Daniel sent us a question that gets to something weird about how we talk about sleep. Two people both have insomnia. One lies in bed for two hours before finally drifting off. The other is out cold by ten thirty but wide awake at three in the morning staring at the ceiling. Completely different biology. The FDA just approved a new dual orexin receptor antagonist in March specifically targeting sleep maintenance, and the treatment pipeline is splitting in ways most clinicians haven't absorbed yet. So where do we even start?
Let's start with the labels, because the labels themselves tell a story about how medicine has been confused about this. Sleep-onset insomnia and sleep-maintenance insomnia. SOI and SMI in the literature. The DSM, the big psychiatric diagnostic manual, lumps them together under insomnia disorder and has done so for years. But here's the thing that drives me a little crazy. The ICD-ten, the international classification of diseases, actually codes them separately. G forty seven point zero zero versus G forty seven point zero one. So you have two different coding systems from the same global health apparatus disagreeing on whether these are the same thing. And that coding discrepancy matters because it shapes what gets reimbursed by insurance and what gets studied in clinical trials.
The billing codes know they're different but the diagnosis manual pretends they aren't. That feels about right for modern medicine.
That's exactly the problem. And the prevalence numbers make the distinction even more important. About thirty percent of adults report some kind of insomnia symptoms, but the distribution is not even. Sleep-onset insomnia peaks in younger adults, roughly twenty to thirty five years old. Sleep-maintenance insomnia dominates in older adults. Once you're over fifty, the prevalence of SMI alone hits somewhere between twenty five and thirty percent. And roughly forty percent of chronic insomniacs have a mixed presentation, both onset and maintenance problems.
If you're twenty five and can't fall asleep, you're in one bucket. If you're fifty five and wake up at three in the morning every night, you're in another. And the biology underneath those two things is fundamentally different.
And this is where the field is finally catching up. For decades we treated insomnia as a monolith. You couldn't sleep, we gave you a sleeping pill. End of story. But the neurobiology of sleep onset versus sleep maintenance involves different circuits, different neurotransmitters, different patterns of brain activity. It is not one problem with one solution.
Alright, so walk me through the biology. What's actually happening in the brain of someone who can't fall asleep versus someone who can't stay asleep?
Let's start with sleep-onset insomnia. The dominant model is hyperarousal, and I want to be careful with that word because it sounds psychological. It's not. It's neurobiological. We can measure it. In someone with SOI, cortisol levels are elevated in the evening hours when they should be dropping. If you put electrodes on their scalp and measure EEG activity thirty minutes before their intended bedtime, you see elevated power in the beta and gamma frequency bands. Beta is roughly thirteen to thirty hertz, gamma is above thirty hertz. Those are frequencies associated with active cognition, with attention, with alertness. The brain should be winding down into alpha and theta rhythms as sleep approaches. In SOI patients, it's not. The engine is still revving.
The racing mind isn't a metaphor. There's literal electrical racing happening.
And functional MRI studies show where that racing is happening. In SOI patients, during the pre-sleep period you see persistent hyperconnectivity in the dorsolateral prefrontal cortex and the anterior cingulate. The dorsolateral prefrontal cortex is the seat of executive function, planning, worry, rumination. The anterior cingulate is involved in error detection and conflict monitoring. So you have the planning center and the what's-wrong center lit up and talking to each other while you're trying to sleep. That's a terrible combination.
This is specific to sleep onset. If you look at someone with sleep-maintenance insomnia, their pre-sleep EEG looks normal?
There was a study by Buysse and colleagues in twenty twenty three that used quantitative EEG to compare the two groups. The SOI patients had elevated beta power thirty minutes before bedtime. The SMI patients had perfectly normal pre-sleep EEG. They fell asleep fine. But then after three to four hours, they showed abrupt transitions from NREM sleep to wakefulness. No gradual lightening, no slow drift upward. Just pop, awake. Something is destabilizing their sleep maintenance specifically in the second half of the night.
If SOI is too much wake signal at bedtime, what's broken in SMI?
Two things are going wrong, and they interlock. The first is circadian misalignment. In SMI patients, the circadian signal for sleep maintenance arrives too early relative to their desired wake time. Specifically, the melatonin offset and the core body temperature nadir happen hours earlier than they should. There was a twenty twenty four meta-analysis from the European Sleep Research Society that found SMI patients have a dim light melatonin onset, the DLMO, that is on average two point three hours earlier than matched controls relative to their desired bedtime. So their internal clock is essentially saying it's morning at three AM.
Two point three hours earlier. So their body is literally trying to start the day in the middle of the night.
That's the circadian piece. The second piece is homeostatic sleep drive failure. Sleep pressure builds up during the day through the accumulation of adenosine in the brain. Adenosine is the molecule that makes you feel sleepy. It binds to adenosine A one and A two A receptors, and the longer you're awake, the more adenosine accumulates. During sleep, it gets cleared out. But in aging brains, and remember SMI is much more common in older adults, adenosine receptors get downregulated. You literally have fewer receptors available to detect the sleep pressure signal. So even if adenosine is accumulating normally, the brain isn't reading the signal as strongly. The result is a weaker drive to stay asleep, especially in the second half of the night when the homeostatic drive is naturally declining anyway.
You've got a circadian clock that thinks it's morning too early, and a sleep pressure system that's not generating enough pressure to override that.
The orexin system ties this all together. Orexin, also called hypocretin, is a neuropeptide produced in the lateral hypothalamus. It promotes wakefulness and suppresses REM sleep. In a healthy sleeper, orexin neuron firing drops to near zero during sleep, especially during the second half of the night. But here's the critical difference. In SOI, the orexin system is hyperactive at bedtime. Too much wake signal right when you're trying to transition to sleep. In SMI, orexin activity rebounds prematurely in the second half of the night. The system should stay quiet until morning, but instead it reactivates around three or four AM.
Same molecule, same system, but the timing of the dysfunction is completely different. That's elegant.
It's elegant and it has direct treatment implications. Which brings us to the neurotransmitter level, and this is where things get pharmacologically specific. Both SOI and SMI involve reduced GABAergic tone, but the receptor subunits affected are different. GABA-A receptors are pentameric, they're made up of five subunits, and different combinations of subunits produce receptors with different properties and different locations in the brain. SOI shows more dysfunction in alpha-one subunit-containing receptors. Alpha-one is heavily expressed in the cortex and is involved in the initiation of sleep. SMI shows more issues with alpha-two and alpha-three subunits, which are more involved in maintaining deep sleep and preventing fragmentation.
This explains why different sleeping pills work for different people.
Zolpidem, Ambien, is highly selective for alpha-one-containing GABA-A receptors. It's great at initiating sleep but has a half-life of about two and a half hours. So it's basically tailor-made for sleep-onset insomnia and nearly useless for sleep maintenance. Eszopiclone, Lunesta, has broader subunit affinity, alpha-one, alpha-two, alpha-three, and a longer half-life of about six hours. It shows better outcomes for SMI. The pharmacology maps directly onto the subunit biology.
The drug that works for one person is not just less effective for the other. It's targeting the wrong mechanism entirely.
And this is why treating insomnia as one condition has been such a clinical failure. Giving zolpidem to an SMI patient is like prescribing a medication that only works for the first two and a half hours of sleep and then leaves them completely unprotected for the remaining five hours. No wonder they still wake up.
That's the biology of the split. But how does that translate into what your doctor actually prescribes? Because I suspect most people are still just getting handed a Z-drug prescription and sent on their way.
The clinical reality is pretty stark. The gold standard treatment for insomnia is cognitive behavioral therapy for insomnia, CBT-I. It is effective for both SOI and SMI, but it works through different mechanisms for each. For SOI, the heavy hitters are stimulus control and sleep restriction. Stimulus control means you break the conditioned association between the bed and wakefulness. You only go to bed when sleepy, you get out of bed if you can't sleep within about twenty minutes, you use the bed only for sleep, no phone, no worrying. Sleep restriction temporarily limits time in bed to consolidate the homeostatic drive. Both of these directly target the conditioned hyperarousal that defines SOI.
For SMI, the most effective components are different. Cognitive restructuring around middle-of-the-night anxiety becomes central. People with SMI often develop catastrophic thinking about waking up. I'm going to be exhausted tomorrow, this is ruining my health, I'll never sleep through the night again. That anxiety further activates the arousal system and makes falling back asleep harder. So you target the catastrophic cognitions. Also, scheduled awakenings can be effective. Counterintuitively, you have the patient set an alarm to wake up briefly, then go back to sleep. This can reduce the spontaneous awakenings by taking control of the timing.
Like adopting a feral cat.
I'm not sure I follow the cat analogy, but yes, you're essentially domesticating the awakening. The problem is access. There are only about five hundred behavioral sleep medicine specialists in the entire United States. For a condition that affects roughly thirty percent of adults. Most people who need CBT-I cannot get it from a properly trained provider. There are digital CBT-I programs now, apps like Sleepio and Somryst, which have decent evidence and are scalable, but they don't do the phenotype-specific tailoring that a trained specialist would do.
Five hundred specialists for tens of millions of patients. That's not a gap, that's a canyon. So what about the medication side? What's the current state of play?
The pharmacopeia is bifurcating, but most prescribing hasn't caught up. Let me lay out the current landscape. For SOI, the Z-drugs, zolpidem and zaleplon, are the most commonly prescribed. Short half-life, alpha-one selective, good for sleep initiation, minimal next-day sedation. Zaleplon has an ultra-short half-life of about one hour, so it can even be used for middle-of-the-night dosing if there are at least four hours left before wake time. Benzodiazepines like temazepam work for both onset and maintenance but carry clear tolerance and dependence risks. They're not first-line anymore.
For SMI specifically?
This is where the orexin antagonists have changed the game. The DORAs, dual orexin receptor antagonists. Suvorexant, brand name Belsomra, was the first approved in twenty fourteen. It blocks both orexin one and orexin two receptors, suppressing the wakefulness signal. Its half-life is about twelve hours, which is great for sleep maintenance but can cause next-day sedation. Daridorexant, brand name Quviviq, was approved more recently with a half-life of about eight hours, which is closer to the sweet spot for SMI. Enough duration to protect the second half of the night, short enough to mostly clear by morning.
Then the new approval in March of this year.
The FDA approved a new DORA, and the details suggest it's been optimized specifically for sleep maintenance with an improved half-life profile and possibly better orexin two receptor selectivity. The orexin two receptor in particular seems to be more important for sleep-wake transitions, so targeting it more selectively could maintain efficacy while reducing some of the orexin one mediated side effects like next-day drowsiness. This is the first insomnia drug class that was designed from the ground up for sleep maintenance rather than sleep initiation.
The Z-drugs were built for onset, and they've been used off-label for maintenance because there wasn't anything better. Now we're getting drugs built for maintenance.
The pipeline takes this separation even further. Let me walk through what's coming, because it's genuinely exciting. First, orexin receptor inverse agonists are entering Phase Three trials. The current DORAs are antagonists, they block orexin from binding. Inverse agonists go a step further, they bind to the receptor and actively reduce its baseline activity below its natural resting state. For SOI, where the orexin system is hyperactive at bedtime, an inverse agonist could more precisely tune down that excess activity without completely suppressing the system. Think of it as a dimmer switch rather than an on-off switch.
A dimmer switch versus a light switch. That's a meaningful difference if you want to fall asleep without feeling like you've been chemically clubbed.
Second area, GABA-A positive allosteric modulators targeting alpha-two and alpha-three subunits specifically. There are compounds like L-eight three eight four one seven analogs that avoid the alpha-one subunit entirely. Why does that matter? Alpha-one is associated with the amnestic effects, the dependence liability, and the next-day cognitive impairment of zolpidem. By targeting only alpha-two and alpha-three, these new compounds could promote sleep depth and maintenance without the baggage. For SMI patients, this could be a game-changer.
You get the sleep maintenance benefit without the we made you forget the last three hours side effect.
Third, there's a new melatonin receptor agonist beyond ramelteon. Ramelteon was the first MT one MT two agonist approved for insomnia, but it's really only effective for sleep onset. The successor compound, building on the tasimelteon profile, is in Phase Two B specifically for SMI with a circadian component. It targets both melatonin receptor subtypes more potently and has a longer duration of action. The idea is to shift and stabilize the circadian phase so the sleep maintenance signal arrives at the right time.
The fourth area you mentioned earlier, adenosine A two A receptor agonists?
This is the most conceptually elegant approach for SMI in aging populations. Remember, SMI is associated with adenosine receptor downregulation. The brain isn't reading its own sleep pressure signal. Caffeine works by blocking adenosine receptors. An adenosine A two A agonist would do the opposite, it would directly stimulate those receptors, essentially mimicking the natural sleep pressure signal. It's like giving the brain back the volume knob for sleep drive. This is still in early preclinical work, but the logic is compelling.
You're not sedating the brain. You're boosting its own sleep pressure machinery.
That's the dream, pardon the pun. And there's also non-pharmacologic stuff in the pipeline that's worth paying attention to. Transcutaneous auricular vagus nerve stimulation, taVNS. This is a non-invasive approach where you stimulate a branch of the vagus nerve through the skin of the outer ear. A twenty twenty-five pilot study showed a forty percent reduction in wake after sleep onset, WASO, in SMI patients. That's comparable to some pharmacologic interventions, and the side effect profile is essentially nil.
How does stimulating the vagus nerve through your ear improve sleep maintenance?
The mechanism isn't fully worked out, but the vagus nerve projects to the nucleus of the solitary tract in the brainstem, which connects to the locus coeruleus and the raphe nuclei, which are key nodes in the arousal and sleep regulation networks. Stimulating the vagus appears to shift autonomic balance toward parasympathetic dominance, reduce sympathetic tone, and dampen the hyperarousal that fragments sleep. It's basically a remote control for the autonomic nervous system.
You're calming down the fight-or-flight system through a nerve in your ear. That's wonderfully weird.
Another non-pharmacologic approach, phase-locked auditory stimulation during slow-wave sleep. This is where you play very quiet tones timed precisely to the up-phase of slow oscillations in deep sleep. The auditory stimulation enhances the amplitude of those slow waves, which deepens sleep and makes it more resistant to fragmentation. A twenty twenty-four trial showed a twenty two percent improvement in sleep maintenance efficiency using this approach. You wear a headband that reads your EEG and delivers the tones at exactly the right moment.
Your sleep is being DJed by an algorithm. Herman Poppleberry's side hustle goes clinical.
I would be excellent at this. But yes, the principle is that you're reinforcing the brain's own slow-wave activity rather than imposing sleep chemically from the outside. It's a fundamentally different treatment philosophy.
All of this is fascinating at the bench, but what does it mean for someone lying awake at three AM right now? What can people actually use?
Let's start with what doesn't work, because there's a big misconception to address. Over-the-counter melatonin. Standard melatonin has a half-life of about forty five minutes. It is gone from your system in a few hours. It is primarily useful for phase shifting, not for maintaining sleep. For sleep-onset insomnia, it has marginal benefit if taken one to two hours before bed, and that benefit is mostly for people whose onset problem is driven by a delayed circadian phase. For sleep-maintenance insomnia, standard melatonin is essentially useless. It's cleared by the time the maintenance problem kicks in.
That's going to annoy a lot of people who've been taking melatonin for years.
The evidence is what it is. Now, there is one nuance. Low-dose sustained-release melatonin, something like zero point five milligrams in an extended-release formulation, has some evidence for SMI. The low dose avoids desensitizing the melatonin receptors, and the sustained release provides coverage into the second half of the night. But it's not a strong effect. The real behavioral intervention for SMI is timed light exposure. Morning bright light therapy strengthens the circadian signal and helps anchor the sleep-wake cycle. For someone with SMI whose internal clock is drifting early, morning light essentially tells the brain this is when morning actually starts, not three AM.
The advice is not take a pill. It's get sunlight in your eyes first thing in the morning.
For SMI specifically, yes. And there's a practical tracking recommendation. If you're not sure which phenotype you have, keep a sleep diary for two weeks that records two numbers. Sleep latency, how long it takes you to fall asleep, and wake after sleep onset, WASO, how many minutes you spend awake after initially falling asleep. If your sleep latency is high and your WASO is low, you're in the SOI camp. If your latency is normal but your WASO is consistently above thirty minutes, you're in the SMI camp. If both are elevated, you're in the mixed group.
That distinction changes what you should ask for?
If you're an SMI patient and your doctor reaches for the zolpidem prescription pad, you should be asking why not a DORA. Daridorexant or the newly approved agent are designed for exactly your problem. If you're an SOI patient, CBT-I with a sleep restriction component is probably going to do more for you than any pill long-term. But you have to know which bucket you're in to have that conversation. Most patients don't. Most doctors don't ask.
There's also a regulatory angle here that you mentioned earlier. The FDA advisory committee in twenty twenty five discussed requiring separate endpoints for SOI and SMI in insomnia trials. What would that actually mean?
It would force drug developers to pick a phenotype. Right now, an insomnia trial can recruit anyone with insomnia disorder and measure global outcomes like total sleep time or the Insomnia Severity Index. That lumps SOI and SMI patients together and muddies the results. If the FDA starts requiring separate sleep onset and sleep maintenance endpoints, then a drug developed for SMI would have to prove it actually improves WASO in an SMI population, not just that it improves sleep generally in a mixed group. That would reshape the entire development pipeline.
Which means drug companies would have to figure out which biology they're targeting before they run the trial, not after.
And that's where the genetics come in. GWAS studies, genome-wide association studies, are showing that SOI and SMI have distinct heritability patterns. SOI is linked to clock gene variants, things like PER three, which is involved in circadian rhythm regulation. SMI is linked to GABA receptor subunit genes, which makes perfect sense given the alpha-two and alpha-three subunit involvement we discussed. These are not the same condition with different timing. They are genetically distinguishable.
Eventually we might see polygenic risk scores guiding treatment. Your genes suggest you're an SMI type, so we start with a DORA rather than a Z-drug.
That's the direction. We're not there yet, but the pieces are falling into place. The biology is clarifying, the pharmacology is bifurcating, the diagnostics are improving. What's lagging is clinical practice. Most primary care physicians are still prescribing the same medications they learned about in medical school, which for many of them predates the orexin antagonist era entirely.
The pipeline is splitting but the prescription pad hasn't caught up. That's a good summary of where we are.
I want to be clear about one thing. None of this means that sleep hygiene doesn't matter. Consistent bedtimes, dark cool bedroom, no caffeine after noon, all of that. But for someone with genuine SMI driven by circadian misalignment and homeostatic failure, sleep hygiene alone is like telling someone with pneumonia to dress warmly. It's not wrong, it's just insufficient.
The distinction between insufficient and wrong is where a lot of medical advice lives. So let me ask the question I think a lot of listeners are sitting with. If I wake up at three AM every night and can't get back to sleep, what's the actual first step I can take tomorrow?
Tomorrow morning, get fifteen to twenty minutes of outdoor light exposure within thirty minutes of waking up. Not through a window, glass filters out some of the relevant wavelengths. Actually go outside. That starts the process of anchoring your circadian phase. Then start tracking. Write down what time you went to bed, roughly how long it took to fall asleep, what time you woke up, and crucially, how many minutes you were awake during the night. Do that for two weeks. If you're seeing a consistent pattern of WASO above thirty minutes, you have data to bring to a conversation with a clinician.
If you can't get to a sleep specialist, which statistically you probably can't given the five hundred figure, what do you ask your primary care doctor?
You say, I've tracked my sleep for two weeks, my main problem is staying asleep, not falling asleep, and I'd like to discuss whether a dual orexin receptor antagonist might be appropriate for my sleep maintenance insomnia. Use those exact words. Sleep maintenance insomnia. It signals that you understand the distinction and you're asking about the drug class designed for your specific problem. Most primary care doctors can prescribe daridorexant or suvorexant. They just don't reach for them first because the Z-drugs are more familiar.
Name the problem correctly, bring data, ask about the right drug class. That's more actionable than most medical advice.
If you want to go the non-medication route, look for a digital CBT-I program. Sleepio has randomized controlled trial data behind it. It's not as good as working with a specialist who can tailor the protocol to your phenotype, but it's far better than nothing and it's accessible. The key is to do the full program, not just cherry-pick the parts that sound easy.
Treatment is bifurcating, the biology is clarifying, and the practical advice for patients depends entirely on which type they have. Let me ask you the forward-looking question. Will the next edition of the DSM finally split insomnia disorder into its constituent phenotypes?
I think it has to. The ICD eleven is already hinting at this with its specifiers. The evidence for distinct neurobiology is too strong to ignore. And the practical pressure from drug development will push it along. If we have drugs that work for one phenotype and not the other, the diagnostic system has to reflect that or we're practicing bad medicine. My guess is the DSM six, whenever that arrives, will at minimum include specifiers for onset versus maintenance, and possibly split them entirely.
If the genetics hold up, we might eventually be screening for insomnia risk before symptoms even appear.
That's the long game. A patient comes in for a routine physical, their polygenic risk score flags elevated SMI risk, and their clinician starts discussing preventive light exposure and sleep scheduling before the three AM awakenings ever start. That would be a real shift from treating insomnia as an acute complaint to managing it as a chronic condition with identifiable risk factors.
From waiting until you're miserable to getting ahead of the biology. That's where all of medicine should be heading.
On that note, I think we've covered the ground. Two insomnias, two biologies, two treatment pipelines, and a diagnostic system still struggling to catch up.
Now: Hilbert's daily fun fact.
Hilbert: In the nineteen sixties, linguists documenting Inuktitut dialects in the Canadian Arctic discovered that a single verb could contain enough polysynthetic morphology to express what would require an entire paragraph in English, including obligatory suffixes that specify whether the speaker witnessed the action themselves or merely heard about it from someone else. Behavioral anomaly: despite this grammatical requirement, speakers routinely violated it when describing events that happened while they were supposed to be hunting, adding a suffix that ambiguously implied direct observation while grammatically hedging against accusation of fabrication.
There's a grammatical mood for I definitely saw this but please don't check my story.
The linguistic equivalent of plausible deniability.
This has been My Weird Prompts. Thanks to our producer Hilbert Flumingtop. If you enjoyed this episode, leave us a review wherever you listen, it helps other people find the show.
Or visit myweirdprompts dot com for every episode and the full archive. We'll be back soon.
