Lenvatinib plus pembrolizumab in melanoma brain metastases: an antiangiogenic resensitization bet that produced almost no intracranial responses

What was tried

Yale University, with drug support from Merck Sharp & Dohme, ran a multicenter Phase 2 trial (NCT04955743) testing whether adding the multikinase VEGFR inhibitor lenvatinib to the PD-1 antibody pembrolizumab could drive intracranial responses in patients whose brain metastases had progressed during or after prior anti-PD-1 or anti-PD-L1 therapy. The study opened on 2022-02-09 with two non-randomized, open-label cohorts: melanoma and renal cell carcinoma. The primary endpoint was best brain metastasis response rate by investigator assessment. Eighteen patients enrolled, sixteen with melanoma and two with renal cell carcinoma, against an actual primary completion date of 2025-05-28. Lenvatinib (CHEMBL1289601) is an approved oral inhibitor of VEGFR1 to VEGFR3, FGFR1 to FGFR4, PDGFRalpha, KIT and RET, marketed since 2015. Pembrolizumab supplied the checkpoint backbone.

The biological hypothesis

The rationale rested on antiangiogenic resensitization. VEGF signaling through KDR (VEGFR2) supports an immunosuppressive tumor vasculature, restricts T cell trafficking, and sustains regulatory and myeloid populations. Blocking VEGFR was expected to normalize vessels and reopen response to PD-1 blockade in patients who had already failed it. The combination had precedent in other settings: pembrolizumab with the VEGFR inhibitor axitinib improved survival in renal cell carcinoma (KEYNOTE-426, Rini et al., NEJM 2019), and nivolumab with cabozantinib did the same (CheckMate 9ER, Choueiri et al., NEJM 2021). The open question this trial asked was specific to the central nervous system, where the blood-brain barrier, distinct vascular biology, and a microglia-dominated immune compartment make extrapolation from systemic disease unreliable.

What actually happened

The melanoma cohort produced one complete intracranial response and no partial responses among sixteen patients, an intracranial CR plus PR rate of 1/16, or 6.3%. Two patients had stable disease, nine had progressive disease, and four were not evaluable. Median progression-free survival in the melanoma cohort was 2.1 months and median overall survival was 5.4 months. The renal cohort, at only two patients, recorded one complete response and one stable disease, with median PFS of 9.6 months and OS of 17.5 months, numbers too small to interpret. Serious adverse events affected nine of sixteen melanoma patients, with seven deaths recorded in that cohort, consistent with the natural history of progressing melanoma brain metastases rather than a new toxicity signal. ClinicalTrials.gov records the stop reason verbatim as "Lack of activity in the drugs." All counts above are taken from the posted results section of NCT04955743.

Failure mechanism, best guess

The most parsimonious reading is compartment-specific efficacy failure in a population selected for prior checkpoint resistance. These patients had already progressed on PD-1 blockade, so the trial was testing whether VEGFR inhibition could reverse established resistance inside the brain. Two features argue against it. First, KDR has no germline genetic anchor to melanoma. The Open Targets association score for KDR and melanoma is 0.46, with a genetic association datatype score of zero and the signal carried by literature and somatic and clinical evidence. The angiogenic dependency is real biology but it is acquired and contextual, not a validated driver of this disease. Second, the central nervous system compartment behaves differently from the periphery. Recent work shows melanoma brain metastases are shaped by a microglia-dominated niche whose reprogramming, not vascular normalization alone, governs immunotherapy response (Rodriguez-Baena et al., Cancer Cell 2025), and brain metastasis organotropism imposes microenvironmental constraints that systemic combinations do not address (Yuzhalin and Vasudev, Cold Spring Harbor Perspectives in Medicine 2020). Adding a VEGFR inhibitor to a failed checkpoint backbone did not supply the missing signal. The nine progressive-disease events and the 2.1-month median PFS point to primary resistance rather than early stopping artifact.

How to prevent this next time

Two quantitative tools would have de-risked this design before nineteen patients were exposed. First, a Bayesian predictive probability of success monitored after the first stage. With a null intracranial response rate of 5% and a target of 25%, a Simon two-stage design rejects the agent if too few responses appear early. The melanoma cohort, with one response in sixteen and nine progressions, would have crossed almost any reasonable futility boundary at the first interim. The predictive probability of eventual success is

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

and after the observed interim it would have fallen close to zero. Second, an explicit power and base-rate calculation. Detecting a rise from 5% to 25% intracranial response with 80% power at a one-sided 0.05 level needs roughly thirty evaluable patients per cohort, so sixteen melanoma patients with four not evaluable was underpowered from the start, and the historical base rate for VEGFR plus checkpoint combinations against established checkpoint resistance is low. A biomarker enrichment strategy keyed to intratumoral T cell presence or microglial state, rather than histology alone, would have concentrated any real signal. The single highest leverage change would have been a pre-specified Bayesian futility boundary at the first interim, tied to intracranial response, so the program could be stopped after the first handful of progressions rather than after a full cohort.

What this means for similar programs

Antiangiogenic plus checkpoint combinations carry approvals in renal and endometrial cancer, and that success tempts extension into checkpoint-resistant and central nervous system settings. The KDR record argues for caution. The target lacks genetic anchoring to melanoma, the resistance population removes the easiest path to benefit, and the brain compartment adds biology that vascular normalization does not cover. Programs aiming at brain metastases should treat intracranial activity as a separate hypothesis with its own enrichment and its own futility analysis, not as a corollary of systemic data. The FAERS profile for lenvatinib, with 85% of 15,246 reports coded serious and dominated by hypertension and fatigue, also means any marginal intracranial benefit must clear a real tolerability cost.

Open questions

Would a checkpoint-naive brain metastasis population have responded, isolating resistance from compartment as the cause? Does intratumoral or microglial profiling identify a subset where VEGFR inhibition restores response? Is the renal signal, one response in two patients, worth a properly sized cohort, or is it noise? And does vascular normalization in the brain require a different schedule or a different VEGFR agent than an approved systemic dose.

Sources

1. • ClinicalTrials.gov, NCT04955743, protocol and posted results (enrollment, intracranial response counts, PFS, OS, adverse events, stop reason). https://clinicaltrials.gov/study/NCT04955743 - Open Targets Platform, KDR (ENSG00000128052) association with melanoma (EFO_0000756), score 0.46, genetic association datatype 0. https://platform.opentargets.org/target/ENSG00000128052 - ChEMBL, lenvatinib (CHEMBL1289601), approved 2015, multikinase VEGFR inhibitor. https://www.ebi.ac.uk/chembl/ - OpenFDA FAERS, medicinalproduct lenvatinib, 15,246 reports, 13,028 serious. https://api.fda.gov/drug/event.json - Rini BI et al. Pembrolizumab plus Axitinib versus Sunitinib for Advanced Renal-Cell Carcinoma. N Engl J Med 2019. https://- Choueiri TK et al. Nivolumab plus Cabozantinib versus Sunitinib for Advanced Renal-Cell Carcinoma. N Engl J Med 2021. https://- Rodriguez-Baena FJ et al. Microglial reprogramming enhances antitumor immunity and immunotherapy response in melanoma brain metastases. Cancer Cell 2025. https://- Yuzhalin AE, Vasudev A. Brain Metastasis Organotropism. Cold Spring Harb Perspect Med 2020. https://.