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

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.


27
The sequential activation of neurons is implicated in a wide variety of behaviors, ranging from episodic memory 28 encoding and sensory processing to the voluntary production of skilled motor behaviors 1-10 . Neural sequences 29 develop through experience and have been described in several brain areas, including the motor cortex, 30 hippocampus, cerebellum, and the basal ganglia 3,7,11-20 . Although computational models provide important insights 31 into circuit architectures capable of sustaining sequenced activity 9,10,20-24 , our understanding of sequence initiation 32 and termination is still limited.

34
The precise neural sequences associated with birdsong provide a useful biological model for examining this issue. Motor planning and preparation activity are associated with accurate production of volitional motor movements 8,30 43 but are poorly described in the context of precise neural sequences or in the production of song. Here we show that 44 ~50% of HVC RA neurons are active during periods associated with preparation to sing and recovery from singing. One 45 population is only active immediately preceding and following song production, but not during either singing or non-46 vocal behaviors. A second population of neurons exhibits ramping activity before and after singing and can also 47 participate in precise neural sequences during song performance. Recordings from downstream neurons in the 48 motor cortical nucleus RA reveal neural activity prior to song initiation and during song termination. The control of 49 respiratory timing is essential for song 31 , and our measurements of respiratory activity suggest that pre-singing 50 activity in HVC RA neurons functions to coordinate changes in respiration necessary for song initiation. From these 51 findings, we argue that subpopulations in HVC encode the neural antecedents of song that drive recurrent pathways 52 through the brainstem to prepare the motor periphery for song production.

56
We used miniscope calcium imaging to examine the activity of populations of HVC RA neurons in singing zebra 57 finches 32,33 . A total of 223 HVC RA neurons were imaged during production of 1,298 song syllables from 6 birds (30 58 song phrases across 18 imaging trials, EDTable 1). To selectively target HVC RA neurons, we combined retrograde viral 59 expression of cre recombinase from injections into RA with viral expression of cre-dependent GCaMP6s from 60 injections into HVC (Figure 1a-b and legend, see Methods) 33 . We confirmed the identity of imaged neurons using 61 conventional retrograde tracing, anatomical measures of neuronal features, and post-hoc histological verification.

62
We found that this approach exclusively and uniformly labeled populations of HVC RA neurons (Figure 1c-f).

64
To elicit courtship singing, we presented male birds with a female and imaged HVC RA neurons during song 65 performance (EDVideo 1). Given the slow decay times of calcium signals relative to singing behavior, we defined 66 neuronal activity by the rise times of calcium events that were >3 standard deviations (SD) above baseline (Figure 1g showing HVC RA neuron somata (green) and their outputs (magenta) to the downstream motor nucleus RA. AAV9-Flex-CAG-79 GCaMP6s was injected into HVC (green syringe) and AAV9-CAG-Cre was injected into RA (magenta syringe) to selectively label 80 HVC RA neurons. c) In vivo two-photon maximum density projection of retrogradely labeled HVC RA neurons expressing GCaMP6s.

116
To better characterize these newly discovered activity profiles, we indexed the song and peri-song activity of all 117 HVC RA neurons throughout a day of singing (phrase index: range -1 to +1, with neurons exclusively active outside of 118 singing scoring -1 and neurons active only during singing +1, Figure 2c-d, see Methods). We found that HVC RA phrase 119 indices were not uniformly distributed (χ 2 (7, N = 223) = 46.3, p = 7.6 x 10 -8 , Chi-square goodness of fit test), with a 120 significant fraction (36%) falling at the extremes of this scale (Figure 2d) Another possibility is that peri-song activity is unrelated to singing and merely reflects low levels of spontaneous 148 activity intrinsic to HVC RA neurons. This was not the case, however, as HVC RA neurons were largely inactive outside of 149 pan-song intervals and were significantly more active during the peri-song periods than baseline (baseline calculated 150 from periods ≥10 s removed from periods of singing or calling, p = 8.4 x 10 -5 , Chi-square = 18.78 Friedman test, 151 baseline = 12.9% ±5.7 SD of fluorescence values normalized to song, pre-song = 26.9% ±14.7, and post-song = 30.2% 152 ±10.7). In addition, we examined the amplitudes of peri-song calcium events and found that they were larger than 153 events occurring during song (t = 3.2769, p = 0.0012, two-tailed t test, 279 fluorescence peaks measured from 154 neurons active during both peri-song and song, pan-song neurons with phrase indices between -0.18 to 0.18). We 155 also asked whether peri-song activity might relate to factors other than singing. We examined trials in which birds did 156 not sing to female birds but found no HVC RA neuron that responded solely to presentation of the female or during 157 non-song related movements of the head, beak, or throat, such as during eating, grooming, and seed-shelling ( Figure   158 2e, EDFig. 6). Indeed, our populations of HVC RA neurons only became substantially active prior to singing or calling.

159
Moreover, we found peri-song activity in the moments before and after undirected song bouts produced in isolation 160 (EDFig. 6 -7), suggesting that peri-song activity is unlikely to be solely associated with extraneous courtship behaviors 161 such as courtship dance.

163
We next examined the possibility that HVC RA neurons play a role in motor planning or preparation as birds prepare to 164 sing. We found pre-song activity in 28/30 song phrases analyzed. In the two instances when we did not detect any 165 pre-song activity, less than 6 HVC RA neurons were active within our imaging window during singing, indicating that  Figure 5c). We measured changes in RA 293 neuron activity in the period just prior to song initiation and just after the conclusion of each song bout (between 0.5 294 and 2.5 s before/after the first/last song syllable). As expected, we did not find substantial differences in spike rates 295 between non-singing and singing states (not shown) but found that the CV ISI for RA neurons changed significantly 296 during song, reaching higher values during pre-song, song, and post-song epochs as compared to spontaneous 297 activity (Figure 5d, p<0.

311
Pre-song activity could reflect motor planning (changes in network activity independent of changes in the motor 312 periphery) and/or motor preparation that functions to coordinate changes in the motor periphery as birds prepare to 313 sing. Song is a respiratory behavior that is primarily produced during expiration and silent intervals in the song 314 correspond to mini-breaths, which are rapid, deep inspirations 31,47 . How birds plan to sing or prepare the respiratory 315 system to sing is poorly understood, but there is evidence that prior to song onset, oxygen consumption decreases 316 and respiratory rate increases 43 . To explore the time course of changes in respiratory patterns in more detail, we 317 used air sac pressure recordings in singing zebra finches (Figure 6a-c become heterogeneously active prior to time-locked sequential activity during song performances (Figures 2 & 3).

351
Preparatory activity in HVC RA neurons precedes the pre-bout activity observed in other classes of HVC neurons in 352 zebra finches and Bengalese finches (Figure 4), suggesting that HVC RA neurons may seed network wide changes in 353 activity 39-41 . The rigid stereotypy of singing behavior enables comparisons from different levels of the nervous system 354 and periphery. We find that preparatory activity in HVC RA neurons drives descending motor commands via RA and 355 motor movements that set the stage for producing song (Figures 5 & 6)    window. The field of view that matched the 2-photon images was identified and the focal plane that enabled the 444 largest number of neurons to be in focus was selected. Dental acrylic was used to fix the baseplate in the desired 445 position and any exposed skull was covered with dental acrylic. Once the dental acrylic dried, the microscope was 446 removed from the baseplate and the bird was allowed to recover overnight. About 30 minutes before the birds' 447 subjective daytime, the microscope was attached to the counterbalance (Instech) with enough cable to allow the bird 448 to move freely throughout the cage. The microscope was then secured to the baseplate with a setscrew. The bird 449 was allowed to wake up and accommodate to the weight of the microscope over the next 2-3 days. imaging was performed at 30 frames per second (fps), at 1080x1920 resolution, Gain was set to 4, and Power was set 462 to 90% for all birds, behavioral videos were collected at 24 fps. Calcium imaging data and behavioral data was 463 synchronized using start of calcium imaging on a frame by frame basis.

530
We measured the SNR of events occurring during song for a subset of pan-song and song neurons. The SNR was 531 calculated as a ratio of peak fluorescence for each song event per neuron to the average fluorescence from baseline 532 period within the trial (as above, 5s of fluorescent activity that was ≥10s removed from periods of singing or calling).

533
We determined the average SNR for each neuron and examined differences between pan-song and song neurons.

551
Recorded signals were bandpass filtered (0.3-5kHz) and negative signal peaks exceeding 4 SD of spontaneous activity 552 (no-song period separated more than 10 seconds from nearest song bouts) were interpreted as multi-unit spikes. In 553 total 38, 62, and 212 song onsets, and 42, 23, and 280 offsets were identified in these birds, respectively. We 554 produced firing rate-traces from each electrode channel with 10ms resolution and averaged them across song 555 renditions. After smoothing by moving average with 500ms window, the averaged firing rates were normalized into 0 556 to 1 as spanning between respective mean firing rates during spontaneous activity and singing (0-3s after song onset) 557 to remove any bias among channels to obtain the general trend of onset-and offset-related firing across channels 558 and birds.

559
The significance of activity elevation during pre-song, song, and post-song periods from the spontaneous activity 560 level was tested by Wilcoxon signed-rank test with significant level at 0.05 after Bonferroni correction for multiple 561 comparison.

563
Onset of pre-song and offset of post-song activity were estimated for each channel as the smoothed spike rate 564 trajectory was exceeded a threshold which was defined as mean + 2 SD of the spontaneous spike rate.

577
To analyze the variation in inter-spike-interval (ISI) in different time periods (Figure 5), we restricted our analysis to 578 cases in which we collected at least one recording that included the relevant song epoch. "Spontaneous" epochs 579 were sampled from neural activity recorded more than 10 sec after the nearest song bout. "Pre-song" activity was 580 sampled from between 2.5 and 0.5 s prior to the first song syllable or introductory note. "Song" activity was sampled 581 from the onset of the first song syllable until the offset of the last syllable in a bout. "Post-song" activity was sampled 582 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

588
Subsyringeal air pressure was recorded from six birds in directed singing conditions. Directed song was defined as a 589 female presented in an adjacent cage during a two-hour recording period. Data from four of the birds were re-590 analyzed from a previously published study (Cooper & Goller, 2006) and data from two additional birds were 591 collected to replicate the effects observed in the previously collected data 70 . As described in (Secora et al. 2004), 592 each bird was accustomed to carrying a pressure transducer that was held in place on the bird's back with an elastic 593 band 71 . To facilitate relatively free lateral and vertical movement in the cage, the weight of the transducer was offset 594 by a counter-balance arm. Subsyringeal air pressure surgery was performed after birds sang while carrying the 595 pressure transducer. Prior to insertion of the air pressure cannula, animals were deeply anesthetized as verified by 596 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 597 micro-dissecting forceps, and a flexible cannula (silastic tubing, OD 1.65 mm, 6.5 cm length) was inserted into the 598 body wall and suture was tied around the cannula and routed between the 2 nd and 3 rd ribs to hold it in place. The 599 skin was sealed to the cannula with tissue adhesive. The free end of the cannula was attached to the pressure 600 transducer. This allowed for measurement of relative subsyringeal air pressure changes inside the thoracic air sac 601 before, during, and after spontaneously generated song events. Birds were monitored following surgery until they 602 perched in the recording chamber.