INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

The processing of sensory information is often modified according to changing behavioral requirements. Behavioral state-dependent modulation of sensory input includes subpopulation-specific loss or reduction of sensory input, changes in strengths of connectivity, modification of bursting patterns of activity, and attentional mechanisms that modify sensory tuning properties (e.g., McCormick and Bal 1997; Ramirez 1998; Suga et al. 2000; Weimann et al. 1991). In the bird song system, auditory activity is profoundly influenced by behavioral state, and the forebrain nucleus HVc has been identified as a possible locus of the state-dependent modulation (Dave et al. 1998; Nick and Konishi 2001; Schmidt and Konishi 1998). Here we report that the activity of distinct classes of HVc neurons is modulated according to behavioral states presumed to be associated with distinct components of sensorimotor learning and sensory processing.

Lesion and electrophysiological studies have suggested multiple roles for HVc in processing auditory information under diverse behavioral states. HVc lesions produce deficits in conspecific song perception (e.g., Brenowitz 1991; Gentner et al. 2000). HVc auditory responses in anesthetized birds are tuned to the bird's own song (BOS), suggesting direct processing of auditory feedback during singing (see Margoliash 1997). In sleeping birds, neurons in the robust nucleus of the archistriatum (RA) exhibit burst patterns spontaneously and in response to playback of BOS that match the burst patterns seen in premotor activity during singing. HVc, which projects to RA, also bursts during sleep, and these bursts are correlated with (and typically lead) RA bursts (Dave and Margoliash 2000).

Some HVc multiunit studies, however, have failed to find auditory responses in awake birds. McCasland and Konishi (1981) observed suppression of physiological activity in response to external auditory stimulation while canaries were singing. Schmidt and Konishi (1998) reported little evidence of auditory response in HVc of awake (nonsinging) zebra finches. These negative results are difficult to reconcile with HVc's posited roles processing auditory feedback during singing and in conspecific song recognition. Indeed, other studies have reported multiunit (canaries: McCasland and Konishi 1981; white-crowned sparrows: Margoliash 1986) and single-unit (zebra finches: Yu and Margoliash 1996) song-selective auditory responses in awake quiescent birds.

None of the extracellular HVc recording studies determined from which neuronal classes recordings were obtained. HVc comprises six or more morphological classes of neurons (Fortune and Margoliash 1995; Nixdorf et al. 1989) that can be categorized as interneurons (HVc-In) and projection neurons; individual projection neurons exclusively target either RA (HVc-RAn) or Area X (HVc-Xn) (Fig. 1). Studies in vitro (Dutar et al. 1998) and in anesthetized birds (Mooney 2000) have shown that HVc-RAn, HVc-Xn, and HVc-In are physiologically distinct and differ widely in their auditory responses (Mooney 2000). This raises the possibility that undetected differential sampling biases for populations of HVc neurons as well as species differences contributed to the conflicting extracellular results.

View larger version (22K): [in this window] [in a new window]   Fig. 1. Schematic of the song system. DLM, medial nucleus of dorsolateral thalamus; HVc, here used as its proper name; lMAN, lateral magnocellular nucleus of the anterior neostriatum; NIf, nucleus interfacialis; RA, robust nucleus of the archistriatum; nXIIts, tracheosyringeal portion of the hypoglossal nucleus; X, Area X.

In the present study, we recorded from single HVc units in anesthetized zebra finches in conjunction with antidromic stimulation from RA and Area X, identifying parameters that distinguish HVc neuronal classes based on extracellularly recorded spike waveforms. We observed distinct physiological properties for putative interneurons and projection neurons that were consistent with previous reports (Mooney 2000). We then recorded HVc single units and multiunits in unanesthetized birds over long periods during sleep and awake quiescence, describing the circadian regulation of auditory activity. The data demonstrate that auditory responses in distinct classes of interneurons are differently regulated by behavioral state.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

All animal procedures were approved by an Institutional Animal Care and Use committee. Zebra finches (Taeniopygia guttata) were obtained as juveniles from a commercial vendor (Magnolia Bird Farms, Anaheim, CA) and then raised to adulthood in our colony (>120 days posthatch). Each bird was isolated in a sound-attenuation chamber, and a sample of songs was recorded. Stimuli comprising introductory syllables and two to four motifs were extracted from the recordings most representative of the bird's song following well-established procedures (e.g., Janata and Margoliash 1999). Briefly, songs were collected from a bird isolated in a sound-attenuation chamber, digitized, and saved to hard disk (sampling rate = 20 kHz). An exemplar that well represented the bird's most typical song was extracted, high-pass filtered at 500 Hz, and scaled to 70-dB root-mean-squared amplitude. In the days prior to days of experiments, birds were transferred to a 16:8 L/D schedule. The long days facilitated birds falling asleep during nighttime recordings.

Electrophysiological recordings from behaving animals

Our procedures for chronic recordings have been described in detail elsewhere (Dave et al. 1999). Briefly, a recording device carrying four glass-coated Pt-Ir electrodes was implanted under modified Equithesin (0.85 g chloral hydrate, 0.21 g pentobarbital, 2.2 ml 10% ethanol, and 8.6 ml propylene glycol, to a total volume of 20 ml with water) anesthesia over HVc. During recording sessions starting 2-4 days later, a flexible cable connected the headgear to an overhead commutator to allow the bird free movement within the cage. Differential recordings were used to minimize movement artifacts. Recording sites were obtained by audiovisual monitoring of the recordings while using a drive screw to manually advance the electrodes. Birds were manually restrained during this procedure, then carefully released into the cage while trying to maintain unit isolation. HVc is intensively active during singing so that prior experiments that required single-unit isolation during singing also required high signal-to-noise ratio (S/N) (Yu and Margoliash 1996). By contrast, in the design of these experiments, HVc activity was recorded primarily during auditory playback or behavioral quiescence. By decreasing the electrode impedance to 800 k to 1.2 M at 1 kHz, we were able to achieve sufficient unit isolation while maintaining the recordings over longer periods of time. Nevertheless, in some cases we were able to maintain recordings with single-unit isolation during vocalization as well as quiescence.

Sessions to compare waking and sleeping states began 30-120 min before the bird's normal nighttime. Auditory stimuli included BOS, BOS played in reverse (REV), and conspecific songs (CON) chosen to match BOS in duration and root-mean-squared mean amplitude. Stimuli were presented randomly interleaved at 20- or 30-s intervals during the initial waking period and continuing throughout the night. As previously reported, birds rapidly habituated to this schedule of song playback (Dave et al. 1998). Ten to 30 min after lights were doused, once the bird had been quiescent for several minutes, activity in HVc entered a characteristic bursting mode. This distinct state was never observed in an awake bird, and it disappeared whenever the bird was disturbed or became active. We used the presence of this bursting state as an assay for sleep. Bursting is correlated with other attributes of sleep including increased auditory responses (Dave et al. 1998). We also observed increased auditory responses during sleep (see RESULTS), which have been shown to be strongly correlated with electroencephalographic (EEG) measures of sleep (Nick and Konishi 2001).

Daytime recording sessions began at various times during the day, with the bird awake and alert throughout. In all cases, the awake, alert state of the bird was confirmed by visual observation (sometimes with the aid of a video camera) of the bird's posture and open eyes. In these sessions, we only recorded sites with at least one unit that could be well isolated. Auditory stimuli were presented 15-40 times each at intervals of 10-12 s. Calls of a conspecific female were presented to elicit vocalizations such as distance calls or song, and we looked for characteristic recruitment of HVc neurons to verify that we were recording from the intended structure. In the vicinity of HVc, song-selective responses to BOS are thought to be restricted to HVc proper and not the underlying cell-sparse "shelf" region (Margoliash 1983), and vocalization-related activity is also thought to be restricted to HVc neurons (e.g., McCasland and Konishi 1981). We cannot rule out the possibility that a small fraction of our data derive from shelf recordings. If so, however, we might have expected to observe differences in response properties for more ventral neurons, but we failed to observe any such differences.

Electrical stimulation experiments

After extracting song stimuli from a bird's recorded vocalizations (as described above), we anesthetized the bird with 45-60 µl of Equithesin after 1 h of food and water deprivation. The bird was immobilized in a stereotaxic frame using ear bars and a beak holder, the top layer of the skull was removed, and a pin was implanted caudal to the bifurcation of the midsagittal sinus.

On the days of recordings (3-5 days after implantation of the pin), birds were once again deprived of food and water for 1 h, and then anesthetized with 70-110 µl of 20% urethan in three to four doses administered intramuscularly over a 1-h period. Once anesthetized, the bird was wrapped in a cloth jacket and placed on a foam cushion, and the head was immobilized by fastening the pin to a frame. The bottom layer of the skull was removed to open a small window over the coordinates for each targeted structure (HVc, RA, and Area X). A thermal probe was inserted into the cloaca and attached to a thermostat (YSI, Dayton, OH, thermostat model 73A, probe model 451). A constant body temperature of 41.5°C was maintained by passing current through a length of nichrome wire looped through the foam cushion. During a few initial experiments, constant body temperature was not maintained. In these cases, the shapes of some of the widest spike waveforms recorded from HVc took on a "shouldered" appearance, manifested as a deflection in the postpeak negative-going slope. However, the shapes of other spike waveforms as well as other physiological measures such as ongoing firing rate and auditory response strength did not vary appreciably from those observed in temperature-regulated preparations. All neurons recorded, including those with "shouldered" waveforms, were included in our analysis.

Electrical stimuli were delivered with custom-built bipolar stimulating electrodes consisting of a pair of parallel glass-coated Pt-Ir electrodes spaced 300-500 µm apart, with resistances of 100-250 k. Stimulating electrodes were placed in Area X and RA of one hemisphere after locating the target nuclei with a separate recording electrode used to assess spontaneous rates and BOS responses. The recording electrode was withdrawn before the stimulating electrode was inserted. Recordings from HVc were made using either glass-coated Pt-Ir electrodes (1-2.5 M) or glass micropipettes pulled to a final resistance of 3-5 M and filled with 2M potassium acetate. Signals were amplified, filtered (300 Hz to 5 kHz band-pass), digitized at 20 kHz, and saved to computer disk. Recording electrodes were moved through the ipsilateral HVc until the spiking activity of single neurons could be isolated.

Glass electrodes tended to provide higher S/N but lower long-term stability as compared with metal electrodes. The higher-resistance electrodes (glass and higher resistance metal) seemed to strongly favor isolation of X-projecting neurons, while lower-resistance electrodes (lower resistance metal) showed greater success in isolating interneurons. With all electrodes, we found RA-projecting cells to be difficult to isolate and hold. This problem was exacerbated by the low firing rates of RA neurons, but attempts to use RA stimulation as a search stimulus helped to identify only a very few RA-projecting neurons, despite frequent success in activating other cell types trans-synaptically.

Once sufficient isolation and S/N were achieved, we attempted to stimulate the HVc neuron being recorded by passing 200-µs biphasic pulses of 20-800 µA through the Area X or RA stimulating electrode. If a spike was elicited, the current was adjusted to the lowest level that produced consistent stimulation with the shortest possible latency. The observation of spikes after the electrical stimulus could be indicative of antidromic activation (direct stimulation of the recorded cell's axon) or trans-synaptic activation (stimulation of other cells that eventually synapse onto the recorded cell). Antidromic activation therefore indicates that the recorded cell is a projection neuron, whereas trans-synaptic activation could likely occur in any cell type in HVc as a result of the highly interconnected nature of HVc circuitry (Fortune and Margoliash 1995).

When spikes occurred with a short and consistent latency, we performed collision tests to determine whether spikes were propagating antidromically from the target nucleus. The output of a threshold discriminator was used to trigger electrical stimulation in the target nucleus with 1- to 20-ms delay after an endogenously fired spike; collisions were judged to occur only if we found a delay that reliably prevented spikes elicited by stimulation, while longer delays failed to prevent spikes. In a few cases, where collisions appeared to occur but an extremely low ongoing firing rate yielded too few tests to be confident in their reliability, we attempted stimulation with trains of pulses at high frequencies. Direct antidromic stimulation should continue to reliably elicit spikes even at high frequencies, while trans-synaptic stimulation should fail given insufficient time for synaptic recovery after a previous stimulation. Accordingly, in these cases, only if spikes reliably followed all pulses in a 500-Hz train did we judge the stimulation to be definitively antidromic. Because there is no way of externally verifying the likelihood of false identification of projection neurons using our methods, we were generally conservative in our determination of whether stimulation was antidromic. Once the cell type was identified, if the S/N remained sufficiently high, 2-5 min of spontaneous spiking activity were recorded as were responses to playback of 20 repetitions of BOS (and, in some cases, REV) at 10- to 12-s intervals.

HVc neurons are often anatomically organized into clusters. Gap junctions have been reported in HVc neurons that could support strong electrical coupling (Burd and Nottebohm 1985; Gahr and Garcia-Segura 1996). In theory this could potentially complicate the interpretation of antidromic stimulation experiments if a neuron not projecting to one of HVc's efferent nuclei (Area X or RA) was strongly coupled to a neuron projecting to an efferent nucleus and if the second neuron was the one that was being activated by electrical stimulation of the efferent nucleus. This would result in mislabeling a neuron as projecting to one nucleus when in reality the neuron projected to the other nucleus or was an intrinsic interneuron. Whereas we could not directly assess whatever potential role electrical coupling had in confounding our analysis, in any case, it was insufficient to abolish the compelling differences between putative interneurons, which were not antidromically stimulated, and projection neurons, which were (see RESULTS).

At the termination of each experiment, the stimulation sites were marked with small electrolytic lesions. Birds were deeply anesthetized with an overdose of pentabarbitol sodium and perfused transcardially with heparinized saline followed by formalin. Brains were stored for several days in 30% sucrose formalin before being cut in 50-µm sections on a freezing stage microtome and stained with cresyl violet. Sections were examined to verify that the stimulating electrodes were placed in the correct target location and that all recorded cells were in HVc. For chronic experiments, the multiple penetrations over days along the same track tended to cause a large lesion over time. In these experiments, the value of the histology tended to be limited to assessing the depth of the penetration relative to the ventral border of HVc, and we relied more on motor-related activity as our principal criterion for assessing the location of recording sites.

Analysis of spike waveforms

Only those recordings confirmed to be in the targeted structure from histological verification of electrode location or, in the case of some HVc recording sites, the presence of motor-related activity during vocalizations are included in this study. Stimulus presentations during which the bird vocalized were discarded. Raw neural data were DC filtered, and single units were identified using the Spikesort program (Lewicki 1994). Spike models generated by this program with 0.01-ms precision were used to estimate spike durations. To standardize these measurements, spike waveforms were inverted as necessary to conform to a canonical spike shape, which began with a large negative going deflection followed by the largest positive going deflection, which functioned as the center peak (Figs. 2 and 3). The zero time was defined as the time of maximum amplitude of the center peak, and the maximum amplitude at t = 0 was normalized to 1.0. The spike widths reported here are defined as the width of the largest positive-going spike at 25% of peak amplitude. In a few cases from the chronic data, spikes exhibited an initial, small positive deflection as well as the above features. In these cases, ignoring the initial deflection when matching to the canonical spike shape gave the best overall fit.

Cells were initially sorted on spike width, but it was observed that other spike characteristics might help to better distinguish spike classes. We took seven measurements from the spike models; in Fig. 2, these correspond to the peak and zero-crossing times denoted by  and amplitudes denoted by . The seven spike variables were used in a principal components analysis (PCA) as computed by the Systat program (Systat Software, Richmond, CA) to identify their contribution to the variance among spike waveforms. Only those principal components with eigenvalues >1in each case, the first two principal componentswere utilized visualizing the data as scatterplots. PCA identifies those factors that best describe the variance in the data but does not measure clustering within the data. We noted, however, that clusters were visually apparent in the scatterplots when considering the distribution of spike widths and projection class as determined by electrical stimulation. These clusters were also consistent with differences in a host of other physiological properties including spontaneous rates, responses to BOS, modulation by behavioral state, etc. (see RESULTS).

View larger version (8K): [in this window] [in a new window]   Fig. 2. Spike waveform classification variables. A representative spike model of an Area X-projecting HVc neuron. , the peak and zero crossing times measured for the spike clustering analysis. The two peak amplitudes indicated () were also measured. The large positive peak was centered at time = 0 and normalized to 1 for all spike models. , the measurement of spike width at 25% peak amplitude.

Analysis of neural responses

Single units were analyzed as time-of-event data (point processes); multiunit recordings were DC filtered, rectified, integrated, and averaged. In cases where long stretches of spontaneous data were not recorded (principally short-term recordings from awake birds), ongoing discharge rates were determined by averaging the discharge rates from 1 s immediately preceding each stimulus presentation and 3 s beginning 2 s after stimulus offset. In other cases (principally in recordings from sleeping birds), we also determined ongoing discharge rates from much longer (>30 s) stretches of activity recorded in the absence of any acoustic stimulation.

The statistical significance of a single-unit response to a song stimulus was tested using a paired t-test comparing the spontaneous and response firing rates for all stimulus repetitions. Units were considered responsive if the response rates were different from the spontaneous rate at P < 0.05. Multiunit responses were subjected to a similar analysis: responses were considered significant if a t-test showed the mean response to be different from zero at P < 0.05. Response magnitudes are reported for multiunits as a fractional change from ongoing discharge and for single units as changes in spikes/s relative to ongoing discharge. Song selectivity was measured using the d'_A-B metric (Green and Swets 1966), defined as: d'_A-B = 2(R_A  R_B)/(s + s), where R_A and R_B are the absolute mean response magnitudes to stimuli A and B, respectively, and s² is the variance of the corresponding R value. A neuron was considered selective for stimulus A over stimulus B if its d'_A-B value was >0.5, as per Solis and Doupe (1997). Unless otherwise noted, we tested the significance of differences in population means (reported as mean ± SD) using unpaired t-tests. To assess the similarity of response patterns to song playback between pairs of recording sites, we calculated peristimulus time histograms (PSTHs) of the single-unit activity for each cell during stimulus playback using 10-ms bin widths, starting 20 ms after stimulus onset and ending 20 ms after stimulus offset to account for response latency in HVc (see Margoliash and Fortune 1992). We then calculated the linear correlation coefficient r for the sets of bin counts obtained from each pair of histograms.

The level of bursting activity in a single unit's spontaneous discharge was measured using a burstiness index, defined as a number between 0 (all spikes regularly spaced) and 1 (all spikes occurring simultaneously, a limit condition) (see Margoliash et al. 1994). This was calculated by summing from the second to penultimate spikes the lesser of the two interspike intervals associated with each spike, and dividing by the duration of the recording; this result was subtracted from 1 to obtain the index.

Unless otherwise indicated, all analyses were performed in the Matlab program (The MathWorks, Natick, MA).

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

A total of 40 single units and 13 multiunits were recorded from 42 sites in the HVc of 11 unanesthetized birds. In some cases, we were able to maintain recordings across the transitions from sleeping to waking states (n = 6 single units, 13 multiunits) and from waking to sleeping states (n = 9 single units, 9 multiunits). Of these sites, 5 single units and 9 multiunits yielded data across both transitions; thus we were able to compare sleeping and waking states for a total of 10 unique single units and 13 unique multiunit sites. In other experiments, we focused on isolating cells during 1- to 4-h sessions during birds' normal waking hours. Recordings of this relatively limited duration were more easily obtained, increasing our yield, and we achieved interpretable recordings from 30 HVc single units in nine birds.

The interpretation of the HVc recordings in unanesthetized animals is aided by knowledge of which projection class of HVc neurons or interneurons each unit corresponds to. To develop criteria to make these assessments in extracellular recordings, we conducted electrical stimulation experiments in anesthetized birds (see METHODS). These experiments are described in the following section.

Antidromic stimulation experiments in anesthetized birds

Of the numerous physiological properties that help to distinguish between categories of HVc cells (Mooney 2000), spike width and shape, spontaneous activity, and phasic/tonic auditory response properties were accessible to our extracellular recordings. We therefore analyzed these parameters in a dataset of 159 HVc neurons in 19 anesthetized adult male zebra finches where single-unit isolation sufficient to obtain reliable (high S/N) spike models was achieved.

SPIKE MORPHOLOGY DISTINGUISHES DIFFERENT CLASSES OF HVC NEURONS. Spike durations ranged widely (0.14-0.90 ms). Visual inspection of the spike models suggested classification into three morphological groups: waveforms with very narrow spikes (0.14-0.19 ms, n = 7), symmetrical peaks, and no prolonged, negatively directed tail (Fig. 3B); those with spikes of intermediate width (0.20-0.35 ms, n = 55), a steeper positive-going than negative-going slope, and a longer-lasting tail (Fig. 3C); and those which had wider spikes (0.35-0.90 ms, n = 97) and rounded peaks, but also had extended tails (Fig. 3, D-F).

View larger version (29K): [in this window] [in a new window]   Fig. 3. Spike shapes obtained from HVc neurons in anesthetized zebra finches. A: scatterplot of spike waveform models according to the first two factors extracted from principal components analysis (PCA; see METHODS). Colors denote classification according to spike width as illustrated in remaining panels: red, narrow-spike models (width: 0.14-0.19 ms); green, intermediate-spike models (width: 0.20-0.35 ms); blue, wide-spike models not confirmed as projection neurons (width: 0.39-0.90 ms); black, confirmed X-projection neurons (width: 0.38-0.86 ms); purple, confirmed RA-projecting neurons (width: 0.36-0.56 ms). B-F: overlaid models for each class of neurons, color-coded as in A. The black, red, and purple traces in C represent spike models that clustered with spikes of intermediate width (represented in green) according to the PCA scatterplot but were classified as other cell types (indicated by their respective colors) based on spike width, as noted in the text. In the top right of each box is the number of cells for that category.

SPIKE SHAPE CLASSES CORRESPOND TO HVC PROJECTION NEURONS AND INTERNEURONS. For many neurons (n = 98), we attempted to elicit spikes through electrical stimulation of both RA and Area X. In cases where spikes were elicited from electrical stimulation in one or both sites with a consistent latency, we then attempted to collide the electrically stimulated spikes with endogenous spikes to determine whether we were antidromically stimulating a projection neuron. Figure 4 shows an example of such an experiment. For this cell, stimulation in RA failed to consistently elicit spikes (not shown). Spikes were reliably elicited from Area X stimulation that occurred 4.3 ms after a spontaneous spike (Fig. 4C), whereas spikes consistently failed when Area X was stimulated 2.0 ms after a spontaneous spike (Fig. 4B). The waveforms were quite stable throughout the experiments (Fig. 4D). The failure to elicit a second action potential only when stimulating from Area X immediately following a first, spontaneously generated spike identifies this cell as an Area X projecting HVc neuron. There were no cases of verified antidromic stimulation from both Area X and RA, a finding that is consistent with the anatomical data that each HVc projection neuron targets only Area X or RA (e.g., Nixdorf et al. 1989). This also argues against contamination of our data set by strong electrical coupling between neurons (see METHODS), at least for electrical coupling between different classes of projection neurons.

View larger version (13K): [in this window] [in a new window]   Fig. 4. Collision test for an X-projecting HVc neuron. A: schematic illustration of the antidromic stimulation experiment. B-D: 200-µs, 200-µA biphasic pulses were made to Area X while recording an HVc single unit (SU). The stimulus artifact is truncated to allow visualization of the neuronal waveforms. The output of a threshold discriminator was used to trigger stimuli 2.0 ms (B) or 4.3 ms (C) after each spike. Note that the stimulus always elicits a spike at the longer latency and never at the shorter latency. Otherwise, the traces show little variation from trial to trial (superimposed traces, D).

View larger version (33K): [in this window] [in a new window]   Fig. 5. Responses to bird's own song (BOS) in anesthetized birds for different classes of HVc neurons. A: a putative interneuron (spike duration = 0.24 ms) with a tonic response to BOS playback; 3 raw traces and a peristimulus time histogram (PSTH) showing the average response to 20 repetitions of BOS playback are aligned above a spectrograph of the stimulus. B: a confirmed X-projecting neuron from the same bird (spike duration = 0.50 ms) with a reliable phasic response to BOS. C: a confirmed RA-projecting neuron from a different bird (spike duration = 0.40 ms) with a less reliable phasic response to BOS.

Chronic recordings in behaving animals

BIAS TOWARD INTERNEURONS. The stimulation experiments facilitated analysis of chronically recorded data from unanesthetized animals. Across all HVc single-unit recordings, including all experiments in which we attempted to maintain recordings for long periods of time across behavioral state transitions, 38/40 single-unit spike waveforms had shapes and durations (0.14-0.25 ms) typical of the interneurons identified in anesthetized animals. These 38 neurons also tended to have relatively high spontaneous rates and tonic responses to BOS characteristic of putative HVc interneurons. Only in the experiments that focused on isolating cells during relatively short sessions during birds' normal waking hours did we obtain two recordings of HVc single neurons whose spike waveforms fit the profile of projection neurons, apparently X-projecting neurons (see following text). Thus it appears that the requirements of these experiments for isolating and maintaining single units over long-duration recordingsespecially sleep-wake transitionscoupled with the low spontaneous rates of HVc projection neurons, strongly biased our recordings toward interneurons. The use of a manually actuated microdrive that required restraining the bird while isolating cells (see METHODS) is likely to have further contributed to recording bias. The recording bias we observed is counter-intuitive in that projection neurons have larger somata (Nixdorf et al. 1989) and may be expected to be easier to record from; knowledge of this bias facilitated our interpretation of the data (see DISCUSSION).

INTERNEURON ACTIVITY IN HVC IS HOMOGENEOUS DURING SLEEP. A total of 10 single interneurons and 13 multiunits were recorded in seven sleeping birds. At each site in HVc, ongoing activity during sleep was characterized by strong bursts of activation of many neurons (Fig. 6A). It was often difficult to maintain single-unit isolation against this background of activity (but see Fig. 6B). Individual bursts typically lasted 30-200 ms, and trains of bursts often lasted 2-4 s. We quantified bursting activity of single units using a normalized index (see METHODS). Single units recorded during sleep tended to burst, so that the average bursting index was high (0.54 ± 0.07, n = 10). In contrast, bursting activity in awake birds was relatively uncommon (Fig. 6, A and B), with the mean bursting index significantly lower than that during sleep (0.38 ± 0.20; P < 0.01).

View larger version (59K): [in this window] [in a new window]   Fig. 6. Comparison of HVc activity during sleep and while awake. A, left: three representative raw traces of a multiunit recording site in HVc recorded ~20 min after lights were turned off and the bird had been quiescent for 15 min; right: three raw traces of the same site 5 min before lights were turned off. All traces are aligned to the auditory song stimulus played back to the bird during recording shown as an oscillograph (bottom). Note the lack of auditory responses or ongoing bursting in the awake recordings, compared with the recordings during sleep, in which spontaneous bursts are seen () as well as bursting in response to song playback. Such strong bursting was only seen in recordings from sleeping birds. B, left: a well-isolated single unit with bursting during song playback and ongoing discharge during sleep. Right: the same unit shows a modest auditory response when awake, and bursting is greatly reduced.

View larger version (78K): [in this window] [in a new window]   Fig. 7. Multiunit HVc activity during song stimulus playback throughout the night and into the following day for 2 sites, 1 per bird. Each row represents 10 presentations of BOS, BOS presented in reverse (REV), or conspecific song (CON). Multiunit neuronal recordings for each group of 10 repetitions were rectified, smoothed with a 20-ms integration window, and averaged. The resulting traces were converted to a gray-scale image, with white representing the lowest level of activity seen during stimulus presentation and black representing the highest. Because stimuli were presented in random order, the amount of real-time (wall clock time) that expired during 10 repetitions of a given stimulus could vary. Thus although experiment start and end times are aligned, the position of the time lights were turned off (filled arrow) and the time lights were turned on (open arrow) vary slightly among the images. A: a multiunit HVc site with a strong response to BOS but only during sleep. Responses during sleep to REV and CON songs are slightly suppressed. B: a multiunit HVc site in a different bird which retained a strong, selective song response during waking hours. Note the clear suppression of activity in response to reversed and conspecific songs during sleep.

View larger version (49K): [in this window] [in a new window]   Fig. 8. PSTHs (10-ms bins) of single-unit HVc responses to BOS during sleep and during the day. A: highly correlated HVc single-unit responses to BOS playback of 2 pairs of recording sites in 2 birds during sleep. Left: response histograms for 2 units recorded on different nights with a linear correlation coefficient r = 0.795. Right: another highly correlated pair of BOS responses (r = 0.724), this time in 2 single units recorded simultaneously during sleep in another bird. B: less correlated single-unit BOS responses in awake birds. Left: responses in 2 simultaneously recorded single units showing strong auditory responses but very low correlation (r = 0.075) as was typically seen in our awake HVc recordings. Right: responses during the day of the same pair of single units represented in the histograms immediately above in A. This pair of units shows an unusually high level of correlation for an awake bird (r = 0.486) that is nonetheless much lower than the correlation seen during sleep.

TRANSITION TO WAKEFULNESS AND EMERGENCE OF HETEROGENEOUS PATTERNS. Sites recorded during sleep had dramatically different responses when birds were awake. The transition to the waking state involved loss of ubiquity and strength of HVc auditory responses, loss of selectivity, and loss of similarity in temporal response patterns. Auditory responses in the awake bird were less common than during sleep: only 31% (4/13) of HVc multiunit sites and 60% (6/10) of single units recorded during sleep exhibited statistically significant responses to BOS during the day (Fig. 7A). In all cases, auditory responses to BOS in awake birds were weaker than the responses at the same sites observed during sleep; the mean multiunit response (over baseline) declined by 77%, whereas the mean firing rate response across all single units declined by 65% (by 12.7 ± 9.2 spikes/s). Nevertheless, for some sites, the awake responses were very strong (Figs. 7B and 8B). There was also an overall reduction of inhibition such that the mean of daytime responses to conspecific and reversed songs was close to zero, and individual sites could exhibit overall weak excitation as well as inhibition; these differences were significant (Table 2). Thus while HVc responses in awake birds remained largely BOS-selective, the degree of selectivity was decreased, as was reflected by the d' selectivity measure (d'_{BOS-REV:sleep} = 3.52 ± 1.79, d'_BOS-REV:wake = 1.70 ± 1.67, P < 0.005, n = 8; d'_{BOS-CON:sleep} = 2.67 ± 0.97, d'_BOS-CON:wake = 1.23 ± 1.48, P < 0.001, n = 7). The observed decrease in selectivity was not the result of state-dependent fluctuations in the mean ongoing spiking rate, which did not differ between the sleeping (10.5 ± 5.7 spikes/s) and waking (10.5 ± 8.8 spikes/s) states. Spontaneous rates of neurons in RA measured by excluding bursts during sleep decrease as compared with spontaneous rates in awake birds (Dave et al. 1998). Bursts during sleep in RA are easy to identify, whereas bursts in HVc are not as unambiguous, and we did not attempt to measure HVc spontaneous rates outside of bursts. Thus the lack of change in HVc reported here may be the result of a combination of two counteracting effects in awake animals, increased spontaneous rates outside of bursts, and loss of bursts.

Distinct neuronal populations and pathways

The data suggest that we obtained nearly all of our HVc single-unit recordings from interneurons and that we recorded from two or more classes of interneurons with differing state-dependent properties. Units with and without auditory responses could occur at the same recording site, verifying that HVc neurons were not simply undergoing global sensory gating (Fig. 9A). In daytime recordings, units with excitatory responses to BOS had significantly lower rates of ongoing discharge (4.52 ± 4.83 spikes/s) than did other units (units with no response or suppressed response: 9.49 ± 5.02 spikes/s, P < 0.01). Our experience was that there were many auditory, slowly firing neurons in the awake HVc (e.g., Fig. 9B), but those recordings were difficult to maintain, whereas faster-firing neurons that tended to have weaker or no auditory responses often yielded more stable recordings.

View larger version (55K): [in this window] [in a new window]   Fig. 9. Single-unit HVc auditory responses during waking hours. A: responses to BOS of 2 single units recorded simultaneously from the same electrode. Left: PSTHs (10-ms bins) showing activity during playback of the BOS stimulus (spectrograph aligned below). Top: note that the unit represented here shows a clear excitatory response to BOS playback, whereas the other, simultaneously recorded unit (bottom) shows no significant response. Middle: spike models obtained for each single unit. Lines with double arrows indicate spike width at 25% peak amplitude, with values of 0.20 ms (top), typical of many HVc neurons with auditory responses during wakefulness, and 0.15 ms (bottom), typical of the narrow spike models obtained for HVc neurons with no wakeful auditory responses. Right: a raw trace showing activity of both units during one repetition of BOS stimulus playback (spectrograph aligned below). The high spontaneous firing rate of the nonauditory unit (smaller spikes) is representative of many HVc single units without wakeful auditory responses. B: responses to BOS of 2 simultaneously recorded single units on different electrodes in another awake bird. Left: PSTHs show significant excitatory responses to playback of BOS (spectrograph aligned below) in both units. Middle: spike models for both units, each with a width of 0.16 ms at 25% peak amplitude. Right: raw traces showing activity for each unit during BOS playback.

The differences in firing rate and wakeful auditory responses are related to the distinction between the narrow-spike and the intermediate-spike classes of interneurons described earlier in anesthetized preparations. Neurons with waveforms matching the profile of the narrowest class of interneuron (symmetrical peak, no extended negative tail deflection) had spike durations ranging from 0.13 to 0.16 ms (Fig. 10B); neurons with greater spike durations (0.18-0.23 ms) had the asymmetrical peaks and short negative tail that distinguished the wider-spike class of interneurons (Fig. 10C). The two putative projection neurons had the predicted wide, rounded spike waveforms (cf. Figs. 10D and 3E). We related the spike shapes from chronic recordings to those obtained from anesthetized preparations with a PCA using both data sets. A scatterplot of the first two principal components (explaining 86% of the total variance) showed that the waveforms obtained from awake animals clustered in a similar manner to the narrow and intermediate waveforms in anesthetized animals (Fig. 10A). The clusters obtained from awake recordings were slightly shifted with respect to the clusters obtained from anesthetized recordings, probably reflecting the briefer spike durations seen in awake recordings. It is not clear whether the discrepancy in spike widths is due to a physiological effect of anesthesia or to sampling biases which differ between recording techniques.

View larger version (27K): [in this window] [in a new window]   Fig. 10. Spike shapes obtained from HVc neurons in both awake and anesthetized birds. A: scatterplot of spike waveform models according to the first 2 factors extracted from PCA (see METHODS). Neurons recorded in awake animals are represented by solid dots; those recorded in anesthetized animals are represented by open circles. Colors denote classification according to spike width as illustrated in remaining panels: red, narrow-spike models; green, intermediate-spike models; blue, wide-spike models. B-D: overlaid spike models for each class of neurons as recorded in awake animals, color-coded as in A.

In awake birds, the narrow-spike putative interneurons showed a significantly lower mean BOS response (2.1 ± 3.5 spikes/s) than did the wider-spike putative interneurons (7.2 ± 6.1 spikes/s; P < 0.01). The difference in auditory response properties between the two interneuron classes is even more strikingly demonstrated by the distribution of auditory responses: while only half of the narrow-spike interneurons showed any significant auditory responses (11/22), all of the wider-spike interneurons showed significant BOS responses (16/16, P < 0.005, Wilcoxon rank sum test). The narrow-spike neurons also had a higher mean firing rate (8.3 ± 6.5 spikes/s) than the wider-spike neurons (5.6 ± 5.0), but this difference was not statistically significant (P > 0.16).

The strong relationship between auditory responses and other physiological characteristics only held during wakefulness. During sleep, all single units were strongly auditory regardless of spontaneous rate or spike duration. However, both the shape and duration of the spike waveforms remained constant across the sleep-wake transition, indicating that our use of spike shape to identify cell classes was valid across natural behavioral states.

Comparisons of activity between awake and anesthetized birds

The interpretation of HVc auditory response properties has relied on studies mostly conducted in anesthetized animals. Given the sensitivity of HVc auditory activity to behavioral state, it is valuable to compare the physiological properties in the anesthetized state to those in naturally occurring behavioral states. To this end, we compared the firing properties of the putative interneurons in our single-unit recordings from asleep or awake, behaving birds with those of interneurons (here defined as those neurons with spike widths <0.35 ms) in recordings from anesthetized birds.

In general, HVc interneurons in anesthetized birds shared similar properties with those in asleep birds, although generally at a depressed firing rate. Interneurons in anesthetized recordings were highly bursty; their mean bursting index (0.68 ± 0.15, n = 56) was actually higher than that measured in sleeping birds (0.55 ± 0.07, n = 11; P < 0.05). As in sleeping birds, auditory responses to BOS playback were common under anesthesia with 52/56 interneurons showing significant excitatory responses. These responses were fairly tonic and qualitatively similar to those seen in interneurons reported by Mooney (2000). However, the absolute magnitude of BOS responses was significantly lower in anesthetized birds (6.9 ± 6.7 spikes/s, n = 52) than in asleep birds (17.1 ± 9.2 spikes/s, n = 10, P < 0.001) and did not differ significantly from that of auditory interneurons in awake birds (5.7 ± 5.4 spikes/s, n = 26). This decrease in magnitude from the sleep state may be explained in part by the overall depressed ongoing firing rates of interneurons in the anesthetized state (3.1 ± 3.1 spikes/s) as compared with the sleeping state (10.2 ± 6.0 spikes/s, P < 0.01). Auditory responses in anesthetized birds also appeared to lack the exquisite selectivity of those in sleeping birds as most REV responses were excitatory rather than inhibitory (10/13); however; the mean d'_BOS-REV (2.32 ± 1.40, n = 13) did not differ significantly from that in sleeping birds (3.52 ± 1.79, n = 9, P = 0.09) or in awake birds (2.25 ± 1.99, n = 21, P = 0.91).

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

We have directly demonstrated that behavioral state differentially modulates subpopulations of HVc neurons. Our antidromic stimulation experiments indicate that most single units recorded using the chronic recording techniques employed herein were interneurons. While all putative interneurons burst frequently during sleep and are responsive to song stimuli, those with the narrower spike shape tend to lose auditory responses during wakefulness while those with the broader spike shape tend to retain modest, less selective responses during wakefulness. Furthermore, during sleep the auditory responses of all HVc interneurons possess a remarkable similarity not seen in awake responses.

There are at least two morphological classes of HVc-RAn and HVc-Xn, and several classes of HVc-In (Fortune and Margoliash 1995; Nixdorf et al. 1989). Physiological evidence for cell classes in HVc has been reported in extracellular (Margoliash 1983; Margoliash and Fortune 1992) and intracellular (Lewicki 1996) recordings describing a subset of low spontaneous rate, phasically responding auditory neurons with nonlinear spectral and temporal facilitatory properties. Differences in intrinsic properties and auditory responsiveness have also been described in intracellular recordings comparing HVc-RAn, HVc-Xn, and HVc-In (Mooney 2000). We characterized neurons with narrow- and medium-width spikes as interneurons. This is consistent with a broad set of data, including correlations and inferences within the data set from electrical stimulation reported here, and with prior studies. The physiological properties of the cell classes we report are also consistent with those described in a recent paper that used antidromic stimulation to identify HVc cell classes in sleep-induced birds (Hahnloser et al. 2002). If neurons with narrow- and medium-width spikes are indeed interneurons, this would be the first demonstration of physiological variation within any of the three broadly defined categories (i.e., HVc-RAn, HVc-Xn, and HVc-In) of HVc neurons.

Auditory responses have been reported in studies of single HVc neurons in awake zebra finches (Yu and Margoliash 1996), and multiunits in white-crowned sparrows (Margoliash 1986) and canaries (McCasland and Konishi 1981). Perhaps all those recordings also suffered substantial (although unrecognized) sampling biases. The strong auditory responses in awake white-crowned sparrows and canaries as compared with zebra finches may also indicate potential species differences. Preliminary results from chronic recordings in awake song sparrows also indicate much stronger HVc multiunit auditory responses than observed in zebra finches (Nealen and Schmidt 2002). In the present study, we had particular difficulty holding auditory single units, so that nonresponsive sites dominated our awake multiunit recordings. This difficulty may explain the failure to find robust auditory responses in HVc in awake zebra finches in a recent study that relied exclusively on multiunit recordings (Schmidt and Konishi 1998).

Zebra finch adults gradually become less reliant on auditory feedback to maintain song as they age (Lombardino and Nottebohm 2000; Nordeen and Nordeen 1992). The birds in this study ranged from ~120 days of age (the beginning of sexual maturity) to 300 days of age. In all our birds, HVc neurons exhibited bursting and auditory responses during sleep, and there were no obvious age-related effects on the prevalence of auditory responses during wakefulness, although there was considerable variation across individuals in the relatively small sample sizes we achieved. This suggests the hypothesis that the variation in auditory responsiveness across individuals is related to individual variation in response to deafening (Brainard and Doupe 2001; Nordeen and Nordeen 1992).

State-dependent functional reorganization of HVc

A description is beginning to emerge of the complex functional modification of HVc at the circuit level that is related to a bird's varying behavioral requirements. A recent intracellular study of HVc in anesthetized birds concluded that while all classes of HVc neurons access auditory input, they are arranged in an ascending hierarchy of feed-forward and feedback connections among HVc-RAn, interneurons, and HVc-Xn, with the more selective responses of HVc-Xn presumably sculpted by inputs from HVc-RAn and interneurons (Mooney 2000). Although prior studies of auditory input to HVc have focused on possible direct or indirect projections of field L (auditory cortex analog) onto HVc (Fortune and Margoliash 1995; Kelley and Nottebohm 1979; Vates et al. 1996), recent anatomical and physiological studies have highlighted a parallel auditory pathway with nucleus interfacialis (NIf) serving potentially as a principal source of auditory input to HVc (Boco and Margoliash 2001; Janata and Margoliash 1999; Vates et al. 1996). The physiological data (also collected in anesthetized birds) demonstrate that NIf neurons are auditory and selective for BOS but are less selective than HVc neurons. Furthermore, ongoing bursting activity in NIf is strong and is synchronized with activity in HVc (Janata and Margoliash 1999).

The current data help to elaborate this model of HVc circuitry. There are at least two classes of putative interneurons in HVc, one retaining auditory responses during wakefulness and the other typically losing auditory responsiveness on waking. Furthermore, while most neurons in RA recorded with a similar methodology as employed in the present study fail to exhibit any auditory responses in awake zebra finches (Dave et al. 1998; Dave and Margoliash 2000), many neurons in Area X recorded with the same methodology exhibit auditory responses during wakefulness (unpublished observations). This suggests that HVc-Xn but not HVc-RAn may retain auditory responses in awake birds. In the proposed hierarchical scheme (Mooney 2000), our data suggest that state-dependent modulation has its greatest effects on lower levels of the hierarchy (i.e., HVc-RAn and nonresponsive interneurons). Thus the loss of auditory responses in a subpopulation of HVc neurons during wakefulness could help explain the weaker, less selective, and less similar awake responses of other HVc neurons. The model proposed by Mooney (2000) predicts that the removal of the inhibitory influence HVc-RAn exert via interneurons could result in less selective responses of HVc-Xn. Our data suggest that behavioral state modulation has such an effect on responsive interneurons (including the "intermediate-spike" class); in the Mooney (2000) model, only a single class of interneurons is recognized. In our model of behavioral state modulation, when birds are asleep, all classes of HVc neurons receive strong NIf input, which synchronizes population activity in HVc. This model is consistent with gating directly influencing intrinsic HVc auditory processing (Dave et al. 1998; Schmidt and Konishi 1998). Indeed, manipulations of neuromodulatory systems within HVc result in differential changes in auditory responses of RA and HVc neurons that mimic the effects of gating (Dave et al. 1998; Shea and Margoliash 1999, 2000).

Is sleep like singing?

For singing birds, or birds attending to songs of conspecifics, selective auditory responses to song can clearly have behavioral significance. The behavioral significance of auditory responses during sleep is less clear. These responses may be an epiphenomenon of a system entering an "unprotected" state that would not normally be activated at night by sensory input (Dave et al. 1998). If activity at night has behavioral relevance, it must be carried not by auditory responses but by bursting patterns of activity during undisturbed sleep. RA neurons exhibit spontaneous bursting during sleep and this bursting mimics the patterns the same neurons express during singing (Dave and Margoliash 2000). These data suggested a model in which patterns of singing are memorized during the day, and during sleep, bursting throughout the song system that mimics singing ("replay") consolidates a subset of those patterns (see also Margoliash 2001). The strong bursting of HVc activity we observed in sleeping birds supports this model. The apparent recruitment during HVc bursts of a large population of otherwise refractory neurons that we observed in multiunit recordings suggests that HVc projection neurons also participate in bursting. Furthermore, when recently developed statistical tools were applied to data collected in this study from sleeping birds, a match between the patterns in spontaneous activity and in response to playback of the bird's own song was observed in HVc neurons (Chi et al. 2002). Thus aspects of replay during sleep also exist in HVc.

During sleep, HVc neurons burst and are auditory. In awake animals, HVc neurons burst most strongly during singing. If similar mechanisms generate bursting during sleep and during singing, this suggests that gating mechanisms also serve to regulate access to highly selective auditory input to the vocal motor and anterior forebrain pathways when the bird sings. Auditory gating related to singing should have the temporal specificity of song itself. A previous study concluded that auditory responses in HVc are suppressed during and immediately after song production (McCasland and Konishi 1981). It is unclear whether the techniques employed in that studyplayback of song not temporally locked to singingcould detect syllable-by-syllable auditory gating that might accompany singing. When auditory feedback that was temporally locked to singing was presented to singing adult birds, song was eventually disrupted (Leonardo and Konishi 1999; Brainard and Doupe 2000). Additionally, the heterogeneous effects of behavioral state change on different populations of HVc neurons observed in the present study suggest that the suppression of auditory responses during singing may not obtain across all HVc neurons.

Awake auditory activity in HVc can support song perception

The awake auditory activity we observed for many HVc neurons lends supports to the long-standing proposition that auditory responses of HVc neurons contribute to perception of conspecific songs (Margoliash 1986; Margoliash and Konishi 1985). This hypothesis has received support in the form of studies demonstrating the contribution of various nuclei to song perception, including lesion studies in HVc (Brenowitz 1991; Del Negro et al. 1998; Gentner et al. 2000) and lesion and volumetric studies of nuclei in the anterior forebrain pathway (Burt et al. 2000; Hamilton et al. 1997; Scharff et al. 1998), that receive their auditory input from HVc (Doupe and Konishi 1991). Data showing that area X neurons also exhibit auditory responses during wakefulness (unpublished observations) also support this hypothesis.

The hypothesis that song system auditory responses contribute to song perception by acting as an autogenous template (see Margoliash 1986) has been difficult to reconcile, however, with the high degree of selectivity for BOS observed in HVc (Margoliash 1983, 1986; Margoliash and Fortune 1992) and the AFP (Doupe 1997; Doupe and Konishi 1991). Previous studies characterizing song system neurons as highly song selective were conducted mostly in anesthetized birds where, as in sleep, strong correlated responses to BOS overwhelmingly dominate (e.g., Sutter and Margoliash 1994). In contrast, in the present study we observed a decrease in auditory selectivity accompanied by loss of inhibitory components within HVc in awake birds, resulting in relatively stronger responses for conspecific songs. The similarity of auditory responses across HVc interneurons also sharply decreased between sleep and waking periods, which may reflect a decrease in the synchronization of auditory responses across HVc. Increased synchronization of neuronal populations has been hypothesized as a mechanism underlying perceptual phenomena in sensory systems (Gray and Singer 1989). We propose that desynchronization may also have perceptual significance in sensorimotor systems. The behavioral data suggest that the auditory neurons in HVc and the AFP participate in the perception of conspecific songs. The de-correlation of HVc neurons during wakefulness would yield greater dynamic range if neuronal synchronization were involved in the perception of a diverse set of conspecific songs.