Updated November 5, 2026: Odilia Lu, the study’s first author, commented to Behavioral Wire after publication.
Five laboratories gave mice the same dose of psilocybin and ran the same behavioral tests. Three acute effects reproduced in all five: more head twitches, less time in the center of an open field, and less freezing when a fear memory was retrieved, each within minutes of the injection. Nothing measured a day or more later reproduced consistently, including the antidepressant-like tests. In the pooled data, the persistent results that reached significance were two, both in mice given corticosterone: less plus-maze distance traveled, and higher sucrose preference that the authors themselves discount. For anyone weighing a single-lab mouse result, what traveled between labs was acute, and on the authors’ simulations a lone persistent positive is a reason to repeat the experiment elsewhere.
Odilia Lu and colleagues report the work in Nature Neuroscience, published online November 2, 2026, with Robert Malenka as last author. The labs, at Stanford, Berkeley and UCSF, agreed on dose (2 mg/kg, injected intraperitoneally), strain (C57BL/6J), age, both sexes and protocols, and left apparatus and time of day free to vary. It is a coordinated replication, not five independent attempts: the labs shared a drug supply, a mouse vendor and analysis scripts. Mice were randomly assigned and manual scoring was blinded. Each experiment used about 200 mice, roughly 10 per group per lab. The paper reports no sample-size calculation and describes no preregistration.
“People should approach single-laboratory reports of persistent behavioral effects with both openness and skepticism. We should still be open to the possibility that these drugs may exert profound, long-lasting behavioral changes. But, we should be skeptical and critically evaluate the rigor of studies, including their experimental design and controls. We should also avoid overinterpreting the findings or presenting psychedelics as a blanket cure-all. Claims about their therapeutic potential should be proportional to the strength of the evidence.”
— Odilia Lu, first author of the study, in comments to Behavioral Wire
What held across labs
Open-field avoidance was the cleanest result: center time fell in all five labs and by 3.0 percentage points pooled (95 percent confidence interval 2.0 to 4.0). Fear retrieval gave the same answer, and freezing to the first cue the next day was high in both groups, so psilocybin suppressed the expression of fear without impairing the memory. Head twitches, the rodent proxy for hallucinogenic activity, showed the drug was active in every setting.
| Assay and timing | Labs with a significant effect | Pooled result |
|---|---|---|
| Head twitch, first 20 min | 5 of 5 | Increased |
| Open-field center time, about 10 min | 5 of 5 | Reduced |
| Fear expression at retrieval, about 5 min | 5 of 5 | Reduced |
| Plus-maze open arms, about 10 min | 3 of 5 | Reduced |
| Novel-object exploration, about 10 min | 3 of 5 | Reduced |
| Open field, plus maze, novel object, 24 h | 0, 1 and 1 of 5 | None |
| Tail suspension, 24 h | 0 of 5 | None |
| Repeated forced swim, day 3 | 1 of 5 | None |
| Repeated tail suspension, day 3 | 1 of 5, higher immobility | None |
| Fear extinction, 24 and 48 h | 1 of 5 | None |
| Social preference, 24 h | 0 of 5 | None |
| Social reward learning, adult mice | 0 of 5 | None |
| Chronic corticosterone: sucrose preference | 2 of 5 | Increased, discounted by the authors |
| Chronic corticosterone: forced swim | 2 of 5, opposite directions | None |
Counting labs is a summary, not the paper’s own diagnostic. Source: Lu et al., Nature Neuroscience, Figures 1 through 8 and Supplementary Table 2.
Read down the table and the results split by timing: every acute effect was significant in at least three labs, and nothing measured later was significant in more than two. The plus-maze and novel-object tests are the weaker acute rows. The plus-maze misses were the labs with the lowest baseline open-arm time, one of them moving the opposite way, and novel-object results ran from significant decreases to nonsignificant increases.
How much the labs differed
Baselines differed widely. Saline-treated mice spent about 100 seconds immobile on day 1 of the repeated forced swim test in one lab and nearly 190 in another. The paper handles this by expressing each lab’s psilocybin data in standard deviations of that lab’s own saline controls, and on that scale the authors call the open-field effect consistent across labs. It reports no formal lab-by-treatment test, and its pooled models give each lab a random intercept only, so consistency rests on overlapping confidence intervals. Method variables such as lighting and time of day did not significantly change psilocybin’s effect, though the authors caution that methods were not sampled evenly across labs.
The antidepressant-like tests show the problem most clearly. In the repeated forced swim design, mice swim on day 1, receive the drug on day 2 and swim again on day 3, and the rise in immobility is the design’s stress induction. That rise reached significance in only two of five labs. Psilocybin reduced day-3 immobility in one, Stanford, which also had the lowest agreement between video scorers (0.47, against 0.56 to 0.87 elsewhere). In a repeated tail suspension version the stress induction worked in all five labs, but immobility was higher after psilocybin in four, significantly at Stanford (about 17 seconds, P = 0.04, unadjusted). That direction is the opposite of antidepressant-like, and the pooled difference was not significant (about 5 seconds, interval -0.6 to 9.8).
Did a stress model rescue a persistent effect?
The consortium tested the objection that unstressed mice are the wrong preparation by giving mice corticosterone for 21 days before psilocybin. The stress itself worked unevenly across labs. Psilocybin produced no persistent anxiolytic effect and no pooled forced swim effect. Sucrose preference rose in two labs and pooled, but the authors discount it: three labs saw no increase, and the rise appeared to be driven by reduced water consumption (P = 0.07, short of significance) rather than more sucrose drunk. The one other pooled result was a drop in plus-maze distance traveled, which sits alongside isolated anxiogenic-like effects in single labs. The stress model did not rescue a persistent effect.
What the simulation shows, and what it leaves open
The authors resampled each lab’s standardized effects 10,000 times and counted how often a repeat experiment would reach P < 0.05 for the main drug effect. That measures the chance of a significant result, not the chance of reaching the correct conclusion. For the acute effects it was high. For the persistent effects, which were small (about one standard deviation of saline behavior or less), it never exceeded 60 percent even at 40 animals per group. Their count was six isolated positives in close to 100 independent experiments, about the five that a 5 percent false-positive rate predicts, and they conclude that persistent effects are small or nonexistent at this dose.
“Our simulations suggest that acute effects were more reproducible across labs because they had larger effect sizes, while persistent effects were less consistent because their effect size was smaller. If the persistent effect sizes are truly this small under the conditions we used, I think we first need to identify the experimental conditions that might enhance or reliably produce them before building a large body of mechanistic research around these supposed behavioral effects.”
— Odilia Lu, first author of the study, in comments to Behavioral Wire
Small is not absent, and one harmonized dose cannot test other conditions. The paper’s discussion cites earlier work in which fear-extinction facilitation peaked at 1 mg/kg and was absent at 2 mg/kg, social reward learning was reopened by 0.3 mg/kg given two weeks before conditioning, and anxiolytic effects appeared at 1.5 and 3 mg/kg in the novelty-suppressed feeding test. Here, adults given 2 mg/kg four days before conditioning showed no social reward learning in any lab. This design is the right one for asking whether a result travels between labs and the wrong one for asking whether it exists at another dose or time point.
Limits
It used one dose, strain and route and mostly one follow-up point, and tested behavior, not mechanism. It reports no positive-control antidepressant, so a null cannot be told apart from an assay that would miss an effective one. The authors note that the tail suspension and forced swim tests have some predictive validity for drugs that act like standard antidepressants and limited predictive power for novel ones. They float one human implication, that suppressed fear expression might matter for exposure therapy if it translates; nothing here tests it.
Malenka advises MapLight Therapeutics, Definium Therapeutics and Bayshore Global Management, and Boris Heifets advises Journey Clinical and Osmind and consults for Arcadia Medicine, Tactogen and Vida Ventures, per the paper’s competing-interests statement. The other authors declare none. The work was funded by an anonymous philanthropic donor and an NIH grant to Heifets and Malenka.
“I would encourage labs that have reported persistent therapeutic effects of psilocybin to collaborate with other labs to directly test the reproducibility of their findings using their original experimental conditions. Published methods sections do not always provide enough detail to fully reproduce an experiment. … Our best starting point is to work with laboratories that have already reported positive findings and systematically investigate the conditions under which those effects emerge.”
— Odilia Lu, first author of the study, in comments to Behavioral Wire
At 2 mg/kg, five coordinated labs reproduced psilocybin’s acute effects and no persistent one; smaller or differently dosed effects remain open.