This is a secondary analysis of previously published studies. The details for subject characteristics, nocturnal hypoxia model, polysomnography and assessment of sleepiness, attention and working memory were described elsewhere.10 In brief: eleven healthy subjects (7 men, 27.0 ± 1.5 years old, body mass index of 23 ± 0.9 kg/m²) without significant medial and sleep disorders were recruited. Subjects were exposed during sleep to nine hours of continuous mildly hypocapnic hypoxia. Initial acclimatization consisted of the first night at sea level, the second at 7,700 feet and the third at 10,000 feet and then the experimental hypoxic condition was given for 14 consecutive nights at 13,000 feet using a normobaric “altitude tent system” (Colorado Altitude Training, Colorado Springs, Colorado, USA). Subjects were at sea level during the day. FiO₂ (0.13) and CO₂ (0.1–0.52%) level was independently verified. Standard polysomnography using Embla (Embla, Denver, Colorado, USA) system was performed at baseline and 3 points during the course of the study (night 3, night 7, and night 14). Polysomnography included recording the EEG, electrooculogram, chin and tibialis electromyogram, thermistor, nasal pressure, thoracic and abdominal effort, ECG, and finger oxymetry. Conventional sleep staging and events scoring including arousal and sleep disordered breathings followed the standard methods.10 Vigilance and working memory assessment consisted with a 10-minute verbal 2-back task and the Psychomotor Vigilance Task (PVT). Performance was measured during the daytime between 08:00 and 10:00 hours to minimize circadian effect at four time points (day 4, 8, 9 and 15). The number of lapses on the PVT and percentage of correct response and mean response time on 2-back task were the primary variables of functional status. Besides, subjective sleepiness and mood was assessed on a scale of 0 (very sleepy or depressed mood) to 10 (usual sleepy or normal mood state).

Delta spectral power was performed using commercially available polysomnography software, RemLogic 1.0 (Embla, Denver, Colorado, USA) at four time points at the baseline and night 3, 7, and 14. Each time point corresponded to preexposure, early-acclimatized, midexposure and postexposure. Spectral power analysis was performed in the central electrode (C4-A1) during stable N2 and N3 sleep using ‘sleep power band analysis’ (RemLogic 1.0) that shows the relative contributions of each EEG spectrum in the given period. Spectral power band consisted with delta (0.5–4 Hz), theta (4–8 Hz), alpha (8–12 Hz), sigma (12–14 Hz), beta (14–25 Hz), and higher bands (> 25 Hz). The relative power in the delta band was the primary measure of interest for analysis and expressed as percentage (%). Stable sleep was defined as sleep periods with significant fragmentation of the EEG (no wake epochs or respiratory arousals) and fluctuations in electromyography and cardiorespiratory signals (Fig. 1). The time threshold was two or five continuous minutes for N2 and 2 minutes for N3.

Details of our CPC method have been published.17 In brief, using a single lead ECG, an automated beat detection algorithm detects beats and classifies them as either normal or ectopic. The algorithm determines amplitude variations in the QRS complex that occur due to shifts with respiration in the cardiac electrical axis relative to the electrodes, and due to changes in thoracic impedance as the lungs fill and empty. Fluctuations in the mean cardiac electrical axis (typically between 1 degree and 12 degrees peak-to-peak) correlate with phasic changes in the respiratory cycle. From these amplitude variations a surrogate ECG derived respiratory signal (EDR) is obtained. A time series of normal-to-normal sinus (NN) intervals and the time series of the EDR associated with these NN intervals is then extracted from the RR interval time series. Outliers due to false or missed R-wave detections are removed using a sliding window average filter with a window of 41 data points and rejection of central points lying outside 20% of the window average. The resulting NN interval series and its associated EDR are then cubic spline resampled at 2 Hz. The cross-spectral power and coherence of these two signals are calculated over a 1024 sample (8.5 minute) window using the Fast Fourier Transform applied to the 3 overlapping 512 sample subwindows within the 1024 coherence window. The 1024 coherence window is then advanced by 256 samples (2.1 minutes) and the calculation repeated until the entire NN interval/EDR series is analyzed. For each 1024 window the product of the coherence and cross-spectral power is used to calculate the ratio of coherent cross power in the low frequency (0.01–0.1 Hz) band to that in the high frequency (0.1–0.4 Hz) band. A preponderance of power in the low frequency band tends to be associated with periodic sleep behaviors regardless of etiology, while excess power in the high frequency band is associated with physiologic respiratory sinus arrhythmia and stable/deep sleep. CPC spectrograms reveal that the LFC and HFC regimes closely follow CAP scoring.

Specifically, LFC is generally associated with CAP and HFC with non-CAP. It is also observed that the ratio of power in the very low frequency (0–0.01 Hz) to the combined power in the low and high frequency bands may be used to estimate periods of wake or REM sleep, where a preponderance of power in the very low frequency band is associated with wake/REM periods. Using appropriate thresholds for the power ratios in these bands, sleep demonstrating stability and HFC, instability and LFC coupling, and wake/REM states can be identified (Fig. 2). A further characterization of the low-frequency coupled state is performed by determining the % (time) demonstrating elevated-LFC power. This category of coupling is strongly correlated (> 0.8) with apneas and hypopneas scored in the PhysioNet Sleep Apnea Database (http://www.physionet.org/physiobank/database/apnea-ecg/).21 It provides an objective method to introduce a severity dimension to LFC. It does not necessarily signify apneas per se, as arousals due to other reasons may appear indistinguishable. HFC was a variable of primary interest.

Means and standard deviations were summarized for delta spectral power and CPC variables. Individual responses to nocturnal hypoxia and cognitive performance is known to be significantly different between subjects. Therefore, randomized block analysis of variance was used to assess significance difference between protocol time points with EEG delta power and HFC as dependent variables. Subject and time points of sleep study are parameters for ‘block’. Post hoc Tukey test was implemented for multiple comparisons. Statistical analysis was performed by using SPSS software (SPSS 12.0, SPSS Inc., Chicago, IL, USA). Null hypothesis was rejected at p-value < 0.05.

Sleep indices from the CPC analysis are presented in Table 1. The amount of HFC, i.e., stable sleep, changed across the entire hypoxic exposures (p = 0.02). HFC was different between pre-exposure and mid-exposure (night 7; 52.3 ± 23.6% vs. 39.0 ± 16.7%, post hoc Tukey test, p = 0.02) but returned to the baseline level at the post-exposure (night 14; 45.4 ± 18.2%, p = 0.39). Contrary to the change in HFC, slow wave sleep decreased continuously along the hypoxic exposure, although it was not statistically significant (7.6 ± 5.7% vs. 4.7 ± 2.6%, p = 0.19). Among conventional sleep architecture parameters, the other significant differences across the protocol were in total sleep time (baseline to early acclimatization only), stage 1 sleep (baseline to early acclimatization only), and the American Academy of Sleep Medicine arousal index (significantly different from baseline to early acclimatization). Breathing during sleep is highly susceptible to hypoxic exposures. Respiratory disturbance index increased immediately after hypoxic exposures. It was significantly higher across the hypoxic period than the baseline (p = 0.002)(Table 1). There was no significant difference in sleep efficiency or REM sleep measures across the study duration (Table 1). Delta spectral power band analysis showed maintenance of relative delta power during hypoxic exposures. Relative delta spectral power during stable sleep was not significantly different across the entire hypoxic exposure (p = 0.98, 0.25 for N2; p = 0.66 for N3).

As reported before, there were no significant differences from the baseline across the period of exposure in any of the following variables: subjective assessment of state (a composite of mood and sleepiness), 2-back mean reaction times, percentage of correct responses, and the lapses on PVT lapses (Table 1). Other variables from the PVT including mean, median, 10% slowest reaction times or false starts did not change along the hypoxic exposure.