DSIP (Delta Sleep-Inducing Peptide): the replication record and dosing

It came out of a rabbit whose sleep was electrical

Monnier's group in Basel started in 1963 with a cross-circulation question: if you make an animal sleep, does its blood carry something that will make a second animal sleep? Donor rabbits were held in slow-wave sleep by running current through the intralaminar thalamus. Blood coming out of their brains, drawn from the dural venous sinus, was filtered outside the body, and the filtrate was infused into the brain cavities of recipient rabbits. Slow-wave power in the recipients went up.

By 1977 the active fraction had been purified, sequenced and made synthetically. Schoenenberger and Monnier tested the synthetic nine-residue peptide against five of its own breakdown fragments, two altered versions and a related three-residue peptide, blinded, in 58 rabbits. Only the intact nine-residue version raised delta and spindle activity.

The 1978 follow-up in 61 rabbits put a number on it: average delta activity about 35% above controls given either vehicle or any of the eight other peptides. The structure had to be exact — shorten the chain by one or two residues, swap an amino acid, or use the beta-aspartyl form instead of the alpha, and the effect weakened or disappeared.

Four things about that method have never been answered

• The donor rabbits were not asleep in any normal sense. Their sleep was produced by passing current through a piece of brain. Whatever turned up in that blood is the product of electrically stimulating the thalamus, which is a different claim from a molecule that builds up during ordinary sleep.
• What was measured was a waveform, not sleep. Delta-band amplitude in a near-anaesthetised preparation is a signal on a screen. Nothing in the original work showed a recipient animal sleeping longer or better by any behaviour you could watch.
• The dose curve is a hump, not a slope. Schoenenberger reported bell-shaped curves with different peaks by route, and Scherschlicht's group at Roche found the same hump in rabbits, cats and mice. A compound with a hump-shaped curve is close to impossible to disprove: any failed replication can be blamed on landing on the wrong side of the peak.
• No gene, no precursor, no receptor. The sequence exists. The biological machinery that would build it and read it does not.

The replication record, study by study

Sort that table by laboratory and the pattern jumps out

Every strongly positive human result came from Basel — Schneider-Helmert, working alongside Schoenenberger, who isolated the peptide in the first place. Every group that tested it without that connection got a null or a triviality.

Tobler and Borbély in Zürich concluded flatly that the compound does not qualify as a sleep-promoting substance. Monti in Montevideo found the placebo comparison fall apart once the starting point was accounted for. Bes in Amsterdam found statistically real but weak effects, warned that part of the signal came from drift in the placebo group, and advised against using it.

Three more facts compress the record further.

Nearly every human study used the same single dose, 25 nmol/kg into a vein. No dose-finding study was ever run in a person. Against a hump-shaped dose curve, testing one rung of the ladder over and over cannot rule the compound in or out. It can only tell you about that rung.

Every human study injected into a vein. The subcutaneous shot that is the entire present-day market has never been tested in a human sleep trial.

And the total number of people ever randomly assigned to DSIP against placebo for sleep, in the whole fifty-year literature, is under forty.

In fifty years nobody has registered a trial of it

Asking the ClinicalTrials.gov registry on 4 August 2026 for "delta sleep-inducing peptide" returns zero studies. Asking for "emideltide" returns zero. Asking for DSIP as a registered intervention returns zero.

Not an ongoing trial. Not a completed one. Not a terminated one. Not a withdrawn one. In fifty years, nobody has registered an attempt to settle the question the compound is named after.

The biggest human numbers in the file are about withdrawal, not sleep

In 1983 Dick, Grandjean and Tissot in Geneva gave DSIP into a vein at 25 nmol/kg, as the only treatment, to 67 patients going through withdrawal — 28 from alcohol, 39 from opiates. Twenty-seven per cent were lost or unsuitable to assess. Of the 49 who could be assessed, 48 were reported to benefit, with the physical signs settling immediately and staying settled. Anxiety took longer.

The 1984 extension enrolled 107 inpatients, 47 in alcohol withdrawal and 60 in opiate withdrawal. Symptoms vanished or improved markedly in 97% of the opiate cases and 87% of the alcohol cases. Opiate patients needed more injections and took longer. It was well tolerated apart from headache in a few people.

An independent open trial of DSIP for coming off opioids was published by Backmund's group in Munich in 1998.

Every one of those is an uncontrolled series. No placebo. No blinding. Doctors and nurses rating the outcome. And between 13% and 27% of the people enrolled dropped out of the analysis after the fact.

Here is why that matters more than usual. Acute withdrawal is a syndrome that ends by itself over a few days whether or not you give anything. An open series in a self-limiting syndrome measures the passage of time. Two hundred people getting better on schedule tells you almost nothing that one person getting better on schedule does not.

The reasoning behind the withdrawal work is at least specific and testable. Tissot had shown that morphine, alcohol, pentobarbital and DSIP all produced spindle-heavy slow-wave sleep when injected into the bulbo-mesencephalo-thalamic recruiting system, and that naloxone — the opioid blocker — reversed all four. Young and Key confirmed that reversal independently in rats: DSIP into the lateral ventricle raised REM and non-REM sleep, and pre-treating with naloxone at a dose that hits the mu-opioid receptor blocked it.

That is the strongest mechanical thread in the whole file, and note what it points at: an interaction with the opioid system, not a sleep factor.

Seven patients, four different conditions, no control group

The chronic pain evidence is one pilot study. Larbig and colleagues treated seven people — migraine and vasomotor headache, long-term tinnitus, and psychogenic pain attacks — with DSIP into a vein on five days running, then five more injections every 48 to 72 hours. Pain fell significantly in six of the seven against each person's own baseline stretch, and low mood fell with it.

Seven patients. Four different conditions. No placebo, no blinding, and a within-person comparison against a baseline assessed after the fact, in conditions that swing on their own. It is a hypothesis-generating observation that nobody generated a hypothesis from: in the forty-two years since, no controlled pain trial of DSIP has been run.

The rat says growth hormone and the woman says no

The rat result is genuinely elegant. Iyer, Marks, Kastin and McCann kept male rats awake for four hours on a slowly turning wheel. When they were taken off it, both slow-wave sleep and growth hormone in the blood shot up. Microinjecting a highly specific anti-DSIP antibody into the third ventricle blocked both rises. Ordinary rabbit serum did not.

The human result is the opposite. Giusti's group in Genoa gave 25 µg/kg of DSIP into a vein over thirty minutes to eight healthy women. Baseline growth hormone: unchanged. Prolactin: unchanged. The daily rhythm of both: unchanged. The growth hormone response to an arginine challenge: unchanged. The authors pointed out that the dose they used was already known to alter the ECG — so it was reaching the subject, and it was doing nothing to growth hormone.

Marketing copy that promises a growth hormone release is citing the rat and ignoring the woman.

Luteinising hormone runs the same course. DSIP triggered LH release in ovary-removed rats that had been primed with oestrogen and progesterone, did nothing in unprimed animals, and did nothing at all to pituitary tissue in a dish — meaning any effect is happening in the hypothalamus and depends on the hormonal background.

DSIP-like material has been found sitting in the same storage vesicles as LH-releasing hormone in the rat median eminence, which is suggestive. No endocrine effect of DSIP in a human being has ever been shown by anybody.

The stress work is the most internally consistent animal data

Rats given DSIP into the belly at 40, 120, 360 and 1080 µg/kg before acute and repeated electric foot-shock showed liver malondialdehyde back to normal, liver cell protein production restored, and serum alanine aminotransferase back to normal. The dose curve was again a hump, not a slope: 120 µg/kg normalised fat oxidation damage in acute stress, while 40 µg/kg was the effective dose under chronic stress.

DSIP and its analogue DSIP-12 cut the frequency, severity and length of metaphit-triggered sound-induced seizures in rats, with the analogue beating the parent compound.

Schoenenberger's own characterisation work reported effects on the daily activity rhythm, on brain monoamine levels, on plasma proteins and on cortisol — a spread wide enough that he coined the phrase "programming functions" rather than claiming a mechanism.

Two caveats attach here. Most of this literature comes from Russian, Ukrainian and Serbian laboratories, much of it published in Russian, and very little has been repeated outside those groups. And an effect on rodent liver enzymes under electric shock is a long way from anything you can act on.

Fifty years of looking and nobody has found the receptor

This is the strangest fact about DSIP and the one that should carry the most weight. For a peptide under continuous study since the 1970s, no gene, no precursor protein and no receptor has ever been isolated.

What exists instead:

• Binding sites, never a receptor. Autoradiography in 1984 found tritium-labelled DSIP sticking to the bodies and branches of cultured rat brain-stem neurones, and unlabelled peptide pushed it off. That observation is forty-two years old and was never followed through into a cloned receptor.
• The material shows up in the wrong places. DSIP-like material concentrates in hypothalamic neurosecretory nuclei that have little to do with sleep. Kovalzon's reading is that the material being detected belongs to a different, related molecule nobody has identified — which would explain both the distribution and the erratic biology.
• Artificial versions work better than the real one. Kovalzon's group found that certain man-made DSIP analogues significantly promote slow-wave sleep in rabbits and rats, and that DSIP itself does not. A naturally occurring dermorphin ten-residue peptide sharing five of nine positions promotes slow-wave sleep, while its mirror-image form suppresses it.
• An adrenaline hypothesis that failed its test. Graf and Kastin's 1986 review proposed that DSIP works by changing adrenaline-type transmission. Direct testing found DSIP did nothing to noradrenaline release from rat brain nerve endings at concentrations from 10⁻⁸ to 10⁻⁵ M, while leu-enkephalin in the same preparation did.
• An opioid interaction that passed. The naloxone reversal replicated. It remains the only mechanical claim about DSIP with independent support.

It gets into the brain and then it is gone in four minutes

DSIP is one of the small set of peptides shown to cross from blood into brain by simple diffusion despite carrying a charge and liking water — a finding that made it a standard test compound in blood-brain barrier research. How much reaches dog spinal fluid tracks its blood concentration, its survival time in blood, and how fat-soluble it is.

Then the clearance numbers undo the story. Kato and colleagues built an enzyme immunoassay and measured it directly after injection into a vein:

Breakdown is why. Put DSIP in human or rat blood and what comes back has the fingerprint of free tryptophan — the front residue gets snipped off and the intact peptide stops existing. Phosphorylated and iodine-labelled versions break down more slowly and form complexes, which is the proposed reason they act more strongly and more consistently than the plain peptide.

Now put that number next to the claim being made. A compound that is half gone from your blood in two to four minutes, injected before bed, is absent from your circulation for essentially the entire night it is supposed to be governing.

Two readings survive that, and only two.

Either the injection is a brief trigger that starts something slower downstream — and the one piece of evidence for that is Schneider-Helmert's observation that the improvement carried into a post-treatment placebo night.

Or the reported effects are not the compound.

Nobody has run the experiment that separates those two.

One more thing sits underneath it. The DSIP-like material your body carries in blood, urine and spinal fluid appears to be bound to a larger carrier protein and protected from being chewed up. Injected synthetic peptide is not bound to anything.

What people taking it report, counted: 18 better, 10 worse, 5 nothing, of 35

Forty-five Reddit threads and comments about DSIP are catalogued as sources on this page, pulled on 29 June 2026. Ten of them are protocol questions, vendor posts and guides with no outcome in them. That leaves 35 posts where a named person says what they took and what happened.

Eighteen X posts and two Instagram reels are also catalogued and are deliberately not counted in that 35. They run overwhelmingly positive, and on those platforms a first-hand report and an advertisement are not reliably distinguishable.

This is a curated set, not a survey, and the direction of the curation is unknown. Read it only for the shape: about half the people who bothered to write down what happened reported no benefit or an active harm. That is not the distribution a working sleeping pill produces. It is the same split the controlled trials produced, reached by a completely different route.

The 18 reports of improvement are vivid and specific

First night, tracked on a wearable:

Deep and REM each up by about forty minutes, with the writer noting tolerance appearing after three or four days:

The dream reports are their own category and they are consistent enough to be worth reading as a signal rather than as colour:

Long-run users. Two years on it, off a self-experimentation forum:

Three weeks before anything changed, then months with no tolerance, at 300–500 mcg under the skin:

Sleep going from three to five hours a night to seven to nine, after alcohol withdrawal wrecked it:

The 5 reports of nothing happening are just as specific

Two full vials, dosed under a doctor's direction:

A full 5 mg vial, from a commenter under one of the enthusiastic posts above:

Two months, in someone whose sleep was already good:

125 mcg in the day and about 300 mcg before bed:

The 10 reports where it made sleep worse are the ones vendor pages never carry

Straightforward insomnia from a sleeping compound:

Four hours of sleep, from someone who ran it for sleep-maintenance insomnia:

Deep sleep measured going the wrong way — 45 minutes a night down to 8 by day five:

Worked for sleep and produced a hangover the next day:

Longest uninterrupted sleep the tracker had ever recorded, followed by two hours of lying awake on subsequent nights:

Sleep getting worse deserves more attention than it usually gets, because it is neither the shape of an inert compound nor the shape of a placebo. Two things in the published record line up with it directly.

Schneider-Helmert's 1981 study found a mild waking-up effect in the first hour after injection, before any sleep-promoting effect showed up. People are injecting fifteen to sixty minutes before bed, which puts the dose squarely inside that first hour.

And the dose curve reported throughout the animal literature is a hump. A dose past the peak can plausibly do the opposite of the dose at the peak. That curve has never been mapped in a human being, at any dose, by anyone.

One report describes tolerance, and nothing in the literature addresses it

No human study ran longer than seven nights in a row. Whether an effect survives weeks of nightly use has never been tested by anybody.

One report is mixed, and it is the most precise in the set

Sleep length and deep sleep unchanged, but REM present every night where it had not been before:

Two threads independently reproduce the 1980s withdrawal finding

The most unexpected corroboration in the self-report record is not about sleep at all. People treating opiate withdrawal describe relief from DSIP under the skin and up the nose in terms close to Dick's Geneva series of 1983 — forty-two years later, with no sign of knowing that series exists.

That does not make the withdrawal claim true. Uncontrolled reports of a self-limiting syndrome resolving are still uncontrolled reports. What it does mean is that the strongest signal in the published record and the strongest signal in the self-report record point at the same non-sleep use, which is worth more than another sleep testimonial.

Why none of it settles the question

Sleep is the outcome most sensitive to what you expect, and self-report is the weakest instrument for measuring it. One detail from the controlled record cuts against dismissing the whole thing, though. In the Bes trial the objective polysomnography numbers improved while how the sleep felt did not. That is the reverse of a placebo signature — placebo moves the feeling first. The authors still judged the effect weak, partly attributable to drift in the placebo group, and advised against using it. That position remains more defensible than either the enthusiasts' or the debunkers'.

Sleep does gate pain and repair, and that is not evidence about this compound

If you are here because pain is keeping you awake, or because being awake is making the pain worse, the link is real and it is well documented. It is also a completely different proposition from anything above, and keeping the two apart is the whole job of this section.

Poor sleep predicts new and worsening pain more reliably than pain predicts poor sleep. Population studies following people over time show sleep problems arriving before new pain and before flare-ups. Fine-grained work with objective measures on both sides finds the sleep-to-pain direction is the stronger one.

One night without sleep measurably lowers your pain threshold. Acute sleep deprivation amplifies the pain response in primary somatosensory cortex, blunts it in the striatum and insula, and widens the range of temperatures a person calls painful. Outside the laboratory, ordinary night-to-night changes in one person's sleep quality predict next-day changes in their pain, in both directions.

Two nights of partial restriction is enough to change a pressure pain threshold at the trapezius. Twenty-one healthy volunteers, paired crossover, sleep cut by half for two nights: pressure pain threshold at the trapezius fell, p = 0.017, and the pain-triggered EEG response got bigger.

Broken sleep slows healing in animals. Fragmenting sleep stretched the time to half-closure of a wound from 10.3 to 13.3 days in diabetic mice, with more TNF-α in the wound. Sleep-deprived rats healing burns grew fewer tissue-building cells and fewer new capillaries than controls.

Now the line, drawn as sharply as it needs to be.

All of that is evidence that sleep matters to pain and to repair. None of it is evidence that DSIP improves sleep. The first proposition has population studies, controlled human experiments and animal tissue work behind it. The second has an inconsistent fifty-year record concentrated in one laboratory and a self-report split that is close to a coin flip.

Treating the strength of the first as if it transfers to the second is the exact move the marketing around this compound depends on. The evidence for one thing gets handed over while a different thing is sold. It is worth being able to see it happen, because it happens in one sentence and the sentence sounds reasonable.

If the goal is sleep, here is what actually has the evidence

Cognitive behavioural therapy for insomnia. Twenty randomised controlled trials, 1,162 participants. Time to fall asleep improved by 19.0 minutes (95% CI 14.1–23.9). Time spent awake after falling asleep improved by 26.0 minutes (95% CI 15.5–36.5). Sleep efficiency improved by 9.9% (95% CI 8.1–11.7). The gains held at later follow-up and no harms were reported. It is first-line in the guidelines for a reason, and its two active ingredients — restricting time in bed, and using the bed only for sleep — need no prescription and cost nothing.

Licensed sleeping pills, with their costs stated. The largest network meta-analysis covers 170 trials and 47,950 participants. Benzodiazepines, doxylamine, eszopiclone, lemborexant, seltorexant, zolpidem and zopiclone all beat placebo for short-term treatment, with standardised effect sizes from 0.36 to 0.83. Eszopiclone and lemborexant came out with the best overall profiles. Zopiclone and zolpidem caused more people to quit over side effects than placebo did.

And the one most people have already tried, which does not work. In that same analysis, melatonin, ramelteon and the unlicensed drugs showed no material benefit overall.

Set the comparison out plainly, because it is the comparison that matters. The best-evidenced sleep intervention rests on 1,162 randomly assigned people, with an effect that lasted and no harms recorded. The best-evidenced DSIP sleep trial rests on eight people per arm and concluded that short-term treatment was not likely to be of major benefit.

Dosing, and the gap between the studies and what is sold

Every human trial used 25 nmol/kg into a vein. At a molecular weight of 849 that is 21.2 µg/kg — about 1.5 mg for a 70 kg adult. Schneider-Helmert's 1986 week-long study used 30 nmol/kg. Giusti's endocrine study used 25 µg/kg into a vein over thirty minutes.

The market dose is 100 to 300 mcg under the skin before bed.

That is five to fifteen times lower than the only dose ever given to a human being, by a route never tested in one. Both changes are undocumented. Against a compound whose dose curve is a hump rather than a slope, cutting the dose fifteen-fold is not the cautious choice. It is an untested choice, and it can land anywhere on that curve including flat zero.

Timing in the circulating protocols is 15 to 60 minutes before you intend to sleep. Worth knowing before you set that: Schneider-Helmert's 1981 study found the mild waking-up effect in the first hour after injection, with the sleep-promoting effect only turning up in the second.

Under the skin and up the nose are the two routes in circulation. There is no comparative human data, and nothing has ever been published on how the body absorbs and clears the nasal form.

Mixing a vial is arithmetic

DSIP ships as a dry powder in a sealed glass vial, usually 2 mg or 5 mg. You mix it with bacteriostatic water — sterile water with 0.9% benzyl alcohol in it, the preservative that makes it safe to put a needle into the same vial more than once. The measuring device is a U-100 insulin syringe: 100 marks per millilitre, so one mark is 0.01 mL.

Two lines produce every number you need:

• Concentration in mcg per mL = total mcg in the vial ÷ mL of water added.
• Mcg per mark = concentration ÷ 100.

Worked through: a 5 mg vial holds 5,000 mcg. Add 2.5 mL of bacteriostatic water. 5,000 ÷ 2.5 = 2,000 mcg/mL. 2,000 ÷ 100 = 20 mcg per mark. A 100 mcg dose is 100 ÷ 20 = 5 marks on the barrel.

Adding more water does not change how much peptide is in the vial. It changes only how many marks carry a given dose. Below about five marks, the error in reading a U-100 barrel starts to dominate your dose — which is the practical argument for more water, not less, at the 100 mcg tier.

Let the vial come to room temperature before you add water. Run the water down the inside wall instead of squirting it onto the powder. Swirl until it dissolves and never shake it: the shearing at the air-liquid surface unfolds peptides and makes them clump, and clumping is exactly the mechanism FDA named when it raised the question of whether the immune system would react to it.

Storage has a hard clock

A 5 mg vial at 100 mcg a night is fifty doses, which is far more nights than the four-week window a mixed vial gives you. The arithmetic forces a choice: buy a smaller vial, take a bigger nightly dose, or throw most of what you paid for away.

Timelines, and what would count as an answer

The circulating protocols run five nights on and two off, or three to five nights a week, for four to eight weeks. There is no evidence base under any of those schedules. No human study of DSIP ran longer than seven nights in a row, and none tested taking it on and off.

The trial that would settle it is easy to specify and has never been run: randomised, blinded, placebo-controlled, with overnight sleep recording, in long-term insomnia, DSIP under the skin at 100–300 mcg — the doses that are actually sold — with sleep efficiency and minutes of slow-wave sleep as the primary outcomes. Anything short of that leaves the position exactly where 1980 left it.

Regulatory status: withdrawn by the nominator, not cleared by the agency

Emideltide, the international nonproprietary name for DSIP, appears on FDA's page Certain Bulk Drug Substances for Use in Compounding That May Present Significant Safety Risks, in the table headed "Bulk drug substances nominated but withdrawn". It is one of seventeen entries there, alongside BPC-157, TB-500, KPV, Semax, Selank acetate, Thymosin alpha-1, Epitalon, MOTS-C, Melanotan II, CJC-1295, AOD-9604 and GHK-Cu.

FDA's stated reason, word for word: compounded drugs containing emideltide "may pose risk for immunogenicity for certain routes of administration and may have complexities with regard to peptide-related impurities and API characterization. FDA has not identified safety-related information regarding emideltide for the proposed route... Therefore, the agency lacks sufficient information to know whether the drug would cause harm if administered to humans."

Statement

Official disclosure

FDA — Certain Bulk Drug Substances for Use in Compounding That May Present Significant Safety Risks; table: Bulk drug substances nominated but withdrawn

“Emideltide (DSIP): Compounded drugs containing emideltide may pose risk for immunogenicity for certain routes of administration and may have complexities with regard to peptide-related impurities and API characterization. FDA has not identified safety-related information regarding emideltide for the proposed route of administration. Therefore, the agency lacks sufficient information to know whether the drug would cause harm if administered to humans.”

Aug 4, 2026 · fda.gov

2026-08-04 08:29w_ww3yt0x9 · #6960d5a315ee

What that status means, precisely:

• It is not an approval and it is not a ban. It is the record of a request. Somebody asked FDA to allow emideltide as a bulk substance for pharmacy compounding; FDA put it in category 2 — substances that may present significant safety risks; the person who asked then withdrew the request. FDA's own words on the page are that these substances were "previously in category 2 of the interim policies" and "were withdrawn by the nominators."
• The practical consequence is that emideltide is not on the 503A or 503B bulks lists. A compounding pharmacy cannot lawfully compound it as a bulk drug substance. What is offered for sale is research-use-only chemical, not medicine, and it comes with no assurance of identity, purity or sterility.
• FDA's specific worries are, in order: clumping and whether the immune system would react to it by certain routes, peptide-related impurities, characterisation of the active ingredient, and — the operative one — an absence of safety information for the route being proposed.
• No marketing authorisation for DSIP exists in any country, and none has ever been applied for.

A 2026 review of therapeutic peptides in orthopaedics lists DSIP among the recovery-oriented compounds and states the position for the whole class in one line: the preclinical work looks good, and there are no clinical trials.

Where that leaves the compound

A confirmed nine-residue sequence.

Half gone from the blood in two to four minutes.

No gene, no precursor, no receptor, after fifty years of people looking for all three.

A sleep record whose positive findings are concentrated in the laboratory that discovered it, and whose independent replications are null.

A withdrawal literature covering several hundred patients with no control group in any of it, in a syndrome that ends on its own.

One seven-patient pain pilot, never followed up in forty-two years.

A growth hormone claim that holds in rats and fails in the only eight humans it was ever tested in.

Zero registered trials, ever.

Thirty-five first-hand reports splitting eighteen good, ten bad, five nothing.

And a regulatory file recording a withdrawn request and an agency saying it does not have enough information to know whether the substance would hurt you.

The peptide is named for the one effect its evidence supports least.

The sibling objects for this page, each one inspectable on its own terms: