A single 25 mg dose of psilocybin lowered depression scores by 7.27 points more than an active placebo by day 8 in a randomized trial at the Northern Stockholm Psychiatric Clinic, and the separation widened to 11.03 points by day 15. The same trial scanned its 35 participants before dosing and 34 of them again afterward, looking for the biological change that the field’s standard explanation says should be there.

It did not find it.

The explanation under test is specific enough to be falsifiable, which is why the test is worth reading closely. Depression has been associated with lower synaptic density; in animals, psychedelics increase dendritic spine density within 24 hours and ketamine is reported to rescue lost spines; therefore a single dose might work by rebuilding synapses. The Karolinska group went after the presynaptic half of that claim in an imaging sub-study of its own trial, posted to medRxiv on November 8, and came back with four flat numbers. It registered the analysis plan in November 2023, after scanning ended and before the imaging data were analyzed. The parent trial is peer reviewed, published in JAMA Network Open in May. The imaging analysis is not; it is a preprint, and should be read as one.

What the tracer reads

[11C]UCB-J binds synaptic vesicle glycoprotein 2A, a protein carried in synaptic vesicles and concentrated in presynaptic nerve terminals. PET using it yields an index of presynaptic terminal density. It is not a count of synapses, and it is blind to the postsynaptic compartment, which is where most of the preclinical work on psychedelic and ketamine plasticity lives. Yale’s own group said so in print: the tracer “binds solely to SV2A, localized to pre-synaptic nerve terminals, such that we cannot detect changes in the post-synaptic compartment.”

That boundary matters before any of the numbers do, because it decides what a null result can mean.

The Stockholm numbers

Participants were randomized 1:1 to 25 mg psilocybin or 100 mg niacin, both arms receiving five psychotherapeutic support sessions over 17 days. Scans ran 2 to 7 days before dosing and a target 15 days after. In the PET sample, depression scores fell from 25.9 to 13.2 on psilocybin and from 26.2 to 24.5 on niacin.

Against that, the imaging comparison, adjusted for each participant’s own baseline scan, produced nothing. In total distribution volume, the psilocybin-versus-niacin coefficient was −0.49 in frontal cortex (95% CI −2.42 to 1.44, p = 0.61) and 0.05 in hippocampus (−1.39 to 1.50, p = 0.94). Binding potential, the tracer’s second outcome measure, gave p = 0.93 and 0.86. Adjusting for scan timing, age and clinical covariates changed nothing, and seven exploratory regions showed no treatment effect either.

The sample is small and the measurement demanding. Incomplete arterial blood sampling leaves 15 psilocybin participants in the distribution-volume analyses, and the follow-up scan, targeted at day 15, landed anywhere from day 7 to day 48 on psilocybin. Six of the 34 follow-up scans, three in each arm, fell outside the protocol’s day 8 to 22 window. The main text reports no power calculation, and the authors state the consequence directly: “confidence intervals were wide and small-to-moderate treatment effects cannot be excluded.”

That sentence is the difference between failing to detect something and establishing that it is absent. This article is about the first.

Two earlier human attempts, which are not replications

Copenhagen’s psilocybin study, published in July in Translational Psychiatry, scanned healthy people. Fifteen volunteers received 0.3 mg/kg oral psilocybin in a single-arm, open-label design with structured psychological support and no formal psychotherapy, and were scanned at baseline and one week. The pre-registration specified a one-sided test and sized the study to detect a 5 percent increase in distribution volume, with a planned interim look at 12 participants. At that interim, conditional power was 24 percent under the hypothesized effect and 2 percent under the observed one, and the study stopped for futility. The observed changes were +2.3 percent in frontal cortex and +3.0 percent in hippocampus (one-sided p = 0.29 and 0.33).

Its more interesting result was exploratory and not randomized. The first five participants were dosed inside an MRI scanner; participants 6 through 15 were dosed in a designated room with a curated music playlist. The difference in SV2A change between settings was 3.0 mL/cm³ in frontal cortex (95% CI 0.4 to 5.5, p = 0.03) and 1.9 in hippocampus (−0.3 to 3.9, p = 0.09). The room group’s own increase, around 5 to 6 percent, had confidence intervals crossing zero in both regions, so the study found a difference between contexts and not an established increase in the room group. The authors call the setting comparison preliminary and hypothesis-generating.

Yale ran the ketamine version, published in 2022: 12 patients with MDD, PTSD or both and 9 healthy controls, racemic ketamine at 0.5 mg/kg intravenously over 40 minutes, scanned at baseline and 24 hours. Mean change was 2.4 percent in patients and 0.5 percent in controls, with no significant effect in any primary region and no difference between groups. Four rhesus monkeys scanned at 24 hours and one week showed no change either.

Yale also ran a post hoc median split. In the six patients with the lowest baseline SV2A, density was 7 to 10 percent higher in three regions 24 hours later (p = 0.035 to 0.054), and a larger rise tracked a larger drop in depressive symptoms. The authors flag the problem themselves: in that subgroup, the change in the tracer’s plasma free fraction correlated with the change in distribution volume at r = 0.88 to 0.91, which raises a measurement artifact as a candidate explanation.

These three studies are converging human evidence on one tracer, not replications of each other. The drug, population, design and scan timing (about two weeks, one week, 24 hours) all differ. What they share is a method and a direction of result. Nor are the psilocybin studies fully independent: one author sits on both papers, with affiliations at both the Karolinska and Copenhagen groups. The ketamine study is from a separate Yale team.

Disclosures differ by study. The Karolinska preprint reports foundation and research-council funding, and honoraria from Janssen-Cilag to its senior author. The Copenhagen paper reports public and foundation grants; its senior author reports speaking for Cybin and consulting for Gilgamesh and other companies, and one author’s salary was supported by Compass Pathways, which the paper says had no other relation to the study. The Yale paper reports federal and foundation support and discloses industry ties among several authors, including a co-sponsored patent on intranasal ketamine for depression licensed by Janssen.

The effect on offer is in single-digit percentages

The preclinical result that set this up is itself small. A 2021 pig study from the same Copenhagen group gave 24 awake pigs either 0.08 mg/kg intravenous psilocybin or saline and measured SV2A by autoradiography: hippocampal density was 4.42 percent higher one day after dosing, and 9.24 percent higher in hippocampus and 6.10 percent higher in prefrontal cortex at seven days. Copenhagen used that work to choose its one-week follow-up scan and sized its own study to detect a 5 percent increase.

Tissue autoradiography in pigs and in vivo distribution volume in people are not the same measurement, so the pig figures do not transfer as biology. They do frame the design question. The quantity at issue sits in the single-digit percentages, and three studies went after it with one post-dose scan each, in 15 to 21 treated people. That is a demanding test, and the Karolinska intervals are wide enough to include small-to-moderate effects.

“This does not rule out synaptic plasticity following psilocybin treatment but highlights the limitations of SV2A binding as a proxy marker.”

— Yngwe et al., Karolinska Institutet, medRxiv preprint, November 8, 2026

What the Stockholm secondary finding does and does not support

The paper’s more provocative result is not about change at all. Higher baseline distribution volume predicted greater subsequent improvement on psilocybin: r = −0.63 in both regions at day 8 (pFDR = 0.026), and r = −0.61 and −0.65 at day 15 (pFDR = 0.032), the two timepoints that survived correction. Associations pointed the same way at day 42 and day 365 without surviving correction, and the niacin arm showed no consistent association.

It is tempting to call that a patient-selection biomarker, and the paper will not let you. The formal treatment-by-baseline interaction tests, the tests that would establish the relationship as specific to psilocybin and not a general prognostic signal, were nominal only, all with pFDR at or above 0.060. The authors say plainly that they cannot distinguish the two. A post hoc analysis separating responders from non-responders reached an area under the curve of 0.89 in frontal cortex, but with a bootstrap interval of 0.66 to 1.00 in a small sample. That is a hypothesis with a prospective test still owed to it.

A second result is easy to misreport. In unadjusted analyses, greater reductions in SV2A binding tracked greater improvement on psilocybin, strongest at day 42 (frontal cortex r = 0.71). Those associations attenuated once baseline binding and baseline depression score were accounted for, and none survived correction at any timepoint. Yale’s post hoc subgroup pointed the other way, with increases tracking improvement in six people. Neither result is robust, and they measure different things in different populations.

Where that leaves the mechanism claim

Between them, the authors offer a set of explanations, and none has been tested against the others. Plasticity may be happening in the postsynaptic compartment or in dendritic spines, which this tracer cannot see. The timing may be wrong, since spine changes in animals are most pronounced in the first days and each human study took one snapshot. The effect may exist only in people who start with a deficit, or only in a permissive context, as Copenhagen’s setting comparison hints. Or, as the Karolinska authors themselves raise, the antidepressant effect may run through other pathways, “including expectancy and unblinding effects.”

What the Stockholm result establishes is narrower and more useful than any of those. A randomized, placebo-controlled trial produced a large antidepressant effect, significant through day 42, in patients with recurrent major depression, and the presynaptic marker in the two regions where the hypothesis predicted movement did not move. Either the group-level increase is not there, or SV2A PET at these sample sizes and timepoints cannot resolve it. Both readings are costly for a mechanism story often told as settled, and the second is the more awkward: a mechanism claim that this tracer cannot fail cannot be tested with it.

The Karolinska group’s own conclusion is that the findings give “no support for a simple synaptic loss-restoration model” of psilocybin’s antidepressant effect. A mechanism in which psilocybin opens a window of plasticity without adding net presynaptic terminals would not predict the increase these studies looked for, and nothing here rules it out. It also would not be tested by this tracer, which means the next experiment has to measure something else.