Amish S. Dave, Albert C. Yu, Daniel Margoliash ^*

Neurons of the song motor control nucleus robustus archistriatalis (RA) exhibited far weaker auditory responses in awake than in anesthetized zebra finches. Remarkably, sleep induced complex patterns of bursts in ongoing activity and uncovered vigorous auditory responses of RA neurons. Local injections of norepinephrine suggested that the changes in response strength occur through neuromodulatory control of the sensorimotor nucleus HVc, which projects to RA. Thus, motor access to auditory feedback, which zebra finches require for song learning and maintenance, may be regulated through neuromodulation. During sleep, the descending motor system may gain access to sensorimotor song memories represented as bursting patterns of activity.

Department of Organismal Biology and Anatomy, Committee on Neurobiology, University of Chicago, 1027 E. 57th Street, Chicago, IL 60637, USA.
^*    To whom correspondence should be addressed. E-mail: dan{at}bigbird.uchicago.edu

Song learning requires auditory feedback, and the role of auditory feedback is modulated during development (3). In the forebrain, the nucleus HVc and its afferents are the principal targets of auditory input to the song system (4). Neurons in HVc project to one of two pathways, either the descending motor pathway through a projection to the forebrain nucleus robustus archistriatalis (RA) or the anterior forebrain pathway (AFP) that eventually projects back to RA (Fig. 1A). Whereas HVc and RA are necessary for singing, the AFP is necessary for the development of normal song, but lesions of AFP nuclei in the adult have little effect on singing in zebra finches (5).

We recorded single neurons in the HVc and RA of awake, freely moving animals (6). Numerous previous studies, mostly conducted in urethane-anesthetized animals, have shown that HVc (7, 8) and RA and AFP (8) neurons have auditory responses that are specific for acoustic features of the individual bird's own song (BOS) and are selective for BOS relative to conspecific songs. We also observed such selectivity in the auditory responses of single HVc neurons in awake birds but, surprisingly, failed to observe any auditory response whatsoever in RA neurons recorded under the same conditions (9). The RA neurons exhibited fast regular oscillatory spiking patterns that lacked the occasional bursts observed in recordings from anesthetized birds. The complete absence of an auditory response in RA may have been the result of a neuromodulatory response related to stress induced during a brief period when the animals were manually restrained to achieve single-unit isolation (6) (see below).

Exploring under what behaviorally relevant conditions RA neurons exhibited the auditory responsiveness observed in anesthetized animals, we discovered that when birds fell asleep, RA neurons acquired complex bursting in their ongoing activity and, remarkably, gained auditory responsiveness to BOS. At night, birds prepared for chronic recordings of RA neurons were presented with continuous playback of BOS (6). When the cage lights were turned off, motion in the cage (as judged by lack of audible movements) eventually ceased and the birds fell asleep. Without fail, sleep was accompanied by slower, less regular firing (14 single units, five birds; Fig. 1B) and a dramatic increase in the auditory response to BOS (10 of 14 single units and six multiple units in four birds were tested); the effect was sufficiently reliable to be easily seen in multiple unit traces (Fig. 1D). The RA "sleep" state of ongoing activity and BOS responsiveness was observed whenever we sampled RA during the night. The only exceptions were during brief episodes of audible movements, implying periods of wakefulness, or brief intervals (typically <30 s) during which ongoing activity transiently increased. The latter may reflect changes in stages of sleep, which in birds occur frequently and for brief intervals (10). The transition between sleeping and waking states could be rapid. For example, when a bird was awakened by a loud sound, RA neurons rapidly returned to the awake pattern of regular ongoing activity (Fig. 1C) and loss of auditory responsiveness (Fig. 1E).

To investigate the robustness of the day-night dichotomy and the role of behavioral state in modulating RA auditory responses, we tested RA multineuronal activity in three birds for responses to BOS beginning at night and continuing into the following day (6). In birds varying from posthatch day (PHD) 58, when juveniles are undergoing the process of learning their song, to adulthood, the results consistently demonstrated that during sleep, RA neurons responded strongly to BOS, whereas when animals were awake, responses were much weaker. Over all recording sites, daytime responses averaged 6 ± 8% (SD) of the responses at night or 9.3 ± 10.1% for the eight sites with statistically significant daytime auditory responses (11). In both adults and juveniles, the strongest auditory responses during the day were observed when birds were relatively silent and inactive (Fig. 1E). Restricting the analysis to the single period of behavioral quiescence with the strongest neuronal response during the day, one per recording site, the response was still just 17.1 ± 13.15% of the response at night. In two cases with birds that were accustomed to the chronic recording situation, we housed a female in an adjacent half-cage beginning in the middle of the day. The birds engaged in countercalling, the male's singing and activity levels increased and the male sang "directed" songs toward the female, but there were no changes in the very weak auditory responses. Thus, RA neurons are considerably less sensitive to auditory stimulation during wakefulness than during sleep; this sensitivity may vary with the level of stress or alertness but is not apparently engaged by social interactions.

The properties of RA neurons during sleep are similar to those observed in anesthetized animals, which could provide a useful experimental paradigm. To quantify the differences between awake and anesthetized states, we recorded a sample of RA and HVc neurons in urethane-anesthetized birds. Simultaneous recordings showed that bursting activity in RA and HVc was correlated (12). We also recorded from the RA of two awake-restrained animals and a third animal that had participated in the awake-chronic RA recordings, before and after the animals were anesthetized. In all three birds, RA neurons failed to show auditory responses while the animals were awake but exhibited strong auditory responses after the animals were anesthetized (13). Rates of ongoing activity declined and neurons commenced to burst as animals were anesthetized; this trend followed the differences comparing awake, freely moving animals and anesthetized animals (14). Particularly compelling are the three cases, one per animal, where a single unit was maintained while the bird was anesthetized, showing the decrease in ongoing rate, increase in bursting, and the emergence of auditory responses at the single cell level over the short time interval as the anesthetic took effect (Fig. 2). Thus, the auditory input to RA is presumably latent in awake animals and is unmasked by anesthesia in qualitatively similar ways as it is by sleep.

RA and HVc receive a variety of neuromodulatory inputs, including noradrenergic input from the locus coeruleus, and have high concentrations of adrenergic receptors (15). We postulated that changes in concentrations of norepinephrine (NE) may affect expression of RA auditory responses. To test this hypothesis, we characterized the ongoing activity and auditory response properties of recording sites in RA and HVc before and after pressure injections of 20 mM NE (200 to 250 nl) into RA or HVc of urethane-anesthetized zebra finches. In all birds tested, small injections of NE into HVc abolished or greatly diminished auditory responses in RA (Fig. 3), whereas small injections of NE into RA did not abolish auditory responsiveness of RA neurons (Fig. 4) (16). Complementary effects on bursting were also observed for the two manipulations. Injections of NE into RA did not eliminate bursts in RA ongoing activity, whereas injections of NE into HVc did; injections of either structure increased the rate and regularity of ongoing RA activity (17). After injections of NE into HVc or into RA, we tested HVc for auditory activity and found that it was retained (Figs. 3 and 4). In control (200 nl) injections of vehicle into HVc, we observed no effect on RA auditory responses; large (500 to 1000 nl) injections of NE into or dorsal to RA compromised auditory responses in RA (18). The large injections were associated with reflux along the dorsoventral electrode track (that passed caudal to HVc) and probably involved HVc as well as RA. Although these experiments do not unambiguously confirm the site or pharmacological mode of action underlying these phenomena, they demonstrate that auditory responsiveness in RA is sensitive to neuromodulation and suggest that the principal site of action is other than RA, presumably HVc. NE has additional, local effects on RA activity (19).

Previous studies suggested that the descending motor pathway could participate in song perception if conspecific songs are perceived in terms of the articulatory gestures necessary to produce those songs. This parallels the "motor" theory of speech perception by reference to production (20). The theory for birds, however, was based on recordings in anesthetized animals of auditory responses in the brainstem hypoglossal nucleus that arise from HVc input (21). The present data suggest that the hypoglossal responses are likely to be very weak or nonexistent in awake animals, making this pathway an unlikely candidate to contribute to conspecific song perception.

Adult zebra finches require auditory feedback to maintain their songs (22). During sleep, RA neurons exhibited bursting activity correlated with complex bursts typical of HVc neurons. An intriguing possibility is that auditory feedback processed during the day modifies the bursting behavior of HVc, which would then be communicated to RA during the night. HVc activity patterns, including bursting, are synchronized across a large spatial extent of the nucleus (23). Information encoded in these bursts may stabilize aspects of the vocal motor program encoded in the neuronal population activity patterns. A similar scheme has been suggested for stabilizing patterns of hippocampal neurons recruited during exploration of novel extrapersonal space (24).

Accounts such as the template theory of birdsong learning posit sensory and sensorimotor phases of development but do not address the role of behavioral state. Our data indicate that sensory properties of neurons in the motor pathway for song are sensitive to changes in behavioral state throughout the day as well as during sleep. Indeed, in adults, the auditory response properties of HVc neurons are apparently suppressed during singing (7). Singing recruits most if not all HVc neurons (25) and can involve a preparatory phase characterized by increasing ongoing activity starting up to 5 s before the onset of singing (26). Such dynamic reconfiguration of the network may involve local changes in concentrations of neuromodulators. Neuromodulators exhibit a complex developmental regulation in the song system (27), which could contribute to a possible modulation of the effect of auditory feedback on HVc neurons during the sensorimotor phase of song learning.

REFERENCES AND NOTES

1. M. S. Livingstone and D. H. Hubel, Nature 291, 554 (1981) [CrossRef] [Medline] ; M. Steriade, D. A. McCormick, T. J. Sejnowski, Science 262, 679 (1993) [Abstract/Free Full Text] ; D. A. McCormick and T. Bal, Annu. Rev. Neurosci. 20, 185 (1997) [CrossRef] [Web of Science] [Medline] .
2. C. Smith, Behav. Brain Res. 69, 137 (1995) [CrossRef] [Web of Science] [Medline] ; J. L. Kavanau, Neuroscience 79, 7 (1997) [CrossRef] [Web of Science] [Medline] ; D. Horn, N. Levy, E. Ruppin, Neural Comput. 10, 1 (1998) [CrossRef] [Web of Science] [Medline] .
3. M. Konishi, Z. Tierpsychol. 22, 770 (1965) [Medline] ; in Perception and Experience, R. D. Walk and H. L. J. Pick, Eds. (Plenum, New York, 1978), pp. 105-118.
4. D. B. Kelley and F. Nottebohm, J. Comp. Neurol. 183, 455 (1979) [CrossRef] [Web of Science] [Medline] ; E. S. Fortune and D. Margoliash, ibid. 360, 413 (1995) [CrossRef] [Web of Science] [Medline]; G. E. Vates, B. M. Broome, C. V. Mello, F. Nottebohm, ibid. 366, 613 (1996).
5. S. W. Bottjer, E. A. Miesner, A. P. Arnold, Science 224, 901 (1984) [Abstract/Free Full Text] ; K. W. Nordeen and E. J. Nordeen, Behav. Neural Biol. 59, 79 (1993) [CrossRef] [Web of Science] [Medline] .
6. Techniques for unit recordings in freely moving songbirds have been described elsewhere [A. S. Dave, A. C. Yu, J. J. Gilpin, D. Margoliash, in Methods for Simultaneous Neuronal Ensemble Recordings, M. Nicolelis, S. Simon, J. Corless, Eds. (CRC Press, Boca Raton, FL, in press)]. Single or multiple units were obtained by manually turning a screw to advance the mechanical microdrive while the bird was restrained, releasing the bird, and recording ongoing, auditory and vocal premotor activity. Except where noted otherwise, isolated birds were induced to sing by broadcasting female calls (25). Song stimuli were broadcast while birds were quiescent (not vocalizing, beak-wiping, pecking, or hopping excessively) but apparently alert and attentive, perched with eyes open, and making occasional small head or eye movements in response to sounds. Any neuronal record to broadcast song that was preceded or followed within 30 s by singing was discarded. In a second paradigm, recordings commenced at nighttime. After obtaining a stable multiunit signal, the bird was released into the dark cage. BOS was presented every 30 s throughout the night and following day, while recording neuronal signals and vocalizations. Daytime videotape recordings facilitated visual assessment of the bird's activity.
7. J. S. McCasland and M. Konishi, Proc. Natl. Acad. Sci. U.S.A. 78, 7815 (1981) [Abstract/Free Full Text] .
8. D. Margoliash, J. Neurosci. 3, 1039 (1983) [Abstract] ; ibid. 6, 1643 (1986); D. Margoliash and M. Konishi, Proc. Natl. Acad. Sci. U.S.A. 82, 5997 (1985) [Abstract/Free Full Text] ; A. J. Doupe and M. Konishi, ibid. 88, 11339 (1991) [Abstract/Free Full Text]; D. S. Vicario and K. H. Yohay, J. Neurobiol. 24, 488 (1993) [CrossRef] [Web of Science] [Medline] ; A. J. Doupe, J. Neurosci. 17, 1147 (1997) [Abstract/Free Full Text] .
9. Responses were computed as the percentage of change in the average spiking rate during the stimuli (typically 30 repetitions) relative to the ongoing rate. Ongoing rates were computed from long (300 s) records. Of 29 HVc single units (19 sites, seven birds) that exhibited premotor activity in relation to singing, 19 also responded to broadcast sounds. When those neurons were presented with BOS and reversed BOS (REV) (N = 19 units, 13 recording sites, six birds) or BOS and conspecific songs (CON) (N = 13 units, six recording sites, two birds), they exhibited much stronger responses to BOS (BOS > REV: paired t = 3.42, df = 18, P < 0.004; BOS > CON: paired t = 4.14, df = 12, P < 0.002). In contrast, no RA neurons recorded under the same conditions exhibited any statistically significant response to BOS (N = 28 SU, 33 recording sites, four birds).
10. E. S. Tauber, in Advances in Sleep Research, vol. 1, E. D. Weitzman, Ed. (Spectrum, Flushing, NY, 1974), pp. 133-172.
11. RA activity was recorded from one adult (> PHD 150) (three sites, 3 days), from a PHD 59 to 69 juvenile (four sites, 4 days), and from a PHD 58 to 81 juvenile (eight sites, 7 days). BOS was presented at 30-s intervals starting 3.8 to 8.0 hours before light onset and thereafter throughout the day (6). Multiunit responses were averaged for each 20 repetitions (1-ms bins), and then the mean of the ongoing activity (the average over 2 to 5 s starting 1 s after stimulus offset) was subtracted. At each site, the presence of an auditory response during sleep or wakefulness was determined by comparing distributions (t tests) of the RMS values calculated over the duration of BOS and during ongoing activity. By this measure, 14/14 sites showed auditory activity during sleep (P  0.0054), and 8/14 sites (3/3 birds) showed auditory activity during wakefulness (P < 0.05) (data from one site were lost).
12. Ongoing unit activity from HVc and RA was binned (1-ms resolution) and cross correlated (16 pairwise comparisons, 11 sites, three birds). A bootstrap procedure showed that the peak correlation was significant (P < 0.02) in 15/16 comparisons, with HVc leading RA by 2 to 14 ms in 12/15 cases (two birds). A third bird contributed three cases; for all three, RA led HVc (by 1, 28, and 44 ms). Correllelograms were much broader for these three paired recordings than observed for any of the recordings in which HVc led RA.
13. Firing rates were estimated from peristimulus time histograms (PSTHs) with 50-ms bins. The strength of response was quantified from the PSTH as the ratio of the variance in firing rate during stimulus presentation over the variance in ongoing activity beginning 1 s after the stimulus offset. This measure is sensitive to slight stimulus-driven changes in temporal patterns of spike trains. Nevertheless, 33 recordings sites (28 single units, 15 multiunits) from four adult awake, freely moving birds gave a response strength for BOS of 1.00 ± 0.40 (1 = no response). In contrast, 72 recording sites (71 single units, 33 multiunits) from 12 urethane-anesthetized birds gave a response strength for BOS of 18.71 ± 33.81. This difference was significant (Mann-Whitney U, Z = 9.004, P < 0.0001). Twenty-two recordings sites (14 single units, 15 multiunits) from five awake but restrained birds gave a response strength for BOS of 2.77 ± 2.62. In only one awake-restrained bird were most sites auditory (six single units, six multiunits, seven sites). One of the animals with chronic headgear and two of the awake-restrained animals were subsequently anesthetized, resulting in average increases in response strength of 111, 1860, and 634%.
14. In awake animals, the distribution of interspike intervals (ISIs) of ongoing activity of RA single units had a single mode corresponding to the mean of the regular pattern of firing. In anesthetized animals, ISI distributions typically had two modes, a long-interval (major) mode reflecting the regular firing pattern and a short-interval (minor) mode reflecting the complex bursts. The major mode was fit with a Gaussian, and then the Gaussian was subtracted from the data. The number of residual intervals from 0 to the Gaussian mean 2 SD, as a percentage of the total number of intervals, was taken as the index of bursting. With these procedures, 29 recording sites (33 single units) from five awake, freely moving animals had average ISIs of 30.7 ± 7.4 ms (32.6 Hz) for the major mode. The irregularity of firing for the major mode was 5.6 ± 4.9 ms (determined from the distribution of SD from the Gaussians). In contrast, data from 60 recording sites (68 single units) from 12 urethane-anesthetized birds resulted in an average ISI of 98.9 ± 32.6 ms (10.1 Hz) for the major mode. The irregularity of firing for the major mode was 20.9 ± 17.2 ms. These differences are significant (average ISI: Z = 8.095, P < 0.0001; irregularity: Z = 6.774, P < 0.0001). The index of bursting was 7.5 ± 8.3% for data from awake animals and significantly higher at 12.8 ± 7.4% from urethane-anesthetized birds (Z = 3.724, P < 0.0003).
15. H. Sakaguchi and N. Saito, Brain Res. Dev. Brain Res. 47, 313 (1989) [Medline] ; G. F. Ball, in Hormones, Brain and Behavior in Vertebrates, J. Balthazart, Ed. (Karger, Basel, Switzerland, 1990), pp. 148-167; R. Li and H. Sakaguchi, Brain Res. 763, 239 (1997) [CrossRef] [Web of Science] [Medline] .
16. Two birds received two left and one right hemisphere injections of NE into RA, with 27 single and multiple units recorded in RA before injection and 17 after injection. Using the measure of response strength for single units (13) and a criterion of response strength > 3, we found that before and after injections of NE into RA, there were 23/27 and 16/17 BOS responsive units, respectively, recorded in RA. The average response strength was 19.7 ± 34.8 before and 16.4 ± 16.0 after injections; the difference was not significant (Z = 0.06, P = 0.9519). In contrast, in three birds injected with NE into HVc, we recorded 27 units in RA before and 30 units after injection. There were significant changes in the number of responsive units and the magnitude of the responses. Before and after injections of NE into HVc, there were 21/27 and 5/30 BOS responsive units, respectively, recorded in RA. The average response strength was 8.5 ± 6.6 before and 2.0 ± 1.6 after injections; the difference was significant (Z = 5.090, P < 0.001).
17. For NE injections into RA, the mean of the major mode of the ISI distributions for RA neurons decreased significantly from 98.7 ± 34.7 ms before injections to 63.1 ± 22.0 ms after injections (Z = 3.842, P < 0.0002). The SD of the major mode also decreased significantly from 16.2 ± 16.8 to 3.6 ± 2.3 ms (Z = 6.308, P < .0001), indicating an increase in the regularity of firing. The burstiness changed from 17.1 ± 8.9 to 14.0 ± 5.2%; this change was not significant (Z = 1.146, P = 0.2516). For NE injections into HVc, all three measures of RA ongoing activity changed significantly (P < 0.05) (mean ISI: 123.0 ± 26.9 to 102.4 ± 24.0 ms; irregularity: 29.0 ± 20.9 to 16.3 ± 15.6 ms; bursting: 14.3 ± 7.5 to 9.5 ± 5.7%).
18. In one bird, vehicle (artificial cerebral spinal fluid stained with dextran-rhodamine) was injected into HVc. All RA sites before (N = 7) and after (N = 8) the injection were auditory. Postmortem histology indicated that the injection filled but was restricted to HVc. Three birds received large injections of NE targeted toward RA but that probably involved large parts of the caudal forebrain (see text). On the basis of the criterion of response strength > 3 (16), in the three birds, 28/35 and 8/26 RA units were auditory before and after injections, respectively.
19. D. J. Perkel, Soc. Neurosci. Abstr. 20, 165 (1994) .
20. A. M. Liberman, F. S. Cooper, D. P. Shankweiler, M. Studdert-Kennedy, Psychol. Rev. 74, 431 (1967) [CrossRef] [Web of Science] [Medline] ; A. M. Liberman and I. G. Mattingly, Cognition 21, 1 (1985) [CrossRef] [Web of Science] [Medline] .
21. H. Williams and F. Nottebohm, Science 229, 279 (1985) [Abstract/Free Full Text] ; H. Williams, Behav. Neural Biol. 44, 470 (1985) [CrossRef] [Web of Science] [Medline] .
22. K. W. Nordeen and E. J. Nordeen, Behav. Neural Biol. 57, 58 (1992) [CrossRef] [Web of Science] [Medline] .
23. M. L. Sutter and D. Margoliash, J. Neurophysiol. 72, 2105 (1994) [Abstract/Free Full Text] .
24. M. A. Wilson and B. L. McNaughton, Science 265, 676 (1994) [Abstract/Free Full Text] ; W. E. Skaggs and B. L. McNaughton, ibid. 271, 1870 (1996) [Abstract].
25. A. C. Yu and D. Margoliash, ibid. 273, 1871 (1996).
26. J. S. McCasland, J. Neurosci. 7, 23 (1987) [Abstract] ; A. C. Yu and D. Margoliash, unpublished data.
27. H. Sakaguchi and N. Saito, Brain Res. Dev. Brain Res. 62, 223 (1991) [Medline] ; J. A. Soha, T. Shimizu, A. J. Doupe, J. Neurobiol. 29, 473 (1996) [CrossRef] [Web of Science] [Medline] ; C. F. Harding, S. R. Barclay, S. A. Waterman, ibid. 34, 329 (1998) [CrossRef] [Web of Science] [Medline].
28. D. Margoliash, J. Neurobiol. 33, 671 (1997) [CrossRef] [Web of Science] [Medline] .
29. All procedures were approved by an institutional animal care and use committee. Supported by grants from the Whitehall Foundation, NIH (NS25677), and the Brain Research Foundation. A.S.D. and A.C.Y. were supported by NIH predoctoral fellowships.