from untreated juveniles did not affect baseline EPSC amplitude (101 ± 5% of baseline, n = 6, P = 0.8) nor did it block LTD induction (49 ± 8% of baseline, n = 6, P = 0.0013). These results show that, when song is crystallized prematurely, the capacity of RA collateral synapses to undergo LTD is suppressed prematurely but baseline synaptic transmitter release probability remains unchanged. This finding is consistent with LTD playing a role in learning and its loss in adults playing a role in song crystallization.

Fig. 6.

Testosterone-implanted juveniles do not show the capacity for LTD at RA collateral synapses. (A) Sample traces taken from the experiment in B, showing two EPSCs elicited by collateral synapse stimulation under baseline conditions (black) and after application of the conditioning protocol (gray). (B) EPSC amplitude plotted across one experiment. Time 0 represents the first EPSC recorded after application of the conditioning stimulus. (C) Summary data showing the mean and SEM of nine LTD experiments in testosterone-implanted juveniles (•), and seven experiments from blank-implanted juveniles (○).

Discussion

Here, we describe a unique form of synaptic plasticity in the songbird equivalent of the motor cortex. Collateral synapses formed between RA projection neurons show activity-dependent LTD. Induction of LTD in RA requires the activation of NMDARs, postsynaptic depolarization, and a rise in postsynaptic calcium, and can be reversed by HFS. Furthermore, this form of synaptic plasticity is present only in juvenile birds in the process of learning their song. In adults, whose songs are crystallized and can no longer be modified, the capacity for LTD at RA collateral synapses vanished. Moreover, when we prematurely crystallized song in juvenile birds with testosterone implants, we were unable to induce any change in synaptic strength. These data not only provide a unique example of premotor synaptic plasticity in the songbird, but also implicate LTD at RA collateral synapses in song learning. LTD of RA collateral synapses therefore provides an excellent, experimentally tractable candidate cellular mechanism for mediating sensorimotor learning.

Potential Mechanisms for Song Modification via RA Collateral LTD.

Song acquisition certainly involves a variety of changes in many brain regions. A synaptic plasticity mechanism, however, could contribute directly to the circuit changes underlying learning. How might LTD at RA collateral synapses specifically contribute to song learning? One possibility is that collateral connections in RA form a dense network early in development and are subsequently selectively pruned through LTD during the sensorimotor period of song learning. This hypothesis fits well with existing data on song development and RA activity. Juvenile song is highly variable; an excess of intrinsic connections in RA could participate in creating this variability by providing many alternative paths for descending motor signals to travel before they reach brainstem motor nuclei. LTD induction occurs at a relatively low rate of collateral synaptic stimulation (5 Hz), which falls within the range of the intrinsic activity of RA neurons in vitro (1–16 Hz) (25, 26) and in vivo (4–10 Hz) (27). However, HFS (100 Hz), reverses the change in synaptic strength induced by LTD. Therefore, collateral synapses that are activated by high-frequency bursts, such as those seen in RA during song production (28, 29), would remain unchanged, and those activated at only the lower frequencies of spontaneous firing would be depressed. Eventually this process would leave circuits formed by only those synapses that were active during singing. In this model, storage of the tutor song and evaluation of the quality of the current song with respect to the tutor song would occur elsewhere, likely upstream of RA.

Alternatively, LTD may work in conjunction with other forms of synaptic modification. Although we have tested for the existence of LTP using a few standard protocols (Table S1), it is difficult to entirely rule out the capacity for LTP at collateral synapses. Perhaps RA collateral synapses are already maximally potentiated by 45 dph. Therefore, HFS alone cannot alter EPSC amplitude, but following LTD, HFS is capable of restoring EPSC amplitude to baseline levels. Alternatively, perhaps producing song elements that closely match the tutor song leads to the release of a neuromodulatory signal in RA (15). In the presence of this neuromodulatory signal, activity that would have otherwise led to the induction of LTD may cause no change in synaptic strength, or perhaps even induce LTP, thereby selectively reinforcing circuits involved in the production of sounds that more closely match those of the tutor song.

RA receives the major output of the AFP, which is necessary for song learning (7, 8), yet LTD induction in RA does not require activation of lateral magnocellular nucleus of the anterior nidopallium (LMAN) inputs. Although the mechanism by which the AFP facilitates song learning remains elusive, the AFP clearly generates song variability (9–11). AFP output provides excitatory input to RA, mediated almost entirely by NMDA receptors (20, 30) and infusion of NMDA receptor blockers into RA mimics the effect of an LMAN lesion on song variability (9). Taken together these data suggest that the AFP injects variability into the motor pathway via the NMDA receptor-mediated LMAN input to RA. This increased variability may allow the bird to explore vocal motor space, a necessary component of reinforcement learning (14), without directly participating in the modification of the song motor program. Furthermore, synaptic plasticity has been described in both Area X (19) and LMAN (31), which would allow AFP output to be modified over the course of learning such that less and less variability is introduced as the juvenile song becomes a closer match to the tutor song. Under normal conditions for closed-ended learners, such as the zebra finch, the bird's song will then remain unchanged. However, perturbations of auditory feedback (32, 33), damage to nucleus HVC (11), deafening (34), or syringeal nerve damage (35) can reinstate song plasticity in adult birds. LMAN lesions before deafening prevent song degradation (35–37); however, lesions of LMAN do not induce recovery of song that has been allowed to completely degrade (38). It would therefore be interesting to determine whether the capacity for LTD in RA is reinstated in these birds or whether open-ended learners, such as canaries or mockingbirds, which continue to modify their song throughout their lives, show the capacity for LTD in RA during periods of adult song plasticity. These experiments would help to address the issue of whether LTD at RA collateral synapses plays a specific developmental role in song acquisition, or participates more generally in vocal motor learning throughout life.

Testosterone and Song Crystallization.

A natural increase in testosterone levels has been implicated in the normal crystallization of song (21). We used testosterone to manipulate song development and measured concurrent changes in the capacity for premotor synaptic plasticity. Although testosterone has been used in past studies to alter song development (22, 23, 39, 40), our measurements of song variability (Figs. 4 and 5) represent a unique behavioral description of the effects of testosterone implantation in juvenile zebra finches, and demonstrate premature crystallization without improved matching to the tutor song.

Boosting testosterone levels presents a powerful tool for modifying song behavior, thereby separating the changes in synaptic efficacy linked directly to song crystallization from the effects of general development. However, this technique also brings with it a few caveats. Elevated testosterone in juvenile birds reduced song variability as if the young bird's song had crystallized (Figs. 4 and 5) and eliminated LTD in RA (Fig. 6). This experiment, however, does not establish whether the loss of LTD is caused by blockade or occlusion of the LTD mechanism, which would have important implications for the mechanism of song crystallization. Furthermore, androgen receptors are widely expressed in the avian telecephalon, including RA (41). The activation of these receptors has dramatic effects on the song control circuitry and on behavior (42, 43). It is therefore not clear whether testosterone acts directly to block the LTD mechanism. However, because the crystallized juvenile songs were not more similar to the tutor song than the songs of blank-implanted juveniles, testosterone appears not to act simply by enhancing or accelerating the learning process, which would argue against occlusion underlying the loss of the capacity for LTD.

Systemic testosterone also shortens the decay time of NMDA receptor-mediated EPSPs at LMAN inputs to RA in juvenile birds (40). LTD in RA requires NMDA receptor activation and developmental changes in NMDAR-mediated currents at collateral synapses may also underlie the developmental loss of capacity for LTD in RA. Manipulations that extend the bird's ability to acquire song did not extend the expression of long NMDA receptor decay times, suggesting that alteration of NMDA receptor-mediated transmission at LMAN inputs to RA does not mediate the closure of the critical period for sensorimotor learning (44). Both NMDA receptor activation and postsynaptic calcium are required for LTD in RA. It would therefore be interesting to determine whether NMDA receptor decay times at RA collateral synapses are altered by testosterone. If, as with the LMAN inputs, testosterone causes a reduction in NMDA receptor currents, this could be the mechanism by which testosterone causes a loss of LTD in RA. Either way, an important follow-up experiment will be to determine whether rearing in isolation, which extends a bird's ability to modify its song, also extends the capacity for LTD in RA.

Concluding Remarks.

Collateral synapses in the premotor nucleus RA show activity-dependent LTD, which disappears in adult birds whose song has naturally crystallized (Figs. 2 and 3). LTD in RA also disappears in juvenile birds whose song has been prematurely crystallized (Fig. 6). These experiments strongly suggest a functional linkage between LTD in the RA and song plasticity. To test for causation, a specific blocker of LTD could be infused into RA of juvenile birds during the sensorimotor period to determine whether blocking LTD blocks song learning. Identifying such a compound will require detailed molecular understanding of the LTD mechanism. Our results provide a strong argument for a role of LTD at RA collateral synapses in sensorimotor learning in birds.

Until recently, LTD had often been considered a mechanism for merely resetting—or forgetting—the changes induced by LTP. However, LTD may also encode information independently of other forms of synaptic plasticity (45). LTD demonstrates all of the characteristics laid out by Hebb (46) as necessary for a cellular mechanism of learning: specificity, cooperativity, and associativity. Furthermore LTD in mammals has been implicated in a number of forms of learning (45). Aspects of spatial learning, in particular novel object recognition, preferentially involve hippocampal LTD over LTP (47). Cerebellar LTD has been hypothesized to underlie various forms of motor learning, including the adaptation of the vestibular ocular reflex and associative eyeblink conditioning (48–50). LTD in the basal ganglia has been linked to both drug addiction and motor learning (51, 52). Taken together, these studies suggest that LTD plays an integral role in mediating complex forms of learning. In songbirds, LTD of collateral synapses in nucleus RA provides a unique candidate cellular mechanism to mediate sensorimotor learning.

Methods

Adult male zebra finches (>120 dph) were obtained from a supplier and housed in groups. Young adults (90–110 dph) and juveniles were bred in our colony and remained with their parents and siblings until the day of the experiment or the beginning of song recording. All birds were housed in the same room and kept on a 14-/10-h light/dark cycle. All procedures were performed in accordance with a protocol approved by the University of Washington Institutional Animal Care and Use Committee.

Slice Preparation.

Acute brain slices were prepared as described previously (20). Briefly, each bird was anesthetized with isofluorane, decapitated, and the brain removed and placed in ice-cold, oxygenated solution consisting of (in mM): 119 NaCl, 2.5 KCl, 1.3 MgS0₄, 2.5 CaCl₂, 1 NaH₂PO₄, 16.2 NaHCO₃, 11 d-glucose, and 10 Hepes (290–300 mOsm). Horizontal slices (300-μm thick; Vibratome) were made and transferred to 37 °C artificial cerebrospinal fluid (ACSF), containing (in mM): 119 NaCl, 2.5 KCl, 1.3 MgSO₄, 2.5 CaCl₂, 1 NaH₂PO₄, 26.2 NaHCO₃, and 11 d-glucose (290–300 mOsm), and allowed to recover for at least 45 min, during which time the ACSF cooled to room temperature.

Electrophysiology.

“Blind” whole-cell voltage-clamp electrophysiological recordings (53) were obtained from slices maintained at 30 (±1) °C and superfused in oxygenated ACSF containing 150 μM picrotoxin to block fast GABAergic inhibitory postsynaptic currents. Electrodes were pulled from borosilicate glass (Garner Glass Co.) on a micropipette puller (Model P-97; Sutter Instruments). Electrodes had a resistance of 4 to 7 MΩ; pipette solution contained (in mM): 120 CsCH₃SO₃, 10 Hepes, 8 NaCl, 0.2 EGTA, 2 MgCl₂, 10 phosphocreatine, 5 QX-314, 2 ATP, and 0.3 GTP. EPSCs from RA collateral inputs were activated by antidromic electrical stimulation (100-μs duration) of RA output fibers using a bipolar stainless steel electrode (FHC). Stimulus amplitude was set for each recording at ∼50% of the maximum EPSC amplitude. Cells were held at −70 mV unless otherwise stated.

Given the large number of intrinsic excitatory synapses in RA (16) and the spontaneous activity of RA projection neurons in vitro (25, 26), the following steps were taken to ensure that the EPSCs studied were monosynaptic: (i) the Ca and Mg concentration of the bath solution was each raised to 4 mM to reduce excitability; (ii) recordings with EPSC onset (measured as the point where the EPSC rose above noise) latencies that varied greater than 0.4 ms (total range) at −70 mV were excluded; (iii) EPSCs that did not rise smoothly to peak were excluded; and (iv) EPSCs with variations in amplitude exceeding 50% between consecutive traces (not including failures) during the baseline period were excluded.

Current signals were amplified and low-pass filtered at 10 kHz with an AxoPatch 1D (Axon Instruments), digitized at 50 kHz with a Digidata 1322A (Axon Instruments), and stored on a PC using pClamp software (Axon Instruments). EPSCs were elicited at 0.1 Hz. Drugs used include: APV (Tocris), BAPTA (Tocris), picrotoxin (PTX; Sigma), QX-314 (Tocris), and crystalline testosterone. All drugs were applied to the bath solution unless otherwise stated.

Histology.

All cells were filled with biocytin (Sigma) during recording. After recording, the slice was fixed in 4% paraformaldehyde, cryoprotected, resectioned to 50-μm thickness with a freezing microtome, and processed with Alexa 488 conjugated streptavidin (1:1,000; Molecular Probes). Sections were mounted on slides and viewed with fluorescence microscopy to confirm cell type. Cells were categorized as projection neurons if they displayed spiny dendrites (26).

Song Recording.

For song analysis, birds were isolated in sound attenuation chambers (Acoustic Systems). Song was continuously recorded using Pacific Pro Audio LD-1 microphones (Pacifc Pro Audio). Sounds were first amplified with an Alesis Studio 12R microphone preamp and then digitally recorded using either an Aardvark DirectPro or M-Audio Delta 44 sound card and Sound Analysis Pro (SAP, v1.04) software (24). Song files from the last day of recording were manually separated from cage noises, and then analyzed using Matlab (v. 2007b, Math Works), Excel (Microsoft), and SAP. To measure song variability, song files were first analyzed with SAP using the batch-processing mode. The entropy and amplitude parameters were set manually and a subset of files was visually inspected to ensure that syllables were properly segmented. SAP data were exported to Excel. Scatterplots were created and data for each syllable was separated with the filter function of Excel using parameters set by referencing song files in SAP (37). The CV of pitch was calculated for each syllable individually and then averaged for each bird. Song similarity was measured by comparing one motif from each of the first 10 bouts of song sung by either a testosterone- or blank-implanted bird and their respective tutor.

Surgery.

For some experiments, juvenile birds were implanted subcutaneously with SILASTIC tubing: inner diameter, 1.0 mm; outer diameter, 2.0 mm (VWR). Implants contained either 10 mm of crystalline testosterone (testosterone implant) or 10 mm of silicon (blank implant). Implants were rinsed with ethanol and soaked overnight in 0.1 M PB before implantation.

Data Analysis.

Recordings were excluded from analysis if the series resistance or holding current changed by more than 20% between the baseline and postconditioning periods. Only recordings from morphologically identified RA projection neurons were included for analysis. Sample traces represent the average of five traces and were low-pass filtered at 2 kHz. Measurements of EPSC amplitude were made with Clampfit software (Axon Instruments) and then imported into Excel. EPSC amplitude was normalized and is expressed as percentage of baseline. Measurements were taken by averaging the values from 13 to 17 min postconditioning except for the BAPTA experiments, which were measured from 3 to 7 min to avoid artifacts caused by rundown (54); HFS experiments, which were measured at 3–7 min; and the blank-implanted birds, which include one cell, measured at 3 to 5 min postconditioning, the last time points recorded for that experiment. Paired-pulse ratios were calculated by evoking two EPSCs separated by 40 ms and dividing the amplitude of the first EPSC by the amplitude of the second EPSC. Ratios reported are the mean from 10 consecutive traces. Data were graphed using either Igor (Wavemetrics) or Prism (Graphpad Software Inc.). Data are presented as the mean and SE unless otherwise stated.

Statistics.

Statistical analysis was performed with Prism software. Unpaired t-tests were used to compare paired-pulse ratios, postconditioning EPSC amplitude, and song spectral parameters between experiment groups. One-sample t-tests were used to compare the effects of conditioning with respect to baseline for all experiments. A one-way ANOVA was used to determine differences in pair-pulse ratios and CV of EPSC amplitude. Differences were considered significant when P < 0.05.