Preparing to sing: respiratory patterns underlying motor readiness for song

Abstract

Evidence for motor preparation and planning comes from neural activity preceding neural commands to activate the effectors; such preparatory activity is observed in pallial areas controlling learned motor behaviors. Vocal learning in songbirds is an example of a learned, sequential motor behavior that is a respiratory motor act and where there is evidence for neuromuscular planning. Respiration is the foundation of vocalization, elucidating the neural control of song motor planning requires studying respiratory antecedents of song initiation. Despite the importance of respiration in song production, few studies have investigated respiratory antecedents of impending vocalizations. Therefore, we investigated respiratory patterns in male zebra finches (Taeniopygia guttata) and Bengalese finches (Lonchura striata domestica) prior to, during, and following song bouts. In both species, compared with quiet respiration, song respiratory patterns were generated with higher amplitude, faster tempo, and ∼70% of the respiratory cycle is in the expiratory phase. In female-directed and isolation song, both species show a change in the respiratory tempo and the proportion of time spent inhaling prior to song. Following song, only zebra finches show systematic changes in respiratory patterns; they spend a greater proportion of the respiratory cycle in the expiratory phase for 1 s after song, which is likely due to hyperventilation during song. Accelerated respiratory rhythms before song may reflect the motor preparation for the upcoming song production; species differences in preparatory motor activity could be related to the degree to which motor planning is required; finally, song termination may be dictated by respiratory demands.

NEW & NOTEWORTHY Motor planning for vocal production in birdsong manifests as an adaptation of respiratory characteristics prior to song. The songbird’s respiratory system anticipates the upcoming song production by accelerating the respiratory tempo and increasing the proportion of time spent inhaling.

INTRODUCTION

Preparation for motor action is a fundamental feature of learned motor behaviors. The act of speaking requires a complex interplay between diverse motor systems regulating respiratory, vocal organ, and upper vocal tract movements (1–3). Motor cortex damage in humans leads to a loss of speech, but similar damage in nonhuman primates does not affect their vocalizations (4); because nonhuman primates do not show vocal learning as humans do, this pattern suggests that these descending motor cortical circuits are specialized for producing learned vocalizations (5, 6). There are two regions of activation in motor cortex during language production; one region is associated with laryngeal muscle activation and the other somatic respiratory circuits (4, 7, 8). Although much is known about the descending control of speech from cortical and subcortical circuits, less is known about the preparation and planning underlying for the impending vocalization. Motor planning deficits are characteristic of autism, speech apraxia, and developmental coordination disorder (9–12). Developing our understanding of normal and disordered speech and language requires study of motor preparation in nonhuman animals. Songbirds are an animal model for vocal learning because humans and songbirds share similarities in the developmental features of vocal imitation, genes for vocal learning, sound production mechanism, and analogous pallial circuits controlling song learning and production (13–19). However, only recently have studies in songbirds begun to evaluate how birds prepare to sing (20–23).

Learning to control sound production effectors is a necessity for an animal to exhibit vocal learning. Mammalian and avian vocal organs (the larynx in humans and the syrinx in aves) produce sound by converting aerodynamic energy into acoustic energy (15, 24, 25). A positive air pressure head is generated by contraction of expiratory muscles (26). In birds, this muscle activation leads to sternal compression of the rib cage, and by extension, the interconnected air sacs that are closely associated with the body wall (for review, see Ref. 27). The process of air sac compression is frequently compared with a “bellows” like action; but it should be noted that this compression is not uniform, rather it is generated with a rapid, millisecond scale temporal modulation of muscle activation and corresponding air pressure modulation (26, 28, 29). In oscines (songbirds), the airflow passes two soft tissue masses at the tracheal bronchial junction, there is one labial pair within each bronchus and this allows for two-voice sound production. The soft tissues are set into vibratory motion through a combination of Bernoulli forces and elastic recoil (15, 24, 30, 31). Thus, for an animal to produce sound using myoelastic-aerodynamic mechanism of sound production it is a prerequisite that pallial circuits must transiently modulate the reflexive medullary respiratory rhythm to produce sound using volitional respiration (5, 27, 32, 33). Respiratory feedback must constrain sound generation, not only how long sounds can be produced but also when sound production can be initiated based on current respiratory conditions. Therefore, exploring how the nervous system readies the respiratory system for phonation is central to developing our understanding of motor preparation underlying vocal production (20), and more generally for understanding the evolution of pallial control of learned vocal behaviors (7, 34).

Few studies have evaluated changes in respiration before singing. Early work in Waterslager canaries (Serinus canari) noted an increase in respiratory cycle tempo before song (35). However, the temporal changes in respiratory cycles leading up to song onset were not quantified. Studies investigating oxygen consumption or heart rate in zebra finches (Taeniopygia guttata) showed increases in respiratory tempo before singing (34, 36). But it is not clear whether the increase in oxygen consumption or heart rate before song was due to increased respiratory activity in preparation for song or movements associated with the presentation of the conspecific for female-directed singing. Consistent with the suggestive evidence of preparatory peripheral motor patterns, there are changes in neural activity in the avian song system, including the avian basal ganglia and pallial areas RA (robust nucleus of the arcopallium) and premotor cortical analog HVC (letters used as proper name) (20–23, 37). In vivo single-photon calcium imaging and multiunit electrophysiological recordings in zebra finches and Bengalese finches (Lonchura striata domestica) suggests there is a progression of preparatory activity transitioning from pallial premotor to motor areas and then to downstream respiratory motor systems. Cell-type-specific calcium imaging identified a subclass of neuron projecting from HVC to RA that were active 3–5 s before song but not during song; further extracellular RA neural activity began to change toward burst firing patterns typical of song circa 0.5–2.5 s before song. Finally, there was an acceleration of the respiration rhythm in the last second before song initiation (20).

Although these earlier studies have provided evidence of preparatory neural and respiratory activity, there are multiple issues that potentially limit the conclusions of these experiments. First and foremost, the previous research has not controlled for the presence of conspecifics used to promote vocalizations. Zebra finches sometimes engage in dance movements during and preceding song. The visual display includes hopping movements in a side-to-side fashion as they approach the female, and they also engage beak swiping behaviors (38). Furthermore, the female may produce calls before song. The role of these movements and auditory stimulation has not been controlled for in these previous studies and both could affect the ongoing respiratory tempo. Second, while preparatory neural activity in Bengalese finches has been investigated, corresponding changes in respiratory patterns have not been studied. Therefore, this experiment sought to explore these issues by comparing directed and undirected singing conditions to control for the male dance movements and female-generated auditory stimulation. In addition, peri-song respiratory features were quantified in Bengalese finches and zebra finches to expand the investigated species. Finally, a qualitative analysis of respiration was undertaken to provide a cycle-by-cycle analysis of pre-song respiratory patterns that foretell song initiation. Here, we define preparatory behavior to be changes in respiration that could facilitate the initiation of singing. We find that regardless of singing context or investigated species, birds show acceleration in the respiratory tempo and a switch toward increasing time spent inhaling, because song is produced with an accelerated rhythm and distinct inspiratory pressure patterns, the results are highly suggestive of respiratory behavior that is preparatory in nature and can be foretelling of song initiation.

METHODS

Ethics Statement and Animal Husbandry

All experiments were conducted after receiving approval from the Institutional Animal Care and Use Committee at Texas Christian University. Eight zebra finches and 10 Bengalese finches were initially used in this study, and successful air pressure recordings that were used in the subsequent analyses came from six birds of each species; however only four of the Bengalese finches sang in the directed condition (described later). All finches were maintained on a 14:10-h light cycle, given seed and water ad libitum, vegetables mixed with vitamins every other day, and egg once a week. Before experimental testing, birds were housed in communal cages with 4–12 birds per cage. During experiments, the birds were individually housed in small cages (31.8 × 10.5 × 25.4 cm) contained in a sound-attenuating box (78.7 × 33 × 33 cm). All four sides of the sound-attenuating box were lined with 1-in. thick acoustic foam (Auralex Acoustics, Indianapolis, IN) to dampen acoustic reflections. A microphone was suspended 14 cm above or in front of the center of the perch.

Surgical Procedure

Air pressure was recorded by inserting one end of a small cannula (Silastic tubing, 0.76 mm I.D., Dow Corning, Midland, MI) into the left or right anterior (or cranial) thoracic air sac. Under isoflurane (1%–2%) anesthesia, a small opening was made in the body wall below the last rib. The cannula was sutured to the body wall and a small amount of tissue adhesive was used to seal the skin to the tubing (34, 39). The free end of the cannula was connected to a pressure transducer (Fujikura XFHM 02PGR; Santa Clara, CA). After surgery, topical application of a 50/50 mix of lidocaine (2%) and bupivacaine (0.5%) was used as a postsurgical analgesic. The pressure transducer was held in place by a Velcro elastic backpack and the transducer was centered between the wings. A counter-weighted balance arm was used to offset the weight of the pressure transducer and accompanying wires. Birds were monitored after surgery until they could perch, hop, and move freely in their cage while carrying the pressure transducer and associated wires.

Data Collection

Signals from the pressure transducer mounted on the bird’s backpack were connected to a DC amplifier (Gain = 100, 6 kHz low-pass filter, Brownlee Precision, Model 440, Santa Clara, CA) and song was recorded with respiration. A microphone (Earthworks TC20, or SR20, Milford, NH) was placed 14 cm in front or above the bird. The microphone signal was amplified and high-pass filtered (300 Hz) before digitization. Song and air pressure were digitized with a National Instruments Analog to Digital conversion Board (National Instruments, NI USB-6251, Austin, TX) and Avisoft Recorder software (Avisoft Bioacoustics, Berlin, Germany). Data were stored in multichannel, 16-bit WAV files with a 22.05 kHz sample rate. Song and air pressure were digitized whenever air pressure exceeded a user-defined threshold. The threshold was set at circa twice the peak of the quiet respiration. Five seconds (s) of data were saved before the threshold onset, and 5 s after the threshold was no longer exceeded. Thus, files minimally contained 10 s of respiration and the microphone recording. Files without song but with periods of interspersed calling were collected, as well. This data collection procedure resulted in a collection of data files that included song, calls, defecation, and periods of quiet respiration surrounding these events.

Directed Song Procedure

Following a previously established protocol for directed song recordings with air pressure measurements (34), directed song was elicited by placing a female bird in a small cage across from the male bird for 2 h. Undirected song was collected by having the bird sing in isolation for 2 h before or after the female was presented for directed singing. The 4-h recording period began within 3 h of room illumination, and the order of treatment conditions (directed and undirected) was counter-balanced across birds.

Song Identification

Figure 1A illustrates an air pressure recording of a zebra finch song bout. The zebra finch song bout consists of one or more repetitions of the bird’s motif. A motif is composed of a repeated sequence of song syllables (3–8 syllables). Bengalese finch song is much more variable than zebra finches but is produced with a semi-fixed sequence of song syllables that are repeated multiple times to form a song bout (39, 40). To replicate previous work, songs selected for the analysis met the criterion that there were no calls produced for 4 s before song onset (20). The exclusionary criterion was used to eliminate variable changes in respiration due to producing calls before song. Ambient pressure was estimated as the midpoint between the positive and negative pressure peaks during quiet respiration (Fig. 1, A–C, red line). Supra-atmospheric pressure corresponds to expiration and inspiration is defined as periods of subatmospheric pressure (Fig. 1, B and C). Song onset was defined as the first inspiratory pulse preceding the first introductory note of the song; the inspiratory pulse onset is defined as subatmospheric pressurization before the production of the expiratory pulse generating the vocal introductory note of the bird’s song (20). Following existing nomenclature, syllables were defined as expiratory pulses producing the unique acoustic structure of the bird’s song (36) and silent gaps corresponded to mini-breaths (Fig. 1, C and D). Mini-breaths are shorter duration and generated with a greater negative pressure than quiet inspiration (35) (Fig. 1C). Song termination was defined as the period of respiration occurring after the song bout. Only songs that had 4 s of respiration without calls or other vocal behaviors before song and 3 s of data without calls following song were analyzed for the peri-song analysis of respiratory cycles.

Figure 1.

Air-sac pressure measurements of respiratory activity. A: a representative example of the air-sac pressure pattern of a zebra finch is shown. Periods of quiet respiration before (pre-song, green background) and after the song (post-song, orange background) surround the song bout. Ambient air pressure is indicated by a horizontal red dashed line. B: pre-song air-ac pressure during quiet respiration is shown in greater detail. A respiratory cycle is defined from the beginning of an inspiratory pulse (IP_i) to the end of the consecutive expiratory pulse (EP_i). Inspiratory pulses correspond to subatmospheric air sac pressure and expiratory pulses are supra-atmospheric pressure events. Two inspiratory pulses (IP_i and IP_i+1) are displayed in blue. An example of an expiratory pulse is indicated in dark green (EP_i). C: the part of the air-sac pressure time series during song is displayed in greater detail, with the corresponding spectrogram of the song shown in the lower panel. Each of the expiratory pulses (an example shown in green) generates one syllable of the song motif. Inspiratory pulses during song are called mini-breaths (MB, shown in blue), because of they are short duration and larger amplitude compared with peri-song respiration.

Following the approach used in previous work, a respiratory cycle was defined as beginning with inspiration and concluded with the following expiration (20). Onset of the respiratory cycle occurred when there was 5 ms or more of continuous data below ambient pressure and expiration was the continuous period of supra-atmospheric pressurization. The duration of the respiratory cycle, duty cycle (the percentage of cycle time exhaling), and average amplitude (full wave rectified signal) was calculated for each cycle. Songs without calls for 4 s before song onset were visually identified from the collection of all files and used for the analysis of respiratory timing before song. The measured parameters for respiratory cycles were calculated for up to 4 s before song onset, during song, as well as after song ended.

Peri-Song Respiratory Tempo Analysis

The song files selected for a repeated-measures analysis were aggregated into 1 s bins for up to 4 s before song, and for 3 s time bins following song termination. A single average value was calculated for each bird within each time bin. For each investigated species, the average cycle duration, duty cycle, and average amplitude were statistically evaluated using a one-way repeated-measures analysis of variance (ANOVA), with a Greenhouse–Geisser correction for heterogeneity of variance when Mauchly’s test of sphericity was significant. If the omnibus test was statistically significant, pair-wise comparisons were made across 1 s time bins with the least-significant difference method employed because of the a priori hypothesis that there would be a change in respiratory patterns. Statistical significance of α = 0.05 was used unless otherwise indicated, and the effect size using partial eta squared (${\eta }_{\text{p}}^2$) is reported (IBM SPSSS Statistics, v.26). In this analysis, the first 4 s before song were analyzed as well as the last 3 s following song termination. The data include six zebra finches in both directed and undirected singing conditions, four Bengalese finches that sang directed song, and five Bengalese finches that sang in the undirected condition. Two zebra finches (ZF1 and ZF2) and two Bengalese finches (BF1 and BF2) sang more vigorously than the other birds, which allowed for a more detailed investigation into cycle-by-cycle characteristics of these individuals. A cycle-by-cycle analysis consisted in the classification of each respiratory cycle with respect to the beginning of the song bout. The notation “−1” means the first respiratory cycle before the song bout, and in general “−m” means the m-respiratory cycle pre-song. Respiratory cycles greater than 5 s from song start were used as control data files; these files came from a combination of data files including calling and singing episodes with the restriction that quiet respiration was not temporally associated with song onset. A Kruskal–Wallis analysis of variance was carried out between the respiratory durations of the control group and the respiratory cycles before song. In the cases for which Kruskal–Wallis yielded a significant result, a Dunn’s test was pursued to explore the significance of the comparison of the cycle “−1” respect to the control respiration with a Bonferroni correction for multiple comparisons. This analysis was performed in Python using SciPy Stats (41) and scikit-posthocs (42).

Ordinal Pattern Analysis of Respiratory Cycles

From the air-sac pressure time series, the duration of each respiratory cycle was computed. This resulted in a sequence of time ordered events: {ΔT_i}_{1 ≤ i ≤ n}, in which ΔT_i represents the duration of the ith respiratory cycle in the time series. A symbolic analysis known as ordinal pattern analysis was carried out (43, 44). Each consecutive sequence of L elements (L = 2 or 3, in this study), like the {ΔT₁, ΔT₂} or the {ΔT₁, ΔT₂, ΔT₃} sequence, was ordered from smallest to largest element in the sequence. For example, for length 3, if ΔT₂ < ΔT₁ < ΔT₃, then this sequence corresponds to an ordinal pattern (word of length 3) “1 0 2.” In this word, the element “i” represents the (i + 1)th element of the sequence, so “0,” “1,” and “2” represent ΔT₁, ΔT₂, and ΔT₃, respectively. The process is repeated for every sequence of three consecutive elements extracted from the time series, and the probabilities of observing each word were calculated.

By analyzing the ordinal patterns of length 2, the relative durations of two consecutive cycles were quantified and compared with a control case outside of song. Only two different words can be generated, “0 1” and “1 0.” The word “0 1” represents a case for which the first element of the sequence of cycle durations is shorter than the following, and the word “1 0” shows the opposite case. This mechanism to create a word out of the sequence only relies on the relative comparison between the elements of the sequence. When words of length 3 are considered, six different ordinal patterns are possible, corresponding to all the permutations of the elements 0, 1, and 2. The sequences that occurred at least 5 s before song were used to calculate control probabilities.

A χ² test was used to statistically test for differences in the probability of occurrence of the words generated with the respiratory cycles and control probabilities. Control probabilities were generated in two different ways. The first control was generated by a uniform distribution of the words, and in the second method, the control probabilities were generated from data 5 s or more from song.

Quantification of Phase Synchronization across Multiple Song Bouts

To quantify the degree of similarity between song starts across different bouts, a phase synchronization was analyzed (45, 46). Each of the song bouts was considered as the output of an oscillatory system, then an order parameter was introduced following Kuramoto’s order parameter (47). The air-sac pressure time series for each of the song bouts was translated in time to place the start of the expiration of the first syllable of the first motif at time zero. The air-sac pressure data were then converted into a phase variable by means of an analytical signal method. The instantaneous phase of the air-sac pressure time series was determined by a Hilbert transformation of the signal (48). In this way, a phase θ_i(t) is obtained for each of the air-sac pressure time series with a song bout. To quantify the phase coherence across the multiple presentations of the song at different times, the following order parameter was introduced:

$$r\left(t\right)=\left|\frac{1}{N}\sum _{(k,j)}^{}{e}^{i({\text{θ}}_k\left(t\right)-{\text{θ}}_j\left(t\right))}\right|$$

The summation runs through all possible pairs of different air-sac pressure time series with song bouts, and N is the total number of pairs. This order parameter approaches 1 for times in which the instantaneous phases of the air-sac pressure are similar, and 0 when they are not similar. For this analysis, the exclusionary criteria based on the lack of calls for 5 s before song onset was not used.

Song Respiratory Relationships

To examine respiratory correlations present during song between different elements of the respiratory program (expiratory pulses and mini-breadths), two different analyses were carried out. In the first analysis, for each animal, the air-sac pressure corresponding to each syllable was segmented out of the time series and mean values over this set were used to graphically represent the syllable. The assumption here is that each mini-breath is composed of three sources of variance: 1) the inspiration allows for the recovery from the lost air supply that was used to produce the preceding syllable, 2) the inspiration is preparatory in nature for the upcoming exhalation that will produce the next song syllable, and 3) error variance. The dataset of the expiratory pulses (generating song syllables) and mini-breaths (gaps) from multiple song renditions for each of the birds was analyzed by using linear mixed-effects models (49, 50). Linear mixed-effects models (LME) are designed for the analysis of nonindependent data. In our case, we want to extract correlations between sequences of mini-breaths and expiratory pulses. Because the data are coming from different animals, we account for the dependency by treating it as a random effect. LME models were implemented in Python using Statsmodels (51).

The second analysis was carried out for ZF1, ZF2, BF1, and BF2. In this case, we considered each presentation of the song motif (data not aggregated) so that we could examine the variability present in each syllable and the correlations between air pressure variability and motif sequence. The goal here was to understand motor control during motif production and to explore characteristics of air pressure that could be predictors of song termination.

Temporal Modulation of Expiratory Pulses

To explore species differences in song motor production, features of the expiratory pulses generating song syllables in zebra finches and Bengalese finches were compared. Five repetitions of each syllable of the bird’s motif were saved and the average duration (s), amplitude (a.u.), and temporal modulation were measured. Similar to analysis done by Méndez et al. (52) to quantify the degree of temporal modulation in a syllable as a proxy for motor complexity, the expiratory pulse corresponding to each syllable was segmented out of the air-sac pressure time series ({p_i}_{1 ≤ i ≤ N}: segment of time series). An ordinary least squares regression was implemented to fit a quadratic polynomial (f) to the segment of the air-sac pressure. The average per point of the residual sum of the squares was used to quantify the temporal modulation:

$$\frac{1}{N}\sum _{i=1}^N{(f_i-p_i)}^2$$

where f_i is the evaluation of the fitted function f at the point i and p_i is the air pressure at point i.

RESULTS

In this study, we used respiratory dynamics to explore the existence of preparatory respiratory activity for song generation and the changes in respiratory features during song in zebra finches and Bengalese finches. Simultaneous recordings of sound and air-sac pressure were used. When respiration close to song production was analyzed, the period of quiet respiration before song was described as pre-song, while the quiet respiration after song as post-song (Fig. 1A). During quiet respiration (Fig. 1B), the periods of time while the air-sac pressure were lower than ambient pressure corresponded to inspiratory pulses (IP_i) and above ambient pressure to expiratory pulses (EP_i). During song production (Fig. 1, C and D), expiratory pulses had larger amplitude and were longer in duration, and mini-breaths were shorter duration and larger amplitude compared with peri-song respiratory patterns. Air pressure patterns allowed for the analysis of motor patterns generating quiet respiration and also to clearly separate respiratory events generating song.

Peri-Song Respiratory Analysis

Respiratory data from different zebra finches (n = 6) were aggregated and several features were chosen to reflect the changes between pre-song, song, and post-song periods in directed and undirected singing contexts (Fig. 2). To complement the visualization of the significance of the data shown in Fig. 2, Supplemental Table S1 contains the means and standard error of the means for each of the quantifiers for the considered time windows. Furthermore, the unaggregated data for each individual animal is displayed in Supplemental Fig. S1. Figure 2, A–C, shows the aggregated data (mean and box-plot) of duration, duty cycle, and the average amplitude of the air-sac pressure before, during, and following directed song. The respiratory cycles occurring in the last second before song production were shorter in duration [F(1.470,7.348) = 7.67, P = 0.02, ${\eta }_{\text{p}}^2$ = 0.605]. Post hoc comparisons showed that the −1 s time bin was significantly different than −2, −3, and −4 s (P = 0.03, respectively). The duty cycle also approached a significant reduction in the breaths leading up to song [F(1.436,7.182, = 4.077, P = 0.074, ${\eta }_{\text{p}}^2$ = 0.449]. The average amplitude, in contrast, did not change significantly across time [F(1.956,9.829) = 0.0325, P = 0.726, ${\eta }_{\text{p}}^2$ = 0.051].

Figure 2.

Zebra finch respiratory antecedents and corollaries of song-related respiration. Respiratory patterns are characterized using duration (A: directed song, D: undirected song), duty cycle (B: directed song, E: undirected song), and air sac pressure average amplitude (C: directed song, F: undirected song) of the respiratory cycles. Green background indicates pre-song data, while post-song data is shown with an orange background. For each animal, the data were classified based on the time from the beginning or end of the corresponding song. If the respiratory cycle happened within the first second before the song, this event was included in “−1 s.” In general, if a respiratory cycle was between m and m − 1 seconds before song, then the event was included in “−m s.” Similarly for post-song, if a respiratory cycle happened between m and m + 1 s, the event was included in “m s.” Considering all these events for each animal, an average per animal was calculated. Box-plots show the distributions of average values across each time window for the different animals (n = 6). The group mean is indicated by a blue point in each time window.

In the undirected singing condition (Fig. 2, E and F), cycle duration decreased significantly in the lead-up to song [F(1.126,5.632), = 6.503, P = 0.04, ${\eta }_{\text{p}}^2$ = 0.565], with the −1 s time bin significantly faster than −2 and −3 bins (P = 0.04, 0.01, respectively), and approached significance for −4 s bin (P = 0.057). The duty cycle = changed significantly as respiration moved toward song [F(3,15) = 11.79, P = 0.001, ${\eta }_{\text{p}}^2$ = 0.702]. The last second before song was significantly different than the −4 s time bin (P = 0.007). Similar to the directed singing context, the average amplitude of quiet respiration did not change significantly in the lead-up to song [F(1.069,5.345) = 0.872, P = 0.399, ${\eta }_{\text{p}}^2$ = 0.148].

The same analysis was carried out for Bengalese finches. The group average results are shown in Fig. 3 and individual birds are displayed in Supplemental Table S2 and Supplemental Fig. S2. Bengalese finches did not sing readily in the presence of the female bird; the data presented are from the four birds that sang in the directed context. The respiratory tempo in the lead-up to song did not increase significantly [F(3,9) = 2.302, P = 0.180, ${\eta }_{\text{p}}^2$ = 0.404]; however, duty cycle decreased significantly [F(3,9) = 4.318, P = 0.038, ${\eta }_{\text{p}}^2$ = 0.590], with the last second time bin significantly compared with the −2 s bin (P < 0.047) and the −4 s time bin approached significance (P = 0.066). Average amplitude did not change significantly in breaths leading up to song [F(3,9) = 1.757, P = 0.225, ${\eta }_{\text{p}}^2$ = 0.369]. Although tempo did not change significantly, this may be due to an accelerated respiratory rhythm during baseline breathing due to the presence of a female conspecific and the small sample size limiting statistical power.

Figure 3.

Bengalese finch respiratory antecedents and corollaries of song-related respiration. Respiratory patterns are characterized using duration (A: directed song, D: undirected song), duty cycle (B: directed song, E: undirected song), and average amplitude (C: directed song, F: undirected song) of the respiratory cycle. Green background indicates pre-song data, while post-song data is shown with an orange background. Data were aggregated across different animals (n = 4) and times. For each animal, the data were classified based on the time from the beginning or end of the corresponding song. If the respiratory cycle happened within the first second before the song, this event was included in “−1 s.” In general, if a respiratory cycle was between m and m − 1 s before song, then the event was included in “−m s.” Similarly for post-song, if a respiratory cycle happened between m and m + 1 s, the event was included in “m s.” Considering all these events for each animal, an average per animal was calculated. In these plots, the distributions of average values across each time window for the different animals are represented as box-plots. The group mean is indicated by a blue point in each time window.

In the undirected context, Bengalese finches (n = 5) showed an acceleration of their respiratory tempo [F(3,15) = 5.085, P = 0.013, ${\eta }_{\text{p}}^2$ = 0.504], with the last second significantly faster than the −3 and −4 s time bins (P = 0.025, 0.018, respectively). The duty cycle decreased significantly as well [F(3,15) = 12.85, P = 0.001, ${\eta }_{\text{p}}^2$ = 0.711], and the last second time bin was significantly lower than all preceding times (P = 0.005). Average amplitude did not change significantly as respiration approached song initiation [F(3,15) = 0.754, P = 0.754, ${\eta }_{\text{p}}^2$ = 0.131].

When song respiration was compared with the preceding quiet respiration, it was clearly observed that song was composed of shorter respiratory cycles (Figs. 2, A and D, and 3, A and D) than quiet respiration in both investigated species. Furthermore, there was an increase in the fraction of time spent exhaling during a cycle (duty cycle, Fig. 2, B and E) and vocalizations were created by larger amplitude expiratory pulses that resulted in larger average air-sac pressure (Fig. 2, C and F and Fig. 3, C and F). After song performance in the directed song condition (Fig. 2, A–C, right), there was no significant change in cycle duration [F(2,10) = 0.509, P = 0.616, ${\eta }_{\text{p}}^2$ = 0.92], whereas duty cycle changed significantly [F(2,10) = 6.553, P = 0.015, ${\eta }_{\text{p}}^2$ = 0.567]. In the first second following song, duty cycle was greater than the following seconds (P = 0.05). Similar to pre-song, average amplitude did not change significantly following song termination [F(2,10) = 2.203, P = 0.161, ${\eta }_{\text{p}}^2$ = 0.306]. In the undirected condition (Fig. 2, D–F, right), cycle duration increased significantly across time bins following song termination [F(2,10) = 4.842, P = 0.034, ${\eta }_{\text{p}}^2$ = 0.492]. The first second after song was significantly faster than the cycle duration in the third second following song termination (P = 0.029). In the undirected condition, duty cycle and average amplitude did not change significantly [F(2,10) = 1.514, P = 0.266, ${\eta }_{\text{p}}^2$ = 0.232, F(2,10), = .391, P = 0.686, ${\eta }_{\text{p}}^2$ = 0.073]. In Bengalese finches singing in the directed and undirected conditions, post-song cycle duration [F(2,6) = 0.338, P = 0.726, ${\eta }_{\text{p}}^2$ = 0.101, directed; F(2,8) = 0.907, P = 0.441, ${\eta }_{\text{p}}^2$ = 0.185, undirected], duty cycle [F(2,6) = 0.650, P = 0.555, ${\eta }_{\text{p}}^2$ = 0.178, directed; F(2,8) = 0.573, P = 0.586, ${\eta }_{\text{p}}^2$ = 0.125, undirected], and average amplitude [F(2,6) = 0.006, P = 0.994, ${\eta }_{\text{p}}^2$ = 0.002, directed; F(2,8) = 0.687, P = 0.687, ${\eta }_{\text{p}}^2$ = 0.146, undirected] did not change significantly.

The aggregate data demonstrated that the most consistent and robust changes in respiration across species occurs in the lead-up to song production. To understand in greater detail the observed trend toward a faster respiratory rhythm closer to song production observed in the aggregate data, an analysis of the respiratory cycles on a cycle-by-cycle basis was carried out. In this analysis, the respiratory cycles were not aggregated, instead each cycle was numbered with respect to the start of the song bout. For example, the cycle “−1” refers to the respiratory cycle immediately before the song bout. A control dataset was created by considering the respiratory cycles occurring at least 5 s before the song bout. To carry out this analysis, large datasets for two zebra finches and two Bengalese finches that sang robustly were used (ZF1, ZF2, BF1, and BF2). The cycle-by-cycle analysis of these animals indicated that all animals with the exception of one of the Bengalese finches have a cycle “−1,” which was shorter than the control (Fig. 4, A–C) (ZF1: Kruskal–Wallis H = 35.4, P = 1e-6, Dunn’s test for “−1” and control P = 2e-7. ZF2: Kruskal–Wallis H = 14.5, P = 0.01, Dunn’s test for “−1” and control P = 0.003. BF1: Kruskal–Wallis H = 35.2, P = 1e-6, Dunn’s test for “−1” and control P = 1e-4.). Supplemental Fig. S3 shows the data for BF2. This observation was in agreement with the previous observation made with the aggregated data. Figure 4D shows the occurrence of each of the respiratory cycles (blue scatter points) for the zebra finch represented in panel A.

Figure 4.

Prior to directed song onset, the last respiratory cycle before song has the largest changes in cycle duration. A–C: box-plots comparing the duration of quiet respiration occurring at times at least 5 s before song (control) to the duration of quiet respiration of the respiratory cycles closer to the song start. In these plots, the last respiratory cycle before directed song is represented by “−1,” the previous cycle by “−2,” etc. The results for two zebra finches are shown in A and B, and for a Bengalese finch in C. For each of these cases, the duration of the last cycle before song is significantly different to the control (Kruskal–Wallis ANOVA followed by Dunn’s test for ZF1: P = 2e-7, ZF2: P = 0.003, BF1: P = 8e-5). There is a clear overlap between the two distributions. This point was analyzed in detail for the animal in A, and the results are shown in D–F. D: blue scatter points represent the duration of pre-song quiet respiration events versus the time before the beginning of song at which each cycle starts. A windowed average of the data (time windows of 0.2 s) is represented in red, showing the existence of shorter respiratory cycles closer to the beginning of song. E: average pressure (P_avg.) versus duration for each respiratory cycle. Red scatter points indicate the control events, and blue is used to indicate the last respiratory cycle before song. F: kernel density estimation (KDE) of the respiratory duration. Red: control, blue: last respiratory cycle before song.

Even though the distribution of respiratory durations from the “−1” cycles were for most of the animals statistically different from the duration for the control sets, clear overlaps between the distributions were observed. This overlap is not only present for the respiratory durations but also for the air-sac pressure amplitude (Fig. 4E). Another way to clearly visualize the overlap between the “−1” cycle and control distributions is by using kernel density estimations (51) for each of the datasets (Fig. 4F). The variability in normal quiet respiration leads to the overlap in the distributions. Thus, within the range of normal respiratory behaviors, it is possible that the faster rhythms are selected rather than being actively generated for the purpose of respiratory preparation for the upcoming vocal acts.

Ordinal Pattern Analysis

The existence of correlated sequences of respiratory cycles was explored by means of ordinal pattern analysis (Fig. 5). Two different ordinal pattern lengths were analyzed (2 and 3). Figure 5 shows the results of applying ordinal pattern analysis to two zebra finches (Fig. 5, C and D, ZF1 and ZF2 respectively) and two Bengalese finches (Fig. 5, E and F, BF1, and BF2, respectively). To analyze the significance of these observations, two different sets of ordinal patterns were used. In the first case, ordinal patterns were generated out of a uniform distribution, and in the second case, the patterns are created considering periods of quiet respiration 5 s or greater from song. The results of χ² tests are summarized in Table 1. For all the animals, the most likely patterns present in our datasets were the “1 0” and the “2 1 0” patterns that point toward a possible acceleration of respiratory rhythm close to song. In several of the comparisons, this observation reached statistical significance (Table 1).

Figure 5.

Ordinal pattern analysis of respiratory cycles. A: an example of an air-sac pressure time series of a zebra finch during quiet respiration (control data). The horizontal dashed red line indicates ambient pressure. Several consecutive respiratory cycles are color coded and the durations of the cycles are named ΔT_i. For this time series, sequences of three consecutive respiratory cycles are shown (sequence 1, 2, and 3). Each of these sequences is converted into an ordinal pattern (word). B: the corresponding words of length 3 for the sequences displayed in (A) are shown. Considering the first sequence {ΔT₁, ΔT₂, ΔT₃}, the (1 0 2) word was generated because ΔT₂ < ΔT₁ < ΔT₃. For each sequence, a geometrical representation of the sequence is also illustrated to represent the relative amplitude of each element of the sequence. Each dot represents a ΔT_i and consecutive elements are connected by a line. These dots are color coded according to (A). In C–F, the probabilities of finding each possible word of length 2 and 3 are shown for two zebra finches (C and D) and two Bengalese finches (E and F). In all the word probabilities plots, red indicates quiet respiration temporally distant from song occurrence (control) and blue indicates the word probabilities generated by considering the sequence of 2 or 3 respiratory cycles previous to song start. The horizontal black line indicates equal probabilities for all the words. For each animal, the frequencies of the words obtained from respiratory cycles previous to song start were compared with the control data and to uniform distribution of word frequencies using χ² tests. ZF1 comparison against uniform distribution: for L = 2, P = 2e-4, and for L = 3, P = 7e-9. ZF1 comparison against quiet respiration ≥ 5s song: for L = 2, P = 1e-4, and for L = 3, P = 3e-6. ZF2 comparison against uniform distribution: L = 2, P = 0.2, and L = 3, P = 0.02. ZF2 comparison against quiet respiration ≥ 5 s song: L = 2, P = 0.1, L = 3, P = 0.1. BF1 comparison against uniform distribution: L = 2, P = 0.005, and L = 3, P = 7e-5. BF1 comparison against quiet respiration ≥ 5 s song: L = 2, P = 0.01, and L = 3, P = 2e-4.

Table 1.

Word analysis chi square results

		Word Length 2	Word Length 3
Zebra finch 1	Uniform distribution	P = 0.0002	P = 0.7e-9
	Quiet respiration	P = 0.0002	P = 0.3e-6
Zebra finch 2	Uniform distribution	P = 0.2	P = 0.02
	Quiet respiration	P = 0.1	P = 0.1
Bengalese finch 1	Uniform distribution	P = 0.005	P = 7e-5
	Quiet respiration	P = 0.005	P = 0.1

Phase Synchronization of Song

Zebra finches produce a highly stereotyped sequence of expiratory pulses and mini-breaths that creates a song motif, and the first motif of song bout is preceded by one or more introductory notes. We explored whether the respiratory phase generating introductory notes are synchronized before song initiation. The first motif of the bout was aligned such that the beginning of the first expiration of the first syllable of the motif corresponded to time zero. The song alignment is illustrated in Fig. 6A. Polar plots of the phases of air-sac pressure (for different presentations) for times before and after zero showed that the phase synchronization starts at time zero (Fig. 6, C and D) but not before (t = −0.2 s). This observation for particular times was then represented for all times close to song production by the calculation of the order parameter r (Fig. 6, C and D).

Figure 6.

Increasing respiratory synchronization during song initiation in zebra finches. A: the zebra finch song bouts were aligned so that the beginning of the expiration of the first syllable of the motif (start time zero); several song bouts are shown to represent this process. The vertical dashed red line indicates start time zero. B: air-sac pressure patterns were converted to respiratory phases by a Hilbert transformation. The respiratory phases for each song bout produced by the zebra finch are shown at two different times, before the motif (t = −0.2 s) and after (t = 0.1 s), showing a consistent production of motor gestures only at the beginning of the motif production. Each red dot represents data from a song bout. C: vector strength (order parameter) is shown to indicate the degree of phase locking between the multiple presentations of the song for the zebra finch. D and E: the vector strength and the respiratory phases before and after the onset of the song motif for another zebra finch.

Song Respiratory Relationships

Due to the species differences observed in post-song respiratory patterns, potential differences in song respiratory patterns were explored. Quiet respiration is produced by highly variable respiratory cycle durations, but it is well known that, especially in zebra finches, there is a switch to highly stereotyped sounds and respiratory motor patterns during song. To examine this point in greater detail, we analyzed the way in which the consecutive leading mini-breaths, expiratory pulses, and the following mini-breaths are correlated in zebra finches. We examined these correlations with an aggregated and more global approach integrating data from multiple animals, and for the zebra finches with larger dataset (ZF1 and ZF2), we also performed an analysis of the specific correlations and variability for the particular animal.

For the first analysis, we identified repeated syllables in zebra finches (n = 6) and Bengalese finches (n = 5) and the associated leading and following mini-breath. To understand the correlations between expiratory pulses and the leading and following mini-breaths, we calculated several linear mixed-effects models. Table 2 shows a summary of the results for the different models. Both Zebra finches and Bengalese finches presented statistically significant correlations between the expiratory pulses and the leading mini-breath (Fig. 7 and Table 2). Zebra finches also showed significant correlations between the expiratory pulses and the following mini-breaths. The general observation that the leading inspiration is correlated with the following syllable production may reflect motor planning within the song to produce the upcoming vocalization and recovery from the lost air supply. To quantify differences between zebra finch and Bengalese finch song respiratory patterns, the duration, amplitude, and temporal modulation of the expiratory pulses were compared. For the mini-breaths, only duration and amplitude were compared between the species. Zebra finches produce syllables with longer duration, higher amplitude, and more temporal modulation than Bengalese finches (Fig. 8). These differences in respiratory patterns between the species may generate the species differences in the post-song respiratory patterns observed in the aggregate data.

Table 2.

Linear mixed-effects model within song respiratory patterns results

Species	Dependent Variable	Independent Variable	Slope [95% Confidence Interval]	P Value
Zebra finch	EP Avg Amp	Lead MB Avg Amp	−0.778, [−1.423, −0.134]	P = 0.018
Zebra finch	EP Avg Amp	Follow MB Avg Amp	−0.811, [−1.407, −0.215]	P = 0.008
Zebra finch	Follow MB Avg Amp	Lead MB Avg Amp		P = 0.1
Bengalese finch	EP Avg Amp	Lead MB Avg Amp	−1.027, [−1.768, −0.285]	P = 0.007
Bengalese finch	EP Avg Amp	Follow MB Avg Amp		P = 0.1
Bengalese finch	Follow MB Avg Amp	Lead MB Avg Amp	0.463, [0.236, 0.690]	P = 0.0001

EP Avg Amp, expiratory pulse average air pressure amplitude; Follow MB Avg Amp, following mini-breath average air pressure amplitude; Leading MB Avg Amp, lead mini-breath average air pressure amplitude.

Figure 7.

Preceding and following inspiratory pressure patterns are highly correlated with the vocal expiration during song. The average amplitude of the expiratory pulses (EP P_avg.) versus the average amplitude of the leading inspirations (Lead-MB P_avg.) are displayed for zebra finches in A and for Bengalese finches in D. The plots of the average amplitude of the expiratory pulses (EP P_avg.) versus the average amplitude of the following inspirations (Follow-MB P_avg.) are shown for zebra finches in B and for Bengalese finches in E. In C and F, plots of the average amplitude of the following inspiratory pulses (Follow-MB P_avg) versus the values for the leading inspiratory pulses (Lead-MB P_avg.) are shown for zebra finches and Bengalese finches respectively. In all the panels, each scatter point indicates the average over multiple presentation of one syllable. Different animals are represented in different colors. The results of linear mixed-effects (LME) models, when significant, are displayed as dashed black lines in each panel. Details of the LME models for each panel. For all the models, the animal ID was a random effect. A: Lead-MB P_avg. independent variable, EP P_avg. dependent variable. Coefficient: −0.778, P = 0.018, 95% confidence interval: [−1.423, −0.134]. B: Follow-MB P_avg. independent variable, EP P_avg. dependent variable. Coefficient: −0.811, P = 0.008, 95% confidence interval: [−1.407, −0.215]. C: Lead-MB P_avg. independent variable, Follow-MB P_avg. dependent variable. P = 0.1. D: Lead-MB P_avg. independent variable, EP P_avg. dependent variable. Coefficient: −1.027, P = 0.007, 95% confidence interval: [−1.768, −0.285]. E: Follow-MB P_avg independent variable, EP P_avg. dependent variable. P = 0.1. F: Lead-MB P_avg independent variable, Follow-MB P_avg dependent variable. Coefficient: 0.463, P < 1e-4, 95% confidence interval: [0.236, 0.690].

Figure 8.

Differences between zebra finch and Bengalese song respiratory patterns. Violin plots for the following characteristics are shown: mini-breath (MB) durations (A), mini-breath amplitude (B), syllable duration (C), syllable peak amplitude (D), and syllable air-sac pressure fluctuation (E). Zebra finch (ZF) data is displayed in light blue, and Bengalese finch (BF) data in pink. The results of the statistical comparisons between the species (Mann–Whitney test) are shown in each panel.

Figures 9 and 10 illustrate the two song motifs and the sequential syllables (letters) and the motif of the bird’s song for ZF1 and ZF2. We limited the sequential analysis to zebra finches because Bengalese finch song is composed of syllables with a variable transition probability between syllables. Greater variability in the respiratory cycle was observed at the beginning or at the end of the motif, but not during the production of intervening syllables. For example, the scatterplot in Fig. 9B graphically illustrates that variability in the respiratory duration increased dramatically when the last syllable of the motif was considered (purple points). The last syllable variability could be related to the end of the song, where the bird either terminates the song bout or transitions back to the first syllable of the motif. The last syllable is coded as the end of the song bout (squares) and end of the motif (circles). When the last syllable of the motif is considered, an increase in the variability in the following mini-breath duration was observed in both conditions. The variability in amplitude of the expiratory pulse and the following mini-breath shows a similar pattern as the duration of the pulses (Fig. 9C). The leading mini-breath is the most variable before the first syllable of the song in both duration and amplitude (Fig. 9, D and E). The coefficients of variation for each of the syllables of the motif are shown as insets in each of the panels. As song progresses from the first syllable to the middle and then toward the last syllable, there appears to be a transition from a more variable, to stereotyped, to variable motor program. Thus, even though the motor performance is highly stereotyped, there is a range of variation within the motif that has been uncovered by this analysis.

Figure 9.

Does variability in song respiratory cycles correspond to song termination? A: the song motif of a zebra finch (ZF1): air-sac pressure at the top and the spectrogram below. Different syllables are indicated with color-coded letters from A to G. The colors are used for the remaining panels of this figure to indicate the different syllables. Horizontal red dashed line indicates ambient pressure. B: for each expiratory pulse (EP) corresponding to each of the song syllables, the duration of the mini-breath after the expiratory pulse (Follow-MB) is plotted versus the expiratory pulse duration. Each dot corresponds to multiple presentations of the song. Syllable G could be the final syllable of the song bout (squared scatter point) or within a song bout (circular scatter point). Inset: coefficient of variation for the mini-breath after the expiratory pulse (CV Follow-MB) versus coefficient of variation of the expiratory pulses durations (CV EP) for each syllable. C: the average of the pressure during the mini-breath after each expiratory pulse (Follow-MB P_avg.) versus the average of the pressure of each expiratory pulse (EP P_avg.). Inset: coefficient of variation for these two quantities for each syllable. D and E: similar to B and C but considering the mini-breath before each of the expiratory pulses (Lead-MB). Syllable A could be in the first motif of the bout (squared scatter point) or during a motif not at the beginning of the bout (circular scatter point).

Figure 10.

Does variability in song respiratory cycles correspond to song termination? A: the song motif of another zebra finch (ZF2): air-sac pressure at the top and the spectrogram below. Different syllables are indicated with color-coded letters from A to E. The colors are used for the remaining panels of this figure to indicate the different syllables. Horizontal red dashed line indicates ambient pressure. B: for each expiratory pulse (EP) corresponding to each of the song syllables, the duration of the mini-breath after the expiratory pulse (Follow-MB) is plotted versus the expiratory pulse duration. Each dot corresponds to multiple presentations of the song. Syllable E could be the final syllable of the song bout (squared scatter point) or within a song bout (circular scatter point). Inset: coefficient of variation for the mini-breath after the expiratory pulse (CV Follow-MB) versus coefficient of variation of the expiratory pulses durations (CV EP) for each syllable. C: the average of the pressure during the mini-breath after each expiratory pulse (Follow-MB P_avg.) versus the average of the pressure of each expiratory pulse (EP P_avg.). Inset: coefficient of variation for these two quantities for each syllable. D and E: similar to B and C but considering the mini-breath before each of the expiratory pulses (Lead-MB). Syllable A could be in the first motif of the bout (squared scatter point) or during a motif not at the beginning of the bout (circular scatter point).

DISCUSSION

This study provides evidence for the existence of preparatory respiratory activity for song generation in two songbird species, zebra finches and Bengalese finches, in two distinct singing contexts. The song respiration was anticipated by a short period of respiratory adaption. This adaptation began to manifest clearly one respiratory cycle before the starting of the song. The most striking evidence of this adaptation was the acceleration of the respiratory rhythm, and also a transition toward spending more time inhaling than exhaling. Thus, animals switched to both a faster motor pattern more closely resembling song respiratory rates and with a bias toward inhalation allowing for the air supply necessary for producing a vocalization. Song termination patterns were not consistent between the two species. Zebra finches showed pronounced changes in post-song respiration, but a similar pattern was not observed in Bengalese finches. This species difference could be related to difference in song respiration, perhaps due to hyperventilation that occurs during a song bout in zebra finches (36).

In humans and other animal species, motor production is commonly anticipated by premotor activity that is not directly related to impending movement (53–56). Such preparatory activity is predictive in many cases of the upcoming motor act or cognitive decision (56, 57). For example, human neuroimaging has revealed that patterns of activity in frontoparietal network can predict an upcoming movement or a cognitive decision up to 7 s before the action or decision occurs (56, 57). Motor preparation has been revealed in a variety of species. For example, rodents take a deep inspiration before producing an ultrasonic vocalization (58). In zebra finches, introductory notes precede the first motif of the song; the gaps between introductory notes get shorter closer to the motif onset (59). Here, by examining the pressure patterns during quiet respiration, we show that this motor planning occurs up to three respiratory cycles before song. Although the highly variable pre-song respiration illustrates the challenges with using respiration to predict song occurrences, it is clear that there are highly significant changes in respiration that occur before song in both investigated songbird species.

The acceleration of respiratory rhythm for song preparation was observed in different behavioral singing contexts. The presence of an accelerated respiratory rhythm that occurs when the male bird is singing in isolation demonstrates that changes in respiration are not occurring as a result of auditory stimulation. Furthermore, dance movements associated with approaching the female bird are not driving the change in respiration (38). Finally, both the aggregated analysis was then followed by detailed characterization of two animals for each of the species that sang vigorously during experimental testing. In the aggregated dataset, a significant acceleration of the respiratory rhythm during the last second before song starts when birds sang to female and when they sang in isolation was observed (Figs. 2 and 3). The case studies in the subset of birds pointed toward the last respiratory cycle before song as the most pronounced and consistent change accounting for this respiratory acceleration (Fig. 4). To further test the sequential change in song acceleration, an ordinal pattern analysis of the case studies showed that the acceleration in a few cases started with a sequence of three respiratory cycles before song; the pattern “2 1 0” occurred before song ∼40% in these few instances (Fig. 5).

Just as quiet respiration is quite variable in duration and duty cycle of the rhythm, the pre-song period was also highly variable. The durations and duty cycles of the last respiratory cycles before the introductory notes were within the range of the variable respiratory patterns produced during quiet respiration outside of singing. During the highly variable quiet respiration, birds generated respiratory cycles with similar characteristics as the last cycles before song (Fig. 4, E and F). To develop algorithms that predict song onset, discriminatory patterns of change in respiration are required. The use of duty cycle and tempo are problematic because the changes that occur before song are also observed during quiet respiration. The overlap in the distributions of patterns during quiet respiration and peri-song respiration poses a challenge to the development of algorithms that predict song onset. One interesting option is that the acceleration in respiration that occurs during quiet respiration might reflect song initiation failures. It is entirely possible that birds prepare to sing, but do not ultimately initiate song during the overlap in song tempo between pre-song and quiet breathing respiration. Further quantification of highly variable respiratory patterns combined with neural recordings may help to differentiate between song-related preparatory patterns and variation in respiration that is entirely unrelated to song.

The transition from pre-song to song is characterized by respiratory tempo variability, as evidenced by the lack of a phase similarity across different presentations of the song bouts (Fig. 6). A high degree of similarity (synchronization index) started at the production of the first syllable of the motif in zebra finches. Intra-song motor control is not equally stereotypical. During the production of the motif, correlations between the consecutive respiratory elements of the song motif were observed (Fig. 8). Even during the production of the highly stereotyped motif, however, an increased variability was observed at the beginning and at the end of the motif sequence (Figs. 9 and 10). This positional variation in motor variance may be related to intrinsic biases in how syllables types are sequenced during the process of song learning (60). However, it is unlikely that the variability in song production is related to song termination, because the last expiratory pulse was not more variable when it preceded an upcoming syllable or the end of the song. It is perhaps more likely that the identifying signatures of song termination is related to chemosensitive feedback rather than motor production variance.

A central feature of learned vocalizations is that volitional respiration must reshape reflexive respiratory patterns to enable higher pressure and fine-grained temporal modulation of airflow, along with the faster duration of the respiratory cycle required to produce the intended vocalization. In songbirds, RA is a motor cortical analog that can directly modulate downstream respiratory centers. The projections from RA to the brainstem nuclei retroambigualis and parambigualis medularis can directly modulate expiratory and inspiratory phases of respiration (61, 62). Electrical stimulation of the pallial and brainstem respiratory areas during song causes an abnormal cessation of ongoing singing (63). HVC is a key premotor area for controlling the timing of song production (64–67), and one class of projection neurons from HVC innervates RA. Electrical stimulation of HVC neuronal populations reliably changed respiration in awake behaving birds (canaries and zebra finches) (33). The pattern of electrical stimulation in HVC entrained respiratory rhythms at a new phase relationship modulated by the stimulation pattern. This illustrates that HVC can directly modulate brainstem respiratory circuitry; this effect is likely occurring because of projections to RA. Further evidence for this view comes from the finding that neural activity in a subclass of neurons in HVC that project to RA (HVC_RA neurons) is correlated with the transition between quiet respiration and song (20). HVC_RA preparatory activity could be responsible for the respiratory acceleration observed in this study. Thus, for song initiation via volitional respiration, the electrical activation of HVC and correlated activity of HVC_RA neurons implicate this pallial structure as the likely driver of the changes in peri-song respiration that was observed in the current study.

Language production in humans requires a transfer from reflexive to volitional respiration. As a person speaks, the duration of the expiration producing sound is predicted by the depth of the inspiration preceding the vocalization (68, 69). Similarly, we observe that the magnitude and duration of the mini-breaths was highly correlated with the following vocalization as well as with the recovery from sound production (Fig. 7). The findings in this work and previous research illustrate a putative neural network that allows for the transfer between reflexive and volitional respiratory patterns in songbirds, an animal model used to explore speech production. In mammals, there is a phase of reflexive inspiration that restricts when such modulation of respiratory rhythmicity is physiologically possible (70). Similar restrictions on modulation of brainstem circuitry may also be true of avian respiratory patterns. Therefore, feedback from respiratory patterns via proprioceptive and chemosensitive feedback likely influences HVC neural activity to guide not only how long birds can sing (71, 72) but also when the bird can initiate their song. Such feedback is likely relayed through vagal afferents that terminate in the nucleus tractus solitarius (32, 73, 74). The brainstem modulation places the constraints on the neural circuits that control song via pallial connections back to the brainstem. Changes in respiratory patterns that proceed song are likely frequently occurring even in the absence of singing, the difficulty is disambiguating those changes that are related to song versus those changes that occur as a function of variable respiratory patterns for other non-song behaviors.

In summary, this study provides evidence that the neuromuscular control of song begins at least one respiratory cycle before song. In all birds and in all singing contexts, there were changes in respiratory rhythm and phase relationships that consistently occurred before song that were distinct from quiet respiration. The pattern of respiration described here awaits future investigation in other species. In this study, we found that zebra finches showed more pronounced changes in song respiration compared with Bengalese finches and they produce more complex expiratory pulses. Thus, the degree of anticipatory song respiration may be related to variable respiration patterns that are used to generate the multitude of song patterns observed in oscine birds. Furthermore, this pattern may be universal in other animals producing learned vocalizations, but whether such preparatory respiratory patterns occur in the seconds before language production awaits investigation. A complete understanding of motor preparation for vocalization requires integrating top-down control of vocal circuits with bottom-up feedback taking into account the respiratory constraints that determine when song can be initiated and how long a sequence of vocalizations can be produced.

DATA AVAILABILITY

Data will be made available upon reasonable request.

SUPPLEMENTAL DATA

Supplemental Fig. S1: https://doi.org/10.6084/m9.figshare.21546936.v1.

Supplemental Fig. S2: https://doi.org/10.6084/m9.figshare.21546945.v1.

Supplemental Fig. S3: https://doi.org/10.6084/m9.figshare.21546951.v2.

Supplemental Table S1: https://doi.org/10.6084/m9.figshare.21546957.v1.

Supplemental Table S2: https://doi.org/10.6084/m9.figshare.21546960.v1.

GRANTS

This research was funded by TCU Invests in Scholarship under Grant No. 66038 and National Institute of Neurological Disorders and Stroke (NINDS) R01NS108424 (to B. G. Cooper).

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the authors.

AUTHOR CONTRIBUTIONS

B.G.C. conceived and designed research; J.D. and B.G.C. performed experiments; J.M.M., J.D., and B.G.C. analyzed data; J.M.M., J.D., and B.G.C. interpreted results of experiments; J.M.M. and B.G.C. prepared figures; J.M.M. and B.G.C. drafted manuscript; J.M.M. and B.G.C. edited and revised manuscript; J.M.M., J.D., and B.G.C. approved final version of manuscript.