Vidutolimod in first-line metastatic melanoma: the randomized TLR9 bet that was called off

What was tried

NCT04695977 was a randomized, open-label Phase 2/3 study of first-line intratumoral vidutolimod (CMP-001) plus intravenous nivolumab compared with nivolumab alone in unresectable or metastatic melanoma. The dosing plan gave vidutolimod 10 mg intratumorally weekly for seven doses with nivolumab 360 mg every three weeks, and the primary endpoint was confirmed objective response rate by blinded independent central review (ClinicalTrials.gov, NCT04695977). The sponsor was Regeneron Pharmaceuticals, which acquired the originator Checkmate Pharmaceuticals in 2022. The study was terminated with a recorded reason of "Business Decision," an actual enrollment of 20 participants, and a primary completion date of 22 July 2024 (ClinicalTrials.gov, NCT04695977).

The biological hypothesis

Vidutolimod is a CpG-A Toll-like receptor 9 agonist packaged inside a virus-like particle. Injected into a tumor, it was designed to activate plasmacytoid dendritic cells, drive a type I and type II interferon response, and convert an immunologically cold lesion into one infiltrated by antitumor T cells (Ribas et al., Cancer Discovery 2021, 10.1158/2159-8290.CD-21-0425). The combination logic was that local innate activation would restore sensitivity to PD-1 blockade. Preclinical work supported this: intratumoral TLR9 agonism plus anti-PD-1 overcame interferon-pathway and antigen-presentation resistance through natural killer and CD8 T cell activation (Torrejon et al., Cancer Discovery 2020, 10.1158/2159-8290.CD-19-1409). The Open Targets association between TLR9 and melanoma is moderate at 0.355, carried mostly by the clinical evidence component at 0.567 rather than by human genetics (Open Targets Platform).

What actually happened

The supporting clinical data came from single-arm Phase 1b experience in PD-1-refractory melanoma. The final analysis reported a best objective response rate of 23.5 percent for vidutolimod PS20-A plus pembrolizumab, 11.5 percent for the PS20-B formulation plus pembrolizumab, and 20.0 percent for vidutolimod monotherapy (Milhem et al., Cancer 2025, 10.1002/cncr.70022). Grade 3 or higher treatment-emergent adverse events occurred in 55.3 percent of part 1 patients and 37.5 percent of part 2 patients, with no treatment-related deaths (Milhem et al., 2025). An earlier Phase 1b cohort had shown durable responses in about 25 percent of patients alongside an induced interferon-gamma signature and rising CXCL10 (Ribas et al., 2021). The first-line randomized study that would have tested whether any of this beat nivolumab alone enrolled 20 patients before it was stopped (ClinicalTrials.gov, NCT04695977).

Failure mechanism, best guess

Two facts sit uncomfortably together. First, the response rate depended on the formulation: 23.5 percent with PS20-A versus 11.5 percent with PS20-B, a difference driven by polysorbate concentration rather than biology (Milhem et al., 2025). A signal that moves with an excipient is a signal that has not been pinned to mechanism. Second, the only randomized Phase 3 readout in the TLR9-agonist class had already failed. Tilsotolimod plus ipilimumab produced an objective response rate of 8.8 percent versus 8.6 percent for ipilimumab alone in ILLUMINATE-301, after the same combination had reached 22.4 percent in the single-arm ILLUMINATE-204 study (Journal of Clinical Oncology 2025, 10.1200/JCO.24.00727). The pattern is a 20-something percent single-arm signal that lands at control parity once a randomized comparator is added. The recorded "Business Decision" is best read as a sponsor declining to fund a randomized confirmation whose class prior had turned strongly negative. The Claidex Mechanism Risk Score for TLR9 in melanoma is 25 out of 100 (yellow), with the genetic-deficit term contributing 9.7 of 15 because human genetic support for the target in this disease is thin.

How to prevent this next time

Two quantitative tools would have changed the decision calculus before the randomized study opened.

First, a Bayesian predictive probability of success built on a class-informed prior. The single-arm vidutolimod responses should not have been read at face value once a sibling agent had collapsed from 22.4 percent to 8.8 percent in randomization (10.1200/JCO.24.00727). Formally,

PP(success)=∫θ​P(final success∣θ,interim data)⋅π(θ∣interim data)dθ

With a prior centered on the randomized class result, the posterior probability of beating first-line nivolumab falls well below a go threshold, because most of the posterior mass for the true added effect sits near zero.

Second, an explicit power calculation tied to a realistic comparator. First-line nivolumab monotherapy delivers an objective response rate near 45 percent. To detect an increase to 60 percent with two-sided alpha of 0.05 and 80 percent power, the two-proportion sample size formula returns roughly 170 patients per arm, about 340 total. The terminated study enrolled 20 (ClinicalTrials.gov, NCT04695977), so it was never positioned to resolve a plausible effect, and the implied effect needed to power a smaller trial was larger than the class had ever shown under randomization.

The single highest leverage change would have been to gate the first-line randomized study on a formulation-locked, randomized Phase 2 readout against nivolumab rather than on formulation-sensitive single-arm response rates.

What this means for similar programs

Intratumoral innate agonists keep generating attractive single-arm response rates that do not survive randomization. The shared failure mode is attribution: an injected lesion regresses, but the comparator-adjusted systemic benefit is small. Programs in this class should treat any uncontrolled response rate as a hypothesis, hold formulation fixed from the first efficacy cohort, and budget for a randomized Phase 2 gate before committing to registrational design. Targets with moderate clinical but weak genetic association, like TLR9 in melanoma at 0.355 (Open Targets), deserve extra scrutiny on the comparator question.

Open questions

Did the PS20-A versus PS20-B response gap reflect a true dose-exposure relationship that a locked formulation could have exploited, or noise in small cohorts? Would a biomarker-enriched population, selected by baseline non-inflamed tumor status and on-treatment CXCL10 induction, have separated responders from the comparator? Is there any intratumoral TLR9 setting, such as neoadjuvant resectable disease, where local activation translates into a randomized survival benefit rather than an injected-lesion effect?

Sources

1. • ClinicalTrials.gov, NCT04695977 (design, enrollment 20, termination reason "Business Decision," primary completion 22 July 2024). - Milhem MM, et al. Intratumoral vidutolimod as monotherapy or with pembrolizumab in PD-1-resistant melanoma: final Phase 1b analysis. Cancer 2025. https://(ORR 23.5/11.5/20.0 percent, grade 3+ TEAEs 55.3/37.5 percent). - Ribas A, et al. Overcoming PD-1 blockade resistance with CpG-A TLR9 agonist vidutolimod. Cancer Discovery 2021. https://(25 percent durable responses, interferon-gamma signature, CXCL10). - Torrejon DY, et al. Overcoming genetically based resistance to PD-1 blockade. Cancer Discovery 2020. https://(TLR9 agonism plus anti-PD-1, NK and CD8 mechanism). - ILLUMINATE-301: tilsotolimod plus ipilimumab in refractory melanoma. Journal of Clinical Oncology 2025. https://(ORR 8.8 versus 8.6 percent, prior 22.4 percent single-arm). - Open Targets Platform: TLR9 (ENSG00000239732) and melanoma (EFO_0000756), overall association 0.355. - openFDA FAERS: vidutolimod, 2 reports, no disproportionality signal. Available from: https://clinicaltrials.gov/study/NCT04695977.