Ulixertinib in metastatic uveal melanoma: an ERK inhibitor with no responders in stage 1

What was tried

Dana-Farber Cancer Institute ran NCT03417739, a single-arm, open-label Phase 2 study of BVD-523, now ulixertinib, in metastatic uveal melanoma. The design was a Simon two-stage plan with a planned sample size of 25 and an interim efficacy evaluation after 13 patients. Dosing was the recommended Phase 2 dose of 600 mg twice daily orally in 28-day cycles. Enrollment was 13 participants at two sites, starting 26 March 2018, with actual primary completion on 5 May 2020 and completion on 16 July 2025. The record was updated on 27 July 2026 with posted results.

Eligibility required confirmed stage IV uveal melanoma with measurable disease by RECIST, an ECOG performance status of 2 or better, and life expectancy over six months. Prior MEK inhibition and other MAPK-targeted agents were permitted in any number, and prior ERK inhibition was the single exclusion on that axis. The primary endpoint was overall response rate by RECIST 1.1 to 52 weeks. Secondary endpoints included overall survival, time to tumour progression, change in DUSP6 expression as a pharmacodynamic readout, and whole exome sequencing of paired biopsies taken before treatment and on day 12 to 16.

Ulixertinib is registered in ChEMBL as CHEMBL3590106, a small molecule at maximum phase 2, synonyms BVD-523 and BVD-ERK.

The biological hypothesis

More than 80% of uveal melanomas carry an activating GNAQ or GNA11 mutation, and the resulting constitutive G alpha q signalling drives the MAPK cascade through phospholipase C beta and protein kinase C. ERK1 and ERK2, encoded by MAPK3 and MAPK1, are the terminal kinases of that cascade. Ulixertinib is a reversible, ATP-competitive inhibitor of both, so it was positioned as the node that feedback reactivation could not bypass.

The Open Targets pattern shows how much of this rationale was pathway logic rather than target-disease evidence. GNAQ scores 0.625 against uveal melanoma on somatic mutation evidence of 0.814, and GNA11 scores 0.625 on the same footing. MAPK1 scores 0.041, built from literature at 0.131 and clinical evidence at 0.061, with no somatic mutation component. The disease has strong genetic definition and the drugged node inherits none of it.

What actually happened

Thirteen patients enrolled between April 2018 and April 2019. Median age was 64 years, range 34 to 76, and 9 of 13 were female. Liver metastases were present in 84.6% and lung metastases in 30.8%.

The posted primary result is an overall response rate of 0%, 95% confidence interval 0 to 20.5. Best response was stable disease in 4 patients and progressive disease in 7, with 2 unevaluable after withdrawal. Median time to tumour progression was 2.0 months, 90% confidence interval 1.8 to 3.6. Median overall survival was 6.9 months, 90% confidence interval 3.2 to 8.3, with eight deaths from disease progression. The trial stopped at the interim futility boundary.

Toxicity behaved as the class predicts. Serious adverse events occurred in 5 of 13 participants, including transaminase elevations in 2 each and single events of anaemia, amylase elevation, hyponatraemia, pruritus and maculopapular rash. The most common events of any grade were diarrhoea in 10 of 13, nausea in 8, and fatigue, anorexia and acneiform rash in 7 each. No deaths were attributed to treatment.

The pharmacodynamic readout is the most informative number in the record. Median change in DUSP6 expression between the pre-treatment and on-treatment biopsies was 311.08 normalised NanoString counts, with a 95% confidence interval from -32.73 to 2514.83. DUSP6 is a canonical transcriptional output of active ERK, so sustained suppression of ERK activity would be expected to lower it. The measured direction is upward and the interval crosses zero, so the trial did not demonstrate durable suppression of ERK transcriptional output in tumour tissue.

Failure mechanism, best guess

The archetype is an efficacy failure, and the mechanism underneath it is a translational mismatch between a genetically defined disease and a pathway-defined target. The GNAQ and GNA11 mutations are near-universal, but they sit three nodes upstream of ERK, and G alpha q signalling also runs through protein kinase C, YAP and other effectors that ERK inhibition leaves intact. MEK inhibition had already made that point. The randomised SUMIT study of selumetinib with dacarbazine did not improve progression-free survival, and the systematic review of MEK inhibitors in this disease reached the same conclusion for the class.

Two design choices compressed the information returned. Prior MAPK-targeted therapy was allowed without stratification, so a population already selected for MAPK-pathway resistance could dominate a 13-patient stage. The DUSP6 result was wide and directionally unfavourable, which leaves the study unable to separate a target-engagement failure from a target-irrelevance failure. Those explanations carry opposite implications for the next molecule.

How to prevent this next time

The posted dataset supports a base-rate argument rather than a Bayesian endpoint model, because there is no comparator arm and the counts are single digits.

First, the class prior should have been the design input. Two published negative reads on MEK inhibition were available before enrollment opened, and both located the failure in the pathway rather than in the specific node. A study of a downstream node in the same cascade should have been powered as a mechanism test with a target-engagement gate, not as a response-rate screen.

Second, the pharmacodynamic endpoint should have gated the design rather than sitting among the secondary measures. A prespecified rule requiring suppression of ERK transcriptional output in a minimum number of paired biopsies before continuing to stage 2 would have converted an uninformative negative into a mechanistic one.

Third, prior MAPK exposure should have been a stratification factor or an exclusion, since allowing it without limit into a 13-patient interim stage puts the futility decision at the mercy of enrolment order.

The single highest leverage change would have been a mandatory paired-biopsy target-engagement gate on ERK transcriptional output, since without it the trial cannot distinguish inadequate ERK inhibition from correct ERK inhibition in a disease that does not depend on ERK.

What this means for similar programs

Open Targets lists six clinical candidates against MAPK1, including ravoxertinib, ASN007, temuterkib, MK-8353 and KO-947, none of which has advanced past Phase 2 in that record. Ulixertinib retains activity where the MAPK lesion is proximal, as in the report of RAF-independent MEK mutations in refractory histiocytic neoplasms responding to ERK inhibition. The distinction that matters is the distance between driver lesion and drugged node.

For metastatic uveal melanoma the field moved elsewhere. Tebentafusp, directed at gp100 in HLA-A*02:01 positive patients, showed an overall survival benefit in a randomised Phase 3, which reset the comparator for any pathway-directed agent here. The Claidex mechanism risk score for MAPK1 is 40, in the yellow band, with the largest contribution from the genetic deficit term, since the target carries almost no genetic association with the disease it was tested in.

Open questions

• Did the paired biopsies show ERK target engagement at the protein level, and can the sequencing data separate engagement failure from pathway irrelevance?
• Do the 4 patients with stable disease share a genotype, such as a secondary MAPK lesion, that the whole exome data could identify?
• Would vertical blockade, for example protein kinase C inhibition combined with ERK inhibition, produce activity that single-node inhibition cannot?

Sources

1. • ClinicalTrials.gov record and posted results for NCT03417739, retrieved 28 July 2026. https://clinicaltrials.gov/study/NCT03417739 - Buchbinder EI, et al. A Phase II Study of ERK Inhibition by Ulixertinib (BVD-523) in Metastatic Uveal Melanoma. Cancer Res Commun. 2024, volume 4, issue 5, pages 1321-1327.- Carvajal RD, et al. Selumetinib in Combination With Dacarbazine in Patients With Metastatic Uveal Melanoma: A Phase III, Multicenter, Randomized Trial (SUMIT). J Clin Oncol. 2018, volume 36, issue 12, pages 1232-1239.- Steeb T, et al. How to MEK the best of uveal melanoma: A systematic review on the efficacy and safety of MEK inhibitors in metastatic or unresectable uveal melanoma. Eur J Cancer. 2018, volume 103, pages 41-51.- Nathan P, et al. Overall Survival Benefit with Tebentafusp in Metastatic Uveal Melanoma. N Engl J Med. 2021, volume 385, issue 13, pages 1196-1206.- Diamond EL, et al. RAF-independent MEK mutations drive refractory histiocytic neoplasms but respond to ERK inhibition. Cancer Cell. 2026, volume 44, issue 1, pages 203-220.- Germann UA, et al. Targeting the MAPK Signaling Pathway in Cancer: Promising Preclinical Activity with the Novel Selective ERK1/2 Inhibitor BVD-523. Mol Cancer Ther. 2017, volume 16, issue 11, pages 2351-2363.- Open Targets Platform, association scores for uveal melanoma (MONDO_0006486) and targets MAPK1 (ENSG00000100030), GNAQ and GNA11, retrieved 28 July 2026. https://platform.opentargets.org/target/ENSG00000100030 - ChEMBL entry for ulixertinib, CHEMBL3590106, retrieved 28 July 2026. https://www.ebi.ac.uk/chembl/compound_report_card/CHEMBL3590106/ - openFDA FAERS drug event endpoint, query for ulixertinib, retrieved 28 July 2026. https://open.fda.gov/apis/drug/event/.