Analysis of breathing data from plethysmography is usually time-consuming, laborious, and often involves measurements of rate of breathing (f_R), tidal volume (V_T), and minute ventilation (V_E). f_R is usually calculated in intervals when the animal is asleep or stationary from the time of peak-to-peak inspiratory or expiratory signal. V_T is often measured by integration of signal over a specified period of time. Therefore, analysis of minute-to-minute changes in breathing may be difficult and take more time. In addition, the fact that conventional analysis is usually subjective might affect the reproducibility of reported results. To overcome this problem, we wrote a custom, open-source Python script using Neurokit2, NumPy, and Pandas software packages (McKinney, 2010; van der Walt et al., 2011; Makowski et al., 2020) to analyze breathing signals in awake, freely-moving animals (Figure 1). While using this script, the user has the option to define the start and end intervals for baseline as well as experimental challenges, which the script uses to import the data and perform analysis. To validate our script, we benchmarked the data analyzed against a conventional method (Sheikhbahaei et al., 2018; Sheikhbahaei et al., 2017) by analyzing simple respiratory data (f_R,V_T, and V_E) in conscious marmosets and rats (Figure 2). We did not identify any differences in values of f_R, V_T, and V_E between those generated using our script or by the conventional method (n = 3 per species) (Figure 2—figure supplement 1 and Figure 2—figure supplement 2). We then used our toolkit to further analyze other breathing behaviors in both male and female marmosets at room air and during acute exposure to hypoxia and hypercapnia (see below).

The f_R at room air (normoxia/normocapnia) was similar in female (79 ± 7 breaths min^–1, n = 8) and male (78 ± 8 breaths min^–1, n = 8) adult marmosets (p = 0.88, Mann–Whitney test) (Figure 2—figure supplement 3). The V_T, calculated from trough to peak amplitude and normalized to body mass, was similar in female (0.43 ± .10 a.u.) and male (0.53 ± 0.08 a.u.) adult marmosets as well (p = 0.37, Mann–Whitney test). Additionally, baseline V_E was similar in female (35 ± 10 a.u.) and male (42 ± 8 a.u.) marmosets (p = 0.38, Mann–Whitney test) (Figure 2—figure supplement 3). Two marmosets (one male and one female) showed prolonged breath holding (11 ± 2 breaths hr^–1 for 4.3 ± 0.1 s).

We also measured changes in f_R, V_T, and V_E before, during, and after acute hypercapnic challenge (6% CO₂ in the inspired air). The magnitude of change in f_R, V_T, and V_E was similar between females and males during hypercapnia (n = 4 per sex, Figure 3—figure supplement 1), so we grouped them for further analyses. Increasing CO₂ inside the chamber increased f_R (87 ± 8 vs. 74 ± 8 breaths min^–1 in baseline, p = 0.039, Wilcoxon matched-pairs signed rank test), V_T (1.04 ± .11 vs. 0.4 ± 0.05 a.u. in baseline, p = 0.008, Wilcoxon matched-pairs signed rank test) and V_E (81 ± 11 vs. 32 ± 5 a.u. in baseline, p = 0.008, Wilcoxon matched-pairs signed rank test) (Figure 3).

Hypercapnic-induced increase in f_R was mainly due to decrease in time of inspiration (T_I) (0.26 ± 0.02 vs. 0.36 ± 0.03 s at baseline, p = 0.008, Wilcoxon matched-pairs signed rank test) rather than time of expiration (T_E) (0.48 ± 0.06 vs. 0.58 ± 0.09 at baseline sec, p = 0.078, Wilcoxon matched-pairs signed rank test). As expected, respiratory flow (R_F) was also increased during hypercapnia (4.1 ± 0.4 vs. 1.2 ± 0.2 a.u. in baseline, p = 0.008, Wilcoxon matched-pairs signed rank test) (Figure 3, Figure 3—figure supplement 2).

Subsequently, we measured regularity of respiration via cycle-to-cycle dispersion of T_TOT in baseline and hypercapnic condition as shown in Poincaré plots (Figure 3). We quantified the regularity of breathing (Sheikhbahaei et al., 2017) by SD1 and SD2 (see Materials and methods and Soni and Muniyandi, 2019). The baselines SD1 and SD2 were greater than those during hypercapnia (132 ± 17 vs. 550 ± 115 a.u. in baseline, p = 0.008, and 198 ± 34 vs. 758 ± 138 in baseline, p = 0.008, respectively; Wilcoxon matched-pairs signed rank test) (Figure 3 and Figure 3—figure supplement 3).

We then measured changes of f_R, V_T, and V_E during acute systemic hypoxic challenges (10% O₂ in the inspired air) with respect to the baseline. Similar to hypoxia, the magnitude of the change in f_R, V_T, and V_E was not different in females and males during acute hypoxia (n = 4 per sex) (Figure 4—figure supplement 1), therefore we combined all the data from both sexes. In the first minute of the hypoxic challenge, V_T and V_E increased by 17% ± 12% and 17% ± 14%, respectively (Figure 4). This initial increase in ventilation may be due to hypoxic-induced carotid body activation. We then analyzed breathing behaviors 5 min after changing the inspired O₂ from 21% (room air) to 10%. Hypoxic conditions did not elicit overall changes in f_R (74 ± 5 vs. 82 ± 7 breaths min^–1 in baseline, p = 0.3, Wilcoxon matched-pairs signed rank test), but decreased V_T (0.39 ± 0.08 vs. 0.54 ± 0.11 a.u. in baseline, p = 0.078, Wilcoxon matched-pairs signed rank test) and V_E (29 ± 6 vs. 46 ± 12 a.u. in baseline, p = 0.043, paired t test) (Figure 4).

Then, we also calculated changes in T_I, T_E, and R_F during hypoxic challenge with respect to baseline. Since T_I, T_E, and R_F were not different in females and males during hypoxia (n = 4 per sex) (Figure 4—figure supplement 2), we combined their data. While hypoxia did not change Τ_Ι (0.34 ± 0.02 vs. 0.30 ± 0.02 s in baseline, p = 0.46, Wilcoxon matched-pairs signed rank test) and T_E (0.55 ± 0.1 vs. 0.50 ± 0.1 s in baseline, p = 0.4, Wilcoxon matched-pairs signed rank test), R_F was decreased during hypoxia (12 ± 2 vs. 16 ± 3 a.u. in baseline, p = 0.008, Wilcoxon matched-pairs signed rank test) after 5 min of challenge (Figure 4C and D).

We also measured the effects of acute hypoxia on the regularity of breathing (Figure 4E–G). We combined the data from male and female marmosets as there were no sex differences measured for irregularity of breathing (Figure 4—figure supplement 3). We quantified the regularity of breathing by generating Poincaré plots and measuring SD1 and SD2. The baselines for SD1 and SD2 were similar during hypoxia (330 ± 39 vs. 374 ± 42 in baseline, p = 0.4, and 419 ± 52 vs. 488 ± 68 in baseline, p = 0.6, respectively; Wilcoxon matched-pairs signed rank test) (Figure 4G).

We then measured changes in respiratory features in the 5 min immediately following the hypoxic challenge (post-hypoxic challenge). Though we saw no changes in f_R (73 ± 4 vs. 82 ± 7 breaths min^–1 in baseline, p = 0.11, Wilcoxon matched-pairs signed rank test) and V_T (0.47 ± .10 vs. 0.54 ± 0.12 a.u. in baseline p = 0.15, Wilcoxon matched-pairs signed rank test), V_E decreased (34 ± 7 vs. 46 ± 12 a.u. in baseline, p = 0.078, Wilcoxon matched-pairs signed rank test) relative to baseline (Figure 4A and B). We also calculated changes in T_I, T_E, and R_F, immediately following the hypoxia challenge. While we observed no change in T_E (5.4 ± 0.5 vs. 5.0 ± 0.6 a.u. in baseline, p = 0.46), there was an increase in T_I (3.4 ± 0.2 vs. 3.0 ± 0.2 a.u. in baseline, p = 0.055) and a decrease in R_F (1.3 ± 0.2 vs. 1.8 ± 0.4 a.u. in baseline, p = 0.078) after hypoxia challenge (Figure 4C and D).

The constant f_R and decrease in R_F during hypoxia and post-hypoxic challenge suggests that the metabolic rate might decrease during acute hypoxic challenge. We then calculated the metabolic rate (M_R) in marmosets during hypoxic challenge. Our data suggest that M_R had a profound decrease (~50%) during hypoxia when compared to the baseline (Table 1). Therefore, we calculated ventilatory efficiency as V_E/M_R to understand the changes in ventilation in response to CO₂ production. Our analysis suggested that the ventilatory efficiency during hypoxic challenge was not different from the baseline (22 ± 6 vs. 19 ± 6 a.u. in baseline, p = 0.6, Wilcoxon matched-pairs signed rank test), however it was lower during post-hypoxic challenge (13 ± 3 vs. 22 ± 6 a.u. in hypoxia, p = 0.055, Wilcoxon matched-pairs signed rank test) (Figure 4H).

Since incidences of sighs, apneas, and sniffing could contribute to the irregularity of respiration, we measured the frequencies of these essential features of breathing behavior. Sighs can be generated within the inspiratory rhythm-generating circuits of the preBötzinger complex (preBötC) (Sheikhbahaei et al., 2018; Li et al., 2016; Lieske et al., 2000; Borrus et al., 2020; Toporikova et al., 2015; Vlemincx et al., 2013), and may be modulated by excitatory signals from central chemocenters (Sheikhbahaei et al., 2018; Sheikhbahaei et al., 2017; Souza et al., 2018; Souza et al., 2019; Li et al., 2016). In female adult marmosets, sigh frequencies were not different when compared to those in male animals during the baseline in room air (11 ± 1 vs. 12 ± 2 hr^–1 in male) (Figure 5—figure supplement 1). In rodents, both hypoxic and hypercapnic challenges increased frequency of sighs (Li et al., 2016; Sheikhbahaei et al., 2018). Consistent with those results, hypoxia increased sigh events by 5.5 folds in marmosets (71 ± 10 vs. 11 ± 1 hr^–1 in room air, p = 0.008, Wilcoxon matched-pairs signed rank test). Similarly, hypercapnia also increased sigh frequency (68 ± 3 vs. 12 ± 1 hr^–1 in room air; p = 0.008, Wilcoxon matched-pairs signed rank test) (Figure 5A).

We also analyzed high-frequency breathing (sniffing) in marmosets. During the hypoxic challenge, the sniffing rate did not change with respect to baseline (74 ± 13 vs. 101 ± 38 hr^–1 in baseline, p = 0.84, Wilcoxon matched-pairs signed rank test) (Figure 5B). However, during hypercapnic challenge, rate of sniffing was less than that in room air (27 ± 16 vs. 100 ± 38 hr^–1 in baseline, p = 0.078, Wilcoxon matched-pairs signed rank test) (Figure 5B).

Spontaneous and post-sigh apneas have been reported in rodents, rabbits, humans, and other animals (Yamauchi et al., 2008; Franco et al., 2003; van der Heijden and Zoghbi, 2018; Bongianni et al., 2010; Li et al., 2006; Ramirez et al., 2013; Sheikhbahaei et al., 2017). We did not find differences in the apnea index between female and male marmosets (Figure 5—figure supplement 1). Apneas decreased during hypoxic challenge (37 ± 12 vs. 79 ± 20 in room air, p = 0.039, Wilcoxon matched-pairs signed rank test) (Figure 5C). During hypercapnic challenge, rate of spontaneous apneas also decreased drastically relative to that in room air (9 ± 6 vs. 129 ± 31 hr^–1 in room air, p = 0.008, Wilcoxon matched-pairs signed rank test) (Figure 5C).

Lastly, to understand if hypoxia or hypercapnia have any effect on the animal’s activity, we measured the movement of marmosets in the plethysmograph during both challenges. In measurements of large changes in position from one quadrant of the chamber to another, we saw no changes during hypoxia (4.1 ± 1.8 vs. 5.4 ± 1.6 quadrant changes per minute at baseline, n = 3, p = 0.99, Wilcoxon matched-pairs signed rank test) or hypercapnia (4.0 ± 0.5 vs. 4.7 ± 1.4 quadrant changes per minute at baseline, n = 3, p = 0.75, Wilcoxon matched-pairs signed rank test). Similarly, we observed no differences in the sum of frame-to-frame Euclidean distances in hypoxic challenge (124 ± 28 vs. 128 ± 27 pixels min^–1 at baseline, n = 3, p = 0.99, Wilcoxon matched-pairs signed rank test) or hypercapnia challenge (126 ± 20 vs. 120 ± 11 min^–1 at baseline, p = 0.99, Wilcoxon matched-pairs signed rank test) (Figure 6).

Currently, the chemosensitivity mechanisms that adjust breathing with respect to the level of PCO₂/pH in the brain are centered around neurons and astrocytes in the retrotrapezoid nucleus (RTN) and medullary raphé (Kumar et al., 2015; Teran et al., 2014; Guyenet et al., 2019; Gourine et al., 2010). However, other data support a hypothesis that distributed chemosensitive regions in the medulla act as central respiratory chemosensors and are responsible for mounting of about 70% of the hypercapnic respiratory response (the mechanism that adjusts breathing in accordance with increase in PCO₂) (Nattie, 1999; Nattie, 2000; Nattie, 2001; Spyer and Thomas, 2000; Nattie and Li, 2009). Specialized peripheral chemoreceptors located in the carotid bodies (and aortic bodies in some species) are responsible for the remaining 30% of hypercapnia-induced augmentation of breathing. In awake, freely behaving marmosets, hypercapnia increased both breathing rate (f_R) and tidal volume (V_T) (Figure 3). However, the augmentation of ventilation (V_E) was mainly due to increase in V_T (by ~160%) rather than f_R. These data are comparable to data obtained from rodents (Bhandare et al., 2020; Sheikhbahaei et al., 2018) and human (Duffin et al., 2000; Ogoh et al., 2009; Serebrovskaya, 1992; Maxwell et al., 1986). In this study we used hyperoxic hypercapnia. In humans and rodents, hyperoxia is proposed to suppress the activity of carotid bodies (Chavez-Valdez et al., 2012; Gonzalez et al., 1994; Bates et al., 2014). By extrapolation, since marmosets lack aortic bodies (Clarke and de B Daly, 2002), we assumed that hyperoxia attenuates marmoset’s carotid body activity, and therefore, the hypercapnic ventilatory response presented here may represent the central CO₂ chemosensitive activity. Hypercapnia also increases frequency of sighs (i.e., augmented breath) in rodents (Forsberg et al., 2016; Ramirez, 2014). Consistent with these data, we also found that hypercapnia increased sigh frequency in marmosets (Figure 5A). Nevertheless, our data suggest that the common marmoset is a good animal model for studying respiratory responses to hypercapnia. However, more experiments are required to show that increases in CO₂ actually activate classical chemosensitive regions in marmosets.

The hypoxic ventilatory response (HVR) in common marmosets was noteworthy, as there was little or no increase in f_R during hypoxic exposure (Figure 4A). We believe the level of O₂ during hypoxia was sufficient to elicit HVR, as a similar level of O₂ (10% O₂) decreased the peripheral oxygen saturation (SpO₂) to 89% in humans after 180 s (Gerlach et al., 2021). In addition, increases in sigh rate and the existence of post-hypoxic depression (see below) strongly suggest that the respiratory circuits were activated by the hypoxic challenge to prevent hypoxic ventilatory decline (HVD).

Although hypoxic conditions in marmosets’ natural habitat (sea-level forests of the Amazon) are rare, hypoxia might occur as a result of disease or during sleep. Acute HVR is likely biphasic in mammals (Easton et al., 1986; Rehan et al., 1996; Martin et al., 1998; Vizek et al., 1987; Waites et al., 1996; Gozal and Gaultier, 2001; Greer and Funk, 2013). During acute hypoxia, ventilation shows an initial increase followed by a subsequent decline to a value at or above the baseline (i.e., HVD). This biphasic hypoxic response has been reported in humans, rats, and other mammals (Eden and Hanson, 1987; Martin et al., 1990; Fung et al., 1996; Dahan et al., 1996; Vizek and Bonora, 1998). However, earlier reports suggest that there is considerable interindividual variation in HVR in humans (Hirshman et al., 1975; Weil and Zwillich, 1976). Recent data in awake adult humans showed no increase of f_R during acute hypoxia (Gerlach et al., 2021), suggesting that any changes in ventilation may be due to changes in V_T, not f_R (Tarbichi et al., 2003). Our data support these reports, as we see variable responses to acute hypoxia in marmosets as well as a slight increase in V_T and V_E during the first minute of HVR followed by a decrease in V_T and V_E as hypoxia continues (Figure 4). However, ventilatory efficiency (V_E /M_R) was not affected by hypoxia (see below). It is possible that the large gas-exchange capacity of marmosets’ lungs (due to the increased oxygen diffusion capacity) (Barbier and Bachofen, 2000) maintains the adequate blood oxygenation, and therefore, blunts the HVR during hypoxia.

Hypoxia increases sigh frequency in mammals, even in animals whose carotid bodies are non-functional (Bartlett, 1971; Schwenke and Cragg, 2000; Cherniack et al., 1981; Sheikhbahaei et al., 2018). Consistent with these data, hypoxia also increased sigh frequency in marmosets. In addition, the fact that sigh frequency, but not breathing frequency, increased during hypoxic challenge, supports the hypothesis that distinct cells may be responsible for the generation of rhythmic sighs and normal breathing (Toporikova et al., 2015; Li et al., 2016; Sheikhbahaei et al., 2018). Recent data from behaving rats suggest that purinergic signaling from astrocytes (numerous star-shaped glial cells) in the respiratory rhythm-generating circuits of the preBötC may play a significant role in regulation of sigh generation (Sheikhbahaei et al., 2018).

On the other hand, the mechanism of HVD is not fully understood. It is proposed that desensitization of peripheral chemoreceptors might play a role (Bascom et al., 1990), though significant evidence suggests that, at least in rodents, astrocytes in the preBötC are capable of acting as central respiratory oxygen chemosensors (Sheikhbahaei et al., 2018; Angelova et al., 2015; Rajani et al., 2018). Moreover, preBötC astrocytes might contribute to the HVD via release of adenosine triphosphate (ATP) (Sheikhbahaei et al., 2018). Existence of ATP receptors in the brainstem respiratory regions in marmosets (Yao et al., 2000) further strengthens this hypothesis in primates. In addition to the preBötC, RTN, rostral ventrolateral medulla, and the nucleus of the solitary tract in the brainstem are proposed to have oxygen sensing capabilities (Accorsi-Mendonça et al., 2015; Mazza et al., 2000; Uchiyama et al., 2020). However, more research is required to understand if this ‘distributed central oxygen chemosensors’ hypothesis (SheikhBahaei, 2020) can be generalized to primates.

Decrease of post-hypoxic ventilation in human is also reported (Tarbichi et al., 2003). This post-hypoxic depression is also illustrated in conscious (Angelova et al., 2015; Sheikhbahaei et al., 2018) and anesthetized (Rajani et al., 2018) rats. Similarly, we observe such a respiratory response in marmosets. However, we did not detect any sex differences during post-hypoxic recovery from hypoxia as reported in rat’s in vitro models (Garcia et al., 2013). Our data are consistent with that reported in humans, namely that there are no differences in ventilation between sexes during post-hypoxic response (Tarbichi et al., 2003).

We also examined marmoset activity during hypoxic challenges. Our data, however, suggest that animal activity was not affected by hypoxia (or hypercapnia). This suggests that decreases in metabolic rate during hypoxic challenge are not due to decrease in spontaneous activity, but may be accounted for by changes in other processes with metabolic demand such as thermoregulation or cardiovascular activities.

Other than an increase in ventilatory response during hypoxia, mammals can reduce oxygen demand by optimizing and decreasing their rates of metabolism (Dzal et al., 2015). During acute hypoxia, adult marmosets decreased their metabolic rates (M_R) by ~50%, which is similar to data reported in other primates (pygmy marmosets [Tattersall et al., 2002] and humans [Robinson and Haymes, 1990]), but two to three times more than the calculated rates from cats (Gautier et al., 1989) and rats (Mortola et al., 1994). This decrease in metabolism together with increase in sigh frequency might be sufficient for homeostatic control of blood oxygen during acute hypoxia in primates. We also analyzed ventilatory efficiency (V_E/M_R) to understand the changes in ventilation in response to CO₂ production. Although we saw a slight increase in ventilatory efficiency, acute hypoxia did not have a significant effect on V_E/M_R (Figure 4H). We believe ventilatory efficiency gives a more comprehensive view on ventilation compared to just measuring V_E. It also suggests that acute hypoxia does not increase ventilation in the common marmoset.

We acknowledge that our characterization of breathing behaviors in the common marmoset is not complete and more experiments are needed to fully characterize hypoxic breathing behaviors in common marmosets. For instance, the respiratory response to hypoxia in rodents is non-linear, as a decrease in inspired O₂ to 15% elicits a minimal ventilatory response, but a decrease to 10% elicits a strongly robust one (SheikhBahaei, 2017; Hosford et al., 2020; Sheikhbahaei et al., 2018). In addition, a decrease in metabolic rate strongly suggests that the core body temperature is affected by hypoxia (Morgan et al., 2014). Since we did not measure core body temperature in the marmoset, we reported the tidal volume as arbitrary units. Future experiments (using telemetry probes or other devices) to measure core body temperature in marmosets are needed to accurately measure changes in tidal volume during stepwise changes of inspired O₂ or CO₂. However, our data suggest that the analysis toolbox presented in this study is a powerful means to analyze breathing data in awake animal models under different experimental O₂ and CO₂ conditions.

We used 16 common marmosets (C. jacchus) (8 males, 8 females; 394 ± 5 g; 40 ± 1 months) and three male Sprague-Dawley rats (320 ± 11 g) for measuring and defining breathing behaviors. All experiments were performed in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals. The experiments on marmosets and rats were approved by the Animal Care and Use Committee (ACUC) of the Intramural Research Program (IRP) of the National Institute of Mental Health and ACUC of the IRP of National Institute of Neurological Disorders and Stroke, respectively. Animals were housed in temperature-controlled facilities on a normal light-dark cycle (12 hr:12 hr, lights on at 7:00 AM). They lived in paired or family-grouped housing and were given food and water ad libitum.

Marmoset respiratory activity was measured using whole-body plethysmography in a room with ambient temperature of 27–28°C. Awake animals were placed in the Plexiglas chamber (~3 L) which was flushed with 21% O₂, 79% N₂, at a rate of 2.2 L min⁻¹ during measurements of baseline respiratory behavior (Figure 1). Concentrations of O₂ and CO₂ in the chamber were monitored using a fast-response O₂/CO₂ analyzer (ML206, AD Instruments). All experiments were performed at the same time of day (between 10:00 and 14:00 hr) to account for possible circadian changes in base level physiology (Iizuka et al., 2010). For measuring the respiratory behaviors during hypoxia, following a 40 min baseline period, the chamber was flushed with 10% O₂, 90% N₂, at a rate of 2.2 L min^–1. After 10 min of exposure to hypoxic conditions, the gas concentration in the chamber was changed to room air for another 10 min (Figure 1—figure supplement 1). Marmoset respiratory activity was also measured during exposure to hypercapnic conditions. Following a 40 min baseline period, the chamber was flushed with 6% CO₂, 60% O₂, 34% N₂, at a rate of 2.2 L min^–1. After 10 min of exposure to hypercapnic conditions, the chamber was then flushed with room air for another 10 min. Hyperoxic condition (60% O₂) was used to prevent any hypoxia associated with hypercapnia as used routinely in rodents (Teppema et al., 1997; Sheikhbahaei et al., 2018). Respiratory data were acquired with Power1401 (CED; RRID: SCR_017282) interface and transferred to Spike2 software (CED; RRID: SCR_000903). To prevent any acclimatization confound, each animal was placed only once in the plethysmography chamber and randomly assigned to either hypoxia or hypercapnia experiment.

Similarly, we used whole-body plethysmography to record respiratory activity in unrestrained conscious adult rats as described before (Sheikhbahaei et al., 2017). Briefly, adult rats were placed in a Plexiglas recording chamber (~1 L) that was flushed continuously with room air (21% O₂, 79% N₂; temperature 22–24°C), at a rate of 1.2 L min⁻¹. The animals were allowed to acclimatize to the chamber environment for ~60 min. Resting breathing activity was then recorded for 10 min. Respiratory activity in all the animals was assessed at the same time of the day (between 10:00 AM and 2:00 PM) to take into the account circadian variations of the physiological parameters. Data were acquired using Power1401 interface and analyzed offline using either Spike2 software (CED) or our in-house script presented in this paper.

For measuring metabolic rate (M_R) in marmosets, we calculated CO₂ production using the following equation and expressed as percent: M_R = ΔCO₂ × F_R/body mass, where ΔCO₂ is the peak changes in the [CO₂] in the chamber as measured by the gas analyzer. F_R is the flow rate through the plethysmography chamber (i.e., 2.2 L min^–1), and body mass is marmoset body mass (g).

We tracked 10 points on the marmoset head and body (n = 3 animal per challenge) from an overhead view of the plethysmograph using WhiteMatter e3Vision cameras (e3Vision camera; e3Vision hub; White Matter LLC). We used DeepLabCut version 2.10.2 for pose estimation of these features (Nath et al., 2019; Mathis et al., 2018). We labeled 656 total frames from 16, 20–30 min videos recorded at 60 fps (95% was used for model training). We used ResNet-50-based neural network with default parameters for four iterations with five shuffles, and the test error was: 29.8 pixels, train: 2.4 pixels, with 0.6 p-cutoff, test error was: 14.0 pixels, train: 2.4 pixels (image size 600 by 800 pixels).

Below-threshold feature coordinates were then filled using methods from the B-SOiD Python toolkit (Hsu and Yttri, 2021). We used the average position of five points on the head for further analysis after qualitative assessment of consistent labeling accuracy. By dividing the labeled images in quadrants along the X- and Y-axes (X = 400 pixels, Y = 300 pixels), we counted the number of times large changes in position (i.e., movement) occurred. Quadrant positions were down-sampled to 2 s to avoid counting quadrant changes from when the animal paused near the dividing lines. Additionally, successive Euclidean distances were calculated for each point across each frame of the videos to produce total movement. Total linear distance was then divided by length of condition in minutes to obtain rate of activity in each condition.

All animals in the study were included in the analysis. Plethysmography data were imported to Python using Neo Python package (Van Rossum and Drake, 2011; Garcia et al., 2014). We wrote a custom Python script using methods from Neurokit2, NumPy, and Pandas software packages (McKinney, 2010; van der Walt et al., 2011; Makowski et al., 2020). Areas of the signal with frequencies above 300 cycles per minute (~3.3 Hz) were excluded from analysis, as they were likely artifact resulting from movement inside the chamber. To ensure that we captured the full change in ventilation, we used steady-state responses to hypoxia and hypercapnia and analyzed the data 5 min after the start of each challenge. Neurokit2 methods were used for signal cleaning and extraction of instantaneous frequency, T_TOT (total time of breath), T_I (time of inspiration), T_E (time of expiration), and amplitude (i.e., tidal volume [V_T]) from trough to peak of the signals (see Figure 2). The calculated V_T was normalized to the body mass (g) of each animal. Mean inspiratory flow rate (R_F) was defined as the ratio of V_T to T_I (V_T/T_I). During hypoxia and hypercapnia challenges, the respiratory signals were analyzed in 1 min epochs to consider local changes in respiration parameters.

High-frequency breathing (i.e., sniffing) was defined as any breathing frequency between 250 cycles (2.5 Hz) and 300 cycles per minute. Apneas were defined by breathing cycles with T_TOT greater than three times the average for each animal. Augmented breaths (i.e., sighs) were readily identifiable by using the criteria described in rats (Sheikhbahaei et al., 2018; Sheikhbahaei et al., 2017) and measured during the baseline and experimental conditions.

Two measures of rate variability were also calculated as described elsewhere (Soni and Muniyandi, 2019). SD1 is a measure of dispersion of T_TOT perpendicular to the line of identity in the Poincaré plots, therefore demonstrating short-term variability. SD2 is a measure of dispersion of T_TOT along the line of identity in the Poincaré plots, demonstrating long-term variability in respiratory rate.

where SD is the standard deviation of successive differences in T_TOT and SDT_TOT is the standard deviation in T_TOT.

All data were tested with Shapiro-Wilk test for normality and statistically compared by t test, Wilcoxon matched-pairs signed rank test, or Mann–Whitney U rank test as appropriate in Prism 9 (GraphPad, Inc; RRID: SCR_002798). Data are reported as mean ± SEM.