Novartis stopped MGY825, and NRF2-mutant lung cancer lost its only clinical-stage inhibitor-direction program

What was tried

Novartis ran NCT05275868, a first in human, open label Phase 1 study of MGY825 in adults with advanced non-small cell lung cancer. The design was single group, non-randomised and unmasked, with a dose escalation and two expansion groups. Escalation and expansion group 1 required a locally confirmed NFE2L2, KEAP1 or CUL3 mutation, while expansion group 2 enrolled irrespective of mutation status. All had progressed after platinum based chemotherapy and a PD-1 or PD-L1 antibody, and had to accept paired screening and on treatment biopsies. Primary endpoints were safety and tolerability only, with overall response rate by RECIST 1.1 among the secondary endpoints.

The study opened on 5 October 2022 and closed with an actual enrolment of 41. Actual primary completion and completion were both 13 October 2025, three years in. ClinicalTrials.gov posted the terminated status on 4 September 2026 with the sponsor reason quoted verbatim: "The trial was terminated due to a business decision and not as a result of any safety concerns." No results were posted. LARVOL VERI reports planned enrolment had earlier been 140 and anticipated primary completion July 2027, putting the study at 29 percent of its intended size and 21 months short of its planned readout.

Novartis has not published the molecular mechanism of MGY825. The registry lists NFE2L2, NRF2, KEAP1 and CUL3 as keywords and uses mutations in those genes as the enrolment filter. Commercial pipeline databases annotate the compound against the same axis, with no peer reviewed mechanism paper behind that annotation.

The biological hypothesis

KEAP1 normally delivers NRF2 to a CUL3 based ubiquitin ligase for degradation. Loss of function in KEAP1 or CUL3, or gain of function in NFE2L2, stabilises NRF2 and leaves the antioxidant transcriptional programme switched on. In lung cancer that state is common and travels with bad outcomes. Pathway mutations define a molecular subset of rapidly progressing lung adenocarcinoma (10.1016/j.jtho.2019.07.003), associate with worse chemotherapeutic response (10.1158/1078-0432.CCR-19-1237), and have been called fuel for a superresistant phenotype (10.1016/j.lungcan.2021.07.006). The eligibility criteria map onto that literature: platinum refractory, checkpoint refractory, mutation selected.

The preclinical case rested on constitutive NRF2 activity creating a dependency rather than only a resistance mechanism. A CRISPR screen identified redox vulnerabilities specific to KEAP1/NRF2 mutant NSCLC (10.1016/j.redox.2022.102358), and NRF2 activation induces NADH reductive stress that becomes a metabolic vulnerability (10.1016/j.cmet.2023.01.012).

Open Targets scores the NFE2L2 and non-small cell lung carcinoma association at 0.475, and the composition matters more than the headline: literature 0.980, somatic mutation 0.564, affected pathway 0.330, clinical evidence 0.122, and no genetic association datatype at all.

What actually happened

The trial stopped without a registry results posting, a published dose escalation report or a disclosed recommended Phase 2 dose. openFDA FAERS returns no reports naming MGY825, so no independent safety read exists.

What is visible is the shape of the retreat. Enrolment was cut and the completion date pulled forward before the terminated status appeared, the pattern of a programme wound down rather than one hitting a stopping rule. Trade coverage in November 2025 reported that this discontinuation left Bayer's BAY 3605349 as the last molecule with activity on NRF2 in clinical development.

Open Targets annotates two clinical stage molecules against NFE2L2: omaveloxolone, approved for Friedreich ataxia, and bardoxolone methyl at Phase 3. Both are NRF2 activators, and neither speaks to an oncology hypothesis needing the opposite direction of effect. ChEMBL returns the same count and no entry for MGY825.

Failure mechanism, best guess

The archetype recorded here is strategic_reprioritization, because that is what the sponsor stated and because no efficacy readout exists that could support an efficacy_failure label. A business decision arriving after enrolment was cut by 71 percent and the completion date advanced by 21 months usually encodes an internal look at data the outside world never sees. Claidex records the stated archetype and flags that the cause could be efficacy, portfolio economics, or both.

The structural mechanism is easier to see. NFE2L2 encodes a transcription factor, and its tractability annotations describe chemical matter rather than a validated modality. A 2025 review sets out how much of thirty years of NRF2 chemistry went into activation rather than tumour pathway shutdown (10.1038/s41573-025-01145-0). A single asset against a mechanism with no clinical precedent carries mechanism risk a safety focused Phase 1 cannot retire, plus portfolio risk, because an asset with no class comparator is the cheapest thing on a pipeline review to cut.

How to prevent this next time

No endpoint level data were released, so no posterior, power calculation or effect size estimate is supportable from the public record. What remain are design levers, three of them available inside this protocol.

Pre-commit a go or no-go on pharmacodynamics rather than response. The protocol already mandated paired biopsies, and NRF2 target gene output in those samples separates "the molecule did not engage" from "engagement did not translate". Only the second finding should retire the mechanism.

Treat the two expansion groups as the enrichment experiment they already are. Group 1 was mutation selected and group 2 was all comers, a biomarker hypothesis in miniature that deserved a prespecified analysis and a public report.

Price the base rate correctly at portfolio review. The number carrying clinical risk is the clinical evidence score of 0.122 with no genetic association evidence, not the overall 0.475 that literature volume inflates.

The single highest leverage change would have been a prespecified, publicly reported pharmacodynamic readout from the mandated paired biopsies, so that a business decision to stop could not also erase the mechanism question.

What this means for similar programs

The Claidex graph holds no other NFE2L2 claim. The mechanism risk score of 24 out of 100 sits in the green band: phase burden 3.53 from a single Phase 1, archetype severity 2.00 from a non-efficacy stop, recency 4.25, genetic deficit 7.87, and saturation 6.77 from three distinct programmes. A low score here reflects a thin record, not a safe target.

The pattern recurs at adjacent targets in the graph. Ceralasertib against ATR and BBI-355 against CHEK1 both ended as Phase 1 strategic discontinuations rather than efficacy failures, leaving each mechanism untested. Groups pursuing KEAP1 and NFE2L2 altered lung cancer may find the synthetic lethal framing more tractable than direct pathway inhibition, and KEAP1 and STK11 alterations have been reported to enhance vulnerability to ATR inhibition in KRAS mutant NSCLC (10.1016/j.ccell.2025.06.011).

Open questions

Was a recommended Phase 2 dose declared? Did expansion group 1 differ from group 2 on any endpoint? Did the paired biopsies show pathway modulation? Did the enrolment cut from 140 to 41 precede or follow an internal efficacy look? Novartis has answered none of these publicly.

Sources

1. - ClinicalTrials.gov, NCT05275868: https://clinicaltrials.gov/study/NCT05275868 - Open Targets Platform API: NFE2L2 (ENSG00000116044) and MONDO_0005233 - ChEMBL API, mechanism records for NFE2L2 target entries - openFDA FAERS drug event API (no records for MGY825) - J Thorac Oncol 2019, - Clin Cancer Res 2020, - Lung Cancer 2021, - Redox Biol 2022, - Cell Metab 2023, - Nat Rev Drug Discov 2025, - Cancer Cell 2025, - ApexOnco, "Time is running out for the NRF2 mechanism", 13 November 2025 - LARVOL VERI drug record, MGY825.