EFTA01076046

The Journal of lietrosoence. F &nary 17. 2010 • 30(71:2783-2791 • 2783 BehaviorallSystems/Cognitive Neuronal Stability and Drift across Periods of Sleep: Premotor Activity Patterns in a Vocal Control Nucleus of Adult Zebra Finches Peter L Rauske,I Zhiyi Chi? Amish S. Dave,' and Daniel Margoliashi Departments of 'Organismal Biology and Anatomy and 2Statistics, University of Chicago, Chicago, Illinois 60611 How stable are neural activity patterns compared across periods of sleep? We evaluated this question in adult zebra finches, whose premotor neurons in the nucleus robustus arcopallialis (RA) exhibit sequences of bursts during daytime singing that are characterized by precise timing relative to song syllables. Each burst has a highly regulated pattern of spikes. We assessed these spike patterns in singing that occurred before and after periods of sleep. For about half of the neurons, one or more premotor bursts had changed after sleep, an average of 20% of all bursts across all RA neurons. After sleep, modified bursts were characterized by a discrete, albeit modest, loss of spikes with compensatory increases in spike intervals, but not changes in timing relative to the syllable. Changes in burst structure followed both interrupted bouts of sleep (1.5-3 h) and full nights of sleep, implicating sleep and not circadian cycle as mediating these effects. Changes in burst structure were also observed during the day, but far less frequently. In cases where multiple bursts in the sequence changed in a single cell, the sequence position of those bursts tended to cluster together. Bursts that did not show discrete changes in structure also showed changes in spike counts, but not biased toward losses. We hypothesize that changes in burst patterns during sleep represent active sculpting of the RA network, supporting auditory feedback-mediated song maintenance. Introduction court females with "directed" singing: precisely structured, regu- Sleep-dependent behavioral plasticity has been observed in a lar songs comprising introductory notes followed by a sequence broad range of perceptual, motor, and higher-level cognitive of syllables organized into a "motif." Directed songs are even tasks in studies in adult humans (Kami et al., 1994; Stickgold et more highly regulated than the undirected songs males otherwise al., 2000; Fischer et al., 2002; Walker et al., 2002; Fenn et al., 2003; sing (Sossinka and Bohner, 1980; Kao et al., 2005; Glaze and Wagner et al., 2004; Brawn et al., 2008). Electrophysiological Troyer, 2006). studies support a role for active processes during sleep affecting Associated with directed singing are highly structured bursts memory consolidation in humans (Maquet et al., 2000; Peigneux of activity in presumptive projection neurons in the nucleus ro- et al., 2004; Reis et al., 2009), and behavioral and electrophysio- bustus arcopallialis (RA) (Yu and Margoliash, 1996). Each spike logical studies in animals implicate sleep in plastic mechanisms. burst has submillisecond precision in its timing relative to its Sleep modulates plastic changes in ocular dominance histograms corresponding syllable within the motif (Chi and Margoliash, in the developing visual cortex of young cats (Frank et al., 2001; 2001; Leonardo and Fee, 2005). These bursts have a well-defined Aton et al., 2009), the emergence of song system neuronal burst- number of spikes in a well-defined temporal pattern, both of ing in juvenile birds at the onset of song learning (Shank and which vary across bursts emitted at different times in the song. A Margoliash, 2009), and experience-dependent changes in the given burst thus has a specific identity associated with onset time, correlations of activity patterns of rat hippocampal neurons (Poe number of spikes, and pattern of spikes. RA neurons show highly et al., 2000). regulated oscillatory spontaneous activity, become completely These results emphasize changes measured in populations of suppressed about 50 ms before onset of song, and may achieve neurons. Sleep-dependent changes in the individual activity pat- instantaneous firing rates of almost 800 Hz during singing. Thus, terns of single neurons during behavior are not well defined, the nervous system expresses almost the entire dynamic range however, and thus there is little data on the stability of single available to precisely modulate the activity of single RA neurons neuron activity patterns across periods of sleep. In this study, we during singing. address this issue in the birdsong system. Male zebra finches We took advantage of the reliability and precision of this sys- tem to examine neuronal stability over extended periods of time. Ikuwed June 5,2009: tedsed Dec. 13.2609; 5,00165 150.11,2010. RA extracellular recordings can be stable with high signal-to- lkswalmassappaled in part to/ KalIcnai Ir60cats °Math GT/Ell M1159831 ten Mast;awful loPaa noise ratio (SNR), but the technical challenge of maintaining Arnida.Penry D.I.gattaret end Sitpilin D. Shea 55tongalitittv.5 ct thls mint600n. high-quality recordings over the required durations and behav- (cert:pc00fficesti0510 te addressed to DI. Wet L.Pathlo.Ser6o0MccartifoimaxeRcgrankkitateatio0 Inuct6e of 010460 3451. Swerlx Sued, SW< 1476.(hog0116051S brt01. iors in freely moving animals required by this design limited the 1)01.1015230/151.0050 3112402010 size of the dataset. Nevertheless, we were able to directly compare Com6ght 02010 the authors D2706174/10/10/761-12515.000) premotor activity of the same single neurons before and after EFTA01076046 2781 • .I. ileurosci., Febuary 17,2010 • 30I71:1783-2791 Rauske et al. • Xeuronal Stability and Mt across Sleep periods ofsleep. To the best ofour knowledge, such comparisons recordings until single-unit isolation was lost. Auditory stimulation en- have not been reported in any premotor system. abled us to verify the responsiveness to the bird's own song that RA neurons exhibit exclusively during sleep (Dave et al., 1998). Further- Materials and Methods more, this was a preliminary experiment to test the hypothesis that sleep- To examine the effects of sleep on the stability of premotor burst patterns related changes in singing behavior result from drift arising from neural in RA neurons, we recorded neuronal activity in three types of experi- replay during sleep activity without concomitant auditory feedback mental sessions: short-sleep (or interrupted-sleep) sessions, long-sleep (Deregnaucourt et al., 2005). We hypothesized that playback would pro- (or normal circadian-sleep) sessions, and awake-only sessions. For both vide structured activity during sleep, possibly preventing sleep-related types of sessions including sleep, we recorded the activity of the same changes, but failed to see systematic differences between short-sleep (au- single RA neurons whilebirdssang or produced learned calls (sec below) ditory stimulation) and long-sleep (no stimulation) sessions (see Re- both before and after the period of sleep. We developed algorithms to sults), a null result with respect to the sleep-drift hypothesis. We do not identify changes to burst patterns across periods of sleep, as well as sta- consider this hypothesis further in this study. tistical techniques to compare the frequency of such changes with that In some additional, exceptional cases = 5 neurons, 3 birds), we observed in the absence of sleep. successfully gambled on our ability to maintain stable unit isolation Eleetrophysiology and design of the experiments. All animal procedures across a full night of sleep (8 or 10 h), maintaining the normal light/dark were approved by an Institutional Animal Care and Usc Committee. cycle. All but one of these cases involved single-unit isolation, with the Adult male zebra finches (n = 13) were habituated to either a 16/8 h or exception being a site in which a pair of units could be reliably distin- 14/10 h light/dark cycle. We found no systematic differences between the guished from background activity but not from each other; this "double- two conditions, and combine the data for aggregate statistical analyses. unit" site was treated similarly to single units in our analysis. No auditory The birds were implanted with microdrives with electrodes targeting PA; stimuli were presented during sleep for these sites, but ongoing activity the implant design and surgical procedures have been described in detail was sampled throughout the night to verify the presence of bunting previously (Dave et al., 1999). Briefly, a recording device carrying four activity in RA that indicated the bird remained asleep. When the next glass-coated Pt-Ir electrodes (impedance, 1.2-2.0 MS/ at 1 kHz) was day's light cycle began, recordings continued until unit isolation was lost. implanted under modified Equithesin anesthesia over RA. During re- Finally, we augmented this data set with additional recordings during cordingsessions starting 2-4 d later, a flexible cable connected the head- vocalizations in recording sessions that did not indude sleep (see below). gear to an overhead commutator to allow the bird free movement within Analysis of deep. Sleep was objectively defined behaviorally (eye clo- the cage. Differential recordings were used to minimize movement arti- sure, body posture), and we also developed a quantitative measure of facts. Recording sites were obtained by audiovisual monitoring of the spontaneous bursting in RA neurons to use as an assay for sleep. We first recordings while using a drive screw to manually advance the electrodes. established a baseline for a neuron's spontaneous spiking activity during Birds were manually restrained during this procedure, then carefully periods before and after darkness when the bird was awake and active, released into the cage while trying to maintain unit isolation. but not vocalizing. We used 1-4 min segments of neuronal activity both Recording sessions began at various times during the day, and we before and after darkness, dividing the spiking activity into 3 s segments. recorded only sites with at least one unit that could be well isolated. In all For each segment, the distribution of interspike intervals (1S1s) was ap- cases, a conspecific female was introduced into an adjacent half-cage to proximately Gaussian because of the highly regular spiking activity of RA elicit directed singing and calling. [In male zebra finches, contact or neurons in awake, nonvocalizing birds. We cakulated for each segment's so-called "long" calls are learned vocalizations whose production in- ISI distribution the mean (IS1-MEAN) and standard deviation (1SI-SD). volves PA premotor activity (Zann, 1985; Simpson and Vicario, 1990), The resulting range of values across all awake segments for each single and they are treated equivalently with song syllables in this study.' unit provided an estimate of the baseline variability in spiking activity in After collecting high SNR spike data during vocalizations comprising the awake bird. at least 10 song motifs and/or contact calls, or in the normal circadian To quantify the amount of sleep during darkness, we similarly divided rhythm depending on experimental design (see below), the cage lights spiking activity into 3 s segments, calculating the ISI-MEAN and IS1-SD were doused. After the bird was quiescent for several minutes, activity in for each segment. Any segment whose ISI-MEAN and 151-517 both fell RA entered a characteristic bursting mode. This distinct state was never within a 95% confidence interval as determined by the baseline awake observed in an awake bird, and bursting disappeared whenever the bird distributions was labeled "awake"; all other segments were labeled as was disturbed or became active. Spontaneous bursting in RA and its `sleep" (for exampk,see supplemental Fig. I, available at unvw.jneurosci.org efferent sensorimotor control nucleus (HVC) has come to be used as an as supplemental material). Such labeling agreed well with visual inspec- assay for sleep. It is reliably associated with the onset of sleep postures and tion of spiking activity, with segments including sleep-typical depressed strong, selective auditory responses (Dave et al., 1998; Dave and Margo- firing rates and/or bunting reliably labeled as sleep. Video surveillance Hash, 2000; Nick and Konishi, 2001; Hahnloser ct al., 2002, 2006; Cardin under infrared illumination verified that the bird was quiescent with and Schmidt, 2003; Rauske a al., 2003; Shank and Margoliash, 2009) closed eyelids in >95% of sleep-labeled segments. We had not developed and has been correlated with EEC measures of sleep (Nick and Kon- reliable EEG recording techniques and an understanding of sleep staging ishi, 2001; Hahnloser et al., 2006; Shank and Margoliash, 2009). Dur- in zebra finches except toward the end of these studies (Low et al., 2008); ing recording sessions including 1.5-3 h darkness (labeled "short sleep"; nevertheless, our analysis reliably distinguished sleep from wakang. n = 10 neurons, 4 birds), we recorded continuously from the isolated RA Song syllables, spike bursts, and a definition of bunt opts. Vocalizations single units, enabling us to estimate the amount of time birds actually and onset and offset times for each syllable were identified by manual slept by examining the bursting activity (or lack thereof) during the dark inspection of spectrographs. A syllable was defined as a stereotyped vocal period. We used a quantitative measure of spontaneous PA bursting as a gesture containing no silent interval 710 ms: in addition to the tradition- sleep assay, described below. During some recording sessions, we also ally defined song syllables that comprise song "motifs" (stereotyped se- verified by direct observation (infrared monitoring) that the bird's eyes quence of syllables), wealso included introductory notes at the beginning were closed and respiration slowed when PA activity indicated sleep of singing bouts and isolated "long" calls, both of which recruit RA (Dave et al., 1998). bunting activity, in our definition of "syllable" for this study. Syllable During the short-sleep recording sessions, we also presented playback onset times and spike times were merged for each site to create a raster of the bird's own song. Recordings of the bird's own song were scaled to plot of spiking activity associated with each syllable type. For each sylla- 70 dB root-mean-squared amplitude and presented randomly at 10-30 s ble, we included the spiking activity beginning 50 ms before syllable onset intervals beginning immediately after turning out the lights. After 50- and ending with the syllable offset. 250 repetitions of song playback, we recorded 20-60 min of ongoing We used simple thresholding techniques to identify spike times for spiking activity while the bird remained asleep. Thereafter, the lights most, extremely well-isolated single units. For a few sites with more were then turned back on, rousing the bird, after 1.5-3 h of sleep. Birds ambiguous isolation, we used theSpiicesort program, which uses a Bayes- then directed singing toward the adjacent female, and we continued ian approach to identify putative spikes with distinct spike-shape models EFTA01076047 Rauske et al. • Neuronal Stability and Drift across Sleep 1. Neurovi, February 17,2010 .30(71:2783-2794. 278S (Lewicki, 1994). To confirm single-unit isolation in all cases, we visually pain, while preserving each individual burst's interspilce intervals. inspected overbid waveforms from all identified spike times to confirm Pre-sleep and postsleep burst stacks were then aligned with each other that spike shapes were consistent throughout our recordings, and we according to similar procedures, with all of the bunts in each stack used ISI distributions to confirm the hallmarks of single-unit isolation in shifted as a whole so that the relative timing within each stack was RA (i.e., an approximately Gaussian distribution of ISIs during behav- preserved. ioral quiescence and a lack ofISIs <1ms). For the majority ofsites (28 of The L, metric used in the L,-MIN method measures the difference 42 single units), we were able to confidently identify 100% of all spikes between two spike sequences obtained by averaging over all spikes the after manual inspection. The remaining single-unit sites, as well as the temporal difference between each spike and its closest corresponding "double-unit" site, included a small number ofambiguous spikes, so we spike in the other sequence, so that optimal alignment would be achieved estimate that we achieved 98-99% correct classification. In these cases, by minimizing this measure. To generate a cross-correlation measure for the ambiguities were attributable to either the extreme attenuation of the CC-MAX method, we used the biweight kernel F(x) = (I — (W0)2] 2 spike amplitude during bursting (Yu and Margoliash, 19%) or sporadic for all < D, where D is a time window corresponding to the temporal background spikingactivity that could not be reliably distinguished from precision of the cross-correlation measure (set to 1.5 ms, a value chosen attenuated spikes in the recordings with the lowest SNR. These sites, to approximate the apparent temporal precision of RA premotor spike however, did not show any greater or lesser stability of temporal patterns patterns).The total CC ofspikc trains S,,..., Sk was defined as!,,,,K(S„ ofspike bursts—the principal dependent variable of this study—than did S,), where K(S„ = !Rs — t) over s in S, and tin The alignment those sites with completely reliable spike identification. maximized the total CC by shifting each spike train S, while preserving RA activity during singing is characterized as having high-frequency each individual bunt's interspilce intervals. bunts of spikes organized into trains of bursts. Each burst in the train of Once bursts within a stack were aligned, fine temporal structure was bunts is distinguished from the others both by the pattern of spikes and expressed as the tightly aligned spikes across renditions. A "feature" the timingof the bunt relative to the syllable (Yu and Margoliash, 1996; within a bunt was defined as a canonical spike, i.e.,a spike produced with Dave and Margoliash, 2000; Leonardo and Fee, 2005). In this study, we reliable timing relative to the other spikes in the bunt across many or all defined a bunt as a sequence of consecutive spikes with all interspike renditions. To identify and quantify features, we defined for each group intervals <10 ms. This simple definition reliably identified all bunts of ofspike trains an adjusted rate function,R(t) = — t11D), where two or more spikes ofan RA neuron during singing. In all cases, we also s is the time of an individual spike within the spike train S, and G(x) = could readily identify a canonical sequence ofbursts for each syllable (Yu (1 — x2) for all Ix] < D and 0 for all Ix] > D with the predefined time and Margoliash, 19%). Aligning multiple renditions of the sequences of window D = 1.2 ms. (Note that D = 1.2 ms results in a more precise bunts relative to the onset of a given syllable (as in a raster plot) created firing rate estimate than the 1.5 ms time window used for the original stacks ofbursts, with each "burst stack" associated with a particular time bunt alignment, achieving a coarse-to-fine alignment procedure.) We relative to syllable onset and a particular temporal pattern of spikes. We then identified peaks in the rate function. This method captures the identified 2.1 ± 1.3 bursts for each syllable across all the neurons, with changes in features we visually observed but is sensitive to the definition some syllables not eliciting any bunts and one neuron reliably emitting ofpeaks in the rate function, for example, slight changes in the temporal eight bunts for a particularly long and complex syllable. jitter of a given spike. The principal data set consisted of 115 distinct burst stacks emitted For a sample of N presleep spike trains, time T was identified as a during singing both before and after sleep by 15 RA neurons (seven feature location ifit satisfied four criteria: (1) the averaged adjusted rate birds). To compare the stability of temporal structure in premotor activ- function had a local peak at time Tli.e.,r(7) Z r(s) for s between T ± D, ity in the absence of sleep, we also examined the activity of RA neurons wherer( 7) = mean(R( Mover the sample]; (2) the peak at time Twas of recorded in periods of singing and/or calling that did not include sleep. significantly high amplitude compared with the variability of the rate We included in this data set the same 15 neurons used in the sleep function, Ir(T) 2 0.3 + , (0.975) X o (7), where a ( = SD(R( 7)) analysis, separating out the pre-sleep activity and postsleep activity into over the sample, and t,,,_, the inverse s-distribution function with N — I distinct sessions, each ofwhich did not include sleep (i.e., 115 bunt stacks from presleep recordings, and 115 burst stacks from postsleep record- degrees of freedom]; (3) the variability of spike times within the pre- ings, for a total of 230 burst stacks). To expand our data set to include defined time window around T was sufficiently low, IsN_ ,(0.975) X a sessions oflonger duration without sleep, we included the additional 28 ( 7)5 D, where a (7) = SD (spike times between T ± 0)1; and (4) the PA neurons recorded from 10 birds (six new, four that were also repre- average value of the adjusted rate function on either side of the peak fell sented in our sleep-inclusive recordings) in experiments where the lights off sufficiently quickly such that If] 5 2 ms, where 1equals the maximal were not turned out and the birds remained awake throughout, yielding interval containing T over which r(s) 2 r(T)13. Under these criteria, an additional 321 burst stacks. Thus, this "augmented" data set com- —65% of all spikes in premotor bunts were identified with located fea- prised a total of 551 distinct burst stacks recorded from 13 birds. tures (4.8 ± 3.1 total spikes/burst; 3.1 ± 1.8 features/burst). During one awake-only session, we also briefly recorded one putative We judged each burst stack as having a `structural change" across the RA interneuron characterized by a low baseline firing rate and an espe- sleep interval if three criteria were met. Pint, features in the presleep and cially narrow spike width (0.13 ms peak to trough,compared to a range of postsleep adjusted rate functions did not align well. Each feature was 0.19-0.41 ms for all other RA neurons we recorded), but we did not evaluated to determine whether we could rule out the existence of a include this unit in our analyses because of insufficient spike isolation corresponding spike in the corresponding stack (i.e., presleep vs during singing. postsleep). If for any feature there was no corresponding feature in the Analysis of burst structure and definition of features and structural corresponding stack within 0.25 ms, and there was no other peak within changes. To evaluate changes to the temporal structure of premotor 0.5 ms in the opposite stack's rate function with a magnitude statistically bunts across many renditions, we (I) developed a procedure to align all indistinguishable from that of the feature being evaluated, then the burst presleep or postsleep bunts for a given bunt stack, (2) generated func- stack was judged to meet this criterion. Second, there was a statistically tions that captured the temporal features of the aligned bursts, and (3) significant change in mean spike count of at least 0.5 spikes/burst. This evaluated the significance of any temporal or spike count differences criterion arises from the observed loss (or, rarely, gain) in spikes across between presleep and postsleep groups of spikes. sleep intervals (see Results). Third, to reduce the effect of artifacts in To optimally align bunt renditions within a presleep or postsleep alignment, structural changes were flagged only when the first two crite- bunt stack, we used two procedures: / 1-distance minimization (l.,- ria were satisfied under both L,-MIN and CC-MAX alignment proce- MIN), as described by Chi and Margoliash (2001), and cross-correlation dures. Overall, under both alignment procedures, 37 burst stacks maximization (CC-MAX). In both cases, the alignment of spike se- satisfied the first criterion, and 60 satisfied the second, with 33 satisfying quences was accomplished by iteratively shifting each burst rendition to both. Thus, changes in spike timing typically were associated with either globally minimize the summed L, distances (L,-MIN) or maxi- changes in spike rate, but the reverse was not generally the case. Those mize summed cross-correlation measures (CC-MAX) across all bunt bunt stacks found to undergo structural changes under these criteria EFTA01076048 2786 • 1. Neurosci., February 17,2010 • 3017):2783-2794 Rauske et al. • Neuronal Stability and Drift across Sleep corresponded well to those burst stacks that appeared to have altered pre-sleep . post-sleep spiking patterns under visual inspection. A Analysis for separator intern& other than sleep. Sleep is a natural sepa- !HUHU rator between groups of vocalizations, but we also explored whether . • II•01. • . . changes to premotor activity occurred at times other than sleep. To this ep ON' iN • ON end, for each burst stack we sought to identify the interval between con- OD MO • • II • •• PIMIC • • • -.1k I • secutive renditions of bunts that was most likely to correspond to a Man • • • • .. •• • .. .. ay. . so , . . . • . loss change in burst structure, referring to the interval thus identified as the • •• •• • • ,•. • • . similar `separator interval." We began by measuring the similarity ofall possible • • x • ••. • .. ... • . post-sleep • r 1 pairings ofindividual bursts within each burst stack, using theL, distance • • • • II.1 more metric described above; greater L, distance implies less similarity. Then, • • • I.' similar we considered each interval between bursts as a candidate separator in- terval, except that we excluded the first four and last four such intervals to avoid boundary effects. For each candidate interval, we divided the bursts 0.9 across-group into preinterval and postinterval groups, and from the collection of can- didate intervals we identified the one interval that maximized the differ- mean LI ence between the mean L, distances of across-group comparisons distance (preintenel vs postinterval) and within-group comparisons (preinterval 0.5 ten roux vs preinterval, or postinterval vs postinterval). This procedure tended to 10 30 identify two groups ofmost-similar bursts,one exclusivelybefore and the interval other exclusively after the interval, dividing the bunt stack at a moment time in time that often corresponded to a visibly noticeable change in burst structure (Fig. IA). B 15 We also used a modified procedure better suited to quantify a subset of the transitions in bunt structure. For these cases there was a distinct transition between distinct states, but with one of the states exhibiting less variability than the other. In these cases, the candidate interval that % of maximized the difference between mean L, distances did not always bursts correspond to the visually observed transition. We found that for these cases, the transition typically coincided with a candidate interval that maximized the differences comparingL, variances for across-group and postinterval (or preinterval) candidate intervals as well as maximizing the differences between L, means for across-group and preinterval (or postinterval) candidate intervals. Therefine, in these cases, we designated the interval thus defined as the separator interval; in all other cases, we simply used the interval maximizing the L,-mean distances between separator interval/ across-group and within-group comparisons as the separator interval. total # burst renditions Estimating occurrences of sleep-separator comparisons attributable to chance. Finally, we also developed a statistical procedure to compare the Figure 1. Identifying intervals with possible changes to premotor burst patterns. A, Corn- location of separator intervals in recordings that did and did not include parisons of pairs of burst renditions for a burst stack recorded before and after a period of sleep, sleep. To this end, we first identified separator intervals for the awake- using the L, distance metric. Each row and column represents a single burst rendition Dem. only recording sessions. Then, for each bunt stack, we calculated the seated by rasters to the left of rows a above columns). Each colored box represents Mel, proportion of bursts occurring before the separator to the total number &stance between the bursts denoted by the given row and column according to a color map ofbursts. A histogram of the resulting distribution suggested a quadratic with red indicating the highest distances (less similarity) and blue representing the lowest L, distribution, so we used a quadratic fit to generate a baseline probability &stances (more similarity). The sleep Interval is denoted by black dashed Ines. Note that burst density function (PDF) (Fig. I B). comparisons above and to the left of the sleep.interval Imes (I.e., comparisons between The PDF estimate allowed us to test the hypothesis that the distribu- preskep and postsleep bursts) show less similarity than 63 comparisom between bursts taking tion ofL,-optimized separator intervals was the same in awake-only and place exclusively before or after sleep. The graph at the bottom right shows the mean L, dis- sleep-inclusive recordings. In a bootstrap procedure, we sampled 115 tances between all pairs of burst renditions taking place across the separator interval(red line) fractions from the PDF and respectively multiplied these by the total &exclusively before& after the separator interval (blue line) for all possibk intervals. The sleep number ofrenditions for each of the 115 burst stacks in our data set to get intervalis &noted by the dashedline, where thedifferenceinmeanLi &stance between these a random separator interval. We repeated this procedure 10,000 times to twogroupsreaches a <leas peak; such a peak defines the optimized separator interval. 8, Esti- obtain a distribution of simulated separator intervals. This distribution mated probability distreutem function for the location of the optimized separator interval in was used to evaluate the likelihood that the number ofseparator intervals recording sessions that did not include sleep (551 burst sucks). the histogram shows the dia. we observed to correspond with the period of sleep (either exactly or tribution ofopeimized separator intervals relative to the total number &burst renditions within within one interval) would occur simply by chance. each recording session. A quadratic fit (line) was used to determine thePDF. The quadratic shape of the PDF can be explained as follows. The greater likelihood of locating separator intervals near the endpoint of an ings was maintained through a period of sleep and subsequent experiment rather than in the middle is most likely attributable to the vocalizations (Table I) (see Materials and Methods). It is likely exaggerated effect ofoutliers on small groups ofbunts. Theasymmetrical that all of these cells were projection neurons targeting the brain- shape of the PDF (see Results) may reflect a slightly increased variability stem, given their fast (>30 Hz), regular baseline spiking activity across bunt renditions later in experiments, when more time sometimes passed between singing bouts as birds became desensitized to the pres- and bursting activity during singing (Spiro et al., 1999; Leonardo ence of the adjacent female and tended to sing less frequently. and Fee, 2005). Each cell reliably burst with consistent timing relative to specific vocalizations such as a particular syllable or Results call; thus, raster plots of the neuronal activity aligned to vocaliza- We recorded from 43 RA single units while birds vocalized (sang tion onsets produced "stacks" of bursts, which were the basis for or called; including one "double unit" that we treat equivalently our analysis (see Materials and Methods). In the 37 cells for which to the single units in our analyses). Only a subset of these record- we recorded singing, there were 10.1 ± 4.2 unique burst stacks EFTA01076049 Rauste et al. • Neuronal Stability and Drift across keep J. Reurovi, February 17, 2010.30(71:2783-2794.2787 Table 1. Distribution of recordings across different experimental conditions cases. The changes persisted for as long as we could hold the Recording session type Birds Neurons Burst stacks recording—in one case, for several hours after waking (Fig. 3). Whatever the cellular or network effects that led to these changes, Sleep inclusive Tall) 7 15 115 Sleep inclusive (shod) 4 lob 83 they achieved their suprathreshold effects during sleep or imme- Sleep inclusive (long) 3 5 32 diately after awakening, and persisted thereafter. No sleep° 13 43° 551 To quantitatively assess these changes in burst structure, we 'Includesboth the pleserpenbt and pssuletpenls pleas el tit serpent:Mitt sessions as &Una no-slett aligned all presleep and postsleep bursts for each of the 115 burst moons. stacks using two algorithms—Le-distance minimization and 'Includes cee"dothkunt °turned as a snub nese:con outatubsts. cross-correlation maximization—converted these into probabi- listic rate functions, located features within those functions, and identified reliable changes in features, which we call "structural per song, and 1.9 -± 1.1 unique burst stacks per call. For the changes" (see Materials and Methods). Using these criteria, we remaining six cells for which we only recorded calls, there were found that for recordings spanning a sleep interval, 33 of 115 one to two unique burst stacks per call. burst stacks showed structural changes (Fig. 4). These 33 burst We examined the effects of sleep on vocalization-related neu- stacks were distributed across 10 of 15 neurons recorded across ral activity in adult male zebra finches under two protocols: short, sleep, with each neuron exhibiting one (four neurons), two (three interrupted periods of sleep and full, uninterrupted nights of neurons), or six (two neurons) bursts with structural changes, sleep. The distribution of recording sessions according to exper- but with one neuron exhibiting 11 bursts with structural changes. imental protocol is reported in Table I. Birds in the short-sleep The statistical significance of all the results that follow was main- design (at = 10 neurons in 4 birds) experienced a period of dark- tained even with the neuron with 11 bursts with structural ness lasting 90-179 min (average, 136 -± 31 min), whereas the changes removed. birds in the second design (n = 5 neurons in 3 birds) experienced Considering the sequence of syllables within motifs or the a full 8 -10 h of darkness. For all birds, during the first 2-10 min sequence of bursts within a syllable, there was no apparent ten- of darkness, birds typically rapidly transitioned between short dency for structural changes to be associated with bursts that periods of wake and sleep. Initially, in some cases, sudden awak- occurred in any particular syllable within the motif, or within any enings apparently resulted from playback of the bird's own song particular burst within a syllable. There was also no clear differ- that was presented during short-sleep sessions (see Materials and ence in the number of changes in burst stacks associated with Methods), but birds quickly habituated to the song playback and contact calls (5 of 21; 24%) compared to those associated with began to reliably sleep through the stimulus. Birds were judged to song syllables (28 of 94; 30%; p = 0.79, Fisher's exact test). begin an extended period of sleep when a full minute passed with We tested the hypothesis that the two different experimental no two consecutive 3 s intervals classified as "awake" (see Mate- designs affected the rate of occurrence of structural changes. rials and Methods) (see supplemental Fig. 1, available at www. There were no significant differences in the frequency of burst jneurosci.org as supplemental material). Based on the measure of changes between short-sleep and long-sleep birds. The frequen- spontaneous activity, the onset of extended periods of sleep began cies of structural changes [27% (22 of 83) short sleep, 34% (1 I of 10.5 -± 8.8 min (range, 1.3-29.5 min) after the start of subjective night, and represented 78.8 -± 10.5% (range 67.8 —94.9%) of the 32) long sleep; p = 0.40; x2 = 0.701 were similar under both total dark phase after sleep onset. Thus, whereas the recording experimental conditions, suggesting that the truncated period of situation probably disrupted the animal's total sleep and sleep sleep and the presence of auditory stimulation were not signifi- architecture to some degree, each animal experienced a consid- cant factors in driving premotor plasticity in RA neurons. erable amount of sleep including extended periods of uninter- Although rare, there were also examples of structural changes rupted sleep. that occurred during the subjective day. For this analysis, we used an augmented data set including a number of daytime-only re- Premotor bursts change after sleep cordings (see Materials and Methods). To quantitatively assess The structure of RA premotor bursts is highly conserved in songs the rate of structural changes while the birds were awake, we directed toward females, as a bird repeats the same stereotyped simply chose the longest vocalization-free interval with a suffi- syllable within and across songs (Yu and Margoliash, 1996). We cient number of vocalizations (ri a 8) both preceding and follow- frequently observed, however, changes to the structure of these ing the interval, and compared the bursts before and after this bursts after sleep. The only systematic change across time de- interval. Across 551 distinct burst stacks in the augmented data scribed previously in RA premotor bursts is the submillisecond set (43 neurons, 13 birds; see Materials and Methods) (see Table magnitude temporal drift in the timing between bursts (Chi and 1 ), the rate of structural changes that occurred in recordings that Margoliash, 2001). In contrast, in the present data, many clear did not span a sleep interval was much lower (3.3%; 18 of 551) and persistent changes in premotor patterns associated with the than the rate of changes across the sleep interval (28.7%; 33 of sleep interval were apparent by visual inspection beginning with 115), and this difference was significant ( p < 0.001; 1(2 = 87.0). the first song renditions after sleep; these were marked by the Finally, since the recordings that included a period of sleep elimination or, rarely, addition of spikes in the postsleep pattern were generally the ones of greatest duration, we also tested (Fig. 2). In cases where bursts had fewer spikes after sleep, the whether the more frequent occurrences of changes to burst pat- decrease in spike number was often accompanied by an increase terns in these recordings resulted from the additional passage of in interspike interval (Fig. 2A, B). Typically it was difficult or time rather than the presence of sleep. We selected the longest impossible to identify a specific spike from the presleep bursts (100-360 min) awake-only recording sessions (n = 8, with 71 that was eliminated after sleep. Instead we saw a restructuring of distinct burst classes). For each of these recordings, we defined an the entire burst. artificial separation interval (1.5-3 h) of similar duration to the Once a change occurred, it tended to be stable. Clear changes dark periods in the shorter-sleep recording sessions and then to burst structure after full nights of sleep were obvious in many compared the two sets of recordings. Even when controlling for EFTA01076050 27118 • .I. Neurosci., FebnNry 17,2010.3017):2783-2794 Retake et al. • Neuronal Stability and Drift across Sleep the passage of time, burst patterns under- B went many more changes across sleep than across wakefulness. During short sleep-inclusive recording sessions, ap- proximately one-fourth (22 of 83) of bursts exhibited structural changes, most of which were obvious under visual inspec- tion. In the recordings with the imposed artificial separation interval, no structural 71 1-7 changes were obvious under visual in- spection, and a much lower number of bursts (4 of 71; 6%; p C 0.001, Fisher's before exact test) exhibited structural changes. sleep ..4.4,44444.wavywo Changes in premotor activity occur across sleep-inclusive intervals The preceding results show that using sleep as the separator interval reliably identifies changes in RA burst patterns, after I but this does not rule out the possibility sleep that the changes actually tended to occur just before sleep (which could occur if the bird could anticipate the onset of the sleep period) or just after sleep. We formulated this hypothesis rigorously as a test of C whether there are previously undetected i dui lia it 11Slit changes to premotor burst patterns ex- pressed as a transition between distinct [111101 11 11,111! III 111111 states, in which a previously stable temporal spiking pattern is replaced with a different • pattern which then persists throughout before the subsequent songs. We then compared sleep those transitions to the occurrence of sleep. To test this hypothesis, for each of the 115 distinct burst stacks associated with the 15 neurons, we identified a separator ECM interval using an algorithm designed to identify the interval most likely to rep- after sleep 144 resent a transition to an altered burst pattern (see Materials and Methods), ig- r • 10 noring when sleep actually occurred. A large proportion of the separator intervals Ron2. Changestotemporal structure of RA premotor buntsacrossa period of sleep. A, Recordoms from an RA neuron that thus identified occurred close to sleep: 30 produced fewer spikes inpremotor bursts associated veith a song syllable alien 2 h skep pftiod. Top, pectrograph of song aligned with simultaneous recoil of premotor neuronal activity. Middle, Recordings of neuronal activity d ring three renditions of the of 115 separator intervals (seven neurons, song syllable. Bottom, Neuronal activity during three more renditions of the same song viable ahe sleep. The kftmost vertical four birds) either coincided with sleep or dashed hoe follows the fourth spike in al presleep bursts but precedes the fourth spike in all postsleep bursts, and the rightmost fell between the last two bursts before dashed line does the same with the eighth spikes. Note that whereas both weleep and postsleep bursts inconsistently Mdude an sleep or the first two bursts after sleep. To extra spite at the end, the postsleep hunts consistently produce one spike fewer than presleep bursts, oith an accompanying gap assess whether the degree of coincidence kit the middle of each bwst (arrows).8, Recordings from another RA neuron that produced fewer premotor spikes during produc- between the separator intervals and sleep tion eta contact call after a 2.5 h sleep period. The dashed line separates fourth spikes n each burst as per A.C. Recordings from a was attributable to chance, we generated third RA neuron that showed extra spikes in two remoter burstsassodated with a song syllable aftera 2 h sleep periodkale bars: predictions of the underlying distribution A, top, 250 ms; bottom, 10 ms;I,top, 100 at; bottom, 10 ms; C, top, 300 ms; bottom, 25 at. of separator intervals using the premotor activity of RA neurons recorded in the contiguous periods that we observed in our actual recordings. Furthermore, the same did not include sleep (see Materials and Methods). We then ran- sampling procedure applied to the 93 burst stacks not showing domly sampled from the predicted distribution 10,000 times for exact matches between separator intervals and sleep yielded an each of the set of 115 burst stacks to estimate the distributions of average of 3.8 ± 1.9 examples of separator intervals occurring separator intervals predicted by chance if the presence ofsleep did within -±1 interval of sleep. In contrast, there were eight such not bias the locations of the separator intervals. examples in the actual data, and only 3.7% of the random trials Overall, in the simulated data, the average number of exact had eight or more such examples (Fig. 5A). In total, the 30 sleep- matches between separator intervals and sleep was 2.5 -± 1.6 burst separator coincidences (exact or within one interval) we observed stacks (of 115 total), with a maximum of 11 such matches—just in the actual data set were much more than was ever observed in halfof the 22 exact matches between separator intervals and sleep the simulated data (maximum, 17). These distinctions were also EFTA01076051 Rauske el al. • Neuronal Stability and Drift across Sleep .I.I0eurovi, February 17, 2010 30(7):2783-2794 • 2789 Structural changes tend to duster across song In those neurons exhibiting structural changes in multiple burst stacks, we ob- served that changed bunts had a signifi- before cant tendency to cluster together. Across sleep the five neurons with changes to more than one burst stack within the song mo- tif, the spread of changed bursts (i.e., the interval between the first and last changed bursts) covered an average of 61% of the total number of bursts within each motif. after However, given the number of changed sleep bursts for each neuron, a random distri- bution of burst changes (determined by shuffling the distribution of changes within each neuron 100,000 times) would be expected to yield an average spread of 74 -± 7% of the total, and spreads as low as or lower than the observed 61% occurred in only 1.9% of the shuffled trials. The clustering was even more striking when excluding the neuron in which nearly ev- ery burst changed: in this restricted data set, the average spread between the first and last changed bursts covered 53% of the total number of bunts. This compares Figure 3. Persistent structural change to premotor bursts in a pair of RA neurons after a full night of sleep. Left, Raster to an expected coverage of 69 -± 9% in the plots of spiking activity during production of the song motif before(top)and after (middle)sleep. This site was the •double corresponding shuffled trials, with only unit' described in the text, in which the activity of a pair of neurons could reliably be distinguished from the background 1.6% of shuffled trials showing coverage activity but not horn each other; the data were treated equivalently to single unit data. Note the stability of the burst as low as actually observed. Across all neu- patternswithin the before.sleep and after•sleep groups, even over long periods of time (before sleep, 171 min; after sleep, rons with multiple bunt changes, nearly 459 min). Rasters are aligned within each group using the L rminimizatim method, and the groups are aligned with a one-third (6 of 19) of all intervals between spectrographof the song motif (bottom). Right, Finer temporal detail. The persistent loss of spikes from the middle bursts is dear, and the temporal pattern of the bursts fails to return to the presleep pattern even after several hours of postsleep changed bursts included an unaltered singing. Scale bars: Left, 100 ms; right, 25 ms. burst. Overall, these data indicate that the mechanism of change was biased to act maintained when assessing coincidence between the separator over temporally restricted portions of the motor program. interval and sleep with a 30 s criterion (Fig. 5B). A similar analysis comparing short-sleep recording sessions to awake-only record- Loss of spikes after sleep ing sessions of similar duration confirmed this result (see supple- We now turn to describing the changes in the structure of spike mental material available at www.jneurosci.org). Finally, the bunts in singing before and after sleep. A striking characteristic of distribution of separator intervals drawn from the actual data singing after sleep was the widespread reduction of overall splic- that did not coincide with sleep (i.e., the 85 of 115 separator ing activity of RA neurons. Most (28 of 33) of the structural intervals differing from sleep by at least two intervals) did not changes were characterized by a loss of spikes (average loss of exhibit significant difference from the corresponding simulated 1.48 -± 0.98 spikes/burst; range, 0.55-4.70), with the remaining distribution ( p = 0.14, Kolmogorov—Smirnov test). Considered five changed bunt stacks gaining spikes (average gain of 1.07 -± together, the overall distribution of separator intervals drawn 0.62 spikes/burst; range, 0.56-2.10). Across all burst classes with from the actual data was significantly different from the simula- structural changes, there was an average loss of 1.10 -± 1.31 tion distribution ( p C 0.01, Kolmogorov—Smirnov test). spikes/burst (representing 16.7 -± 19.9% of the total number of We note that our technique for identifying separator intervals presleep spikes/burst). The bunt durations did not decrease sig- depends on statistical comparisons of the L, measure across pop- nificantly however (average change —0.20 -± 4.18 ms; p = 0.79, ulations of bunts, and therefore the estimates of the timing of paired r test), because spike loss was associated with a significant putative changes to burst structure will exhibit noise that arises average increase of 0.55 -± 1.22 ms in the interspike intervals ( p from rendition-to-rendition variability of premotor bursts in RA 0.05, paired t test comparing bunts before and after sleep; both neurons. In fact, simply using sleep as an indicator for possible bunt durations and spike intervals were normally distributed). changes proved a more effective strategy for identifying structural In contrast to changes in firing rates after sleep, spike loss was changes, as only 27 of 115 bunt classes exhibited such changes not clearly present in structural changes during daytime singing. across algorithm-identified separator intervals, compared with A small majority of such burst stacks (II of 18) showed reduction the 33 changes observed across sleep. Since sleep was more reli- in spike number, with an average change across all 18 of these able for locating structural changes than was the algorithmic ap- bunt classes of -0.39 -± 1.10 spikes/burst, a statistically insignif- proach ignoring sleep, this supports the conclusion that under icant change ( p = 0.15, t test). This average change represented the conditions of our experiment, sleep itself induced discrete an overall loss of only 4.4% of the total number of spikes for these changes in premotor activity. bunts. Thus, there is some tendency in daytime recordings for EFTA01076052 2790 • 1. Neurosci„ Fettuary 17, 2010 • 30171:2781-2794 Rauske et al. • Neuronal Stability and Dnfl across Sleep I 3 : [:::- -•• is:i.i iI,,. •: 314 * 4 •- -3 4 S 1c tl CI '4 :.' i dd • i it f c•c 1 •• '3 .,. • • .4 v -.S. 791n**;:-. 1 ;1 • • .*:i f g I t .. `: I :• CM' I 41. . ••• • tH nit.' gli •••• f Si-:11: I II t...... 1,3: -- ... pre-sleep ... post-sleep Figure 4. Changes to premotor burst stricture in RA neurons. Raster plots for 33 premolar bursts that exhibited structural changes across sleep are shown. Presleep rasters ate in red, and postsleep rasters are inblue. Scale bats, S ms. the same phenomenon as observed across sleep, but it is much spikes/burst, representing 0.5 -± 12.9% of presleep spikes/burst; weaker. p = 0.75, paired r test). The preceding analyses were restricted to bunts exhibiting Bursts that had not undergone structural changes nevertheless structural changes; we now broaden consideration to all bursts. had a strong tendency to show changes in spike counts after sleep. When considering all bursts, there was a significant net loss of Many bunts exhibited a change in the tendency for a given spike spikes after sleep (0.30 -1- 0.97 spikes/burst; p < 0.01), resulting in to occur (often apparently the last spike), or for a change in the an overall reduction of 6% from presleep spike counts. This ap- rate of activity in more variable sections of the bursts. Such parent global reduction, however, was entirely accounted for by changes were not well characterized by the feature-based analysis the spike losses observed in conjunction with structural changes. but resulted in spike count changes. Comparing mean spike There was no significant tendency for bursts without structural counts of presleep burst renditions with mean spike counts of changes to gain or lose spikes (average net increase, 0.02 -± 0.52 postsleep burst renditions, we observed that 45% (37 of 82) of EFTA01076053 Rauske et al. • Keurenal Stability and Drift across Sleep J. Neurosci., February 17, 2010 • 30(7):2783-2794 • 2791 A 16 stronger across sleep than in daytime recordings, where spike counts were significantly more stable: only 17% of all burst stacks (95 of 551) had significant spike count differences across the % random trials daytime separator intervals used for our prior analyses (p < 0.001; x2 = 38.0). The mean magnitude of change was also greater across sleep (0.69 -± 0.84 spikes/burst) than across wak- 0 20 40 ing periods of similar duration (0.38 -± 0.46 spikes/burst; p # sleep-separator coincidences (within 1 rep) 0.01, t test). We conclude that there is an ongoing process of increases and B 12 decreases in spike counts of RA bursts, but only over the sleep interval are these changes strongly biased toward spike loss and % random trials decreasing spike counts. Spike loss must be compensated for by an as yet unidentified regenerative process that increases spike counts (see Discussion). o o 20 40 sleep-separator coincidences (within 30 sec) Temporary shifts in burst timing after sleep Finally, we observed one additional form of neuronal variability. Figure S. Bias ofLi-optimized separators toward sleep-ndusive ntervals. A, Histogram of The period immediately after birds awoke was occasionally asso- the distribution of separator-sleep coincidences (defined as zero- or one-syllable renditions ciated with a temporary increase in duration of intervals between between the f;optimized separator and dark phase) generated from 10,000 randomlydeter- bursts within syllables. This was observed for 14 of 55 interburst mined separators drawn for each burst dass from the PDF shown n Figure 18. The 30 of 115 intervals (10 of 31 syllables, 4 of 12 neurons), with all shifts tend- coincidences actualy observed in the data (arrow) were much mote than the mean number ing toward longer intervals (average increase, 3.54 ± 2.24 ms), predicted by chance alone. 8, Histogram of separator-sleep coincidences according to Wa- rta& time (defined as <30 s between the 1,-optimized separator and dark phase) generated and the magnitude of the shift representing 10.2 -± 9.4% of the from the same 10,000 randomly determined sets of separators used in A. The 39 of 11S exam- mean presleep duration of that interval. In contrast, only one ples observed in the data (arrow) were more than was ever observed under random samping. burst interval exhibited a significant change during singing that immediately preceded the subjective night, a decrease of 0.81 ms (8.7%) from the av- long calls song syllables erage preceding interval. In most (10 of 14) cases, interburst intervals rapidly re- covered to presleep values. The period of relaxation to presleep intervals was 323 ± 370 s. The appearance of changes in inter- burst timing was not correlated with the appearance of changes in burst structure. Bursts bordering intervals with significant postsleep drift had similar frequency of structural changes (25%; 6 of 24) compared Pre with other bursts (structural changes 27 of 91; 30%; p = 0.80, Fisher's exact test). This observation indicates that changes to burst post structure are not the result of changes to song tempo expressed at the timescale of in- terburst interval plasticity. Particularly favorable examples were cases of changes in interburst intervals that were observed for long calls, which birds often produce within seconds upon waking, whereas even in favorable cases 'g they only begin to sing minutes later. Four of these cases (out of seven total interburst intervals, three sites, two birds) represent dramatic examples of plasticity immedi- figure6. Temporary postsleep plastidty in interburst bming diving calfing.Left,Raster pkitsof nemotor activity al three sites (three ately after wakening (Fig. 6). In these separate nights)in one bird duringyroduction &distance calls.11ote that theadnityafter sleep (below the horizontal dashed lines)initially shows greater interburst intervals, whkh then gradualy relax toward the presleep values. BOIL Raster plots of gremotot manly at the cases, many (18 —115) long calls were pro- same siteduring production °Infected song syllatres.Brackets indicate peiodsofonnap6ng sinOing and callingadnityaftersleep;note duced starting 15-97 s after waking and that the Fostsleep increase inintedourg intervalsduring cal production hasmostlyeatishedby thetimesingog commentesSedebak2Oms. continuing for 141-542 s preceding the first postsleep song. The overall magni- tude of change in spike timing at the onset burst stacks not exhibiting structural changes nevertheless un- of the light phase was 6.2 ± 2.2 ms. By the time singing com- derwent significant changes (unpaired t test) in their mean spike menced, interburst intervals associated with long calls had al- counts. There was no bias toward loss or gain of spikes. The ready returned to presleep durations. Burst intervals associated overall tendency for spike counts to change was significantly with song syllables recorded at those same sites exhibited only EFTA01076054 2792 • .I. Neurosci., Fekeuary 17,2010 • 3017J:2783-2794 Rauske et al. • Neuronal Stability and Dolt awns Sleep modest increases of 1-3 ms followed by gradual change toward RAps at the very fine time scale we observed for single neurons. presleep interval durations (7 intervals) or no significant changes Information carried by neuronal replay could serve as the sub- (14 intervals). These data help to emphasize that the process un- strate for fine-tuning neural networks during sleep, where the derlying structural changes proximate to sleep produced stable spontaneous replay of premotor bursts has been observed in RA changes that are distinct from the temporary plasticity manifest (Dave and Margoliash, 2000) and may play similar roles in other in the period immediately after wakening. The song and burst systems (Wilson and McNaughton, 1994; Qin et al., 1997; timing variability suggest that adult zebra finches exhibit a brief Nadasdy et al., 1999; Hoffman and McNaughton, 2002; Pennartz period of performance variability immediately after waking that et al., 2004; Ji and Wilson, 2007; Peyrache et al., 2009). is akin to performance variability termed sleep inertia in humans (Dinges, 1990). The period of sleep inertia is shorter in finches than in humans (Jewett et al., 1999). Circuit models ofRA burst changes It is helpful to consider models of how RA bursts are generated. Discussion Sleep-mediated changes in burst structure could reflect changes We demonstrated that periods ofsleep are commonly associated in inputs to RA and/or changes in RA local circuits. During sleep, with small but secure changes in the burst structure of RA neu- spontaneous RA activity is strongly driven by HVCactivity (Dave rons. RA burst changes occurred even when sleep was artificially and Margoliash, 2000; Hahnloser et al., 2006). HVC-RAns have curtailed, indicating that the process ofchange is related to sleep been described as sparsely firing, each neuron emitting a highly itself and is not a manifestation of circadian cycles. This also regulated single short burst once per motif (Hahnloser et al., argues against nonspecific effects (such as movement artifacts or 2002; Kozhevnikov and Fee, 2007). A recent "clock" model posits electrochemical changes at the electrode tip associated with long- that RAp activity is dominated by clock-like, feedforward excita- duration recordings) driving the observed changes in burst struc- tion from HVC (Leonardo and Fee, 2005), with a given popula- ture. Nonspecific effects should increase monotonically with the tion of HVC-RAns broadly distributed in HVC projecting onto amount ofmovement (most prominent during daytime singing) single RAps, and a subset of HVC-RAns active at any moment in or duration of recordings, but we observed no such correlations. time. Changing one or a few HVC-RAns, representing a particu- The spontaneously tonically firing neurons we recorded from lar time point in the clock, could affect a specific burst of an RA are qualitatively similar to the presumptive RA projection neu- neuron. rons (RAps) recorded during singing but held for shorter periods A second model of RA bursting arises from the observation of time (Yu and Margoliash, 1996; Dave and Margoliash, 2000; that each RAp spontaneously oscillates at a given frequency (Yu Leonardo and Fee, 2005). In contrast, RA interneurons are likely and Margoliash, 1996), which arises from the interaction of in- to fire more sporadically (Spiro et al., 1999; Leonardo and Fee, trinsic RAp subthreshold oscillations with network properties 2005) (see Materials and Methods) and are very rarely isolated in (Mooney, 1992). The "reconfiguration" model posits that during chronic recordings. Our experimental design was challenging— singing, different RA neurons (viewed as simple oscillators) are maintaining recordings over two singing periods separated by dynamically coupled and uncoupled by the action of interneu- sleep. Our sample size is correspondingly small, collected over rons, resulting in transient local networks with complex bursting several years ofrecordings.If our sample ofRAps is unbiased, this duringsinging such as is actually observed for RA neurons. In this implies that approximately halfof all RA projection neurons alter model, HVC inputs would select different local RA circuits via their burst patterns on a nightly basis, and that these changes are changes in the synaptic weights onto RAps, thereby influencing expressed in >40% of the bursts in those neurons. RA burst patterns. Long-distance inhibitory interactions within The nucleus RA represents the sole forebrain output of the RA could help support rapid coupling of different sets of RA song system, projecting to brainstem nuclei that control the syr- projection neurons (Spiro et al., 1999). The principal distinction inx and regulate respiratory rhythm. Sparse singing activity in from the clock model ofLeonardo and Fee (2005) is that local RA RA-projectingHVC neurons (HVC-RAns) probably represents a circuits contribute to the structure and the timing of RA bursts. time code (Hahnloser et al., 2002; Long and Fee, 2008), which is We note that the two models are not mutually exclusive, and converted to a denser representation representing notes (parts of alternate models may obtain (Trevisan et al., 2006). Additionally, syllables) in RA (Yu and Margoliash, 1996; Leonardo and Fee, neither model considers a recently described class of RAps that is 2005). Accumulated continuously throughout adult life, it seems reciprocally connected with HVC (Roberts et al., 2008). unlikely that the observed level of RA nocturnal spike loss could To directly compare the two models, we consider the case represent uncompensated noise or drift without changes in syl- where the spike loss we observed arises from loss of synaptic lable morphology or larger circadian changes in adult song than drive from HVC-RAns onto RAps. (Alternatively, spike loss have been observed (Deregnaucourt et al., 2005; Glaze and Troyer, could arise from changes local to RA.) How would reduction in 2006). Instead, we hypothesize that RAp nocturnal variation is a HVC drive manifest as variation in RA bursting? Structural component of adult song maintenance, an active process me- changes were characterized by spike loss, not changes in burst diated by auditory feedback (Nordeen and Nordeen, 1992; timing. If several HVC-RAns represent any given moment in Leonardo and Konishi, 1999; Andalman and Fee, 2009; Sober and time, this is consistent with the clock model, so that loss of some Brainard, 2009). The apparent role of nighttime sleep in song HVC-RAn input changes the magnitude but not the timing of the maintenance as revealed by song deterioration after adult deaf- drive into an RAp. Spike loss was also associated with compensa- ening is consistent with this hypothesis (Derkgnaucourt et al., tory changes in interspike intervals tending to affect the entire 2005). RA bursting during sleep adaptively changes with devel- burst, yet HVC-RAn bursts are much shorter than the duration of opmental song learning, providing additional if indirect support many RAp bursts. Either the timing of HVC-RAns "tile" the du- for this hypothesis (Shank and Margoliash, 2009). In songbirds, ration of an RAp burst and several HVC-RAns (representing syringeal muscles are "superfast," exhibiting functional modula- nonidentical points in time) change, or several simultaneous tion at frequencies exceeding 200 Hz (Elemans et al., 2008), and HVC-RAn inputs (one or more of which is lost) initiate an RAp could be sensitive to changes distributed over a population of burst but do not regulate it thereafter. EFTA01076055 "tusk et al. • Neuronal Stability and Drift across Skop 1. Neurosei., February 17,2010 30(7):2783-2794 • 2793 Models of PA bursting also need to account for the distribu- Brainard MS, Doupe AI (2000) Auditory feedback in learning and mainte- tion of bursts that changed in a given RAp. The percentage of nance of vocal behaviour. Nat Rev Neurosci 1:31-40. neurons with changed bursts decreased monotonically from zero Brawn TP, Fenn KM, Nusbaum HC,Margoliash D (2008) Consolidation of sensorimotor learning during deep. Learn Mem 15:815-819. changed bursts per neuron to the one cell with 11 changed bursts, Cardin IA, Schmidt MF (2003) Song system auditory responses are stable and there was a tendency for changed bursts to be temporally and highly tuned during sedation, rapidly modulated and unselective restricted in song but not necessarily temporally contiguous. For during wakefulness, and suppressed by arousaL 1Neurophysiol 90:2884- the clock model, temporal contiguity of changes could result if 2899. any plastic event in HVC that occurs during sleep were to prop- Chi Z,Margoliash D (2001) Temporal precision and temporal drift in brain agate down the chain ofHVC-RAns acsnriated with that RAp. To and behavior of zebra finch song. Neuron 32:899-910. fit the observed data, it would also be necessary for the plastic Dave AS, Margoliash D (2000) Song replay during sleep and computational rules for sensorimotor vocal learning. Science 290:812-816. changes to skip some HVC-RAns while continuing to propagate Dave AS, Yu AC, Margoliash D (1998) Behavioral state modulation of au- down the chain. For the reconfiguration model, an attractive ditory activity in a vocal motor system. Science 282:2250-2254. hypothesis is that during sleep a small number of RAps exhibit Dave AS, Yu AC, Gilpin II, Margoliash D (1999) Methods for chronic neu- robust changes in somatic or proximate dendrite conductances ronal ensemble recordings in singing birds. In: Methods for simultaneous (e.g., the RAp with most ofits bursts changed), and that the effect neuronal ensemble recording (Nicoklis MAL, eel), pp 101-120. Boca on other RAps depend on how many circuits the two share Raton, FL• CRC. Deregnaucoun 5, Mitra PP, Feher O, Pytte C, Tchemichovski O (2005) throughout song. To account for the temporal contiguity of the How sleep affects the developmental learning of bird song. Nature burst changes requires another assumption, most parsimoni- 433:710-716. ously that nearby RA neurons tend to participate in more shared Dinges DF (1990) Are you awake? Cognitive performance and reverie dur- circuits. ing the hypnopompic state. In: Sleep and cognition (Bootzin R,KihIstrom 1, Schacter D, eds), pp 159-175. Washington, DC American Psycholog- Homeostatic regulation of electrical excitability ical Association. Spike elimination during sleep must be compensated by spike Elemans CP, Mead AF, Rome LC. Goller F (2008) Superfast vocal muscles gain, otherwise over time RAps would cease firing. Spike gain control song production in songbirds. PLoS One 3:e2581. Fenn KM, Nusbaum HC, Margoliash D (2003) Consolidation during sleep could occur during the day, for example during undirected sing- of perceptual learning of spoken language. Nature 425:614-616. ing not sampled in our experiments. Undirected singing is char- Fischer S. Hallschmid M, Elsner AL, Born I (2002) Sleep forms memory for acterized by increased temporal variability in acoustic output finger skills. Proc Natl Acad Sci USA 18:11987-11991. (Kao et al., 2005; Olveczky et al., 2005), and greater variability in Frank MG, Issa NP, Stryker MP (2001) Sleep enhances plasticity in the de- activity of the neurons of the lateral magnocellular nucleus of the veloping visual cortex. Neuron 30:275-287. nidopallium (LMAN) (Hessler and Doupe, 1999; Leonardo, Glaze CM, Troyer TW (2006) Temporal structure in zebra finch song: Im- 2004) that help regulate auditory-feedback-mediated song mainte- plications for motor coding. 1 Neurosci 26S91-1005. Hahnloser RH, Kozhevnikov AA, Fee MS (2002) An ultra-sparse code nance (Brainard and Doupe, 2000). Adult neurogenesis is prom- underlies the generation of neural sequences in a songbird. Nature inent in the population of HVC-RAns (Alvarez-Buylla and Kim, 419:63-70. 1997; Scott and Lois, 2007; Scotto-Lomassese et al., 2007). Incor- Hahnloser RH, Kozhevnikov AA, Fee MS (2006) Sleep-related neural activ- poration ofnew functional synapses could increase excitability of ity in a premotor and a basal-ganglia pathway of the songbird. I Neuro- RA neurons. By this scenario, RA neurons could lose excitatory physiol 96:794-812. input over multiple nights and then suddenly regain it (presum- Messier NA,DoupeAi (1999) Social context modulates singing-related neu- ral activity in the songbird forebrain. Nat Neurosci 2:209-211. ably at night). The latter, hypothesized event would be rare, and Hoffman KL, McNaughton BL (2002) Coordinated reactivation of distrib- difficult to detect in electrophysiological recordings. uted memory traces in primate neocortex. Science 297:2070-2073. The mechanisms of spike loss, and presumptive spike gain, Jewett ME, Wyatt 1K, Ritz-De Cecco A, Khalsa SB, Dijk DI, Czeisler CA may be specific to the song system, but regulation of levels of (1999) Time course of sleep inertia dissipation in human performance spiking activity may be a more general phenomenon. It has been and alertness. I Sleep Res 8:1—8. proposed that homeostatic mechanisms regulate the overall level li D, Wilson MA (2007) Coordinated memory replay in the visual cortex of excitation in the nervous system (Turrigiano and Nelson, and hippocampus during sleep. Nat Neurosci 10:100-107. 2004). Spike loss duringsleep could be a superthreshold manifes- Kao MH, Doupe AI, Brainard MS (2005) Contributions of an avian basal ganglia-forebrain circuit to real-time modulation of song. Nature tation of such mechanisms, shedding excess spikes that accumu- 433:638-643. late during daytime activity (Vyazovskiy et al., 2009) to maintain Karni A, Tonne D, Rubenstein BS, Askenasy IL Sagi D (1994) Dependence a static level ofoverall network excitability, and is consistent with on REM sleep of overnight improvement of a perceptual skill. Science recent evidence for a net potentiation of synaptic strength in rat 265:679-682. cortex and hippocampus during wakefulness and net depression Kozhevnikov AA, Fee MS (2007) Singing-related activity ofidentified HVC during sleep (Vyazovskiy et al., 2008). If spike loss of the same neurons in the zebra finch. 1Neurophysiol 97:4271-4283. magnitude as we have observed in RAps occurs in other systems, Leonardo A (2004) Experimental test of the birdsong error-correction model. Proc Nad Acad Sci USA 101:16935-16940. it may be difficult to detect with single-cell recordings in those Leonardo A, Fee MS (2005) Ensemble coding of vocal control in birdsong. systems, where spiking activity is less precise than that of RAps. Neurosci 25:652-661. Leonardo A, Konishi M (1999) Decrystallization of adult birdsong by per- References turbation of auditory feedback. Nature 399:466-470. Alvarez-Buylla A,KimIR (1997) Birth, migration, incorporation, and death Lewicki MS (1994) Bayesian modeling and classification of neural signals. of vocal control neurons in adult songbirds. I Neurobiol 33:585-601. Neural Comp 6:1005-1030. Andalman AS, Fee MS (2009) A basal ganglia-forebrain circuit in the song- Long MA, Fee MS (2008) Using temperature to analyse temporal dynamics bird biases motor output to avoid vocal errors. Proc Natl Acad Sci U S A in the songbird motor pathway. Nature 456:189-194. 106:12518-12523. Low PS, Shank SS, Sejnowski TI, Margoliash D (2008) Mammalian-like fea- Mon SI, Seibt 1, Dumoulin M, Iha SK, Steinmetz N, Coleman T, Naidoo N, tures of sleep structure in zebra finches. Proc Natl Acad Sci U S A 105: Frank MG (2009) Mechanisms of sleep-dependent consolidation of 9081-9086. cortical plasticity. Neuron 61:454-466. Maquet P, Laureys S. Peigneux P, Fuchs S. Petiau C, Phillips C, Aens I, Del EFTA01076056 2791 • J. Neurosci., Felxuary 17,2010 • 3(471:1783-2794 Rauske et al. • Xeuronal Stability and Drys across Skep Fiore G, Degueldre C. Meulemans T. Luxen A, Franck G, Van Der Linden Scott BB, Lois C (2007) Developmental origin and identity of song system M, SmithC, Cleeremans A (2000) Experience-dependent changes in ce- neurons born during vocal learning in songbirds. I Comp Neural rebral activation during human REM sleep. Nat Neurosci 3:831-836. 502:202-214. Mooney R (1992) Synaptic basis for developmental plasticity in a birdsong Scotto-Lomassese 5, Rochefort C, Nshdejan A, ScharffC (2007) HVC inter- nudeus. I Neurosci 12:2464-2477. neurons are not renewed in adult zebra finches. Eur I Neurosci Nadasdy Z, Hirase H, Czurko A, Csicsvari 1. Buzsaki G (1999) Replay and 25:1663-1668. time compression of recurring spike sequences in the hippocampus. Shank SS. Margoliash D (2009) Sleep and sensorimotor integration during Neurosci 19:9497-9507. early vocal teaming in a songbird. Nature 458:73-77. Nick TA, Konishi M (2001) Dynamic control of auditory activity during Simpson HR. Vicario DS (1990) Brain pathways for learned and unlearned sleep: correlation between song response and EEG. Proc Nall Acad Sci vocalizations differ in zebra finches. 1Neurosci 10:1541-1556. USA 98:14012-14016. Sober SI, Brainard MS (2009) Adult birdsongis actively maintained by error Nordeen KW, Nordeen El (1992) Auditory feedback is necessary for the maintenance of stereotyped song in adult zebra finches. Behav Neural correction. Nat Neurosci 12:927-931. Biol 57:58-66. Sossinka R, Bohner 1 (1980) Song types in the zebra finch (Poephila gurrata Olveaky BP, Andalman AS, Fee MS (2005) Vocal experimentation in the cautworis). Z Tierpsychol 53:123-132. juvenile songbird requires a basal ganglia circuit. PLoS Bio13:e153. Spiro 1E, Dalva MB, Mooney R (1999) Long-range inhibition within the Peigneux P, Laureys S, Fuchs S. Collette F, Perrin F, Reggers 1, Phillips C, zebra finch song nucleus RA an coordinate the firingof multiple projec- Degueldre C, Del Fiore G. Aerts 1, Luxen A, Maquet P (2004) Are spatial tion neurons. 1Neurophysiol 81:3007-3020. memories strengthened in the human hippocampus during slow wave Stickgold R, lames L. Hobson IA (2000) Visual discrimination requires sleep? Neuron 44:535-545. sleep after training. Nat Neurosci 3:1237-1238. Pennartz CM, Lee E, Verheul L Lipa P, Barnes CA, McNaughton BL (2004) Trevisan MA, Mindlin GB, Colter F (2006) Nonlinear model predicts di- The ventral striatum in off-line processing: ensemble reactivation during verse respiratory patterns of birdsong. Phys Rev Lett 96:058103. sleep and modulation by hippocampal ripples. I Neurosci 24:6446 -64%. Turrigiano GG, Nelson SB (2004) Homeostatic plasticity in the developing Peyrache A, Khamassi M, Benchenane K, Wiener SI, Battaglia FP (2009) nervous system. Nat Rev Neurosci 5:97-107. Replay of rule-learning related neural patterns in the prefrontal cortex Vyazovskiy VV,Cirelli C.Pfister-Genskow NI, Faraguna U, Tononi G (2008) during sleep. Nat Neurosci 12:919-926. Molecular and electrophysiological evidence for net synaptic potentiation Poe GR, Nitz DA, McNaughton BL, Barnes CA (2000) Experience- in wake and depression in sleep. Nat Neurosci 11:200-2008. dependent phase-reversal of hippocampal neuron firing during REM Vyazovskiy VV, Olcese U, Lazimy YM, Faraguna U, Esser SK, Williams 1C, sleep. Brain Res 855:176-180. Cirelli C. Tononi G (2009) Cortical firing and sleep homeostasis. Neu- Qin YL, McNaughton BL, Skaggs WE, Barnes CA (1997) Memory repro- ron 63:865-878. cessing in conicocortical and hippocampoconical neuronal ensembles. Wagner U, Gais 5, Haider H, Verleger R, Born 1 (2004) Sleep inspires in- Philos Trans R Soc Lond B Biol Sci 352:1525-1533. Rauske PL, Shea SD, Margoliash D (2003) State and neuronal dass- sight. Nature 427:352-355. dependent reconfiguration in the avian song system. I Neurophysiol Walker MP, Brakefield T, Morgan A, Hobson IA, Stickgold R (2002) 89:1688-1701. Practice with sleep makes perfect: sleep-dependent motor skill learn- Reis I, Schambra HM, Cohen LG, Buch ER, Fritsch B, Zarahn E, Celnik PA, ing. Neuron 35:205-211. Krakauer nv (2009) Noninvasive cortical stimulation enhances motor Wilson MA, McNaughton BL (1994) Reactivation of hippocampal ensem- skill acquisition over multiple days through an effect on consolidation. ble memories during steep. Science 265:676-679. Proc Natl Acad Sci USA 5:1590-1595. Yu AC, Margoliash D (1996) Temporal hierarchical control of singing in Roberts TF, Klein ME, Kubke MF, Wild IM, Mooney R (2008) Telence- birds. Science 273:1871-1875. phalic neurons monosynaptically link brainstem and forebrain premotor Zann RA (1985) Ontogeny of the zebra finch distance call:I.Effects ofcross- networIcs necessary for song. I Neurosci 28:3479-3489. fostering to Bengalese finches. Z Tierpsych 68:1-23. EFTA01076057 Supplemental Information Separator interval duration and sleep In the sleep-inclusive recordings, the dark phase differed from other inter-rendition intervals not only in the presence of sleep, but in its much greater duration. To determine whether the additional passage of time, rather than the presence of sleep, could be responsible for the more frequent occurrences of sleep-separator coincidences or changes to burst patterns in sleep-inclusive recordings, we used the longest (100-360 min) awake-only recording sessions (n=8, with 71 distinct burst types). For each of these recording sites, we defined an artificial separation interval of 1.5-3 h, so that the resulting intervals were of similar duration to the dark periods in the shorter-sleep recording sessions. The beginning and end of the artificial interval were chosen so as to occur between uninterrupted singing bouts and to yield at least 8 renditions of each syllable type before and after the interval. We removed all vocalizations within the artificial interval from this stage of the analysis (Supp. Fig. 2A). We also employed a similar bootstrap method using 10,000 iterations to create a distribution of predicted coincidences between the artificial intervals and the optimized separators, which we used to evaluate the significance of the observed number of coincidences. We used this newly generated distribution of predicted long (awake) interval-separator coincidences to evaluate the number of observed sleep-separator coincidences observed in short- sleep experiments. If the previously observed bias toward sleep-coincident separators was due to the greater duration of the sleep interval when compared with the other intervals, then we should have been able to observe a similar bias toward the long intervals devined for our wake-only sessions. In fact, we found 8 such coincidences, compared with an average of 8.3 +1- 2.5 coincidences in our randomly sampled data—i.e. the same number of coincidences predicted by EFTA01076058 chance (Supp. Fig. 2B). In contrast, in our short-session sleep-inclusive recordings, the number of sleep-separator coincidences (28/83) was more than was ever observed across 10000 random samplings (mean 10.2 +/- 2.9, maximum 20). Supplemental Figure 1. Assessment of behavioral state using RA spiking activity. Representative spiking activity of an RA neuron with the lights on and the bird awake (top); five minutes after lights-out, with the bird transitioning to sleep (middle); and 30 minutes after lights- out, with the bird asleep. Raw recording traces of neuronal activity (18-second duration) are presented on the left, with inter-spike interval (ISI) histograms generated from the same traces on the right. Vertical dashed lines divide the raw traces into the 3-second epochs used to determine behavioral state (see Materials and Methods), and are labeled as awake (W) or asleep (S) based upon the distribution of ISI durations within the epochs. Scale bar, 5 s. Supplemental Figure 2. Comparison of burst changes across sleeping and wakeful intervals of similar duration. (a) Construction of an artificial separation interval, represented by the dashed line in the raster plot on the right, from an awake-only recording session. (b) Histogram of the distribution of separator-artificial interval coincidences (defined as 2 or fewer syllable renditions between optimized separator and artificial interval) generated from 10,000 randomly determined separators drawn for each burst type from the PDF shown in Figure 2b. The 8/71 coincidences actually observed in the data (arrow) were equivalent to the mean number predicted by chance alone. EFTA01076059 W W W W W W rI I I I I 1 1 1 1 1 1 1 1 1 1 I I I I I S EFTA01076060 B 15 % random trials 0 20 40 # interval-separator coincidences EFTA01076061