The compounds filed under sleep, mood and social behavior are the hardest on any research shelf to read honestly. The endpoints are soft, the placebo responses are large, and the delivery route most of the human literature uses is itself contested. The last twelve months delivered a useful crop of results in exactly this space — a negative multisite trial, a reanalysis that pulled a positive signal out of a famously null one, a safety scare that arrived as a conference abstract and was answered by a survey analysis six months later, and preclinical work that quietly concedes the real problem was never the molecule.
Oxytocin's negative column got longer
The headline result is Tiouririne et al., published in Alcohol, Clinical and Experimental Research in July 2026 (PMID 42458234): a randomized, double-blind, placebo-controlled, multisite trial of intranasal oxytocin in alcohol use disorder. One hundred participants, fifty per arm, four clinical sites, twelve weeks.
The primary outcome was the weekly percentage of heavy drinking days across the ten-week maintenance phase. There was no significant between-group difference. Oxytocin was well tolerated, adverse effects were mild, and the authors note a modest numerical reduction in drinking in the oxytocin arm late in the trial — roughly weeks nine through twelve — which they frame as a reason to look at longer treatment duration and responder subgroups rather than as evidence of efficacy.
That framing is the correct one. A trend appearing in the final quarter of a trial that missed its primary endpoint is a hypothesis — it is generated by the same data that failed the test, so it cannot also serve as the test. Same structure as the subgroup signals in our immune-modulating peptides roundup: useful for designing the next study, not a result in its own right.
Alcohol use disorder now joins the list of indications where adequately powered oxytocin trials have returned null primaries.
…and a reanalysis pushed back
The field's most-cited negative is SOARS-B — Sikich et al., New England Journal of Medicine 2021;385:1462–1473 — a 24-week, multicenter, placebo-controlled trial of intranasal oxytocin in 290 children and adolescents with autism spectrum disorder, which found no significant difference from placebo on its primary social-withdrawal endpoint or on secondary social and cognitive measures.
In July 2025 a secondary analysis of that dataset was posted to medRxiv (doi 10.1101/2025.07.20.25331870), applying a data-driven "outcome learning" approach to the trial and its open-label extension. It reports that oxytocin significantly improved learned composite measures of emotional responsiveness and social-emotional reciprocity at small-to-moderate effect sizes, with the effects reappearing in the open-label phase when former placebo participants crossed over.
Three caveats travel with it, none minor. It is a preprint, not yet through peer review. It is a secondary analysis of a trial that missed its prespecified endpoints, using composites derived from the same data rather than specified in advance. And the registered endpoints remain the ones the trial was powered to test.
What it genuinely contributes is a methodological argument: that the instruments used as primary endpoints here may be too coarse to detect the effect. That claim is testable — and it becomes evidence of efficacy only when a trial is built around those composites prospectively and hits them.
The delivery question nobody has closed
Underneath every intranasal oxytocin result sits an unresolved question: how much of the administered peptide reaches central oxytocin receptors, and whether the observed effects require that it does. The proposed route is transport along perineural channels surrounding olfactory and trigeminal nerve fibers, bypassing the blood-brain barrier — which oxytocin crosses poorly. Radiolabel work in animals has found tracer in those nerves and in brain regions along their trajectories, and intranasal administration raises cerebrospinal fluid oxytocin in rhesus macaques. But a Molecular Psychiatry review (2021) summarizing the field's methodological lessons concluded that the extent of brain penetration in humans remains unclear, and that some regional imaging effects may be partly explained by the rise in systemic oxytocin the intranasal route also produces.
If the peripheral rise is doing part of the work, then route and formulation are not implementation details — they are independent variables, and trials that differ on them are not testing the same thing. That is a plausible contributor to the field's replication record, and the reason a null oxytocin trial is weaker evidence against the mechanism than it looks. Our pharmacokinetics primer covers why compartment, not just potency, decides whether a compound is testable at all.
Melatonin: a November signal, a May rebuttal, and the evidence in between
At AHA Scientific Sessions in November 2025, Nnadi et al. presented Abstract 4371606 (Circulation 2025;152 suppl 3): a TriNetX electronic-health-record cohort of 130,828 adults with insomnia, roughly half with recorded long-term melatonin use, matched on more than forty baseline variables. It reported substantially higher hazards of incident heart failure, heart-failure hospitalization and all-cause mortality over five years, and drew heavy coverage. It is a conference abstract, not a peer-reviewed paper, and the AHA's own release flagged the findings as preliminary and non-causal.
On 21 May 2026, Open Heart published an NHANES analysis framed explicitly as a reality check on that signal. In a nationally representative adult cohort (1999–2018), survey-weighted multiple-imputation models found recent melatonin use not significantly associated with prevalent composite cardiovascular disease or heart failure. Its more durable contribution is the error mechanism it names: melatonin users differ sharply from non-users at baseline on sleep disturbance and depression, leaving these associations unusually exposed to confounding by indication, reverse causality and exposure misclassification. People reach for melatonin because something is wrong with their sleep — and disturbed sleep is itself a cardiovascular risk marker. Neither study settles the question; together they are a clean worked example of why a matched EHR cohort and a weighted survey cohort can disagree without either containing an error.
Melatonin's stronger evidence sits elsewhere. A 2026 systematic review and meta-analysis in Sleep Medicine Research (Hanafi et al.) pooled 18 randomized controlled trials covering 784 patients with clinically diagnosed delayed sleep-wake phase disorder, reporting reduced sleep onset latency, an advance in dim light melatonin onset (DLMO — the standard circadian phase marker), improved sleep efficiency and reduced wakefulness after sleep onset. Heterogeneity was moderate to high, several included trials were open-label, and the authors describe the results as initial considerations rather than definitive guidance.
The instructive part is which endpoint moved. The DLMO shift is a chronobiotic effect, not a hypnotic one, and timing relative to a subject's own endogenous melatonin onset drove the result — consistent with melatonin's pharmacology at the MT1 and MT2 receptors, inconsistent with the popular framing of it as a sedative. As our receptor pharmacology primer argues, timing is a pharmacological variable in its own right, not a convenience.
One nomenclature note that recurs constantly here: melatonin is an indoleamine, not a peptide. It sits on this shelf for what it is studied for, not what it is made of.
DSIP: the engineering moved, the human file didn't
DSIP — delta sleep-inducing peptide, the nonapeptide isolated from rabbit cerebral blood in the 1970s — is where the gap between search volume and evidence is widest.
The human record is small, old and unencouraging. Monti's double-blind crossover polysomnography work in chronic insomniacs (International Journal of Clinical Pharmacology Research, 1987) concluded that sleep improvement under DSIP was of little clinical significance. Bes et al. (Neuropsychobiology, 1992), a double-blind matched-pairs study across five laboratory nights in 16 chronic insomniacs, found weak objective effects and no meaningful subjective change, concluding short-term DSIP treatment was unlikely to be of major therapeutic benefit. No modern adequately powered human trial has superseded either.
What did move is the delivery side. A 2024 open-access paper in Frontiers in Pharmacology (Mu et al., doi 10.3389/fphar.2024.1439536) describes a DSIP–CBBBP fusion peptide secreted from Pichia pastoris — DSIP joined to a blood-brain-barrier-crossing sequence — tested in a PCPA-induced insomnia mouse model, where it outperformed DSIP alone on neurotransmitter readouts.
Read that carefully: a new molecule in a rodent model, not a revision of DSIP's human record. But the design choice is the tell. When a group's response to a decades-old peptide is to bolt a transport sequence onto it, the implicit diagnosis is that the limitation was getting to the target, not what happens on arrival — the same diagnosis hanging over intranasal oxytocin, and over VIP with its roughly one-minute plasma half-life.
What this shelf looks like on a COA
Three specifics worth carrying into sourcing for this category:
Oxytocin is a cyclic, C-terminally amidated nonapeptide with an intramolecular Cys1–Cys6 disulfide. Two near-mass impurities follow: the free-acid form (one mass unit lighter than the amide) and the reduced, open-chain form (two units heavier). Both are trivially resolvable by mass spectrometry — if anyone looks. Disulfide integrity is an identity question no purity percentage answers.
DSIP carries a tryptophan at position one, so UV quantitation at 280 nm is viable for it — unlike Selank, Epithalon or KPV, which carry no tryptophan, tyrosine or cystine at all. Our piece on net peptide content covers why that changes which number on a COA means anything.
Melatonin, being a small molecule, should not be characterized with peptide-shaped tests. A peptide-mapping or amino-acid-analysis line on a melatonin COA is a template artifact, not characterization.
The through-line
Across all four compounds the pattern holds: the mechanisms are not what's in dispute — the measurements are. What remains unsettled is whether the compound reaches the relevant compartment, whether the instruments can detect the effect, and whether observational data can be cleaned of the reason people took it in the first place. That is more tractable than "the mechanism is wrong," and it is why this category keeps producing trials that read as neither positive nor conclusively negative.
It is also why the cognitive-mood label is a research question rather than a chemical class — a peptide hormone, an indoleamine, a nonapeptide of unsettled mechanism and the Russian regulatory-peptide branch from our Selank vs Semax comparison share almost nothing structurally. Sort by mechanism and evidence maturity, not by shelf.
FAQ
Does a null primary endpoint mean the compound doesn't work? It means that compound, at that exposure, in that population, on that instrument, over that duration, did not separate from placebo. Each qualifier is a place a real effect can hide — and equally, a place an explanation can be retrofitted for an inactive compound. Nulls become evidence of absence when they stack across varied designs.
Why treat a positive reanalysis more skeptically than a positive trial? Because the outcome was chosen after seeing the data. Prespecification bounds how many ways a result can be looked for; data-driven composite selection removes that bound, so conventional p-values understate the false-positive rate. A reanalysis is a proposal for the next trial, not a substitute for one.
Every result described above is a finding in the published literature or a trial-participant outcome, not a property being attributed to any material sold here. All compounds referenced are research materials. Browse the full compound library or read more on analytical methods.
This article is educational and for the laboratory research community. Trulogic Labs products are sold for laboratory and research use only and are not for human consumption.