Seltorexant

JNJ-42847922; MIN-202; selective orexin-2 (OX2R) receptor antagonist. NOT a dual orexin antagonist — suvorexant, lemborexant and daridorexant block both receptors

Seltorexant (JNJ-42847922; MIN-202; selective orexin-2 (OX2R) receptor antagonist. NOT a dual orexin antagonist — suvorexant, lemborexant and daridorexant block both receptors) is a sleep research compound. Selective antagonist at the orexin-2 receptor (OX2R). Orexin-A and orexin-B, the two wake-promoting peptides made in the lateral hypothalamus, signal through OX1R and OX2R; orexin-A engages both, orexin-B is OX2R-preferring, and OX2R is the receptor that carries arousal drive into the histaminergic and monoaminergic wake-promoting nuclei. Blocking it withdraws the wake signal instead of adding sedation, which is mechanically different from every GABA-A hypnotic — there is no chloride channel, no benzodiazepine site. Reported selectivity is an approximate 2-log ratio over the human orexin-1 receptor, and that single difference is the whole argument for the molecule: the three approved orexin drugs block both receptors. Target specificity was established genetically rather than by selectivity ratio alone — the compound promoted sleep in wild-type mice and had no effect in OX2R knockout mice. It did not raise dopamine in rat nucleus accumbens and did not produce place preference, in contrast to zolpidem run alongside it.

How Seltorexant works

Selective antagonist at the orexin-2 receptor (OX2R). Orexin-A and orexin-B, the two wake-promoting peptides made in the lateral hypothalamus, signal through OX1R and OX2R; orexin-A engages both, orexin-B is OX2R-preferring, and OX2R is the receptor that carries arousal drive into the histaminergic and monoaminergic wake-promoting nuclei. Blocking it withdraws the wake signal instead of adding sedation, which is mechanically different from every GABA-A hypnotic — there is no chloride channel, no benzodiazepine site. Reported selectivity is an approximate 2-log ratio over the human orexin-1 receptor, and that single difference is the whole argument for the molecule: the three approved orexin drugs block both receptors. Target specificity was established genetically rather than by selectivity ratio alone — the compound promoted sleep in wild-type mice and had no effect in OX2R knockout mice. It did not raise dopamine in rat nucleus accumbens and did not produce place preference, in contrast to zolpidem run alongside it.

Proposed benefits

Researched for insomnia disorder and for depressive symptoms in major depressive disorder with insomnia symptoms, by selectively blocking the orexin-2 receptor that carries wake-promoting drive.

Where to get Seltorexant

The evidence for Seltorexant

Graded by what exists behind each claim.

✅ Clinically validated

• The sleep effect is the best-established thing about this molecule, and it was measured with polysomnography rather than questionnaires. In insomnia disorder without psychiatric comorbidity, 364 participants were randomized across seltorexant 5 mg, 10 mg, 20 mg, placebo and zolpidem 5-10 mg for 14 days. On night 1, latency to persistent sleep improved against placebo at 10 mg (least-squares mean ratio 0.64, 90% CI 0.51-0.81) and 20 mg (0.51, 0.41-0.64), and 20 mg beat zolpidem (0.71, 0.57-0.88). Wake after sleep onset over the first 6 hours moved the same way. By night 13 the 10 mg and 20 mg improvements were maintained while zolpidem's had diminished — which is the more interesting result, because tolerance is the standard objection to a hypnotic.
• The 5 mg arm is the one that matters commercially and it is the one that failed. Across that same trial, the dose-response was significant, and the doses reported as separating from placebo were 10 mg and 20 mg. 5 mg is the strength on sale. A trial that tests three doses and finds the lowest one inert has said something specific about the lowest one.
• An earlier crossover put numbers on the same effect twice over. In insomnia disorder, 40 mg for five days improved sleep efficiency by 5.8% (SD 9.2) after a single dose and 7.9% (SD 9.8) after multiple doses, both p < 0.001, with total sleep time up 27.7 and 37.9 minutes and latency to persistent sleep down 18.8 and 29.9 minutes. In antidepressant-treated patients with persistent insomnia, single doses of 10, 20 and 40 mg all shortened latency to persistent sleep against a placebo least-squares mean of 61.05 minutes, all p < 0.001 — and 40 mg was not better than 20 mg, which is the first sign of the flat top that runs through this whole program.
• The depression program is real, phase 3 met its primary endpoint, and the peer-reviewed record still lags the press release. Johnson & Johnson reported on 29 May 2024 that study MDD3001 — seltorexant 20 mg once daily added to an SSRI or SNRI in adult and elderly patients with MDD with insomnia symptoms — produced a statistically significant improvement in MADRS total score at day 43. That topline was presented at a congress and announced by the sponsor. A primary peer-reviewed publication of MDD3001 was not located as of 29 September 2026, so the phase 3 result is cited here as a sponsor announcement, which is what it currently is.
• The randomized evidence that IS peer-reviewed for mood is phase 2b and phase 1b, and it is honestly mixed. The phase 2b adjunctive study's primary endpoint was MADRS change at week 6: the 20 mg group showed a least-squares mean difference against placebo of -4.5 (90% CI -6.96 to -2.07, P = .003) at week 3 but -3.1 (-6.13 to -0.16, P = .083) at week 6, i.e. the primary timepoint did not clear a conventional threshold. The signal concentrated where the mechanism predicts: in patients with baseline Insomnia Severity Index >= 15 the difference was -4.9 (-8.98 to -0.80), versus -0.7 (-5.16 to 3.76) in those below 15. A phase 1b monotherapy study in an enriched sample of placebo lead-in non-responders found HDRS-17 change at week 5 of -7.0 (SD 5.04) on 20 mg, -5.5 (4.34) on 40 mg and -4.4 (3.67) on placebo, p = 0.0456, with the benefit again greater at higher baseline insomnia severity (nominal p = 0.0059) and still present when sleep items were removed from the scale (nominal p = 0.0289).
• Against an active comparator it won on tolerability, not on efficacy. A 24-week exploratory study against quetiapine extended release (seltorexant 20 or 40 mg versus quetiapine-XR 150 or 300 mg, both added to ongoing SSRI/SNRI) reported treatment-emergent adverse events in 65.4% on seltorexant versus 80.8% on quetiapine-XR and all-cause discontinuation of 41.2% versus 47.1% (p = .5355), with the seltorexant 20 mg group numerically better on MADRS at weeks 18 and 24. The sponsor's 22 September 2025 release on the 26-week phase 3 MDD3005 reported response at 26 weeks of 57.4% versus 53.4% for quetiapine-XR — numerically higher and not statistically significant — with MADRS total score changes from baseline of -23.0 and -22.7, somnolence in 6% versus 24%, and mean weight change of +0.5 kg versus +2.1 kg.

📊 Correlative data

• The entire human case rests on trials, so there is little genuinely correlative material — and that is worth stating rather than padding. The one association that recurs across three independent studies is internal to those trials: the antidepressant effect is moderated by how badly the patient was sleeping at baseline. Higher baseline insomnia severity predicted larger mood improvement in the phase 2b adjunctive study, in the phase 1b monotherapy study, and it is the population the phase 3 program was deliberately enriched for.
• That moderation is the page's central unresolved question, and it is not resolvable from the data as published. A drug that fixes sleep in depressed people with broken sleep will improve depression scales partly because those scales contain sleep items and partly because sleep loss worsens mood. The phase 1b study addressed the first half — the benefit survived removing sleep items from HDRS — at nominal significance in an enriched sample of 86. Nobody has addressed the second half, which would need a non-orexin hypnotic matched for sleep improvement as the comparator.
• The biomarker observation that points at a route other than sedation. In the phase 1b monotherapy study the waking cortisol response fell in the 20 mg arm and not in the 40 mg or placebo arms. One arm of one study, and it is the kind of finding that either replicates into a mechanism or evaporates — but it is the only published human signal suggesting the mood effect runs through something other than more hours asleep.

🧪 Theoretical / extrapolated

• The selectivity argument, stated precisely. Orexin-A and orexin-B drive wakefulness through OX1R and OX2R. OX2R is the receptor feeding the histaminergic and monoaminergic arousal nuclei; OX1R is more associated with reward, appetite and stress circuitry. Blocking OX2R alone is meant to buy sleep without the OX1R consequences — and, on one reading, without the REM distortion of a dual blocker. Reported selectivity is roughly 2 logs over human OX1R.
• The preclinical package is unusually clean on target specificity. In rats, single oral doses of 3-30 mg/kg in the light phase dose-dependently reduced latency to non-REM sleep and prolonged non-REM time in the first 2 hours, with REM minimally affected; the effect held over 7 days of repeated dosing at 30 mg/kg and all sleep parameters returned to baseline after discontinuation, which is the rodent version of no rebound. The compound promoted sleep in wild-type mice and had no effect in OX2R knockout mice — target engagement proven by genetics rather than by a binding ratio.
• The abuse-liability argument is mechanistic and was tested against the obvious comparator. Seltorexant did not increase dopamine release in rat nucleus accumbens and did not produce conditioned place preference, while zolpidem in the same program did. Withdrawing a wake signal is not the same lever as potentiating GABA-A, and this is the experiment that shows the difference behaviorally.
• The literature contradicts itself on whether OX2R selectivity spares REM sleep, and the contradiction is the most useful thing on this page. One rat study found an OX2R antagonist increased non-REM duration with REM unaffected, and that adding an OX1R antagonist on top reduced REM latency and increased REM at the expense of non-REM — framed explicitly as dysregulation, and the strongest published case for going OX2R-selective. A separate program reached the opposite conclusion with a different OX2R-selective molecule, reporting that OX2R antagonism alone increased both non-REM and REM across rats, dogs and humans, similar to dual blockade. Both are real results with real molecules. What follows is that 'selective, therefore preserves sleep architecture' is a hypothesis this compound's own class has not settled.
• And this compound's own human data lands on the awkward side of that argument at the high dose. The 40 mg arm of the phase 1b study shortened REM onset latency and increased stage N1 sleep; the 20 mg arm did not. Whatever the class argument, dose appears to decide whether architecture is preserved.

How to read these tiers: they say how much human evidence exists, not how well something works — and ✗ flags harm, never a disappointing trial. How the evidence tiers work →

What Seltorexant actually does

This molecule takes a foot off the accelerator instead of pressing a brake, and that is not a metaphor — it is the difference between two receptor families. Two peptides made in the lateral hypothalamus, orexin-A and orexin-B (also called hypocretin-1 and hypocretin-2), hold the brain awake by signaling through two G-protein-coupled receptors, OX1R and OX2R. Orexin-A engages both; orexin-B is the OX2R-preferring ligand. OX2R is the receptor that carries arousal drive into the histaminergic and monoaminergic wake-promoting nuclei, and it is the only one this compound is built to occupy.

The selectivity is the entire design claim, and it is a number rather than an adjective. Bonaventure 2015 characterized the compound under its development code JNJ-42847922 — chemically [5-(4,6-dimethyl-pyrimidin-2-yl)-hexahydro-pyrrolo[3,4-c]pyrrol-2-yl]-(2-fluoro-6-[1,2,3]triazol-2-yl-phenyl)-methanone — and reported an approximate 2-log selectivity ratio versus the human orexin-1 receptor. The three orexin drugs that reached approval — suvorexant, lemborexant and daridorexant — block both receptors. One receptor instead of two is the whole argument for this one, and the paper states the stated purpose plainly: selective OX2R antagonists “may offer specificity and a more adequate sleep profile by preserving normal sleep architecture”.

Nothing in this mechanism touches GABA, and the program tested what that buys. Zolpidem and the benzodiazepines potentiate chloride flux at GABA-A; there is no such site here, and no channel. Bonaventure 2015 ran the behavioral consequence rather than asserting it: seltorexant did not increase dopamine release in rat nucleus accumbens and did not produce place preference after subchronic conditioning, “indicating that the compound lacks intrinsic motivational properties in contrast to zolpidem” — zolpidem being run in the same program as the comparator.

The proof that the sleep comes from OX2R is genetic, not pharmacological, and that is a stronger kind of evidence than a binding ratio. A selectivity number tells you what a compound prefers in a dish. The knockout tells you what the effect requires in an animal: the compound promoted sleep in wild-type mice and had no effect in OX2R knockout mice Bonaventure 2015. Delete the receptor and the drug does nothing, which is what target specificity actually looks like.

Occupancy arrives fast and leaves fast, and that shape is deliberate. Ex vivo receptor binding showed the compound quickly occupied OX2R binding sites in the rat brain after oral administration and rapidly cleared from the brain Bonaventure 2015. A hypnotic whose job is to cover a sleep period and be gone by morning wants exactly that curve, and the human half-life below confirms it was achieved rather than intended.

Cell, rodent, human — and where it stops

The chain here runs cell → rodent → dog → human and it is complete, which is rare on this site — so the interesting question is not whether it transferred but what changed on the way.

Rodent, and note the design detail that makes it a real test. Bonaventure 2015 gave rats single oral doses of 3–30 mg/kg during the light phase — that is, against the rat's own sleep period, when there is little sleep left to gain, which is the hard direction to show an effect. It dose-dependently reduced the latency to non-REM sleep and prolonged non-REM sleep time in the first 2 hours, while REM sleep was minimally affected. The reduced onset and increased duration were maintained upon 7-day repeated dosing at 30 mg/kg, and “then all sleep parameters returned to baseline levels following discontinuation” — the rodent version of no rebound insomnia, which is the standard objection to any hypnotic and is usually asserted rather than measured.

The species jump is a kinetics problem, not a potency problem, and it runs the opposite way to the usual one. The rat doses are 3–30 mg/kg; the human doses that produce somnolence are 20–60 mg total, roughly 0.3–0.9 mg/kg at 70 kg — one to two orders of magnitude lower. The same paper explains it: the compound rapidly cleared from the rat brain, so a rodent needs a large bolus to hold occupancy across a sleep bout. The lesson is not to scale this compound's rodent mg/kg into a human dose, since human dosing was established directly and lands far below where allometry would have put it. That is reasoning about why the numbers differ, not a measured species comparison.

Human, sleep, three trials with numbers. De Boer 2018 ran a two-way crossover of 40 mg for five days against placebo with 8 h polysomnography in insomnia disorder, 27 completing: sleep efficiency up 5.8% (SD 9.2) after a single dose and 7.9% (SD 9.8) after multiple doses, p < 0.001 at both, total sleep time up 27.7 and 37.9 min, latency to persistent sleep down 18.8 and 29.9 min. Brooks 2019 gave single doses of 10, 20 and 40 mg to 20 antidepressant-treated MDD patients with persistent insomnia in a four-way crossover: against a placebo least-squares mean latency of 61.05 min, ratios were 0.32 (80% CI 0.24–0.44), 0.15 (0.11–0.20) and 0.17 (0.12–0.23), all p < 0.001 — 40 mg no better than 20 mg, the ceiling's first appearance. Then the decisive one: Mesens 2025 randomized 364 participants 1:1:1:1:1 to 5 mg (n = 71), 10 mg (n = 74), 20 mg (n = 71), placebo (n = 75) or zolpidem 5–10 mg (n = 73) for 14 days, 55 sites, 6 countries, adults 18–64 and older adults 65–85 with ISI at or above 15. Night-1 latency improved versus placebo at 10 mg (LSM ratio 0.64, 90% CI 0.51–0.81) and 20 mg (0.51, 0.41–0.64), 20 mg beat zolpidem (0.71, 0.57–0.88), wake after sleep onset moved similarly, and on night 13 the 10 and 20 mg effects held while zolpidem's diminished. The arm that did not separate is 5 mg — the strength on sale.

Human, mood, and this is where the chain gets conditional. Savitz 2021 ran the phase 2b adjunctive study with an adaptive design (2:1:1 to placebo, 20 mg or 40 mg; then 3:3:1 to placebo, 10 mg or 20 mg, with 40 mg dropped after an interim analysis at the 160th patient). The primary endpoint was MADRS change at week 6, and the honest reading is that week 3 hit (−4.5, 90% CI −6.96 to −2.07, P = .003) and week 6 did not (−3.1, −6.13 to −0.16, P = .083). The effect concentrated where the mechanism says it should: −4.9 (−8.98 to −0.80) in patients with baseline ISI at or above 15 versus −0.7 (−5.16 to 3.76) below 15. Mesens 2024 then ran monotherapy in an enriched sample of 86 placebo lead-in non-responders out of 128 enrolled: HDRS-17 change at week 5 was −7.0 (SD 5.04) on 20 mg, −5.5 (4.34) on 40 mg and −4.4 (3.67) on placebo, p = 0.0456 — 20 mg beating 40 mg — with benefit again greater at higher baseline insomnia severity (nominal p = 0.0059) and surviving removal of the sleep items (nominal p = 0.0289). Phase 3 then met its primary endpoint: Johnson & Johnson announced on 29 May 2024 that study MDD3001, seltorexant 20 mg once daily added to an SSRI or SNRI in adult and elderly patients with MDD with insomnia symptoms, improved MADRS total score at day 43. That is a sponsor announcement and a congress presentation; I did not locate a primary peer-reviewed publication of MDD3001 as of 29 September 2026, and it is cited here as what it is.

The obstacle to the mood claim, named precisely. Every mood result is larger in the poor sleepers, which leaves the load-bearing question open: is this an antidepressant, or a sleep drug being measured with instruments that contain sleep items and given to people whose mood is being degraded by sleep loss? Mesens 2024 addressed half of it by removing the sleep items and keeping a nominal effect. Nobody has addressed the other half, and the design that would is stated in the next section.

Seltorexant pharmacokinetics — how much of it actually gets in

Start with what is unusual for anything in this catalog: the half-life is characterized, in humans, from a dedicated study. van der Ark 2018 enrolled five consecutive cohorts of healthy subjects, six on drug and two on placebo per cohort, gave 5–60 mg orally once daily over 10 days, and measured drug in plasma and urine over 24 h on days 1, 5 and 10. The compound was rapidly absorbed after morning dosing, with median t_max 0.5–1.5 h and mean t_1/2 2–3 h.

Exposure stops tracking dose above 20 mg, and that single sentence explains a lot of this page. At dose levels at or above 20 mg, mean C_max and mean AUC increased less than dose proportionally van der Ark 2018. So doubling the dose does not double the drug. The flat top that shows up repeatedly in the efficacy data — 40 mg no better than 20 mg on sleep onset Brooks 2019, 40 mg worse than 20 mg on depressive symptoms Mesens 2024 — has a kinetic contribution and not only a receptor one, and anyone reasoning that more capsules means proportionally more effect is reasoning against the measured absorption.

The arithmetic a person who has to work in the morning should actually do. At a mean t_1/2 of 2–3 h, eight hours after a bedtime dose is between roughly 2.7 and 4 half-lives, i.e. about 85–94% eliminated; twelve hours out it is 94–98%. That is arithmetic from the published half-life, not a measurement of next-day function — and the measurement exists separately, which is the better evidence. van der Ark 2018 ran a battery for vigilance state, sedation and alertness at 4, 6 and 8 h post-dose and found that apart from a mild decrease in attention at doses above 10 mg at four hours, no clinically relevant changes in other central measures were observed. Note what those two facts do together: the residual-effect data is reassuring because the drug is gone, so anything that slows its clearance removes the reassurance.

Where somnolence sits on the dose scale, measured in daytime dosing. Because van der Ark 2018 dosed in the morning, it could read sedation directly rather than inferring it: at doses at or above 20 mg the compound consistently induced somnolence on all study days, and at doses above 5 mg it was identified as sedating by the Addiction Research Center Inventory-49 at four hours post-dose. The on-target effect and the unwanted daytime effect are the same effect at the wrong hour, which is why the short half-life is the safety feature.

What is absent, stated as absence in my search rather than absence in the literature. On 29 September 2026 I did not locate a published food-effect study, a hepatic- or renal-impairment study, or a dedicated CYP interaction study for this compound. The phase 1 measured urinary as well as plasma concentrations, so renal handling was at least observed, and the phase 2b and phase 3 programs dosed it on top of ongoing SSRIs and SNRIs, so co-administration with those has clinical exposure behind it even without a dedicated interaction study. Absence in a search is a statement about the search.

What would have to be true, and how you would know it was not

Four predictions. The first cuts directly against the product as sold, the second is the one most likely to mislead somebody reading their own labs, and each names something measurable with a direction and a window.

1. THE PREDICTION THAT CUTS AGAINST THE PRODUCT, and it comes from the pivotal trial rather than from skepticism. Mesens 2025 tested 5, 10 and 20 mg against placebo and zolpidem with polysomnography in 364 people. The doses reported as separating from placebo on night-1 latency to persistent sleep and on wake after sleep onset were 10 mg and 20 mg. 5 mg — the strength on the bottle — is the arm that did not. So the prediction is concrete: one capsule should produce no detectable change in sleep-onset latency or wake after sleep onset. The instrument is a validated one and it is free: the Insomnia Severity Index at baseline and at 14 days, the same window the trial used, where the accepted threshold for a real change is about 6 points. A wrist actigraph's sleep-onset latency over the same fortnight is the objective companion. If one capsule moves ISI by 6 points, that result contradicts the dose-response of the trial this page is built on, and it should be taken seriously rather than explained away — but the honest prior, from 364 randomized people, is that it will not move.

2. Cortisol drawn on this drug is not a clean baseline, and the direction is down. Mesens 2024 measured serum cortisol and the cortisol waking response and found the waking cortisol response decreased in the 20 mg arm and not in the 40 mg or placebo arms. Anyone investigating an HPA-axis question — morning serum cortisol, or a salivary cortisol awakening response — while taking this should expect a lower reading and should not read it as adrenal insufficiency. The window comes from the drug's own kinetics van der Ark 2018: at a mean t_1/2 of 2–3 h, five half-lives is 10–15 hours, so a draw taken on a full off-drug morning is interpretable and a draw taken 4–6 h post-dose is not. This is the single most likely way this compound produces a confusing lab result.

3. The architecture prediction, which is also a test of the selectivity story. The class contradicts itself: Dugovic 2014 found OX2R blockade increased non-REM with REM unaffected, and that adding OX1R blockade reduced REM latency and increased REM at the expense of non-REM; Gotter 2016 found OX2R blockade alone increased both non-REM and REM from mouse to man. This compound's own human data sides with the second at the high dose: in Mesens 2024 the 40 mg arm showed reduced REM onset latency and increased stage N1 sleep, and the 20 mg arm did not. Prediction: REM onset latency shortens as dose rises above 20 mg, and stage N1 — the lightest sleep — increases with it. Instrument: a single-night home polysomnography or a device that reports REM latency, run dose-matched. If REM latency shortens at 20 mg as well, the “selectivity preserves architecture” claim is empty for this molecule.

4. No rebound on stopping — a prediction in the product's favor that has never been tested in a person. Bonaventure 2015 reported that after 7-day dosing in rats, “all sleep parameters returned to baseline levels following discontinuation”. Prediction: ISI measured one week after stopping should return to its pre-drug value rather than overshoot it. That is the falsifiable half — if insomnia comes back worse than baseline, the rat result did not transfer, which would matter most to the reader, and it is the one thing in this whole file that a single person can test on themselves with a free questionnaire and no equipment.

What nobody has tested yet

Five experiments nobody has run. The first two would settle what this specific product does, and both are cheap next to what has already been spent on this molecule.

1. The head-to-head against a dual antagonist, on architecture. The entire case for OX2R selectivity is that it preserves normal sleep architecture, and the class cannot agree whether it does Dugovic 2014 Gotter 2016. Mesens 2025 ran the comparator that answers a commercial question — zolpidem — and beat it, including on night 13. Nobody has published seltorexant against suvorexant, lemborexant or daridorexant with REM latency, REM percentage and slow-wave time as the endpoints. One polysomnography night per arm would resolve a decade of class argument, and it is the trial with the least commercial incentive behind it, because the sponsor's differentiation does not depend on winning it.

2. Whether 5 mg does anything over weeks. The 5 mg arm was tested once, for 14 days, in one dose-finding trial, on objective endpoints, and did not separate Mesens 2025. No study has given 5 mg for 8 weeks or longer to anybody, on any endpoint. That is precisely the exposure the product on sale creates, and the interesting possibility is not that 5 mg is weakly effective but that a subjective scale over two months could show something a two-week objective endpoint could not. Until somebody runs it, the strength being sold is the least-studied dose in the program.

3. The experiment that separates the antidepressant from the hypnotic. Every mood result is bigger in the worse sleepers Savitz 2021 Mesens 2024. The design that settles it is a three-arm trial: seltorexant, a non-orexin hypnotic titrated to produce the same objective sleep improvement, and placebo, with MADRS excluding the sleep item as primary. Pinter 2026 ran seltorexant against quetiapine-XR, but quetiapine is an antipsychotic with its own mood pharmacology and was not sleep-matched, so it answers a tolerability question instead — which it did, with treatment-emergent adverse events of 65.4% versus 80.8%. The sleep-matched comparison has not been published in any orexin drug.

4. What happens past six months. The longest published exposures are the 24-week exploratory study Pinter 2026 and the 26-week phase 3 MDD3005 the sponsor described in September 2025. Nothing longer than roughly six months exists. For a mechanism that holds down a wake-promoting system nightly, the open questions are whether receptor expression changes, whether the maintained night-13 effect Mesens 2025 is still maintained at month 18, and whether discontinuation after a year behaves like discontinuation after a week. No study is designed to answer any of them.

5. The ECG finding nobody has followed up, and it is this page's most specific gap. Mesens 2025 reported four asymptomatic ECG-related treatment-emergent adverse events leading to discontinuation — three on 5 mg and one on 20 mg. Asymptomatic means they were found because the trial ran ECGs, and the distribution is odd: three of four at the dose that had no efficacy signal, which is exactly the pattern you would expect from chance in a 364-person trial rather than from a dose-dependent drug effect. That is a reason to be unimpressed by it and also the reason nobody has resolved it: no dedicated thorough-QT or cardiac-monitoring study has been published for this compound, so the honest statement is that four events were seen, they look like noise, and the measurement that would prove it has not been run. The person swallowing a capsule is not running an ECG either way.

Seltorexant — its own safety story, not its class's

The first safety fact about this compound is not a side effect, it is a category: nothing has been approved, so nothing has been assessed. Suvorexant, lemborexant and daridorexant have labels — contraindications, interaction tables, a stated population, a boxed set of warnings where warranted. Seltorexant has none of that in any jurisdiction, because it is still in development. That is not the same as being safe and it is not the same as being dangerous; it means the document that normally carries the answers does not exist, and this page is not a substitute for it.

The predicted harm is the on-target effect at the wrong hour, and it has been measured rather than guessed. A drug that blocks the wake signal at 11pm is a drug that can still be blocking it at 8am. van der Ark 2018 tested exactly this by dosing in the morning: somnolence was consistent at doses at or above 20 mg, the compound read as sedating on the ARCI-49 above 5 mg at four hours, and on a battery at 4, 6 and 8 h the only other finding was a mild decrease in attention above 10 mg at four hours. The mitigation is the half-life and it is arithmetic, not advice: mean t_1/2 2–3 h means a bedtime dose is 85–94% gone by eight hours later. Which also identifies what removes the mitigation — anything that slows clearance, and a compound with no published hepatic-impairment or interaction data is a compound where that risk is unquantified.

Observed adverse events, including the ones that cut in the product's favor. Across the mood studies the events reported at 5% or more were somnolence, headache and nausea Savitz 2021, and headache and somnolence were the most common in the insomnia crossover De Boer 2018. In the largest trial, treatment-emergent adverse events were lower on drug than on placebo: 73/216 (33.8%) across the combined seltorexant doses versus 37/75 (49.3%) on placebo and 31/73 (42.5%) on zolpidem, with two serious events, one on seltorexant 20 mg and one on zolpidem Mesens 2025. Against quetiapine-XR over 24 weeks it was 65.4% versus 80.8% Pinter 2026, and the sponsor's 26-week phase 3 reported somnolence in 6% versus 24% and weight change of +0.5 kg versus +2.1 kg. On tolerability this molecule looks good, and saying so is part of being accurate.

The cataplexy question, answered properly rather than either dismissed or inflated. Narcolepsy with cataplexy is a disease of orexin neuron loss — the signal is permanently absent, not transiently blocked — so the mechanistic worry about an orexin antagonist is legitimate and is not the same condition. Gotter 2016 addressed it directly in the species where cataplexy is most readily provoked, reporting that OX2R-selective antagonism increased non-REM and REM sleep in dogs without inducing cataplexy. What remains genuinely open is architecture rather than cataplexy: at 40 mg this compound shortened REM onset latency and increased stage N1 Mesens 2024, which is the direction a dual antagonist goes Dugovic 2014, so the selectivity protection is dose-limited on its own data.

The risk specific to buying this as a capsule is the monitoring, not the molecule. This drug's development population is people with major depressive disorder, and the trials were built accordingly: suicidal ideation and behavior were assessed with the Columbia Suicide Severity Rating Scale van der Ark 2018 Brooks 2019, and safety assessments included serial ECGs, vital signs and clinical laboratory work — which is how Mesens 2025 detected four asymptomatic ECG-related discontinuations nobody would otherwise have known about. Somebody taking capsules for low mood and bad sleep has none of that, and the deficit is not theoretical: the reason the trials measured it is that the population being treated is a population where mood can deteriorate. The mitigation that follows from the same mechanism is a scale, not a supplement: if the reason for taking it is sleep, the Insomnia Severity Index at baseline and 14 days answers whether it is working; if the reason is mood, the honest reading of this page is that the dose sold is below the one the mood trials used and the monitoring the mood trials required is absent.

Finally, you cannot verify what you bought. The product page carries a strength, a count and a storage range, and no CAS number, molecular formula, molecular weight or purity figure — nothing to check a capsule against. For a molecule whose whole clinical case is dose-dependent, with a studied window between 10 and 20 mg, an unverifiable 5 mg claim is the input to every calculation on this page.

Sources read for this page

• Bonaventure P, Shelton J, Yun S, Nepomuceno D, Sutton S, Aluisio L, Fraser I, Lord B, Shoblock J, Welty N, Chaplan SR, Aguilar Z, Halter R, Ndifor A, Koudriakova T, Rizzolio M, Letavic M, Carruthers NI, Lovenberg T, Dugovic C. Characterization of JNJ-42847922, a Selective Orexin-2 Receptor Antagonist, as a Clinical Candidate for the Treatment of Insomnia. Journal of Pharmacology and Experimental Therapeutics 2015;354(3):471-82 · PMID 26177655
• De Boer P, Drevets WC, Rofael H, van der Ark P, Kent JM, Kezic I, Parapatics S, Dorffner G, van Gerven J, Benes H, Keicher C, Jahn H, Seiden DJ, Luthringer R. A randomized Phase 2 study to evaluate the orexin-2 receptor antagonist seltorexant in individuals with insomnia without psychiatric comorbidity. Journal of Psychopharmacology 2018;32(6):668-677 · PMID 29848147
• van der Ark PD, Golor G, van Nueten L, Nandy P, de Boer P. Multiple daytime administration of the selective orexin-2 receptor antagonist JNJ-42847922 induces somnolence in healthy subjects without residual central effects. Journal of Psychopharmacology 2018;32(12):1330-1340 · PMID 30182786
• Brooks S, Jacobs GE, de Boer P, Kent JM, Van Nueten L, van Amerongen G, Zuiker R, Kezic I, Luthringer R, van der Ark P, van Gerven JM, Drevets W. The selective orexin-2 receptor antagonist seltorexant improves sleep: An exploratory double-blind, placebo controlled, crossover study in antidepressant-treated major depressive disorder patients with persistent insomnia. Journal of Psychopharmacology 2019;33(2):202-209 · PMID 30644312
• Savitz A, Wajs E, Zhang Y, Xu H, Etropolski M, Thase ME, Drevets WC. Efficacy and Safety of Seltorexant as Adjunctive Therapy in Major Depressive Disorder: A Phase 2b, Randomized, Placebo-Controlled, Adaptive Dose-Finding Study. International Journal of Neuropsychopharmacology 2021;24(12):965-976 · PMID 34324636
• Mesens S, Kezic I, Van Der Ark P, Etropolski M, Pandina G, Benes H, Savitz A, Drevets WC. Treatment effect and safety of seltorexant as monotherapy for patients with major depressive disorder: a randomized, placebo-controlled clinical trial. Molecular Psychiatry 2024; published online 11 December 2024 (doi 10.1038/s41380-024-02846-5) · PMID 39663378
• Mesens S, Krystal AD, Melkote R, Xu H, Pandina G, Saoud JB, Luthringer R, Savitz A, Drevets WC. Efficacy and Safety of Seltorexant in Insomnia Disorder: A Randomized Clinical Trial. JAMA Psychiatry 2025; published online 13 August 2025 (doi 10.1001/jamapsychiatry.2025.1999) · PMID 40802194
• Pinter C, Thase ME, McIntyre RS, Cooper K, Xu H, Pandina G, Savitz A, Drevets WC. Safety, tolerability, and preliminary efficacy of seltorexant versus quetiapine extended release as adjunctive therapy in major depressive disorder: a randomized, flexible-dose, 6-month, parallel-group, exploratory study. International Journal of Neuropsychopharmacology 2026; published online 9 March 2026 (doi 10.1093/ijnp/pyag009) · PMID 41795948
• Dugovic C, Shelton JE, Yun S, Bonaventure P, Shireman BT, Lovenberg TW. Orexin-1 receptor blockade dysregulates REM sleep in the presence of orexin-2 receptor antagonism. Frontiers in Neuroscience 2014;8:28 · PMID 24592208
• Gotter AL, Forman MS, Harrell CM, Stevens J, Svetnik V, Yee KL, Li X, Roecker AJ, Fox SV, Tannenbaum PL, Garson SL, Lepeleire ID, Calder N, Rosen L, Struyk A, Coleman PJ, Herring WJ, Renger JJ, Winrow CJ. Orexin 2 Receptor Antagonism is Sufficient to Promote NREM and REM Sleep from Mouse to Man. Scientific Reports 2016;6:27147 · PMID 27256922

Seltorexant — interference & stacking

Bloodwork to run alongside Seltorexant

Baseline first, then again at 8–12 weeks.

The Sleep Quality & Recovery panel covers these in one order — 11 markers, $172.35 with the discount applied.

Check results you already have → · All 103 markers A–Z

Seltorexant — frequently asked questions

What Seltorexant is used for

Seltorexant appears under 2 goals in the goal router.

Related Sleep compounds

← Explore the full Protocol Vault

↑ Back to on this page