For most of a decade the effort to separate what psychedelics do for depression from what they do to perception has been a chemistry problem. Build an analog that engages the serotonin 2A receptor the useful way and not the other way, and the drug can be prescribed without a supervised dosing room.
A paper published October 23 in Nature Neuroscience proposes a second route, and it comes from one of the labs that spent years on the first. Keep the existing drug. Change when the doses arrive.
The group, led by Sheng Wang at the Shanghai Institute of Biochemistry and Cell Biology and Jianjun Cheng at ShanghaiTech, reports that giving mice 20 very small LSD doses inside a single hour suppressed the head-twitch response, the behavior that normally signals a psychedelic effect in rodents, while an antidepressant-like effect persisted for at least a week. In three macaques, a comparable intravenous schedule gave partial protection against the motor impairment a larger LSD dose otherwise produces.
The result is interesting for a reason the summary tends to flatten: the two best-known psychedelics did not behave the same way, and the suppression of the psychedelic behavior looks like desensitization, the oldest known limitation of these drugs, put to deliberate use.
What the animals received
The regimens need stating precisely, because the shorthand everyone will reach for is wrong.
Mice given the LSD schedule received 0.00375 mg/kg, injected 20 times at three-minute intervals. The whole sequence runs about an hour and delivers a cumulative 0.075 mg/kg. Mice given the psilocybin schedule received 0.2 mg/kg, injected 10 times at 30-minute intervals, roughly five hours, cumulative 2 mg/kg. Where a later challenge dose was used, it was 0.15 mg/kg LSD for the behavioral assay, 0.015 mg/kg LSD for the antidepressant assays, or 2 mg/kg psilocybin. The macaques received 0.003 or 0.006 mg/kg LSD intravenously, 10 or 20 times at 15-minute intervals, before a 0.06 mg/kg challenge.
This is not microdosing. Microdosing means an intermittent sub-perceptual dose taken every few days over weeks. What this paper describes is within-session repeated low-dose loading: a dense burst of injections inside one sitting, designed so receptor occupancy accumulates toward the level a single large dose would reach without ever producing that dose’s peak concentration. The paper’s own framing is that LSD’s slow dissociation from the receptor lets the schedule approach “the receptor occupancy achieved by an acute high dose while avoiding its high peak concentration.”
The distinction matters beyond pedantry. Twenty injections in an hour is a pharmacology experiment, not a protocol anyone could carry into a clinic.
LSD and psilocybin parted company
Mean seconds immobile in the tail-suspension test seven days after treatment, in corticosterone-exposed male mice, n = 8 per group. Individual mice varied widely, from 0 to 197.5 seconds across the groups shown. Tail suspension screens for antidepressant-like activity and is not a model of human depression. No single 0.075 mg/kg LSD arm was tested, so single-dose and repeated-schedule bars are not dose-equivalent comparisons. Doses are mg/kg. ×20 is 0.00375 mg/kg LSD injected 20 times and ×10 is 0.2 mg/kg psilocybin injected 10 times. Each compound had its own saline-control group. Values are rounded to 0.1 second.
Source: He et al., Nature Neuroscience, 2026, deposited source data for Fig. 5, seven-day panels; group means calculated from the individual-animal values. Exact statistical comparisons are given in the text.
The paper’s deposited source data, which carries the group-level numbers behind the figures, shows a split the abstract compresses into a single clause.
In the corticosterone model of depression-like behavior, with eight male mice per group, the repeated low-dose LSD schedule given alone, with no later challenge, cut immobility in the tail-suspension test from 112.4 seconds in saline-treated animals to 39.9 seconds at 24 hours, close to a single conventional 0.015 mg/kg dose at 23.6 seconds. The chart shows the seven-day result, when the repeated schedule and the single conventional dose were again similar. A single injection of the schedule’s size did nothing at either time. Sucrose preference, a measure of anhedonia, moved the same way, from 47.5 percent to 63.5 percent at 24 hours and from 53.4 percent to 77.7 percent at seven days. A second depression model, chronic unpredictable mild stress, showed a similar pattern in the extended data.
Psilocybin did not reproduce that. Its repeated low-dose schedule, given alone, left immobility at 114.9 seconds at 24 hours against 120.3 in controls, and at seven days it sits with the controls in the chart. The effect appeared only once a high dose followed, whether that dose was 2 mg/kg psilocybin or 0.015 mg/kg LSD.
The statistics support the split. Each arm is compared with the corticosterone-exposed control group, and the figures print exact P values. Repeated low-dose LSD differed from those controls at both time points (P < 0.0001 at 24 hours, P = 0.0002 at seven days), and a single injection of the same size did not (P = 0.7882 and P = 0.8968). Repeated low-dose psilocybin alone did not differ from controls (P = 0.9982 at 24 hours, P > 0.9999 at seven days), and each version that ended with a high dose did (P < 0.0001). All mice were male, and the authors report randomized treatment and blinded analysis. The companion extended-data panels state ordinary one-way ANOVA; the post hoc procedure behind the P values is not stated in the material available.
The authors attribute the difference to binding kinetics. In their receptor-binding assays LSD stayed bound after washout and psilocin, the active form of psilocybin, did not, so successive small LSD doses stack into sustained occupancy while the same psilocybin schedule never accumulates and the later large dose has to do the work.
That is a coherent mechanistic story, and it is also a warning against treating the finding as a class effect. One of these two compounds supports the strategy on its own. The other needs the very exposure the strategy was built to avoid.
One design gap follows from the doses. The repeated LSD schedule delivers a cumulative 0.075 mg/kg, five times the 0.015 mg/kg single dose used as the conventional comparator. The single-dose arms are 0.015 mg/kg and 0.00375 mg/kg, the size of one injection. No arm in the figures, extended data or source data gives the 0.075 mg/kg total as a single injection, so the antidepressant-like data cannot separate the effect of the schedule from the effect of a larger total exposure. The head-twitch comparison asks a different question: animals pretreated with the schedule responded less to a 0.15 mg/kg challenge than saline-pretreated animals did. That shows pretreatment changes the response to a later dose. It does not show the schedule outperforms an equal single dose.
The head-twitch suppression looks like desensitization
The rodent readout for psychedelic-like activity is the head-twitch response, a brief head shake that tracks activation of the 5-HT2A receptor, and repeated low-dose pretreatment suppressed it. The paper’s own word for this is desensitization.
Extended Data Figure 6 is titled “Desensitization of HTR by repeated low-dose psychedelics is independent of β-arrestin.” β-arrestin is a protein recruited by the activated receptor that drives a signaling branch separate from the one that triggers calcium release. The suppression still occurred in mice lacking β-arrestin 1 and in mice lacking β-arrestin 2, with four animals per group.
The receptor measurement is thinner than it first reads. Maximum binding of a radiolabeled ligand in frontal cortex was 59.75 ± 13.13 in saline-treated mice and 51.38 ± 3.74 after the repeated schedule (mean ± SEM, three mice per group). The authors call the levels comparable and report no test. With three animals and a spread that large, the measurement cannot rule out a meaningful change in receptor number, and it says nothing about whether the receptors still signal. The supportable reading is that no loss was detected.
Tolerance to LSD in people was described in the 1950s and has generally been filed under limitations, the reason repeated doses do less than the first. What this paper proposes is to use it: desensitize the head-twitch response, then collect an antidepressant-like effect that persists.
Whether the two readouts travel on different pathways rests on the authors’ knockout mice. The seven-day antidepressant-like effects of LSD and of the group’s engineered compound were not detectable in mice lacking β-arrestin 2, and the repeated LSD schedule had no antidepressant-like effect in mice lacking the receptor itself. Those are germline knockouts, with the gene absent in every tissue from birth, so they indicate the pathway is required in these assays. They do not establish that β-arrestin signaling is sufficient to produce an antidepressant effect, and the paper does not claim it.
What three macaques can and cannot tell you
Beam-walking speed in cm/s at the first post-challenge test, with each animal’s baseline in parentheses. Individual animals, n = 3 (two male, one female), not averaged. LSD doses are mg/kg per injection, ×10 and ×20 are the number of injections, and each LSD schedule was followed by a 0.06 mg/kg LSD challenge.
| Animal | Saline ×20 | LSD 0.003 ×10 | LSD 0.006 ×10 | LSD 0.006 ×20 |
|---|---|---|---|---|
| Male 1 | 0 (78.3) | 0 (84.4) | 0 (80.7) | 0 (132.4) |
| Male 2 | 0 (90.3) | 0 (99.4) | 0 (82.9) | 8.1 (65.3) |
| Female | 0 (88.3) | 0 (58.4) | 17.1 (100.0) | 65.6 (114.1) |
Food-tracing correct rate (percent) at the first post-challenge test, baseline in parentheses. Males only, n = 2.
| Animal | Saline ×20 | LSD 0.003 ×10 | LSD 0.006 ×10 | LSD 0.006 ×20 |
|---|---|---|---|---|
| Male 1 | 5.4 (100) | 30 (85.3) | 0 (86.7) | 100 (84) |
| Male 2 | 0 (96.7) | 16.1 (90) | 23.3 (92.6) | 72.5 (94.4) |
Source: He et al., Nature Neuroscience, 2026, deposited source data for Fig. 4e and 4f and Extended Data Fig. 8b and 8c.
The primate work is the part most likely to be misread, which is why the tables keep every animal separate.
Three cynomolgus macaques, two male and one female, were tested on beam-walking speed and crossings, and the two males on a food-tracing accuracy task. In single-dose experiments, LSD at 0.06 mg/kg and psilocybin at 1 mg/kg impaired beam walking, and the 5-HT2A antagonist MDL100907 attenuated those deficits, the word the extended-data legend uses. The schedules were then tested ahead of the 0.06 mg/kg LSD challenge.
Ten injections did not protect. At the first post-challenge test, beam speed and crossings were zero in all three animals after 0.003 mg/kg ten times and in two of three after 0.006 mg/kg ten times, and food tracing fell well below baseline. Twenty injections of 0.006 mg/kg, a cumulative 0.12 mg/kg or twice the challenge dose, did better, but unevenly: the female kept walking, one male stayed at zero and the other was close to it. All three were walking again at 30 minutes. Food tracing stayed at 100 and 72.5 percent correct in the two males tested.
That is partial protection, from one schedule, carried mostly by one animal on the beam. Three animals cannot say how often the schedule would protect, or whether the differences among them reflect the schedule or the individual.
The abstract summarizes the macaque result as “no detectable psychedelic-associated behavioral disruption.” The title of the figure says the schedule “attenuates” the responses, and the animal-level values support the narrower word.
Every one of those endpoints measures motor or cognitive performance. None measures subjective experience. Normal performance on a beam does not establish the absence of a psychedelic subjective state, in a macaque or in anyone else. An animal can cross a beam at normal speed and still be perceiving something unusual, and no behavioral assay available in a monkey can distinguish those cases. The supportable version of the primate finding is that one schedule partly limited measurable impairment in a three-animal study, which is a safety-relevant observation and not a demonstration that the experience was gone.
The mouse head-twitch assay carries the same boundary. It is the field’s standard proxy for psychedelic activity because it tracks potency well across compounds, not because it reports what the animal perceives.
One group, two strategies
In January 2022 Wang and Cheng were among the authors of a Science paper on structure-based discovery of non-hallucinogenic psychedelic analogs. That paper’s approach was to design molecules that keep the therapeutic signaling and drop the rest.
It would be tidy to call the new paper a pivot from molecules to schedules. It is not. Its mechanistic spine still runs through engineered compounds, including IHCH-8134 and IHCH-8152, used to establish that β-arrestin-biased activation drives the sustained effect, and the schedule is built on that framework and applied to LSD and psilocybin. The second route uses existing compounds rather than a new chemical entity, and the regimen itself is unproven in people.
Several relationships bear on how to weigh the work. Eight of the authors, including Wang and Cheng, disclose patent filings through their universities “around the new 5-HT2AR agonists.” The declaration does not say whether the filings extend to the schedule. The macaque experiments were run in cooperation with Prisys Biotechnologies, a Shanghai company whose animal-care committee approved the primate protocol, and one author is affiliated with it. Funding came from Chinese government science agencies and academy programs.
No registered human trial tests this
A search of ClinicalTrials.gov on October 25, 2026, for LSD and psilocybin studies using terms for repeated, low-dose, pretreatment, priming and tolerance returned conventional single-dose studies, LSD microdosing studies and multiple-dose low-dose psilocybin studies. None of the records reviewed describes within-session loading followed by a high-dose challenge. One psilocybin record mentions “priming and repeat low-dose” dosing without a schedule that can be compared. A registry search covers only registered work.
The translational problem is not only that the work is preclinical. It is that the regimen has no obvious human form. Twenty intraperitoneal injections at three-minute intervals is not a dosing schedule anyone can translate directly, and the macaque version required intravenous access and a five-hour session. Any human test would need a formulation or infusion protocol that reproduces the occupancy curve, and no such protocol appears in the paper or the registry records reviewed.
Nor should the result be read as evidence that people could obtain antidepressant benefit without a psychedelic experience. The mouse assays are tail suspension and sucrose preference, which are screens for antidepressant-like activity, not models of human depression; plenty of compounds that pass them fail in people. The behavior that was suppressed is head-twitch, which is a proxy. The primate evidence is performance. None of those links subjective experience to therapeutic outcome in either direction.
That link is the field’s open question. A decade of trials has not resolved whether the subjective experience causes the clinical benefit, correlates with it, or merely accompanies it. A paper showing that two readouts can be pulled apart in a mouse does not answer it, though it does supply a tool for asking.
What this work changes is narrower and still worth noting. Tolerance has been treated as the thing that limits how often a psychedelic can be given. Here desensitization is being used on purpose, and the evidence that it operates through a mechanism distinct from the one carrying the antidepressant effect comes from the authors’ own knockouts in mice. If that separation survives in people, the design question stops being which molecule to build and starts being how to schedule one that already exists.