How sleep affects the developmental learning of bird song
1. How sleep affects the
developmental learning of
bird song
S. Degregnaucourt et al. Nature 2005
introduced by K. Sasahara
2010 10 4 1
2. Introduction
• Sleep facilitates memory consolidation.
• Reactivation
Hippocampal areas activated during spacial learning were
reactivated during sleep in humans.
• Replay
Some neurons of premotor nuclei show spontaneous bursting
during sleep, similar to the pattern of activity in awake singing birds.
• Bridging between sleep and developmental learning
• A model system
Songbirds learn a correspondence between vocal-motor output
and auditory feedback during development.
2010 10 4 2
3. govern this skill in songbirds have been de- Fig. 1. An example of
scribed (3). We here report on conditions that training. (A) Acclima-
bring vocal learning under fine experimental tion to the training
Materials and Methods
apparatus from days
control and provide a detailed acoustic anal- 30 to 42 after hatch-
ysis of the sound transformations that under- ing, in the presence of
lie the learning process. a plastic model of an
Zebra finch (Taeniopygia guttata) males adult male (on middle
develop their song between 35 and 90 days perch). (B) Untutored
after hatching, a time known as the sensitive subsong was recorded
on day 43. Spectral
period for vocal learning (4). This song con- derivatives provide a
sists of complex sounds (“syllables”) separat- representation of song
ed by silent intervals (5). A song motif is that is similar but su-
composed of dissimilar syllables repeated in perior to the tradi- Model tutor Pupil
•
a fixed order (5). When a young male zebra tional sound spectro-
Male zebra finches
finch is reared singly in the company of an
adult male, it develops a song that is a close
gram. Instead of pow-
er spectrum versus
time, we present di- Target
copy of the sounds and temporal order of that
•
rectional derivatives
Continuous recording with a
male’s song (4, 6). Acquisition of the audi- (changes of power) on
tory memory of the model song can start as a gray scale so that
early as 25 days after hatching, but this onset the detection of fre-
model tutoring
can be delayed by withholding exposure to
the model (7, 8). Once acquired, a stored
quency contours is lo-
cally optimized. This
was particularly useful
representation of the model song can be con-
2 days of training
•
for the analysis of ju-
verted to a motor imitation. This conversion
Continuous measurements
venile song. (C) The
has been modeled by assuming simple Heb- keys were then uncov-
bian and reinforcement learning rules (9). ered. The bird learned
Nevertheless, past technical limitations en- to peck on either one
•
of the keys to induce a
Syllable features
countered when studying early song develop-
ment have left much of the fine-grained struc-
short song playback
from the plastic mod-
e.g., duration, pitch, AM, FM, entropy ...
ture of the imitation process unexplored. el. (D) Song playback 3 days of training
was composed of two
renditions of the song
•
1
Field Research Center, The Rockefeller University, motif (the “model”)
Similarity to the target
Millbrook, NY 12545, USA. 2Bell Laboratories, Lucent
Technologies, Murray Hill, NJ 07974, USA.
depicted. The overall
daily exposure was
*All authors contributed equally to this work. limited to 28 s. As
†To whom correspondence should be addressed. E- shown, the bird’s song
mail: tcherno@mail.rockefeller.edu had changed by (E) the second and (F) the third day of training.
2564
O.Tchernichovski et al. Science 2001
30 MARCH 2001 VOL 291 SCIENCE www.sciencemag.org
2010 10 4 3
4. bserved strong daily oscillations in syllable features, Supplementary Data).
ncedly in variance features that capture the richness of The decrease in syllable structure in the morning may suggest that
cture within a syllable (Supplementary Table 1). vocal changes during the day progress with the overall develop-
Tracing Vocal Changes
ws how values of Wiener entropy variance (EV) mental trend, whereas vocal changes after night-sleep oppose the
m day to day during development (Fig. 3a, b). Strong developmental trend. To test this hypothesis we assigned a sign to
ions were observed shortly after training started vocal changes during night-sleep, by reference to the overall
and decreased thereafter (Fig. 3e, f). Note that the developmental trend: positive if in the same direction, negative if
V after night-sleep was opposite to the overall trend of opposite. For example, if syllable mean pitch decreased during
• Three clusters emerge and stabilize
with development
• Variability decreases in the morning
vocal changes. a, Spectral derivatives29 of adult song motif with three emerged shortly after training. d, Plotting two-dimensional distribution (duration versus
othed histogram4 syllable durations46 in an adult bird (^95%
2010 10 of Wiener EV) shows syllable types as clusters (unclustered syllables are not shown). 4
5. many samples should approach zero. As shown in Fig. 2c, same conditions but were not trained did not show a significant
across the 12 birds (including all features of all syllables) trend (Fig. 2e, f). The magnitude of post-sleep deterioration
ed that the net effect of vocal changes after night-sleep was differed significantly across the groups: highest when training
. Furthermore, deterioration was observed in both mean started early, lower when training started later, and lowest in the
Vocal Changes During Night-sleep
iance feature values (Supplementary Table 1) and was untrained birds (Fig. 2f, median test, P , 0.05). Overall, song was
l from day 50 to day 55. To separate the effect of training less structured and more primitive after night-sleep but only during
at of chronological age, we repeated the experiment in birds days of rapid learning.
Full bars: model tutor
Striped bars: live tutor
• Baseline < Day-to-day < Pre-post sleep
• Isolates < 60d < 43d
• Adults still retain the day-to-day
variability, suggesting learning plasticity
ocal changes10 4
2010 during night-sleep. a, Vocal changes (absolute values; night-sleep with reference to the overall developmental trend, in birds trained from day 43 5
7. of the similarities within a 1-h moving window. The record laboratory . The bird did not attempt to sing but exhibi
similarity at a given time is defined as the best 95th centile achieved activities including eating and calling. When singing resu
so far. Improvements in record similarity are computed with 4 h after lights were turned on) we would have expected
Progression of Song Learning
reference to developmental time, not to the previous hour, thus
excluding improvements that are due solely to recovery (Fig. 4b, c).
Strong improvements in record similarity were achieved during
ρ=0.6, P=0.038
• Eventual similarity
∝ Magnitude of daily oscillation
× Total developmental change
Record similarity
(the best 95th centile so far) • Morning intense singing guides
95th centile strong improvements in
similarity.
• Morning less structured
(more plastic) sounds gives an
opportunity to explore
learnability.
Figure 5 Vocal changes after song prevention and sleep manipulation. a, R
song structure could be circadian (scenario 1) or it might require singing activ
2010 10 4 Figure 4 Progression of song learning. a, Relation between magnitude of post-sleep 2). b, Vocal changes (in reference to the developmental trend) were measured7
8. owards structure. Alternatively, vocal changes might be caused by circadian of song prevention (Fig. 5d). The difference between the effects of enon has not b
record changes in hormonal state. To test both hypotheses, we prevented night-sleep and those of song prevention was statistically significant stronger in) dev
ng the the bird from singing for 2 h during one morning (n ¼ 6, range: (Wilcoxon, P ¼ 0.02). be the requirem
centile
record
hieved
d with
Causes of Post-sleep Deterioration
50–57 days) by taking the cage from the sound box into the
laboratory35. The bird did not attempt to sing but exhibited usual
activities including eating and calling. When singing resumed (3–
4 h after lights were turned on) we would have expected to see at
Finally, we tested for causal relations between sleep and deterio-
ration of song structure by inducing sleep during the day. Melatonin
can induce sleep in zebra finches, and electrophysiological record-
ings suggest that song replay occurs readily in this induced sleep
ments spannin
We advance
related oscillati
the observation
r, thus state27. We allowed each bird (n ¼ 6, age 54–63 days) to recover its imitation. The
4b, c). song structure for 4 h after night-sleep and then injected 3 mg of plasticity and
during melatonin27. All birds fell asleep within 15 min, and slept for 2–3 h plasticity migh
Sleep inertia?
during 4 ^ 0.5 h of vocal rest. When singing resumed we observed
song deterioration in all birds, comparable to that observed after
model through
structure.
Circadian changes?
night-sleep (Fig. 5e, Wilcoxon, n.s.). Finally, it is
such as simula
Vocal changes during sleep could relate to synaptic21 or cellu-
lar36,37 remodelling that might occur during sleep. Regardless of the parameters. Th
come from me
Lack of practice?
specific mechanism, our results indicate the involvement of an
active process, perhaps neural song-replay during sleep. How might ness of steel thr
the bird’s ‘internal singing’26,27 during sleep give rise to the observed reheating and c
plasticity? If the lack of auditory feedback during replay during sleep
Sleep-related activity? Methods
has a similar effect to that of abolishing32,33,38 or perturbing39
Animal care
auditory feedback, we would expect to see drifts in song structure
e.g.,
after sleep replay. We analysed post-deafening deterioration in eight
All experiments wer
Health and have bee
adult birds and found that the deterioration of intra-syllabic
Neural song replay without auditory of the C
Committee
temporal structure was similar to that observed after night-sleep
in young birds (Fig. 6). Therefore, song replay along with lack of
feedback Experimental des
auditory feedback could by itself explain the decrease in song
We used 50 zebra fi
breeding colony. Co
previously17. All bir
Recordings from
colony of the Wesle
parents and siblings
together in groups o
Experimental grou
Training from day 4
playbacks17, starting
per song model). Bi
12 h:12 h LD.
Training from da
day 60 with the thre
Training from da
from day 90 with the
in acoustic isolation
Figure 5 Vocal changes after song prevention and sleep manipulation. a, Recovery of
Post-sleep Post-deafening
Figure 6 Comparison of post-sleep deterioration and post-deafening deterioration.
a, Post-sleep changes in birds trained from day 43. b, Post-deafening changes (2–4
No training: six
from day 30 to day
Singing prevent
song structure could be circadian (scenario 1) or it might require singing activity (scenario weeks after deafening). Each slice indicates the mean relative contribution of each feature 12 h:12 h LD (days 0
-sleep 2). b, Vocal changes (in reference to the developmental trend) were measured after night- to the overall effect. The sign indicates whether the feature value increased or decreased. day 90 (five birds pe
sleep plus 2 h of song prevention (yellow bar, mean ^ s.e.m.). Results were similar to Var, variance. bird (with its cage)
2010 10 4 8
9. Conclusions and Discussions
• Sleep-related oscillation in developmental learning
• Post-sleep deterioration vs. morning intense practice
• Oscillation in behavioral performance
• Bumblebees, juvenile rats, newborn babies
• Not observed in many other procedural and declarative learning
skills
• Optimization algorithm using non-monotonic
parameter trajectories
e.g., Simulated tempering: Heating vs. reheating
2010 10 4 9
10. Further Reading
Vol 458 | 5 March 2009 | doi:10.1038/nature07615
LETTERS
Sleep and sensorimotor integration during early vocal
learning in a songbird
Sylvan S. Shank1 & Daniel Margoliash1,2
Behavioural studies widely implicate sleep in memory consolida- Within each bird there was some variation in the amount of high-
tion in the learning of a broad range of behaviours1–4. During frequency activity of RA cells on nights after the onset of song learn-
sleep, brain regions are reactivated5,6, and specific patterns of ing, but the tendency towards shorter ISIs was apparent in most cells
neural activity are replayed7–10, consistent with patterns observed (Fig. 1c).
in previous waking behaviour. Birdsong learning is a paradigmatic Emerging RA bursting activity, furthermore, was shaped by the
model system for skill learning11–14. Song development in juvenile specific tutor song that a bird heard. Nightly mean ISI distributions
zebra finches (Taeniopygia guttata) is characterized by sleep- were calculated for all RA neurons recorded for each bird after tutor
dependent circadian fluctuations in singing behaviour, with song exposure (which showed little difference from night to night;
immediate post-sleep deterioration in song structure followed Fig. 2c), and nightly mean distributions were averaged together to
by recovery later in the day15. In sleeping adult birds, spontaneous generate one mean curve per bird. For the resulting post-exposure
bursting activity of forebrain premotor neurons in the robust curves, within the high-frequency range (ISIs #40 ms), the shapes—
nucleus of the arcopallium (RA) carries information about day- as assessed using Pearson correlation coefficients—were more sim-
time singing16. Here we show that, in juvenile zebra finches, play- ilar in birds hearing the same tutor song than in birds hearing dif-
back during the day of an adult ‘tutor’ song induced profound and ferent tutor songs (see Supplementary Information). This grouping
tutor-song-specific changes in bursting activity of RA neurons of ISI distribution shapes by tutor song can be visualized by compar-
during the following night of sleep. The night-time neuronal ing the average ISI distributions for individual birds on nights before
changes preceded tutor-song-induced changes in singing, first tutor song exposure to those on nights after tutor song exposure
observed the following day. Interruption of auditory feedback (Fig. 2a, black and coloured lines, respectively). The differences
greatly reduced sleep bursting and prevented the tutor-song- between groups can be visualized by comparing global average ISI
specific neuronal remodelling. Thus, night-time neuronal activity distributions—one for each group of birds hearing a given tutor song
is shaped by the interaction of the song model (sensory template) (Fig. 2b).
and auditory feedback, with changes in night-time activity preced- Once a bird was exposed to a tutor song, a prototypical post-
ing the onset of practice associated with vocal learning. We hypo- exposure ISI distribution shape was quickly obtained and then main-
thesize that night-time bursting induces adaptive changes in tained. To quantify this, we compared (using Pearson correlations)
premotor networks during sleep as part of vocal learning. By this the nightly ISI distributions (#40 ms) for each bird before and after
hypothesis, adaptive changes driven by replay of sensory informa- song exposure to the corresponding global mean curve (Fig. 2b),
tion at night and by evaluation of sensory feedback during the day excluding data from the bird being analysed from the global mean
interact to produce the complex circadian patterns seen early in distributions. Before tutor song exposure, both the within- and
vocal development. between-group comparisons (Fig. 2d, black and grey dots, respect-
To explore the role of sleep in the early phases of song learning, we ively) had large variability and were not significantly different from
2010 10 4 characterized the properties of single RA neurons in head-fixed, each other on any night (P 5 0.25 to P 5 0.73). By the first night after 10