Effects of Chronic Sleep Fragmentation on Wake-Active Neurons and the Hypercapnic Arousal Response

Abstract

Study Objectives:

Delayed hypercapnic arousals may occur in obstructive sleep apnea. The impaired arousal response is expected to promote more pronounced oxyhemoglobin desaturations. We hypothesized that long-term sleep fragmentation (SF) results in injury to or dysfunction of wake-active neurons that manifests, in part, as a delayed hypercapnic arousal response.

Design:

Adult male mice were implanted for behavioral state recordings and randomly assigned to 4 weeks of either orbital platform SF (SF4wk, 30 events/h) or control conditions (Ct4wk) prior to behavioral, histological, and locus coeruleus (LC) whole cell electrophysiological evaluations.

Measurements and Results:

SF was successfully achieved across the 4 week study, as evidenced by a persistently increased arousal index, P < 0.01 and shortened sleep bouts, P < 0.05, while total sleep/wake times and plasma corticosterone levels were unaffected. A multiple sleep latency test performed at the onset of the dark period showed a reduced latency to sleep in SF4wk mice (P < 0.05). The hypercapnic arousal latency was increased, Ct4wk 64 ± 5 sec vs. SF4wk 154 ± 6 sec, P < 0.001, and remained elevated after a 2 week recovery (101 ± 4 sec, P < 0.001). C-fos activation in noradrenergic, orexinergic, histaminergic, and cholinergic wake-active neurons was reduced in response to hypercapnia (P < 0.05-0.001). Catecholaminergic and orexinergic projections into the cingulate cortex were also reduced in SF4wk (P < 0.01). In addition, SF4wk resulted in impaired LC neuron excitability (P < 0.01).

Conclusions:

Four weeks of sleep fragmentation (SF4wk) impairs arousal responses to hypercapnia, reduces wake neuron projections and locus coeruleus neuronal excitability, supporting the concepts that some effects of sleep fragmentation may contribute to impaired arousal responses in sleep apnea, which may not reverse immediately with therapy.

Citation:

Li Y; Panossian LA; Zhang J; Zhu Y; Zhan G; Chou YT; Fenik P; Bhatnagar S; Piel DA; Beck SG; Veasey S. Effects of chronic sleep fragmentation on wake-active neurons and the hypercapnic arousal response. SLEEP 2014;37(1):51-64.

INTRODUCTION

Obstructive sleep apnea (OSA) is characterized by recurrent, brief sleep state-dependent periods of complete or partial collapse of the upper airway, where pharyngeal collapse results in both hypoxemia and hypercapnia.¹ Hypercapnia is an important stimulus for arousal and resolution of apneic events.¹ Delayed arousals will increase the severity of hypoxemia. Individuals with OSA show blunted hypercapnic ventilatory drive in wakefulness that improves over time with therapy.^2,3 Children with OSA who are otherwise healthy can show impaired arousal responses to hypercapnia.⁴ Mechanisms for impaired hypercapnic arousals in OSA are not known.

Work in animal models supports the concept that two physiological disturbances in OSA, intermittent hypoxia (IH) and sleep fragmentation (SF), influence both wakefulness and ventilatory drive. Exposure to long-term IH in rodents results in spatial learning and memory decrements^5–10 and irreversible wake impairments.^11,12 IH neurobehavioral deficits are associated with significant neuronal injury in specific brain regions.^6,8,13–16 Lasting wake impairments in IH are associated with loss of catecholaminergic wake neurons, without signifi-cant injury to other wake-active neuronal populations, including the orexinergic and histaminergic wake neurons.¹⁷ IH, however, does not appear to impair either ventilatory drive or hyper-capnic responses. In fact, IH may enhance ventilatory drive by increasing carotid sinus nerve and sympathetic activity.^18–21 In contrast, SF may impair ventilatory responses. In wakefulness following 24 h of SF, rats have attenuated hypercapnic ventilatory responses, despite normal resting arterial blood gases.²² Monitoring electroencephalographic waveforms after SF, Bowes et al. found normal hypercapnic ventilatory responses but an elevated arousal threshold for arterial carbon dioxide,²³ and also for laryngeal stimulation and hypoxia following short-term SF.²⁴ Thus, it appears that short-term SF can impair arousal responses to hypercapnia, a response that would lengthen apneas and result in more severe hypoxemia and acid-base dyshomeostasis. Elicited arousals from sleep have been shown to markedly increase firing rates in at least one group of wake-active neurons, locus coeruleus neurons.²⁵ We hypothesized that frequently repeated elicited arousals in chronic sleep fragmentation, modeling OSA, might impose metabolic stress on wake-active neurons, impairing the wake neuron response to hypercapnia.

In this series of studies, we first determined the long-term (4 week) effectiveness of a SF protocol previously validated for short-term SF.²⁶ We then characterized the effects of SF4wk on wakefulness and arousal responses to hypercapnia and tactile stimuli and then examined the effects of SF4wk on wake neuron excitability, axonal projections and the c-fos response to a hypercapnic challenge. The work presented identifies reduced excitability in locus coeruleus wake-active neurons and finds hypercapnic arousal impairments extending into recovery upon long-term sleep fragmentation.

METHODS

Mice, Animal Care, and Study Overview

Adult male mice (C57Bl/6J, 10-12 weeks) were housed on a 12-h lights-on: 12-h lights-off cycle and fed ad libitum standard rodent chow and water. Ambient temperature and humidity were maintained between 21-23°C and 35% to 60%, respectively. The methods and study protocols conformed to the revised National Institutes of Health Office of Laboratory Animal Welfare Policy and were approved in full by the University of Pennsylvania Institute for Animal Care and Use Committee. Mice were randomized to 4 weeks of control rested conditions (Ct4wk, n = 17), sleep fragmentation for 4 weeks without recovery (SF4wk, n = 22), or SF4wk with a 2 week recovery (SF4wkRec, n = 5). Sleep-wake recordings were performed before and during the SF and Ct conditions. Following sleep-wake recordings, mice were examined for wakefulness and arousal responses to tactile and hyper-capnic stimuli, histological studies examining the wake-active neurons, and/or brain slice recordings of locus coeruleus neurons to assess excitability.

Surgical Implantation of Sleep Recording Electrodes and Electrophysiological Recordings

Under general anesthesia, mice were implanted with chronic sleep-wake recording electrodes, as previously detailed,²⁷ with the addition of application of a dental adhesive (Super-Bond, Sun Medical) for long-term recordings.²⁸ Mice were given a 1 week recovery with littermates prior to single housing and connecting recording cables. Mice then had an additional week to adjust to single housing and counter-weighted cables prior to recordings. Frontal EEG and nuchal EMG signals were filtered, amplified, digitized, and recorded as previously described.²⁷

Behavioral State Analysis

Raw EEG and EMG data were exported to SleepSign (version 3.0, Kissei) for analysis. Sleep-wake states were classified as wake, NREM, or REM sleep using 4-second (4sec) epochs to allow for detection of brief arousals. Wake was defined by low amplitude, fast desynchronized frequency EEG and relatively high amplitude EMG; NREM sleep was defined by EEG delta frequencies (0.5-4Hz) comprising > 30% of EEG waveforms/epoch with associated lower amplitude EMG (moving average adjusted per animal by scorer), and REM sleep was defined as delta frequencies comprising < 20% of waveforms/epoch and theta (5-10 Hz) comprising > 30% of the EEG in the epoch with a low EMG. Once automatically scored by the software program, each epoch of each 24-h recording was reviewed and corrected by a trained scorer blinded to the condition.

Wake, NREM and REM sleep times were measured as the total time in each stage for 24 h, the 12-h lights-on period, and the 12-h lights-off period (n = 9-13/group). Brief arousals were defined as ≤ 3 epochs (12 sec) of wake preceded by ≥ 3 epochs of sleep. Wake bouts were defined as ≥ 4 epochs of wake, preceded by ≥ 3 epochs of sleep. Sleep bouts (either REM or NREM) were defined as ≥ 3 consecutive sleep epochs. The arousal index was defined as the combined frequency of brief arousals and wake bouts per hour of sleep. Arousal and bout indices were calculated as means/24 h. Values are expressed as mean ± SE. Data were analyzed using Microsoft Excel 2007, Stata/IC 11.2 and Prism 6.0b. Differences in sleep parameters across time were assessed for SF mice across baseline (SFBL), day 1 (SF1d), and 4 weeks (SF4wk), and in Ct mice across baseline (CtBL) and 4 weeks (Ct4wk) using two-way repeated-measures ANOVA, followed by Tukey HSD posttests to analyze each pair of comparisons, with values reported as Studentized Range Statistic q (range = 3, df = 26). To distinguish SF effects on sleep/wake from effects of chronic tethered sleep recording and/or single housing, Ct4wk mice were compared to SF4wk using a non-paired t-test, Bonferroni corrected for the number of variables in each analysis.

EEG Spectral Analysis

Fronto-parietal EEG waveforms from 24-h recordings were subjected to spectral analysis at baseline and at 4 weeks for each behavioral state (wake, NREM sleep and REM sleep). A frequency distribution of μv² for 0.25 Hz bins across 0.5-20 Hz was generated and normalized to total power for that behavioral state for each animal. Frequency distributions for each behavioral state were averaged in SF (n = 5) and control mice (n = 5) at each time point. Spectral distributions were analyzed using 2-way ANOVA for time and SF condition. To examine decline in slow wave activity across the lights-on period, we calculated the delta power in NREM sleep for each hour of the lights-on period, and expressed these values relative to the average delta power for the last 4 h of the lights-on period, as previously described.^27,31 Relative delta power values were analyzed using 2-way ANOVA for time and SF condition with Tukey post hoc tests.

Sleep Fragmentation Apparatus

SF was induced using methods validated by Sinton et al.²⁶ using an orbital rotor (MaxQ 2000; Thermo Scientific, Marietta, OH) with speed set at 110 rpm, and a repeated cycle of 10 sec on, 110 sec off continuously across 4 weeks controlled by a timer (H3CR-F8-300, OMRON Corporation, Kyoto, Japan). The auditory stimulus used in the Sinton model was omitted to avoid awakening Ct mice housed in the same room. This arousal frequency (30/h) was chosen to be potentially representative of arousal indices in humans who have moderate-to-severe sleep apnea. Four to 8 cages were placed on the enlarged rotor platform (65 cm × 120 cm). Prior to study, we ensured that all mice were able to groom, eat, and drink during orbital rotor movement of the platform, yet would arouse from sleep upon orbital rotor movement. Water bottles with long nozzles equipped with ball valve tips were used to prevent leakage with platform movement. To ensure that mice were eating and drinking sufficiently, weights were obtained at 1 week and 4 weeks of the SF or Ct period and compared using two-way ANOVA for time and SF condition.

Plasma Corticosterone Levels

As an index of stress in the SF and sleep recording paradigm, plasma corticosterone was measured in SF and Ct conditions 2 h after lights-on. Groups of mice studied included: (1) unimplanted colony mice housed with littermates, n = 4; (2) Ct1wk (n = 5); (3) SF1wk (n = 5); (4) Ct4wk (n = 5); (5) SF4wk (n = 5). Once deeply anesthetized (200 mg/kg intraperitoneal pentobarbital), 50-100 μL of blood was spun to separate plasma for measurements using a ³H radioimmunoassay kit (MP Biomedicals), as directed. Values were measured against a standard curve that was created using a set of known corticosterone concentrations ranging 0.025 ng/0.5mL to 1 ng/0.5mL. Data were analyzed using two-way ANOVA for SF condition and time into study.

Multiple Sleep Latency Testing (MSLT)

For the MSLT, the rotor table was stopped 30 min prior to the onset of the 12-h lights-off period, and an MSLT was performed as previously detailed with mice remaining in their home cages and the first nap beginning at lights-off.³² Sleep latency was calculated as the latency to 30 sec of uninterrupted sleep. In addition, we calculated the latency to the first 4 sec epoch of sleep,³³ hereafter termed “microsleep” latency. Sleep-onset and microsleep-onset latency times for the 4 naps were averaged for each mouse and analyzed using one-way ANOVA for SF condition (Ct4wk vs. SF4wk), Bonferroni corrected for measuring 2 latencies. A second test of sleep latency was the average time to sleep onset for all sleep periods in a 24-h recording.³³ This microsleep onset latency was determined for BL and 4 week recordings for Ct and SF groups, using two-way ANOVA for sleep condition and time into study.

Hypercapnic Arousal Latency and Hypercapnic c-fos Activation Response

To assess the effect of SF4wk on hypercapnic arousal latency, the rotor platform was stopped 1 hour prior to trials. Two lines of gas mixtures (“room air”: 21% O₂, remainder N₂ and “hyper-capnia”: 21% O₂, 5% CO₂, remainder N₂) were joined with a Y-piece attached to the cage. Stop valves allowed either gas mixture to be bled into the cage. Cage tops were modified to provide for efficient exchange of gas mixtures. Throughout experimentation, flow rates into the cage were held constant at 1 L/min. Following 30 sec of NREM sleep observed on electro-graphic recordings, the room air mixture was switched to hyper-capnia, and the latency to arousal after reaching an ambient CO₂ in the home cage of > 4% (range 4% to 5%) was detected as the first 4 sec epoch scored as wake. Four successful hypercapnic arousals from NREM sleep were measured for each mouse during the mid-4 h of the lights-on period. A third group of mice were allowed a 2 week recovery after SF and were examined for CO₂ arousals as above.

To determine whether the impaired arousal response was secondary to a generalized arousal impairment or specific to impaired chemosensitivity, SF or Ct conditions were resumed for 24 h. We then measured the threshold to a normocapnic pressure arousal by delivering pressurized air 1 inch from the mouse, via a sideport, at incremental pressures (pounds/square inch, PSI) until the mouse awakened. Briefly, using the same cage set-up as described for CO₂ responses, we bled in 21% O₂, balanced with N₂ into the animal's home cage at 1 PSI until the animal entered NREM sleep for 30 sec. Once in NREM sleep, using a regulator, the pressure was increased by 1PSI every 2 sec until the animal awoke (arousal threshold). Trials were repeated in each animal 4 times within the mid 2-h of the lights-on period. Latencies were compared in a non-paired t-test for SF4wk and Ct4wk.

Several groups of wake-active neurons, including histaminergic, noradrenergic, orexinergic and serotoninergic neurons, demonstrate an increase in nuclear c-fos in response to hyper-capnic challenge and may contribute to hypercarbic arousal responses.^34–37 To examine the c-fos response to hypercapnia, we delivered an exchange of 21% oxygen (O₂) and 5% carbon dioxide (CO₂) with remainder nitrogen (N₂) at 1L/min into the home cages of experimental mice. Humidity across the exposure was 30-35% and ambient temperature was 21-23°C. Ambient CO₂ level was measured continuously with a capno-graph (Capstar-100, CWE) and ranged 4% to 5% throughout the exposure. Mice were exposed to 2.5 h of hypercapnia and then deeply anesthetized with pentobarbital prior to transcardial perfusion of paraformaldehyde and procurement of the brain, as described below in immunohistochemistry.

Immunohistochemistry

Immunohistochemistry was performed to examine the c-fos response to wakefulness and to examine LC and orexinergic projections to the cingulate cortex. Paraformaldehyde perfused brains were cryopreserved, coronally sectioned (40 μm) and placed in 1:6 series, as previously described.³⁸ C-fos antibodies and antibodies used to detect wake-active neurons: LC (tyrosine hydroxylase, TH), serotoninergic (serotonin, 5-HT), orexinergic (orexin-A, Orex-A), histaminergic (histidine decarboxylase, HDC), and cholinergic (choline acetyltransferase, ChAT) neurons were recently detailed.³⁸ C-fos antibody was labeled with Alexafluor 594, and the wake-active neuronal identifier was labeled with Alexafluor 488 (Molecular Probes), as recently detailed.¹² C-fos wake-active neuron data were analyzed as percentage of strong nuclear labeling for n = 5/group using two-way ANOVA for SF and hypercapnia conditions. C-fos was also examined in the cingulate cortex using DAB to label NeuN positive neurons and nickel to label c-fos. The density of TH and orex-A labeled projections into cingulate cortex 1 (Cg1) and orexinergic projections into the LC was measured in SF4wk and Ct4wk mice by labeling TH or orex-A fibers with alkaline phosphatase Blue Substrate (Vector Labs). Light microscopic images were imported into Image J, where a grid with 225 800 μm² gradations was placed over Cg1 to count all labeled neurite segments crossing opposite sides of one gradation.³⁸ Counts were obtained for 3 sections/animal. Mean counts/section were analyzed using one-way ANOVA for differences in neurite segment counts for SF4wk vs. Ct4wk mice (n = 5/group). Similarly, orexinergic projections into the LC were measured and analyzed, using a grid of 100 800 μm² gradations placed over the LC nucleus and dendritic field.³⁹ A complementary analysis was used to measure the percentage of the LC area covered by orexinergic neurites. Using ImageJ, images were converted to 8-bit gray scale and then transformed to binary images to obtain the % orexinergic projection coverage in the LC. Mean data per animal were analyzed for Ct4wk (n = 5) and SF4wk (n = 5) as a one-way ANOVA.

Brain Slice Locus Coeruleus Neuron Recordings

Whole cell LC recordings were performed to determine the effect of SF4wk on LC neuron excitability. Following SF4wk and Ct4wk conditions, mice were decapitated with brains immediately immersed in 2-4°C oxygenated Ringer's and prepared as 200 μm horizontal slices that were continuously perfused with oxygenated aCSF (1.5 mL/min). Composition of the aCSF (in mM) was: 124, NaCl; 2.5, KCl; 1.25, NaH₂PO₄; 2.0, MgSO₄; 2.5, CaCl₂; 10, dextrose; and 26, NaHCO₃. Sections were gradually warmed to 32-33°C for recording of one cell/slice. Pipette resistance was 6-10 MΩ when filled with an intracellular solution (in mM): 130 K-gluconate, 5 NaCl, 10 Na-phosphocreatinine, 1 MgCl₂, 0.02 EGTA, 10 HEPES, 2 MgATP, 0.5 Na₂GTP, and 0.1% biocytin (pH 7.3). Inhibitory postsynaptic currents (IPSCs) were recorded in voltage-clamp mode (V_m = -70 mV). Rheobase was determined by injecting current in 10pA steps, and recordings were collected with a Multiclamp 700B amplifier, Digidata 1320 A/D converter, and Clampex 9.0 software (Molecular Devices). LC neurons in slice were identified by proximity to the 4th ventricle, translucency of the LC nucleus, and soma ≥ 20 μm and after recordings confirmed with biocytin/TH labeling. Baseline whole cell recordings were obtained upon stability of firing rate for 10 min (± 0.2Hz). In current clamp mode neurons were further selected for prominent afterhyperpolarization (AHP), characteristic of LC neurons.⁴⁰ Current-voltage (I-V) was recorded from 100 pA with increments of 20 pA. The time constant τ was measured from the membrane potential response to -20 pA current injection. Membrane potential was held at approximately -50 mV to prevent spontaneous firing. Spike frequency adaptation and the number of action potentials (F) generated in response to increasing depolarizing current pulses (0 pA to 220 pA in 20 pA steps, F-I plot) was measured.⁴¹ Recordings were analyzed using pCLAMP software (Axon Instruments) and Clampfit 10.0 (Molecular Devices) as previously detailed.^42,43 F-I slope and AHP amplitude were used to characterize excitability. Secondary measures of neuronal characteristics were time constant τ, capacitance, resistance, action potential (AP) threshold, AP duration, and resting membrane potential (RMP).

RESULTS

General Health and Corticosterone Levels across SF

Physical appearance did not vary in either group across time. Mice were provided with nesting materials, and all SF and Ct mice built nests. There were no differences in weights across groups of mice relative to single housing and tethered sleep cables or SF (Figure 1A). SF4wk mice weighed 28.8 ± 0.5 g, and Ct4wk mice weighed 27.9 ± 0.2 g, t = 1.7, N.S. Similarly, there were no differences in corticosterone across groups (Figure 1B). Thus, long-term SF and chronic cable recordings do not influence corticosterone levels.

Figure 1.

Weight and corticosterone effects of sleep fragmentation (SF). Mice were assessed for total body weight and plasma corticosterone levels to assess for stress effects of the implemented SF paradigm. Values were compared to colony mice who were group-housed, non-implanted and did not have SF. The Ct1wk group were mice implanted for sleep recordings, and singly housed without SF. Ct4wk mice were implanted for sleep recordings, singly housed for 4 weeks without SF. SF1wk were mice implanted for sleep recordings, and singly housed with sleep fragmentation for 1 week; and SF4wk mice were implanted for sleep recordings, and singly housed with sleep fragmentation for 4 weeks. Shown are mean ± SE for n = 5/group. No significant differences across groups were noted for either weights or plasma corticosterone levels.

Rotor-Platform Sleep Fragmentation (SF) Is Effective Long-Term

Overall, there were significant differences in the total arousal index (the number of brief arousals and wake onset bouts/h) across treatment (F = 14.5, P = 0.0008). Data are summarized in Figure 2. Tukey post hoc analysis revealed an increase in the total arousal index relative to baseline for both SF1d (q = 5.5, P < 0.01) and SF4wk (q = 6.6, P < 0.01), Figure 2A. In contrast, in Ct mice there was no change in arousal index with time CtBL vs. Ct4wk, P = 0.31. There were no differences between SFBL and CtBL in sleep/wake parameters. In Ct mice, neither brief arousal or wake bout indices changed across time. SF increased the number of wake bouts/24h at SF1d relative to baseline (q = 4.6, P < 0.01, Figures 2B and 2C), while at SF4wk, there was no difference in the wake bout index compared to baseline (q = 0.06, N.S.). In contrast, brief arousals were not increased at SF1d (q = 1.4, N.S.), yet brief arousals were increased at SF4wk (q = 9.97, P < 0.01). In support of disrupted sleep, NREM sleep bout durations were also abbreviated upon SF (F = 6.03, P = 0.021). Specifically, NREM bout durations, while similar to controls at baseline (SFBL 76 ± 6 min vs. CtBL 82 ± 6 min, P = 0.45), were reduced at SF1d (62 ± 4 min, q = 3.64, P < 0.05) and SF4wk (58 ± 3 min, q = 4.65, P < 0.01), but remained stable in Ct4wk mice (80 ± 7 min, P = 0.79), relative to CtBL. Representative 30-min hypnograms are presented in Figures 2D-2F. In summary, the rotor platform effectively fragments sleep for at least 4 weeks, initially increasing the number of wake bouts and over time increasing brief arousals only.

Figure 2.

Orbital shaker platform movement effectively fragments sleep for 4 weeks. (A-C) Shown are mean ± SE for control mice (light gray columns) and SF mice (black columns) for three specific measures of sleep disruption: (A) total arousal index (number of arousals per hour of sleep); (B) wake bout index (number of wake bouts/hour of sleep), and (C) brief arousals (number of arousals with wake lasting < 12 sec per hour of sleep). Lines denote significant differences, as detailed in the results. (D) Representative hypnograms for the first 30 min of the lights-on period show the effect of shaker table movement on sleep wake activity, comparing baseline to the first day of sleep fragmentation and at SF4wk. W, wakefulness; R, rapid-eye movement sleep; and N, non-rapid eye movement sleep. Many arousals on day 1 result in wake bouts, and in contrast, most arousals by SF4wk are brief. Lines denote statistical difference across groups: *P < 0.05; **P < 0.01; ***P < 0.001.

Sleep Wake Characteristics across Time and SF

There was no difference in wake time in CtBL vs Ct4wk mice, P = 0.11 (n = 9, summarized in Figure 3A). In mice exposed to SF (n = 13), at 1 day there was a trend towards increased wake time during the 12-h lights-off period, from 482 ± 14 min at baseline to 509 ± 21 min SF1d, N.S. (or 68.3% ± 2.0% baseline to 71.9% ± 3.0% SF1d), Figure 3A. By SF4wk, the lights-off period wake time had declined relative to SF1d to 445 ± 21 min (or 63.9% ± 3.0%), P < 0.05, and was not different than SFBL wake time. Similarly, there was no difference in wake time between Ct4wk and SF4wk mice for 12-h lights-on (P = 0.06), 12-h lights-off (P = 0.51), or 24 h (P = 0.09). While there was no significant effect on NREM sleep times (Figure 3B), lights-on REM sleep time/12 h demonstrated a significant decline across SF, 60.11 ± 6.4 min SFBL vs. 43.0 ± 7.6 min at SF4wk (or 8.57% ± 0.91% SFBL vs. 6.27% ± 1.10% at SF4wk), (q = 3.60, P < 0.05) without a change for lights-off REM sleep (Figure 3C). There were no other differences in sleep-wake times observed. The lights-on: lights-off diurnal ratio for total sleep time in Ct4wk mice was 1.64 ± 0.1 and in SF4wk mice was 1.47 ± 2, t = 0.2, N.S. In summary, SF4wk reduces REM sleep time during the lights-on period without effects on total sleep and wake times.

Figure 3.

Sleep wake behavioral state times across rotor platform sleep fragmentation. (A-C) 12-h lights-on (light gray columns) and 12-h lights-off (black columns) provide composite histograms of the percent time spent in wake, NREM, and REM sleep expressed as mean ± SE across various time points and SF conditions. Shown from the left are: Control baseline (CtBL); Ct mice at 4 weeks (Ct4wk); SF baseline (SFBL); SF first day (SF1d) and SF at 4 weeks (SF4wk). Lines denote statistical difference across groups: *P < 0.05.

Effects of Sleep Fragmentation on EEG Power Spectra across Behavioral States

Sleep recordings were performed on a series of SF4wk mice (n = 5) for BL, SF1d and SF4wk and Ct mice (n = 5) for time points CtBL and Ct4wk to assess both the NREM sleep delta decline across the lights-on period and spectral power distribution in the all 3 behavioral states across a 24-h period. Overall, there was a significant time of day effect on delta power, P < 0.001. In Ct mice delta power for each time of day point was unchanged between CtBL and Ct4wk. In contrast, SF mice showed reduced relative delta power in NREM sleep in the first 2 h of the lights-on period, P < 0.05 (Figure 4A, 4B). EEG spectral frequency distribution for each behavioral state was also assessed in the same groups of SF and Ct mice at BL and 4 weeks. There were no significant differences in the distribution of relative EEG power at any of the 0.25-Hz intervals in SF or Ct mice (Figure 4C-4H). Observing a trend towards increased delta in waking in SF4wk mice (Figure 4C), a secondary analysis examined differences within the delta power range (0.5-4Hz) in wakefulness across groups and also found no significant differences in delta/total power. In summary, SF reduces delta decline in NREM sleep across the lights-on period and has minimal effects on behavioral state power spectra frequency distribution.

Figure 4.

The effects of chronic SF on behavioral state EEG spectra over time. (A-B) Time course across the 12-h lights-on period for NREM sleep delta power (0.5-4 Hz) normalized to the average delta power in the last 4 h of the lights-on period. (A) Shown are mean ± SE relative delta power values/h for mice exposed to SF (n = 5), assessed at 3 time points: baseline (BL, red circles), SF day 1 (SF1d, blue squares) and at 4 weeks (SF4wk, black triangles) and non-linear best-fit lines for each data set time point. Asterisks denote significant differences between SF1d and SF4wk for Zeitgeber 1 h and 2 h (P < 0.05). (B) Relative delta power mean ± SE for control mice (Ct, n = 5) at time points: baseline (BL, red circles) and 4 weeks (Ct4wk, black triangles). A best-fit nonlinear line is drawn for each time point. (C-E) Mean ± SE EEG power for each 0.25Hz frequency normalized to total EEG power across 0.5-20Hz for waking (C), NREM sleep, (D) and REM sleep (E). Three time points were assessed in SF mice (n = 5): BL (red circles), SF1d (blue squares), and SF4wk (black triangles). (F-H) Mean ± SE behavioral state relative power for control mice at 2 time points: BL (red circles) and Ct4wk (black triangles).

Effects of Chronic Sleep Fragmentation on Sleep Latency

SF4wk reduced the lights-off MSLT sleep latency, P < 0.05, Figure 5A. The latency to microsleep (first NREM sleep epoch) was also reduced in mice exposed to SF4wk relative to Ct4wk, P < 0.01 (Figure 5B). We next analyzed the microsleep onset latency averaged across all spontaneous sleep attempts in the 12-h lights-on and lights-off periods. There were no differences in microsleep onset latencies between CtBL and Ct4wk mice. In contrast, we found a significant reduction in the microsleep onset latency in SF4wk mice relative to SF BL for the lights-on period at 4 weeks, P < 0.001, and no significant reduction for the lights-off period, where microsleep onset latencies showed greater variation (Figures 5C, 5D). Thus, SF reduces sleep latencies measured with both the MSLT and microsleep-onset latencies.

Figure 5.

Chronic sleep fragmentation shortens sleep latencies. Four measures of sleepiness in mice that have been described in the literature are presented with mean ± SE for Ct4wk and SF4wk mice. (A) Results from a multiple sleep latency test (MSLT) (n = 9/group) performed at the onset of the lights-off period, where sleep latency is defined as the onset to 30 seconds of consecutive sleep. (B) MSLT data analyzing the latency to microsleep onset as the latency to the first epoch (4 sec) scored as NREM sleep in the MSLT. For Ct4wk mice the sleep latency is the same as the latency including microsleeps, suggesting there are no microsleeps. In contrast, the SF4wk mice show shorter latencies to microsleeps, suggesting that these events do occur even at the onset of the lights-off period. (C-D) These latencies are average values obtained spontaneously during a 24-h recording across the lights-on (C) and lights-off (D) periods from the onset of any wake bout (> 12 sec wake) to the first sleep epoch (4 sec) in Ct (light gray columns) or SF group (black columns) at baseline (BL) and after Ct4wk or SF4wk (4wk). Lines denote significant differences: *P < 0.05; **P < 0.01; and ***P < 0.001.

Hypercapnic arousal threshold is increased in mice exposed to SF4wk and persists despite a 2 week recovery

Overall differences were observed in the arousal latency in response to hypercapnia, P < 0.0001, as presented in Figure 6. Mice exposed to SF (n = 5), relative to Ct mice (n = 5), showed markedly prolonged latencies to arousal in hypercapnia (SF4wk 154 sec ± 6, vs. Ct4wk 64 sec ± 5, t = 11.3, P < 0.001). Following 2 week recovery, latencies remained increased in SF4wk mice at 109 sec ± 8, t = 5.2, P < 0.01, but were reduced relative to SF4wk, t = 6.2, P < 0.01 (Figure 6A). SF4wk mice showed an increased tactile arousal threshold relative to Ct4wk mice, t = 10.9, P < 0.001 (Figure 6B). In summary, SF4wk results in delayed hyper-capnic and tactile arousal responses; although improved, the delayed hypercapnic arousal in SF4wk mice remains 2 weeks into recovery.

Figure 6.

Chronic sleep fragmentation delays the hypercapnic arousal and increases the tactile arousal threshold. (A) Control (Ct) and sleep fragmentation (SF) 4wk mice and SF4wk after 2 weeks recovery (SF4wkRec) were exposed to 5% hypercapnia while latencies to arousal were measured for 4 exposures in NREM sleep. (B) Air delivered at incremental pressures into the cages (increasing pressure per square inch, PSI) of SF4wk mice relative to Ct4wk required a higher pressure for arousal, n = 5/group. Lines denote significant differences between groups: **P < 0.01; ***P < 0.001.

SF influences the c-fos response to hypercapnia in several groups of wake neurons

Overall there were significant CO₂ and SF effects on c-fos expression in selected wake groups of wake-active neurons, P < 0.0001. Data were Bonferroni corrected to t = 3.1 for P < 0.05. Data are summarized separately for each wake-active cell group, as described below and summarized in Figure 7A.

Figure 7.

C-fos activation is blunted in mice exposed to SF in select groups of wake-active neurons. (A) Histograms are presented for 5 of the wake active neuronal groups showing mean c-fos % labeling in response to normocapnia (white columns) vs. hypercapnia (dark gray) in Ct4wk and SF4wk mice. Lines denote significant differences in c-fos values within a nucleus: *P < 0.05; **P < 0.01; ***P < 0.001. (B) Representative confocal microscopic images of orexinergic (red) neurons and c-fos (green) labeling in coronal lateral hypothalamic sections from mice (Ct4wk and SF4wk) under conditions of normocapnia, room air (RA) and hypercapnia (CO2). C-fos labeling is far weaker and few dendrites are appreciable in sections from the SF mice. Arrows delineate the c-fos+ orexinergic neurons in each image. Calibration bar, 50 μm.

Orexinergic neurons in Ct4wk mice showed greater percentage of nuclear c-fos than SF4wk for normocapnic conditions, t = 3.6, P < 0.05. Ct4wk mice showed a strong c-fos response to hypercapnia in orexinergic neurons, t = 3.1, P < 0.05, while SF4wk mice showed no response to CO₂, t = 1.4, N.S. Consequently, c-fos activation in hypercapnia was higher in orexinergic neurons in Ct4wk mice than in SF4wk mice, t = 6.8, P < 0.001. Representative images are presented in Figure 7B.

For LC noradrenergic neurons, under normocapnic conditions c-fos varied across Ct4wk and SF4wk, t = 7.9, P < 0.01. In response to CO₂, both Ct4wk and SF4wk mice showed an upregulation of c-fos (Ct4wk, t = 4.9, P < 0.01 and SF4wk, t = 3.3, P < 0.05). There was a very large difference in LC c-fos across Ct4wk and SF4wk under hypercapnic conditions, t = 9.5, P < 0.05.

Serotoninergic neurons showed low levels of c-fos under normocapnic conditions in both Ct4wk and SF4wk mice, without a difference across groups, t = 1.0, N.S. Both Ct4wk and SF4wk groups showed robust increases in c-fos in response to hypercapnia: Ct4wk t = 6.7, P < 0.01, and SF4wk t = 5.5, P < 0.01, so that there was no difference in the c-fos response to hypercapnia between Ct4wk and SF4wk, t = 2.3, N.S.

Histaminergic neurons showed no difference between Ct4wk and SF4wk for normocapnic conditions, t = 2.3, N.S. Histaminergic neurons in Ct4wk mice showed a strong response to CO₂, t = 4.7, P < 0.01, while there was no apparent CO₂ response for histaminergic neurons in SF4wk mice, t = 2.3, N.S., so that in comparing Ct4wk and SF4wk CO₂ c-fos responses for histaminergic neurons, a far larger response was observed in Ct4wk mice, t = 4.6, P < 0.01.

Under normocapnic conditions, percentage cholinergic neurons with c-fos labeling was similar for Ct4wk and SF4wk, t = 1.3, N.S. Ct4wk showed an increase in c-fos in response to hypercapnia, t = 6.4, P < 0.01. In contrast, there was no increase in c-fos for SF4wk mice, t = 2.3, N.S. As a consequence under hypercapnic conditions, c-fos was higher in Ct4wk mice than in SF4wk mice, t = 5.3, P < 0.01.

Having identified reduced c-fos expression in many wake neurons, we sought to distinguish reduced c-fos in wake neurons with reduced c-fos as a more generalized phenomenon. Thus, we counted c-fos+ neurons in layers II/III of the cingu-late cortex to where noradrenergic fibers project. Here, no SF effect was observed in cingulate c-fos, N.S., as summarized in Figure 7. Thus, wake-active neurons, aside from 5-HT neurons, show reduced c-fos activation in response to hypercapnia in mice exposed to SF4wk, relative to Ct4wk mice.

Sleep fragmentation is associated with reduced orexinergic and locus coeruleus axonal projections in the prefrontal cortex and orexinergic projections to LC

We next examined axonal projections for locus coeruleus and orexinergic neurons. Tyrosine hydroxylase projections into the anterior cingulate cortex (Cg) are almost exclusively from the locus coeruleus,⁴⁴ and a loss of projections would impair arousal.⁴⁵ Using a Bonferroni-corrected non-paired two-way t-test, we found significant reductions in both tyro-sine hydroxylase projections to the cingulate 1 region (Cg1), primarily at level II/III (t = 8.8, P < 0.001) and orexinergic projections into the deep Cg1, below level VI (t = 4.5, P < 0.05). The percentage area over the LC nucleus covered by orexinergic projections was reduced in SF4wk mice, relative to Ct4wk mice (t = 3.9, P < 0.05, Figure 8). Consistent with axonopathy, orexinergic projections consisted of beaded, curled axons with smaller boutons. In summary, SF4wk appears to reduce axonal projections from both locus coeruleus and orexinergic neurons, and induces morphological changes consistent with axonopathy.

Figure 8.

Chronic sleep fragmentation (SF) induces axonopathy in locus coeruleus and orexinergic neurons. (A) Substrate Blue labeling of the TH projections in the cingulate cortex in the area measured for analysis of axon projection density as segments crossing gradations. Representative sections from a Ct mouse (left) and a SF4wk mouse (right) are provided to illustrate loss of TH fiber immunointensity and projections at all levels (I-VI) of the cortex. Cortical layers in the section are marked as I, II/III and IV. (B) Averages ± SE for tyrosine hydroxylase-positive projections (left panel) measured in the anterior cingulate bilaterally in 5 control (Ct4wk, gray) mice and 5 mice exposed to SF4wk (black columns); orexinergic fibers measured similarly in the cingulate cortex (middle panel) and the density of orexinergic projections in the LC, expressed as % of area covered (right panel). Data are presented as the numbers of 10mm segments estimated for 1 mm² anterior cingulate 2 cortex/section. Lines denote significant differences: *P < 0.05; ***P < 0.001. (C) An example of the orexinergic projections (substrate blue) into the mid locus coeruleus region, where typically the most dense orexinergic boutons reside. Upper panels show lower magnification to show density, calibration bar, 25 μm; lower panels at higher power illustrate the fine axons, beading of axons and loss of boutons. Calibration bar, 10 μm.

Sleep fragmentation reduces excitability of locus coeruleus neurons

Complete recordings were successfully accomplished in LC neurons later confirmed to be TH-positive LC neurons in Ct4wk (n = 17) and SF4wk (n = 14). Data and statistics are summarized in Table 1. A representative image of a confirmed positive LC neuron with biocytin after recording is shown in Figure 9. SF4wk influenced LC neuron excitability. Specifically, F-I slopes (mean ± SE) were reduced in SF4wk relative to Ct4wk by 40%, t = 3.2, P < 0.01, and AHP was increased in SF4wk by 100%, t = 5.4, P < 0.0001. A secondary analysis (time constant τ, capacitance and resistance) was performed to determine whether electrophysiological changes were consistent with neurite loss. Consistent with a loss of total surface area, SF4wk mice had an increased τ and reduced capacitance with increased resistance. Time constant τ was increased 7 msec in SF4wk relative to Ct4wk, t = 3.3, P < 0.01. Moreover, resistance increased 33% in SF4wk, t = 2.3, P < 0.05, and capacitance decreased by 35% in SF4wk relative to Ct4wk, t = 7.2, P < 0.0001. Thus, the electrophysiological properties of LC neurons in mice exposed to SF4wk show reduced excitability and are consistent with reduced neurite complexity.

Table 1.

Effects of SF4wk on locus coeruleus neuron electrophysiology

Figure 9.

Sleep fragmentation impairs locus coeruleus neuron excitability. Whole cell patch clamp recordings were performed in mice after SF4wk (n = 7) and Ct4wk (n = 7). (A) A representative verification of biocytin-labeled (red) recorded neuron within a 200 μm slice showing co-localization of tyrosine hydroxylase (TH, green) within the locus coeruleus. (B) Representative time constant tau for Ct4wk (red arrow) and SF4wk (blue arrow). Statistics are provided in Table 1. (C) Representative raw data traces of action potentials (APs) induced by depolarizing current pulses in a Ct4wk mouse and a SF4wk mouse. The SF4wk mice have a lower number of APs induced. (D) Mean ± SE frequency-current (F-I) graph demonstrating group differences in excitability in response to 100-220 pA current injection. (E) Summary mean ± SE (F-I) slope data for Ct4wk (red) and SF4wk (blue); **P < 0.001.

DISCUSSION

Sleep fragmentation (SF) is an integral physiological characteristic of obstructive sleep apnea (OSA), even in its mildest of forms. It is estimated that over 5% to 9% of the adult and 2% of the child population in developed countries has OSA.^46–48 Because OSA develops slowly and is frequently present for years prior to diagnosis, SF is present in large numbers of individuals for long periods of time prior to therapy. Yet, we know very little of the consequences of long-term SF on brain function and health.

In the present study, we explored the effectiveness of orbital movement for long-term SF. Several different approaches to sleep disruption have been developed in animal models to begin to identify important neurobehavioral consequences of SF.^{24,26,49–54} One method that avoids physical movement to disrupt sleep is optogenetic activation of a select population of wake-active neurons. This may be accomplished with opto-genetic activation of either orexinergic or noradrenergic wake neurons.^55,56 Whether this method could be implemented for long-term SF is not known. Several other approaches induce arousals very effectively by encouraging ambulation at specified intervals. These paradigms use either a treadmill or a bar that sweeps along the bottom of the cage, prompting the animal to step over the bar at fixed intervals.^50–53,57 One potential downside is that the sweeper or treadmill movements make nest building and sleeping in a nest more difficult. Consequently, some researchers use the sweeper models for 12 h/d to allow unperturbed sleep/nest time; yet, in OSA, all sleep is fragmented. Sinton et al. developed the orbital movement model of SF used in the present study to which they added a noise during platform movement to augment SF.²⁶ They validated the effectiveness of orbital movement SF for 48 h, and in the present study we extend this effectiveness to 4 weeks, showing minimal effects on total sleep times. Further, this model appears to be minimally stressful as it allows SF mice to be maintained in normal mouse cages with usual bedding and nesting materials and does not force ambulation, and corticosterone levels are unaffected.

During obstructive sleep disordered breathing events in OSA, arousal to ventilatory disturbances, including hypercapnia, is an important protective mechanism to restore ventilation and acid-base status and to minimize hypoxemia and hemodynamic fluctuations.⁵⁸ Hypercapnia is an important contributor to the apnea arousal response. In the present studies we observed a marked delay (> 1 min) in the hypercapnic arousal latency in mice exposed to SF4wk. Previously, Liu et al. examined hypoxic and hypercapnic ventilatory responses immediately after SF24hr in adult rats and identified an impaired hyper-capnic ventilatory response.²² The impaired hypercapnic ventilatory response was short-lived, resolving 2 h after recovery from SF. Homeostatic sleep drive may have contributed to the response, as treating rats with an adenosine A1 antagonist prevented this impaired hyper-capnic ventilatory response. The SF duration in the present studies was far longer, and a 2 week recovery after SF4wk was allowed in one group of mice to assess reversibility. We reasoned that if the impaired response was secondary to increased sleep homeostatic drive, 2 weeks of uninterrupted sleep should allow the arousal latency to fully normalize. Yet, after 2 weeks of recovery the arousal threshold in the present study was still significantly prolonged, supporting the concept that a slow-to-resolve change in the central nervous system underlies the prolonged hypercapnic arousal. Children with OSA (presumably for months or years) can demonstrate an increased hyper-capnic threshold to arousal from sleep and a longer latency to arousal.⁴ Surgical correction of OSA with tonsillectomy and adenoidectomy gradually shortens the time to the hypercapnic arousal and lowers the end-tidal CO₂ levels at arousal.⁴ Berry et al. found that withdrawal of positive airway pressure therapy in patients effectively treated for OSA results in a progressive lengthening of apneic events (duration of apnea before arousal) over several days of treatment withdrawal, suggesting that OSA can also acutely raise arousal thresholds.⁵⁹ Integrating short-term SF and OSA with our present long-term SF results and clinical OSA, we propose a model (Figure 10), whereby dysfunction of hypercapnia-responsive wake-active neurons becomes a more important influence with longer durations of sleep disruption, and this influence may take longer to resolve.

Figure 10.

Proposed model for sleep fragmentation (SF) effects on arousal response. SF, both acutely (24 h) and chronically (weeks), results in increased sleep drive which plays a role in delayed arousal responses to both tactile stimuli and hypercarbia. Chronic SF also leads to dysfunction of wake-active neurons, which further contributes to impaired arousal and sleepiness.

To establish whether SF4wk can directly impair the function of wake-active neurons, we examined whole cell LC neuronal recordings in slice. Significant reductions in LC neuron F-I slopes and increased AHP amplitude in SF4wk relative to Ct4wk strongly support impaired excitability. The findings of increased resting membrane potential (relative depolarization), increased resistance, and lack of effect on the action potential amplitude or firing rate in SF4wk mice relative to controls are inconsistent with an adenosine effect. Adenosine applied in bath hyperpolarizes LC neurons and reduces spontaneous firing rates.⁶⁰ Similarly the response is not consistent with a corticotropin releasing factor stress response, which would increase the spontaneous firing rate of LC neurons⁶¹ and over time expand the LC neuronal dendritic field which would reduce capacitance.⁴⁴ Collectively the findings are most consistent with neuronal dysfunction.

The EEG power spectral responses to SF are consistent with dysfunction of the locus coeruleus. Specifically, we found little change in behavioral state power spectra frequency distribution, yet a dampening in the delta decline across the lights-on period. Similarly, neurotoxin lesioning of the locus coeruleus has no appreciable effect on the frequency distribution of the waking, NREM sleep or REM sleep EEG power, yet blunts the NREM sleep delta decline across the lights-on period.⁶² Mice lacking the major norepinephrine-synthesizing enzyme dopa-mine beta hydroxylase have increased delta power (1-4 Hz) in waking and REM sleep.⁶³ While our SF mice showed trends in this direction, the findings were not statistically significant. It is possible that with larger sample sizes, an increase in waking relative delta power would be discerned.

C-fos protein is translocated to the nucleus upon cellular activation or activation of specific signal transduction pathways.⁶⁴ Consequently, c-fos is used as an indirect measure of neuronal activity. There are important limitations in that not all cells express c-fos with all types of activation, and c-fos can remain elevated when protease degradation is impaired, or when intra-cellular calcium is high.^65,66 By focusing on groups of neurons with known c-fos responses to wakefulness and hypercapnia and by comparing normocapnic c-fos responses to hypercapnic c-fos responses, we can conclude that nuclear translocation of c-fos in wake-active neurons does not occur as readily in SF4wk, relative to Ct4wk mice. Normocapnic levels for most wake-active neuronal groups did not vary between SF4wk and Ct4wk mice. Consistent with previous studies,^67–69 serotoninergic, noradrenergic, histaminergic and orexinergic neurons demonstrated increased c-fos nuclear translocation in response to hypercapnia in Ct4wk mice. In the present study hypercapnia revealed large differences in nuclear c-fos across SF4wk and Ct4wk mice. Collectively, these findings support impaired responsiveness to hypercapnia. The impaired hypercapnic and tactile arousal latencies, and reduced LC neuron excitability support a generalized dysfunction of arousal systems.

SF4wk mice evidenced loss of LC and orexinergic labeled axons. Impaired connectivity via axon loss could also contribute to the observed impaired arousal responses. We postulate that the mechanisms underlying axonopathy in chronic SF involve metabolic/oxidative stress to neurons that are repeatedly activated upon arousal. Spontaneous discharge of LC neurons is sleep state-dependent, with rates reduced in NREM sleep and further reduced in REM sleep. However, tone-elicited arousals in NREM sleep result in large increases in LC neuron activity (approximately 10-fold), twice as large as LC responses in waking.²⁵ The more pronounced neuronal firing response to elicited arousals in NREM sleep relative to wake may reflect reduced inhibitory autoreceptor tone in NREM sleep. LC neuronal activation in elicited arousals from REM sleep has not been characterized but is likely to cause even greater neuronal activation, as autoreceptor tone would be further reduced. Prolonged sleep disruption in rodents also increases glutamate release and firing rates of cortical neurons.^70,71 In a similar fashion, it is likely that the increased activation of neurons during SF is in part a consequence of increased glutamate neurotransmission. Repeated neuronal depolarization could result in calcium dyshomeostasis that would, in turn, promote mitochondrial injury and oxidative and endoplasmic reticulum stress, processes implicated in axonopathy by disrupting axonal transport.^72,73 Nair et al. demonstrated impairments in spatial learning and memory retention in mice subjected to 14 days of sleep fragmentation for 12 h per day; the SF mice had upregulation and activation of NADPH oxidase in the cerebral cortex.⁵³ Sportiche et al. found spatial learning and memory deficits lasting at least 2 weeks after SF, supporting the concept that neuronal dysfunction induced by SF may be slow to resolve and potentially partially irreversible.⁷⁴

In summary, the present studies validate an orbital movement SF paradigm for long-term SF studies. While SF4wk has minimal effects on sleep/wake power spectra and total times spent in each stage, SF4wk has a pronounced effect on LC neuron excitability and results in reduced axonal projections into key wake-related regions, suggesting that connectivity may be impaired by SF. SF induces impairments (delays) in hypercapnic arousal that only partially correct with a 2 week recovery period. Delayed arousals in OSA are expected to worsen oxygenation, alter acid-base homeostasis, and potentially cause hemodynamic disturbances in sleep. As the severity of hypoxemia in OSA is linked to cardiovascular morbidity and mortality,⁵⁸ identifying the mechanisms underlying the impaired response is of utmost importance.

DISCLOSURE STATEMENT

This was not an industry supported study. This work was supported in part by NIH (HL079555, HL096037, HL07713) and National Natural Science Foundation of China (Grant Number: 81070070, 81100991, 81200061) and State-Sponsored Post-graduate Study Abroad Programs from China Scholarship Council. Dr. Veasey is on the Galleon Pharmaceutical Scientific Advisory Board. The other authors have indicated no financial conflicts of interest.