Experiments were conducted using CD1 mice and protocols were approved by the Animal Care and Use Committee at Seattle Children's Research Institute and at The University of Chicago in accordance with the National Institutes of Health guidelines. All animal subjects were housed at 21°C and in a 12/12 h light cycle where the light phase was from 07:00 to 19:00. Animals had access to food and water ad libitum.

Neonatal mice (beginning from postnatal day 0–2) and their dam were exposed to CIH. CIH occurred during the light cycle and lasting for 8 h/day (i.e., 80 intermittent hypoxia cycles/day), for 10 consecutive days. Litters were culled down to a total of eight pups prior to CIH exposure.

As described previously (Garcia et al., 2016) a single hypoxic bout was achieved by flowing 100% N₂ into the chamber for ~60 s created a hypoxic state where the nadir O₂ value reached 4–5% O₂ (for 5–7 s). The hypoxic bout was followed by a 300 s air break achieved by flushing with air (~21% O₂). The return to air restored a normoxic state (18–20%) within 60 s following hypoxia. Environmental CO₂ did not rise >0.02% during any phase. Both pups and their dam were exposed to the CIH paradigm.

Transverse brainstem slices containing the preBötC were prepared from either CIH-exposed mice (12–48 h after the end of the CIH paradigm) or naïve (i.e., control) CD1 mice (postnatal day 10–12). Mice were rapidly decapitated or anesthetized with isofluorane and then decapitated at which point the brainstem was removed and prepared for slicing. The isolated brainstem was glued to an agar block (dorsal face to agar) with the rostral face up toward the vibratome blade and submerged in artificial cerebrospinal fluid (aCSF, ~4°C) equilibrated with Carbogen (95% O₂–5% CO₂). One hundred to two hundred micrometer serial transverse slices at a 20° angle were made in a rostral to caudal direction until disappearance of the parafacial group and appearance of the inferior olive, nucleus ambiguus, the hypoglossal nucleus, and the opening of the fourth ventricle as also described previously by others (Ballanyi and Ruangkittisakul, 2009). A 550 μm thick rhythmic slice containing the preBötC was made. Our approach encompasses the preBötC as we have previously documented (Viemari et al., 2013). A preBötC network in the brainstem slices was considered intact if transmission of the preBötC rhythm to the hypoglossal pool was consistent (i.e., >40% of preBötC bursts transmitted to the motor pool) and/or a synchronous preBötC rhythm was detected bilaterally in each preBötC prior to the start of our recordings.

The composition of artificial cerebral spinal fluid (aCSF) was (in mM): 118 NaCl, 3.0 KCl, 25 NaHCO₃, 1 NaH₂PO₄, 1.0 MgCl₂, 1.5 CaCl₂, 30 D-glucose. The aCSF had an osmolarity of 308 ± 2 mOSM and a pH of 7.40 to 7.45 when equilibrated with gas mixtures containing 5% CO₂ at ambient pressure. Rhythmic activity from the preBötC was induced by raising extracellular KCl to 8.0 mM. Baseline conditions were made by equilibrating aCSF with carbogen (95% O₂, 5% CO₂) while hypoxic conditions were made by aerating with 95% N₂, 5% CO₂. Exposure to hypoxia lasted for 600 s. Despite the equilibration of aCSF with 0% O₂, hypoxic media contained some O₂ (Garcia et al., 2010; Hill et al., 2011).

Extracellular population activity was recorded with glass suction pipettes (tip resistance <1 MΩ) filled with aCSF, and were positioned over the ventral respiratory column containing the preBötC. Signals were amplified 10,000X, filtered (low pass, 1.5 kHz; high pass, 250 Hz), rectified, and integrated using an electronic filter. Extracellular recordings were acquired in pCLAMP software (Molecular Devices, Sunnyvale, CA) and were analyzed post-hoc in Clampfit software (version 10.2). In some cases, population recordings from two preBötC brainstem slices were recorded simultaneously within a single recording chamber. In situations where two slices were used, the slices were positioned in a staggered arrangement such that media flow was not obstructed for either preparation.

Whole animal plethysmography was performed under unrestrained conditions in mice (P10 to P13) using a commercial environmental chamber (Buxco Research System). Briefly, normoxic/normocapnic air (19 to 21% O₂, 0% CO₂) was replaced by hypoxic breathing gas (10% O₂, 0% CO₂) for 5 min. We measured respiratory frequency (breath per min), tidal volume (μL per breath), and minute ventilation (μL^.g^−1.min⁻¹). Measurements were performed during (a) control period under air, prior to hypoxia, (b) the second minute of hypoxia, (c) the fifth minute of hypoxia, (d) the post-hypoxic period (2 and 4 min following the end of hypoxia when normoxia was restored in the chamber), and (e) 10 min later. Mice were kept thermoneutral at 33°C throughout the plethysmography session.

To resolve how CIH affects the dynamical system properties of rhythm generation we examined (1) burst-to-burst variability using the irregularity score (IrS; Zanella et al., 2014); (2) the magnitude of fluctuation in the network rhythm was defined by its fluctuation value (χ); and (3) the predictability and thus, the complexity of the network rhythm throughout the time series using sample entropy (sENT; Richman and Moorman, 2000). For further detail see Figure 1.

To calculate IrS of period (IrS_P) for a single cycle we used the following formula (Figure 1B; see also Telgkamp et al., 2002): IrS_{P − N} = 100 × ABS (P_N - P_{N − 1})/P_{N − 1}, where IrS_{P− N} is the IrS_P of the Nth cycle, P_N is its period, P_{N − 1} is the period of the preceding cycle, and ABS is the absolute value. We also apply this formula to assess the irregularity of burst amplitude IrS_AMP by replacing in the previous formula the period by the amplitude of the integrated and rectified inspiratory burst (with unchanged time constants throughout the study). The mean irregularity score for all cycles was calculated for each rhythmic activity generated by a given slice.

The network population rhythm is the result of phase-locking activity between weakly coupled rhythmically active neurons, which includes bursting pacemakers and non-pacemaker neurons (Pena et al., 2004; Carroll et al., 2013). Changes in intrinsic neuronal and/or synaptic properties by CIH have the potential to alter the inherent fluctuations of the network rhythm (i.e., periodic fluctuations in the amplitude and/or period (illustrated with red or green lines that follow the peak amplitudes in Figure 1A).

If CIH caused changes to the periodic fluctuations of amplitude and period that span across more than two cycles in the time-series (Figure 1A), the burst-to-burst window used to calculate the irregularity score (Figure 1B) would provide limited insights into the phenomenon. To address this limitation of irregularity score, we calculated the fluctuation value (χ) and sample entropy (sENT) of metrics in the time series of each rhythm. While sENT is a modification of approximate entropy used for characterizing the complexity of physiological time-series signals (Richman and Moorman, 2000) that describes the frequency of repetitive fluctuations in the time series, we introduce χ to characterize the gain of these repetitive fluctuations in the time series (see Figures 1C,D).

To calculate the fluctuations, a Butterworth filter was applied to event values plotted sequentially and local minima and maxima were detected. The fluctuation value (χ) for a given metric was calculated as the difference between local maxima and minima, χ_i, where i is single value of fluctuation and n- total number of fluctuations in an 300–400 s window. The average fluctuation was calculated as 1n∑i=1nxi. Thus, χ_i was used to identify whether the magnitude of fluctuations in a given metric of rhythm generation changed over time. To calculate χ of amplitude, amplitude events were normalized to its average. sENT was used to identify whether the overall predictability in the time series as given metric of rhythm generation changed over time. sENT was calculated as -logNxNy where N_x and Ny are the number of pairs of dimension 3 and 2 with distance less than two standard deviations of a given time series data set. Thus, a lower value of sENT corresponded to a time series containing repetitive patterns (i.e., less structural predictability) and thus, complexity in the time-series. Both χ and sENT were calculated from a 300 to 400 s window. In some cases, too few events were available to calculate either χ or sENT for a given parameter and were not included in the respective analyses (see Table 1 for n-values).

Analysis for hypoxic rhythm generation from the preBötC was divided into early hypoxia and steady-state hypoxia. Early hypoxia was defined as the initial 200 s interval beginning when O₂ in the circulating aCSF began to drop due to a switch to the hypoxic gas mixture. Kaplan-Meier estimator functions were used to describe the probability of stable rhythmogenesis during the early hypoxia phase. The endpoint for this analysis was defined as the first occurrence of an interburst interval (i.e., cycle period) during hypoxic augmentation, which was at least two times greater than the mean period of rhythmogenesis prior to hypoxia. Rhythms not demonstrating the defined endpoint during hypoxic augmentation were censored. Differences between the Kaplan-Meier estimator functions were determined using the Gehan-Breslow-Wilcoxon test. Steady-state hypoxia was defined as the final 400 s interval of hypoxia (Hill et al., 2011; Garcia et al., 2013). During this interval the oxygen sensitive current measuring bath O₂ was clearly reduced from the baseline, stable and asymptotic. To determine whether changes in the fictive gasping rhythm was uniquely impacted by CIH, we calculated the normalized values for a given metric using the following:

where X_normalized represents the normalized value of metric “X” during hypoxia.

X_hypoxia is the value of the metric “X” and X_baseline is the value of metric “X” during baseline prior to hypoxia.

A 2-way ANOVA was used to assess differences in respiratory metrics during the peri-hypoxic interval in plethysmography experiments. Unless noted otherwise, comparisons between two groups were made using unpaired t-test. Differences between two means with similar variance was defined by a P < 0.05. An F-test for unequal variance was also performed and in cases where the P-value for the F-test was P < 0.05, we considered that differences existed between groups due to unequal variances. All statistical comparisons were made in Prism 5 (GraphPad Software).

To identify how CIH affected rhythmogenesis during baseline conditions (i.e., in Carbogen), recorded rhythmogenesis from the isolated preBötC taken from either control mice (n = 31) or mice exposed to CIH (n = 39). We calculated the irregularity score (IrS), fluctuation value (χ), and sample entropy (sENT) for both burst amplitude and period. CIH affects the irregularity score of the network rhythms (Figure 2). The IrS of amplitude (IrS_AMP) of control rhythms was 0.16 ± 0.01 (n = 31) while the IrS_AMP of following CIH was 0.21 ± 0.02 (n = 39). The IrS_AMP between the two groups was different from one another and exhibited unequal variances (Figure 2B; P = 0.03; F-ratio = 3.266, P = 0.001). Although, the IrS of period (IrS_P) of control and CIH rhythms were similar (Figure 2C; control: 0.24 ± 0.01, n = 39 vs. CIH: 0.29 ± 0.02 n = 31; P = 0.08), an unequal variance existed between groups (F-ratio = 2.834; P = 0.004) suggesting that the effect CIH on the IrS_P was largely variable and different to the control group.

We next determined whether CIH affected the complexity of rhythmogenesis by calculating the fluctuation value, χ, and the entropy value, sEnt, during baseline conditions for both groups. The χ of period (χ_P) and of the amplitude (χ_AMP) were different between the two groups (Table 1). Neither sENT of amplitude (sENT_AMP) nor sENT of period (sENT_P) were different between the respiratory activity of controls vs. mice exposed to CIH (Table 1). Thus, CIH affects the dynamics of rhythm generation beyond burst-to-burst irregularity by increasing the gain of the fluctuations in rhythmogenesis, yet does not change predictability and complexity of the time series for these metrics.

To determine whether CIH affected the central respiratory network activity during hypoxia, we examined how the respiratory rhythmic activity in preBötC slices from control (n = 27) and CIH-exposed mice (n = 22) respond to hypoxia and reoxygenation. While control and CIH groups displayed qualitatively similar stereotypical responses to hypoxia and reoxygenation (Figure 3A), we found significant quantitative differences during the transition to and in hypoxia (Figure 3B) but not during post-hypoxic recovery (data not shown). During the transition to hypoxia (i.e., the first 200 s of exposure), 33% (n = 9 of 27) of control rhythms vs. 64% (n = 14 of 22) of CIH-exposed rhythms showed no doubling of period. Using the doubling of period as an endpoint, for Kaplan-Meier estimator functions revealed that exposure to CIH significantly increased the probability for stable rhythmogenesis (Figure 3B). Later during steady-state hypoxia (i.e., the final 400 s of hypoxia), 18% (n = 5 of 27) of control networks failed to generate respiratory rhythmic activity which contrasted the continued rhythmogenesis produced by all CIH-exposed inspiratory networks (n = 22; Figure 4A). When comparing the change between the control and CIH groups, normalized instantaneous frequency (f_inst) during hypoxia was reduced by −17 ± 12% in the control group; whereas, in the CIH group in normalized f_inst during hypoxia increased by +18 ± 11% (Figure 4B). Normalized network burst amplitude during hypoxia was reduced by −59 ± 5% in the control group while it was only reduced by −46 ± 3% in the CIH group (Figure 4C) later during hypoxia. Interestingly, we observed an unequal variance between control and CIH exposed mice with regards to the normalized amplitude (F-ratio = 3.175, P = 0.01), IrS_AMP (F-ratio = 3.082, P = 0.013), and IrS_P (F-ratio = 3.175, P = 0.010) where the variance of the dataset during hypoxia was consistently larger in the control group vs. the CIH-exposed group. However, during hypoxia, the χ and sENT values from the CIH for both period and amplitude were not different from the respective control group (Table 1). These findings together indicate that CIH differentially affects rhythmogenesis in well-oxygenated states and during hypoxia, and suggests that the CIH-induced effects are state-dependent.

Because of the differences during hypoxia, and the documented riluzole sensitivity of the preBötC during hypoxia (Pena et al., 2004), we sought to assess how CIH impacted riluzole sensitivity of the preBötC following CIH. Riluzole (10 μM) caused several observable changes in rhythmogenesis that were different in control (n = 9) vs. in mice that were exposed to CIH (n = 16; Figure 5A). In riluzole, f_inst was greater (P = 0.003) in control (0.40 ± 0.04) vs. networks from mice exposed to CIH (0.21 ± 0.03). The differential effect of riluzole on f_inst was also reflected by the greater reduction in the change of f_inst from the CIH group (Figure 5B). Differences were also evident in the change of IrS_AMP between groups. Specifically, the change in IrS_AMP did not exceed 100% in the control group when exposed to riluzole. Yet, in the CIH group, riluzole increased IrS_AMP by >100% in 3 of 16 recordings of respiratory rhythmic activity. When comparing these groups, the mean effect of riluzole was a decrease in IrS_AMP by -24 ± 14% (Figure 5C, P = 0.04) and an increase in IrS_P +49 ± 17% (Figure 5D, P = 0.030) for the CIH exposed group. Although, riluzole did not produce differences in χ_AMP between control and CIH, the riluzole exposure increased χ_P of rhythmogenesis following the CIH (Table 1). These findings suggest that CIH enhances a riluzole sensitive mechanism to reduces both burst-to-burst variability and the gain of fluctuations in rhythmogenesis. Interestingly, riluzole reduced both sENT_P and sENT_AMP in the CIH exposed group (Table 1) and suggests that networks exposed to CIH possess a riluzole sensitive mechanism, which decreases the degree of overall predictability of rhythmogenesis. Thus, together these observations support the notion that CIH fundamentally changes rhythmogenesis from the preBötC.

To access how changes caused by CIH affect the ventilatory response in the intact, alert animal, we examined the response to hypoxia and reoxygenation in both control and CIH mice. Both groups exhibited qualitatively similar stereotypical responses to hypoxia (10% O₂ inspired) and reoxygenation (Figure 6A). Although, no differences during the perihypoxic interval were found between the control and CIH groups in the overall pattern of minute ventilation (Figure 6B) or tidal volume (Figure 6C), a difference between groups was observed in respiratory frequency (Figure 6D). Comparisons specifically during air and hypoxia reveal, however, that CIH led to a decrease in frequency yet an increase in tidal volume (Table 2).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.