Sleep physiology / 03

The airway closes, and four signals find out about it at different times.

Between the back of the nose and the top of the larynx there is no bone and no cartilage — only a muscular tube that has to stay open on muscle tone alone. Sleep turns that tone down. Choose how collapsible the tube is, and watch what happens to the air moving, to the effort spent trying to move it, to the oxygen in the blood, and to the sleep itself, when it gives way.

Why the tube closes at all

Everywhere else the airway is held open by something rigid. The nose has bone; the larynx and the trachea have cartilage rings, drawn at the bottom of the figure. The stretch in between — the pharynx — has neither, and it cannot, because it has to be able to collapse: that is how swallowing and speech work. It is held open instead by around twenty small muscles, the genioglossus chief among them, pulling the tongue and the pharyngeal walls forward against the suction of each breath in.

Awake, this is effortless and invisible. A reflex senses the negative pressure of each inspiration and stiffens the airway a fraction of a second before that pressure peaks. Sleep switches the reflex off. Tone drops at sleep onset and drops again in REM, and from then on the pharynx is defended only by whatever background activity is left. Whether that is enough depends on how much closing pressure the anatomy generates: a set-back jaw, a large tongue, tonsils, fat in the neck, fluid that has shifted out of the legs since the person last stood up. Moving the collapsibility control from mild to severe is raising that closing pressure further past what sleeping tone can resist.

Mild, moderate, severe — what the words mean on the page

The three settings are not three amounts of the same event. They are the three things a pharynx can do, and they look different on the page because they are different. On mild the tube narrows and never closes: every event in the window is a hypopnea, air still moves through a slot too small to be enough, and the oximeter gives up three to eight points each time. On moderate the same airway crosses the line and back — hypopneas and complete apneas in the same four minutes, from one breath to the next, which is exactly what a report showing both looks like. On severe there is no partial version left: the airway shuts, every time, and the arousals stack up behind it.

The bands under the traces say which is which, and the difference is not a matter of degree. A hypopnea is a reduction — thirty percent or more, for ten seconds, and it has to cost something: three points of saturation, or an arousal. An apnea is the airflow gone, ninety percent or more of it, for ten seconds. That is the whole distinction, and it is why the strip is worth reading before the numbers.

Reading the four traces

The recording opens before the sleeper is asleep, on the shaded stretch at the left. That part is the control: regular airflow, regular effort, a saturation that does not move. It is there because none of what follows means anything without it — a desaturation is only a desaturation relative to a number the sleeper was holding a minute earlier. Watch tone come off at the dashed line, the pharynx narrow into snoring, and only then the first event.

Airflow is the trace that says what happened to the breath. It is drawn here in three sizes, because those are the three things the scoring recognises: a normal breath, a breath cut to about a third of normal — a hypopnea — and no breath at all, an apnea. A real recording is messier than this. Breaths shrink and flatten through every size in between, and it takes a technologist years to read at a glance; the model rounds those in-between breaths to the nearest of the three so that the events stand apart from the breathing around them. What does not change is the rule underneath: once a floppy tube has narrowed past a certain point, pulling harder does not buy more air. Set the airway to mild and roll the sleeper onto their side and nothing is left but snoring.

Effort is the trace that decides what the event is called. Through a complete obstructive apnea the chest goes on moving, and moves more: carbon dioxide is accumulating, drive is climbing, and each attempt is bigger than the last. That crescendo against a closed airway is the signature of obstruction. A central apnea looks identical on the flow channel and flat on this one — nothing is blocked, the brain has simply stopped asking. The effort belt is how the two are told apart, and it is the main reason an oximeter alone cannot do it.

Oxygen arrives late and leaves late. There is a lung full of air to get through first, then eight or nine seconds for blood to travel from lung to fingertip, then the oximeter's own averaging window on top. So the desaturation belonging to an event usually reaches its lowest point after that event has ended: the airway is open again, the sleeper is breathing hard, and the number on the screen is still falling. Drag the playhead to the end of an apnea and watch how far the oximeter still has to go.

The EEG is the channel that says what any of it costs. The other three describe a breathing problem; this one is the reason it is a sleep problem. At four minutes to the page you are not reading individual waves — you are reading the thickness of the band. Asleep it is slow and shallow, with the occasional spindle. Then the airway shuts, effort climbs, and for three or four seconds the trace goes fast and dense: the brain has surfaced far enough to put tone back into the pharynx. Compare it to the shaded stretch at the far left, before the sleeper was asleep. It is the same trace. That is the point — every one of those marks is a few seconds of being awake, and there may be four hundred of them in a night, and not one will be remembered in the morning. The complaint that gets someone into clinic is usually this line, not the oxygen one.

The close-up beside the figure shows the same channel the way a technologist reads it, fifteen seconds at a time. Asleep, the trace rolls along in slow waves about a second apart. Then, at the moment marked on the lane below, the slow waves stop and the trace turns into a dense, fast scribble — the same rhythm as the waking stretch at the start of the recording — and three or four seconds later the slow waves resume as if nothing had happened. That is what an arousal is: not waking up, but the brain surfacing far enough to put tone back into the pharynx. Pause the recording just after an event ends and drag slowly through it to watch the change.

The saturation curve does not fall in a straight line either, and that is the oxyhemoglobin dissociation curve doing its work. From 97% down to about 92% the fall is gentle; past 90% it steepens sharply. Two apneas of identical length can cost four points or fourteen depending on where on that curve they begin — part of why the same airway is a more serious problem at altitude, or with lung disease, or in REM.

What ends the event

Nothing on the traces reopens the airway. What reopens it is an arousal: a few seconds of cortical activation that restores dilator tone the way waking would, without the sleeper ever knowing. The airway snaps open, there is a snort, breathing overshoots, carbon dioxide is blown off below where it started — and with the stimulus gone, tone falls and the tube closes again. Left alone this cycles every thirty to fifty seconds, for hours.

That is why the arousal threshold control changes the illness rather than just the numbers. Set it low and the sleeper is rescued early: events are short, oxygen barely moves, the oximetry looks almost reassuring — and there are a great many arousals, so the night is shredded — watch the EEG lane fill up while the oximeter barely moves. That is the person with no dramatic desaturations who is nonetheless exhausted, and whose home oximetry came back reassuring. Set it high and you get the opposite: long events, deep desaturation, an alarming-looking oximetry trace, and comparatively intact sleep. The same airway, two different illnesses, and they do not respond to the same things.

The last two controls behave the way the clinic does. Move from NREM to REM and events lengthen without the anatomy changing at all, because REM withdraws most of the tone the airway was relying on. Roll from the back onto the side and the gravitational load comes off the tongue and the soft palate; on the moderate airway that alone takes a montage carrying apneas down to hypopneas, and on the mild one it empties the page, which is the entire premise of positional therapy. Set the state to Awake and everything stops — which is the part people find hardest to believe about their own airway. It is perfectly competent all day.

Putting the mask on

Turn on CPAP and watch what changes and what does not. The anatomy is the same. The muscle tone is the same — as low as it ever was in sleep, lower still in REM. What has changed is the pressure inside the tube. Whether a floppy tube stays open is a contest between the pressure pushing in on it from the tissues around it and the pressure inside it holding it out, and a mask that blows air in under a few centimetres of water simply wins that contest from the inside. It is a splint made of air. Nothing is being treated; the airway is being held open for the night and let go in the morning, which is why the machine has to be worn every night and why the apnea is back the first night it is not.

Then watch the four traces. The airflow recovers its full, rounded shape. The effort trace settles back to matching it breath for breath, because the chest is no longer pulling against a closed door. The oximeter goes flat. And the EEG — the one that mattered — loses its arousals: in the close-up the slow waves just keep rolling, uninterrupted, which is the thing the sleeper has not had in years. Now turn the pressure down. At too low a setting the events do not vanish but shrink: apneas become hypopneas, and hypopneas become snoring. Switch to REM and a pressure that held in NREM will not hold any more. That is why a titration night exists, and why the number on the machine is set high enough for the worst stretch of the night rather than the average of it.

A model, not a recording. The oscillation is not scripted: it emerges from a negative-feedback loop with a transport delay in it, in which collapse reduces ventilation, carbon dioxide rises, chemical drive raises both dilator tone and respiratory effort, an arousal fires once effort crosses the threshold, the airway opens, ventilation overshoots, and drive falls back below what the airway needed. Two elements are scripted. Sleep onset: dilator tone is held at its waking value for the opening seconds, then released the way it subsides through N1, which is what puts a stretch of ordinary breathing at the left of every setting. And a slow fluctuation in tone across about seventy seconds, standing in for the drift in sleep depth that a night actually has — without it every cycle is identical to the last, and a mild airway either obstructs on every one of them or on none, when what mild disease does is cross its threshold and come back. Arousal answers to chemical drive together with the effort of breathing against an inadequate airway, and the second route is not blunted in REM the way the first one is; both routes end hypopneas as well as apneas, which is why the mild setting has arousals in it at all. Patency follows the Starling-resistor account of Gold and Schwartz, with a critical closing pressure set by the collapsibility control and dilator tone set by state. Wall adhesion scales with how hard the pharynx is being squeezed, so a marginal airway settles part-open while a crowded one shuts outright — the difference between a hypopnea and an apnea, in one term. Flow through the narrowed tube is rounded to one of three levels — unrestricted, about thirty percent of a normal breath, or nothing — rather than following the continuous Starling ceiling a real pharynx obeys. The continuous version was more faithful, and it produced airflow traces that only a technologist could read, with breaths tapering through every size between normal and absent; this exhibit exists to make the events legible to the person who has them. Arousal from an obstructed breath fires on a rising crossing of the chemical threshold, or once effort against the obstruction has accumulated for a delay the arousal-threshold control sets. Ventilation follows inspired volume alone, since nothing can be exhaled that was not first inhaled. The four controls correspond loosely to the trait framework of Eckert and colleagues (2013), in which anatomy, muscle responsiveness, arousal threshold and loop gain vary independently between people carrying the same apnea–hypopnea index. Oxygen is carried through a single alveolar compartment and a Hill dissociation curve with P₅₀ 26.6 mmHg and n 2.8, then delayed eight seconds for circulation and smoothed to mimic an oximeter's averaging. Events are scored against a moving flow envelope on the AASM criteria, including the requirement a scorer actually applies to a hypopnea — a three percent desaturation or an arousal, without which it is not scored at all. CPAP enters the loop in one place: the mask pressure is subtracted from the critical closing pressure, since collapse is decided by the pressure across the wall of the tube and positive pressure inside it is worth exactly what it applies. One centimetre of water is worth 0.033 of a tone unit, calibrated so that, on the back at the default arousal threshold, the mild airway clears at 4 cm, the moderate at 5 and the severe at 9 in NREM, and each needs a few centimetres more in REM. The EEG channel is illustrative: it is generated from the arousals the model has already produced, and it does not feed back into anything. The close-up reads the same generated signal at full rate; in NREM its background is a slow rhythm at about one a second with spindles and K-complexes on a fixed schedule, and an arousal replaces it with the waking rhythm for three and a half seconds. No one would read an index off four minutes of recording, so the readout counts what is on the page rather than extrapolating from it; the events-per-hour band belongs to the severity you selected, and the window is a run of the kind that severity produces rather than an average of a night. The numbers are illustrative and internally consistent rather than calibrated against any patient population, and the anatomy is schematic and not to scale. Nothing here diagnoses anything.