Intrinsic motivation for singing in songbirds is enhanced by temporary singing suppression and regulated by dopamine

Behaviors driven by intrinsic motivation are critical for development and optimization of physical and brain functions, but their underlying mechanisms are not well studied due to the complexity and autonomy of the behavior. Songbirds, such as zebra finches, offer a unique opportunity to study neural substrates of intrinsic motivation because they spontaneously produce many renditions of songs with highly-quantifiable structure for vocal practice, even in the absence of apparent recipients (“undirected singing”). Neural substrates underlying intrinsic motivation for undirected singing are still poorly understood partly because singing motivation cannot be easily manipulated due to its autonomy. Also, undirected singing itself acts as an internal reward, which could increase singing motivation, leading to difficulty in measuring singing motivation independent of singing-associated reward. Here, we report a simple procedure to easily manipulate and quantify intrinsic motivation for undirected singing independent of singing-associated reward. We demonstrate that intrinsic motivation for undirected singing is dramatically enhanced by temporary suppression of singing behavior and the degree of enhancement depends on the duration of suppression. Moreover, by examining latencies to the first song following singing suppression as a measure of singing motivation independent of singing-associated reward, we demonstrate that intrinsic singing motivation is critically regulated by dopamine through D2 receptors. These results provide a simple experimental tool to manipulate and measure the intrinsic motivation for undirected singing and illustrate the importance of zebra finches as a model system to study the neural basis of intrinsically-motivated behaviors.


Results
Long-term suppression of undirected singing increases singing motivation. In young adult male zebra finches (87-119 days post-hatch [dph]), we examined how temporary suppression of undirected singing affects singing motivation by assessing singing behavior immediately after the singing suppression period. Undirected singing was suppressed by turning off the light in the sound recording chambers, during the daytime, for a short (30 min) and a long (5 h) period (separated by a 2.5-h light period; Fig. 1B). To eliminate potential influences of circadian rhythm, the order of the 30-min and 5-h lights-out (LO) periods was switched every 1-3 days (individual birds received both 30-min and 5-h LO 7-27 times [median = 16.0]). No song production was observed during the LO periods regardless of their duration.
Interestingly, when we compared singing immediately after the offset of 30-min LO and 5-h LO periods, we found that birds sang much sooner and more intensely after 5-h LO ( Fig. 1C-F, Supplementary Fig. 1). On average, instantaneous singing rates (calculated using 2-min time bins) rapidly increased, to almost double baseline singing rates, only after 5-h LO, and then gradually returned to baseline levels ~ 1 h after LO offset (Fig. 1C-D). Also, the latency to the first song after end of LO, which we refer to as "first song latency, " was significantly shorter after 5-h LO compared to 30-min LO in 6 out of 7 birds ( Fig. 1E; p < 0.05 for each bird; p = 0.016 for group data [n = 7 birds, W = 28], Wilcoxon signed-rank test). Moreover, the mean singing rates measured over a 30-min period starting at the first song following each LO period (green bars in Fig. 1C), which we refer to as "initial singing rates, " were significantly higher after the 5-h LO than after the 30-min LO in 5 out of 7 birds ( Fig. 1F; p = 0.016 for group data; W = 0). Because birds with higher singing motivation should show shorter first song  30-min LO (blue area) and 5-h LO (red area) were given with a 2.5-h intervening light period, followed by a 4-h light period. The order of 30-min LO and 5-h LO was switched every 1-3 days. Each row indicates the schedule on one day. (C) Raster plot of song bouts produced before and after 30-min LO (top) and 5-h LO (bottom) and corresponding singing rate histograms (bin size is 2 min) in a representative bird. Green horizontal lines indicate the 30-min periods in which the initial singing rates were measured for the 1st experimental day (see "Methods"). Note that the bird started singing sooner and with higher rates after 5-h LO compared to after 30-min LO. (D) Time course of instantaneous singing rate before and after 30-min (blue) and 5-h LO (red), normalized to the mean singing rate before LO (mean ± SEM, n = 7 birds). (E) The first song latencies after 5-h LO and those after 30-min LO in the bird shown in (C) (top) and in all the birds examined (n = 7 birds, bottom). Gray lines on top indicate data from single days, and the black line indicates their mean ± SEM. Circles on bottom indicate mean data from individual birds and bars and whiskers indicate mean + /− SEM across birds; filled circles indicate statistical significance for individual birds (p < 0.05). Arrow indicates the bird shown in the top. As a group, first song latencies were significantly shorter after 5-h LO compared to 30-min LO (*p = 0.016). (F) Initial singing rates after 5-h LO and those after 30-min LO in the bird shown in (C) (top) and in all the birds examined (n = 7 birds, bottom). Conventions are same as in (E). Initial singing rates were significantly greater after 5-h LO compared to 30-min LO (*p = 0.016).   (Fig. 2B, E). These results demonstrate that intrinsic motivation for undirected singing can be easily manipulated over a wide range simply by changing the duration of LO periods, providing a useful tool for studying the neural mechanisms of singing motivation in songbirds.
Singing motivation can be enhanced by singing suppression even under light conditions. The amount of singing in most songbirds is also regulated by circadian rhythms. Birds often sing a lot early in the morning after a long dark period 31 . Given that we suppressed singing by turning off the lights, we cannot exclude the possibility that a prolonged dark condition, and not song suppression, influences singing motivation and subsequent singing. To test this possibility, we suppressed singing under light conditions by physically interfering with singing posture using a method previously reported 32 (see "Methods"). Singing was suppressed for short (10-s) and long (5-h) time periods with a schedule similar to that in the LO experiments described in Fig. 1B (individual birds received both 10-s and 5-h singing obstruction (SO) 16-18 times [median = 17.5]; Fig. 3A). We observed gradual decreases in first song latencies over days after both 10-s and 5-h SO periods (Fig. 3B), presumably reflecting habituation of the birds to the SO procedure. Nevertheless, there was a strong trend of enhanced singing motivation after long suppression compared to short suppression just like the LO experiments: for individual experimental days, first song latencies tended to be shorter after 5-h SO than after 10-s SO (Fig. 3C, D top). As a group, first song latencies were significantly shorter after 5-h SO than after 10-s SO (Fig. 3D bottom; n = 6 birds; p = 0.03, W = 21, Wilcoxon signed-rank test). Likewise, initial singing rates were significantly greater after 5-h SO than after 10-s SO ( Fig. 3E; n = 6 birds; p = 0.03, W = 0). These results indicate that singing motivation can be enhanced by singing suppression even under light conditions, thus confirming that singing suppression, and not dark conditions per se, is critical for enhancing the intrinsic motivation for undirected singing.
Suppression-dependent enhancement of singing motivation depends on age. All the experiments described so far were done with relatively young adult zebra finches (87-119 dph). To examine age dependence of the suppression-induced enhancement of singing motivation, we suppressed singing in older birds (219-897 dph) using the LO procedure described in Fig 4D; p = 1, W = 8 for the group data). In fact, the extent to which singing motivation was enhanced after 5-h LO as compared to 30-min LO, as measured by relative change in mean first song latencies (see "Methods"), was strongly correlated with age ( Fig. 4E; n = 10 birds, r = 0.94, p = 3.8 × 10 -5 ).  www.nature.com/scientificreports/ These results show that the enhancement of singing motivation after song suppression was strongly dependent on the age of the bird: younger the bird, greater the enhancement and older the bird, lesser the enhancement. Previous studies have shown that the daily amount of undirected song is relatively high in young birds and gradually decreases with age, even from young adults to old adults 7,8 , suggesting an age-dependent decline in overall motivation for undirected singing. We hypothesized that such an age-dependent decline of overall singing motivation contributed to the observed age dependence of singing motivation enhancement after longterm singing suppression. To test this hypothesis, we measured baseline singing rate (mean singing rates over 1-h periods immediately before LO periods) in both young and old adult birds as a measure of baseline singing motivation and compared it with age and with the magnitude of singing motivation enhancement. In accord with a previous study showing an age-dependent decrease in daily singing amounts with age ranges similar to those of our birds 7 , baseline singing rates immediately before LO periods gradually decreased with age (Fig. 4F, n = 10 birds, r = − 0.84, p = 2.3 × 10 -3 ). We found that these baseline singing rates are positively correlated with the magnitude of singing motivation enhancement caused by singing suppression (Fig. 4G, n = 10 birds, r = 0.73, p = 0.017): younger adult birds exhibited higher baseline singing rates and greater increases in singing motivation after singing suppression compared to older birds. These results are consistent with the hypothesis that the age-dependent decline of overall singing motivation contributes to the age dependence of singing motivation enhancement caused by long-term singing suppression.
Effects of dopamine and opioid receptor antagonists on intrinsic motivation for undirected singing, quantified as first song latencies. Previous studies have demonstrated the regulation of undirected singing by at least two neuromodulators, dopamine and opioids. Singing rates were significantly correlated with levels of dopamine-or opioid-related signals 16,[21][22][23][24][25][26] . Although these correlative results suggest the involvement of dopamine and opioids in undirected singing, whether those neuromodulators critically contribute to the motivation to sing (a state of "wanting") and/or to intrinsic reward associated with the act of singing (a state of "liking") remains unclear, as singing rates analyzed in previous studies are closely linked to both motivation and reward processes. For example, an increase in singing rate could reflect an increased level of singing motivation and/or an increase in singing-associate reward. It is difficult to tell these two apart by just measuring singing rates. However, our measure of first song latency is more likely to reflect singing motivation, as it is unlikely to be influenced directly by any processes following the act of singing, such as singing-associated reward, especially when singing is suppressed for a relatively long time. Taking advantage of this measure, we assessed the direct contributions of dopamine and opioids to undirected singing motivation (independent of singing-associated reward) by administering antagonists of these neuromodulators.
We systemically injected antagonists of dopamine or opioid receptors at the end of 5-h LO periods (30 min before the offset of LO periods) and assessed their effects on first song latencies during the subsequent light period (7-h duration; Fig. 5A). Doses of the drugs were based on previous studies using the same drugs in songbirds and chickens 16,[33][34][35] . Compared with vehicle injections, we found that low (0.2 mg/kg) and high (1 mg/kg) doses of a dopamine D1 receptor antagonist, SCH23390, moderately but significantly increased the first song latencies after 5-h LO ( Fig. 5B; n = 9 birds, p = 0.004, W = 0 for both low and high doses, Wilcoxon signed-rank test with a Holm-Bonferroni correction for multiple comparisons, corrected significance level α = 0.0083; effect size [Hedges' g] = 1.69 for low dose and 0.67 for high dose). In contrast with SCH23390 injections, injections of a dopamine D2 receptor antagonist, haloperidol, dramatically increased the first song latencies in a dose-dependent manner; first song latencies were much longer with a higher dose (1 mg/kg) but not lower dose (0.2 mg/kg) when compared with vehicle injections (Fig. 5C; n = 8 birds, p = 0.95, α = 0.05, W = 27, effect size = 0.11 for lower dose; n = 9, p = 0.008, corrected α = 0.0125, W = 0, effect size = 2.16 for higher dose). In addition, 3 out of 9 birds with the higher dose haloperidol injections did not produce any songs during the post-LO periods (7-h duration), whereas all birds with vehicle injections produced songs during the same period (dashed line in Fig. 5C bottom). These results suggest that dopamine plays a critical role in regulating intrinsic motivation for undirected singing through D2 receptors. Although D1 receptors also appear to be involved in undirected singing motivation, given the relatively small effect of SCH23390 and its potential influence on D2 receptor signaling 36 , further studies are needed to determine the contribution of D1 receptors (see also Discussion). We also tested the possibility that these dopamine receptor antagonists affect general motor behavior by examining the number of hopping and flying over a 30-min period following 5-h LO, but found no significant effects for either drug at any dose (Supplemental Fig. 2). Thus, the increasing effects of these drugs on first song latencies are not simply due to decreases in general motor behavior.
In contrast to the dopamine D2 receptor antagonist haloperidol, the general opioid receptor antagonist naloxone did not have strong effects on first song latencies. Although there appear to be a trend of longer first song latencies with naloxone injections (both the low [2 mg/kg] and high [10 mg/kg] doses) compared to those with vehicle injections (effect size = 1.03 and 0.99 for low and high doses, respectively), the differences were not statistically significant (Fig. 5D; n = 9 birds, p = 0.04, corrected α = 0.0167 and W = 5 for lower dose; p = 0.07, corrected α = 0.025 and W = 7 for higher dose). Naloxone did not significantly affect non-vocal behaviors over a 30-min period following 5-h LO either ( Supplementary Fig. 2), as reported in a previous study that used similar doses and experimental paradigms 16 . Effects of dopamine and opioid receptor antagonists on initial singing rates. We also examined the effects of dopamine and opioid receptor antagonists on the initial singing rate, which, unlike the first song latency, is likely to reflect not only intrinsic singing motivation but also singing-associated reward. We found that neither low (0.2 mg/kg) nor high (1 mg/kg) dose of SCH23390 had significant effects on initial singing rates ( Fig. 6A; n = 9 birds, p = 0.3, corrected α = 0.0125 and W = 13.5 for lower dose; n = 9 birds, p = 1.0, corrected , sharply contrasting with their increasing effects on the first song latency (Fig. 5B). Because singing rates directly reflect the degree of singing motivation (and probably indirectly reflect singing-associated reward), these results suggest that singing motivation that has been suppressed by this drug mostly recovers by the time the birds produce the first song following the LO period. Given that the drug and the vehicle were injected 30 min prior to the offset of the LO period and that many birds resumed singing ~ 20 min after the LO offset (see Fig. 5A and B), it is likely that the drug effect to suppress singing motivation mostly wears off during this period. In contrast, the high dose (0.2 mg/kg), but not the low dose (1 mg/kg), of haloperidol significantly decreased the initial singing rate ( Fig. 6B; n = 9 birds, p = 0.03, corrected α = 0.01 and W = 41 for lower dose; n = 9 birds, p = 0.004, corrected α = 0.008 and W = 45 for higher dose; if no songs were produced during the 7-h, post-LO periods, the data of the initial singing rate was assigned zero). This inhibitory effect of the high dose of www.nature.com/scientificreports/ haloperidol on initial singing rate is not surprising given its long-lasting inhibitory effect on singing motivation reflected by markedly-prolonged first song latencies (Fig. 5C): it is likely that singing motivation is still partly suppressed even after the birds resumed singing following LO periods, resulting in reduced initial singing rates. We also found that neither the low (2 mg/kg) nor high (10 mg/kg) dose of naloxone had significant effects on initial singing rates ( Fig. 6C; n = 9 birds, p = 0.9, corrected α = 0.025 and W = 24 for lower dose; n = 9 birds, p = 0.7, corrected α = 0.017 and W = 19 for higher dose). Although these results appear to be inconsistent with a previous study showing dose-dependent inhibitory effects of similar doses of naloxone on undirected singing 16 , this discrepancy can also be explained by the time gap between the drug injection and the measurement of singing rates in our experiments as explained for our SCH23390 experiments above. This interpretation is supported by the fact that naloxone crosses the blood brain barrier and exits the brain rapidly 37 .

Discussion
In the present study, we highlight the advantage of undirected singing in zebra finches as a model system to study intrinsic motivation for a complex, learned behavior. We found that temporary suppression of spontaneous undirected singing, either by turning off the ambient light or by physically obstructing singing, dramatically increased intrinsic motivation for singing as quantified by singing latencies and frequencies. This increase in motivation depended on the duration of singing suppression: longer the suppression of singing, sooner and more intense was the singing after release from suppression. We also found that suppression-induced enhancement of singing motivation depended on age, presumably due to an age-dependent decline of baseline singing motivation. Finally, we revealed that singing motivation is critically regulated by dopamine through D2 receptors. Taken together, our findings provide a simple and useful experimental tool to manipulate and measure singing motivation independent of singing-associated reward and offer novel insights into the mechanisms underlying intrinsic motivation for vocal practice in songbirds. www.nature.com/scientificreports/ Spontaneous and intense singing of songbirds in the absence of apparent recipients, such as undirected singing of zebra finches, has long been of immense interest to biologists as the function of undirected singing remains unclear. Accumulating evidence in zebra finches and Bengalese finches indicates that undirected singing serves, at least in part, as vocal practice by which birds develop and optimize song structures 12,[38][39][40][41] . Because such vocal practice would ultimately increase the effectiveness of song during future interaction with a mate, maintaining a high level of motivation for undirected singing to routinely optimize song structure appears to be critical for reproduction success. Our findings of enhanced singing motivation after long periods of singing suppression may be explained by this model: immediate and intense singing following long periods of singing suppression may compensate for the loss of vocal practice during singing suppression and enable birds to quickly and thoroughly re-evaluate and re-optimize song structure to prepare for future courtship activity. Future research should examine whether daily undirected singing is critically required for the maintenance of song structure by suppressing undirected singing for multiple days and assessing its effect on the detailed structure of song. Additionally, similar mechanisms could account for the observed age-dependent declines of overall singing rate and of singing motivation enhancement caused by singing suppression. Although young adult zebra finches maintain and optimize song structure through vocal practice, song plasticity gradually declines with age 10,11,42 . Given the reduced plasticity of song structure in older adult birds, it is possible that less vocal practice is sufficient to maintain song structure even after relatively long singing suppression. Consistent with this line of thinking, birds may decrease undirected singing motivation and daily song amount with age by constantly evaluating how much vocal practice is required to maintain song structure and by adaptively adjusting singing motivation in order to decrease energy consumption and/or the risk of predation caused by undirected singing. Alternatively or in addition, because the act of undirected singing is rewarding 13 , lower singing rates resulting from lower singing motivation in old birds may simply be caused by age-dependent decline of intrinsic reward associated with undirected singing which could be genetically encoded in the brain.
Enhancement of singing motivation depending on duration of singing suppression is reminiscent of the socalled "Lorentz' psychohydraulic model" 43 , a classic model of animal motivation. In this model, a fluid representing action-specific energy (i.e. motivational drive) builds up in a reservoir over time if the behavior is not being executed, leading to the eventual opening of a valve of the reservoir depending on external factors that modulate the valve opening; the flow of action-specific energy out of the reservoir into a bucket underneath represents the execution of the behavior with fixed action patterns. Although this model is over-simplified in comparison with the modern view of motivation, suppression-induced enhancement of undirected singing motivation that we observed nicely fits the main concept of this model: the fluid in the reservoir represents motivation for undirected singing and accumulation of the fluid by long-term suppression of singing results in immediate and robust expression of singing behavior when the suppression was released; the external factor represents ambient light in our LO experiments. Thus, our results underscore the importance of undirected singing as a tractable model system for studying the fundamental mechanisms of animal motivation.
Our results of pharmacological manipulation of dopamine or opioid signaling provide a significant advance in our understanding of the mechanisms underlying the spontaneous production of undirected song. Previous studies have demonstrated that both dopamine and opioids are associated with the production of undirected song 16,[21][22][23][24][25][26]33 , and opioids have recently been demonstrated to contribute to the process of singing-associated regard (a state of "liking") 27 . However, it was unclear whether dopamine contributes to the motivation to sing (a state of "wanting") and/or singing-associated reward. Using a measure "first song latency, " which quantifies the degree of singing motivation independent of singing-associated reward, we revealed that dopamine contributes to the regulation of singing motivation through D2 receptors. Although our results of naloxone treatment are not clear and do not allow for a strong conclusion regarding the contribution of opioids to singing motivation, the critical contribution of opioids to singing-associated reward shown by a recent study 27 and the contribution of dopamine to singing motivation shown by our current results highlight striking parallels with the mechanisms of reward-associated behaviors in mammals. A large body of research in mammals indicates that dopamine primarily underlies anticipatory, motivated state whereas opioids underlies hedonic pleasure or reward (for review, 4,[18][19][20]. Future studies should identify which neural circuits are modulated by dopamine to regulate singing motivation. Previous studies demonstrated that dopamine-related signals are correlated with the production of undirected singing in several brain areas including Area X, the medial preoptic area (mPOA), the ventral tegmental area (VTA), and septum 21,22,26,44 , but it remains unclear whether those areas are involved in singing motivation. Although dopaminergic inputs from VTA to a song-specialized basal ganglia nucleus Area X are activated during undirected singing 21,44 , they are unlikely to underlie singing motivation based on the following findings: (1) lesions of the VTA-Area X projections or of Area X itself do not abolish undirected singing 45-50 ; (2) neither stimulation nor inhibition of VTA-Area X axon terminals alters undirected singing rate 51 ; and (3) the VTA-Area X projections fire during, but not before, singing to encode performance error regarding ongoing song quality [51][52][53] , which is similar to reward prediction error in mammalian VTA dopamine neurons observed during reward-seeking tasks 54 . Independent of reward-dependent firing of VTA dopaminergic neurons, gradual increases in extracellular dopamine levels have recently been observed in mammalian striatum when animals approach a reward site 30,55 , suggesting a critical role for slow time scale dopamine releases in motivation for reward-seeking behavior. Similar mechanisms might be involved in regulating intrinsic motivation for undirected singing in songbirds.
Relative contributions of different subtypes of dopamine receptors to undirected singing motivation is also needed to be determined. Although our systemic injections of a D1 receptor antagonist SCH23390 significantly increased the first song latencies (Fig. 5B), given that SCH23390 can also block GIRK channels 56 , which are downstream targets of dopamine D2 receptor signaling 36 , there is a possibility that the increases in the first song latencies following SCH23390 injections are an off-target effect and not directly related to D1 receptor signaling. www.nature.com/scientificreports/ Testing the effects of other types of D1 receptor antagonist such as SCH39166 as well as D1 receptor agonists would be needed to tease out the relative contributions of different dopamine receptors to singing motivation. Intrinsic motivation has long been a major topic of interest in human psychology, but intrinsically motivated behavior was first acknowledged in the study of animal behavior 4,57 . Since then, a variety of behaviors have been investigated across a range of species, including social play behavior in rodents 58,59 . Nevertheless, neural mechanisms underlying intrinsic motivation are still largely unclear, in part due to variability and complexity of behaviors that prevent us from examining a direct link between behavior and underlying neural activity. Undirected singing in zebra finches, investigated in the present study, is a complex but highly stereotyped and quantifiable motor skill that is learned and maintained by relatively simple neural circuits specialized for singing. Given the tractable nature of zebra finch song, as well as accumulating knowledge about neural mechanisms of song production and learning, undirected singing in zebra finches provides an excellent model system to study neural circuit mechanisms of intrinsic motivation. Thus, our simple procedure to manipulate intrinsic motivation for undirected singing and the new insights into the neuromodulatory system for regulating singing motivation provide an important first step toward understanding more detailed neural mechanisms underlying intrinsic motivation for complex, learned motor behaviors.

Methods
This study was carried out in compliance with the ARRIVE guidelines (http:// www. nc3rs. org. uk/ page. asp? id= 1357).  (16-bit). Recorded data were down-sampled to a sampling rate of 32-kHz. Recording was triggered if the program detected five consecutive sound notes, each of which was defined based on sound magnitude, duration, and intervening gap duration. Recording ended if a silent period lasted longer than 0.5 s (i.e. each song file contains a single "song bout" that is separated from other bouts by > 0.5-s silent periods). Songs were recorded throughout the day, and all song recordings were of undirected song (i.e. no female was present). Birds with sufficient singing rates (> 300 song bouts per day) were used for our experiments.

Subjects.
Singing suppression. Singing was suppressed by turning off the light in the sound-attenuating chambers using digital timers in most experiments. The duration and schedule of LO periods varied depending on experimental paradigms (from 30 min to 10 h; see Results). In a subset of birds, we suppressed singing by attaching a detachable weight (17-20 g) on the birds' necks 32 . This procedure suppresses singing by preventing birds from taking the singing posture without affecting their daily behaviors. The weight was usually supported by the floor and not carried by the bird's neck and therefore birds continue all their daily behaviors, such as drinking, eating, grooming, and calling. The weights were attached either for 5 h or only transiently (for ~ 10 s). No birds produced any song motifs during either LO periods or periods with the weight attached. Song analysis. The first song latency was measured as the time interval from the offset of a singing-suppression period to the onset of the first song recorded. We visually inspected spectrograms of the sound files recorded after the singing-suppression periods to find the first file that included at least one song motif. To quantify singing motivation enhancement using this measure, first song latencies across all LO periods with the same durations (30 min or 5 h) were averaged and percent differences between them were calculated as follows: where Latency 30-min and Latency 5-hr are mean first song latencies after 30-min LO and after 5-h, respectively. This measure will be positive when Latency 5-hr is shorter than Latency 30-min , and vice versa.
To measure singing rates during the periods before and after LO, we screened all sound files recorded during those periods to exclude non-song files using a semi-automated method. Song motifs of adult zebra finches have highly stereotyped temporal structure, which is clearly distinct from that of other sounds such as calls, introductory notes, and cage noises. We, therefore, sorted song files (sound files that include at least one full motif of song) and non-song files by focusing on the temporal structure of two acoustic features, sound amplitude and Weiner entropy and by comparing them between a canonical song motif and all sound files as follows. Temporal trajectories of those features were calculated using Sound Analysis Tools for Matlab 60 for all sound files examined. The canonical song motif was made by averaging amplitude envelopes or entropy trajectories of the most stereotyped part of 10 randomly selected motifs, and the cross-correlation function was calculated between the canonical motif and all sound files; because temporal structure of song motifs varies slightly across renditions, the canonical motif was allowed to undergo ± 10% proportional changes in their temporal pattern. We then plotted www.nature.com/scientificreports/ the maximum correlation coefficients (mCCs) of amplitude envelopes against mCCs of entropy trajectories for each bird (Supplementary Fig. 3). In these plots, most song files formed a clear cluster around the high mCCs, whereas non-song files were scattered around the low mCCs area. Our preliminary inspections of a subset of files showed that most non-song files had low mCCs in both amplitude envelope and envelope trajectory. Therefore, we set thresholds of mCCs in both features and labeled files with mCCs below the thresholds as non-song files. Finally, we visually inspected spectrograms of non-song files and discarded those that did not include any song motifs; files that included song motifs were put back to the song dataset to be analyzed. The initial singing rate was measured as the mean singing rate over a 30-min period starting at the onset of the first song produced after an LO period (the timing of the 30-min period varied across trials depending on the first song latencies). Instantaneous singing rate was measured by counting the occurrence of song bouts (i.e. the number of song files) over 2-min bins and averaged across trials. Baseline singing rates were measured as mean rates over 1-h periods immediately before the LO periods.
In the experiments with four different LO durations, if songs were not produced during the post LO period, the first song latency was assigned a time of 2 h (120 min) and the initial singing rate was assigned zero. Likewise, in the experiments with drug injections (Fig. 5), if birds did not produce any songs post-LO periods (7-h duration), the first song latency was assigned a time of 7 h (420 min) and the initial singing rate was assigned zero.

Drug injections.
For pharmacological manipulations of dopamine or opioid signaling, dopamine or opioid antagonists or the corresponding vehicle was injected into the pectoral muscle at 30 min prior to the offset of LO periods once a day (see Fig. 5A). Injected drugs and their doses were as follows: the dopamine D1 receptor antagonist R(+)-SCH23390 (Millipore Sigma, D054) dissolved in 0.9% saline (0.2 and 1.0 mg/kg); the dopamine D2 receptor antagonist haloperidol (Millipore Sigma, H1512) stored as stock solution in DMSO at − 20 °C and diluted in 0.9% saline before injection (0.2 and 1.0 mg/kg); the opioid receptor antagonist naloxone hydrochloride dihydrate (Millipore Sigma, N7758) dissolved in 0.9% saline (2 and 10 mg/kg). Doses were selected based on literature in songbirds and chickens 16,[33][34][35] . Four birds were tested with lower doses of all three drugs above; other 4 birds were tested with higher doses of those drugs; other 5 birds were tested with both lower and higher doses of those drugs. For each drug at each dose (and corresponding vehicle), injection was made at least twice in each bird, and the results (first song latencies and initial singing rates after 5-h LO) were averaged across injections; the same number of vehicle injections were made. Multiple injections of the same drug were made with inter-injection intervals > 4 days to prevent possible desensitization to the drug. Each drug-injection day was followed by a washout day with no injections. To assess the effect of drugs on general motor behavior, individual birds were videotaped for 30 min immediately after the offset of 5-h LO periods, and the number of hopping and flying was counted by an observer who was blind to the treatment (drug or vehicle) given to the birds.

Statistical analysis.
To analyze the effect of singing suppression on subsequent singing behavior, we compared first song latencies and initial singing rates between 30-min LO and 5-h LO for each bird using a Wilcoxon signed-rank test (α = 0.05); we used a Wilcoxon signed-rank test for group data as well. We examined the effects of singing suppression, with 4 different durations, on singing behavior using one-way ANOVA. To examine the effects of drug administrations, we compared birds' behaviors (first song latencies, initial singing rates, and general motor behavior) between those after drug administrations and those after vehicle administrations using a Wilcoxon signed-rank test with a Holm-Bonferroni correction, in which the significance threshold (alpha) for rejecting the null hypothesis varies depending on the p-value based rank of individual comparisons and thus is not determined a priori 61 . All statistical analyses were performed using Matlab (RRID: SCR_001622).