1. Neuroscience
Download icon

Transitioning between preparatory and precisely sequenced neuronal activity in production of a skilled behavior

  1. Vamsi K Daliparthi
  2. Ryosuke O Tachibana
  3. Brenton G Cooper
  4. Richard HR Hahnloser
  5. Satoshi Kojima
  6. Samuel J Sober
  7. Todd F Roberts  Is a corresponding author
  1. UT Southwestern Medical Center, United States
  2. The University of Tokyo, Japan
  3. University of Zurich/ETH Zurich, Switzerland
  4. Texas Christian University, United States
  5. Neuroscience Center Zurich (ZNZ), Switzerland
  6. Korea Brain Research Institute, Republic of Korea
  7. Emory University, United States
Research Article
  • Cited 0
  • Views 985
  • Annotations
Cite this article as: eLife 2019;8:e43732 doi: 10.7554/eLife.43732

Abstract

Precise neural sequences are associated with the production of well-learned skilled behaviors. Yet, how neural sequences arise in the brain remains unclear. In songbirds, premotor projection neurons in the cortical song nucleus HVC are necessary for producing learned song and exhibit precise sequential activity during singing. Using cell-type specific calcium imaging we identify populations of HVC premotor neurons associated with the beginning and ending of singing-related neural sequences. We characterize neurons that bookend singing-related sequences and neuronal populations that transition from sparse preparatory activity prior to song to precise neural sequences during singing. Recordings from downstream premotor neurons or the respiratory system suggest that pre-song activity may be involved in motor preparation to sing. These findings reveal population mechanisms associated with moving from non-vocal to vocal behavioral states and suggest that precise neural sequences begin and end as part of orchestrated activity across functionally diverse populations of cortical premotor neurons.

https://doi.org/10.7554/eLife.43732.001

Introduction

The sequential activation of neurons is implicated in a wide variety of behaviors, ranging from episodic memory encoding and sensory processing to the voluntary production of skilled motor behaviors (Fee et al., 2004; Fiete et al., 2010; Hahnloser et al., 2002; Li et al., 2015; Lynch et al., 2016; Markowitz et al., 2015; Okubo et al., 2015; Peters et al., 2014; Rajan et al., 2016; Svoboda and Li, 2018). Neural sequences develop through experience and have been described in several brain areas, including the motor cortex, hippocampus, cerebellum, and the basal ganglia (Barnes et al., 2005; Dragoi and Buzsáki, 2006; Foster and Wilson, 2006; Harvey et al., 2012; Jin et al., 2009; Li et al., 2015; Luczak et al., 2007; Mauk and Buonomano, 2004; Peters et al., 2014; Pfeiffer and Foster, 2013; Pfeiffer and Foster, 2015; Schwartz and Moran, 1999). Although computational models provide important insights into circuit architectures capable of sustaining sequenced activity (Churchland et al., 2010b; Fiete et al., 2004; Fiete et al., 2010; Haga and Fukai, 2018; Harvey et al., 2012; Kumar et al., 2010; Rajan et al., 2016), our understanding of sequence initiation and termination is still limited.

The precise neural sequences associated with birdsong may provide a useful biological model for examining this issue. Premotor projection neurons in the cortical vocal region HVC (HVCRA neurons, see Figure 1 legend for anatomical abbreviations) exhibit precise sequential activity during song and current evidence suggests that this activity is acutely necessary for song production (Hahnloser et al., 2002; Kozhevnikov and Fee, 2007; Long and Fee, 2008; Long et al., 2010; Scharff et al., 2000). HVCRA neurons are thought to only be active during vocal production in waking adult birds, yet ~50% of recorded HVCRA neurons do not exhibit any activity during singing (Hahnloser et al., 2002; Hamaguchi et al., 2016; Kozhevnikov and Fee, 2007; Long et al., 2010; Lynch et al., 2016), leaving the function of much of the HVCRA circuitry unresolved.

Figure 1 with 1 supplement see all
Population imaging of song-related HVCRA sequences.

(a) Diagram showing three distinct projection neuron targets of the vocal premotor nucleus HVC. The projection neurons connecting HVC to the downstream motor nucleus RA (HVCRA neurons) are shown in green. Auditory region Avalanche (Av), robust nucleus of the arcopallium (RA), nucleus HVC of the nidopallium (HVC). (b) Schematic showing HVCRA neuron somata (green) and their outputs (magenta) to the downstream motor nucleus RA. AAV9-Flex-CAG-GCaMP6s was injected into HVC (green syringe) and AAV9-CAG-Cre was injected into RA (magenta syringe) to selectively label HVCRA neurons. (c) In vivo two-photon maximum density projection of retrogradely labeled HVCRA neurons expressing GCaMP6s. Scale bar = 100 µm. (d) A cross-section of HVC showing GCaMP6s-labeled HVCRA neurons (green). The dashed line indicates where the cranial window was made over HVC. Note the lack of labeling in the region directly ventral of HVC, known as the HVC shelf. Scale bar = 100 µm. (e) A sagittal section of HVC showing HVCRA neurons (green) and retrogradely labeled HVCX neurons (red). The dashed line indicates the border between HVC and HVC shelf. Scale bar = 50 µm. (f) Whisker and scatter plots of soma diameters of GCaMP6s expressing cells show that retrogradely labeled HVCRA neurons (green) have smaller diameters than neurons labeled using only direct viral injections (AAV9-CAG-GCaMP6s) into HVC (mixed population neurons, black). Boxes depict 25th and 75th percentiles, whiskers depict SD. HVCRA: N = 21 neurons; mean diameter = 14.0 ± 3.8 µm (SD); Mixed: N = 52 neurons; mean = 26.9 ± 4.9 µm; t = −10.7, p=2.0×10−16, two-sample t test. (g) Example calcium traces from 2 HVCRA neurons in a bird that sang five consecutive motifs. Shown are the background-subtracted traces (black) and the inferred calcium traces (green). The magenta overlays indicate the rise time (intervals between onset and peak times) of the recorded calcium transients. The horizontal dashed line (gray) denotes 3 SD above baseline activity. The bars above the spectrogram denote cage noise associated with birds hopping or flapping their wings (yellow) or production of song motifs (red). (h) Motif-related activity of 36 HVCRA neurons across five motifs. Each row shows activity of a neuron from one trial. The dashed magenta lines separate different neurons. Empty spaces indicate trials wherein neurons were not active (no event, NE). The inset shows a zoom-in of activity from three separate HVCRA neurons.

https://doi.org/10.7554/eLife.43732.002

Neuronal activity related to motor planning and preparation has been associated with accurate production of volitional motor movements (Churchland et al., 2010a; Svoboda and Li, 2018) but is still poorly described in the context of initiating precise neural sequences for motor behaviors, like those exhibited in HVCRA neurons. Although it is not known if HVCRA neurons exhibt activity related to motor planning and preparation, previous studies have identified anticipatory or preparatory activity in other classes of HVC neurons and in other regions of the songbird brain (Goldberg et al., 2010; Goldberg and Fee, 2012; Kao et al., 2008; Keller and Hahnloser, 2009; Rajan, 2018; Roberts et al., 2017). HVC contains interneurons and at least three classes of projection neurons, including neurons projecting to the striatopallidal region Area X (HVCX), neurons projecting to a portion of the auditory cortex termed Avalanche (HVCAv), and the aforementioned HVCRA neurons that encode precise premotor sequences necessary for song production (Akutagawa and Konishi, 2010; Mooney and Prather, 2005; Roberts et al., 2017). Multi-unit recordings from HVC, which are typically dominated by the activity of interneurons, show increases in activity tens to hundreds of milliseconds prior to singing (Crandall et al., 2007; Day et al., 2009; Rajan, 2018). Calcium imaging from HVCAv neurons and electrophysiological recordings from HVCX neurons indicate that they also become active immediately prior to song onset (Rajan, 2018; Roberts et al., 2017). These data are consistent with recordings from the downstream targets of HVCAv and HVCX neurons. Portions of the auditory cortex (Keller and Hahnloser, 2009) and the basal ganglia pathway involved in song learning show changes in activity immediately prior to singing (Goldberg et al., 2010; Goldberg and Fee, 2012; Kao et al., 2008). Given this background, and that ~50% of HVCRA neurons may not exhibit any activity during singing (Hamaguchi et al., 2016; Long et al., 2010), we sought to examine if the precise neural sequences associated with song arise as part of larger changes in activity among populations of HVCRA neurons.

To examine the neural circuit activity associated with the initiation and termination of singing, we imaged from populations of HVCRA neurons in freely singing birds. We show that ~ 50% of HVCRA neurons are active during periods associated with preparation to sing and recovery from singing and that their activity presages the volitional production of song by 2–3 s. One population of HVCRA neurons is only active immediately preceding and following song production, but not during either singing or non-vocal behaviors. A second population of neurons exhibits ramping activity before and after singing and can also participate in precise neural sequences during song performance. Recordings from downstream neurons in the motor cortical nucleus RA reveal neural activity prior to song initiation and following song termination. The control of respiratory timing is essential for song (Schmidt and Goller, 2016), and our measurements of respiratory activity suggest that one function of pre-singing activity in HVCRA neurons is to coordinate changes in respiration necessary for song initiation. From these findings, we reason that subpopulations of HVCRA neurons are involved in motor planning and motor preparation, encoding the neural antecedents of song that drive recurrent pathways through the brainstem to prepare the motor periphery for song production.

Results

Activity sequences in populations of HVCRAneurons

We used miniscope calcium imaging to examine the activity of populations of HVCRA neurons in singing zebra finches (Chen et al., 2013; Ghosh et al., 2011). A total of 223 HVCRA neurons were imaged during production of 1298 song syllables from six birds (30 song phrases across 18 imaging trials, Supplementary file 1). To selectively target HVCRA neurons, we combined retrograde viral expression of cre recombinase from bilateral injections into RA with viral expression of cre-dependent GCaMP6s from injections into HVC (Figure 1a–b and legend, Figure 1—figure supplement 1; see Materials and methods) (Chen et al., 2013). We confirmed the identity of imaged neurons using conventional retrograde tracing, anatomical measures of neuronal features, and post-hoc histological verification. Although this did not label all RA projecting neurons in HVC, we found that this approach exclusively and uniformly labeled populations of HVCRA neurons (Figure 1c–f).

To elicit courtship singing, we presented male birds with a female and imaged HVCRA neurons during song performance (Video 1, see Materials and methods for definitions song). On average, birds engaged in singing in 24 s (±49 s) of a female bird being presented. Given the slow decay times of calcium signals relative to singing behavior, we defined neuronal activity by the rise times of calcium events that were >3 standard deviations (SD) above baseline (Figure 1g, average rise time: 0.112 ± 0.047 s SD, see Materials and methods). The activity of individual HVCRA neurons was time-locked to a moment in the birds’ song, with different neurons active at different moments in the song motif (Figure 1g–h; onset jitter = 55.0 ± 60.9 ms, populations imaged at 30 frames per second). We found that the sequential activity of HVCRA neurons roughly coded for all moments in the song motif (Figure 1h). These results provide the first glimpse of activity across populations of identified HVCRA neurons during singing and support the idea that sparse and precise neuronal sequences underlie the sequential structure of birdsong (Amador et al., 2013; Hahnloser et al., 2002; Long et al., 2010; Lynch et al., 2016; Picardo et al., 2016).

Video 1
Synchronized video of calcium imaging and behavior in a bird singing to a female.
https://doi.org/10.7554/eLife.43732.005

Peri-Song activity in populations of HVCRAneurons

The sequence of syllables in zebra finch song is stereotyped and unfolds in less than a second. Like other rapid and precise motor movements, song may benefit from motor planning and preparatory activity unfolding on much longer timescales than the synaptic delays associated with descending motor commands, which in zebra finches are estimated to be ~25–50 ms (Amador et al., 2013; Fee et al., 2004); however, HVCRA neurons have been hypothesized to represent temporal sequences for songs and calls, and to remain quiescent at other times (Hahnloser et al., 2002; Hamaguchi et al., 2016; Kozhevnikov and Fee, 2007; Long et al., 2010). We examined the activity of HVCRA neurons prior to song onset, in between song bouts, and immediately after singing (Figure 2a–c, Figure 2—figure supplement 1). Song onset was defined by the onset of introductory notes that preceeded the bird’s song phrase. A song phrase was defined as one or more repetitions of a bird’s motif with less than 2 s between onset of the next song bout (see Materials and methods for detailed definitions of phrases, song, and peri-song behavior). For any given song phrase, we observed significant activity surrounding singing behavior, indicating that HVCRA neuronal activity is not restricted only to the production of precise neural sequences for song in waking birds. 30 out of 30 song phrases from six birds exhibited ‘peri-song’ activity, defined as the 5 s intervals before and after singing, plus silent gaps within song phrases. Slightly fewer HVCRA neurons were active during peri-song intervals (54.3% of neurons) than during singing (59.9%), (t = −1.1, p=0.29 two-sample t test; Figure 2b). Of the neurons that were active during peri-song intervals, 40.6% were active prior to song onset, 17.8% were active following song offset, and 41.6% were active before and after song (N = 197 neurons, six birds, Figure 2—figure supplement 2). When we examined the timing of peri-song events, we found no correlation in the timing of pre-song and post-song event onsets in neurons that were sparsely active both before and after song (Figure 2—figure supplement 3). Although neuronal populations displayed considerably more calcium events during song (t = 5.635, p=5.65×10−7 two-sample t, normalized song event rate = 24.09 events/s±15.95 SD, normalized peri-song event rate = 6.97 events/s±3.72 SD), a substantial fraction of all recorded calcium events occurred within the 5 s intervals before or after song (997/2366 or 32.5% of all calcium events).

Figure 2 with 11 supplements see all
Most HVCRA neurons exhibit peri-song activity.

(a) Normalized calcium transients from 67 simultaneously recorded HVCRA neurons during production of a song phrase. The red dashed lines delimit five consecutive motifs. (b) Percentages of active neurons during peri-song (54.3 ± 17.8%, SD, purple) and song (59.9 ± 22.5%, gray) are similar (30 phrases, t = −1.1, p=0.29 paired two-sample t test). Box plots show the median, 25th and 75th percentiles with whiskers showing ±1.5 interquartile range (IQR). (c) Sample neurons with diverse phrase indices ranging from −1 to 1 and their corresponding calcium traces during 6 motifs over three bouts. Dashed lines indicate bout onsets and offsets. Bars above spectrogram indicate the presence of cage noise related to hopping and wing flapping (yellow) or female calling (FC, orange). (d) Histogram of phrase indices for all 223 neurons from six birds. (e) Undirected song from a different male showing periods of cage noise or hopping behavior (yellow) and feeding behavior (green). Blue boxes indicate the male calling. Red dashed lines indicate onsets and offsets of song bouts.

https://doi.org/10.7554/eLife.43732.006

To better characterize the activity profiles of HVCRA neurons, we indexed the song and peri-song activity of all HVCRA neurons throughout a day of singing (phrase index: range −1 to +1, with neurons exclusively active outside of singing scoring −1 and neurons active only during singing +1, Figure 2c–d, see Materials and methods). We found that HVCRA phrase indices were not uniformly distributed (χ2 (7, N = 223)=46.3, p=7.6×10−8, Chi-square goodness of fit test), with a significant fraction (36%) falling at the extremes of this scale (Figure 2d), and that neurons with different phrase indices were anatomically intermingled throughout HVC (Figure 2—figure supplement 4). Most neurons (64.1%) displayed sparse heterogenous activity during peri-song periods and transitioned to temporally precise activity during singing (referred to here as ‘pan-song neurons’; Figure 2—figure supplement 5; Supplementary file 2). At the extremes of the phrase index scale (phrase indices −1 or +1), we found that 18.4% of neurons were exclusively active during peri-song intervals (phrase index = −1, referred to here as ‘peri-song neurons’; Supplementary file 2), while 17.5% participated exclusively in neural sequences during singing (phrase index =+1, referred to here as ‘song neurons’; Supplementary file 2).

These results reveal that a substantial portion of HVCRA neurons are active outside of the precise neuronal sequences associated with song, expanding our view of the potential functional role of this neuronal population. That more than half of all HVCRA neurons can be active during peri-song intervals raises the prospect that precise neural sequences emerge as part of changing network dynamics across subpopulations of HVCRA neurons. These results may also lend insight into why approximately half of HVCRA neurons recorded using electrophysiological methods appear to be inactive during song (Fee et al., 2004; Hamaguchi et al., 2016; Kozhevnikov and Fee, 2007; Long et al., 2010). Previous multi-unit recordings in zebra finches and mockingbirds, which are dominated by activity of interneurons or neurons projecting to the basal ganglia, have identified ‘anticipatory’ activity in HVC hundreds of milliseconds prior to song onset, but the role of this activity and whether HVCRA neurons are active prior to song onset have not been examined (McCasland, 1987; Rajan, 2018; Rajan and Doupe, 2013).

These results also raise questions as to why previous studies in adult zebra finches have not identified peri-song activity in HVCRA neurons. Although speculative, several reasons may account for this. First, previous calcium imaging experiments have not restricted GCaMP expression to HVCRA neurons, but rather relied on either non-selective labeling of neuronal populations in HVC (Katlowitz et al., 2018; Liberti et al., 2016; Markowitz et al., 2015; Picardo et al., 2016) or have been restricted to imaging small populations of other classes of HVC neurons (Roberts et al., 2017). Second, electrophysiological studies of identified HVCRA neurons have been mostly confined to recording one neuron at a time. These experiments have focused on understanding coding during song production and at least initial studies used short (~500 ms) buffering windows triggered by singing behavior (Hahnloser et al., 2002). Therefore, sparse heterogenous activity occurring seconds before or after song could be simply overlooked or could appear irrelevant unless viewed through the lens of population dynamics. Third, it is possible that there are incongruencies between spiking activity and calcium signals, as has been shown in the feed-forward inhibitory control of bursting activity in the striatum (Owen et al., 2018). Although several studies have examined the relationship between calcium responses and suprathreshold activity in HVC projection neurons (Graber et al., 2013; Peh et al., 2015; Picardo et al., 2016), this relationship has not been tested during peri-song periods in freely singing birds.

It is also possible that peri-song activity is unrelated to singing and merely reflects low levels of spontaneous activity intrinsic to HVCRA neurons. This was not the case, however, as HVCRA neurons were largely inactive outside of peri-song intervals and were significantly more active during the peri-song periods than baseline (baseline calculated from periods ≥ 10 s removed from periods of singing or calling, p=8.4×10−5, Chi-square = 18.78 Friedman test, baseline = 12.9% ± 5.7 SD of fluorescence values normalized to song, pre-song = 26.9% ± 14.7, and post-song = 30.2% ± 10.7). In addition, we examined the amplitudes of peri-song calcium events and found that they were larger than events occurring during song (t = 3.2769, p=0.0012, two-tailed t test, 279 fluorescence peaks measured from neurons active during both peri-song and song, pan-song neurons with phrase indices between −0.18 to 0.18). We also asked whether peri-song activity might relate to factors other than singing. We examined trials in which birds did not sing to female birds but did not find HVCRA neurons that responded solely to presentation of the female or during non-song related movements of the head, beak, or throat, such as during eating, grooming, and seed-shelling (Figure 2e, Figure 2—figure supplement 6, Figure 2—figure supplement 7, Figure 2—figure supplement 8). Indeed, populations of HVCRA neurons only became substantially active prior to singing or calling. Moreover, in a single bird in which we were able to image neuronal activity during undirected singing, that is song produced when the bird was alone in its cage, we also found peri-song activity in the moments before and after (Figure 2—figure supplements 69), suggesting that peri-song activity is unlikely to be solely associated with extraneous nonvocal singing behaviors such as courtship dance.

We next examined the possibility that HVCRA neurons play a role in motor planning or preparation as birds prepare to sing. We found pre-song activity in 28/30 song phrases analyzed. In the two instances when we did not detect any pre-song activity, less than 6 HVCRA neurons were active within our imaging window during singing, indicating that the lack of activity was likely the result of under sampling from the population. Electrophysiological recordings in young zebra finches have identified HVC neurons that mark the onset of song bouts (Okubo et al., 2015), ‘bout neurons’ that burst immediately prior to vocalizations. The vast majority of pre-song activity we describe occurs hundreds of milliseconds to seconds prior to vocalizations, suggesting a role in planning or preparation to vocalize (96.2% of calcium events occurred more than 100 ms prior to vocal onset and 65.6% occurred more than 1 s prior to vocal onset). Zebra finches often sing a variable number of introductory notes prior to the first motif of a song bout. Pre-song activity (prior to introductory notes) could be related to the number of introductory notes to be sung, but we found no correlation between pre-song event rates and the number of introductory notes (Figure 2—figure supplement 10). In 27/28 song phrases we found increases in population activity greater than 3 SDs above baseline predicted song onset within the following 4–5 s (2.44 s ± 1.0 s SD, 28 song phrases from five birds). Calcium activity reached two-thirds of the maximum pre-song activity only prior to song onset or prior to short vocalizations (Figure 2—figure supplement 11). Together, these results indicate that HVCRA neuron activity is predictive of the voluntary production of courtship song and suggests a role for this network in motor planning and in preparation to sing.

Peri-Song and Pan-Song neurons

A substantial fraction of all imaged neurons (41/223 neurons) were active exclusively before or after singing (Figure 2d, neurons with a phrase index of −1). These peri-song neurons exhibited sparse heterogeneous activity before song phrases and were occasionally active in the silent intervals between song motifs (Figure 3a–c, and Figure 3—figure supplement 1). Both the number of active peri-song neurons and the density of calcium events increased 1–3 s prior to song onset, with the event rate peaking 1.5 s prior to singing (Figure 3b–c,i) and then declining sharply in the last second before song onset (Figure 3i). Most of the neurons we imaged, pan-song neurons (143/223 neurons), exhibited sparse, heterogeneous activity before and/or after song and exhibited time-locked sequences during singing (Figure 3d–f). Pan-song neurons exhibited substantial increases in their activity in the last ~2 s prior to song onset. Their activity continued to increase as the activity of peri-song neurons began to ramp-off prior to song onset (Figure 3g,I, and Figure 3—figure supplement 2). Differences in pre-song activity profiles between peri-song and pan-song neurons (Kolmogorov-Smirnov, K-S test, k = 0.26, p=0.056) may support temporally coordinated network transitions as birds prepare to sing, suggesting that neural sequences for song could emerge as part of changing network dynamics in HVC.

Figure 3 with 3 supplements see all
Description of peri-song and pan-song neuron activity.

(a) 21 peri-song neurons from one bird singing 3 bouts containing six motifs (1 st bout: three motifs; 2nd bout: two motifs; 3rd bout: one motif, the dashed red lines indicate the onset and offset of the three bouts). Each row shows song-aligned calcium events (CEs N = 54 CEs; average rise time = 0.18 ± 0.09 s SD). The shaded horizontal bars separate different neurons. One CE is seen to overlap with the beginning of the song phrase. The onset time for this event is 170 ms before song, but the rise time is slow and extends to 100 ms after song onset. Below the CE raster plot is a peri-event histogram with the event rate in 200 ms bins shown for the trial above. (b) Song-aligned CEs in peri-song neurons 5 s before and after phrase onset (41 neurons, 190 CEs). The activity rate peaks ~ 1.2 s before phrase onset. (c) The number of active peri-song neurons in 200 ms bins before and after phrase onset (41 neurons, 169 CEs). (d) Song-aligned activity of pan-song neurons (same trial as shown in panel a), N = 29 neurons, 253 CEs). (e) Peri-event histogram of pan-song neurons (143 neurons, 1,333 CEs). (f) The number of active pan-song neurons in 200 ms bins (143 neurons, 853 CEs). (g) Pre-song event rate for all neurons. The event rate was calculated by counting event onsets in 100 ms bins and then smoothed with a 1 s moving window. 28 trials are shown from five birds; the black line indicates the average event rate. The black triangles mark the peak event rate occurring 0.6 s before song onset and 2.5 s when the event rate reaches 3 SD above the baseline event rate, respectively. Baseline event rate was determined by measuring the average event rate from −5 to −4 s before song onset. Shaded region indicates standard deviation. (h) Post-song event rate for all neurons. 27 trials are shown from five birds. The black triangles mark when the event rate reaches 3 SD above the baseline event rate. Baseline event rate was determined by measuring the average event rate during −5 to −4 s before song onset. (i) Pre-song event rate for peri-song and pan-song neurons calculated as calcium events in moving 1 s windows. The black line indicates average event rate. The black triangle indicates peak event rate occurring 1.5 s before phrase onset. (j) Same as i), but post-song event rates for peri-song and pan-song neurons. The black triangle indicates peak event rate occurring 0.5 s after phrase offset.

https://doi.org/10.7554/eLife.43732.020

During production of song motifs, pan-song neurons exhibited sequential bursts of activity that roughly coded for all moments in the bird’s song, similar to sequencing previously described in song neurons. We found that pan-song and song neurons had a similar probability of being active within each motif (song neurons probability of at least one calcium event per motif P(motif)=0.72 ± 0.32, pan-song neurons P(motif)=0.66 ± 0.29, K-S test, p=0.11; Figure 3—figure supplement 3); however, we also noted that the probability of being active was lower than in electrophysiological recordings (Hahnloser et al., 2002; Kozhevnikov and Fee, 2007). This likely reflects limitations in event detection using single-photon calcium imaging. To better understand this, we calculated signal to noise ratios (SNRs, signal defined as peak fluorescence of calcium events during song) between pan-song neurons and song neurons during singing (SNR song neurons = 913.8 ± 405.4 (7 neurons), pan-song neurons = 638.3 ± 248.2 SEM, n = 36 neurons). We found no difference in SNR between these neurons (two-sample t test, p=0.6; SNR calculated from 156 calcium events (28 from song neurons and 128 from pan-song neurons), suggesting that although we are underestimating activity during singing, these limitations are unlikely to obscure differences between pan-song and song neurons.

In addition to preceding song onset, neurons also marked the end of song phrases. Within 5 s after song offset, peri-song neurons exhibited a sharp increase in activity followed by a gradual ramp-off (Figure 3a–c,h,j, and Figure 3—figure supplement 2) whereas pan-song neurons only exhibited a ramp-off (Figure 3d–f,h,j). Although the distribution of post-song activity differed between peri-song and pan-song neurons (K-S test, k = 0.2692, p=0.0373), both populations returned to baseline activity over similar timescales. The function of post-song activity is unclear but may provide a circuit mechanism for birds to rapidly re-engage in song performances given appropriate social feedback or context. Courtship singing is tightly coupled to social interaction with female birds and it is common for male birds to string two or more song phrases together during courtship song (Williams, 2004) (see Figure 2—figure supplements 1 and 6). Post-song activity could also reflect moments when birds are unable to continue singing due to hyperventilation induced by the rapid respiratory patterns associated with song (Franz and Goller, 2003). To explore this idea, we examined whether the duration of song phrases was correlated with the number of active neurons in the post-song period, but did not find a significant correlation (r2 = 0.03). Although the function of post-song activity is unclear, our results indicate that pre-song activity forecasts impending song and suggest a previously unappreciated role for the HVCRA network in planning or preparing to sing.

Common preparatory activity in premotor circuits across multiple species

To examine whether preparatory activity is a common circuit mechanism for the production of birdsong, we recorded HVC and RA neural activity in another songbird species, Bengalese finches (Lonchura striata domestica). Motor planning and preparation facilitate the accurate execution of fast and precise movements, which are common to the songs of zebra finches and Bengalese finches, however, syllable sequences in Bengalese finches are less stereotyped than those in zebra finches (Okanoya, 2004). Using multi-channel neural recordings in HVC, we identified robust preparatory activity several hundreds of milliseconds prior to song onset (Figure 4a–g). The pre-song and song–related multiunit spike rates were significantly above baseline (Wilcoxon signed-rank test after Bonferroni correction, n = 29 MU sites. pre-song: z = 4.62, p=0.030; song: z = 4.70, p<0.001), whereas the post-song spike rates was not (z = 2.17, p=0.089). Pre-song activity increased above baseline −1.47 ± 1.05 s prior to song onset, a timescale that closely matched the timing of peak calcium-event rates in peri-song neurons in zebra finches. The offset timing of post-song activity was 0.42 ± 0.23 s. This indicates that preparatory activity in HVC is a common network motif important for song generation and the onset of precise neural sequences.

Pre-song and post-song firing in HVC of Bengalese finches.

(a) Schematic of recording site. (b) An example of song initiation and termination (dashed lines indicate phrase onset and offset, pre and post-song period marked in purple and song marked in red) and simultaneously recorded HVC multiunit activity on two electrodes (channels #4 and #15, bird p15o56). Green raster plots represent detected spikes on the two electrodes. (c,d) Phrase onset (c) and offset (d) related average multiunit activity obtained from an example electrode channel (vertical dashed lines indicate phrase onset and offset, respectively). The spike rate was averaged across multiple song onsets or offsets and was first calculated in 10 ms bins (gray thin line) and then smoothed with a 500 ms window (bold line). Upper and lower horizontal dotted lines show mean spike rates during singing and baseline, respectively. Arrowheads indicate onset (c) and offset (d) timings of spike rate, as assessed by crossing of a pre-defined threshold (red line). (e,f) Normalized multiunit activity related to phrase onset (e) and offset (f). Before averaging, the spike rate trace of each electrode channel was normalized such that 0 corresponds to the mean rate during baseline and 1.0 to the mean rate during singing (see Materials and method). The bold line shows an average across all electrodes and birds (n = 29 channels). Purple area indicates ±1.0 SD. Arrowheads show mean onset (e) and offset (f) timing of pre-song and post-song activity, respectively. (g) Mean multiunit spike rates during spontaneous (black), pre-song (purple with diamonds), song (gray with diamonds), and post-song (purple with diamonds) periods. Pre-song and post-song periods are indicated by the horizontal bars in panels e and f. Box plots show the median, 25th and 75th percentiles with whiskers showing ±1.5 IQR.

https://doi.org/10.7554/eLife.43732.028

HVC contains multiple cell types, including interneurons and at least three different classes of projection neurons (Mooney and Prather, 2005; Roberts et al., 2017). Multichannel recordings in HVC provide an important read-out of the network activity prior to song onset but alone are insufficient to assess whether this preparatory activity influences descending cortical pathways involved in song motor control. Therefore, we recorded single unit electrophysiological activity from the downstream targets of HVCRA neurons within the cortical premotor nucleus RA. Projection neurons in RA are tonically active at baseline (Figure 5a, ‘Spontaneous’) and exhibit precise bursts of activity during singing (Figure 5a, ‘Song’), a transition well captured by changes in the coefficient of variation of the inter-spike intervals (CVISI, Figure 5c). We measured changes in RA neuron activity in the period just prior to song initiation and just after the conclusion of each song bout (between 0.5 and 2.5 s before/after the first/last song syllable). As expected, we did not find substantial differences in spike rates between non-singing and singing states (not shown) but found that the CVISI for RA neurons changed significantly during song, reaching higher values during pre-song, song, and post-song epochs as compared to spontaneous activity (Figure 5d and Figure 5—figure supplement 1, p<0.005, two-sided K-S tests). This suggests that HVCRA sequences associated with preparation to sing propagate to downstream premotor circuits prior to song onset and that HVC continues to influence descending motor pathways following song cessation.

Figure 5 with 1 supplement see all
Pre-song and post-song firing in RA of Bengalese finches.

(a) Example extracellular recording from a single RA neuron. Colored lines highlight four epochs (pre-song (purple), song (red), and post-song (purple)) relative to the beginning and ending of a song phrase (see main text). Gray areas indicate discontinuities in time (pauses between ‘spontaneous’ epoch and song initiation and within the middle portion of the song bout). (b) Schematic of recording site. (c) We quantified inter-spike-intervals (ISIs) and computed the coefficient of variation (CV) in each epoch. (d) We found significantly higher ISI variability in the pre-song epoch (purple with diamonds) compared to spontaneous (p<0.005, two-sided K-S test). Box plots show the median, 25th and 75th percentiles with whiskers showing ±1.5 IQR.

https://doi.org/10.7554/eLife.43732.030

Peripheral preparation to sing

Pre-song activity could reflect motor planning (changes in network activity independent of changes in the motor periphery) and/or motor preparation that functions to coordinate changes in the motor periphery as birds prepare to sing. Song is a respiratory behavior that is primarily produced during expiration and silent intervals in the song correspond to mini-breaths, which are rapid, deep inspirations (Hartley and Suthers, 1989; Schmidt and Goller, 2016). How birds plan to sing or prepare the respiratory system to sing is poorly understood, but there is evidence that prior to song onset, oxygen consumption decreases and respiratory rate increases (Franz and Goller, 2003). To explore the time course of changes in respiratory patterns in more detail, we used air sac pressure recordings in singing zebra finches (Figure 6a–c). During singing, birds significantly accelerated the respiratory rhythm and marginally shifted towards longer periods of expiration during each cycle (Figure 6b, respiratory duration pre-song = 0.38 ± 0.04 s (SD), song = 0.18 ± 0.03 s, post-song = 0.39 ± 0.06 s: F(2,10) = 46.63, p<0.001; duty cycle (% of time in expiration) pre-song = 58% ± 3 (SEM), song = 57% ± 2.5, post-song = 61% ± 2.4: F(2,10) = 3.56, p=0.07). We found that significant changes in respiration also preceded song onset. Respiratory cycle duration significantly accelerated in the last second prior to song onset with relative decreases of expiratory phases (respiratory duration: F(3,15) = 7.67, p=0.02, duty cycle: F(3,15) = 4.077, p=0.07; Figure 6d). Following song termination, birds immediately returned to longer respiratory cycles but during the first second post-song, they spent more time exhaling compared to inhaling, a behavior likely involved in helping to recover from singing-related hyperventilation (respiratory duration: F(2,10) = 0.509, p<n .s., duty cycle: F(2,10) = 6.553, p<0. 01; Figure 6e). These changes in respiration during the last second before and first second following song support the idea that HVCRA neurons provide descending motor commands that coordinate transitions between non-vocal and vocal states by coordinating respiratory patterns. The lack of changes at earlier time-points prior to singing also indicate that pre-song activity 1–3 s prior to song onset may reflect motor planning or the decision to sing, rather than respiratory preparation. Together these, findings support the idea that HVCRA neurons could function in aspects of motor planning as well as preparation.

Air sac pressure recording in zebra finches.

(a) Waveform of pressure changes during non-singing and singing periods. Waveforms above the horizontal line (suprambient pressurization) indicate expiration and below the line (subatmospheric pressurization) indicate inhalation. Song start was identified by the presence of introductory notes preceeding the song phrase. M1-3 corresponds to three repetitions of the bird’s motif. Inset illustrates measurements for respiratory cycle duration and duty cycle (% time in expiration) and the first two introductory notes. (b) Respiratory cycle duration and (c) duty cycle of expiratory phase before (Pre), during (Song), and after (Post) song production (N = 6 birds). (d) Plots of average respiratory cycle durations and (e) duty cycles during pre-song and post song periods (N = 6 birds). Longer duty cycles correspond to increased periods of expiration. Data in panels b-e is derived from the same six birds.

https://doi.org/10.7554/eLife.43732.033

Discussion

Previous studies suggested that the HVCRA network functions as a time-keeper, encoding motif-level temporal representations of song via propagation of precisely timed neural sequences (Hahnloser et al., 2002; Kozhevnikov and Fee, 2007; Long and Fee, 2008; Long et al., 2010; Lynch et al., 2016; Markowitz et al., 2015; Picardo et al., 2016). Central to this view is that HVCRA neurons are active during singing and hence behave in primarily two modes, inactive or propagating neural sequences. Our principal result is that neural sequences among HVCRA neurons emerge as part of orchestrated population activity across a larger network of HVCRA neurons. This activity can be correlated with motor planning and preparation prior to song initiation. Peri-song and pan-song HVCRA neurons forecast the start of singing. Peri-song HVCRA neurons are inactive during song, whereas pan-song neurons become heterogeneously active prior to time-locked sequential activity during song performances (Figure 2 and 3). Moreover, we find that preparatory activity in HVCRA neurons precedes the pre-bout activity described in other classes of HVC neurons and other portions of the song system previously described in zebra finches (Danish et al., 2017; Goldberg et al., 2010; Goldberg and Fee, 2012; Kao et al., 2008; Rajan, 2018; Roberts et al., 2017; Vyssotski et al., 2016; Williams and Vicario, 1993) and described here in Bengalese finches (Figure 4). This suggests that HVCRA neurons may seed network wide changes among other classes of HVC neurons and the song system more generally.

Indeed, the rigid stereotypy of singing behavior enables comparisons from different levels of the nervous system and periphery. We find that preparatory activity in HVCRA neurons drives descending motor commands via RA and motor movements that set the stage for producing song (Figure 5 and 6). Because song is a self-initiated, volitional behavior, our findings further indicate that the HVCRA network either functions as a sensitive read-out of the decision to sing or as an integral factor in the decision itself. Finally, previous studies have shown that about half of HVCRA neurons are inactive during song (Hahnloser et al., 2002; Hamaguchi et al., 2016; Kozhevnikov and Fee, 2007; Long et al., 2010). We find that approximately half of the HVCRA neuronal network is active at peri-song intervals, perhaps accounting for previous recordings from neurons that are inactive during singing. We therefore propose that one important function of HVCRA neurons is to plan and prepare for the upcoming song performance.

Together, these findings support a simple model for song and neural sequence initiation. Preparatory activity in populations of HVCRA neurons drives descending motor commands via RA and its connections to the ventral respiratory group and syringeal motoneurons in the medulla (Andalman et al., 2011; Goller and Cooper, 2004; Roberts et al., 2008; Sturdy et al., 2003; Suthers et al., 1999; Wild, 1993). Given the sparsity of HVCRA neuron activity and the convergence of HVCRA input to RA, it is likely that only population activity, like that described here, is sufficient to drive bursting in RA. These motor signals increase respiratory rate, bringing it closer to the high rate needed to coordinate production of song syllables. Because the initiation of singing requires precise coordination between respiratory state and descending motor commands, we hypothesize that recurrent projections from the brainstem update activity in HVC and trigger the initiation of neural sequences once the periphery is readied for the first respiratory cycle for song (Ashmore et al., 2005; Hamaguchi et al., 2016; Schmidt et al., 2012). Circuitry related to recurrent projections into HVC increase their activity tens of milliseconds prior to song onset (Danish et al., 2017; Vyssotski et al., 2016; Williams and Vicario, 1993) and thus may provide final inputs that help mediate the transition from heterogenous preparatory activity to the precisely sequenced activity that underlies song.

Absent from this model is how activity in peri-song and pan-song neurons is first initiated seconds before song onset. HVC receives input from cholinergic neurons in the basal forebrain (Li and Sakaguchi, 1997; Shea et al., 2010; Shea and Margoliash, 2003), noradrenergic neurons in the locus coeruleus (Appeltants et al., 2000), and dopaminergic neurons in the midbrain (Hamaguchi and Mooney, 2012; Tanaka et al., 2018), any or all of which could play potent roles in shifting the excitability of subsets of HVCRA involved in song preparation and initiation. Neuromodulatory inputs may also play a role in shifting the excitability of the entire song system as birds prepare to sing. Further studies will be needed to understand the function of neuromodulatory inputs to HVCRA neurons and the rest of the song circuitry in song initiation.

Also absent from our model are specific predictions about the role peri-song neurons and pan-song neurons may play in motor planning. Although it is not yet known whether activity occuring among peri-song and pan-song HVCRA neurons prior to song onset is necessary for song initiation or production, previous studies have identified pyramidal tract neurons (PTupper) in the anterior lateral motor cortex that play a specific role in motor planning (Economo et al., 2018). These neurons are active prior to movement onset and show decreased activity during movement, similar to the activity of HVCRA peri-song neurons described here. PTupper neurons innervate the thalamus rather than medulla and are hypothesized to encode cognitive signals related to motor planning (Economo et al., 2018). Likewise, RA provides inputs to the medulla and the thalamus (Goldberg and Fee, 2012; Roberts et al., 2008; Vates et al., 1997; Wild, 1993). However, it is currently not known if subpopulations of RA neurons project exclusively to the thalamus or if peri-song and pan-song HVCRA neurons have unique downstream targets within RA. Future experiments, requiring novel closed-loop manipulations to exclusively disrupt the activity of functionally identified peri-song or pan-song neurons, independent of activity associated with the motif-level temporal representations of song, and detailed circuit mapping studies will be needed to dissect the function of these newly identified subpopuations of HVCRA neurons. When possible, such experiments will undoubtedly lend insights into whether subpopulations of HVCRA neurons are involved in aspects of motor planning independent of motor preparation or song performance.

Neural activity associated with motor planning and preparation has been observed in motor and premotor cortices for a variety of different motor tasks in rodents and primates (Chen et al., 2017; Churchland et al., 2010a; Churchland et al., 2006a; Churchland et al., 2006b; Economo et al., 2018; Inagaki et al., 2019; Kaufman et al., 2014; Li et al., 2015; Li et al., 2016; Murakami and Mainen, 2015; Murakami et al., 2014; Tanji and Evarts, 1976), including vocalizations (Gavrilov et al., 2017). This activity is thought to reflect the decision to perform movements and is characterized by a high degree of variability from trial to trial. Changes in circuit dynamics function to shift the initial state of a network to levels that enable efficient and accurate motor performances and this activity can start to unfold seconds prior to movement initiation (Churchland et al., 2010a; Inagaki et al., 2019; Murakami and Mainen, 2015; Svoboda and Li, 2018). HVC is proposed to be analogous to the mammalian motor cortex (layer III neurons of the primary motor cortex) (Pfenning et al., 2014), or premotor cortex (Bolhuis et al., 2010). Our observation of preparatory activity seconds before the onset of courtship song in two different songbird species suggests that pre-movement activity is a common mechanism for ensuring the accurate production of volitional behaviors. In line with recordings of pre-motor activity in mammals, individual HVCRA neurons exhibit a high degree of trial to trial variability. Although zebra finch courtship song is famously stereotyped, there is a measurable degree of variability in the structure and duration of song motifs, bouts and phrases from trial to trial. Recording of preparatory activity across larger populations of HVCRA neurons may be useful for decoding this trial to trial variability in song structure, ultimately providing a predictive readout of impending behaviors.

Sequential activation of neurons is thought to provide computational advantages for encoding temporal information associated with episodic memories or behavioral sequences (Fiete et al., 2004; Kumar et al., 2010; Rajan et al., 2016). Neural sequences in HVC provide one of the cleanest examples linking brain activity with a naturally learned and volitionally produced skilled motor behavior (Fee et al., 2004). Our study provides a glimpse of how these sequences emerge through temporally coordinated transitions within a potentially hierarchically organized network and suggests a general framework for initiating the production of skilled motor behaviors.

Materials and methods

Key resources table
Reagent type
(species) or resource
DesignationSource or referenceIdentifiersAdditional information
Strain, strain background (adeno-associated virus)AAV9.CMV.HI.eGFP-Cre.WPRE.SV40James M. WilsonAddgene viral prep # 105545-AAV9; http://n2t.net/addgene:105545; RRID:Addgene_105545
Strain, strain background (adeno-associated virus)AAV9.CAG.Flex.GCaMP6s.WPRE.SV40Chen et al., 2013Addgene viral prep # 100842-AAV9; http://n2t.net/addgene:100842; RRID:Addgene_100842
Strain, strain background (adeno-associated virus)AAV9.CAG.GCaMP6s.WPRE.SV40Chen et al., 2013Addgene viral prep # 100844-AAV9; http://n2t.net/addgene:100844; RRID:Addgene_100844
Commercial assay or kitMiniature MicroscopeInscopixhttps://www.inscopix.com/nvista
Software, algorithmMatlabMathworkshttp://www.mathworks.com/products/matlab/; RRID:SCR_001622
Software, algorithmCalcium AnalysisPeters et al., 2014
Software, algorithmCNMFPnevmatikakis et al., 2016https://github.com/epnev/ca_source_extraction

Software and data availability

All custom analysis codes and calcium imaging data are publically available as a Github repository (https://github.com/TRobertsLab/HVCRA_PreparatoryActivityData) (TRobertsLab, 2019https://github.com/elifesciences-publications/HVCRA_PreparatoryActivityData).

Animals

Experiments described in this study were conducted using adult male zebra finches and Bengalese finches (>90 days post hatch). During experiments, birds were housed individually in sound-attenuating chambers on a 12/12 h day/night schedule and were given ad libitum access to food and water. All procedures were performed in accordance with established protocols approved by Animal Care and Use Committee’s at UT Southwestern Medical Centers, Texas Christian University, Emory University, and the Korea Brain Research Institute.

Defining song and Peri-Song behavior

Zebra finch song is composed of a song motif which contains a stereotyped set of song syllables. The song motif is often produced two or more times in immediate succession – this is referred to as a song bout. Song bouts are often preceded by one or more introductory syllables. When a bird strings together more than one song bout, this is referred as a song phrase. In this study we defined song bouts separated by less than 2 s of silence as being part of a single song phrase, while any singing following silent gaps greater than two seconds from the last song phrase were considered as the beginning of a new song phrase. For analysis of pre-song activity we only analyzed data in which birds were not singing for a minimum of 5 s prior to production of their first introductory note or song motif. For analysis of post-song activity we only analyzed data from birds that did not engage in further singing for a minimum of 5 s following production of the last syllable in their song phrase or bout.

Viral vectors

The following adeno-associated viral vectors were used in these experiments: AAV2/9.CAG.Flex.GCaMP6s.WPRE.SV40 (University of Pennsylvania Vetor Core; Addgene Catalog #: 100842-AAV9), AAV2/9.CMV.EGFP.Cre.WPRE.SV40 (University of Pennsylvania Vector Core; Addgene Catalog #: 105545-AAV9) and AAV2/9.CAG.GCaMP6s.WPRE.SV40 (University of Pennsylvania Vector Core; Addgene Catalog #: 100844-AAV9). All viral vectors were aliquoted and stored at −80°C until use.

Imaging equipment

Head-mounted miniaturized fluorescent microscopy in freely behaving singing birds was conducted with an nVista system (Inscopix). Two-photon microscopy was conducted with a commercial microscope (Ultima IV, Bruker) running Prairie View software using a 20x (1.0 NA) objective (Zeiss) with excitation at 920 nm (Mai Tai HP DS, Newport). Two-photon and initial single photon imaging was conducted in lightly anesthetized head-fixed animals. Two-photon and single-photon images of HVC were acquired through the cranial window using a sCMOS camera (QImaging, optiMOS) and these images were used to guide placement of the baseplate for the miniaturized single-photon microscope (Inscopix). CAD files for head holders and stereotaxic devices are available upon request.

Stereotaxic surgery

All surgical procedures were performed under aseptic conditions. Birds were anesthetized using isoflurane inhalation (~1.5–2%) and placed in a stereotaxic apparatus. Viral injections were performed using previously described procedures (Roberts et al., 2012, Roberts et al., 2017) at the following approximate stereotaxic coordinates relative to interaural zero and the brain surface (rostral, lateral, depth, in mm): HVC (0, 2.4, 0.1–0.6); and RA (−1.0, 2.4, 1.7–2.4). The centers of HVC and RA were identified with electrophysiology. For calcium imaging experiments, 1.0–1.5 μL of Cre-dependent GCaMP6s was injected at three different sites into HVC and 350 nL of Cre was injected into RA. Viruses were allowed to express for a minimum of 6 weeks before a cranial window over HVC was made.

Cranial windowing, Imaging and Baseplate Implantation

Briefly, a unilateral square craniotomy (~3.5×3.5 mm) was created over HVC and the dura was removed. A glass coverslip was cut to match the dimensions of the craniotomy and held in place with a stereotaxic arm as Kwik Sil was applied to the edges of the cranial window. Dental acrylic was applied over the Kwik Sil and allowed to slightly overlap with the glass coverslip to ensure the window would not move and would apply the appropriate amount of pressure to the brain. An aluminum head post was affixed to the front of the bird’s head to enable head-fixed imaging under the 2-photon microscope and to enable head-fixation for baseplate implantation. Following verification of labeling, identification of HVC boundaries, and high-resolution images of neurons under the 2-photon microscope, the bird was lightly anesthetized with isoflurane and the miniaturized fluorescent microscope (Inscopix) was placed over the cranial window. The field of view that matched the 2-photon images was identified and the focal plane that enabled the largest number of neurons to be in focus was selected. Dental acrylic was used to fix the baseplate in the desired position and any exposed skull was covered with dental acrylic. Once the dental acrylic dried, the microscope was removed from the baseplate and the bird was allowed to recover overnight. About 30 min before the birds’ subjective daytime, the microscope was attached to the counterbalance (Instech) with enough cable to allow the bird to move freely throughout the cage. The microscope was then secured to the baseplate with a setscrew. The bird was allowed to wake up and accommodate to the weight of the microscope over the next 2–3 days. After the end of imaging experiments, the baseplate was removed and stereotactic injections of AlexFluor 594 were made into Area X. After 1 week, birds were given an overdose of Euthasol, perfused and tissue was sectioned.

GCaMP6s imaging using a miniaturized fluorescent microscope

The miniaturized fluorescent microscope (Inscopix) was not removed following successful baseplate implantation and remained attached to the birds’ head until either the cranial window closed or 7–10 days had passed. The counterbalance was adjusted based on the observed behavior of the bird and its ability to move freely. The female was not housed in the cage with the male bird, but instead was introduced to the males during a minimum of 3 morning and afternoon sessions to evoke directed song. Video recording was first started followed by 5 to 10 s of spontaneous recording with the miniaturized fluorescent microscope. The female bird was placed in the cage as quickly and as with little disruption as possible for each session. If the male bird did not sing within a minute of the females’ presence, the session was stopped, and the female was removed. All trials were recorded on video, and audio was recorded using Sound Analysis Pro (SAP) software and the HD video camera microphone. Calcium imaging was performed at 30 frames per second (fps), at 1080 × 1920 resolution, Gain was set to 4, and Power was set to 90% for all birds, behavioral videos were collected at 24 fps. Calcium imaging data and behavioral data was synchronized using start of calcium imaging on a frame by frame basis.

Calcium image processing and analysis

Calcium images were collected using the miniaturized fluorescent microscope developed by Inscopix (Ghosh et al., 2011). Video recordings of birds behavior and singing were manually synchronized on a frame by frame basis to the onset of calcium imaging, visualized by turning of the blue LED on the miniscopes. For processing of calcium imaging data the FOV (Field of View) was spatially cropped to exclude pixels that did not include neurons or observable changes in fluorescence. Next, the preprocessing utility within the Mosaic data analysis software was used to spatially bin the images by a factor of 2 to reduce demands on computer memory and enable faster data processing. The TurboReg implementation within Mosaic was used to perform motion correction. A reference image was created using a maximum intensity projection of the dataset and the images were aligned in the x and y dimension to the reference image. Imaging datasets with translational motion greater than 20 pixels in either the x or y dimensions were excluded from further data analysis. Post-registration black borders were spatially cropped out. The resulting spatially-cropped, preprocessed, and motion corrected calcium imaging datasets were exported for further analysis in custom Matlab scripts.

We performed ROI-based analysis on the motion-corrected calcium imaging datasets using previously described methods (Peters et al., 2014) (seeSource code 14). ROIs were manually drawn around identifiable soma and a secondary ROI that extended six pixels around the boundaries of the neuronal ROI was used to estimate background fluorescence (i.e. neuropil or other neurons). The pixel values were averaged within the neuronal and background ROIs, and background fluorescence signal was subtracted from neuronal signal. An iterative procedure using custom Matlab scripts were used to estimate baseline fluorescence, noise, and active portions of the traces (Peters et al., 2014). A subset of calcium images were re-analyzed using previously described constrained non-negative matrix factorization (CNMF) methods, but calcium fluorescence traces were identical to the traces pulled out by the ROI-based analysis (Pnevmatikakis et al., 2016). Calcium traces generated by ROI-based analysis were further deconvolved to produce inferred calcium traces using the pool adjacent violators algorithm (PAVA) (Friedrich et al., 2017). The deconvolved calcium traces were normalized to values between 1 and 0 to enable visualization of activity across different neurons during the same trial. Calcium transients that were 3 SD above baseline activity were recorded as events. The corresponding onset times and the rise times to peak fluorescence of individual calcium transients were correlated with synchronized behavior.

All calcium events were first categorized as falling into peri-song or song behavioral epochs on a frame by frame basis. Peri-song was limited to the 5 s period before the onset of vocalizations, including introductory notes, and the 5 s period after the offset of the last syllable. These event counts were used to assign a Phrase index to all imaged neurons. Neurons that had fewer than two calcium events recorded over a day of singing were excluded from further analysis because sparsity of calcium events could spuriously identify neurons as peri-song or song exclusive. We combined the number of calcium events from neurons imaged across multiple trials during the same day. Neurons imaged across multiple days were treated as unique neurons. The phrase index was calculated as a ratio of the total number of song events imaged from a neuron during a day subtracted by the number of peri-song events to the total number of calcium events. This bounded the phrase index to values of −1 (peri-song exclusive) and +1 (song exclusive). We used the phrase index to examine the timing properties of neurons active only during peri-song, song, or both behavioral epochs.

To examine the distribution of calcium events, we generated histograms with bin sizes of 200 ms. Peri-song event rates were first calculated in 100 ms bins and then a plotted using a 1 s moving average window that reaches a minimum of 500 ms at boundaries (−5,+5 s, and at song onset and offset). The average event rate and standard deviation was calculated using all pre and post phrase event rates from all birds.

Fluorescence analysis across intervals

Average fluorescent changes were measured for each neuron across baseline, pre-song, post-song, and song behavioral periods. Baseline was defined as a behaviorally quiet period covering 5 s of fluorescent activity that was ≥10 s removed from periods of singing or calling. Pre-song was 5 s before phrase onset and post-song was 5 s after phrase offset. The background subtracted fluorescent traces were used to measure average fluorescence across the above intervals for all phrases and all birds. Averaged fluorescent values were than normalized to the average fluorescence measured during song.

Comparison of Signal-to-Noise ratio and calcium event peak magnitudes

We measured the SNR of events occurring during song for a subset of pan-song and song neurons. The SNR was calculated as a ratio of peak fluorescence for each song event per neuron to the average fluorescence from baseline period within the trial (as above, 5 s of fluorescent activity that was ≥10 s removed from periods of singing or calling). We determined the average SNR for each neuron and examined differences between pan-song and song neurons.

Peak magnitudes during peri-song and song periods of pan-song neurons were measured using normalized deconvolved fluorescent traces. The peak values for each pan-song neuron (with phrase indices between −0.18 to +0.18) during peri-song and song periods were used to evaluate potential differences between calcium events occurring outside of song versus during song.

Neural recordings

Multiunit recordings of HVC neurons were collected from three adult (>90 days old) male Bengalese finches. All procedures were approved by the Korea Brain Research Institute. An array of 16 tungsten microwires (175 µm spacing, OMN1005-16, Tucker Davis Technologies) was implanted into left HVC. The location of HVC was identified by searching for spontaneous spike bursts and for antidromic response to stimulation in RA. The extracellular voltage traces of all channels from birds singing alone (without presentation of female) were amplified and recorded with an interface board (RHD2132, Intan Technologies) at a sampling rate of 25 kHz. The interface board was tethered to a passive commutator (Dragonfly Inc) via a custom-made light-weight cable. In total we obtained HVC recordings from 35 electrode channels in three birds (15, 7, and 8 channels, respectively), all of which showed spontaneous bursts typical of HVC neurons. The impedance of successful electrodes was around 100–300 kΩ.

Recorded signals were bandpass filtered (0.3–5 kHz) and negative signal peaks exceeding 4 SD from spontaneous activity (spontaneous activity was measured from time points greater than 10 s from the nearest song bout) were interpreted as multi-unit spikes. In total 38, 62, and 212 song onsets, and 42, 23, and 280 offsets were identified in these birds, respectively. We produced firing rate-traces from each electrode channel with 10 ms resolution and averaged them across song renditions. After smoothing with a 500 ms moving average window, the averaged firing rates were normalized between 0 and 1 to enable comparisons of recordings from different channels and to obtain the general trend of onset- and offset-related firing across channels and birds.

The significance of activity elevation during pre-song, song, and post-song periods from the baseline was tested by Wilcoxon signed-rank test with significant level at 0.05 after Bonferroni correction for multiple comparison. Onset of pre-song and offset of post-song activity were estimated for each channel as the smoothed spike rate trajectory was exceeded a threshold which was defined as mean +2 SD of the spontaneous spike rate.

Single-unit and multiunit recordings of RA neurons were collected from six adult (>140 days old) male Bengalese finches as described previously (Tang et al., 2014). All procedures were approved by the Emory University Institutional Animal Care and Use Committee. Briefly, an array of four or five high-impedance microelectrodes was implanted above RA. We remotely advanced the electrodes through RA using a miniaturize motorized microdrive and recorded extracellular voltage traces as birds produced undirected song (i.e., no female bird was present). We used a previously described spike sorting algorithm to classify individual recordings as single-unit or multiunit (Sober et al., 2008). In total, we recorded 19 single units (multiunit recordings were not analyzed further in this study). Based on the spike waveforms and response properties of the recordings, all RA recordings were classified as putative projection neurons (Leonardo and Fee, 2005; Sober et al., 2008; Spiro et al., 1999).

Analysis of chronic recording data

To analyze the variation in inter-spike-interval (ISI) in different time periods (Figure 5), we restricted our analysis to cases in which we collected at least one recording that included the relevant song epoch. ‘Spontaneous’ epochs were sampled from neural activity recorded more than 10 s after the nearest song bout. ‘Pre-song’ activity was sampled from between 2.5 and 0.5 s prior to the first song syllable or introductory note. ‘Song’ activity was sampled from the onset of the first song syllable until the offset of the last syllable in a bout. ‘Post-song’ activity was sampled from between 0.5 and 2.5 s after the offset of the last syllable in a bout. In some cases, we did not have sufficient data available from all epochs for all neurons (note the variation in the number of neurons included in the analyses shown in Figure 5d).

Air sac recording procedures

Subsyringeal air pressure was recorded from six adult male zebra finches in directed singing conditions. Directed song was defined as a female presented in an adjacent cage during a two-hour recording period. Data from four of the birds were re-analyzed from a previously published study (Cooper and Goller, 2006) and data from two additional birds were collected to replicate the effects observed in the previously collected data (Cooper and Goller, 2006). As described in (Secora et al., 2012), each bird was accustomed to carrying a pressure transducer that was held in place on the bird’s back with an elastic band (Secora et al., 2012). To facilitate relatively free lateral and vertical movement in the cage, the weight of the transducer was offset by a counter-balance arm. Subsyringeal air pressure surgery was performed after birds sang while carrying the pressure transducer. Prior to insertion of the air pressure cannula, animals were deeply anesthetized as verified by an absence of a toe-pinch response. A small opening in the body wall below the last rib was made with a fine pair of micro-dissecting forceps, and a flexible cannula (silastic tubing, OD 1.65 mm, 6.5 cm length) was inserted into the body wall and suture was tied around the cannula and routed between the 2nd and 3rd ribs to hold it in place. The skin was sealed to the cannula with tissue adhesive (Nexaband). The free end of the cannula was attached to the pressure transducer. This allowed for measurement of relative subsyringeal air pressure changes inside the thoracic air sac before, during, and after spontaneously generated song events. Birds were monitored following surgery until they perched in the recording chamber.

The voltage output of the pressure transducer was amplified (50–100 x) and low-pass filtered (3 kHz cutoff; Brownlee, Model 440, Neurophase, Santa Clara, CA). Respiration was recorded for five seconds prior to and following singing epochs using a National Instruments analog-to-digital conversion board (NI USB 6251, Austin, TX) controlled by Avisoft Recorder software (Avisoft Bioacoustics, Berlin, Germany). Data were collected in wav file format, 16 bit resolution, with sampling rates varying from 22.05 to 40 kHz. Songs were selected for analysis that contained at least 3 s of uninterrupted quiet respiration prior to and following song. Songs that were preceded by calls, drinking, defecation, or movement-related activity were excluded from the analysis.

Air sac data classification

Air pressure was analyzed as respiratory cycles, which was defined as an inspiration followed by expiration. The onset of inspiration was identified as subambient air pressurization and the return to ambient pressure following the expiratory phase of the cycle. The cycle duration (s), duty cycle (% time spent in the expiratory phase of respiration), and average rectified amplitude (a.u.) was calculated for each cycle. Song respiration was analyzed prior to song onset and following song termination. Song onset was defined as the inspiration preceding the first introductory note; using this marker, the onset time for each respiratory cycle in the pre-song recording period was determined. The conclusion of song was defined as the termination of the expiration generating the last song syllable in the bird’s song bout. The timing of the respiratory cycles following song were identified relative to the song termination marker.

Statistical analyses of respiratory data

For statistical analyses of the respiratory data, each bird contributed a single average value for each measured parameter (cycle duration, duty cycle, average amplitude). A repeated measures ANOVA was used to determine how respiration changes prior to and following song. For each bird, ten to twenty songs were identified for the statistical analysis (see above for criteria). The average for the pre- and post-song (3–5 s) for each measured parameter for each bird was calculated. To evaluate the time course of change in respiratory patterns preceding and following song, the average for one second bins for each bird were used in the repeated measures ANOVA. In cases where the assumption of sphericity was violated, the Greenhouse-Geisser correction for the degrees of freedom was used. All p values reported are based on this correction. An a priori alpha level of 0.05 was used for determining statistical significance.

References

  1. 1
  2. 2
  3. 3
  4. 4
  5. 5
  6. 6
  7. 7
  8. 8
  9. 9
  10. 10
  11. 11
  12. 12
  13. 13
  14. 14
  15. 15
  16. 16
  17. 17
  18. 18
  19. 19
  20. 20
  21. 21
  22. 22
  23. 23
  24. 24
  25. 25
  26. 26
  27. 27
  28. 28
  29. 29
  30. 30
  31. 31
  32. 32
  33. 33
  34. 34
  35. 35
  36. 36
  37. 37
  38. 38
  39. 39
  40. 40
  41. 41
  42. 42
  43. 43
  44. 44
  45. 45
  46. 46
  47. 47
  48. 48
  49. 49
  50. 50
  51. 51
  52. 52
  53. 53
  54. 54
  55. 55
  56. 56
  57. 57
  58. 58
  59. 59
  60. 60
  61. 61
  62. 62
  63. 63
  64. 64
  65. 65
  66. 66
  67. 67
  68. 68
  69. 69
  70. 70
  71. 71
  72. 72
  73. 73
  74. 74
  75. 75
  76. 76
  77. 77
  78. 78
  79. 79
  80. 80
  81. 81
  82. 82
  83. 83
  84. 84
  85. 85
  86. 86
  87. 87
    The neuromuscular control of birdsong
    1. R Suthers
    2. F Goller
    3. C Pytte
    (1999)
    Philosophical Transactions of the Royal Society of London. Series B: Biological Sciences 354:927–939.
    https://doi.org/10.1098/rstb.1999.0444
  88. 88
  89. 89
  90. 90
  91. 91
  92. 92
  93. 93
  94. 94
  95. 95
  96. 96
    Birdsong and singing behavior
    1. H Williams
    (2004)
    Annals of the New York Academy of Sciences 1016:1–30.
    https://doi.org/10.1196/annals.1298.029
  97. 97

Decision letter

  1. Erich D Jarvis
    Reviewing Editor; The Rockefeller University, United States
  2. Eve Marder
    Senior Editor; Brandeis University, United States
  3. Franz Goller
    Reviewer; University Utah, United States

In the interests of transparency, eLife includes the editorial decision letter and accompanying author responses. A lightly edited version of the letter sent to the authors after peer review is shown, indicating the most substantive concerns; minor comments are not usually included.

Thank you for submitting your article "Transitioning between preparatory and precisely sequenced neuronal activity in production of a skilled behavior" for consideration by eLife. Your article has been reviewed by Eve Marder as the Senior Editor, Erich Jarvis as the Reviewing Editor, and two reviewers. The following individual involved in review of your submission has agreed to reveal his identity: Franz Goller (Reviewer #2).

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

Summary:

This study by Daliparthi et al., presents an interesting new finding, that a subpopulation of neurons in the motor forebrain nucleus HVC of songbirds has exclusive premotor and/or postmotor activity, that does not appear to correlate with the actual vocalizations produced. To make these discoveries, the authors used GCaMP calcium imaging of HVC neurons that project to RA song nucleus of singing zebra finches, multiunit electrophysiological recordings in HVC of Bengalese finches, and respiratory activity in air sacs that is consistent with a pre-motor preparation before singing. Their interpretation is that the peri-song activity in HVC (as they call it) could be responsible for initiating and recovering singing behavior, or motor planning of the behavior. This is a technical tour de force study, especially the Ca2+ imaging in a specific neurons type, of awake singing birds.

Although the reviewers were excited with the approach used, there were mixed opinions on the novelty of the findings. The main novel finding that the authors say that they have is pre and post singing activity in the HVC song nucleus, and their interpretation that this cannot be controlling the motor act of producing song syllables/sequencing. It seems clear that there is a mismatch between what the authors say is published in the literature, and what is in the literature. There are studies that have shown pre and post singing neural firing, not just in HVC or HVCRA projecting neurons, but in other song system neurons as well. Where the real novelty may lie, is how much more have these authors learned about this? This study may be one of the first to find neurons that fire exclusively before or after singing, but not during. And if this is the case, what insights have the authors gained that others have not gained in the past studies? But this is not what the authors have done. Rather, they appear to present a misinterpretation of their findings relevant to the current literature. This results in the authors' overselling their story and underselling the past literature, taking a strawman approach. Further, such a big difference may not be necessary for the study to be considered important. It appears to the editors that the paper will require significant revisions and more balanced justifications to be accepted.

Essential revisions:

The authors must carefully reconsider and present what is novel relevant to the current literature, and their interpretation of those findings (particularly in reference to Okubo et al., 2016; Day et al., 2009; Goldberg et al., 2010 and Kao et al., 2008). The authors first say that studies in the past have not found pre-song activity in HVC, and that it was thought to only function at the moment of singing for sequences of vocalizations, citing single unit studies. Then they more correctly claim later in the paper, that electrophysiology studies have found premotor activity in HVC, using multi-unit recordings. What the authors need to do is be fair to the field of the neurobiology of motor behavior of song behavior, in particular, and mention a well-known principle in neuroscience, that neural activity in brain areas that controls movement of muscle activity precedes the movement/muscle activity. This is one reason why HVC was called a premotor nucleus.

The authors introduced their topic in a very wide context of neurons controlling sequencing behavior in the brain, across vertebrates. But then neither in the Results section or Discussion section did they interpret their findings in that wider context. So, the reader is left with a myoptic view of the HVC song nucleus, and nothing of relevance to anything else. The authors should in the discussion say how they interpret their findings in the control of learned movements generally. HVC has been proposed to function like cells in the mammalian motor cortex (layer III neurons of M1 for example Pfenning et al., 2014) or premotor cortex (Brocas area for example Bolhuis et al., 2010). Based on these comparisons, would the authors propose or predict that mammalian motor cortex show preparatory activity seconds before a movement is performed? Hasn't such analyses been undertaken already in mammalian motor cortex for movement, at the multiunit and single unit levels? If the authors are going to brush broad strokes in their Introduction, they must complete that in the Discussion section.

The authors look like they combined multiple studies from different labs without an initial plan for those studies to be combined. What this leads to are different studies with different variables and definitions. This includes different definitions of a song bout, song phrase, pre- and post-singing times for defining a bout and for analyses; different species (zebra finch vs Bengalese finch); and different social context (with a female or singing alone). It is well known that these factors make a difference in analyses and neural activity in the brain. The authors need to be more consistent in their behavioral definitions, pre- and post-singing time periods across studies, and the social context used. They should consider doing some of their analyses over again with the same timing of 1, 2, or 5 seconds pre and post singing to make them more consistent, and not one set of times (5 seconds) for the GCaMP imaging and another (2 seconds) for the multiunit electrophysiology. Also, in some of their analyses they seem to not include introductory notes in the bouts/phrases, and sometimes do. If they do not, then they could get an artificial error of thinking they found activity before singing when it is really activity occurring during singing the introductory notes. In the video the HVC neurons seem to become active during production of the introductory notes.

[Editors' note: further revisions were requested prior to acceptance, as described below.]

Thank you for resubmitting your work entitled "Transitioning between preparatory and precisely sequenced neuronal activity in production of a skilled behavior" for further consideration at eLife. Your revised article has been favorably evaluated by Eve Marder (Senior Editor), and two reviewers. We apologize for the time this taken, but we were waiting for input from one of the relevant parties and now are just proceeding without that input.

Both reviewers feel that the manuscript has been substantially improved but one of the reviewers still is concerned about some issues.

Two important (yet addressable) concerns remain.

1) A previous concern was how bout onsets were defined and how cage noise was defined. The authors responded that the reviewers should 'rest assured' that the bout onsets are correctly labeled.

The spectrograms in Figure 2A,C,E and many of the supplementary Figures (e.g. Birds 162048, 160046) still all have what look like vocalizations prior to the red lines marking the bout onset. Specifically, there are apparent harmonics in the spectrogram in the 2-8khz range that look a lot like calls/intro notes in both Figure 2C,E under areas called cage noise. Cage noise, like song, can be broadband, but unlike song has lots of power concentrated at <1kHz and – also unlike song, rarely has nice harmonics with fundamentals at ~kHz ranges. But the cage noise parts of most spectrograms does not have this spectral profile – unless the spectrogram is being bandpass filtered in a way that is not find mentioned in the Materials and methods section.

Also, in the supplemental figure showing Directed Singing 162048, the sound trace underneath the mic voltage signal is totally flat for the first seconds and the HVC activity is totally silent. Then there is a sudden onset of power in the mic signal, which by eye looks to be totally synchronous with the first HVCRA transient – the mic signal is then noisy for the next several seconds when there is HVC activity. A similar pattern is observed in 160046. Thus, in these examples HVC 'pre-bout' activity is associated with noises on the mic signal (reported to be cage noise). Overall, HVC activity is rarely observed when the mic signal is flat and is much more commonly observed when there is time-aligned power on the mic signal (called cage noise).

While I hesitate to be to nit-picky about this, the entire paper is about pre- and post-bout Ca2+ transients. Yet most spectrograms appear (by eye) to have vocalizations preceding and following defined bout onsets and offsets. Since we know that HVCRA neurons can discharge during calls and intro notes, a lot depends on the authors totally nailing beyond a shadow of a doubt what constitutes a bout onset and what constitutes peri-bout concentrated cage noise.

Guidance: One way to resolve this issue would be to simply plot beneath the unfiltered mic signal a mic signal that is equal to the ratio of power in the 0.2-1 Hz range to the power in the 2-8 kHz range. Cage noise is easily distinguishable from real vocalizations by its excess power at low (<1kHz) frequencies. To address this specific reviewer concern, expanded views of the spectrograms highlighted in the comments above would also help.

Whatever the authors do, the paper would be stronger – and better for posterity – if cage noise was convincingly/objectively defined and clearly distinguished from calls, and if the showcased examples of peri-bout HVC Ca2+ transients were not compromised by what many readers are bound to think are spectral elements in peri-bout periods that look like vocalizations.

2) While the authors have done a much better job in the introduction and discussion contextualizing their findings in the literature, in some places they are still mis-characterizing past work. I make some recommendations below that I hope will be considered.

Subsection “Peri-Song Activity in Populations of HVCRA Neurons”: The authors claim that the pre-bout burst in HVC is a "newly discovered activity profile." But that's not accurate. In fact, the very first example of a singing related projection neuron in Picardo et al., (Figure 1E) is a neuron that exhibits a pre-bout transient that precedes bout onset by several seconds – just like examples in this study. If the authors want to counter that the Picardo study imaged both HVCX and HVCRA neurons while this one specified HVCRA neurons; they may be technically accurate but it's still misleading to report a 'new activity profile' – especially to someone who is familiar with the Picardo work (which includes pre-bout activity seconds in advance of the bout), the Okubo and Rajan papers which also include pre-bout activity.

Subsection “Peri-Song Activity in Populations of HVCRA Neurons”: I think including the word 'exclusively' is unnecessary and not quite accurate description of the papers cited. Just because a paper focuses entirely on the role of HVCRA activity in patterning vocalizations does not mean that the paper requires HVCRA activity to discharge exclusively during vocalization.

Subsection “Peri-Song Activity in Populations of HVCRA Neurons”: The authors wonder why previous studies did not observe peri-bout activity. Yet other studies did observe peri-song activity; just less of it. Again, see Picardo et al., Figure 1; Okubo et al., etc, Rajan…

Subsection “Peri-Song Activity in Populations of HVCRA Neurons”: The authors' explanations/conjectures here about why potentially pre- and post-bout discharge was 'missed' are too speculative and not quite accurate. The triggering with a buffer is not how ephys studies of HVC were done. In original experiments they were done with audio monitors so sparse discharges would not be missed. Antidromic identification also does not bias for discharge during singing (thus the discovery that many HVCRA neurons are silent). And during intracellular recordings from HVC in awake, singing birds – every recording is so precious and hard-earned and eyeballed in real time on a computer/oscilloscope that there is no way that pre-bout discharge would have been missed. In these studies, half of the HVCRA neurons were either silent or active during introductory notes (though this was not emphasized in the papers).

A real possibility that could be included here is that Ca2+ transients do not equal spiking and vice versa. Dopamine 'ramps' are observed in Ca2+ imaging of neurons and in fiber photometry but have never been observed in DA spiking (there was a long discussion of this issue at CoSyne 2018 meeting, it is covered in recent review by Josh Berke (Berke, 2018), and is relevant to ramping signals observed in the recently posted biorxiv paper from the Witten group (https://www.biorxiv.org/content/10.1101/456194v1)).

Also, inhibition can independently regulate Ca2+ influx and spiking – resulting in non-congruence between spikes and GCAMP Ca2+ signals (nicely shown in Owen, Berke and Kreitzer, 2018);

Discussion section: Get rid of 'exclusively'.

Discussion section: Get rid of 'only'.

Discussion section: Again the statement that pre-bout Activity in HVC was not previously observed is not accurate; see Picardo et al., Figure 1.

In the interest of expediency, I have thought about a faster way that the authors could address the issue of cage noise and bout onset definition. It seems to me that if the authors simply provide the wav files that correspond to all of the spectrograms in Figure 2 and Figure 2—figure supplement 1 than any reviewer and reader could just download the song and listen to it him/herself.

[Editors' note: further revisions were requested prior to acceptance, as described below.]

Thank you for resubmitting your work entitled "Transitioning between preparatory and precisely sequenced neuronal activity in production of a skilled behavior" for further consideration at eLife. Your revised article has been favorably evaluated by Eve Marder (Senior Editor) and Erich Jarvis (Reviewing Editor). The manuscript has been improved but there are some remaining issues noted by the Reviewing Editor that need to be addressed before acceptance, as outlined below:

There is a remaining issue with interpretation and semantics of what is pre- and post-vocalization activity measured in HVC and the song system relative to prior studies, which is what has been causing confusion to the reviewers. From a semantic understanding, pre-song activity in HVC-RA or any neurons of the song system could be considered 1 ms, 10ms, 100ms, 1s, 10s, or 10 min before vocalizing. In all cases, activity increases occur before vocalizing. All are premotor changes in neural firing. The difference that the authors find is that some neurons continue to fire during the motif or syllable production that has been found in previous studies and others do not, the new finding of this study. These former are what the authors appear to characterize as 'during song neurons', even if they fire 1-500 ms before or after the sound itself. The later fire 500ms to 3,000ms before sound is produced, and some of these shut off during the moments of syllable production. The authors assume that the idea of the 1-500ms range of premotor activity is "baked in", and therefore they call them 'during song production/singing neurons'. But what is baked-in the neuroscience field is that there is premotor activity in motor pathways in the manuscript before sound production or other behavior production. Can the authors see the confusion here?

To be less confusing to readers, the authors should include a table that defines the different neuron categories according to their temporal firing patterns, and then use those definitions consistently throughout the paper. Example terms that need such definitions in the table are: peri-song neurons (presumably within 5-0.5s firing before or after the first or last song syllable, respectively); pre-song neurons (presumably firing ~0.5s to 1ms before each syllable or the first song syllable); post-song; pan-song, etc. The real novelty of this study is finding neurons of the peri-song neurons as defined above, and not even just HVC-RA projecting neurons, but the fact that any such neurons exist at all in the song system. The paper needs to highlight this novelty more clearly. To do so, this difference in timing of neuron population firing and pattern should be stated in the Abstract, end of Introduction and in sentences of the Results section when this topic is discussed.

More specifically, what appears most novel about the current study is that the authors find subpopulations of HVC-RA projection neurons that: (a) fire 1-3 seconds before singing starts [or after it ends]; (b) some of which turn off a few ms before or right at the time singing starts; and (c) are not correlated with as of yet identified pattern of singing. The other types of neural firing patterns, including pre- and post-song production, have been found in prior studies, and are confirmed in this study. This needs to be stated much more clearly in the manuscript.

Introduction: Making too strong of a statement that HVC-RA neurons are exclusively active during vocalization in awake birds. Even in awake birds, physiology and immediate gene studies have shown some activity to hearing songs in other species (even if birds own song) or eating (even if difficult to interpret gene expression studies). Instead of saying exclusively, how about predominantly.

Results section: The authors claim that the pre-bout burst in HVC is a "newly discovered activity profile." But that's not accurate when considering the 500 ms window of pre-bout activity. Here is where the authors would benefit by using more well-defined terminology that they show in a table as well as define the timing of what they mean by pre-bout activity.

Discussion section: It appears the Bengalese finch electrophysiology peri-song activity is within a shorter time window of several 100ms rather than 1-3 seconds as found with Ca2+ imaging in zebra finches. Is this a species difference, a physiology versus GCAMP sensing difference, or some other difference?

The song system diagram of Figure 1A is appears to be simplified too much. It does not show MAN, the loop with AreaX and DLM, and the feedback to RA (as well as to HVC). Such additions to the figure would not cause more confusion, and in fact would add clarity and has been what is commonly shown in birdsong diagrams for several decades now. If it would make it easier for the authors, they could use and modify one of the figures showing these connections in the adobe illustrator versions on the Jarvis Lab website: http://jarvislab.net/summary-figure-originals/ These are more polished than the hand drawn one of this paper.

As the authors wrote in one of their responses, they should mention that the variability of peri-song neuron activity on different bouts is similar to variable pre-movement activity found primate M1 neurons. Such a comparison broadens the impact of the paper.

https://doi.org/10.7554/eLife.43732.052

Author response

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

We thank the reviewers for their comments. We have endeavored to address all of them and provide written responses to each below. We have been asked to address three essential editorial revisions to our manuscript that involve adding additional discussion of the literature to the Introduction, the Discussion section, and to provide a more seamless presentation of results for some of our data. We have directly addressed these requests outlined by the editor. We have (1) added a discussion of previous literature showing premotor activity in HVC-X, HVC-interneurons, and other portions of the song system into the Introduction section of the manuscript, (2) broadened our Discussion section in order to interpret our findings in the context of research in the mammalian motor cortex and the initiation of neuronal sequences more generally, and (3) endeavored to be more consistent in the descriptions, analysis and presentation of the research conducted across the multiple laboratories that have contributed to this study. In addition to these revisions we have carefully gone through each comment by the reviewers and editor and provide detailed replies to each.

Summary:

This study by Daliparthi et al., presents an interesting new finding, that a subpopulation of neurons in the motor forebrain nucleus HVC of songbirds has exclusive premotor and/or postmotor activity, that does not appear to correlate with the actual vocalizations produced. To make these discoveries, the authors used GCaMP calcium imaging of HVC neurons that project to RA song nucleus of singing zebra finches, multiunit electrophysiological recordings in HVC of Bengalese finches, and respiratory activity in air sacs that is consistent with a pre motor preparation before singing. Their interpretation is that the peri-song activity in HVC (as they call it) could be responsible for initiating and recovering singing behavior, or motor planning of the behavior. This is a technical tour de force study, especially the Ca2+ imaging in a specific neurons type, of awake singing birds.

We thank the editor and reviewers for the praise of the calcium imaging approaches used here and for our findings more generally. These technically challenging experiments allowed us – for the first time – to examine population level HVC-RA neuron activity in freely-behaving zebra finches and we believe that these findings provide a new view for how precise neuronal sequences can arise in the brain.

Although the reviewers were excited with the approach used, there were mixed opinions on the novelty of the findings. The main novel finding that the authors say that they have is pre and post singing activity in the HVC song nucleus, and their interpretation that this cannot be controlling the motor act of producing song syllables/sequencing. It seems clear that there is a mismatch between what the authors say is published in the literature, and what is in the literature. There are studies that have shown pre and post singing neural firing, not just in HVC or HVCRA projecting neurons, but in other song system neurons as well. Where the real novelty may lie, is how much more have these authors learned about this? This study may be one of the first to find neurons that fire exclusively before or after singing, but not during. And if this is the case, what insights have the authors gained that others have not gained in the past studies? But this is not what the authors have done. Rather, they appear to present a misinterpretation of their findings relevant to the current literature. This results in the authors' overselling their story and underselling the past literature, taking a strawman approach. Further, such a big difference may not be necessary for the study to be considered important. It appears to the editors that the paper will require significant revisions and more balanced justifications to be accepted.

In our original version of the manuscript we did not mean to imply that pre and post singing activity in other classes of neurons in HVC or other areas of the song system does not occur. We are aware of this literature and both cited and discussed it in the Results section and Discussion section of the original manuscript. We believe the confusion has arisen from the distinction between previous recordings of different classes of HVC projection neurons and interneurons, whereas our work has focused exclusively on HVC-RA neural activity. We have added a discussion of this previous literature in the Introduction and more discussion of it in the Discussion section. In addition, in the revised manuscript we have tried to clarify the writing to focus the reader on the fact that the calcium imaging is focused on HVC-RA neurons. Our original argument is that we have a novel observation of activity related to motor planning and preparation in this population of neurons.

Reviewers seem to claim that pre- and post-singing activity has previously been reported in HVC-RA neurons. We have been unable to find strong support for this in the literature. It is exceedingly difficult to record from HVC-RA neurons – this is why most people do not do this and why there are only a handful of papers that have successfully recorded from identified HVC-RA neurons. The reason they are difficult to record from is that (according to the published literature) these neurons do not fire action potentials unless the bird is singing. HVC-RA neurons are not spontaneously active, necessitating antidromic identification to identify the neurons prior to recording electrophysiological activity of these cells in vivo. Moreover, it is well documented that different classes of HVC projection neurons or interneurons have different firing patterns during singing (see for example Kozhevnikov and Fee, 2007) – therefore recordings from other classes of neurons (e.g., HVC-X, or HVC-Av) or multi-unit recordings, which are dominated by interneuron activity, cannot be used to infer the activity profile of HVC-RA neurons.

To thoroughly evaluate what is known about HVC-RA neuron activity in zebra finches we have systematically reviewed the previously published literature:

In Hahnloser, Kozhevnikov and Fee, 2002 the first recording of HVC-RA neurons in singing birds; Kozhevnikov and Fee, 2007; and in Long, Jin and Fee, 2010 and Hamaguchi, Tanaka and Mooney, (2016) the authors use intracellular recordings in sing zebra finches.

In Okubo et al., (2015) and Lynch et al., (2016) the Fee lab reports on extracellular recordings from RA-projecting and X-projecting neurons in juvenile and adult HVC. In many instances in these manuscripts they do not directly state if the neurons are identified other than to say that they think they are projection neurons. An important conclusion from these papers is that activity patterns in HVC changes drastically over development as birds learn to sing, transitioning from more rhythmic activity to less rhythmic song locked activity. This implies that activity in juvenile birds cannot be used as a measure for how the precise neural sequences exhibited in adult birds are initiated. Nonetheless, pertinent to the editors’ statement that people have shown that HVC-RA neurons are active before and after song, in the Okubo paper they describe pre and post song activity in juvenile birds.

Please note that the authors refer to these neurons as projection neurons, not HVC-RA neurons. In the parlance of this study, this means that they are grouping data from some antidromically identified neurons projecting to either Area X or to RA and some neurons that are simply presumed to be projection neurons based on their firing patterns. It is not reasonable to draw from this statement that HVC-RA projecting neurons are active immediately before and/or after song. Therefore, we have carefully gone through the data in this manuscript as well as the published dissertation of Okubo in order to better examine the idea that HVC-RA neurons have been shown to be active before and after song. In these documents there is the report of a single identified HVC-RA neuron in a 65dph juvenile bird that is active only at the offset of song bouts and it appears to be within 100-200ms of bout offset (the Okubo paper only examines activity within 300ms of the beginning or ending of song – a very different timescale than we have examined which includes the observation that activity significantly increases 2.5 seconds prior to song onset, see Figure 1E and ED Figure 2H,K,L). We did not find any mention of HVC-RA neurons being active prior to song bouts. Moreover, the data regarding bout onset and offset neurons appears to be exclusive for a juvenile bird and is not reported in adults. It should also be noted that a main point of the Okubo study is that the rhythmic firing observed in juveniles can be used to slowly build the precise neural sequences seen in adult birds. In particular, Okubo hypothesizes in both the Nature paper and the Ph.D. thesis that the bout related activity could be useful for incorporating new song syllables from the edges of song during song learning. This activity is not hypothesized to function in motor planning or motor preparation, which is conceptually different than the main point of our study.

The most comprehensive study to date documenting pre-song activity in HVC is the recent paper from Rajan, (2018). Rajan details the activity of interneurons and X-projecting neurons in HVC prior to song onset (~217-775 ms before song onset). Rajan, (2018) does not report on the activity of HVC-RA neurons; however, in the Discussion section a review of the literature is provided that appears to affirm the arguments we make here. In the Discussion section Rajan, (2018) states:

“In the current study and in previous studies, pre-bout activity of HVCRA projecting neurons has not been characterized. However, it has been reported that a fraction of HVCRA projecting neurons do not have any motif related activity (Hahnloser et al., 2002; Kozhevnikov and Fee, 2007; Long et al., 2010). Whether such neurons have pre-bout activity needs to be determined by recording from HVCRA projecting neurons before bout onset. Given the high connection probability between HVCRA projecting neurons and interneurons (Kosche et al., 2015), it is possible that a small number of HVCRA neurons with increased pre-bout activity can drive increases in a large fraction of interneurons as seen in this study. Additionally, slice recordings have shown that HVCX projecting neurons are also a source of excitatory input to interneurons, albeit less frequently than HVCRA projecting neurons (Mooney and Prather, 2005). They could also be a source of the increased pre-bout activity seen in interneurons as the time-scales of increases in activity were similar for HVCX neurons and interneurons (Figure 4). … Assuming the presence of specialized HVCRA neurons with similar preparatory roles, such neurons would be very different from the HVCRA neurons that are thought to encode “time” in song (Fee et al., 2004; Long et al., 2010). In addition to differences in firing patterns (“ramping-up” like activity vs. stereotyped sparse bursting), the connectivity of these neurons would also be different reflecting their different functional roles.”

Last, we also examined computational models of song production as it relates to HVC. From these models we were not able to find mention of HVC-RA neurons functioning in a capacity associated with motor preparation or motor planning. We then examined how these papers modeled the start of song/neural sequences with the idea that they could provide more general perspective on how neural sequences in HVC are hypothesized to be initiated by the field. For example, do they arise from preparatory activity among HVC-RA neurons or do they tend to be associated with moments in song and initialized by extrinsic excitatory drive (for example). From this analysis it appears that computational models also restrict themselves to the activity associated with triggering precise neural sequences at the start of song, rather than HVC-RA transitioning from phases associated with heterogenous activity for planning and preparation, spanning several seconds, prior to launching into precise song sequences. The summary of these computational models is presented in Table 1.

Table 1

ReferenceCell TypeComputational ModelSequence Initiation
Fiete et al., 2010HVC-RA neurons and a global inhibitory unit representing the pool of HVC interneuronsHeterosynaptic competition combined with spike-time-dependent plasticityRandom barrage of external input
Li and Greenside, 2006Either Leaky integrate-and-fire (LIF) neurons or single-compartment conductance based (HH, for Hodgkin-Huxley like) neurons.1-dimensional, homogenous, excitatory chain of nonbursting neuronsA brief high-frequency burst.
Gibb, Gentner, and Abarbanel 2009HVC-RA and HVC interneurons using single compartment Hodgkin-Huxley neurons.Chain of bistable clustersBurst was initiated with a 3-4ms DC current pulse into 50% of HVC-RA neurons.
Jin and Ramazanoğlu, 2007HVC-RA neuronsTwo-compartment model of HVC-RA neurons, minimal conductance-based modelBurst propagation was initiated by stimulating eight spikes in neurons in the first group of the chain, via current injection lasting 10ms.

The robust level of HVC-RA activity in the time periods leading up to song vocalization that we describe suggest that these neurons may a play a significant role in motor planning and preparation – an important distinction from the current view of HVC-RA neuron function. The reviewers are correct in saying that neural activity that might be associated with motor planning and preparation has been described in other HVC neurons and in other regions of the songbird brain. However, the preparatory activity described in our study has important distinctions from these previous studies. First, we are examining neural activity in populations of HVC-RA neurons. These neurons are necessary for production of learned song and are widely thought to carry a timing signature for song by controlling learned, sequential motor patterns. The anterior forebrain pathway is not necessary for an adult bird to produce its learned song (lesions or inactivation of lMAN, the output of the AFP, do not disrupt the birds ability to sing – this has been shown by Brainard and Doupe and the Nordeen’s among others) and lesions of HVC-X neurons (Scharff et al., 2000) suggest that HVC-X neurons (unlike HVC-RA) are not necessary for producing song. It is not known whether HVC-RA neurons are involved in motor planning or motor preparation (see text above and quoted Discussion section of Rajan, 2018). Second, we show that subpopulations of HVC-RA neurons are active on the order of seconds before song vocalization. Previous work has identified preparatory activity in other cell types and regions on much shorter timescales: HVC-X and interneurons (~217-775 ms before song onset; Rajan, 2018), subpopulations of Area-X neurons projecting to DLM (~0.56 s before song onset; Goldberg et al., 2010), LMAN neurons during undirected song (~600 to 100 ms before song onset; Kao et al., 2008), multi-unit HVC activity in juveniles with immature song (no clear time scale, Day et al., 2009), and putative projection neurons in HVC of juveniles with immature songs (~0 to 300 ms before and after song onset, Okubo et al., 2015). These previous findings are summarized in Table 2.

Table 2

ReferenceCell TypeDescription of ActivityTiming
Okubo et al. 2015Putative HVC projection neurons: N= 285; 137/285 bout onset neurons; 98/285 bout offset neurons; 50/285 active both before and after bouts. 1 identified HVC-RA neuron firing at the end of bouts. At least 3 identified HVC-X neurons (2 bout onset, 1 bout offset).Significant pre and post-bout activity in the 300ms window before and after bouts. Both pre and post-bout peaks occurred between 0 to 100 ms before or after bout onset or offset, respectively.Activity hundreds of milliseconds before or after song vocalizations in mixed populations of neurons.
Kao et al., 2008LMAN neurons: N= 18; elevated activity before undirected song onset. N=16; decreased activity following cessation of undirected song.In 18 neurons, “firing rate in a 500ms interval from 600 to 100ms before undirected song was greater than the average spontaneous firing rate”. In 16 neurons, firing rate “was lower than the average level of spontaneous activity” following song offset.Activity in LMAN neurons hundreds of milliseconds before or after song vocalization, but only in undirected social context.
Goldberg et al., 2010Area X-DLM neurons (HF-2 neurons): N=37 neuronsIn HF-2 neurons, “firing rate gradually increased before singing (0.56 ± 0.22s), and slowly returned to baseline following singing (1.64 ± 1.2s).”Activity in AreaX-DLM neurons exhibited pre-song activity hundreds of milliseconds before song onset, and seconds after song offset.
Day et al., 2009Multiunit activity in HVCElevated activity in juvenile birds before and after song vocalization.Unclear
Rajan and Doupe, 2013HVC-X neurons: N= 30; 12 antidromically identified, 18 putative.
Putative Interneurons: N= 16
Activity of putative interneurons and HVC-X neurons shown to be correlated with onset times of introductory notes.Activity time-locked to introductory note vocalization.
Vyssotski et al. 2016NIF neurons projecting to HVC16 Nif-HVC neurons fire on average above baseline levels 70 ms prior to song onset and it decays to baseline about 30 ms prior to song offset.Activity is time locked to introductory note type. Authors did not look beyond 500 or 1000 ms before song onset.
Rajan 2018HVC-X neurons: N=39; 13 antidromically identified, 26 putative.
Putative interneurons: N=17
4 HVC-X neurons increased their activity (-928 ms to -129ms prior to song bout onset). 3 HVC-X neurons decreased activity -528ms to -47ms). 13 interneurons increased activity (~-674ms to -117ms).Activity before any vocalization including introductory notes.

In our study, we report pre-song neural activity in HVC-RA neurons that is 3 SD’s above baseline 2.5s prior to song onset. Indeed, a core point of our manuscript is that neurons known to sit atop the descending motor pathway necessary for birds to produce their song (HVC-RA neurons) also exhibit activity seconds prior to song onset that is consistent with a role in both motor planning and preparation and may provide a sensitive read-out of the decision to start singing. In addition, we identify for the first time HVC-RA neurons that are active before singing, but inactive during singing. This new functional population of neurons appear to play an exclusive role in motor planning and preparation. We think this is important for several reasons. First, it suggests strong commonalities with the mammalian motor system which exhibits activity associated with motor planning and preparation. Neurons in the mammalian cortex have also been found to be involved in preparation independent of performance (Economo et al., 2018), as described here in populations of HVC-RA neurons. Second, our findings indicate that we can observe the activity of a small population of neurons and potentially track the decision-making process in the seconds before a zebra finch volitionally begins to engage in courtship singing. Third, our data suggests the intriguing possibility that we can predict if the bird will sing 2-3 seconds prior to singing, a time-frame several times longer than it takes for descending motor commands to influence muscle groups associated with the motor control of song. In sum, our findings expand our knowledge of HVC-RA neuronal activity, how it relates to mammalian motor systems, and point to this network as a core network for examining how birds decide to sing. This opens an important new avenue for future studies in the field.

Essential revisions:

The authors must carefully reconsider and present what is novel relevant to the current literature, and their interpretation of those findings (particularly in reference to Okubo et al., 2016; Day et al., 2009; Goldberg et al., 2010 and Kao et al., 2008). The authors first say that studies in the past have not found pre-song activity in HVC, and that it was thought to only function at the moment of singing for sequences of vocalizations, citing single unit studies. Then they more correctly claim later in the paper, that electrophysiology studies have found premotor activity in HVC, using multi-unit recordings. What the authors need to do is be fair to the field of the neurobiology of motor behavior of song behavior, in particular, and mention a well-known principle in neuroscience, that neural activity in brain areas that controls movement of muscle activity precedes the movement/muscle activity. This is one reason why HVC was called a premotor nucleus.

We have edited the Introduction of the manuscript to include a discussion of this previous work, as requested. We hope that this will better place the functional significance of our results in the broader context of the published literature.

We have added the following paragraph to the Introduction:

“Neuronal activity related to motor planning and preparation has been associated with accurate production of volitional motor movements(Churchland et al., 2010a; Svoboda and Li, 2017) but is still poorly described in the context of initiating precise neural sequences for motor behaviors, like those exhibited in HVCRA neurons. Although it is not known if HVCRA neurons exhibt activity related to motor planning and preparation, previous studies have identified anticipatory or preparatory activity in other classes of HVC neurons and in other regions of the songbird brain(Goldberg et al., 2010; Goldberg and Fee, 2012; Kao et al., 2008; Keller and Hahnloser, 2009; Rajan, 2018; Roberts et al., 2017). HVC contains interneurons and at least three classes of projection neurons, including neurons projecting to the striatopallidal region Area X (HVCX), neurons projecting to a portion of the auditory cortex termed Avalanche (HVCAv), and the aforementioned HVCRA neurons that encode precise premotor sequences necessary for song production(Akutagawa and Konishi, 2010; Mooney and Prather, 2005; Roberts et al., 2017). Multi-unit recordings from HVC, which are typically dominated by the activity of interneurons, show increases in activity tens to hundreds of milliseconds prior to singing(Crandall et al., 2007; Day et al., 2009; Rajan, 2018). Calcium imaging from HVCAv neurons and electrophysiological recordings from HVCX neurons indicate that they also become active immediately prior to song onset (Rajan, 2018; Roberts et al., 2017). These data are consistent with recordings from the downstream targets of HVCAv and HVCX neurons. Portions of the auditory cortex (Keller and Hahnloser, 2009) and the basal ganglia pathway involved in song learning show changes in activity immediately prior to singing (Goldberg et al., 2010; Goldberg and Fee, 2012; Kao et al., 2008). Given this background, and that ~50% of HVCRA neurons may not exhibit any activity during singing (Hamaguchi et al., 2016; Long et al., 2010), we sought to examine if the precise neural sequences associated with song arise as part of larger changes in activity among populations of HVCRA neurons.”

We apologize for any confusion about the claims in the Introduction of our manuscript and have worked to make it as clear as possible. In our original submission we state that past studies have not found pre-song activity in HVC-RA neurons. In the Introduction of our original submission we state:

“HVCRA neurons are thought to be exclusively active during vocal production, yet ~50% of recorded HVCRA neurons do not exhibit any activity during singing (Hahnloser et al., 2002; Hamaguchi et al., 2016; Kozhevnikov and Fee, 2007; Long et al., 2010; Lynch et al., 2016), leaving the function of much of this circuit unresolved.”

To avoid any confusion, we have now amended this statement to read:

“HVCRA neurons are thought to be exclusively active during vocal production in waking adult birds, yet ~50% of recorded HVCRA neurons do not exhibit any activity during singing (Hahnloser et al., 2002; Hamaguchi et al., 2016; Kozhevnikov and Fee, 2007; Long et al., 2010; Lynch et al., 2016), leaving the function of much of the HVCRA circuitry unresolved.”

HVC-RA neurons are at the top of the descending motor pathway necessary for birds to produce their song. From our reading of the literature, it was not previously known if these neurons exhibit activity associated with motor planning or motor preparation. We only cite single unit studies because those are the only studies that use antidromic identification to record from identified HVC-RA neurons and as we mentioned above, it is inappropriate to try and infer the activity of HVC-RA neurons using multi-unit recordings (biased toward interneuron activity) or recordings from other classes of HVC neurons (which have different activity profiles during song) or neurons in other regions of the song system because they are not necessary for adult song production. We hope that the additional text in the Introduction and Discussion section of our revised manuscript help to clarify distinctions between our work and what has been previously published and the relevance of this research to the field.

The authors introduced their topic in a very wide context of neurons controlling sequencing behavior in the brain, across vertebrates. But then neither in the Results section or Discussion section did they interpret their findings in that wider context. So, the reader is left with a myoptic view of the HVC song nucleus, and nothing of relevance to anything else. The authors should in the discussion say how they interpret their findings in the control of learned movements generally. HVC has been proposed to function like cells in the mammalian motor cortex (layer III neurons of M1 for example Pfenning et al., 2014) or premotor cortex (Brocas area for example Bolhuis et al., 2010). Based on these comparisons, would the authors propose or predict that mammalian motor cortex show preparatory activity seconds before a movement is performed? Hasn't such analyses been undertaken already in mammalian motor cortex for movement, at the multiunit and single unit levels? If the authors are going to brush broad strokes in their introduction, they must complete that in the Discussion section.

The Discussion section has been expanded and our research is framed in the larger context of sequencing behavior and results are interpreted in the larger context of learned motor movements.

The authors look like they combined multiple studies from different labs without an initial plan for those studies to be combined. What this leads to are different studies with different variables and definitions. This includes different definitions of a song bout, song phrase, pre- and post-singing times for defining a bout and for analyses; different species (zebra finch vs Bengalese finch); and different social context (with a female or singing alone). It is well known that these factors make a difference in analyses and neural activity in the brain. The authors need to be more consistent in their behavioral definitions, pre- and post-singing time periods across studies, and the social context used. They should consider doing some of their analyses over again with the same timing of 1, 2, or 5 seconds pre and post singing to make them more consistent, and not one set of times (5 seconds) for the GCaMP imaging and another (2 seconds) for the multiunit electrophysiology. Also, in some of their analyses they seem to not include introductory notes in the bouts/phrases, and sometimes do. If they do not, then they could get an artificial error of thinking they found activity before singing when it is really activity occurring during singing the introductory notes. In the video the HVC neurons seem to become active during production of the introductory notes.

We apologize for the lack of clarity in the original writing of the manuscript. Throughout our manuscript, in both zebra finches and Bengalese finches, we include introductory notes and calls as part of song vocalization, as our hypothesis is that HVC-RA neurons play a role in motor preparation and planning, we wanted to examine activity occurring prior to when the birds produce any vocalizations. We have verified that we include consistent definitions of song bout, song phrase, and pre- and post-singing times across the diverse array of experiments. For consistency, we report the same time window before and after song in whenever possible, and we have added text in the Results section clarifying how these time windows were chosen for different experiments. We have also included supplementary figures when available that describe multi-unit activity in HVC and single-unit activity in RA.

[Editors' note: further revisions were requested prior to acceptance, as described below.]

Both reviewers feel that the manuscript has been substantially improved but one of the reviewers still is concerned about some issues.

Two important (yet addressable) concerns remain.

1) A previous concern was how bout onsets were defined and how cage noise was defined. The authors responded that the reviewers should 'rest assured' that the bout onsets are correctly labeled.

The spectrograms in Figure 2A,C,E and many of the supplementary Figures (e.g. Birds 162048, 160046) still all have what look like vocalizations prior to the red lines marking the bout onset. Specifically, there are apparent harmonics in the spectrogram in the 2-8khz range that look a lot like calls/intro notes in both Figure 2C,E under areas called cage noise. Cage noise, like song, can be broadband, but unlike song has lots of power concentrated at <1kHz and – also unlike song, rarely has nice harmonics with fundamentals at ~kHz ranges. But the cage noise parts of most spectrograms does not have this spectral profile – unless the spectrogram is being bandpass filtered in a way that is not find mentioned in the Materials and methods section.

Also, in the supplemental figure showing Directed Singing 162048, the sound trace underneath the mic voltage signal is totally flat for the first seconds and the HVC activity is totally silent. Then there is a sudden onset of power in the mic signal, which by eye looks to be totally synchronous with the first HVCRA transient – the mic signal is then noisy for the next several seconds when there is HVC activity. A similar pattern is observed in 160046. Thus, in these examples HVC 'pre-bout' activity is associated with noises on the mic signal (reported to be cage noise). Overall, HVC activity is rarely observed when the mic signal is flat and is much more commonly observed when there is time-aligned power on the mic signal (called cage noise).

While I hesitate to be to nit-picky about this, the entire paper is about pre- and post-bout Ca2+ transients. Yet most spectrograms appear (by eye) to have vocalizations preceding and following defined bout onsets and offsets. Since we know that HVCRA neurons can discharge during calls and intro notes, a lot depends on the authors totally nailing beyond a shadow of a doubt what constitutes a bout onset and what constitutes peri-bout concentrated cage noise.

Guidance: One way to resolve this issue would be to simply plot beneath the unfiltered mic signal a mic signal that is equal to the ratio of power in the 0.2-1 Hz range to the power in the 2-8 kHz range. Cage noise is easily distinguishable from real vocalizations by its excess power at low (<1kHz) frequencies. To address this specific reviewer concern, expanded views of the spectrograms highlighted in the comments above would also help.

Whatever the authors do, the paper would be stronger – and better for posterity – if cage noise was convincingly/objectively defined and clearly distinguished from calls, and if the showcased examples of peri-bout HVC Ca2+ transients were not compromised by what many readers are bound to think are spectral elements in peri-bout periods that look like vocalizations.

In the interest of expediency, I have thought about an faster way that the authors could address the issue of cage noise and bout onset definition. It seems to me that if the authors simply provide the wav files that correspond to all of the spectrograms in Figure 2 and Figure 2—figure supplement 2 than any reviewer and reader could just download the song and listen to it him/herself.

We have gone through the figures referred to above and have provided high-resolution spectrograms and audio files for expanded sections before and between vocalizations. This data covers the examples shown in Figure 2 (new data is illustrated in Figure 2figure supplement 7 and associate audio files) and for files associated with directed singing trails 162048 and 160046 (expanded regions now shown in Figure 2—figure supplement 8 and associated audio files). The reviewer is correct that there appear to be harmonic structures within some of the periods before the songs illustrated in Figure 2. Based on careful monitoring of videos from the cage during the song trials from Figure 2 and based on the duration of these harmonics and lack of frequency modulation, we have identified them as female “distance calls” and “tet” calls. Distance calls and tet calls are sexually dimorphic in zebra finches. Both call types function as contact calls used during affiliative interactions, like the social interactions used in our study to elicit courtship song (ie directed song). Please see the recent paper by Julie Elie and Frederidic Theunissen (Elie and Theunissen, 2016) for a thorough description of call types in male and female zebra finches. We apologize for this oversight and have edited the relevant figures to depict where the female calls occurred.

In addition, we examined the frequency distribution of our microphone recordings and bandpass filtered them as described above by the reviewer. First, most of the power seems to be concentrated to less than 5 kHz, however, as the reviewer notes above, cage noise can be broadband and might explain how some cage noise can appear to look like harmonic structures in low-resolution spectrograms. We hope the included audio files enable readers to verify that noisy pre-song components are not vocalizations unless otherwise noted. When there are vocalizations in the audio files, they are easily discernable by ear.

2) While the authors have done a much better job in the introduction and discussion contextualizing their findings in the literature, in some places they are still mis-characterizing past work. I make some recommendations below that I hope will be considered.

Subsection “Peri-Song Activity in Populations of HVCRA Neurons”: The authors claim that the pre-bout burst in HVC is a "newly discovered activity profile." But that's not accurate. In fact, the very first example of a singing related projection neuron in Picardo et al. (Figure 1E) is a neuron that exhibits a pre-bout transient that precedes bout onset by several seconds – just like examples in this study. If the authors want to counter that the Picardo study imaged both HVCX and HVCRA neurons while this one specified HVCRA neurons; they may be technically accurate but it's still misleading to report a 'new activity profile' – especially to someone who is familiar with the Picardo work (which includes pre-bout activity seconds in advance of the bout), the Okubo and Rajan papers which also include pre-bout activity.

We have changed the first part of this sentence to eliminate the phrase “these newly discovered activity profiles”. The first part of this sentence now reads, “To better characterize the activity profiles of HVCRA neurons, …”.

We do however disagree with the reviewer’s interpretation of the Picardo study. In our opinion it is not possible to know if the activity profile in Figure 1E of the Picardo study shows a similar profile to what we describe in our study. First, it looks like the bird is vocalizing prior to the onset of the song. The bird appears to be producing 2 calls followed by introductory notes before the first motif and the activity of neuron #1 may well be active due to those vocalizations. Second, there is no description in the results or Discussion section of this putative pre-song activity. Therefore, it is not possible to draw conclusions about pre-song activity from this figure panel. Third, it is not known what cell type was being recorded. The study tries to focus on putative projection neurons, but this is largely determined by the activity patterns during singing. Without further support, we believe it is not appropriate to claim that panel 1E in the Picardo study provides the first description of an HVC projection neuron exhibiting pre-song activity.

Subsection “Peri-Song Activity in Populations of HVCRA Neurons”: I think including the word 'exclusively' is unnecessary and not quite accurate description of the papers cited. Just because a paper focuses entirely on the role of HVCRA activity in patterning vocalizations does not mean that the paper requires HVCRA activity to discharge exclusively during vocalization.

We have deleted this word.

Subsection “Peri-Song Activity in Populations of HVCRA Neurons”: The authors wonder why previous studies did not observe peri-bout activity. Yet other studies did observe peri-song activity; just less of it. Again, see Picardo et al., Figure 1; Okubo et al., etc, Rajan…

We have changed this sentence, so it specifies HVCRA neurons and studies of adult birds. It now reads. “These results also raise questions as to why previous studies in adult zebra finches have not identified peri-song activity in HVCRA neurons.”

Subsection “Peri-Song Activity in Populations of HVCRA Neurons”: The authors' explanations/conjectures here about why potentially pre- and post-bout discharge was 'missed' are too speculative and not quite accurate. The triggering with a buffer is not how ephys studies of HVC were done. In original experiments they were done with audio monitors so sparse discharges would not be missed. Antidromic identification also does not bias for discharge during singing (thus the discovery that many HVCRA neurons are silent). And during intracellular recordings from HVC in awake, singing birds – every recording is so precious and hard-earned and eyeballed in real time on a computer/oscilloscope that there is no way that pre-bout discharge would have been missed. In these studies, half of the HVCRA neurons were either silent or active during introductory notes (though this was not emphasized in the papers).

A real possibility that could be included here is that Ca2+ transients do not equal spiking and vice versa. Dopamine 'ramps' are observed in Ca2+ imaging of neurons and in fiber photometry but have never been observed in DA spiking (there was a long discussion of this issue at CoSyne 2018 meeting, it is covered in recent review by Josh Berke (Berke, 2018), and is relevant to ramping signals observed in the recently posted biorxiv paper from the Witten group (https://www.biorxiv.org/content/10.1101/456194v1)).

Also, inhibition can independently regulate Ca2+ influx and spiking – resulting in non-congruence between spikes and GCAMP Ca2+ signals (nicely shown in Owen, Berke and Kreitzer, 2018);

Ok. We have modified this section. We now acknowledge at the beginning that we are speculating. We also specifically cite the Hahnloser, 2002 study when we mention methods for making the initial recordings of HVC-RA neurons using short buffering windows. According to Richard Hahnloser, an author on this manuscript and the first author on the 2002 study, audio monitors were not used in the recordings reported in that study and it is not how he has done these types of recordings in the past. Different studies may have been carried out in different ways over the years, so we have only cited the Hahnloser, 2002 paper in this section. Lastly, we have also added the possibility that there could be incongruencies between spiking activity and calcium signals (citing the Kreitzer paper) and acknowledged that the relationship between spiking and calcium signals has not been tested during peri-song epochs in freely behaving birds.

Discussion section: Get rid of 'exclusively'.

OK.

Discussion section: Get rid of 'only'.

OK.

Discussion section: Again, the statement that pre-bout Activity in HVC was not previously observed is not accurate; see Picardo et al., Figure 1.

I could not find the statement the reviewer is referring to. The sentence in the first paragraph of the Discussion section reads:

“Moreover, we find that preparatory activity in HVCRA neurons precedes the pre-bout activity described in other classes of HVC neurons and other portions of the song system previously described in zebra finches (Danish et al., 2017; Goldberg et al., 2010; Goldberg and Fee, 2012; Kao et al., 2008; Rajan, 2018; Roberts et al., 2017; Vyssotski et al., 2016; Williams and Vicario, 1993) and described here in Bengalese finches (Figure 4).”

[Editors' note: further revisions were requested prior to acceptance, as described below.]

There is a remaining issue with interpretation and semantics of what is pre- and post-vocalization activity measured in HVC and the song system relative to prior studies, which is what has been causing confusion to the reviewers. From a semantic understanding, pre-song activity in HVC-RA or any neurons of the song system could be considered 1 ms, 10ms, 100ms, 1s, 10s, or 10 min before vocalizing. In all cases, activity increases occur before vocalizing. All are premotor changes in neural firing. The difference that the authors find is that some neurons continue to fire during the motif or syllable production that has been found in previous studies and others do not, the new finding of this study. These former are what the authors appear to characterize as 'during song neurons', even if they fire 1-500 ms before or after the sound itself. The later fire 500ms to 3,000ms before sound is produced, and some of these shut off during the moments of syllable production. The authors assume that the idea of the 1-500ms range of premotor activity is "baked in", and therefore they call them 'during song production/singing neurons'. But what is baked-in the neuroscience field is that there is premotor activity in motor pathways in the manuscript before sound production or other behavior production. Can the authors see the confusion here?

1) For all the calcium imaging data (Figure 1, Figure 2, Figure 3 and associated supplemental figures), we include all activity in the 5 seconds prior to the bird vocalizing as ‘pre-song’ and all the activity in the 5 seconds following song offset as ‘post-song’. We do not exclude the 500ms period just before and just after vocalization as separate data – it is all included in the pre or post-song.

2) For the multi-unit electrophysiological recordings from HVC (Figure 4) and RA (Figure 5) we exclude the 500ms just before and just after song from the spike rate (Figure 4G) and CV ISI calculations (Figure 5D) of pre and post song data. The exclusion of the 500ms windows is illustrated in panels 4A and 5A. We did this because activity in HVC and RA are known to increase in the last ~300ms prior to song onset and we did not want this activity to bias our measurements at longer times either before or after singing. In the context of this paper we show that HVCRA neurons are active on the order of seconds before singing. We were interested in seeing if we could detect early or late activity using multi-unit recordings in HVC and RA from another species. If we included the 500ms time-period just around the song in these measurements it would make our data look stronger, but it would not correctly address our question. Therefore, we have taken a conservative approach to analyzing this data. The synaptic delays associated with descending motor commands are thought to be on the order of 25-50ms in the songbird system – well within the 500ms time-window excluded from this analysis.

3) Lastly, in our calcium imaging data set from HVC-RA neurons we found less than 4% of activity falling within last 100ms prior to song onset, suggesting that >96.2% of activity described here falls outside of the time-window thought to be involved in providing descending motor commands controlling vocalization. This supports our arguments that HVC-RA neurons play a role in motor planning or preparation rather than merely reflecting synaptic delays associated with singing. We touch on these ideas in the Results section. In subsection “Peri-Song Activity in Populations of HVCRA Neurons” we say: “The sequence of syllables in zebra finch song is stereotyped and unfolds in less than a second. Like other rapid and precise motor movements, song may benefit from motor planning and preparatory activity unfolding on much longer timescales than the synaptic delays associated with descending motor commands, which in zebra finches are estimated to be ~25-50ms (Amador et al., 2013; Fee et al., 2004)” In subsection “Peri-Song Activity in Populations of HVCRA Neurons” we say: “The vast majority of pre-song activity we describe occurs hundreds of milliseconds to seconds prior to vocalizations, suggesting a role in planning or preparation to vocalize (96.2% of calcium events occurred more than 100 ms prior to vocal onset and 65.6% occurred more than 1 second prior to vocal onset).”

To be less confusing to readers, the authors should include a table that defines the different neuron categories according to their temporal firing patterns, and then use those definitions consistently throughout the paper. Example terms that need such definitions in the table are: peri-song neurons (presumably within 5-0.5s firing before or after the first or last song syllable, respectively); pre-song neurons (presumably firing ~0.5s to 1ms before each syllable or the first song syllable); post-song; pan-song, etc.

We now provide Supplementary file 2 with definitions for the three categories of HVC-RA neurons described in this paper.

The real novelty of this study is finding neurons of the peri-song neurons as defined above, and not even just HVC-RA projecting neurons, but the fact that any such neurons exist at all in the song system. The paper needs to highlight this novelty more clearly. To do so, this difference in timing of neuron population firing and pattern should be stated in the Abstract, end of Introduction and in sentences of the results when this topic is discussed.

We believe that we have highlighted these findings in the submission in the Abstract, Introduction and Discussion section.

In the Abstract we say:

“Using cell-type specific calcium imaging we identify populations of HVC premotor neurons associated with the beginning and ending of singing-related neural sequences. We characterize neurons that bookend singing-related sequences and neuronal populations that transition from sparse preparatory activity prior to song to precise neural sequences during singing.”

At the end of the Introduction we say:

We show that ~50% of HVCRA neurons are active during periods associated with preparation to sing and recovery from singing and that their activity presages the volitional production of song by 2-3 seconds. One population of HVCRA neurons is only active immediately preceding and following song production, but not during either singing or non-vocal behaviors. A second population of neurons exhibits ramping activity before and after singing and can also participate in precise neural sequences during song performance.

At the beginning of the Discussion section we say:

“Our principal result is that neural sequences among HVCRA neurons emerge as part of orchestrated population activity across a larger network of HVCRA neurons. This activity can be correlated with motor planning and preparation prior to song initiation. Peri-song and pan-song HVCRA neurons forecast the start of singing. Peri-song HVCRA neurons are inactive during song, whereas pan-song neurons become heterogeneously active prior to time-locked sequential activity during song performances (Figures 2 and 3). Moreover, we find that preparatory activity in HVCRA neurons precedes the pre-bout activity described in other classes of HVC neurons and other portions of the song system previously described in zebra finches (Danish et al., 2017; Goldberg et al., 2010; Goldberg and Fee, 2012; Kao et al., 2008; Rajan, 2018; Roberts et al., 2017; Vyssotski et al., 2016; Williams and Vicario, 1993) and described here in Bengalese finches (Figure 4). This suggests that HVCRA neurons may seed network wide changes among other classes of HVC neurons and the song system more generally.”

We would also like to point out that Rajan, 2018 found that there is a subclass of HVC-X neurons that is active immediately before and after song, but not active during singing. Therefore, we do not feel that identification of any neuron with this firing pattern, as the Reviewing Editor suggests, is the real novelty of this study. We discussed this in detail in the response to our original submission. What we think is truly novel is that some HVC-RA neurons behave in a much more complicated fashion than previously known. We have identified 3 sub-populations of HVC-RA neurons that have activity patterns going well beyond sequencing of motor acts, the activity of two of these populations can start seconds before singing, and that precise sequences for song are embedded within this complex orchestrated pattern of activity. Moreover, the activity of HVC-RA neurons precedes the activity of all the pre-bout activity in the song system that has been reported to date (as far as we know).

More specifically, what appears most novel about the current study is that the authors find subpopulations of HVC-RA projection neurons that: (a) fire 1-3 seconds before singing starts [or after it ends]; (b) some of which turn off a few ms before or right at the time singing starts; and (c) are not correlated with as of yet identified pattern of singing. The other types of neural firing patterns, including pre- and post-song production, have been found in prior studies, and are confirmed in this study. This needs to be stated much more clearly in the manuscript.

We feel like these points have already been addressed in the paper.

In the Introduction we say:

“Neuronal activity related to motor planning and preparation has been associated with accurate production of volitional motor movements(Churchland et al., 2010a; Svoboda and Li, 2017) but is still poorly described in the context of initiating precise neural sequences for motor behaviors, like those exhibited in HVCRA neurons. Although it is not known if HVCRA neurons exhibt activity related to motor planning and preparation, previous studies have identified anticipatory or preparatory activity in other classes of HVC neurons and in other regions of the songbird brain(Goldberg et al., 2010; Goldberg and Fee, 2012; Kao et al., 2008; Keller and Hahnloser, 2009; Rajan, 2018; Roberts et al., 2017). HVC contains interneurons and at least three classes of projection neurons, including neurons projecting to the striatopallidal region Area X (HVCX), neurons projecting to a portion of the auditory cortex termed Avalanche (HVCAv), and the aforementioned HVCRA neurons that encode precise premotor sequences necessary for song production(Akutagawa and Konishi, 2010; Mooney and Prather, 2005; Roberts et al., 2017). Multi-unit recordings from HVC, which are typically dominated by the activity of interneurons, show increases in activity tens to hundreds of milliseconds prior to singing(Crandall et al., 2007; Day et al., 2009; Rajan, 2018). Calcium imaging from HVCAv neurons and electrophysiological recordings from HVCX neurons indicate that they also become active immediately prior to song onset (Rajan, 2018; Roberts et al., 2017). These data are consistent with recordings from the downstream targets of HVCAv and HVCX neurons. Portions of the auditory cortex (Keller and Hahnloser, 2009) and the basal ganglia pathway involved in song learning show changes in activity immediately prior to singing (Goldberg et al., 2010; Goldberg and Fee, 2012; Kao et al., 2008). Given this background, and that ~50% of HVCRA neurons may not exhibit any activity during singing (Hamaguchi et al., 2016; Long et al., 2010), we sought to examine if the precise neural sequences associated with song arise as part of larger changes in activity among populations of HVCRA neurons.”

Introduction: Making too strong of a statement that HVC-RA neurons are exclusively active during vocalization in awake birds. Even in awake birds, physiology and immediate gene studies have shown some activity to hearing songs in other species (even if birds own song) or eating (even if difficult to interpret gene expression studies). Instead of saying exclusively, how about predominantly.

We have now changed this sentence, so it reads:

“HVCRA neurons are thought to only be active during vocal production in waking adult birds, yet ~50% of recorded HVCRA neurons do not exhibit any activity during singing(Hahnloser et al., 2002; Hamaguchi et al., 2016; Kozhevnikov and Fee, 2007; Long et al., 2010; Lynch et al., 2016), leaving the function of much of the HVCRA circuitry unresolved.”

As we discussed thoroughly in our response to the first round of reviews, there is not any evidence that we have been able to find showing that HVC-RA neurons are active outside of singing in awake adult songbirds. The IEG study referenced by the editor does not show that HVC-RA neurons are active in response to eating, and there are not any electrophysiological recordings of identified HVC-RA neurons from any species showing that they are active outside of song in awake adult birds. We want to be correct and we have extensively searched the literature. If the Reviewing Editor can provide us with literature showing that HVC-RA neurons are active outside of singing in adult waking songbirds we would be happy to change our text, but as it stands, we do not feel that our statement is too strong. Moreover, changing the text to say “predominantly”, as suggested, implies that HVC-RA neurons in adult waking birds are known to be active outside of song.

Results section: The authors claim that the pre-bout burst in HVC is a "newly discovered activity profile." But that's not accurate when considering the 500 ms window of pre-bout activity. Here is where the authors would benefit by using more well-defined terminology that they show in a table as well as define the timing of what they mean by pre-bout activity.

This was corrected in an earlier version of our manuscript and is not included in the current submission.

Discussion section: It appears the Bengalese finch electrophysiology peri-song activity is within a shorter time window of several 100ms rather than 1-3 seconds as found with Ca2+ imaging in zebra finches. Is this a species difference, a physiology versus GCAMP sensing difference, or some other difference?

The differences between the zebra finch and Bengalese finch experiments may arise from using multi-unit electrophysiological recordings, which are dominated by the activity of interneurons, versus the cell type specific calcium imaging methods that allowed us to image exclusively from populations of HVC-RA neurons. We find that pre-song activity increased above baseline -1.47 ± 1.05 s prior to song onset in Bengalese finches, a timescale that closely matched the timing of peak calcium-event rates in peri-song neurons in zebra finches. Previous multiunit recordings from zebra finch HVC show elevated activity prior to song onset on the scale of tens to hundreds of milliseconds prior to song onset, while our recordings from Bengalese reveal increased activity ~1.5 sec prior to song onset. Therefore, species differences in preparatory activity at the level of HVC interneurons may also exist. Controlled multi-unit recordings from zebra finches and Bengalese finches would be needed to fully resolve this question.

The song system diagram of Figure 1A is appears to be simplified too much. It does not show MAN, the loop with AreaX and DLM, and the feedback to RA (as well as to HVC). Such additions to the figure would not cause more confusion, and in fact would add clarity and has been what is commonly shown in birdsong diagrams for several decades now. If it would make it easier for the authors, they could use and modify one of the figures showing these connections in the adobe illustrator versions on the Jarvis Lab website: http://jarvislab.net/summary-figure-originals/ These are more polished than the hand drawn one of this paper.

we have added a supplemental figure that includes the entire song system. We prefer to keep Figure 1A as it is, but the supplemental figure will provide interested readers with a complete diagram of the song system.

As the authors wrote in one of their responses, they should mention that the variability of peri-song neuron activity on different bouts is similar to variable pre-movement activity found primate M1 neurons. Such a comparison broadens the impact of the paper.

We added this to the Discussion section in our first resubmission. The second to last paragraph of the Discussion section reads:

“Neural activity associated with motor planning and preparation has been observed in motor and premotor cortices for a variety of different motor tasks in rodents and primates (Chen et al., 2017; Churchland et al., 2010a; Churchland et al., 2006a; Churchland et al., 2006b; Economo et al., 2018; Inagaki et al., 2019; Kaufman et al., 2014; Li et al., 2015; Li et al., 2016; Murakami and Mainen, 2015; Murakami et al., 2014; Tanji and Evarts, 1976), including vocalizations (Gavrilov et al., 2017). This activity is thought to reflect the decision to perform movements and is characterized by a high degree of variability from trial to trial. Changes in circuit dynamics function to shift the initial state of a network to levels that enable efficient and accurate motor performances and this activity can start to unfold seconds prior to movement initiation(Churchland et al., 2010a; Inagaki et al., 2019; Murakami and Mainen, 2015; Svoboda and Li, 2017). HVC is proposed to be analogous to the mammalian motor cortex (layer III neurons of the primary motor cortex) (Pfenning et al., 2014), or premotor cortex(Bolhuis et al., 2010). Our observation of preparatory activity seconds before the onset of courtship song in two different songbird species suggests that pre-movement activity is a common mechanism for ensuring the accurate production of volitional behaviors. In line with recordings of pre-motor activity in mammals, individual HVCRA neurons exhibit a high degree of trial to trial variability. Although zebra finch courtship song is famously stereotyped, there is a measurable degree of variability in the structure and duration of song motifs, bouts and phrases from trial to trial. Recording of preparatory activity across larger populations of HVCRA neurons may be useful for decoding this trial to trial variability in song structure, ultimately providing a predictive readout of impending behaviors.”

https://doi.org/10.7554/eLife.43732.053

Article and author information

Author details

  1. Vamsi K Daliparthi

    Department of Neuroscience, UT Southwestern Medical Center, Dallas, United States
    Contribution
    Conceptualization, Software, Formal analysis, Investigation, Writing—original draft, Writing—review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-6358-729X
  2. Ryosuke O Tachibana

    1. Department of Life Sciences, The University of Tokyo, Tokyo, Japan
    2. Institute of Neuroinformatics, University of Zurich/ETH Zurich, Zurich, Switzerland
    Contribution
    Formal analysis, Investigation, Writing—review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-4766-4504
  3. Brenton G Cooper

    Department of Psychology, Texas Christian University, Fort Worth, United States
    Contribution
    Formal analysis, Investigation, Writing—review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-3572-8822
  4. Richard HR Hahnloser

    1. Institute of Neuroinformatics, University of Zurich/ETH Zurich, Zurich, Switzerland
    2. Neuroscience Center Zurich (ZNZ), Zurich, Switzerland
    Contribution
    Conceptualization, Supervision, Funding acquisition, Investigation, Writing—review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-4039-7773
  5. Satoshi Kojima

    Department of Structure and Function of Neural Network, Korea Brain Research Institute, Daegu, Republic of Korea
    Contribution
    Supervision, Investigation, Writing—review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-0753-4238
  6. Samuel J Sober

    Department of Biology, Emory University, Atlanta, United States
    Contribution
    Formal analysis, Funding acquisition, Investigation, Writing—review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-1140-7469
  7. Todd F Roberts

    Department of Neuroscience, UT Southwestern Medical Center, Dallas, United States
    Contribution
    Conceptualization, Supervision, Funding acquisition, Investigation, Writing—review and editing
    For correspondence
    todd.roberts@utsouthwestern.edu
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-0967-6598

Funding

National Institute of Neurological Disorders and Stroke (R01NS108424)

  • Brenton G Cooper
  • Richard HR Hahnloser
  • Todd F Roberts

National Institute on Deafness and Other Communication Disorders (R01DC014364)

  • Todd F Roberts

National Science Foundation (IOS-1457206)

  • Todd F Roberts

Swiss National Science Foundation (31003A_127024)

  • Richard HR Hahnloser

Swiss National Science Foundation (31003A_156976)

  • Richard HR Hahnloser

National Institute of Neurological Disorders and Stroke (R01NS084844)

  • Samuel J Sober

National Institute of Neurological Disorders and Stroke (R01NS099375)

  • Samuel J Sober

Korea Brain Research Institute Basic Research Program (18-BR-01-06)

  • Satoshi Kojima

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

The authors thank Joseph Takahashi for generous use of a miniscope for calcium imaging experiments, Jennifer Holdway, Maaya Ikeda, Devin Merullo, Brad Pfeiffer, Alynda Wood, and Lei Xiao for comments on the manuscript and discussions, the Genie Project (Janelia HHMI) for development of calcium indicators, Takaki Komiyama and Andrew Peters for analysis codes, Andrea Guerrero for laboratory support and animal husbandry, and Jacque Dukes for assistance with data analysis.

Ethics

Animal experimentation: Experiments described in this study were conducted using adult male zebra finches and Bengalese finches ( >90 days post hatch). During experiments, birds were housed individually in sound-attenuating chambers on a 12/12 h day/night schedule and were given ad libitum access to food and water. All procedures were performed in accordance with established protocols approved by Animal Care and Use Committee's at UT Southwestern Medical Centers (2016-101562), Texas Christian University, Emory University, and the Korea Brain Research Institute. Research conducted by our colleagues in Korea was under IACUC-15-00028 and research at Emory was under 2003538.

Senior Editor

  1. Eve Marder, Brandeis University, United States

Reviewing Editor

  1. Erich D Jarvis, The Rockefeller University, United States

Reviewer

  1. Franz Goller, University Utah, United States

Publication history

  1. Received: November 17, 2018
  2. Accepted: June 10, 2019
  3. Accepted Manuscript published: June 11, 2019 (version 1)
  4. Version of Record published: June 25, 2019 (version 2)

Copyright

© 2019, Daliparthi et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 985
    Page views
  • 180
    Downloads
  • 0
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Download citations (links to download the citations from this article in formats compatible with various reference manager tools)

Open citations (links to open the citations from this article in various online reference manager services)