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

Neonatal mice (beginning from postnatal day 0 to 2) and their dam were exposed to a CIH paradigm similar to that described by Peng et al. (2004). To reduce the potential effects of maternal stress, litters exposed to CIH were culled down to a total eight pups prior to CIH exposure while control litters were left unculled (unculled litter size = 12–16). In a subset of animals, we compared the growth by the end of exposure to CIH or after 10 days left unexposed. The mass of control subjects [6.58 ± 0.29 g (n = 11)] was not different from that of CIH exposed subjects [7.07 ± 0.38 g (n = 15), P = 0.33; F = 2.24, P = 0.18]. The CIH paradigm was executed during the light cycle and lasted for 8 h/day (i.e., 80 intermittent hypoxia cycles/day) for 10 consecutive days. A single hypoxic cycle was achieved by flowing 100% N₂ into the chamber for approximately 60 s that created a hypoxic environment where the nadir O₂ chamber value reached 4–5% O₂ (for 5–7 s). The hypoxic phase was immediately followed by an air break (300 s). The air break was achieved by flushing the chamber with air (~21% O₂) restoring a normoxic state (18–20%) within 60 s following hypoxia. Environmental CO₂ did not rise greater than 0.02% during any phase. Both pups and their dam were exposed to the CIH paradigm. In a subset of experiments, pups exposed to CIH were treated daily with the cell-permeable superoxide anion scavenger, 5,10,15,20-Tetrakis(1-methylpyridinium-4-yl)-21H,23H-porphyrin manganese(III) pentachloride, (MnTMPyP; 25 mg/kg daily; route: i.p.) throughout the entirety of exposure to CIH or for 10 days without CIH.

CD1 pups (n = 7) that received MnTMPyP treatment over the course of 10 days (from p1 to p10) without CIH exposure had similar masses to both control and CIH (9.53 ± 0.72 g at p10). Electrophysiological experiments were conducted where both preBötC and XIIn were simultaneously recorded (n = 3). The irregularity score of amplitude was 0.11 ± 0.03 and the irregularity score of period was 0.32 ± 0.08 in the preBötC and 100% transmission was observed between preBötC and XIIn (data not shown).

Transverse brainstem slices containing the preBötC (Hill et al., 2011; Garcia et al., 2013b; Zanella et al., 2014) were prepared from either CIH-exposed mice (12–48 h after the end of the CIH paradigm) or naïve CD1 mice (postnatal day 10–12). In brief, the isolated brainstem was glued to an agar block (dorsal face to agar) with the rostral face up and submerged in artificial cerebrospinal fluid (aCSF, ~4°C) equilibrated with Carbogen (95% O₂, 5% CO₂). Serial cuts were made through the brainstem until the appearance of anatomical landmarks such as XIIn and inferior olive. A single slice (550–600 μm) was retained containing the preBötC and XIIn.

The composition of 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 305–312 mOSM and a pH of 7.40–7.45 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. Control oxygen conditions were made by equilibrating aCSF with carbogen (95% O₂, 5% CO₂) while hypoxic conditions were made by aerating with 95% N₂, 5% CO₂. Despite the equilibration of aCSF with 0% O₂, hypoxic media contained some O₂ (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 or the XIIn. Recorded signals were amplified 10,000X, filtered (low pass, 1.5 kHz; high pass, 250 Hz), rectified, and integrated using an electronic filter. 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.

Intracellular recordings were made from putative inspiratory neurons of the preBötC using a multiclamp amplifier (Molecular Devices, Sunnyvale, CA). We defined the putative inspiratory preBötC neurons as neurons that received excitatory synaptic input in phase with the network rhythm—independent of degree of action potentials generated per cycle. These recordings were made in current clamp configuration using the blind patch clamp approach. Recordings were made using borosilicate glass recording electrodes (4–6 MΩ that were pulled using a P-97 Flaming/Brown micropipette puller (Sutter Instrument Co., Novato, CA) and filled with intracellular patch electrode solution containing (in mM): 140 potassium gluconate, 1 CaCl₂, 10 EGTA, 2 MgCl₂, 4 Na₂ATP, and 10 Hepes (pH 7.2). The calculated junction potential of −12 mV was subtracted posthoc from membrane potential.

Both extracellular and intracellular recordings were acquired in pCLAMP software (Molecular Devices, Sunnyvale, CA) and were analyzed posthoc in Clampfit software (version 10.2).

The degree of lipid peroxidation was determined by measuring malondialdehyde (MDA) levels in sectioned tissue samples containing either the preBötC or XIIn. Each sample n contained tissue pooled from two subjects either from the control or the CIH group. MDA levels were assayed as previously described (Garcia et al., 2011; Raghuraman et al., 2011) and expressed as nanomoles per milligram of protein.

Short term variability of a given metric was quantitatively described by the irregularity score (Zanella et al., 2014). Kernel density estimation was conducted and plotted using Matlab. Comparisons between two groups were made using Unpaired Student's t-test with a Welch's correction to avoid assuming equal variances between groups using Graphpad Prism (Graphpad Software Inc., La Jolla CA). Unless otherwise stated, comparisons between two groups were plotted using box-whisker plots where the error bars represent the maximum and minimum values of the data. Numerical data reported were expressed as the mean ± standard error of the mean. Linear regression analysis was used to describe the fidelity of an individual neuron to the respective population rhythm.

Two and three-dimensional graphs of kernel density functions were calculated and plotted using Matlab. Univariate kernel density estimates the probability density function of a single trait or variable for a given data set. We also used bivariate kernel density (KS_2D) to give a bivariate estimation of the probability density function of 2-dimensional data set. KS_2D was defined as: KS2D(x,y)=KS(xi,…n) x KS(yi,…n), where KS(x_i,…n) and KS(y_i,…n) are a univariate kernel density functions of two independent data sets such as (1) irregularity scores of amplitude (x) and period (y) where n is sample size.

The ratio between the integrated burst areas generated in the hypoglossal nucleus (XII) and preBötC were used to examine the input-output (I/O) relationship between the pre-motor network (i.e., preBötC) and the motor pool (i.e., XIIn). We refer to this ratio as the I/O ratio, which as defined by the following equation: I∕O ration=∫BAXIIn∕∫BApreBötCn where ∫BA_XIIn is the integrated burst area of XII of the n^th cycle and ∫BA_preBötCn is the corresponding integrated burst area of preBötC of the n^th cycle. In cases where a preBötC burst was detected but no corresponding burst was detected in XIIn, the integrated burst area of XIIn was assigned a zero value.

The standard deviation perpendicular (SD1) to the identity line and the standard deviation scattered along the identity line (SD2) were used to quantify the degree of dispersion in Poincare plots. SD1 described short term variability while SD2 described long term variability (Brennan et al., 2001).

Extracellular population recordings were made from the preBötC of control (n = 57) and CIH-exposed (n = 58) subjects. Although, all mice had the same genetic background (CD-1), CIH exposure resulted in an unexpected increase in individual variability of period and amplitude of respiratory activity. The central respiratory activity of some CIH-exposed subjects was indistinguishable from control (Figure 1A), but many CIH exposed individuals exhibited central respiratory rhythmic activities that were markedly more irregular in the amplitude and/or period of respiratory population activity (Figure 1B). To quantify these observations, we compared the irregularity score of amplitude (IrS_Amp) and the irregularity score of period (IrS_P) between rhythmic population activities of CIH exposed and control animals. Both the IrS_AMP (control: 0.17 ± 0.01, CIH: 0.24 ± 0.02; P = 0.004) and the IrS_P (control: 0.25 ± 0.01, CIH: 0.33 ± 0.02; P = 0.003) were greater in CIH treated groups. In agreement with our qualitative observations, the variance between the two groups was different for both IrS_P (F-ratio = 3.45, P_Ftest < 0.0001) and IrS_Amp (F-ratio = 2.11, P_Ftest < 0.006). This finding was readily demonstrated in the KS_2D of IrS_AMP and IrS_P of individual rhythms and indicates that CIH dispersed the tight distribution of composite burst-to-burst regularity of the preBötC rhythm that was characteristic for control animals (Figure 1C).

Simultaneous extracellular recordings of the population rhythms from the preBötC and the corresponding XIIn were recorded in brainstem slices from CIH and control subjects. Intermittent failure in transmission of the preBötC rhythm to the hypoglossal network was evident in the CIH group and contrasted the reliability observed in the control group (Figure 2A). In networks with increased amplitude irregularity after CIH exposure, failed transmission to XIIn occurred during network bursts that were characterized by small amplitudes (Figure 2A, yellow boxes). Transmission from the preBötC to XIIn was lower following CIH (Figure 2B; n = 10 CON, n = 16 CIH; P = 0.01, F-ratio = 690, P < 0.0001). Comparing the I/O ratios from networks exposed to CIH with transmission rates <90% (n = 5) to the control group revealed a significant, yet variable, reduction in the input-output relationship between preBötC and XIIn (Figure 2C, P = 0.01; F-ratio = 41.63, P < 0.00001). Poincare plots of the I/O ratio revealed that CIH exposure led to erratic transmission of the preBötC rhythm to XIIn (Figure 2D). Quantitative analysis of short-term variability (SD1: Control = 0.24 ± 0.04, CIH = 0.44 ± 0.08, P = 0.02; F-ratio = 2.64 P = 0.208) and long term variability (SD2: Control = 0.31 ± 0.04, CIH = 0.54 ± 0.09, P = 0.014; F-ratio = 3.05 P = 0.16) in the Poincare plots were different following CIH. Together these data suggests that CIH changed the I/O relationship and this change correlated to the increased irregularity of preBötC amplitude.

To further understand the impact of CIH on the preBötC, we recorded the activity of individual preBötC neurons. Exposure to CIH changed the reliability of action potential generation during the network burst (Figure 3A), which resulted in fewer action potentials generated during the population burst in network obtained from animals that were exposed to CIH (Figures 3B,C; control: 20 ± 5, n = 10; CIH: 8 ± 1, n = 16; P = 0.034, F-ratio = 7.84, P = 0.008). This raised the question whether CIH changed the fidelity of activity between an individual neuron and the network population rhythm.

To account for a potential masking effect due to the differences in absolute output between control and CIH exposed neurons, we examined the relationship between normalized output of an individual neuron to that of the corresponding network burst. This analysis revealed that the output range an individual neuron (i.e., the number of action potentials generated during a network burst) expands following CIH when compared to control neurons (Figure 4A). Specifically, the number of action potentials generated by a single neuron is often attenuated and even absent despite the generation of network activity. Linear regression analysis revealed that both r²-values (control: 0.58 ± 0.08, CIH: 0.35 ±.06; P = 0.034; F-ratio = 1.11, P = 0.90) and fitted slope values (control: 0.75 ± 0.10, CIH: 0.44 ± 0.08; P = 0.021; F-ratio = 1.38, P = 0.99) are reduced following CIH (Figures 4B,C). The smaller r²-value in the CIH group indicated that the number of action potentials are weaker predictors of network activity following CIH. Secondly, the smaller slope also suggested that changes in number of action potentials from these neurons have a smaller impact on network activity following CIH. Thus, the effects of CIH at the single cell level extend beyond a simple reduction of overall neuronal excitability, but rather appears to involve a change in the predictive fidelity of individuals in participating in the network rhythm.

The augmenting effects of CIH on the output from the carotid bodies of the peripheral nervous system can be mitigated with antioxidant treatment and suggested that CIH promotes pro-oxidant state which affects physiology in the nervous system (Pawar et al., 2009). Furthermore, our previous work in the preBötC demonstrated that exogenous reactive oxygen species and oxidative stress significantly affects rhythmogenesis from the preBötC (Garcia et al., 2011). Thus, to determine whether CIH affected the pro-oxidant state within the preBötC, we measured the amount of MDA, a common end product of lipid peroxidation resulting from oxidative stress. In the preBötC, amount of MDA was larger in control (5.21 ± 0.20 nM per mg; n = 4) compared to CIH (7.22 ± 0.35 nM per mg; n = 4). While this difference in the amount of MDA was significant in the preBötC (Figure 5A; P = 0.003; F = 3.14, P = 0.37), no differences were observed in MDA from XIIn (Figure 5B; control: 5.48 ± 0.34 nM/mg, n = 4 vs. CIH 6.50 ± 0.79 nM/mg, n = 4; P = 0.28). Based on these observations, we sought to determine whether the provision of the antioxidant, MnTMPyP (25 mg/kg), affected the outcome of CIH on the isolated rhythmic network. Although MnTMPyP treatment did not appear to affect CIH-mediated period irregularities (i.e., in IrS_P was not different between the CIH and CIH with MnTMPyP, P = 0.24; F-ratio = 1.90, P = 0.28), the antioxidant appeared to prevent amplitude irregularities within the preBötC (Figure 6A; n = 12). IrS_Amp between rhythms from subjects exposed CIH alone and rhythms from subjects receiving MnTMPyP during CIH was different (Figure 6B; P = 0.018; F-ratio = 6.72, P = 0.002). MnTMPyP treatment during CIH also appeared to preserve transmission of preBötC the rhythm to XIIn (n = 10; Figure 6C). This was confirmed by the difference in transmission rates between the CIH and the antioxidant group (Figure 6D; CIH: 89.15 ± 4.12%; MnTMPyP: 99.02 ± 0.471%; P = 0.03; F-ratio = 122.5, P < 0.001).

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.