Emraclidine's M4 bet ended in a 698-patient extension nobody had a reason to finish

What was tried

Emraclidine is an orally available, brain-penetrant positive allosteric modulator of the M4 muscarinic acetylcholine receptor. It was discovered at Pfizer as PF-06852231, carried into Cerevel Therapeutics as CVL-231, and became an AbbVie asset when AbbVie acquired Cerevel. ChEMBL lists it as a small molecule, CHEMBL5314557, with a maximum recorded phase of 2, and its discovery campaign was published in 2024 (doi:10.1021/acs.jmedchem.4c00293).

NCT05443724 was a 52-week, Phase 2, open-label, single-group AbbVie trial measuring long-term safety and tolerability of emraclidine 30 mg once daily in adults with schizophrenia. It enrolled 698 participants and ran from 1 September 2022 to 25 June 2025. Three populations fed into it: 97 rolling over from active emraclidine in the two double-blind parent trials, 49 rolling over from placebo, and 552 enrolled de novo. The primary outcome was the count of treatment-emergent and serious adverse events, not symptom change. The registry lists the stop reason as business reasons rather than safety.

The biological hypothesis

Every approved antipsychotic before 2024 acted at the dopamine D2 receptor or, in the case of xanomeline-trospium, through broad muscarinic agonism. The M4 receptor sits upstream of striatal dopamine release, so selective M4 modulation should dampen the dopaminergic overactivity associated with psychosis without occupying D2 and without the motor and prolactin liabilities that follow D2 blockade. Positive allosteric modulation adds a second argument: it amplifies endogenous acetylcholine tone rather than imposing constant orthosteric drive, which should reduce peripheral cholinergic side effects.

Open Targets scores the CHRM4 and schizophrenia association at 0.7987, with a clinical component of 1.0, an animal model component of 0.959 and a genetic association component of 0.749. That profile describes a target resting on pharmacology and model systems rather than on human genetic causality. Open Targets lists ten CHRM4 drugs, of which only xanomeline reached Phase 3 and none of the others targeted psychosis. The 2022 Lancet Phase 1b report and its accompanying comment set the expectation that selective M4 action could reproduce the xanomeline effect without the peripheral burden (doi:10.1016/S0140-6736(22)01990-0 and doi:10.1016/S0140-6736(22)02421-7).

What actually happened

The extension did not fail. The trials feeding it did. EMPOWER-1 (NCT05227690, N=385) and EMPOWER-2 (NCT05227703, N=391) were six-week, double-blind, placebo-controlled trials in acute exacerbation of schizophrenia, and both missed their primary endpoint of change from baseline in PANSS total score at week 6.

In EMPOWER-1 the least-squares mean change was -13.5 on placebo, -14.7 on emraclidine 10 mg and -16.5 on emraclidine 30 mg. The difference from placebo was -1.2 (95% CI -5.6 to 3.3, p = 0.6007) at 10 mg and -3.0 (95% CI -7.5 to 1.4, p = 0.1765) at 30 mg. In EMPOWER-2 the change was -16.1 on placebo, -18.5 at 15 mg and -14.2 at 30 mg, giving differences of -2.4 (95% CI -7.0 to 2.1, p = 0.2925) and +1.9 (95% CI -2.5 to 6.4, p = 0.3914). The 30 mg dose was numerically better than placebo in one trial and numerically worse in the other. All values are from the posted results on ClinicalTrials.gov and match AbbVie's 11 November 2024 release.

Safety in the extension was unremarkable. Among the 97 active rollover participants there was 1 death, 7 with serious adverse events and 24 with other adverse events at the reporting threshold. Discontinuation was driven by withdrawal of consent (30, 14 and 96 across the three groups) and by the sponsor stopping the study (4, 4 and 200), not by toxicity: adverse events accounted for 8, 2 and 45 withdrawals. Only 170 of 698 participants completed.

Failure mechanism, best guess

The most economical reading is that emraclidine at the doses tested did not produce enough M4-mediated pharmacology in the human striatum to move PANSS, and that the Phase 1b signal was a small-sample estimate rather than a stable effect. Two features of the readouts support this. First, the placebo response was large in both trials, from -13.5 to -16.1 points, which compresses the space available for a drug effect and is a recurring problem in acute schizophrenia trials. Second, the dose ordering was inconsistent: 30 mg beat placebo in EMPOWER-1 and lost to it in EMPOWER-2, which is the pattern expected from noise rather than from a dose-dependent mechanism operating below its effective exposure.

A competing explanation is that selective M4 modulation is insufficient. Xanomeline is a non-selective M1 and M4 agonist, and reviews published while emraclidine was in Phase 2 argued that its effect may require combined M1 and M4 action (doi:10.1093/ijnp/pyaf015). Nothing in the emraclidine dataset separates these accounts, because no target engagement measure was reported alongside the efficacy endpoints. That is the informative gap.

How to prevent this next time

The endpoint-level data here support a small number of concrete design levers, and no formal Bayesian update, because there is no shared prior distribution across the Phase 1b and Phase 2 populations that could be defended from public data.

Measure receptor occupancy in the trial that decides the mechanism. An M4 tracer exists and has been characterised in non-human primates, [11C]MK-6884 (doi:10.1177/0271678X241238820). An occupancy substudy inside EMPOWER at each dose would have separated insufficient exposure from insufficient mechanism, and that distinction determines whether the M4 hypothesis dies or is redosed.

Set a base rate for the placebo arm before sizing. Both trials landed in the range where a 3-point drug effect is invisible at these sample sizes, and the effect that emraclidine produced at its best was 3.0 points.

Test the selectivity assumption directly. A single arm using an M1 and M4 comparator inside the same protocol would have told the field whether the loss came from selectivity or from potency, at the cost of one additional arm.

The single highest leverage change would have been an M4 receptor occupancy substudy embedded in EMPOWER at each dose level, so that the failure could be attributed either to exposure or to the mechanism itself.

What this means for similar programs

CHRM4 now carries a Claidex Mechanism Risk Score of 38 (yellow), driven by a saturation term of 13.34 across 11 distinct programs and a phase burden of 8.05 from one Phase 2 failure. The band warns about crowding rather than biology: the genetic deficit term is only 3.02, because Open Targets still scores the CHRM4 and schizophrenia link at 0.7987.

The practical read for other muscarinic programs is that approving a non-selective agonist did not validate selective M4 modulation, and that any program claiming a cleaner xanomeline should show occupancy and effect in the same study. Dual M1 and M4 programs are not directly contradicted here. M4-only programs now carry the burden of explaining why their exposure differs.

Open questions

What striatal M4 occupancy did emraclidine achieve at 10, 15 and 30 mg, and was any dose near saturation?

Would a longer treatment period than six weeks have produced separation, given that the extension ran 52 weeks without a controlled efficacy comparison?

Do the 552 de novo participants in the extension show a symptom trajectory that differs from the rollover groups?

Is the residual M4 hypothesis testable in a population enriched for cholinergic deficit rather than in unselected acute exacerbation?

Sources

1. ClinicalTrials.gov, NCT05443724, emraclidine 52-week open-label extension, including posted results. Accessed 26 August 2026. https://clinicaltrials.gov/study/NCT05443724 ClinicalTrials.gov, NCT05227690 (EMPOWER-1) and NCT05227703 (EMPOWER-2), posted results. Accessed 26 August 2026. AbbVie, "AbbVie Provides Update on Phase 2 Results for Emraclidine in Schizophrenia," 11 November 2024. https://news.abbvie.com/2024-11-11-AbbVie-Provides-Update-on-Phase-2-Results-for-Emraclidine-in-Schizophrenia Open Targets Platform, CHRM4 (ENSG00000180720), association with schizophrenia (MONDO_0005090), tractability and known drugs. Accessed 26 August 2026. ChEMBL, EMRACLIDINE, CHEMBL5314557. Accessed 26 August 2026. Krystal JH et al. Emraclidine, a novel positive allosteric modulator of cholinergic M4 receptors, for the treatment of schizophrenia. Lancet, 2022.Nair PC, Bastiampillai T. A new cholinergic mechanism for antipsychotics: emraclidine and M4 muscarinic receptors. Lancet, 2022.Davoren JE et al. Design and Synthesis of Clinical Candidate PF-06852231 (CVL-231). J Med Chem, 2024.Yohn SE et al. From theory to therapy: unlocking the potential of muscarinic receptor activation in schizophrenia. Int J Neuropsychopharmacol, 2025.Tong J et al. PET imaging of M4 muscarinic acetylcholine receptors in rhesus macaques using [11C]MK-6884. J Cereb Blood Flow Metab, 2024.openFDA FAERS drug event endpoint, queries for EMRACLIDINE and COBENFY. Accessed 26 August 2026.