Search tips
Search criteria 


Logo of nihpaAbout Author manuscriptsSubmit a manuscriptHHS Public Access; Author Manuscript; Accepted for publication in peer reviewed journal;
J Neurosci. Author manuscript; available in PMC 2010 June 9.
Published in final edited form as:
PMCID: PMC2819911

Millisecond timescale disinhibition mediates fast information transmission through an avian basal ganglia loop


Avian song learning shares striking similarities with human speech acquisition and requires a basal ganglia (BG)-thalamo-cortical circuit. Information processing and transmission speed in the BG is thought to be limited by synaptic architecture of two serial inhibitory connections. Propagation speed may be critical in the avian BG circuit given the temporally precise control of musculature during vocalization. We used electrical stimulation of the cortical inputs to the BG to study, with fine time resolution, the functional connectivity within this network. We found that neurons in thalamic and cortical nuclei that are not directly connected with the stimulated area can respond to the stimulation with extremely short latencies. Through pharmacological manipulations, we trace this property back to the BG, and show that the cortical stimulation triggers fast disinhibition of the thalamic neurons. Surprisingly, feedforward inhibition mediated by striatal inhibitory neurons onto BG output neurons sometimes precedes the monosynaptic excitatory drive from cortical afferents. The fast feedforward inhibition lengthens a single inter-spike interval in BG output neurons by just a few milliseconds. This short delay is sufficient to drive a strong, brief increase in firing probability in the target thalamic neurons, evoking short latency responses. By blocking glutamate receptors in vivo, we show that thalamic responses do not appear to rely on excitatory drive, and we show in a theoretical model that they could be mediated by post-inhibitory rebound properties. Such fast signalling through disinhibition and rebound may be a crucial specialization for learning of rapid and temporally precise motor acts such as vocal communication.

Keywords: Basal Ganglia, Thalamus, birdsong, Electrical stimulation, Neuron, Neurotransmitter


The central nervous system provides several examples of circuits specialized for fast processing because they subserve tasks in which long transmission delays have important behavioral costs. For instance, auditory information reaches the forebrain through a polysynaptic pathway within 10 ms (Heil and Scheich, 1991; Sen et al., 2001; Pollack et al., 2003). Similarly the vestibulo-ocular reflex can cause eye movements within 10–15 ms of head motion (Broussard et al., 1992). Transmission speed is critical in neural systems involved in vocal communication in humans and birds since the duration of perceptually critical sounds in speech and bird songs can be as short as 10 ms (Goller and Cooper, 2004; Glaze and Troyer, 2007; Liberman et al., 1961).

The forebrain nuclei mediating song control form two segregated pathways (Fig. 1A; Nottebohm et al., 1976). The robust nucleus of the arcopallium (RA) receives input from the cortical (see Materials and Methods for discussion of nomenclature) area HVC (proper name) both monosynaptically and polysynaptically, through a basal ganglia (BG)-thalamo-cortical circuit called the anterior forebrain pathway (AFP, Bottjer et al., 1989). This BG pathway is essential for vocal learning (Bottjer et al., 1984; Scharff and Nottebohm, 1991). Song-related signals from HVC propagate through these two pathways (Kimpo et al., 2003), which converge on individual RA neurons (Mooney and Konishi, 1991). For the signals themselves to interact in RA, propagation delays should match.

Figure 1
The response of LMAN neurons to HVC electrical stimulation

Transmission delays within the BG, however, are expected to slow down signal propagation through the AFP. Indeed, activation of the thalamus through the BG first requires disinhibition, the interruption of the tonic inhibition exerted on thalamic neurons by pallidal neurons (Deniau and Chevalier, 1985). In mammals, latencies of over 20 ms have been reported between cortical activation and pallidal inhibition (Nambu et al., 2000). Further delays could be introduced depending on the mechanisms driving thalamic neurons upon disinhibition. In songbirds, postinhibitory rebound in thalamic neurons allows them to respond to disinhibitory signals in the absence of excitation (Luo and Perkel, 1999b; Person and Perkel, 2005). Although faster than in mammals (Thomson, 1988), postinhibitory rebound might impose further delays (>10 ms) between the cortical drive and the thalamic response. Based on time lags in correlated activity in the song system, Kimpo et al. (2003) have hypothesized transmission delays through the AFP of around 60 ms. Troyer and Doupe (2000) proposed a theoretical framework allowing learning despite long AFP delays. However, shorter delays would support a wider range of possible learning mechanisms.

We used electrical stimulation to measure with high temporal precision the time course of information flow through the songbird BG circuit. We report surprisingly fast transmission through the AFP, with latencies as short as 10 ms, relying on a BG disinhibitory circuit specialized for speed. AFP signals may thus overlap in time in RA with signals from the monosynaptic motor pathway, expanding the range of possible mechanisms underlying vocal learning.

Materials and Methods


Adult male zebra finches (Taeniopygia guttata) were obtained from a commercial supplier and used in accordance with an animal use protocol approved by the University of Washington Institutional Animal Care and Use Committee. Animals were housed under a 14/10h light/darkcycle with food and water available ad libitum.


Animals were first food deprived for 30 min and were then given three intramuscular injections totaling 5–6.5 ml/kg of 20% urethane over 1h. Urethane anesthesia preserves normal neurotransmission in various subcortical areas and the peripheral nervous system (Maggi and Meli, 1986). Local anesthetic (1% lidocaine) was injected under the scalp before animals were placed in a stereotaxic apparatus. Small craniotomies were made above the midline reference point, the bifurcation of the midsagittal sinus, and above HVC and the lateral magnocellular nucleus of the anterior neostriatum (LMAN), the medial portion of the dorsolateral nucleus of the anterior thalamus (DLM) or Area X unilaterally. Lidocaine gel was then applied to the incision at 3h intervals.

Electrophysiological recording

Glass pipettes (TW100F-3, World Precision Instrument, Sarasota, FL) were pulled on a micropipette puller (Model P97, Sutter Instrument Co., Novato, CA), and the tips were blunted to achieve 5 to 25Mω impedance. A ground electrode was placed in the cerebellum posterior to the midline reference point. A concentric stimulation electrode (FHC, Brunswick, ME) was placed in HVC (0 mm rostral, 2.4 mm lateral from the midline reference point, 0.5 mm deep). The recording electrode signal was amplified 10x and low-pass filtered at 3kHz (Axoclamp2B amplifier, Molecular Devices, Foster City, CA), passed through a Hum Bug noise eliminator(AutoMate Scientific, San Francisco, CA), and amplified further100x (Model 410, Brownlee Precision, San Jose, CA). Recordings were monitored using an oscilloscope and an audio monitor. We searched for single-unit neuronal activity in Area X, DLM or LMAN using HVC stimulation as a search stimulus. Once a neuron was isolated, the electrophysiological signal was sampled at 20kHz and spike times and raw traces were stored for further analysis (Spike2, Cambridge Electronic Design, Cambridge, UK). Principal components analysis of the spike shapes allowed clear separation from noise and all extracted units obeyed a refractory period of 1ms. In recordings in Area X, neurons displaying spontaneous firing above 25 spikes/s (sp/s) are referred to as pallidal neurons given their similarity with pallidal terminals recorded in DLM (Person and Perkel, 2007), while neurons displaying a firing rate lower than 25 sp/s are called putative interneurons. Recordings were performed during HVC microstimulation (monophasic 0.2 ms single pulses), with various stimulation intensities (10 to 4000 μA). Each pulse of HVC microstimulation saturated the amplifier and occluded spiking activity for 1–2 ms in the recordings, due to the “overshoot” following saturation. Because the duration of this stimulation artifact was much shorter than the latency of the fastest responses recorded, it occluded only spontaneous activity and thus did not alter our analysis.

Previous studies have shown that stimulation with a monopolar microelectrode at 200 μA activates about 50% of the neurons located in a shell of 0.2 mm outside radius (Ranck, 1975; Tehovnik et al, 2006). These values should be considered with caution because the current intensity necessary to activate an axon at a given distance depends on a number of other variables such as the size of the axon or its biophysical properties (e.g. cellular excitability or axon myelination). In addition, concentric bipolar electrodes greatly reduce current spread, especially for higher stimulation intensities (Bagshaw and Evans, 1976; Follett and Mann, 1986), and estimates from monopolar electrodes thus provide an imprecise upper bound of the activated volume. The volume of the “activated shell” for 200 μA (inside and outside radius: 0.1 and 0.2 mm), is 0.029 mm3, and represents about 10% or less of HVC volume (0.2–0.5 mm3, MacDougall-Shackleton et al, 1998). Therefore, 200 μA pulses applied near the center of HVC through concentric bipolar electrodes are expected to activate less than 5% of HVC neurons.

We cannot exclude the possibility that some high amplitude stimulation led to current spread to neighboring structures. However, because of the segregation of the AFP circuit from surrounding tissue, it seems unlikely that occasional activation of neighboring structures would modify the interpretation of the present data.

Paired recordings

Pairs of units with distinct waveforms were recorded simultaneously at 14 DLM sites. These recordings always included a rapidly firing unit with a monophasic spike waveform of smaller amplitude and a more slowly firing unit with a biphasic waveform of larger amplitude. The waveforms differed in amplitude by at least 4-fold, permitting unambiguous classification of spikes. These recordings have been shown to reflect the activity of a presynaptic pallidal terminal and the corresponding postsynaptic thalamic cell (Person and Perkel, 2007). In all cases, the extracted units obeyed a refractory period of 1ms. Since previous recordings of song-related and spontaneous activity gave similar results, they were pooled in the analysis.

Antidromic activation

We recorded high frequency, spontaneously active neurons in Area X thought to be the pallidal projection neurons. We found no significant difference in the spike shapes of high versus low spontaneous activity neurons in Area X. The distributions of spontaneous activity between these neurons and terminals within DLM were not significantly different (Fig. 4A; 73 Area X putative somata averaged 62.8 ± 17.3 sp/s; 37 DLM terminals averaged 60.9 ± 13.6 sp/s). In some experiments, a parallel bipolar stimulating electrode (0.5 mm spacing, FHC, Bowdoin, ME) was placed in DLM for antidromic identification of projection neurons. Continuous analog records were acquired for at least 300 stimuli, each of which consisted of single monophasic pulses delivered at 0.1s intervals (0.1 ms duration, 0.2–2 mA intensity, stimulus isolation unit from ISO-flex, AMPI, Jerusalem, Israel). Latency variability was defined as the standard deviation (SD) of latencies to the first evoked spike. Putative projection neurons were defined as having evoked spikes with small latency variability (<0.2 ms) and collision tests resulting in 100% failure of the antidromic spike.

Figure 4
Comparison of Area X output neuron somata and terminals and the morphological properties of their axons

Drug injections

All drugs were diluted in a 0.9% saline solution with 0.5% dextran-conjugated Alexa-fluor 488 (3000MW; Invitrogen, Eugene, OR). The role of GABAA receptor-mediated synaptic transmission in Area X was examined using micro-injection of the GABAA receptor blocker gabazine (0.1–0.2 mM, Tocris, Ellisville, MO). The role of glutamatergic transmission in Area X and DLM was examined using micro-injection of the AMPA receptor blocker NBQX (1 mM, Tocris) and the NMDA receptor blocker APV (5 mM, Tocris) in 0.2% DMSO. Vehicle solution consisted either of saline or 0.2% DMSO solution. Drugs were pressure ejected from pulled glass pipettes (10–20 μm tip size) using a Pressure system IIe (Toohey Co., NJ; 50 ms pulses at 10–16 psi). Injected volumes were 20–100 nL. When the recording and drug injections were made in the same structure, we aimed to place the tip of the injection pipette 200–300 μm from the tip of the recording pipette.


At the end of each experiment, recording sites were labeled by iontophoretic injections of fluorescent dye (5% Alexa-488- or -568-conjugated 10kDa dextranamine in 0.01M phosphate-buffer PB, pH 7.4, ejected by 5 μA AC current for 5min). Animals were killed by intramuscular injection of sodium pentobarbital (Nembutal) and perfused with 0.9% saline followed by 4% paraformaldehyde as fixative. The brain was then removed, post-fixed in 4% paraformaldehyde for 24h, and cryoprotected in 30% sucrose. 40μm-thicksections were then cut in the parasagittal plane on a freezing microtome and processed for histological examination to verify the location of stimulating and recording electrodes, and drug injection sites. In addition to gross observation of electrode tracts, the brain slices were visualized using a fluorescence microscope allowing better determination of recording location. In the case of DLM recordings, descending axons from Area X were clearly retrogradely labeled in most cases, allowing unambiguous determination of recording site. We were able to recoverdye deposits in DLM after recording in 13 animals (out of 23 animals in which DLM recordings were conducted).

For quantification of Area X projection axon size, bilateral iontophoretic injections of biotinylated dextran amine (BDA; 10000MW, Molecular Probes, Leiden, Netherlands, 10% in 0.1M phosphate-buffer PB, pH 7.4) were made into Area X. After a survival time birds were euthanized with pentobarbital as described above, then perfused transcardially, first with physiological saline followed by fixative containing 2% paraformaldehyde (TAAB, UK) and 0.5% glutaraldehyde (TAAB) in acetate buffer (pH 6.0; 5 min), and finally with fixative containing 2% paraformaldehyde and 0.5% glutaraldehyde in borate buffer (pH 8.5; 50 min). Parasagittal sections (60 μm) containing DLM and Area X were cut with a Vibratome. After pretreatment of sections, injection sites and labeled fibers were visualized with a nickel intensified 3,3′-diaminobenzidine (DABNi) reaction resulting in bluish-black reaction product (sections were first incubated with avidin biotinylated-horseradish peroxidase complex (ABC, Vector Laboratories Burlingame, CA), then developed with DABNi). To examine the GABA content of the labeled axons, the tracer was visualized using a preembedding gold method followed by a postembedding anti-GABA reaction (Bodor et al., 2008). Briefly, sections were first incubated with ABC followed by a signal amplification step using biotinylated tyramide reagent (PerkinElmer Life Sciences, Boston, MA) than incubated in 1 nm gold-conjugated streptavidin (Aurion, Wageningen, Netherlands), postfixed in 2% glutaraldehyde, then silver intensified with Aurion R-Gent intensification kit. Light microscopic images were taken only from DAB stained sections on 100x oil immersion objective.

Electron microscopy

DABNi and gold labeled sections were treated with OsO4 and uranyl acetate, dehydrated in ethanol and propylene oxide and embedded in Durcupan (ACM, Fluka, Buchs Switzerland). From DABNi and gold labeled tissue 60 nm thick (silver color) ultrathin sections were cut, and sections were mounted on copper single slot grids. Finally, sections were stained with lead citrate and washed in distilled water. Postembedding GABA immunostaining was carried out on gold-labeled tissue mounted on nickel grids according to Somogyi et al. (1985). Electron micrographs were taken using a camera (Morada, Olympus) connected to an electron microscope (JEM 1200 EXII, JEOL, Tokyo). In Fig 4D, the tracer BDA was visualized with preembedding streptavidin gold labeling. The ultra-small gold particles were silver intensified (larger magnification of the area indicated by the black square shown in inset). This method reveals morphological details of the large axons, such as the structure of the filaments and myelin.

Data analysis

Spike times were analyzed using Matlab 7.0.1 software (MathWorks, Natick, MA). For each cell, we calculated spontaneous firing rate, interspike interval (ISI) distribution, and Peri-Stimulus Time Histogram (PSTH) of the response to HVC stimulation. While the PSTHs displayed in Fig. 1, ,2,2, ,5,5, and and77 have a 1 ms bin and are not smoothed, further analyses were performed on PSTHs smoothed as follows. For each trial, the firing rate time course was determined with 1 ms resolution by convolving the spike train with a Gaussian kernel of width 1 ms (Baker and Gerstein, 2001). The mean and SD of the spontaneous rate were determined over the 100 ms preceding stimulation. A neuron was considered to display a significant response if at least two consecutive bins of the PSTH were beyond limits defined by the spontaneous mean ± 2.5 SD. Responses were often made up of several components (especially in Area X pallidal cells), some inhibitory and some excitatory. We defined the beginning of the response component as the time of the first of two consecutive bins of the PSTH in which the firing rate fell outside significance limits; similarly, the end occurred when two consecutive bins fell back within significance limits. In DLM and LMAN, the response latency was defined as the latency to the first component of the response, while response duration was summed over all components. For pallidal neurons, we defined the latency to excitatory or inhibitory components as the latency to the first such component. The duration of excitatory or inhibitory components was the sum of the duration over all such components, respectively. Response strength was calculated over a response window as follows. First, the area of the PSTH above (or below) baseline firing rate was calculated as the sum of the differences between the PSTH bins above (or below) baseline firing and the mean baseline firing. These areas were then divided by the population average spontaneous firing rate multiplied by the length of the response window (35 ms in Area X, 60 ms in DLM). The result was called excitation (or inhibition) strength and expressed as a percentage of baseline firing. For DLM neurons, Area X putative interneurons and pallidal neurons displaying only excitatory responses, the lowest stimulation current intensity evoking a reliable response (at least one additional spike in each trial) was selected for further analysis. For pallidal neurons displaying some inhibition in response to stimulation, the lowest current intensity evoking an inhibitory component in their response was selected for further analysis. For responses in DLM neurons, the probability of response was calculated as the probability that a given trial contained at least one spike between 0 and 60 ms after HVC stimulation. The jitter in the response was defined as the standard deviation of the time of the first spike in this window. For paired recordings in DLM, we computed the distribution of time lags between each thalamic spike and the immediately preceding pallidal spike.

Figure 2
The response of DLM neurons to HVC electrical stimulation
Figure 5
Response of pallidal neurons to HVC electrical stimulation
Figure 7
Responses of putative Area X interneuron to HVC electrical stimulation


Numerical values are given as mean ± SD, unless stated otherwise. Response latency, strength and duration before and after drug injections were compared using a paired t-test. In addition, for each cell, spontaneous activity over multiple trials was compared before and after drug injection using a paired t-test. Axon-size measurements were made with IMAGE J 1.40g and datasets were compared using a t-test.

DLM neuron model

We built a single compartment conductance based model of a DLM neuron using the NEURON simulation package (Hines, 1998). The parameters were based on the thalamic relay neuron models of McCormick and Huguenard (1992) and Destexhe et al. (1996). Channel conductances were tuned by hand such that model responses to current injection resembled those recorded (Luo and Perkel, 1999b). Soma diameter was 15 μm. Parameters for voltage-dependent Na and K currents were unchanged from the model of Destexhe et al. (1996) except that “vtraub” was −45 mV. Maximal conductance density was 3 mS/cm2 for Na and 5 mS/cm2 for K. Parameters for low-threshold Ca and H currents followed Destexhe et al. (1996) except that Ca dynamics were excluded and H current voltage-dependence and kinetics were adjusted to match data of Luo and Perkel (2002). Maximal conductance for Ca was 4 mS/cm2 and for H current was 0.1 mS/cm2.

Intrinsic properties were then fixed and a single inhibitory synaptic input was added. Reversal potential was set at −95 mV (Person and Perkel, 2005). The conductance waveform was the difference between two exponentials, in the form


where τrise was 1.3 ms and τdecay was 9 ms (Luo and Perkel, 2002). G was set to give a peak conductance of 12.8 nS (Luo and Perkel, 2002). No short-term synaptic plasticity was included.

Note concerning nomenclature

According to the revised nomenclature of the avian brain (Reiner et al., 2004b), nuclei HVC and LMAN lie in the avian pallium, the structure overlying the BG. Mammalian cortex is also a pallial structure, along with the amygdala and claustrum. Although there remains some controversy regarding the precise evolutionary relationship between mammalian cortex and these avian pallial structures, we refer here to HVC and LMAN as “cortical”. In addition, we refer to the spontaneously active output neurons of Area X as “pallidal cells” because they share many features with mammalian pallidal neurons (Farries et al., 2002). It is important to note that they also differ in some respects from mammalian pallidal neurons (Carrillo and Doupe, 2004; Gale and Perkel, 2009). Such simplified nomenclature is intended to make this article more accessible to researchers interested in BG physiology but unfamiliar with the avian anatomy.


LMAN and DLM neurons respond with very short latency to HVC stimulation

To measure propagation speed through the AFP, we made extracellular single-unit recordings from 26 neurons in LMAN and 50 neurons in DLM displaying a response to electrical stimulation of cortical nucleus HVC.

In both nuclei, spontaneous activity was low (4 ± 6 sp/s in LMAN, 2 ± 3 sp/s in DLM) and irregular when it was present at all. The coefficient of variation (CV) of inter-spike-intervals (ISI) was 1.3 ± 0.5 in 26/26 LMAN neurons and 0.9 ± 0.4 in 36/50 DLM neurons. HVC stimulation evoked increases in firing in all LMAN neurons (Fig. 1B, C). The minimal stimulation current necessary to evoke responses in LMAN was 200 μA. The latency of these responses was surprisingly short in most neurons (Fig. 1E), with a median latency of 16 ms. As a result, the population average PSTH undergoes a sharp rise starting at 11 ms and peaks at 17 ms after HVC stimulation (Fig. 1D). Firing irregularity remained high during the responses, with a CV of firing of 1.3 ± 0.5 in the 0–500 ms window following stimulation. Responses were usually reliable; the response probability was 0.6 ± 0.3 (range 0.2–1). The timing of these responses was both precise and brief, with a 9 ± 5 ms jitter in the time of the first evoked spike. The response duration, defined as the number of consecutive bins beyond significance in the PSTH (see methods), was 27 ± 16 ms. While latency to LMAN responses informed us about the overall transmission delays in the AFP, we decided to investigate DLM responses to reveal the mechanisms allowing fast signal propagation through the BG.

HVC stimulation also evoked rapid increases in firing in all but one DLM neurons (Fig. 2B, C). The minimal stimulation necessary to evoke responses in DLM was 100 μA. The median latency of these responses was 17 ms (Fig. 2D), and the population average PSTH rose at 10 ms and peaked at 18 ms after stimulation (Fig. 2E). The similarity in the median latency in LMAN and DLM is surprising at first glance. However, given the convergence of synaptic inputs from many DLM neurons to a single LMAN neuron, each LMAN neuron is expected to receive input from a pool of DLM neurons. In response to HVC stimulation, the DLM neurons displaying short latencies might drive fast responses in most LMAN neurons. Although DLM responses consisting of single spikes or doublets were stable and consistent over time, they were usually not elicited on each trial. The probability of response among the recorded neurons averaged 0.3 ± 0.3 (range 0.04 – 1, Fig. 2F), and the firing variability thus remained high in response to HVC stimulation (ISI CV of 0.8 ± 0.3). However, the timing of these responses, when elicited, was both precise and brief, with a 10 ± 6 ms jitter in the time of the first evoked spike. The response duration, defined as the number of consecutive bins beyond significance in the PSTH (see methods), was only 10 ± 10 ms.

In summary, despite low response probability, DLM neurons display fast and precise responses to HVC stimulation. The response latencies in LMAN and DLM are shorter than were anticipated given the number of synaptic connections between HVC and DLM and the disinhibitory mechanism involved.

Thalamic firing is dominated by the inhibitory pallidal input

The main input to DLM is GABAergic and arises from the BG (Vates et al., 1997; Luo and Perkel, 1999a; Person et al., 2008). Each DLM neuron receives input from a single or, at most, two large, aspiny Area X neurons (Okuhata and Saito, 1987; Luo and Perkel, 1999a). These neurons display high spontaneous firing rates and are termed pallidal neurons due to their resemblance to neurons in the mammalian pallidum (Farries et al., 2002; Reiner et al., 2004a). The large one-to-one calyx-like terminals formed by Area X pallidal neurons around DLM somata strongly hyperpolarize DLM neurons (Luo and Perkel, 1999b; Person and Perkel, 2005). Moreover, this configuration allows simultaneous recording of pairs of the presynaptic pallidal and postsynaptic thalamic units (Person and Perkel, 2007), and we recorded the activity in 14 such pairs. The spike times of thalamic neurons in these pairs were very strongly constrained by the firing pattern of their presynaptic pallidal terminals. Indeed, in 11/14 pairs, at least 90% of the thalamic spikes occurred more than 10 ms after the preceding pallidal spike (Fig. 3A). As a result, the probability of firing of thalamic neurons was very low 0–10 ms following a pallidal spike. Their firing probability increased dramatically for longer delays after the last pallidal spike (Fig. 3B). This suggests that a thalamic neuron could fire only if the presynaptic pallidal ISI was long enough.

Figure 3
Relation between firing and pallidal input activity in a DLM neurons

Most pallidal neurons project to the thalamus and have very short conduction latency

Since the activity of DLM cells is dominated by their inhibitory somatic input from a single pallidal neuron, we investigated how short latency responses to HVC stimulation in DLM could be mediated by pallidal neurons. We first measured the time it takes action potentials to propagate between Area X and DLM. We recorded from pallidal neurons in Area X (see Methods and Fig. 4A for a comparison with terminals in DLM) and antidromically stimulated in DLM (Fig. 4B). We found that 67% (20/30) of the high-frequency spontaneously active neurons in Area X displayed antidromic activation in response to DLM stimulation, with a latency of 0.85 ± 0.24 ms (latency variability of 62 ± 30 μs). In contrast, axons from HVC to Area X have conduction latency around 5 ms (Hahnloser et al., 2006) over a similar distance as that from Area X to DLM. In summary, at least 2/3 of Area X pallidal neurons appear to send an axon to DLM. Moreover, the axons from Area X to DLM conduct rapidly.

Pallidal projection neurons have unusually large axons

Consistent with the short conduction delays observed between Area X and DLM, the diameter of these axons was very large (Fig. 4). Light microscopy of axons anterogradely labeled from Area X revealed diameters of 2.1 ± 0.7 μm as the axons entered DLM. Within DLM, axon diameters became even larger. Light and electron microscopy (EM) revealed diameters over 2.5 microns, (light: 3.0 ± 0.8 μm, n=63; EM: 2.7 ± 0.6 μm; n=9). In addition, these large Area X axons remain myelinated in DLM (Fig. 4C, D). These large diameters contrast with other axons connecting song-system nuclei. Retrogradely labeled axons from LMAN to Area X, from the same injections used to label axons projecting to DLM, had significantly smaller diameters of 0.68 ± 0.33 μm (p<0.001, n=97). Comparing the conduction velocity of Area X-to-DLM neurons with large axons to smaller gauge axons between LMAN or HVC and Area X (LMAN: Fig. 4E; HVC: unpublished observation) suggests that the large diameter of these axons saves at least 3 ms, or 20% of the transmission delay from HVC to LMAN.

Pallidal neurons display fast inhibition in response to HVC stimulation

To understand how much time processing within Area X contributes to propagation delays between HVC and DLM, we investigated the response of pallidal neurons in Area X to HVC electrical stimulation. As expected, pallidal neurons recorded at the soma in Area X or the terminal in DLM exhibited rapid spontaneous activity (62.4 ± 15.6 sp/s, n=110), which, unlike DLM firing, was quite regular (ISI CV of 0.38 ± 0.16).

In response to low intensity stimulation in HVC (40–100 μA, 0.2 ms pulses), pallidal neurons displayed short duration excitation, with a latency of 8.0 ± 2.3 ms and a duration of 8.6 ± 4.6 ms (Fig. 5A). Note that such low stimulation intensities usually did not evoke any response in DLM. When stimulation intensity was increased (>100 μA), excitation became stronger, and, in many neurons, an inhibitory effect of stimulation appeared (Fig. 5B, C, D). The minimal stimulation current evoking feedforward inhibition in Area X pallidal cells was 200 μA. Surprisingly, inhibition was sometimes fast and often appeared prior to excitation (in 48.8% of the neurons displaying inhibition, Fig. 5D). Short duration excitation is likely mediated by monosynaptic connections from HVC whereas inhibition is likely disynaptic, and observing the inhibitory response prior to excitation was thus unexpected. Such responses were recorded in pallidal neurons recorded either in Area X (somata) or in DLM (terminals). Because responses were very similar in both groups (see Fig. 6 and Table 1), we pooled the data for further analysis. As a result of combined excitatory and inhibitory effects, the first excitatory component of the pallidal neuron response had broadly distributed latencies (median 10 ms, range 4 – 25 ms, n=89, Fig. 5H, Table 2), and the latency to the early inhibitory component was just as short (median 10 ms, range 5–28 ms, n=44, Fig. 5G, Table 2).

Figure 6
Responses to HVC electrical stimulation in pallidal neurons recorded in DLM (terminals) and Area X (somata)
Table 1
Ratio of neurons displaying excitation and inhibition in pallidal neurons recorded in DLM (terminals) and Area X (somata)
Table 2
Response pattern to HVC electrical stimulation in pallidal neurons

Interestingly, relatively high stimulation intensities (>100 μA) are necessary to evoke inhibitory components in the responses of pallidal neurons and to induce responses in downstream nuclei. Such intensities might be necessary to activate simultaneously a physiologically relevant number of neurons in HVC. Indeed, we estimated that a stimulation current of 200 μA (sufficient to evoke an inhibitory component in many pallidal neurons) applied in the center of HVC is activating at most 5% of all HVC neurons (see methods). Because neurons projecting to RA or X and interneurons are distributed homogeneously over the nucleus (Wild et al., 2005), we expect such stimulation to activate 5% of HVC neurons projecting to Area X (HVCX). During singing, almost all (80%) HVCX neurons fire 1–4 bursts of 6 ms average duration during each song motif (Kozhevnikov and Fee, 2007), corresponding to an average activation period of 10–20 ms, or 2–4% of the duration of the motif (about 500 ms). If all HVCX neurons were firing independently, 2–4% of them would thus be simultaneously activated at any given time point in the song. As is apparent from Fig. 2A in Kozhevnikov and Fee (2007), singing-related activity bursts seem often to happen simultaneously in 2–3 out of the 24 HVCX neurons recorded in that bird. The ratio of simultaneously activated HVCX neurons might thus be higher than our estimate (closer to 10%), at least at certain times during singing. The strong synchronized drive sent from HVC to the BG following electrical stimulation with high current (>100 μA) thus seems to involve similar number of HVC neurons as singing-related activation.

We then examined the distribution of the first few (2–3) ISIs evoked by HVC stimulation in pallidal neurons displaying an inhibitory component in their response. Their distribution extended over a range similar to that of ISIs evoked by song playback in those neurons (Person and Perkel, 2007), confirming that HVC electrical stimulation evoked behaviorally relevant activation patterns in these neurons. Because of an increased number of shorter ISIs following stimulation, the median ISI duration was shorter in response to HVC stimulation than at baseline (13 ± 6 ms vs 17 ± 5 ms). However, longer ISIs (>30 ms) were also more likely to occur following the stimulation than during episodes of spontaneous activity (11.8 ± 14.0% vs 7.3 ± 11.8%, p=0.03, Fig. 5E and F). Since longer pallidal ISIs are associated with higher firing probability in DLM neurons, this response pattern in pallidal neurons might evoke increased firing in DLM.

While the exact relationship between the responses to HVC stimulation evoked in a pallidal cell and its target thalamic neuron could not be investigated here, some relationships can be inferred. On one hand, fast responses in DLM neurons, which have latencies shorter than the average pallidal ISI, might only involve the lengthening of a single pallidal ISI and therefore most likely rely on early inhibitory responses in pallidal neurons. On the other hand, slower responses in DLM neurons could be associated with more complex pallidal response patterns. As suggested by Person and Perkel (2007), fast deceleration associated with an excitation-inhibition response profile could activate thalamic neurons, and may thus underlie at least some of the slower responses in DLM neurons. Finally, late excitatory components of pallidal responses might be involved in the truncation of DLM responses, which could serve to maintain or improve timing precision.

Together, our results indicate that Area X pallidal neurons often display a fast inhibition in response to HVC stimulation, and this short inhibitory response would be expected to disinhibit a target DLM neuron, and increase its firing probability.

Putative Area X interneurons exhibit short-latency response to HVC stimulation

The spontaneously active pallidal neurons constitute only a minority of Area X neurons. The vast majority of neurons within Area X are striatal spiny neurons, which are homologous to mammalian medium spiny neurons (Farries et al., 2002; Reiner et al., 2004a) but do not project outside of Area X. Other interneuron types within Area X include fast spiking inhibitory interneurons and cholinergic interneurons (Farries et al., 2002), which, in mammals, display spontaneous activities of 0–10 Hz (Mallet et al., 2006) and 2–10 Hz (Aosaki et al., 1994) respectively. Area X interneuron populations are generally thought to display little or no spontaneous activity (Farries and Perkel, 2002), which distinguishes them from Area X’s only known projection neuron population, the pallidal neurons. We recorded the response to HVC electrical stimulation in 15 neurons with spontaneous firing below 25 sp/s (mean 11.5 ± 7.8 sp/s). These cells displayed very rapid responses to HVC stimulation, with most cells displaying latencies shorter than 5 ms (mean 5.9 ± 3.8 ms, Fig. 7). Their response was long in duration (25 ± 28 ms) and very reliable (response probability of 0.8 ± 0.3). Overall, Area X putative interneurons, at least some of which are inhibitory and project to pallildal neurons (Farries et al., 2005), display very short latency and long duration responses to HVC stimulation, consistent with a role in conveying fast feedforward inhibition onto pallidal neurons that project to DLM.

Area X responses to HVC stimulation are mediated by glutamatergic input

To confirm that the responses to HVC stimulation in Area X neurons are mediated by the known glutamatergic afferents from HVC (Farries et al., 2005), we injected the glutamate receptor blockers NBQX and APV into Area X. Infusion of these drugs completely suppressed all Area X neuron responses to HVC stimulation (Fig. 8A, B), consistent with responses in Area X being mediated by direct excitatory projection from HVC to Area X neurons. Response strength (see methods) was decreased by 98.5 ± 3.0% (130 ± 91% vs 2 ± 3% of baseline, n=4, p=0.03) after drug infusion. In contrast, responses to HVC stimulation in Area X cells did not change after saline injection in Area X (response strength from 56 ± 27% to 52 ± 38%, n=3, p=0.9). We therefore confirm, as suggested by Farries et al. (2005), that excitatory afferents from HVC drive fast excitation in Area X neurons.

Figure 8
Direct excitation from HVC in Area X neurons is mediated by glutamate, while feedforward inhibition of pallidal neurons is mediated by GABAA receptors and shapes direct excitation

Local inhibition in Area X mediates fast inhibition of pallidal neurons

If inhibitory responses to HVC stimulation in pallidal neurons are mediated by Area X GABAergic interneurons, they should be suppressed when GABAergic transmission is blocked in Area X. In pallidal cells, injection of gabazine, a reversible GABAA receptor blocker, into Area X had several effects. First, drug application significantly increased spontaneous activity of the recorded pallidal cells from 48.2 ± 23.8 sp/s to 73.9 ± 36.9 sp/s (n=10, p=0.03). In addition, the strength of the inhibitory component of the response to HVC stimulation was strongly decreased (from 11 ± 5% to 4 ± 6% sp/trial, p=0.01, n=4, Fig. 8C, D) and its duration was not significantly changed (from 6 ± 2 ms to 3 ± 6 ms, p=0.3). In contrast, the excitatory component was significantly stronger (from 63 ± 30% to 206 ± 158%, n=10, p=0.01) and longer in duration (from 13 ± 6 ms to 35 ± 26 ms, p=0.02). Close examination of excitatory response latencies calculated before and after drug infusion indicated that early inhibition, even when it was not visible in the PSTHs summarizing a cell’s response to HVC stimulation, delayed the excitatory response to stimulation in these cells (Fig. 8C, D). As a result, latency of the excitatory component significantly decreased after gabazine injection (from 12.4 ± 3.9 ms to 10.9 ± 4.4 ms, p=0.03). This was also apparent in average PSTHs of all cells recorded after injection of gabazine into Area X aligned to HVC stimulation (Fig. 8E), showing that early inhibition delays the population response among pallidal cells to stimulation while late inhibition shortens the response.

These results indicate that the local inhibitory network in Area X mediates feedforward inhibition onto pallidal neurons. In half (48.8%) of the cells displaying inhibition, feedforward inhibition arrived before direct excitation and could create sufficiently long ISIs in the input to DLM neurons to enable them to fire.

Local inhibition in Area X is necessary for DLM responses

If the short latency response in DLM to HVC stimulation is evoked via a fast inhibition of the Area X pallidal neurons, it should be suppressed when GABAergic transmission is blocked in Area X. To test this hypothesis, we recorded DLM neuron responses to HVC stimulation while pharmacologically manipulating neurotransmission in Area X. Injection of the GABAA receptor blocker gabazine into Area X caused no consistent change in DLM neuron spontaneous firing rate (5 ± 5 sp/s to 2 ± 2 sp/s, n=8, p=0.2). However, the response of all DLM neurons to HVC stimulation was suppressed by gabazine injection in Area X, and response strength decreased by 93.4 ± 8.8% (240% ± 180% vs 17 ± 20%, n=8, p=0.02, Fig. 9A, B). In contrast, after saline injection in Area X DLM responses to HVC stimulation remained unchanged (response strength: 330 ± 220% vs 330 ± 220%, n=3, p=0.9). The local inhibitory network in Area X is thus essential for the response of thalamic neurons to HVC stimulation.

Figure 9
GABAA receptors in Area X are essential for the response of DLM neurons to HVC electrical stimulation

Glutamatergic transmission to the thalamus is not necessary for DLM responses

Our results suggest that the response of DLM neurons is mainly driven by changes in their GABAergic input from pallidal cells. Alternatively, fast transmission between HVC and DLM could be mediated through a polysynaptic excitatory pathway between these structures. Previous anatomical studies have shown that RA, which receives direct excitatory input from HVC (Mooney, 1992; Kubota and Saito, 1991; Stark and Perkel, 1999), projects to DLM (Foster et al., 1997; Vates et al., 1997; Wild, 1993). Furthermore, DLM neurons show AMPA receptor-mediated EPSPs (Luo and Perkel, 1999b). Although the RA-to-DLM connection appears anatomically to be much weaker than the primary connections of the song system (Vates et al., 1997), it is a putative candidate for mediating fast transmission between HVC and DLM.

We tested whether glutamatergic transmission mediates fast responses in DLM by injecting the AMPA receptor blocker NBQX and the NMDA receptor blocker APV. The spontaneous firing rate was significantly decreased in the presence of these drugs (3.6 ± 3.6 sp/s to 1.9 ± 2.5 sp/s, n=7, p=0.03, Fig. 10A). In most DLM neurons, however, the response to HVC stimulation was not affected (Fig. 10B, C). The response strength was not significantly changed (140 ± 120% vs 90 ± 80%, n=7, p=0.4), nor was the response latency (28 ± 22 ms to 27 ± 23 ms, p=0.7). Moreover, when the analysis was restricted to neurons displaying a significant change in spontaneous firing rate after application of glutamate receptor blockers, similar results were obtained. Injection of vehicle solution into DLM (0.9% saline with 0.2% DMSO and 0.5% Alexa488) did not modify either spontaneous activity (1.2 ± 0.8 sp/s to 1.2 ± 0.9 sp/s, n=3, p=0.9), or responses to HVC stimulation (response strength: 220 ± 20% vs 200 ± 60%, p=0.6), ruling out the possibility that our vehicle solution had any effect on DLM neurons. In summary, responses evoked in DLM by HVC stimulation persisted when glutamatergic transmission was blocked in DLM but were altered by manipulations in Area X, which makes a GABAergic projection to DLM.

Figure 10
Glutamate receptors are not essential for the response of DLM neurons to HVC electrical stimulation

Very short latency probabilistic responses to HVC stimulation in a simple DLM neuron model

As shown above, DLM responses to HVC stimulation are mediated by fast disinhibition through Area X. Since these responses do not rely on excitatory inputs to DLM, they most probably involve postinhibitory rebound in the thalamic neurons. However, the response latency to HVC stimulation is sometimes as short as the rebound latency extrapolated from in vitro data (Person and Perkel, 2005). How can the rebound process not add an additional delay? We investigated the mechanisms of rebound-driven responses in the thalamic neurons using a conductance-based model of a DLM neuron. In this model, we generated theoretical responses to the pallidal spike trains recorded in response to HVC stimulation in 23 randomly selected pallidal neurons displaying an inhibitory response component (Fig. 11D). The spike trains drove an inhibitory synaptic conductance in a model DLM neuron. We measured model DLM neuron spike times and analyzed the responses relative to the times of HVC stimulation with the same analyses applied to experimentally recorded spike trains. In 17/23 cases (representing 23 unique pallidal trains), we found a significant increase in DLM model neuron activity following HVC stimulation, and the distribution of the response latencies was very similar to that experimentally recorded in DLM neurons. In particular the distribution was dominated by short latency responses (mean 24.6 ± 32.3 ms, median 14 ms, Fig. 11E). Moreover, like the recorded neurons, the probability of a response among the simulated DLM neurons was low, ranging from 0.04 to 0.9, averaging 0.25 ± 0.29. This simple test reveals that the HVC-evoked inhibition in Area X pallidal cells could evoke rebound firing of suitable probability and latency in a typical DLM neuron.

Figure 11
Working model of information transmission between HVC and DLM


We report here very fast signal propagation through an avian cortical-BG-thalamo-cortical pathway, with overall transmission delays as short as 10 ms. A rapid disinhibitory process mediates this fast propagation; through feedforward inhibition, ISIs of tonically active pallidal neurons lengthen slightly, transiently interrupting the sustained inhibition of thalamic neurons and greatly increasing their firing probability. Large-diameter axons projecting from pallidal neurons to the thalamus may contribute to the speed of this process. Finally, thalamic responses do not depend on glutamatergic transmission, and thus most likely involve postinhibitory rebound. These data reveal a new view of BG transmission whereby pallidal neurons continually send veto signals that arrive just in time to prevent postsynaptic firing. Omission or even slight delay of that inhibitory input triggers a nearly immediate thalamic response.

Mechanism of fast response in DLM

The main input to the thalamic nucleus DLM is GABAergic and consists of large calyx-like terminals of Area X pallidal neurons onto the DLM soma (Okuhata and Saito, 1987; Vates et al., 1997; Luo and Perkel, 1999a). In vitro, pallidal, inhibitory postsynaptic potentials (IPSPs) strongly hyperpolarize DLM neurons, which then generate rebound action potentials with a 20–40 ms latency (Luo and Perkel, 1999b; Person and Perkel, 2005). This latency could be shorter in vivo, where spikes in DLM neurons are consistently preceded by 10–20 ms intervals of silence of the inhibitory presynaptic terminals (Person and Perkel, 2007). These studies suggested that postinhibitory rebound mediates DLM firing in vivo, but stopped short of pharmacological manipulation to test the contribution of pallidal firing to DLM activity.

Here, we provide evidence that feedforward GABAergic inhibition triggered by HVC input onto pallidal neurons is necessary and sufficient to trigger a DLM response. In the absence of HVC input, the rapid and regular spontaneous activity of pallidal neurons induces IPSPs separated by 10–25 ms intervals, imposing a constant veto on DLM neuron firing. Presumably, this inhibition also hyperpolarizes the membrane potential to voltages that assure the availability of low-threshold calcium channels that support postinhibitory rebound.

A synchronized drive from HVC triggers fast feedforward inhibition of pallidal neurons in Area X. If the feedforward inhibition arrives shortly after the last pallidal spike, it will have little effect (Fig. 11B). If that inhibition arrives later in the pallidal ISI, it will delay the next spike (Fig. 11C), and trigger an almost instantaneous spike in the target DLM neuron. Indeed, DLM neuron membrane potential rises following each presynaptic spike until it eventually reaches threshold unless an IPSP arrives to stop it (Person and Perkel, 2005). Note that, in vivo, the rise in membrane potential might result from a combination of postinhibitory rebound and concurrent excitatory inputs, although the latter appear not to be necessary for DLM responses.

Increased firing rates preceding long ISIs in the presynaptic pallidal terminal might be important for DLM neuron firing (Person and Perkel, 2007; Kojima and Doupe, 2009). Consistent with intracellular recordings (Person and Perkel, 2005), we propose that increases in pallidal firing rate, such as those observed during song playback, shorten the delay to DLM rebound spikes and increase the firing probability in response to a given presynaptic ISI. However, our results show that such an increase in activity preceding long ISIs is neither necessary nor sufficient to evoke DLM responses, pointing to a critical role for BG feedforward inhibition.

While we have shown that short pauses (~20–50 ms) in pallidal activity allow fast thalamic responses, longer inactivation of pallidal neurons (>seconds) also leads to increased spontaneous thalamic activity (Kojima and Doupe, 2009). Whether this “ungated” thalamic activity relies on excitatory synaptic drive or on DLM neuron excitability remains to be determined.

Based on our results, information propagating through the AFP might reach RA faster than was anticipated previously (Kimpo et al., 2003). In that study, peaks of correlated activity between HVC and LMAN led to estimated propagation latencies of around 60ms. However, many paired recordings showed only a single broad correlation peak, possibly reflecting faster transmission through the AFP. Our approach using electrical stimulation reduces ambiguity in the signal start time, and led to our observation of much faster propagation speeds than observed previously.

Relation to function

BG circuits in general and the AFP in particular are proposed to introduce variability necessary for exploration during motor learning (Graybiel, 2005; Kao et al., 2005; Ölveczky et al., 2005). Where within the neural circuit does this variability arise? LMAN neuron spontaneous activity is very irregular, suggesting one source. Moreover, we show here that the response of DLM neurons to activation of cortical areas is highly unreliable. The success or failure of a response, determined by the relative timing of the HVC input to the preceding pallidal spike (Fig. 11), could introduce variability in LMAN response timing. Indeed, assuming LMAN neurons receive convergent input from many DLM neurons, they may be activated with different latencies depending on which population of presynaptic neurons is active.

In addition to introducing variability, the AFP may provide patterned signals to guide changes in motor output (Kao et al., 2008). Such signals would most likely come from HVC and be transformed in the AFP, potentially through dopamine-dependent mechanisms. Indeed, dopamine could alter information flow through Area X depending on the social context (Sasaki et al., 2006) through short- or long-term synaptic effects (Ding et al., 2004). In this view, an overlap in RA with the “production signal” sent through the monosynaptic HVC-RA pathway would provide the AFP signal with a simple mechanism to alter online behavior. Because song-related signals are very short (10–20 ms, Hahnloser et al., 2002; Kozhevnikov and Fee, 2007), there is not much time for transmission through the AFP. Our results suggest that latencies through the AFP might be shorter than 20 ms and would thus allow interaction of motor pathway and AFP signals within RA.

Comparison with mammalian BG

Area X differs from mammalian BG in its gross anatomical structure, but displays similar circuitry at a finer scale. Indeed, Area X is comprised primarily of striatal neurons, but consistent with mammalian circuitry it is its pallidal component that directly projects to the thalamus (Bottjer et al., 1989; Luo and Perkel, 1999a; Farries and Perkel, 2002; Carrillo and Doupe, 2004; Reiner at al., 2004a). It contains the anatomical substrate for a disinhibitory pathway homologous to the direct pathway through mammalian BG (Farries et al., 2005; Reiner et al., 2004a), which, as shown here, is functionally critical. Indeed, afferents from the cortical structure HVC drive feedforward inhibition in BG output through striatal neurons and disinhibit thalamic target neurons. Interestingly, avian striatal inhibitory neurons may require high synchronization of their cortical inputs to drive such feedforward inhibition, similar to striatal medium spiny neurons (Charpier et al, 1999). In contrast to mammalian pallidal neurons, Area X output neurons also receive monosynaptic input from cortical glutamatergic afferents (Farries et al., 2005). This input provides a powerful and sensitive drive to pallidal neurons, with broadly distributed latencies, consistent with previous estimations of transmission delays from HVC to Area X (Hahnloser et al, 2006). This connection might provide Area X with a pathway functionally equivalent to the “hyperdirect” pathway through the subthalamic nucleus (STN) in mammals (Nambu et al., 2000), which is not connected with Area X (Person et al., 2008). This analogue of the hyperdirect pathway may increase the selectivity of disinhibitory signals transmitted through the direct pathway both spatially, by broadly inhibiting DLM during focused disinhibition, and temporally, by limiting the duration of the disinhibitory signal (Nambu et al., 2000).

Speed specialization

The AFP may be an example of a BG circuit specialized for rapid processing of cortical inputs. First, BG output axons have very large diameters (3 μm) for vertebrate central nervous system neurons, especially compared to their mammalian counterparts (<1μm in rats, Bodor et al., 2008) or to other axons in the zebra finch brain (<1μm), allowing fast conduction. Secondly, in contrast with the mammalian BG motor loop, in which transmission along the direct pathway is slower than feedforward excitation through the STN (Nambu et al., 2000), feedforward inhibition is surprisingly fast in the song-related BG. Indeed, it arrives fast enough to prevent or slow the rise of monosynaptic excitation from cortical structures. This rapid feedforward inhibition is mediated by GABAergic Area X neurons, most likely interneuron. In particular, it might involve the fast-spiking interneurons, which provide fast feedforward inhibition in the mammalian striatum (Mallet et al., 2005) and are present in Area X (Farries et al., 2002; Reiner et al., 2004a).

The mammalian pallido-thalamic connection displays powerful GABAergic synapses similar to those seen in songbirds (Bodor et al., 2008; Wanaverbecq et al., 2008), the cellular mechanisms described here may well be at least partially available in mammals. While motor output often operates on longer time scales (>100ms) in mammals, specific BG systems involved in fast sensorimotor tasks such as speech in humans or whisking in rats are subjected to similar constraints as the AFP. Although information propagation has not yet been carefully characterized in these circuits, they may display similar specializations for fast transmission as those presented here.


We thank F. Rieke, G. Murphy, L. Acsády, D. Hansel, T. Boraud, and members of the Perkel Lab for their comments on the manuscript. The GABA antibody for postembedding reaction was kindly donated by Dr Peter Somogyi. This work was supported by NIH grant R01-MH066126 to D.J.P and a grant Lavoisier (Ministère des Affaires Etrangères, France) to A.L.


  • Aosaki T, Tsubokawa H, Ishida A, Watanabe K, Graybiel AM, Kimura M. Responses of tonically active neurons in the primate’s striatum undergo systematic changes during behavioral sensorimotor conditioning. J Neurosci. 1994;14:3969–84. [PubMed]
  • Bagshaw EV, Evans MH. Measurement of current spread from microelectrodes when stimulating within the nervous system. Exp Brain Res. 1976;25:391–400. [PubMed]
  • Bodor AL, Giber K, Rovó Z, Ulbert I, Acsády L. Structural correlates of efficient GABAergic transmission in the basal ganglia-thalamus pathway. J Neurosci. 2008;28:3090–102. [PMC free article] [PubMed]
  • Bottjer SW, Miesner EA, Arnold A. Forebrain lesions disrupt development but not maintenance of song in passerine birds. Science. 1984;224:901–3. [PubMed]
  • Bottjer SW, Halsema KA, Brown SA, Miesner EA. Axonal connections of a forebrain nucleus involved with vocal learning in zebra finches. J Comp Neurol. 1989;279:312–26. [PubMed]
  • Broussard DM, Brontë-Stewart HM, Lisberger SG. Expression of motor learning in the response of the primate vestibuloocular reflex pathway to electrical stimulation. J Neurophysiol. 1992;67:1493–508. [PubMed]
  • Carrillo GD, Doupe AJ. Is the songbird Area X striatal, pallidal, or both? An anatomical study. J Comp Neurol. 2004;473:415–37. [PubMed]
  • Charpier S, Mahon S, Deniau JM. In vivo induction of striatal long-term potentiation by low-frequency stimulation of the cerebral cortex. Neuroscience. 1999;91:1209–22. [PubMed]
  • Deniau JM, Chevalier G. Disinhibition as a basic process in the expression of striatal functions. II The striato-nigral influence on thalamocortical cells of the ventromedial thalamic nucleus. Brain Res. 1985;334:227–33. [PubMed]
  • Destexhe A, Bal T, McCormick DA, Sejnowski TJ. Ionic mechanisms underlying synchronized oscillations and propagating waves in a model of ferret thalamic slices. J Neurophysiol. 1996;76:2049–70. [PubMed]
  • Ding L, Perkel DJ. Long-term potentiation in an avian basal ganglia nucleus essential for vocal learning. J Neurosci. 2004;24:488–94. [PubMed]
  • Farries MA, Perkel DJ. A telencephalic nucleus essential for song learning contains neurons with physiological characteristics of both striatum and globus pallidus. J Neurosci. 2002;22:3776–87. [PubMed]
  • Farries MA, Ding L, Perkel DJ. Evidence for “direct” and “indirect” pathways through the song system basal ganglia. J Comp Neurol. 2005;484:93–104. [PubMed]
  • Follett KA, Mann MD. Effective stimulation distance for current from macroelectrodes. Exp Neurol. 1986;92:75–91. [PubMed]
  • Foster EF, Mehta RP, Bottjer SW. Axonal connections of the medial magnocellular nucleus of the anterior neostriatum in zebra finches. J Comp Neurol. 1997;382:364–81. [PubMed]
  • Gale SD, Perkel DJ. Anatomy of a songbird basal ganglia circuit essential for vocal learning and plasticity. J Chem Neuroanat 2009 in press. [PMC free article] [PubMed]
  • Glaze CM, Troyer TW. Behavioral measurements of a temporally precise motor code for birdsong. J Neurosci. 2007;27:7631–9. [PubMed]
  • Goller F, Cooper BG. Peripheral motor dynamics of song production in the zebra finch. Ann NY Acad Sci. 2004;1016:130–52. [PubMed]
  • Graybiel AM. The basal ganglia: learning new tricks and loving it. Curr Opin Neurobiol. 2005;15:638–44. [PubMed]
  • Hahnloser RH, Kozhevnikov AA, Fee MS. An ultra-sparse code underlies the generation of neural sequences in a songbird. Nature. 2002;419:65–70. [PubMed]
  • Hahnloser RH, Kozhevnikov AA, Fee MS. Sleep-related neural activity in a premotor and a basal-ganglia pathway of the songbird. J Neurophysiol. 2006;96:794–812. [PubMed]
  • Heil P, Scheich H. Functional organization of the avian auditory cortex analogue. II Topographic distribution of latency. Brain Res. 1991;539:121–5. [PubMed]
  • Hines ML. The Neurosimulator NEURON. In: Koch C, Segev I, editors. Methods in Neuronal Modeling. Cambridge, MA: MIT Press; 1998. pp. 129–136.
  • Jarvis ED, et al. Avian Brain Nomenclature Consortium. Avian brains and a new understanding of vertebrate brain evolution. Nat Rev Neurosci. 2005;6:151–9. [PMC free article] [PubMed]
  • Kao MH, Doupe AJ, Brainard MS. Contributions of an avian basal ganglia-forebrain circuit to real-time modulation of song. Nature. 2005;433:638–43. [PubMed]
  • Kao MH, Wright BD, Doupe AJ. Neurons in a Forebrain Nucleus Required for Vocal Plasticity Rapidly Switch between Precise Firing and Variable Bursting Depending on Social Context. J Neurosci. 2008;28:13232–47. [PMC free article] [PubMed]
  • Kimpo RR, Theunissen FE, Doupe AJ. Propagation of correlated activity through multiple stages of a neural circuit. J Neurosci. 2003;23:5750–61. [PubMed]
  • Kojima S, Doupe A. Activity propagation in an avian basal ganglia-thalamocortical circuit essential for vocal learning. J Neurosci. 2009;29:4782–93. [PMC free article] [PubMed]
  • Kozhevnikov AA, Fee MS. Singing-related activity of identified HVC neurons in the zebra finch. J Neurophysiol. 2007;97:4271–83. [PubMed]
  • Kubota M, Saito N. NMDA receptors participate differentially in two different synaptic inputs in neurons of the zebra finch robust nucleus of the archistriatum in vitro. Neurosci Lett. 1991;125:107–9. [PubMed]
  • Liberman AM, Harris KS, Kinney JA, Lane H. The discrimination of relative onset-time of the components of certain speech and nonspeech patterns. J Exp Psychol. 1961;61:379–88. [PubMed]
  • Luo M, Perkel DJ. Long-range GABAergic projection in a circuit essential for vocal learning. J Comp Neurol. 1999a;403:68–84. [PubMed]
  • Luo M, Perkel DJ. A GABAergic, strongly inhibitory projection to a thalamic nucleus in the zebra finch song system. J Neurosci. 1999b;19:6700–11. [PubMed]
  • Luo M, Perkel DJ. Intrinsic and synaptic properties of neurons in an avian thalamic nucleus during song learning. J Neurophysiol. 2002;88:1903–14. [PubMed]
  • McCormick DA, Huguenard JR. A model of the electrophysiological properties of thalamocortical relay neurons. J Neurophysiol. 1992;68:1384–400. [PubMed]
  • MacDougall-Shackleton SA, Hulse SH, Ball GF. Neural correlates of singing behavior in male zebra finches (Taeniopygia guttata) J Neurobiol. 1998;36:421–30. [PubMed]
  • Maggi CA, Meli A. Suitability of urethane anesthesia for physiopharmacological investigations in various systems. Part 1: general considerations. Experentia. 1986;42:109–14. [PubMed]
  • Mallet N, Ballion B, Le Moine C, Gonon F. Cortical inputs and GABA interneurons imbalance projection neurons in the striatum of Parkinsonian rats. J Neurosci. 2006;26:3875–84. [PubMed]
  • Mooney R. Synaptic basis for developmental plasticity in a birdsong nucleus. J Neurosci. 1992;12:2464–77. [PubMed]
  • Mooney R, Konishi M. Two distinct inputs to an avian song nucleus activate different glutamate receptor subtypes on individual neurons. Proc Natl Acad Sci USA. 1991;88:4075–9. [PubMed]
  • Nambu A, Tokuno H, Hamada I, Kita H, Imanishi M, Akazawa T, Ikeuchi Y, Asegawa N. Excitatory cortical inputs to pallidal neurons via the subthalamic nucleus in the monkey. J Neurophysiol. 2000;84:289–300. [PubMed]
  • Nottebohm F, Stokes TM, Leonard CM. Central control of song in the canary, Serinus canarius. J Comp Neurol. 1976;165:457–486. [PubMed]
  • Okuhata S, Saito N. Synaptic connections of thalamo-cerebral vocal nuclei of the canary. Brain Res Bull. 1987;18:35–44. [PubMed]
  • Ölveczky BP, Andalman AS, Fee MS. Vocal experimentation in the juvenile songbird requires a basal ganglia circuit. PLoS Biol. 2005;3:e153. [PMC free article] [PubMed]
  • Person AL, Perkel DJ. Unitary IPSPs drive precise thalamic spiking in a circuit required for learning. Neuron. 2005;46:129–40. [PubMed]
  • Person AL, Perkel DJ. Pallidal neuron activity increases during sensory relay through thalamus in a songbird circuit essential for learning. J Neurosci. 2007;27:8687–98. [PubMed]
  • Person AL, Gale SD, Farries MA, Perkel DJ. Organization of the songbird basal ganglia, including Area X. J Comp Neurol. 2008;508:840–66. [PubMed]
  • Pollak GD, Burger RM, Klug A. Dissecting the circuitry of the auditory system. Trends Neurosci. 2003;26:33–9. [PubMed]
  • Ranck JB., Jr Which elements are excited in electrical stimulation of mammalian central nervous system: a review. Brain Res. 1975;98:417–40. [PubMed]
  • Reiner A, Laverghetta AV, Meade CA, Cuthbertson SL, Bottjer SW. An immunohistochemical and pathway tracing study of the striatopallidal organization of Area X in the male zebra finch. J Comp Neurol. 2004a;469:239–61. [PubMed]
  • Reiner A, Perkel DJ, Bruce LL, Butler AB, Csillag A, Kuenzel W, Medina L, Paxinos G, Shimizu T, Striedter G, Wild M, Ball GF, Durand S, Güntürkün O, Lee DW, Mello CV, Powers A, White SA, Hough G, Kubikova L, et al. Revised nomenclature for avian telencephalon and some related brainstem nuclei. J Comp Neurol. 2004b;473:377–414. [PMC free article] [PubMed]
  • Sasaki A, Sotnikova TD, Gainetdinov RR, Jarvis ED. Social context-dependent singing-regulated dopamine. J Neurosci. 2006;26:9010–4. [PMC free article] [PubMed]
  • Scharff C, Nottebohm F. A comparative study of the behavioral deficits following lesions of various parts of the zebra finch song system: implications for vocal learning. J Neurosci. 1991;11:2896–913. [PubMed]
  • Sen K, Theunissen FE, Doupe AJ. Feature analysis of natural sounds in the songbird auditory forebrain. J Neurophysiol. 2001;86:1445–58. [PubMed]
  • Somogyi P, Hodgson AJ. Antisera to gamma-aminobutyric acid. III Demonstration of GABA in Golgi-impregnated neurons and in conventional electron microscopic sections of cat striate cortex. J Histochem Cytochem. 1985;33:249–57. [PubMed]
  • Stark LL, Perkel DJ. Two-stage, input-specific synaptic maturation in a nucleus essential for vocal production in the zebra finch. J Neurosci. 1999;19:9107–16. [PubMed]
  • Tehovnik EJ, Tolias AS, Sultan F, Slocum WM, Logothetis N. Direct and indirect activation of cortical neurons by electrical microstimulation. J Neurophysiol. 2006;96:512–21. [PubMed]
  • Thomson AM. Inhibitory postsynaptic potentials evoked in thalamic neurons by stimulation of the reticularis nucleus evoke slow spikes in isolated rat brain slices I. Neuroscience. 1988;25:491–502. [PubMed]
  • Troyer TW, Doupe A. An associational model of birdsong sensorimotor learning I. Efference copy and the learning of song syllables. J Neurophysiol. 2000;84:1204–23. [PubMed]
  • Vates GE, Vicario DS, Nottebohm F. Reafferent thalamo-”cortical” loops in the song system of oscine songbirds. J Comp Neurol. 1997;380:275–90. [PubMed]
  • Wanaverbecq N, Bodor AL, Bokor H, Slézia A, Lüthi A, Acsády L. Contrasting the functional properties of GABAergic axon terminals with single and multiple synapses in the thalamus. J Neurosci. 2008;28:11848–61. [PubMed]
  • Wild JM. Descending projections of the songbird nucleus robustus archistriatalis. J Comp Neurol. 1993;338:225–241. [PubMed]
  • Wild JM, Williams MN, Howie GJ, Mooney R. Calcium-binding proteins define interneurons in HVC of the zebra finch (Taeniopygia guttata) J Comp Neurol. 2005;483:76–90. [PubMed]