The Phase 2b trial of BPL-003, the intranasal 5-MeO-DMT program Eli Lilly now owns, was published in JAMA on October 26. It met its primary endpoint. But the dose chosen for Phase 3, 8 mg, was not the dose that endpoint tested. The 8 mg dose was tested in the trial’s smallest arm, 46 patients, after an amendment cut that arm’s share of randomization.
That is a reasonable way to choose a dose and a thin way to prove one. The trial found no efficacy edge for 12 mg and a visible tolerability cost, which supports dropping it. It was not designed to show that 8 mg works as well as 12 mg, and the pivotal program is built to keep testing the dose. A second claim from the paper, that most patients were ready for discharge within two hours, needs the same care.
Lilly closed its acquisition of AtaiBeckley on September 11, paying $6.75 a share in cash plus one contingent value right worth up to $2.50. Up to $0.50 of that depends on US approval and DEA rescheduling of BPL-003 before the fifth anniversary of closing.
A result at the low end of the plan
The primary endpoint compared 12 mg with a 0.3 mg low-dose comparator on the change in score on the Montgomery-Åsberg Depression Rating Scale (MADRS), a clinician-rated scale from 0 to 60 on which higher means more severe depression, from baseline to day 29. Patients on 12 mg improved by 11.1 points and comparator patients by 5.5. JAMA’s estimate of the difference, from a model that adjusts for baseline severity, is 5.3 points (95 percent confidence interval 1.8 to 8.9; p = .003). By our arithmetic on the standard deviations JAMA reports, that is about half a standard deviation.
The protocol synopsis posted to ClinicalTrials.gov states what the trial was built to detect: 64 patients per primary-comparison arm gives 80 percent power at an alpha of 0.05 to detect a six-point difference, assuming a standard deviation of 12.0. Six was a planning assumption, not a pass mark, and the interval around 5.3 includes it. It is still a fact about the size of this effect, and it is not in the abstract: the result sits toward the low end of what the trial was designed to find.
The secondary comparison, 8 mg against 0.3 mg, produced a difference of 6.2 points (95 percent confidence interval 2.1 to 10.2). Day-29 response was 35 percent on 8 mg, 34 percent on 12 mg and 12 percent on the comparator; remission was 26, 18 and 8 percent.
The effect arrives quickly and holds, but it does not grow. The sponsor reported day-2 improvements of 8.8 points on 8 mg and 8.9 on 12 mg against 3.9 on the comparator. The gap was about six points at day 8, five to six at day 29 and five to six at day 57, where the sponsor’s figures were 10.8, 10.2 and 5.2 points. The comparator was not inert: it improved by 5.5 points by day 29, with 12 percent response and 8 percent remission.
The 8 mg arm: 46 patients, and why
The comparator arm had 74 patients, the 12 mg arm 73 and the 8 mg arm 46. Ninety percent of patients completed the core study, so attrition does not explain the gap.
The imbalance was designed. Randomization began at 1:1:1 and was changed to 3:1:3 for 0.3 mg, 8 mg and 12 mg in protocol version 3.0, dated May 17, 2024, about eight months after the study opened. The sponsor’s note in the registry gives the reason: to enable recruitment to complete, which cut the planned total from 225 patients to 203. Because the primary comparison was 12 mg against 0.3 mg, the cut fell on the middle dose, and the note says it left the power of that comparison intact. For each new patient, the chance of receiving 8 mg fell from one in three to one in seven. Across the whole trial the allocation averaged about 3:2:3, and the 8 mg arm finished at 46 of the 193 patients analyzed (196 were randomized), under a quarter. JAMA’s Research Summary describes the randomization as 3:2:3 and lists the smaller 8 mg group “after a randomization change” among the trial’s limitations, so the paper discloses it.
The public record does not say who approved version 3.0 or whether anyone had seen unblinded outcomes at the time. The stated reason, recruitment, is a routine one for an allocation change, and the registry’s masking description lists the sponsor among the masked parties, though the protocol’s own list names therapists and the rater instead. The protocol posted to the registry is 19 of its 135 pages, a synopsis that does not describe the decision. The process question is open, and nothing posted contradicts the recruitment reason.
What the smaller arm costs is precision. The model-based day-29 estimates are 12.0 points on 8 mg and 11.2 on 12 mg, with standard errors of 1.61 and 1.27. A rough calculation from those errors, which is ours and not a reported analysis, makes the data compatible with 8 mg doing anywhere from about three points worse to about five points better than 12 mg. “Equivalent efficacy” is the sponsor’s phrase, and a trial built around a different comparison cannot establish it. Nothing here shows that 8 mg beats 12 mg, and nothing shows that 12 mg adds nothing.
What the trial shows more clearly is a gradient on side effects. Across 0.3 mg, 8 mg and 12 mg, nausea ran at 3, 33 and 40 percent, administration-site pain at 8, 17 and 22 percent and raised blood pressure at 4, 13 and 16 percent. Panic attacks occurred in no comparator patient, one patient on 8 mg and six on 12 mg. Anxiety was less orderly, at 8, 4 and 16 percent. The registry reports counts, not tests, and several of these differences rest on single digits. Overall adverse-event rates were 73, 76 and 85 percent, 99.7 percent of events were mild or moderate, and the two serious events in the blinded period were on the comparator and 8 mg, none on 12 mg.
The sponsor’s July 1, 2025 release, nine weeks after the last follow-up visit, read this as a case for 8 mg. The two doses “demonstrated equivalent efficacy,” dose-related increases in side effects suggested 8 mg “was better tolerated,” and the results supported choosing 8 mg for Phase 3, pending consultation with regulators. Those are the sponsor’s readings, and the paper and the release come from overlapping hands. Beckley Psytech funded the trial, eight of the 15 JAMA authors list an AtaiBeckley or Beckley Psytech affiliation, and the first author, David Feifel of the Kadima Neuropsychiatry Institute, was quoted in the release. That is ordinary for an industry Phase 2b and does not impeach the data.
Dropping a dose that adds side effects without a detected gain in efficacy is ordinary development practice. The point is narrower: the randomized, blinded evidence for 8 mg in this indication is one 46-patient arm and a comparison too imprecise to separate it from 12 mg. BPL-003 also has earlier open-label studies, which this article does not assess. Phase 3 will carry more weight. By its stated 2:1:2 ratio, ReConnection-1’s roughly 350 patients would put about 140 on 8 mg, three times the Phase 2b arm.
What a subperceptual comparator can and cannot control
The 0.3 mg arm is labelled in the registry as a sub-therapeutic active placebo comparator, and JAMA calls it subperceptual, meaning chosen not to produce a noticeable effect. That is the difficulty. A comparator built to be imperceptible cannot mask a drug whose defining property is a perceptible effect, and the side-effect spread is itself a cue: nausea in 3 percent of comparator patients against 33 and 40 percent on the active doses is not something a patient needs training to notice.
The trial took the question seriously. Masking covered participants, therapists, the investigator and the MADRS rater. The protocol pre-specified, as exploratory measures, the Bang Blinding Index, which asks participants, clinicians and raters on day 2 to guess the assigned treatment, and the Stanford Expectations of Treatment Scale, which records what patients expect before treatment, along with its relationship to outcomes. Those are the right instruments. The abstract does not report their results, and neither does the registry posting or the sponsor’s topline release. JAMA’s Research Summary does list “functional unblinding was common” among the trial’s limitations.
One design choice does more work than the comparator: the registry says MADRS was rated remotely by independent, centralized raters, so the person scoring the outcome never sat in the dosing room. That helps even if participants knew what they received. The comparator also controls for nasal administration, clinic time and the preparation, supervision and integration around the dose, which is part of why it improved by 5.5 points. It cannot control for knowing you took the drug, a limit psychedelic trials share.
Both pivotal trials replace it with a placebo. That is a more conventional comparator, and it changes the number to beat: the 5.3 points here is a difference from an arm that improved by 5.5, and this trial cannot say how a true placebo arm would have behaved. A placebo produces no acute effect either, so whether participants and raters could tell what they received carries over unchanged.
Two hours, four hours, and the visit count
The access argument is why this asset exists in its current form, and it is the part of the paper most likely to be repeated inaccurately.
JAMA reports that most participants were deemed ready for discharge within two hours of dosing, 85 percent on 8 mg and 78 percent on 12 mg. The sponsor said most were ready at the first assessment, 90 minutes after the dose. The posted protocol says how readiness was measured and what followed. Readiness was assessed every 30 minutes starting 90 minutes after dosing, and the earliest discharge “will be 4 hours after receiving the dose.”
That is the only protocol version posted, version 3.2 for US sites, dated February 24, 2025, so it may not be the text every earlier participant’s site followed. JAMA’s Research Summary does describe the intervention as a single spray followed by more than four hours of observation. The two-hour figure is a readiness assessment, not an observed discharge time, and neither the abstract nor the Research Summary reports when patients left. The finding supports a two-hour operating model as plausible. It does not demonstrate one.
The authors’ comparison is split in two, and it should stay split: a shorter monitoring period than psilocybin, which their abstract names as the logistical hurdle, and fewer clinic visits than esketamine. Against esketamine the monitoring claim does not hold. The Spravato label requires monitoring for at least two hours at every treatment session, followed by a stability assessment, and the sponsor’s own July 2025 release said BPL-003 “could fit within” the existing two-hour in-clinic paradigm Spravato established. That is parity, offered as parity.
The visit count is where the difference sits. Esketamine for treatment-resistant depression is labelled at twice weekly for four weeks, weekly for four more, then every one to two weeks, each session with its own two hours of monitoring. BPL-003 was tested as a single dose. The pivotal program tests one dose, or two 14 days apart, with 8 mg retreatment every 8 or 12 weeks in the open-label extension.
A short monitoring window is not a light treatment. Each dose sat inside three preparation visits of 60 to 90 minutes, a dosing day staffed by a lead therapist, an assistant therapist and a physician in attendance or close by, and three integration visits of the same length. The protocol calls this psychological support, not psychotherapy, and the distinction is correct. It is still six scheduled therapist visits around a single administration, and JAMA notes the trial did not test the support separately.
Patients were also washed off their antidepressants, with washout complete at least two weeks before dosing. Esketamine’s label allows monotherapy or use alongside an oral antidepressant. JAMA lists the required washout among its limits on generalizability.
What Phase 3 is designed to answer
The pivotal program was disclosed in March, after what the company described as a successful end-of-Phase 2 meeting at which the FDA indicated support for advancing into Phase 3. ReConnection-1 plans roughly 350 participants on a single dose of 8 mg, 4 mg or placebo, randomized 2:1:2. ReConnection-2 plans a two-dose design on days 1 and 15, 8 mg against placebo. Both run a 12-week randomized core and a 52-week open-label extension, with primary endpoint MADRS change at week 4. Topline is guided to early 2029.
Twelve milligrams is gone from the registrational program. Four milligrams is new in ReConnection-1. The comparator is placebo. And the company’s stated purpose for ReConnection-1 includes to “further characterize dose–response relationship,” an acknowledgement that the Phase 2b did not settle it.
The 4 mg arm is the one to watch, and not only for this asset. If the acute experience weakens at 4 mg, an arm that still performs like 8 mg would bear on a question the field is circling: whether the subjective experience is doing the therapeutic work or travelling with it. The Phase 2b protocol pre-specified an exploratory comparison of a mystical-experience questionnaire with MADRS change, and those results are not in the abstract.
AtaiBeckley’s last quarterly report before the acquisition closed states that “BPL-003 Phase 3 activities in treatment-resistant depression have been initiated.” Neither ReConnection-1 nor ReConnection-2 appears on ClinicalTrials.gov as of October 27. US registration is due within 21 days of the first participant’s enrollment, so a US start would show there, though a first patient enrolled elsewhere would not.
GH Research’s inhaled version of the same molecule completed an 81-patient Phase 2 trial in treatment-resistant depression in March 2025 and has posted no results to the registry, so the two programs cannot yet be compared on effect size.
What Lilly bought is a fast-acting antidepressant signal that cleared its primary endpoint toward the low end of what the trial was built to detect, a tolerability profile that improves as the dose falls, and a delivery model whose central operational claim has so far been measured as readiness, not practice. Its Phase 3 dose was chosen after a comparison, a 46-patient arm against a 73-patient arm, too imprecise to separate it from the dose that was dropped, and the pivotal trials meant to test that choice are not yet on the registry.