Pomaglumetad, GRM2 and a genetic signal that pointed at the other receptor

What was tried

NCT01129674 was a long-term, open-label, randomised, multicentre study of LY2140023 against atypical antipsychotic standard of care in adults with DSM-IV-TR schizophrenia. It enrolled 560 participants across 81 sites from June 2010 to October 2012, flexibly dosed at 20, 40 or 80 mg. The primary outcome was time to discontinuation due to adverse events through 52 weeks, and the 17 secondaries included PANSS, CGI-S, NSA-16 and C-SSRS. Entry required completion of an acute placebo-controlled LY2140023 study, so the cohort came from the pivotal programme.

The record now lists Denovo Biopharma LLC as lead sponsor. Denovo licensed the compound from Lilly in 2015 to redevelop it as a biomarker-selected therapy. The termination reason was posted on 14 September 2026, recorded verbatim as "The decision to stop the trial was based on efficacy results in the overall schizophrenia participant population." No results have been posted. The 2026 posting is the administrative close-out of a study that stopped fourteen years earlier.

Pomaglumetad methionil is a prodrug of LY404039, a group II metabotropic glutamate agonist, held in ChEMBL as CHEMBL2105707, with the anhydrous form CHEMBL2110730 at maximum phase 3.

The biological hypothesis

Group II metabotropic glutamate receptors, mGlu2 and mGlu3, sit presynaptically and dampen glutamate release. The glutamatergic model of schizophrenia holds that NMDA receptor hypofunction disinhibits cortical glutamate release, and that positive and negative symptoms follow. An mGlu2/3 agonist should restrain that release without touching dopamine D2, promising antipsychotic activity without weight gain, prolactin elevation or extrapyramidal effects.

Which receptor carried the effect was settled early and in one direction. Fell and colleagues used mGlu receptor deficient mice to show that the preclinical antipsychotic pharmacology of LY404039 in phencyclidine and amphetamine models depended on mGlu2 and not mGlu3 (doi:10.1124/jpet.108.136861). The programme was an mGlu2 bet dosed with an mGlu2/3 agonist.

What actually happened

The compound failed to beat comparator or placebo on efficacy across the late-stage programme. In the 24-week Phase 3 comparison against aripiprazole, 678 patients were randomised 516 to pomaglumetad and 162 to aripiprazole. PANSS total improvement favoured aripiprazole, -15.58 plus or minus 1.58 against -12.03 plus or minus 0.99, p=0.045. Serious adverse events ran 8.2% against 3.1%, p=0.032, and discontinuation due to adverse events 16.2% against 8.7%, p=0.020. The metabolic promise held, weight change -2.8 plus or minus 0.4 kg against +0.4 plus or minus 0.6 kg, p<0.001 (doi:10.1155/2014/758212). As adjunctive treatment for prominent negative symptoms, 167 randomised patients showed no significant NSA-16 improvement over placebo added to a second-generation antipsychotic (doi:10.1016/j.schres.2013.08.020).

Lilly then ran an exploratory analysis of the trial database, reporting response in patients early in disease, at three years or less, against those late in disease at ten years or more, using the HBBM study and an integrated analysis of five placebo-controlled trials (doi:10.1016/j.biopsych.2015.03.016). That analysis is the origin of the biomarker-selected programme Denovo acquired. In February 2026 Denovo announced a partnership with Orygen on an open-label Phase 2 feasibility study of DB103, the same molecule, at 40 mg twice daily for 12 weeks in 18 to 25 year olds with first-episode psychosis or clinical high risk.

openFDA FAERS returned no adverse event records for pomaglumetad, pomaglumetad methionil or LY2140023. A control query returned 7,770 records for olanzapine, so the null reflects an investigational compound rather than a broken connector.

Failure mechanism, best guess

The registered reason is efficacy in the overall population. The more useful question is why a target with this much preclinical and clinical investment produced so little.

Open Targets scores the GRM2 association with schizophrenia at 0.335, built from clinical precedence at 0.533 and literature co-mention at 0.359, with no human genetic component. GRM3 scores 0.579 on the same disease, including genetic association at 0.800 from GWAS credible sets. A meta-analysis of 14 GRM3 SNPs across 11,318 cases, 13,820 controls and 486 trios found significant associations at rs2237562, odds ratio 1.06 with 95% confidence interval 1.02 to 1.11, and rs13242038, odds ratio 0.90 with 95% confidence interval 0.85 to 0.96 (doi:10.1038/tp.2017.172).

That is the mismatch. Human genetics implicates the paralogue the knockout work had already excluded from the antipsychotic effect. The clinical precedence that makes GRM2 look supported is circular, generated by this programme. Strip it out and the receptor the drug was built around has literature co-mention and nothing else.

Receptor desensitisation is a second candidate. Sustained agonism at a Gi-coupled presynaptic receptor invites tolerance, and long-term dosing is where the programme did worst. The public record carries no occupancy or desensitisation data to test this, so it stays a hypothesis.

How to prevent this next time

No endpoint-level data are posted for this trial, so no calculation is performed here. The published Phase 3 comparison supplies the numbers above and nothing further is inferred.

The first lever is stripping clinical precedence out of target scores before committing. Clinical precedence measures how much has been spent on a target, not whether it is right. GRM2 at 0.335 drops to literature co-mention alone once the programme's own trials are removed, and that is the number a go decision should have used.

The second lever is paralogue resolution. When genetics points at GRM3 and pharmacology at mGlu2, the correct next experiment reconciles them rather than moving to Phase 3. A selective mGlu2 positive allosteric modulator tested against a selective mGlu3 tool would have separated the hypotheses at a fraction of the cost.

The third lever is enrichment specified in advance. The early-in-disease signal was found inside a failed database rather than pre-specified, which is why it remains unresolved fourteen years later and is only now being tested prospectively.

The single highest leverage change would have been to resolve the mGlu2 against mGlu3 question in humans, with selective tools and a pre-specified early-in-disease population, before committing an mGlu2/3 agonist to Phase 3 in broadly defined schizophrenia.

What this means for similar programs

The Claidex failure graph carries GRM2 at an MRS of 39, band yellow. Saturation contributes 9.48 of 15, reflecting five distinct programmes recorded against the target, and genetic deficit 9.97 of 15. Recency contributes 0.71, because this failure is fourteen years old and the exponential decay prices it accordingly. A target can look quiet in the graph and still be unresolved.

For group II metabotropic glutamate programmes generally, the receptor selectivity question was never closed. A 2026 pairwise and network meta-analysis of mGluR agonists and positive allosteric modulators in schizophrenia appeared under the title "Unmet Promise or Missed Potential?", a fair summary of the state of play (doi:10.1016/j.jpsychires.2026.01.023).

The broader pattern is targets whose association scores are inflated by their own failed clinical history. Where clinical precedence is the largest evidence contributor, check whether that precedence is independent of the programme being evaluated.

Open questions

Does the early-in-disease effect replicate prospectively, and will the Orygen study be powered to say?

Is the responder population defined by illness duration, prior antipsychotic exposure or a genomic marker, and are those the same patients?

Would a selective mGlu2 positive allosteric modulator succeed where a mixed agonist failed, or does the GRM3 genetic signal indicate that the mGlu2 hypothesis was wrong from the start?

Will results ever be posted for a 560-participant trial that closed in 2012?

Sources

1. • ClinicalTrials.gov v2 API, NCT01129674, termination reason posted 14 September 2026. https://clinicaltrials.gov/study/NCT01129674 - Open Targets Platform 26.06, GRM2 and schizophrenia. https://platform.opentargets.org/evidence/ENSG00000164082/MONDO_0005090 - Open Targets Platform 26.06, GRM3 and schizophrenia. https://platform.opentargets.org/evidence/ENSG00000198822/MONDO_0005090 - ChEMBL CHEMBL2105707. https://www.ebi.ac.uk/chembl/compound_report_card/CHEMBL2105707/ - Fell et al., Journal of Pharmacology and Experimental Therapeutics 2008,- Adams et al., Schizophrenia Research and Treatment 2014,- Stauffer et al., Schizophrenia Research 2013,- Kinon et al., Biological Psychiatry 2015,- Saini et al., Translational Psychiatry 2017,- Mishra et al., Journal of Psychiatric Research 2026,- Denovo Biopharma, licensing of late-stage neuroscience drug from Lilly, 2015. https://www.prnewswire.com/news-releases/denovo-biopharma-licenses-late-stage-neuroscience-drug-from-lilly-for-development-as-a-personalized-medicine-300043642.html - Denovo Biopharma and Orygen Phase 2 partnership on DB103, February 2026. https://www.biospace.com/press-releases/denovo-biopharma-announces-partnership-with-orygen-on-phase-2-study-evaluating-db103-pomaglumetad-methionil-in-treating-psychosis - openFDA drug event API, no records for pomaglumetad or LY2140023, queried 16 September 2026.