| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
1 Department of Biology, University of Pennsylvania, Philadelphia, Pennsylvania 19104; 2 Neuroscience Graduate Group, University of Pennsylvania, Philadelphia, Pennsylvania 19104
Submitted 2 January 2003; accepted in final form 26 August 2003
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMETHODSRESULTSDISCUSSIONDISCLOSURESACKNOWLEDGMENTSREFERENCES |
|---|
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONDISCLOSURESACKNOWLEDGMENTSREFERENCES |
|---|
). Certain motor behaviors, such as language, may be primarily under the control of a dominant specialized hemisphere (Geschwindt 1970). In many cases, however, motor behavior requires the finely tuned coordination between motor patterns produced in each hemisphere. A classic example of such coordination is the production of bimanual movements (Kelso et al. 1979). Because sectioning of the corpus callosum causes deficits in bimanual coordination (Brinkman and Kuypers 1973), it has generally been assumed that interhemispheric motor coordination is primarily mediated by the corpus callosum (Donchin et al. 1998). An equally complex motor behavior that requires finely tuned coordination between hemispheres is the production of song in passerine birds (Vu et al. 1998). Interestingly, the forebrain structures that control song production are not connected between hemispheres (Wild 1997) (birds do not have a corpus callosum). Therefore in contrast to mammalian motor systems, the song system provides a powerful model for investigating the mechanisms of interhemispheric motor coordination in the absence of direct callosal projections.
Song production requires the finely tuned coordination of respiratory muscles and those controlling the vocal organ (syrinx) (Goller and Suthers 1996; Suthers 1997). In addition, because the syrinx is a bilateral structure capable of producing separate independent sounds in each syringeal half (Goller and Suthers 1995; Suthers 1990), exquisite coordination is also required between the muscle commands controlling the left and right syrinx. Neural control of respiratory and syringeal muscles is ultimately controlled by a complex network of interconnected forebrain nuclei (Nottebohm et al. 1976, 1982) known collectively as the song system. This system projects to vocal and respiratory centers in the brain stem (Wild 1997) and is bilaterally organized in that each hemisphere contains anatomically identical song control nuclei (Nottebohm et al. 1976).
Nucleus HVc (also known as HVC) is a forebrain song nucleus necessary for the production of song (McCasland 1987; Nottebohm et al. 1976; Yu and Margoliash 1996). It forms an integral part of a song pattern–generating network (Vu et al. 1994) and its activity is highly stereotyped across song renditions (Hahnloser et al. 2002; Hessler and Doupe 1999; Vu et al. 1998; Yu and Margoliash 1996). Control of brain stem vocal control centers is primarily ipsilateral in nature (Wild et al. 2000) but song premotor activity in HVc appears to be highly coordinated between hemispheres (Vu et al. 1994, 1998) and this coordination is most likely achieved by feedback signals that originate in the thalamus or midbrain. Specifically, nucleus robustus archistriatalis (RA), which receives direct motor inputs from HVc, projects in turn to structures (Fig. 1A) that project (indirectly) back to HVc in both hemispheres (Striedter and Vu 1998; Vates et al. 1997). These include the dorsomedialis posterior (DMP) thalamic nucleus (Vates et al. 1997), the midbrain nucleus DM (Striedter and Vu 1998), and nucleus parambigualis (PAm), a brain stem structure containing inspiratory bulbospinal neurons (Reinke and Wild 1998; Striedter and Vu 1998; Wild 1997). Two of these pathways project back to HVc by way of nucleus uvaeformis (Uva) (Striedter and Vu 1998), a thalamic nucleus that exhibits premotor activity bursts during singing and has been hypothesized to play a role in the initiation, and possible coordination, of song production (Williams and Vicario 1993).
|
|
; Vu et al. 1998). However, because these recordings were performed at different times and often in different birds, it was not possible to assess the fine temporal relationship between activity patterns in each hemisphere and address how song motor patterns might be coordinated and synchronized between hemispheres. In the present study, HVc neural activity was recorded simultaneously from multiple electrodes implanted in each hemisphere and shown to be highly synchronized between hemispheres during short portions of the song that were time-locked to key transitions in the song. The lack of direct connections between left and right HVc suggests the existence of a common input that drives HVc activity in each hemisphere. Because zebra finch song is hierarchically organized, characterization of the source and timing pattern of the inputs that synchronize HVc activity are likely to provide important insight into the nature by which higher-order song features may be encoded in the avian song system. |
METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONDISCLOSURESACKNOWLEDGMENTSREFERENCES |
|---|
Adult (>200 days posthatch) male zebra finches (Taeniopygia guttata) were obtained from the lab's breeding colony or from a commercial breeder (Magnolia Bird Farm, Anaheim, CA and Canary Bird Farm, Old Bridge, NJ). In some cases (5/9), small silastic pellets of testosterone proprionate were implanted subcutaneously to increase the frequency of singing behavior. Electrodes were chronically implanted into both the left and right HVc. In some cases (4/9), 2 electrodes (separated by 200–400 µm) were also placed in the same HVc. One electrode was also sometimes deliberately placed just outside of HVc to make sure there was no cross talk between electrodes. All electrodes for each bird were connected to a nanoconnector (Ultimate, Orange, CA or Omnetics, Minneapolis, MN) and the whole assembly was cemented (GripCement, Milford, DE) onto the bird's skull with the nanoconnector placed several mm away from the implanted electrodes. Recording electrodes were fabricated from Formvar-insulated nichrome wires (25 µm bare diameter; A-M Systems, Seattle, WA) whose tips were electroplated with rhodium to lower the tip impedance. Typical impedances of electrodes ranged from 100 k
to 1 M (measured at 1 kHz). A silver wire was inserted under the dura and partly cemented to the skull to ground the animal. To locate the sites of implanted electrode tips, parasagittal sections of 40 µm thickness were cut on a freezing microtome, mounted, and stained with cresyl violet as described previously (Vu et al. 1998). The location of individual recording electrodes was identified based on gliosis that formed around electrode tips, which had been implanted for 4–12 wk. Electrodes placed within the borders of HVc could be unambiguously identified as being placed within the nucleus. The exact location within HVc could only be estimated because of the gliosis scar around the electrode tip. All procedures were approved by an institutional animal care committee.
Neural recording
Several days after surgery, birds were placed in custom-designed Plexiglas recording chambers (manufactured by Herb Adams, Caltech Machine Shop) and attached to a liquid mercury commutator by a flexible cable made from a braided bundle of 7–9 ultraflexible PVC-coated 38-gauge copper wires (Cooner Wire, Chatsworth, CA). The recording chamber was itself placed in a sound-attenuating chamber (Industrial Acoustics Company, Bronx, NY). The flexible cable/commutator arrangement allowed the bird full freedom of movement while providing electrical connections between the electrode and the recording amplifier. In most cases, a female zebra finch was placed in an adjacent cage to induce directed singing. Birds were provided with unrestricted access to food, gravel, and water. Although birds could remain electrically connected for several weeks at a time, activity was typically recorded from each individual bird over only a 1- to 4-day period. In all cases, birds displayed normal feeding, singing, and courtship behavior and showed no signs of discomfort in this recording setup. To test the effect of auditory feedback deprivation on neural activity in HVc, some implanted birds were deafened by cochlear removal (Konishi 1964).
The recording cable was attached to a low-noise, LinCMOS, operational amplifier (TLC27L4B, Texas Instruments, Dallas, TX) that was connected to the bird's head. This amplifier served as a unity-gain voltage follower providing a low-impedance path from the bird's head to the main recording amplifier, thus greatly reducing movement artifacts from the signal. Auditory and neural records were amplified (4-channel differential AC amplifier, A-M Systems, Everett, WA or a BioAMP, Tucker-Davis Technologies, Gainesville, FL), band-pass filtered between 300 Hz and 10 kHz (7-pole antialiasing filter FT6–7, Tucker-Davis Technologies), and digitized at 20 kHz with a 100-MHz, 16-bit data acquisition (DAQ) board (PCI-MIO-16XE-10 from National Instruments, Austin, TX). Acquisition software (Labview, National Instruments, Austin, TX) was custom-written (A. Leonardo, Caltech, Pasadena, CA) and allowed the simultaneous recording of one sound channel and up to 4 neural channels. Individual files were collected automatically by triggering off of specified features (e.g., amplitude, spectral profile) of the bird's vocalization. Great care was taken to ensure that song files used for analysis were completely devoid of vocalizations from the neighboring female. In many cases this meant that a large number of songs could not be used for analysis and explains why only 10–20 songs were analyzed for each bird. Neural and sound files were analyzed post hoc using custom-written Matlab scripts (Mathworks). All data were collected and analyzed using a Dell computer.
Data analysis
SONG ANALYSIS AND ALIGNMENT. The highly stereotyped nature of syllables in zebra finch song allows precise alignment of neural traces across many renditions of the same syllable. To construct a profile of neural activity for individual syllables, syllables of a given type (e.g., syllable A) were aligned to the syllable onset and the corresponding rectified HVc neural traces were smoothed (see following text) and averaged. Syllable onset time was determined manually using a custom-written Matlab script. Similar methods were used to align syllables based on their acoustic offset. In some cases, alignment was performed using the calculated cross-correlation time lag between a reference syllable and individual renditions of that syllable. This method of syllable alignment yielded identical results to those obtained using the manual alignment method. Because the time delay between premotor activity and syllable onset is about 45 ms (Yu and Margoliash 1996; present study), syllable-specific neural records typically started 100 ms before syllable onset and ended at syllable offset. In cases such as the 1st syllable of the song, where the 100-ms period preceding the acoustic onset of the syllable might have included an introductory note, care was taken to reduce this period to 60 ms so as not to include premotor activity associated with the introductory note. The variable duration of intersyllable intervals inevitably caused individual records to contain small portions of the premotor neural trace from adjacent syllables. In all figures, song is represented as either the spectrogram or the derivative of that spectrogram (Tchernichovski et al. 2000). Spectrograms were calculated using scripts written either in Matlab (Leonardo and Konishi 1999) or C++ (Tchernichovski et al. 2000) using a sliding-window (5–8 ms) method in which each time point consisted of the direct multitaper estimate of the power spectrum.
ANALYSIS OF NEURAL ACTIVITY. To assess syllable-specific neural traces across electrode sites, neural waveforms for each rendition of the syllable were rectified and smoothed by convolving traces with Gaussian filters of variable widths (SD = 1, 5, or 10 ms). These different smoothing parameters differentially preserve the temporal structure of the premotor neural trace (Fig. 3A). The similarity between smoothed waveforms was assessed by calculating the linear correlation (Pearson's correlation coefficient). Correlation coefficients for all syllable comparisons were presented as means ± SD and were calculated for each syllable.
|
In some cases, to visualize neural activity associated with an entire sequence of syllables, such as in a motif, profiles of premotor activity were constructed for the first 2 motifs of the song. To construct these profiles, a canonical song was used as a template to concatenate neural activity profiles obtained from individual song syllables (Fig. 2B). These "whole song" neural activity profiles were then normalized by dividing by the maximum activity level for each recording site. It should be emphasized that these "whole song" profiles were used for visual display only and were never used for quantification purposes.
To quantify the dynamic nature of coincident activity in HVc between electrode pairs, a sliding-window cross-covariance method (using the Matlab function xcov.m) was used to compute a profile of song-related coincident neural activity. This method of analysis was only ever performed on neural traces smoothed with a narrow Gaussian function (SD = 1 ms). Cross-covariance, which is simply the cross-correlation of both traces whose means are subtracted, was computed for individual windows (20-ms-long segments), which were then moved piecewise in 4-ms steps. For a typical neural trace of 200 ms, the syllable would therefore be represented by 48 overlapping 20-ms windows. The specific window size was optimized to capture rapid synchronous onsets and burst events on the time scale of those shown in Figs. 5 and 14. The correlation coefficient r for each time lag (k) was then obtained by normalizing each cross-covariance by the product of each autocovariance at time lag 0 ms such that
 |
). A value of r = 1 indicates that both waveforms are perfectly correlated, whereas a value of r = 0 means waveforms are not correlated. Negative r-values describe waveforms that are anticorrelated. The sliding cross-covariance matrix calculated for each syllable rendition was then averaged for all renditions of that syllable to obtain a matrix where each point represented the mean r-value at any given time and time lag. Cross-covariance matrices were color coded to illustrate the different r-values with red representing high levels of correlation (r > 0.5) and blue representing either negative r-values or r-values near zero as shown in Fig. 6.
|
|
|
To correlate neural patterns in both hemispheres with individual acoustic features of the song, matrices were transformed into vectors by taking the maximal r-values within the time-lag range of ±1 ms (see Fig. 8A). These vectors are referred to as covariance profiles. The Pearson correlation was used to compare the similarity between "covariance profiles" of the same syllable in different motifs. Correlation coefficients for all syllable comparisons were presented as means ± SD. Peaks in the "covariance profile," which were arbitrarily defined to be significant if r > 0.5, were then compared with acoustic features that occurred 45 ms after these peaks. This delay is the average premotor delay obtained from all birds in this study.
|
|
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONDISCLOSURESACKNOWLEDGMENTSREFERENCES |
|---|
To study the relationship between song motor patterns in each hemisphere, multiple electrodes (2–4) were implanted in left and right HVc of adult male zebra finches (Fig. 1A). Neural activity was monitored from multiunit clusters of neurons in each hemisphere during singing and quiet non-vocalizing episodes. Neurons in HVc generally exhibited little spontaneous activity during quiet periods but showed a dramatic increase in activity 40–70 ms before song onset (Fig. 1B). With the exception of the first few introductory notes, where onset and offset were clearly demarcated, neural activity remained generally elevated during the entire song (Fig. 1B), although clear modulations in firing patterns could be readily observed (Figs. 2B and 14). After song termination, activity usually rapidly returned to baseline levels, and in some cases (Fig. 1B), was even suppressed for several hundred ms thereafter. The general pattern of HVc song premotor activity was similar to patterns obtained from multiunit recordings in previous studies (McCasland 1981; Vu et al. 1998; Yu and Margoliash 1996).
A defining feature of zebra finch song is its highly stereo-typed production. A typical song consists of several introductory notes, which are followed by a stereotyped sequence of syllables, known as motifs, that are repeated several times throughout a given song bout (Sossinka and Boehner 1980) (Fig. 2A). Individual syllables can often be subdivided into smaller song elements known as notes (Immelmann 1969; Zann 1996). To analyze premotor activity patterns, neural records obtained for each song rendition were aligned to the onset of individual song syllables. Neural traces were first rectified and smoothed before being averaged across all renditions to create a "neural activity profile" for each syllable. These activity profiles were then used to compare syllable-specific neural activity patterns across motifs and electrode pairs.
A graphic representation of the similarity of HVc song premotor activity patterns across motifs as well as across hemispheres is shown in Fig. 2B. In this example, neural activity profiles computed for individual syllables were smoothed with a wide Gaussian filter (SD = 10 ms) and then concatenated to provide a graphic representation of the "neural activity profile" for the entire motif (see METHODS). Smoothing the neural waveform with this filter retains the slowly varying modulations in firing pattern while smoothing out rapidly modulated events (see Fig. 3A). This figure illustrates the highly stereotyped nature of smoothed song premotor firing patterns in HVc between motifs (Fig. 2B). A similar sized smoothing filter was previously used to illustrate the similarity of song premotor patterns in HVc across syllable renditions (Vu et al. 1998).
Quantitative comparison of smoothed neural waveforms (SD = 10 ms) associated with individual syllables shows that the neural pattern is highly correlated for the same syllables produced in the 1st and 2nd motifs (r = 0.73 ± 0.07; n = 34 syllable pairs from 9 birds; Fig. 3B gray bars). No significant correlation is observed when activity is compared between different syllables (r = 0.05 ± 0.09; n = 34 syllable pairs in 9 birds). In contrast, comparisons of syllable-specific neural traces smoothed with a narrower filter (SD = 1 ms; Fig. 3B, black bars) or no filter (Fig. 3B, white bars) give much lower correlation values [F(2,21) = 175; P < 0.001; one-way ANOVA]. Average correlation values are 0.39 ± 0.09 and 0.06 ± 0.02, respectively. These results suggest that the similarity observed between the same syllables produced in different motifs is high only when analyzing slowly varying modulations in the firing pattern.
Slowly varying HVc song motor patterns are highly correlated across hemispheres
In all cases in which recordings were monitored in both hemispheres (n = 9 birds), neural activity profiles in HVc during singing were remarkably similar across hemispheres. In Fig. 2B, smoothed (SD = 10 ms) HVc premotor activity profiles from the left (solid gray) and right (black line) side, for both the 1st and 2nd motifs of the song, are superimposed to show this similarity. Quantitative comparison of smoothed neural activity for individual syllables shows a high level of correlation between left and right HVc (r = 0.73 ± 0.06; n = 34 syllable pairs from 9 birds). No significant correlation is observed when activity in left and right HVc is compared between different syllables (r = 0.06 ± 0.08; n = 34 syllable pairs from 9 birds). In several cases multiple electrodes were successfully implanted in HVc in the same hemisphere (separated by 200–400 µm). Neural activity in ipsilateral electrode pairs showed correlation levels similar to those measured between hemispheres (r = 0.72 ± 0.11; n = 17 syllable pairs from 4 birds). A summary of correlation values for each recorded bird is shown in Fig. 3.
Similar to comparisons made between identical syllables across motifs, correlations of smoothed neural activity between hemispheres decrease significantly as the size of the smoothing filter is narrowed [F(2,24) = 402; P < 0.001; one-way ANOVA]. Average correlation values of neural traces smoothed with a narrow filter (SD = 1 ms) or no filter at all are 0.41 ± 0.06 and 0.07 ± 0.02, respectively, for contralateral HVc electrode pairs and 0.47 ± 0.07 and 0.11 ± 0.04 for ipsilateral electrode pairs. Because electrodes pairs are placed at different sites within HVc, the similarity between ipsilateral or bilateral electrode pairs suggests that HVc neurons exhibit the same slowly varying motor pattern independent of placement within the nucleus. This similarity in activity between recording electrodes is not caused by cross talk between electrode pairs because: 1) placing one of the 2 electrodes a short distance outside of HVc (300–500 µm) fails to show any vocalization-specific neural activity (data not shown), and 2) electrode pairs are only correlated when neural activity is smoothed with a wide filter.
The present results suggest that slowly varying premotor patterns are nearly identical in both hemispheres for the entire duration of the syllable. However, because correlation values are low (0.41 ± 0.06) when neural traces are smoothed with a narrow window (SD = 1 ms), these results also suggest that the fine temporal patterns in the premotor trace are significantly different between electrode pairs. The low correlation values obtained for whole syllable comparisons, however, might potentially mask the existence of short segments in the neural trace that are highly correlated across hemispheres. Neural traces associated with each syllable were therefore divided into short 20-ms segments and the linear correlation between left and right HVc was calculated for each segment. The distribution of correlation values obtained from whole syllables (white bars; n = 987 renditions of 21 syllables in 4 birds) and those obtained from the segment correlations (black bars; n = 21,322 segments obtained from all syllables in 4 birds) are shown in Fig. 4. The distribution of correlation values obtained from the short 20-ms segments is broader than that obtained for the whole syllable correlations and contains many more high correlation values. For the whole syllable comparisons, the number of correlation values >0.7 is only 0.3% (3/992 syllable renditions in 4 birds). This is in contrast to 12.6% (2,691/21,322) for the short segments. These data indicate that HVc premotor activity, although mostly uncorrelated, nevertheless contains a small number of short segments that are highly correlated across hemispheres. These high correlation values are not simply attributed to chance, given that the distribution is statistically distinct from the distribution of correlation values obtained from these same, but randomly paired, segments (dotted white line; Fig. 4; Kolmogorov–Smirnov KSa = 36.74, P < 0.001).
|
Interhemispheric synchronization of song motor activity
The existence of short segments of highly correlated neural activity between electrode pairs suggests that neural activity may be correlated between hemispheres during precisely timed segments in the song neural trace. Figure 5 illustrates an example of such temporally precise correlated premotor activity. Neural activity is elevated during most of the song but displays clear periods where the modulations in the firing pattern are nearly identical (gray boxes labeled 1 and 2) in each hemisphere (left HVc trace in black and right HVc trace in red) and are highly stereotyped across renditions of the syllable. These highly correlated segments of activity are often flanked by periods of activity where the neural discharge pattern appears quite different between the left and right HVc (dotted boxes labeled a and b). Activity during these periods is also less stereotyped across renditions. In this example, the portion of the neural traces highlighted in gray is characterized by discrete bursts of activity that occur at precisely the same time at each electrode. Similar patterns of synchronized and desynchronized neural activity are observed in bilateral and ipsilateral electrode pairs.
To quantify the relationship between activity patterns in each hemisphere during singing, a sliding-window cross-covariance method was used to compute a profile of synchronous neural activity between electrode pairs. The cross-covariance, which is simply the cross-correlation of both traces whose means are subtracted, is computed for individual windows (20-ms-long segments), which are then moved piecewise in 4-ms steps. The specific window size is optimized to capture correlated burst events on the same time scale of those shown in Fig. 5. Neural traces are rectified and smoothed with a narrow filter (SD = 1 ms) and the cross-covariance function for each segment is normalized to the autocorrelation function of that segment at 0-ms time lag (Diggle 1990) (see METHODS) such that the value of the cross-correlation function at any given time lag represents the correlation coefficient r. In the present study, synchronous is defined as events, in the rectified smoothed multiunit neural trace, that are highly correlated (r > 0.5) between hemispheres using this method.
A pseudocolor representation of r-values, obtained from cross-correlating neural traces obtained from left and right HVc, is shown in Fig. 6. This quantitative representation of synchronous activity, shown for both the 1st and 2nd motifs of the song, reveals a pattern of highly correlated activity (red peaks) that is remarkably conserved between motifs. Cross-covariance analysis reveals strongly conserved patterns of synchronous activity between motifs (see quantification in following section) in all recorded ipsilateral (n = 4) and bilateral (n = 9) electrode pairs. This stereotypy of interhemispheric synchrony suggests a possible direct relationship between coincident HVc activity patterns and the production of specific song features.
From a sample of 9 implanted birds, 20 syllables from 8 of these birds (see Table 2) were carefully chosen to compare neural patterns with song acoustic features. All chosen syllables had sharp acoustic onsets that allowed reliable syllable alignment and were 
50 ms long. Although the pattern of coincident activity clearly differs for each individual syllable, 2 general features emerge that are common to all syllables. First, all correlations occur at a 0-ms time lag, suggesting that both hemispheres are driven by a common synchronizing input. Second, neural activity is maximally correlated about 45 ms before syllable onset.
Cross-covariance matrices for all syllables were summed to represent the common features that could be extracted from the correlation analysis between left and right HVc. In Fig. 7A, maximal correlation values at all time lags (y-axis), conditional that they be >0.5, are shown for each time window (x-axis) of the sliding cross-covariance matrix. The overall distribution of these conditional maximal correlations shows that all of these values occur within a time lag of ±2 ms, with the majority of them centered at 0-ms time lag (Fig. 7B). In the time axis (Fig. 7C), the temporal distribution of these correlation values (thick black line) reveals a prominent peak at about 45 ms before syllable onset (peak occurs at –43.4 ms; Fig. 7C). In the interval between this initial peak and a secondary peak that occurs about 10 ms after syllable onset, activity is largely uncorrelated between hemispheres. The initial peak likely represents the onset of premotor activity because it is temporally correlated with the rise in activity (Fig. 7C; filled gray histogram, smoothed with a Gaussian filter whose SD = 1 ms).
|
Because all high correlations (i.e., r > 0.5) were centered around the 0-ms time lag, individual cross-covariance matrices were collapsed into vectors that were constructed by taking the maximal r-value in the time-lag range of –1 to +1 ms. These vectors will be referred to as covariance profiles. A representative example of such a "covariance profile" is shown in Fig. 8A for the same syllable produced in 3 different motifs. For reference, a control "covariance profile" (±2 SD) obtained from correlating normal left HVc activity with the reversed neural trace obtained from the right HVc is also shown in this figure (Fig. 8A; gray area near the 0 correlation value). A direct comparison of the relationship between raw neural activity and "covariance profiles" is shown in Fig. 5. This example illustrates the tight correspondence between visually apparent patterns of coincident activity and the corresponding high r-values obtained from the "covariance profile."
Comparing "covariance profiles" between the 1st and 2nd motifs, for the 20 syllables whose acoustic onset could be reliably defined, shows that these profiles are highly stable across renditions of the same syllable. The linear correlation obtained for left/right HVc "covariance profiles" in the 1st and 2nd motifs is 0.78 ± 0.15 (n = 20 syllables in 8 birds). These correlation values are similar to those obtained from ipsilateral paired recordings (r = 0.75 ± 0.19; n = 11 syllables in 4 birds). In contrast, "covariance profiles" between different syllables in the 1st and 2nd motifs are uncorrelated (r = –0.010 ± 0.22, n = 20 syllables in 8 birds). A summary of average correlation values for each bird is shown in Fig. 8B.
To investigate whether the pattern of correlated activity between hemispheres is linked to the produced vocal output, correlation values were compared under normal conditions and under conditions where neural traces were shuffled across renditions (Fig. 9A). Briefly, neural traces aligned to a given syllable were shuffled pairwise such that the neural trace from the left HVc was correlated with the neural trace from the right HVc from a previous rendition of that syllable. The "covariance profile" obtained from shuffled data were then compared with those obtained under normal conditions. Qualitatively, these profiles appeared remarkably similar (Fig. 9B) with a general tendency for the size of the peaks to be smaller when the data were shuffled (Fig. 9B; dotted line). Quantitatively, comparison of the correlation coefficient obtained from calculating the linear correlation of the "covariance profile" for normal data between the 1st and 2nd motifs was not significantly different (P > 0.05; paired t-test) from correlations obtained between the shuffled 1st motif and a normal 2nd motif (Fig. 9C). The linear correlation coefficient was, respectively, 0.77 ± 0.10 for the shuffled data (white bar; n = 11 syllables in 4 birds) and 0.82 ± 0.09 for the normal data (black bar; n = 11 syllables in 4 birds). Both of these values were significantly greater than the correlation values obtained from comparing "covariance profiles" when one of the neural traces was reversed in time [gray bar; F(2,51) = 187; P < 0.05; one-way ANOVA]. To compare the magnitude of the correlations under both conditions, the total number of r-values >0.5 (hereafter referred to as high r-values) were counted for each "covariance profile" (Fig. 9D). The mean number of high r-values was smaller for the shuffled data with the mean number of high r-values, respectively, 3.52 ± 0.86 for the shuffled data (n = 11 syllables in 4 birds) and 17.77 ± 2.54 for the normal data (n = 11 syllables in 4 birds; P > 0.05, paired t-test).
|
Because correlation values remain relatively elevated and show the same overall profile after the data are shuffled, these results suggest that most of the correlated activity is directly linked to song output. The difference in the overall magnitude of the correlation values between the normal and shuffled traces is difficult to resolve. One possibility is that higher correlations observed during pairwise comparisons represent an inherent synchronization of activity across hemispheres that occurs independently of the bird's vocal output. Alternatively, because each syllable rendition is slightly different from the other, it is possible that the difference in correlation values may also be caused by slight misalignments of neural traces across renditions. Such misalignments would be expected to increase as one moves farther away in time from the point of syllable alignment (i.e., the syllable onset). Supportive evidence for this possibility is shown in Fig. 9B, where correlation values decrease as one moves farther away from the syllable onset time point.
Interhemispheric synchronization occurs in the absence of auditory feedback
Although HVc exhibits premotor activity during singing, the existence of auditory-responsive neurons within HVc (McCasland and Konishi 1981; Schmidt and Konishi 1998; Yu and Margoliash 1996) suggests that the present observations could result in part from neural responses caused by auditory feedback of the bird's own song. To test this possibility directly, several birds (n = 3 birds; see Table 1) were deafened by cochlear removal to prevent auditory feedback. An example of the "covariance profile" calculated from the same left/right HVc electrode pair before (black line) and after deafening (gray line) is shown in Fig. 10. In this example, because the bird tended to sing little in the days after cochlear removal, a 4-wk period separated recordings obtained before and after deafening. The large number of peaks in the "covariance profile" with values of r > 0.5 illustrates that activity is highly correlated between hemispheres in both normal and deafened birds. It also reveals that the overall pattern of synchronous activity remains stable after deafening. Linear correlation of "covariance profiles" generated before and after deafening reveals that the pattern of synchronous activity between left and right HVc is stable in all 3 birds (r = 0.58 ± 0.11; n = 9 syllables with sharp acoustic onsets in 3 deafened birds).
|
|
Correlated HVc activity defines key timing elements such as syllable and note onset
Activity recorded during production of all analyzed syllables (20 syllables; 8 birds) showed left/right HVc "covariance profiles" that were highly stereotyped from one motif to the next (Fig. 8B). In general, the amount of time where activity was highly correlated across hemispheres (i.e., r > 0.5) for a given syllable was positively correlated with syllable duration (Fig. 11; regression = 0.57). This suggests that each syllable-specific motor pattern contains more than simply a segment that is correlated with the acoustic onset of the syllable. Because syllable length is likely related to syllable complexity, this relationship suggests that synchronization may be associated with features that specify syllable substructure. In the sections that follow, representative syllables will be used to illustrate the relationship between neural patterns, "covariance profiles," and the associated vocal acoustic features in 3 different categories of syllables. These categories are 1) syllables with simple harmonic structures (Fig. 12), 2) syllables that are composed of 2 clearly distinguishable notes (Fig. 13), and finally 3) complex syllables with multiple note elements (Fig. 14).
|
|
|
SIMPLE SYLLABLE. In the population of 9 birds recorded for the present study, 4 birds produced songs that contained simple harmonic syllables. All 4 of these syllables showed the same general pattern and 2 of these are shown in Fig. 12 to illustrate the general relationship between neural activity patterns in both hemispheres and syllable morphology. In these examples, the smoothed (SD = 1 ms) neural activity profile (middle) shows a sudden increase in activity in both the left (gray) and right (black) HVc about 45 ms before the syllable's acoustic onset. This correlated increase in activity in both hemispheres is reflected by the 1st large peak [Peak 1 in the left (r = 0.71) and right (r = 0.68) syllable] in the "covariance profile" (bottom). The delay between the correlated increase in activity and the acoustic onset of the syllable is consistent with the delay calculated from the population analysis (Fig. 7C) and likely represents the premotor onset response for this syllable. After the initial increase in activity, neural discharge levels, although modulated, remain relatively elevated and uncorrelated for another 40 ms. Interestingly, about 50 ms after the initial correlated increase in activity in both hemispheres, there is a 2nd correlated increase in activity [Peak 2 in the left (r = 0.62) and right (r = 0.67) syllable].
The relationship between this 2nd correlated increase in activity and the syllable is difficult to assess because shifting the peak by the premotor delay of 45 ms places the pattern of activity squarely in the middle of the syllable. Interestingly, a similar pattern is observed in the complex syllable shown in Fig. 14 where the 1st component of the syllable consists of a simple harmonic note similar in its acoustic structure to the syllables described here. Shifting the 2nd peak in the complex syllable by the same premotor delay of 45 ms also aligns it to the middle of the note (2nd dotted line in Fig. 14). This similarity in neural patterns obtained for acoustically similar elements from different recordings in different birds suggests a possible common encoding scheme for the production of acoustically simple syllables.
TWO-NOTE SYLLABLE. Given the striking relationship between correlated increases in activity in both hemispheres and the acoustic onset of syllables, it was of interest to investigate whether a similar relationship could be observed for the acoustic onset of notes. Although many syllables are made up of 2 or more notes, it is sometimes difficult to determine the precise transition between the notes that make up these syllables. In a few cases, however, clear transitions can be identified. In the 9 birds that were recorded, 4 syllables from 4 different birds were identified as having clear syllable transitions. These syllables were analyzed in detail and 2 of these are shown in Fig. 13. All 4 syllables showed the same general pattern.
In the examples shown below, activity is, as in previous examples, highly correlated with the acoustic onset of the syllable [Peak 1 on the "covariance profile" for both the left (r = 0.59) and right (r = 0.53) syllable]. In both syllables, however, a 2nd correlated increase in activity, shown by the large covariance peak [Peak 2 (r = 0.68) in the left panel and Peak 3 (r = 0.54) in the right panel], occurs 60 ms (left syllable) and 98 ms (right syllable), respectively, after the initial increase in activity. Assuming a premotor lead time of 45 ms, these correlated increases in activity correspond almost exactly to the acoustic onset of the 2nd note in the syllable (dotted line). In one of the syllables (ZF16 Syllable D), there is also a correlated decrease in activity followed by a correlated burst in both hemispheres (Peak 2; r = 0.51). This correlated pattern of activity falls precisely in the middle of the syllable's 1st note and is reminiscent of the pattern observed for the simple harmonic syllables. Taken together, these examples suggest that both syllable and note onset are temporally linked to the correlated activation of HVc activity in both hemispheres.
COMPLEX SYLLABLE. In the previous examples, activity was modulated during production of the syllable but emphasis was directed at correlated activity associated with either the acoustic onset of the syllable or the notes that make up the syllable. In this example, analysis of the neural patterns associated with long (235-ms) complex syllables reveals that the premotor pattern can be highly modulated and that this modulated firing pattern is highly correlated across hemispheres. In Fig. 14, neural activity (top panel) recorded simultaneously from left and right HVc is shown for 22 different renditions of this complex syllable. In this pseudocolor representation of premotor activity, neural traces have been rectified and smoothed with a narrow filter (SD = 1 ms). Red represents high levels of activity and dark blue represents the absence of neural activity. Particularly striking in this example is the relatively large number of discrete neural events characterized by short bursts of activity that are highly stereotyped across renditions. Many of these bursts are quite short (about 5–10 ms in duration) and are reminiscent both in their precision of firing as well as in their duration to the sparse bursts produced by HVc projection neurons during singing (Hahnloser et al. 2002). Because electrodes sample neural activity from multiple neurons, these bursts could consist of multiple neurons activated in near synchrony or single neurons producing a single burst or both. Although the neural pattern is for the most part nearly identical across hemispheres, a few differences (highlighted by the small red arrows in Fig. 14) are apparent. Equally striking in this example are the brief periods of nearly complete suppression (shown in dark blue) of activity in both hemispheres.
Analysis of the "covariance profile" reveals 7 distinct peaks whose r > 0.5. Small black arrows numbered 1–7 show the temporal relationship between the raw neural trace and each associated peak. Shifting each peak by the premotor delay of 45 ms aligns the 1st peak (Peak 1) with the onset of this complex syllable. The 2nd peak (discussed in the previous section) aligns to the middle of the first note in the syllable. Peaks 3 and 4 align, respectively, to the offset of the 1st note and the onset of the 2nd note. Peak 5 appears to be loosely correlated with a possible note transition in this complex syllable, whereas Peak 6 does not seem to be correlated to any obvious feature of the syllable. Peak 7 may be associated with the acoustic offset of the syllable but the exact relationship is complex, given that this peak corresponds to a correlated increase in activity rather than a correlated decrease in activity. Taken together, the relationship between peaks of correlated interhemispheric activity and syllable transitions in this example suggests that synchronous activation of neural activity in both hemispheres may represent timing signals that define transitions in these complex syllables. Some of these transitions may be obvious (Peaks 1, 3, 4, and 5), whereas others (Peaks 2, 6, and 7) may not.
Acoustic offset is not strongly associated with a correlated offset of activity in both hemispheres
The relationship between acoustic offset and neural activity is often difficult to assess. Because syllables within a motif precede other syllables, it is hard to determine the end of the premotor activation period for one syllable and the onset of activity for the next syllable. This problem is clearly illustrated in Figs. 12 and 13 for the rightmost syllables where the large correlated increase in activity at the end of the trace is not associated with the syllable shown in the figure but rather with the syllable that directly follows it. To circumvent this problem, the relationship between premotor activity and acoustic offset was analyzed only during production of the last syllable of the bird's song bout.
On average, neural activity decreases slowly toward baseline about 40 ms before acoustic offset of the syllable (Fig. 15A; filled gray histogram, top panel). This slow decrease in activity is in sharp contrast to the rapid synchronized increase in activity observed at syllable onset (see Fig. 7C). One problem with this average representation of premotor activity is that it smoothes out the considerable variability that is observed across syllables (see examples in Fig. 15, B and C). Some syllables will show rapid decreases in activity about 40 ms before the acoustic offset, whereas others will continue to exhibit bursts of activity past the expected premotor offset period (Fig. 15B). In some cases, these premotor bursts in HVc may even occur past the acoustic offset of the syllable (see Fig. 16).
|
|
In addition to the general variability of the premotor pattern associated with syllable offset, variability is also observed in the degree to which activity is correlated across hemispheres. This variability is shown in the "covariance profiles" for each of the analyzed syllables (Fig. 15A, bottom). In 7/10 birds, correlated activity is observed during a 20-ms period (–35 to –55 ms) centered around the expected premotor offset (double line in Fig. 15A). In 5 of these birds, however, this period of correlated activity is followed by at least one additional period of correlated activity during the period spanning –35 ms and the acoustic offset (dotted line in Fig. 15A) of the syllable (i.e., 0 ms). In the remaining 3 syllables, correlated activity is absent during the expected premotor onset period. Individual examples are shown in Fig. 15, B and C. The 1st syllable (Fig. 15B) shows a correlated decrease in premotor activity in both hemispheres. It is difficult to estimate, however, what part of the neural trace is truly representative of acoustic termination because there is a correlated decrease in activity 38 ms before the syllable offset (Peak 3) and then another correlated decrease in activity about 20 ms later (Peak 4). In the second example (Fig. 15C), activity generally decreases in both hemispheres but is completely uncorrelated. This is illustrated by the absence, in the "covariance profile," of any peaks >0.5 during the 100 ms that precede the syllable's acoustic offset. Remarkably, even though this syllable is aligned at its acoustic offset, neural activity associated with the syllable onset (Peak 1) and the note transition (Peak 2) is still highly correlated.
Taken together, these data
RA 
