Stable NREM Throughput and Respiratory-Control Efficiency: A Mechanistic Hypothesis for the Compression of Sleep Need

Locke

Abstract

Contemporary sleep medicine evaluates sleep primarily through total duration, but duration alone does not distinguish stable, restorative NREM from fragmented, autonomically unstable NREM. This paper proposes the Stable NREM Throughput Hypothesis: restorative output may depend not only on total sleep time or EEG-defined stage percentages, but on the absolute volume and density of high-frequency cardiopulmonary coupling (HFC)-dominant stable NREM. CO₂/H+ respiratory-control stability is proposed as one modifiable route to increasing stable NREM throughput. In individuals with high respiratory-control instability, substantial sleep time may be spent in low-frequency-coupling unstable NREM, requiring compensatory extension of total sleep duration. Conversely, improved respiratory stability may increase restorative density and, in selected individuals, reduce preferred sleep duration without functional impairment. An N-of-1 case observation is presented as a generating signal, and specific experimental designs for prospective validation are proposed. This framework is not a recommendation for sleep restriction; it is a testable hypothesis requiring CPC spectrograms, polysomnography, CO₂ monitoring, and objective daytime performance measures.

Introduction: Total Sleep Duration as an Imprecise Endpoint

The prevailing public health recommendation that adults require seven to eight hours of sleep per night is anchored primarily in large-scale epidemiological studies demonstrating U-shaped associations between self-reported sleep duration and adverse health outcomes, including cardiovascular disease, obesity, diabetes, and all-cause mortality (Cappuccio et al., 2010; Gangwisch et al., 2006; von Ruesten et al., 2012). These findings have been instrumental in establishing clinical guidelines (Watson et al., 2015) and shaping public health messaging.

However, this duration-centric framework contains several underexamined assumptions. First, the epidemiological evidence relies heavily on self-reported sleep duration, which consistently overestimates actual sleep time by 20 to 40 minutes compared to objective actigraphy and polysomnography (Lauderdale et al., 2008). When sleep is measured objectively via commercial wearable devices across large cohorts, the median actual sleep duration falls to approximately 6.7 hours (Zheng et al., 2024, Nature Medicine; N = 6,785 participants, 6.5 million person-nights in the NIH All of Us Research Program). The "seven to eight hour" recommendation may therefore describe time in bed rather than physiological sleep.

Second, and more fundamentally, total duration does not distinguish between sleep that is architecturally efficient and sleep that is fragmented, autonomically unstable, and restoratively impoverished. Two individuals sleeping seven hours may experience profoundly different restorative outcomes depending on the internal stability and composition of their NREM sleep. The duration metric treats these individuals as equivalent, obscuring the variable that may actually govern restorative output: the physiological quality of the NREM sleep obtained.

This paper proposes a reframing: total sleep duration is not an independent biological requirement but a dependent variable, substantially modulated by the restorative throughput of NREM sleep. When NREM sleep is stable, consolidated, and autonomically coherent, the restorative work of sleep completes efficiently, and the total temporal requirement contracts. When NREM sleep is unstable, fragmented by respiratory-driven microarousals and autonomic oscillations, the same restorative work requires a vastly extended temporal window. The population-level average of seven to eight hours may therefore combine true biological sleep requirement with unmeasured variation in architectural efficiency, including respiratory-control stability. In this framework, duration recommendations remain useful public-health heuristics, but they may not resolve the question of whether some individuals can achieve equivalent restorative output with less total sleep when stable NREM throughput is unusually high. This paper does not argue that short sleep is inherently healthy, nor that total sleep duration should be voluntarily reduced. It proposes that sleep duration may be partly compressible in selected individuals when stable NREM density increases enough to preserve absolute restorative output, and that CO₂/H+ respiratory-control stability is one modifiable input into this system.

To prevent misinterpretation, several explicit non-claims should be stated at the outset. This paper does not argue that baseline sleep need can be voluntarily overridden through willpower or behavioral habituation. It does not argue that REM sleep is expendable. It does not claim that the kinetics of homeostatic sleep pressure dissipation (Process S) can be accelerated. It does not propose that the mechanism generalizes beyond individuals with evidence of respiratory-control instability. The hypothesis is that respiratory instability imposes a throughput tax on NREM sleep, forcing the homeostatic clearance machinery to operate on a reduced fraction of total sleep time; removing that tax allows the same fixed-rate machinery to work on a larger proportion of the sleep period. The claim is not faster clearance but less wasted time.

Reframing Sleep Quality: Physiological Stability Over Stage Depth

The Limitations of EEG-Based Sleep Staging

Traditional polysomnographic (PSG) sleep staging, codified by the American Academy of Sleep Medicine (AASM), classifies NREM sleep into stages N1, N2, and N3 based on the morphology and frequency of cortical electroencephalographic (EEG) activity. Under this system, N3 (slow-wave sleep, SWS), defined by high-amplitude delta oscillations (0.5 to 4.0 Hz) occupying at least 20% of a 30-second epoch, is designated the "deepest" and most restorative stage. N2, characterized by sleep spindles and K-complexes but insufficient delta power to qualify as N3, is frequently conceptualized as a transitional stage. This staging hierarchy has shaped decades of sleep research, but it captures only cortical activity and does not directly measure the autonomic, respiratory, or cardiovascular stability that may be equally or more determinative of whether a given epoch of sleep is genuinely restorative.

Cardiopulmonary Coupling: A Multi-System Measure of Sleep Stability

Cardiopulmonary coupling (CPC) analysis, developed by Thomas and colleagues, offers a complementary measurement framework that captures the coherence between cardiac and respiratory oscillations during sleep (Thomas et al., 2005; Thomas et al., 2014). Using single-lead electrocardiogram (ECG) data, CPC computes the cross-spectral power and coherence between heart rate variability (HRV) and ECG-derived respiration (EDR), generating a sleep spectrogram that classifies sleep into distinct physiological states independent of cortical EEG staging.

High-frequency coupling (HFC; 0.1 to 0.4 Hz) serves as the biomarker for stable NREM sleep, characterized by robust parasympathetic dominance, strong respiratory sinus arrhythmia, stable breathing, physiological blood pressure dipping, and an elevated arousal threshold. Low-frequency coupling (LFC; below 0.1 Hz) indicates unstable NREM sleep, associated with sympathetic hyperactivity, cyclic breathing fluctuations, cyclic alternating pattern (CAP) sequences, and elevated microarousal frequency.

A critical finding from CPC validation studies is that the distinction between stable and unstable NREM does not map cleanly onto the N2/N3 EEG boundary. In healthy subjects, the proportion of HFC during visually scored N2 is statistically indistinguishable from that during N3 (70.1% versus 70.7%, P = 0.97; Thomas et al. validation data). This convergence is independently confirmed by computational endotyping: Sands and colleagues demonstrated that loop gain does not differ significantly between N2 and N3 sleep, indicating that respiratory-control instability affects NREM sleep uniformly across the N2/N3 EEG boundary. In a respiratory-stable individual, N2 sleep is physiologically similar to N3 along both autonomic/cardiopulmonary dimensions (CPC) and ventilatory-control dimensions (loop gain). The difference between stable and unstable N2 is greater than the difference between stable N2 and N3. Autonomic stability appears to be a major restorative axis that cuts across the N2/N3 EEG boundary; cortical stage depth remains important, but it is not the only axis of restoration.

Quantifying the Stability Differential

Table 1. CPC-derived sleep stability across populations (Thomas validation cohorts).

Thomas has proposed that clinically excellent sleep is characterized by approximately 50 to 60% stable NREM, approximately 20% REM, and the remainder as naturally unstable NREM (personal communication, 2026). This framework treats stability, not stage depth, as the primary metric of sleep quality.

Mechanistic Background: Respiratory Control and NREM Stability

Loop Gain and the Fragility of NREM Sleep

The stability of NREM sleep is heavily dependent on the behavior of the ventilatory control system during the transition from wakefulness to sleep. Upon sleep onset, the behavioral drive to breathe is withdrawn, and ventilation becomes entirely dependent on metabolic control, primarily the response of central and peripheral chemoreceptors to arterial partial pressure of CO₂ (PaCO₂). Loop gain is the engineering parameter that quantifies the sensitivity and stability of this feedback control system. An individual with high loop gain possesses a hyper-responsive chemoreflex: minor, physiologically normal fluctuations in PaCO₂ during sleep state transitions trigger exaggerated ventilatory responses, producing oscillating patterns of hyperventilation and hypoventilation that can cross the apneic threshold and trigger central respiratory pauses.

Polysomnographic data confirms that loop gain varies across sleep stages, with higher values during N2 compared to REM sleep (mean 0.63 versus 0.51), and increasing loop gain across the night (0.57 in the first third versus 0.64 in the final third). This stage-dependent and time-dependent variation partially explains why NREM sleep is more vulnerable to respiratory disruption than REM, and why sleep stability degrades as the night progresses.

The CO₂ Reserve and the Apneic Threshold

The CO₂ reserve, the margin between eupneic PaCO₂ and the apneic threshold, determines how much ventilatory perturbation can be tolerated before breathing ceases. Dempsey and colleagues established that this margin is narrow during sleep: approximately 3.8 mmHg in pure OSA and as low as 1.6 mmHg in mixed apnea phenotypes, with a general range of 3 to 6 Torr for the onset of apnea during sleep. N1 and N2 are specifically vulnerable stages, while slow-wave sleep is maximally stable. The wakefulness drive to breathe masks the proximity to the apneic threshold during waking hours, creating an asymmetry between waking and sleeping CO₂ tolerance that is central to understanding why respiratory-control instability manifests primarily during sleep.

The Stabilization Requirement: HFC-to-Delta Temporal Lag

Thomas et al. (2014, Sleep Medicine) utilized cross-correlation analysis of 3,150 polysomnographic datasets from the Sleep Heart Health Study. The analysis revealed a moderate but significant overall correlation (r = 0.40, P = 0.001) between delta power and HFC power, with the maximal cross-correlation occurring at a temporal lag of approximately 4 minutes, indicating that CPC coupling ratio fluctuations consistently precede delta power fluctuations. The moderate correlation strength indicates that HFC and delta power capture overlapping but distinct dimensions of sleep physiology. Rather than implying a deterministic link, the approximately 4-minute lag suggests that sustained autonomic stability increases the probability of subsequent delta emergence.

Sub-Threshold Respiratory Instability and Stable NREM Loss

Not all respiratory-control instability produces scoreable apnea, hypopnea, or oxygen desaturation. In a high-controller-gain phenotype, CO₂/H+ fluctuations may generate microarousals, sleep-stage transitions, or autonomic instability without crossing conventional event-scoring thresholds. These sub-threshold events would be expected to reduce HFC-dominant stable NREM while leaving AHI and oxygen saturation deceptively normal. This provides a mechanism by which sleep can remain architecturally inefficient despite apparently controlled obstruction. CPC analysis, which captures the autonomic signature of respiratory-driven instability regardless of whether event-scoring criteria are met, may provide the necessary sensitivity to detect this sub-threshold instability.

Eckert's PALM endotypic framework quantifies this phenomenon: approximately 36% of OSA patients have high loop gain as a primary contributing factor, and among patients with only mild anatomical compromise, approximately 70% are loop-gain-dominant. These individuals may present with controlled AHI on CPAP yet continue to experience unstable NREM and poor restorative output because the underlying controller gain remains elevated.

The Underweighted Substrate: Independent Restorative Functions of Stable N2

Cerebral Protein Synthesis

Picchioni, Schmidt, Loutaev et al. (2023, Journal of Cerebral Blood Flow and Metabolism) utilized positron emission tomography (PET) with L-[1-¹¹C]leucine to measure regional cerebral protein synthesis (rCPS) in 17 human subjects across waking and sleeping conditions. The study found that rCPS was 30 to 39% higher during combined N1/N2 sleep compared to N3 sleep. Remarkably, rCPS during N3 was statistically indistinguishable from values measured during sleep-deprived wakefulness. The differential was most pronounced in the thalamus (+0.50 nmol/g/min) and cerebellum (+0.60 nmol/g/min). The authors hypothesized that the massive, highly synchronous alternating up-states and down-states of delta oscillations during N3 may impose energetic constraints less permissive to continuous ribosomal translation. A critical caveat: N1 and N2 could not be statistically separated in this study, and replication by independent groups is needed.

Adenosine Clearance, Synaptic Downscaling, and Cardiovascular Restoration

Adenosine clearance begins upon sleep onset and accelerates through N1 and N2 as cerebral metabolic demand drops. The enzymatic rephosphorylation of adenosine to ATP occurs continuously throughout NREM sleep, with N3 providing the deepest metabolic rest but N2 contributing meaningful clearance. Sleep spindles and K-complexes during N2 perform targeted, selective synaptic pruning complementary to the broader depotentiation of N3. Blood pressure dipping occurs during HFC-dominant sleep independent of cortical delta waves. Siow et al. (2022, Neurology) demonstrated that absolute N2 duration was independently and positively associated with glymphatic function measured by DTI-ALPS index.

Non-Substitutable Functions of N3

Despite the independent restorative value of stable N2, specific physiological processes remain strictly dependent on N3. Growth hormone secretion is binary and time-locked to N3. Tasali et al. (2008, PNAS) demonstrated that selective acoustic suppression of SWS for three consecutive nights, replacing N3 with N2 while maintaining total sleep time, produced a 25% decrease in insulin sensitivity and a 20% decrease in the glucose disposition index, comparable to gaining 20 to 30 pounds of body weight. The relationship between stable N2 and N3 is therefore synergistic and sequential: stable N2 handles continuous maintenance while providing the autonomic platform for N3 to execute its specialized neuroendocrine functions.

The Stable NREM Throughput Hypothesis

The evidence presented above converges on a central hypothesis: the restorative output of a sleep episode may be more accurately predicted by the absolute volume and proportion of stable NREM (HFC-dominant) sleep than by total sleep duration or the percentage of any individual EEG-defined stage. This framing converges independently with recent computational phenotyping research: Sands and colleagues have characterized the therapeutic goal of lowering loop gain as enabling "sustained, uninterrupted NREM throughput," identifying stable NREM consolidation, rather than AHI reduction per se, as the mechanism by which respiratory stabilization restores neurocognitive function.

To formalize this hypothesis, two metrics are proposed. Stable NREM output is defined as the absolute minutes of HFC-dominant NREM per night. Stable NREM density is defined as HFC minutes divided by total sleep time, estimating the restorative concentration of a given sleep episode. The hypothesis predicts that sleep need may compress only if stable NREM density rises enough that absolute stable NREM output remains sufficient despite lower total sleep time.

A night with less total sleep but high stable NREM density may produce greater restorative output than a longer night dominated by unstable NREM. Total sleep duration is not a fixed biological requirement but a dynamic, compensatory variable that expands to accommodate the wasted time of unstable NREM and contracts when stable NREM density is optimized.

A critical mechanistic distinction must be stated precisely. This hypothesis does not propose that respiratory training accelerates the kinetics of homeostatic clearance. The biological time constants governing Process S dissipation appear to be conserved across phenotypes (Dijk and colleagues). Instead, the hypothesis proposes that respiratory instability imposes a throughput tax: in a high-loop-gain individual, a substantial fraction of NREM time is consumed by sub-threshold ventilatory cycling, autonomic oscillations, and aborted stage transitions rather than by the high-amplitude slow-wave activity that Process S clearance requires. Stabilizing respiratory control is hypothesized to increase the proportion of NREM time that is effective for clearance, operating within the same fixed kinetics rather than altering them. The distinction is between accelerating the engine and removing the drag.

The Inefficiency of Unstable NREM

During unstable, LFC-dominant NREM sleep, multiple restorative processes are impaired simultaneously. Protein synthesis is interrupted by microarousals, with each sympathetic spike triggering a metabolic state shift from anabolic to catabolic. Adenosine clearance may be partially offset by new adenosine generation from sympathetic activation and respiratory muscle effort, creating a homeostatic clearance inefficiency (this clearance inefficiency hypothesis requires direct empirical validation). Cardiovascular dipping fails during LFC-dominant periods. N2-to-N3 transitions are repeatedly aborted, with transition matrix analyses revealing up to 179.8% increases in N2-to-N1/Wake regressions in respiratory-unstable populations. Each failed transition, combined with the approximately 4-minute HFC-to-delta temporal lag, represents a significant temporal cost.

Independent confirmation comes from Process S research. Dijk and colleagues demonstrated that acoustic suppression of slow-wave activity during sleep, without reducing total sleep time or waking the subject, produced increased daytime sleepiness indistinguishable from partial sleep deprivation. Process S clearance stalled because the homeostatic dissipation mechanism requires continuous, high-amplitude cortical slow-wave synchronization to operate efficiently; when SWA expression was suppressed, the uncleared homeostatic pressure carried over into the subsequent waking day despite ostensibly "normal" sleep duration. This finding validates the throughput hypothesis from a completely independent measurement framework: it is not duration but the density and continuity of effective SWA within NREM that determines restorative output.

Natural Short Sleepers and the Biological Precedent

Familial natural short sleepers (DEC2, ADRB1, NPSR1, GRM1, SIK3 mutations; Fu, Ptácek, and colleagues, UCSF) provide biological evidence that total sleep duration is not absolutely fixed. Polysomnographic evaluation of the DEC2 Y362H variant (Pellegrino et al., 2014) revealed that the mutation carrier slept approximately one hour less per night than the non-carrier twin (actigraphy: 4.99 versus 6.05 hours) while maintaining "almost identical amounts of NREM sleep." The primary compressed stage was REM, not NREM. Critically, delta power was not merely preserved but appears amplified across multiple FNSS models: ADRB1 and SIK3 mutant mice exhibit elevated initial slow-wave activity, consistent with intensified homeostatic clearance per unit of sleep time. The Y362H carrier accumulated 40% fewer lapses on the Psychomotor Vigilance Test during 38 hours of forced sleep deprivation compared to the non-carrier twin, confirming that the compressed sleep was functionally sufficient. These findings directly support the throughput concept: the same restorative output achieved in less chronometric time through higher architectural density.

A critical distinction must be acknowledged. Fu and Ptácek draw a firm line between familial natural short sleep (genetic) and behaviorally restricted short sleep: there is no published support from the FNSS research program for the proposition that average individuals can train themselves to need less sleep through behavioral protocols. Habituation to chronic sleep restriction masks subjective fatigue while objective PVT deficits continue to accumulate. This paper does not claim to reproduce FNSS physiology through respiratory training. It proposes a narrower mechanism: that individuals whose NREM sleep is rendered inefficient by respiratory-control instability may recover architectural efficiency through chemoreflex stabilization, a distinct claim from the assertion that baseline sleep need can be voluntarily overridden.

CO₂/H+ Respiratory Stability as a Modifiable Input

If respiratory stability determines the proportion of NREM sleep that is restorative, the trainability of the chemoreflex becomes a critical question.

The Inherent-Trait Challenge

Sands and colleagues conducted a pivotal study in which OSA patients underwent six weeks of optimal CPAP therapy, completely eliminating nocturnal intermittent hypoxia and stabilizing sleep architecture. Following this protocol, elevated daytime loop gain did not normalize. Sands concluded that elevated loop gain is "largely an inherent, non-reversible trait" rather than an acquired consequence of the disease itself. This finding presents a direct challenge to any framework proposing loop gain modifiability and must be addressed honestly.

However, the intervention tested was CPAP, which eliminates downstream obstructive events but does not directly train the chemoreceptors. CPAP stabilizes the airway; it does not expose the chemoreflex to the progressive, controlled CO₂/H+ loading that behavioral training or pharmacological acidosis would provide. Whether direct chemoreflex conditioning through sustained CO₂ exposure can achieve what passive event elimination cannot is the central empirical question this paper raises. Notably, Tirzepatide independently reduced loop gain through non-respiratory pathways (potentially anti-inflammatory or central neural modulation), demonstrating that loop gain is not entirely immutable even if CPAP alone does not normalize it. The question is not whether loop gain is inherent in the absence of targeted intervention, but whether it is modifiable in the presence of one.

Evidence for Chemoreflex Trainability

Bernardi et al. (2000, The Lancet) demonstrated that long-term yoga practitioners exhibit a generalized reduction in baseline chemoreflex sensitivity, distinct from acute effects. Dereli et al. (2019, Frontiers in Neuroscience) demonstrated that sustained hypercapnic exposure upregulates preprogalanin expression in Phox2b-positive RTN neurons, providing a candidate molecular mechanism for durable chemoreceptor adaptation.

Pharmacological and Device-Based Respiratory Stabilization

Multiple intervention modalities that stabilize respiratory control produce consistent improvements in sleep architecture. Acetazolamide reduces loop gain by approximately 40%; Eckert and Edwards confirmed that 500 mg twice daily reduced median loop gain from 3.4 to 2.0 (p < 0.05), producing a 51% reduction in supine NREM AHI. The Ni, Yang, and Thomas (2021) meta-analysis found TST increased by 25.74 minutes, consistent with initial consolidation and debt repayment rather than immediate compression. The conventional view is that acetazolamide's effects do not persist after withdrawal, but limited evidence complicates this: Verbraecken et al. (1998) reported that central apnea index remained suppressed at six months post-discontinuation in non-hypercapnic patients (CAI 25 at baseline, 4 on drug, 3 at six months off drug), a duration far exceeding the drug's pharmacokinetic half-life and the timeframe for renal bicarbonate normalization. Whether this persistence reflects durable neuroplastic adaptation or other mechanisms remains unresolved, but it suggests the distinction between pharmacological buffering and structural recalibration may not be as clean as the conventional view implies.

The distinction between plant gain and controller gain is important here. Acetazolamide's conventional mechanism is CO₂ reserve widening through metabolic acidosis: the drug shifts the eupneic PaCO₂ down while shifting the apneic threshold down even further, repositioning the system relative to its stability boundary. This is commonly classified as plant gain modification, though the effect is more precisely a change in the operating point than a change in the lungs' intrinsic gas exchange efficiency. Whether acetazolamide also produces durable controller gain modification through brain extracellular fluid acidosis and downstream neuroplastic adaptation remains an open question with suggestive but not definitive evidence. Behavioral CO₂ tolerance training, by contrast, targets controller gain directly: the intervention conditions the chemoreceptors to tolerate higher CO₂/H+ levels before firing, as measured by the progressive increase in breath-hold tolerance that Messineo et al. validated as a predictor of loop gain during NREM. Both reduce overall loop gain, but through different mechanistic pathways with different durability profiles. The carotid body accounts for 35 to 40% of eupneic ventilatory drive (Dempsey), and bidirectional chemoreceptor plasticity has been confirmed in both directions: sensitization through chronic intermittent hypoxia and desensitization through sustained hypercapnic exposure.

Exogenous 2.5% CO₂ (Liang, He et al.) produced 91.7% sleep efficiency and significant N3 prolongation. Enhanced Expiratory Rebreathing Space (EERS) improves sleep efficiency from approximately 71% to 75%.

No published study has simultaneously measured CO₂ tolerance and objective sleep architecture (PSG staging, CPC spectrograms) in the same cohort. This represents a critical gap and a direct opportunity for prospective investigation.

Training Dose, Timing, and Dual-Timescale Effects

CO₂ tolerance training should not be treated as a uniformly beneficial stimulus at all times. Intense breath-hold work may produce a chronic adaptive effect (durable chemoreflex desensitization) while also imposing an acute arousal load (transient sympathetic and orexinergic activation). In the generating N-of-1 case, late-day breath-hold sessions appeared to facilitate sleep onset through a transient parasympathetic rebound but later produced early-night instability, consistent with a delayed arousal cascade. Moving intense training to the morning produced additional sleep consolidation. The hypothesis therefore predicts a timing effect: morning CO₂ tolerance training should improve stable NREM throughput over time, while intense evening training may acutely worsen HFC/LFC balance despite improving sleep-onset latency.

N-of-1 Generating Observation

The hypothesis presented in this paper was generated in part by a detailed N-of-1 case observation. A middle-aged male with treated obstructive sleep apnea (AHI under 2 on CPAP) and years of severe residual fragmentation conducted 11 months of progressive CO₂ tolerance training. Buteyko Control Pause increased from approximately 18 seconds to over 40 seconds. Awakenings decreased from 12 to 15 per night to 1 to 3. Subjective sleep need decreased from over 9 hours to approximately 5.5 to 6 hours with improved daytime function. Consumer wearable staging showed 99 minutes of deep sleep (35.6% of 4 hours 38 minutes total) with deep sleep concentrated before midnight. The highlighted wearable night was disrupted by monitoring awareness; the subject was unable to return to sleep for a final cycle. This night illustrates architectural density rather than a claim that sub-five-hour sleep is the stable target. A comparison night from the disease state (February 2025) on the same device showed severely fragmented architecture across 8+ hours. The major planned intervention that changed was breath-hold training; obvious confounders were minimized but cannot be excluded. CPAP was used throughout both periods with AHI under 2.

The N-of-1 chronology contained separable phases that strengthen the causal inference. Removal of chemical sleep aids and stimulation reduction produced modest improvement but did not resolve fragmentation. Progressive CO₂ tolerance training then produced the major architectural shift on a timeline consistent with chemoreflex adaptation. Later timing optimization, moving intense breath-hold work away from the sleep window, produced additional separable gains. This phase dissociation does not prove causality, but it reduces the probability that the observed changes were attributable to a single confound.

This case is presented as a generating signal. Consumer wearable "deep sleep" may covary with autonomic stability rather than strictly reflecting EEG-defined N3 (Thomas, personal communication, 2026). Validation through formal polysomnography with concurrent CPC analysis, CO₂ monitoring, and objective daytime performance testing is required.

Addressing the Counterargument: Short Sleep and Adverse Outcomes

The strongest counterargument is the robust epidemiological evidence linking objective short sleep (under 6 hours) to adverse outcomes. The Penn State Cohort studies demonstrated that objective short sleep with insomnia symptoms is a marker of biological hyperarousal and is associated with significantly increased mortality risk (OR = 4.0 over 14 years in men). This evidence must be taken seriously.

However, this paper does not argue that short sleep is universally safe. It argues that the critical variable is the autonomic and respiratory state underlying the short sleep. Short sleep driven by sympathetic hyperarousal is pathological. Short sleep driven by highly efficient, HFC-dominant stable NREM may represent a fundamentally different physiological state. Existing epidemiological cohorts do not stratify short sleepers by CPC phenotype, and the safety profile of the "high-HFC, short-duration" phenotype has not been investigated. This paper does not recommend sleep restriction; it identifies architectural efficiency as a variable warranting investigation.

Additional constraints on the compression hypothesis must be acknowledged. The circadian pacemaker (Process C) independently constrains sleep duration regardless of homeostatic clearance; even if Process S clears within 4 to 5 hours, the circadian system continues to broadcast sleep-promoting signals for several additional hours, and the compression hypothesis applies most plausibly to the homeostatic component of sleep need. Compressing total sleep also disproportionately truncates late-night REM sleep, which serves indispensable functions in emotional regulation and affective memory consolidation; this paper does not argue that REM can be safely curtailed. Long-term CPAP adherence data presents a further challenge: CPAP dramatically increases HFC but median nightly use stabilizes at approximately 6 hours without evidence of progressive sleep compression, possibly because post-CPAP HFC remains below healthy-control levels or because non-respiratory factors also contribute. Finally, high HFC and short sleep tolerance may both be downstream manifestations of a resilient biology rather than HFC causing efficient restoration; the proposed experimental design (section “Testable Predictions and Experimental Design”) addresses this through within-subject pre/post intervention testing.

A further challenge comes from Process S kinetics. Dijk and colleagues demonstrated that habitual short sleepers and habitual long sleepers have identical time constants for both Process S accumulation and dissipation; short sleepers do not clear homeostatic pressure faster. Instead, they tolerate a higher residual sleep pressure, as evidenced by elevated waking theta activity. When given extended sleep opportunity, habitual short sleepers slept 8.2 hours, confirming that they carry differential sleep debt they simply tolerate during normal life. Dijk has cautioned that because the decay time constant of Process S (approximately 2 to 3 hours) is biologically conserved, truncating sleep inevitably means living at a higher homeostatic asymptote, and there are no truly healthy methods of permanent sleep compression. This paper does not claim that respiratory training accelerates Process S kinetics. It claims that in individuals with high respiratory-control instability, a substantial fraction of NREM time is spent in sub-threshold cycling and autonomic instability rather than in the high-amplitude SWA expression that Process S clearance requires. Stabilizing respiratory control is hypothesized to increase the proportion of NREM time that is effective for clearance, operating within the same fixed kinetics rather than altering them. The distinction is between accelerating the engine and removing the drag.

Respiratory stabilization may have different effects depending on recovery phase. In an acute consolidation phase, total sleep time may increase as sleep debt is repaid. In a later adaptation phase, stable NREM density may rise. Only after debt repayment and architectural stabilization would spontaneous compression be expected. Therefore, increased TST in acute acetazolamide or CPAP studies does not by itself refute the hypothesis; it may reflect an earlier phase of recovery. However, this staged-recovery model remains speculative and requires longitudinal CPC data tracking stable NREM density and preferred sleep duration across months of sustained stabilization.

Testable Predictions and Experimental Design

Prediction 1: Individuals with higher CO₂ tolerance will demonstrate a higher proportion of HFC during NREM sleep, independent of total sleep time, fitness, and anxiety.

Prediction 2: Interventions that reduce loop gain will increase HFC proportion during NREM sleep, measurable by CPC spectrograms.

Prediction 3: Increased stable NREM proportion will correlate with improved next-day function on objective measures independent of total sleep time.

Prediction 4: After chronic respiratory stabilization and sleep debt repayment, preferred sleep duration will decrease in individuals with initially high loop gain without accumulation of new sleep debt.

Prediction 5: N2 spindle density will be higher in stable (HFC) N2 epochs compared to unstable (LFC) N2 epochs.

Prediction 6: Waking EEG theta activity (4 to 8 Hz), an established marker of residual homeostatic sleep pressure (Dijk and colleagues), will decrease after respiratory stabilization, reflecting more complete Process S clearance within the sleep period. This prediction directly tests the "removing drag" mechanism: if respiratory stabilization increases the proportion of NREM time spent in effective SWA expression, more homeostatic pressure should be discharged per night, producing lower residual pressure upon waking.

Study Design

A prospective N-of-1 series or small-cohort study (target N = 20 to 30 participants with treated OSA and residual sleep complaints despite AHI under 5) is proposed, with measurements at baseline, 4 weeks, and 12 weeks of structured CO₂ tolerance training.

Primary Instrumentation

SleepImage or equivalent FDA-cleared CPC software for nightly spectrograms. Full in-laboratory polysomnography at baseline and 12 weeks. End-tidal or transcutaneous CO₂ monitoring. Formal hypercapnic ventilatory response (HCVR) testing.

Loop gain will be quantified using the PUP/PUPpy computational methodology (Sands et al.), which estimates LG₁ (loop gain at 1 cycle per minute) from standard polysomnographic signals using a modified Mackey-Glass differential equation within 7-minute overlapping windows. PUPpy has been validated against the gold-standard CPAP-drop method (r = 0.63) and demonstrates moderate-to-good night-to-night reproducibility (ICC = 0.61). The minimal detectable difference (MDD) for LG₁ on a single-night PSG is 0.17; averaging two consecutive PSG recordings reduces the MDD by approximately 20 to 30%, and this two-night averaging protocol is recommended for the baseline and 12-week assessments to maximize measurement precision.

Primary Endpoints

Change in HFC percentage of total sleep time. Change in absolute HFC minutes per night. Change in HCVR slope. Change in LG₁. Co-primary endpoint: objective daytime function assessed by 10-minute Psychomotor Vigilance Test (PVT) lapses. A forced sleep extension opportunity at 12 weeks (time in bed extended to 9 hours for 3 consecutive nights) could determine whether reduced habitual sleep duration reflects genuine reduced need by testing whether subjects spontaneously wake early when given the opportunity. Statistical approach: multilevel modeling of nightly HFC versus total sleep duration with random intercepts for individuals.

Responder Phenotype

The hypothesis is most applicable to individuals with evidence of respiratory-control instability rather than global sleep insufficiency. The published threshold for "high loop gain" phenotyping is LG₁ > 0.55 (Sands et al.), and this cutoff should define the primary responder subgroup for pre-specified analysis. Additional inclusion criteria: low breath-hold tolerance, unstable NREM on CPC, residual fragmentation despite controlled obstruction, and preserved oxygenation. The hypothesis should not be generalized to baseline hypercapnia, hypoventilation syndromes, uncontrolled obstruction, COPD, or sedative-driven hypoventilation in which CO₂ retention may represent pathology rather than trainable reserve.

Limitations

The hypothesis is generated from mechanistic reasoning, literature synthesis, and an N-of-1 observation, not a controlled prospective study. The N-of-1 relies on consumer wearable staging rather than clinical PSG. Multiple confounders cannot be fully excluded. The pharmacological evidence shows increased TST and improved efficiency, not reduced TST; the extrapolation to eventual compression is plausible but unproven. N3 duration varies substantially with age, sex, and individual traits, complicating claims about fixed requirements. Consumer wearable metrics, PSG staging, and CPC stability represent three distinct measurement domains that covary but are not interchangeable.

The inherent-trait challenge must be addressed directly. Sands demonstrated that six weeks of optimal CPAP did not normalize elevated daytime loop gain, concluding that high loop gain is largely inherent rather than acquired from the disease process. If loop gain is a fixed trait, then the modifiability premise of this paper fails. However, the intervention tested (CPAP) eliminates downstream obstructive events without directly conditioning the chemoreflex. Behavioral CO₂ tolerance training specifically targets controller gain through progressive, sustained CO₂/H+ exposure, a fundamentally different mechanism than passive event elimination. Whether this distinction is clinically meaningful is the central empirical question the proposed study is designed to answer. Notably, recent evidence that Tirzepatide independently reduces loop gain through non-respiratory pathways (SURMOUNT-OSA; Sands et al.) demonstrates that loop gain is not entirely immutable, even if CPAP alone does not normalize it.

The generalizability of the hypothesis across demographic and clinical subgroups remains uncertain; age-related declines in HFC, sex differences in chemoreflex sensitivity, interactions with OSA severity and insomnia phenotype, and the effects of common medications on CPC profiles may all modulate the proposed mechanism. The N-of-1 involves a middle-aged male with treated OSA; whether the mechanism operates similarly in women, older adults, or individuals without sleep-disordered breathing is unknown.

Clinical Implications

If validated, the stable NREM throughput hypothesis suggests that sleep quality assessment may benefit from incorporating CPC-based stability metrics alongside traditional staging. Respiratory-control stability may warrant evaluation as a modifiable contributor to sleep quality in patients with unrefreshing sleep despite adequate duration and controlled obstruction. Sleep duration recommendations may eventually require individualization based on architectural efficiency. The intersection of CO₂ tolerance training and objectively measured sleep architecture represents a fertile and unexplored frontier for clinical research.

References

Original bibliography wording is preserved. Linked source records identify verified bibliographic corrections; searches are labeled as searches.

1. Bernardi L, Spicuzza L, et al. Slow breathing reduces chemoreflex sensitivity. The Lancet. 2000.
   Search PubMed by titleSource record & notes
   Bibliographic note: Original citation not exactly verified. Related verified sources are linked separately; the main link remains a title search.
2. Cappuccio FP, D'Elia L, Strazzullo P, Miller MA. Sleep duration and all-cause mortality: a systematic review and meta-analysis. Sleep. 2010;33(5):585-592.
   Read sourceAbstractOpen full textSource record & notes
3. Chen CR, Xing L, Ptácek LJ, Fu YH. Salt-inducible kinase 3 (SIK3) mutation identifies a novel gene associated with familial natural short sleep. PNAS. 2025.
   Read sourceAbstractOpen full textSource record & notes
   Bibliographic note: The title and lead author initials in the supplied reference are inaccurate. Verified title is The SIK3-N783Y mutation is associated with the human natural short sleep trait; first authors Chen H, Xing Y, Wan C.
4. Dempsey JA, Smith CA, Blain GM, et al. Role of central/peripheral chemoreceptors and their interdependence in the pathophysiology of sleep apnea. Adv Exp Med Biol. 2012;758:343-349.
   Read sourceAbstractSource record & notes
5. Dempsey JA, Veasey SC, Morgan BJ, O'Donnell CP. Pathophysiology of sleep apnea. Physiol Rev. 2010;90(1):47-112.
   Read sourceAbstractOpen full textSource record & notes
6. Dereli AS, Yaseen Z, Carrive P, Kumar NN. Adaptation of respiratory-related brain regions to long-term hypercapnia. Frontiers in Neuroscience. 2019.
   Read sourceAbstractOpen full textSource record & notes
7. Dijk DJ, Beersma DG, Daan S. EEG power density during nap sleep: reflection of an hourglass measuring the duration of prior wakefulness. J Biol Rhythms. 1987;2(3):207-219.
   Read sourceAbstractSource record & notes
8. Dijk DJ, Czeisler CA. Contribution of the circadian pacemaker and the sleep homeostat to sleep propensity, sleep structure, electroencephalographic slow waves, and sleep spindle activity in humans. J Neurosci. 1995;15(5):3526-3538.
   Read sourceAbstractOpen full textSource record & notes
9. Dijk DJ, Aeschbach D. Short and long sleepers: is the quest for sleep need just a matter of time? Sleep. 1996 (data reported in Aeschbach et al., J Neurosci, 1996;16:6465-6475).
   Search PubMed by titleSource record & notes
   Bibliographic note: Original citation not exactly verified. Related verified source is linked separately; the main link remains a title search.
10. Eckert DJ, White DP, Jordan AS, Malhotra A, Wellman A. Defining phenotypic causes of obstructive sleep apnea: identification of novel therapeutic targets. Am J Respir Crit Care Med. 2013;188(8):996-1004.
    Read sourceAbstractOpen full textSource record & notes
11. Gangwisch JE, et al. Short sleep duration as a risk factor for hypertension. Hypertension. 2006;47(5):833-839.
    Read sourceAbstractSource record & notes
12. He Y, Jones CR, Fujiki N, et al. The transcriptional repressor DEC2 regulates sleep length in mammals. Science. 2009;325(5942):866-870.
    Read sourceAbstractOpen full textSource record & notes
13. Lauderdale DS, Knutson KL, Yan LL, Liu K, Rathouz PJ. Self-reported and measured sleep duration: how similar are they? Epidemiology. 2008;19(6):838-845.
    Read sourceAbstractOpen full textSource record & notes
14. Liang M, He X, et al. Effects of controlled carbon dioxide delivery by a novel device on sleep in healthy subjects. 2025/2026.
    Read sourceAbstractSource record & notes
    Bibliographic note: Verified publication is 2026 and authors begin Liang SF, He BT (not Liang M, He X).
15. Messineo L, Taranto-Montemurro L, Sands SA, et al. Breath-holding as a means to estimate the loop gain contribution to obstructive sleep apnoea. J Physiol. 2018;596(17):4043-4056.
    Read sourceAbstractOpen full textSource record & notes
    Bibliographic note: Pharmacological recalibration reference gives a different title and incorrect volume/pages; verified article is 596(17):4043–4056. Maximal breath-hold testing was evaluated as a marker of loop gain, not as a therapeutic training intervention.
16. Ni YN, Yang Y, Thomas RJ. The role of acetazolamide in sleep apnea at sea level: a systematic review and meta-analysis. Journal of Clinical Sleep Medicine. 2021.
    Read sourceAbstractOpen full textSource record & notes
    Bibliographic note: Verified second author is Yang H, not Yang Y in the throughput bibliography.
17. Ohayon MM, Carskadon MA, Guilleminault C, Vitiello MV. Meta-analysis of quantitative sleep parameters from childhood to old age. Sleep. 2004;27(7):1255-1273.
    Read sourceAbstractSource record & notes
18. Pellegrino R, Kavakli IH, Goel N, et al. A novel BHLHE41 variant is associated with short sleep and resistance to sleep deprivation. Sleep. 2014;37(8):1327-1336.
    Read sourceAbstractOpen full textSource record & notes
19. Picchioni D, Schmidt KC, Loutaev I, et al. Increased rates of brain protein synthesis during [N1,N2] sleep. Journal of Cerebral Blood Flow and Metabolism. 2023;43(1):59-71.
    Read sourceAbstractOpen full textSource record & notes
20. Sands SA, Edwards BA, Terrill PI, et al. Phenotyping pharyngeal pathophysiology using polysomnography in patients with obstructive sleep apnea. Am J Respir Crit Care Med. 2018;197(9):1187-1197.
    Read sourceAbstractOpen full textSource record & notes
21. Sands SA, Owens RL, Malhotra A. New approaches to diagnosing sleep-disordered breathing. Sleep Med Clin. 2022;17(4):603-618.
    Read sourceAbstractOpen full textSource record & notes
    Bibliographic note: The cited title and authors resolve to a 2016 article, Sleep Med Clin 11(2):143–152, not the supplied 2022;17(4):603–618 citation.
22. Shi G, Xing L, Wu D, et al. A rare mutation of beta-1 adrenergic receptor affects sleep/wake behaviors. Neuron. 2019;103(6):1044-1055.
    Read sourceAbstractOpen full textSource record & notes
23. Shi G, Yin C, Fan Z, et al. Mutations in metabotropic glutamate receptor 1 contribute to natural short sleep trait. Current Biology. 2021;31(1):13-24.
    Read sourceAbstractOpen full textSource record & notes
24. Siow TK, Toh CH, Hsu JL, et al. Association of sleep, neuropsychological performance, and gray matter volume with glymphatic function. Neurology. 2022.
    Read sourceAbstractSource record & notes
    Bibliographic note: Verified first author initials are TY (not TK); study used a DTI-ALPS proxy rather than a direct measurement of clearance.
25. Tasali E, Leproult R, Ehrmann DA, Van Cauter E. Slow-wave sleep and the risk of type 2 diabetes. PNAS. 2008;105(3):1044-1049.
    Read sourceAbstractOpen full textSource record & notes
26. Thomas RJ, Mietus JE, Peng CK, Goldberger AL. An electrocardiogram-based technique to assess cardiopulmonary coupling during sleep. Sleep. 2005;28(9):1151-1161.
    Read sourceAbstractSource record & notes
27. Thomas RJ, Mietus JE, Peng CK, Guo D, Gozal D, Montgomery-Downs H, Gottlieb DJ, Wang CY, Goldberger AL. Relationship between delta power and the electrocardiogram-derived cardiopulmonary spectrogram. Sleep Medicine. 2014;15(1):125-131.
    Read sourceAbstractOpen full textSource record & notes
28. Tononi G, Cirelli C. Sleep function and synaptic homeostasis. Sleep Medicine Reviews. 2006;10(1):49-62.
    Read sourceAbstractSource record & notes
29. Vgontzas AN, Fernandez-Mendoza J, et al. Insomnia with objective short sleep duration: the most biologically severe phenotype. Sleep Medicine Reviews. 2013.
    Read sourceAbstractOpen full textSource record & notes
30. Verbraecken J, Willemen M, De Cock W, et al. Central sleep apnea after interrupting long-term acetazolamide therapy. Respir Physiol. 1998;112(1):59-70.
    Read sourceAbstractSource record & notes
31. von Ruesten A, Weikert C, Fietze I, Boeing H. Association of sleep duration with chronic diseases. PLoS ONE. 2012;7(1):e30972.
    Read sourceAbstractOpen full textSource record & notes
32. Watson NF, et al. Recommended amount of sleep for a healthy adult. Sleep. 2015;38(6):843-844.
    Read sourceAbstractOpen full textSource record & notes
33. Xing L, Shi G, Mostovoy Y, et al. Mutant neuropeptide S receptor reduces sleep duration with preserved memory consolidation. Science Translational Medicine. 2019.
    Read sourceAbstractOpen full textSource record & notes
34. Zheng NS, Annis J, Master H, et al. Sleep patterns and risk of chronic disease as measured by long-term monitoring with commercial wearable devices in the All of Us Research Program. Nature Medicine. 2024;30(9):2648-2656.
    Read sourceAbstractOpen full textSource record & notes