Tumour paint in the mouth: tozuleristide failed to separate oral cancer from normal mucosa

What was tried

An investigator-initiated Phase 1/2 trial at the University of Washington (NCT05316688), with Blaze Bioscience as collaborator, tested whether tozuleristide could mark oral cavity squamous cell carcinoma and high-grade dysplasia during surgery. The design was single group, open label and diagnostic in purpose. Participants received tozuleristide intravenously over 1 to 5 minutes beginning about 1 hour before surgery, then underwent resection per standard of care with near-infrared imaging on a standard-of-care device.

Eight adults enrolled and all eight completed. Median age was 60.5 years with a full range of 31 to 75, the split was 4 female and 4 male, and all 8 tumours were oral tongue primaries. The primary outcome was the number of adverse events at 7 to 21 days after dosing. Two secondary outcomes carried the scientific question: how many subjects showed no discernible tumour fluorescence against background, and how many achieved negative margins.

The trial ran from 17 April 2024 to completion on 26 September 2025. The record was updated on 16 September 2026 with status TERMINATED and the reason stated plainly: terminated due to suboptimal delineation of fluorescence between tumor and native oral cavity mucosa.

The biological hypothesis

Tozuleristide, also called BLZ-100, is chlorotoxin conjugated to indocyanine green, listed in ChEMBL as CHEMBL3990023. Chlorotoxin is a 36-amino-acid scorpion venom peptide that concentrates in tumour tissue. Parrish-Novak and colleagues published the nonclinical profile in 2017, Farkas and colleagues showed in 2023 that chlorotoxin binds both matrix metalloproteinase 2 and neuropilin 1, and McGonigle and colleagues had earlier reported that neuropilin 1 drives tumour-specific uptake.

Open Targets scores MMP2 against oral cavity squamous cell carcinoma at 0.709, one of the stronger target-disease associations in this corpus, with somatic mutation evidence at 0.974, literature at 0.881 and clinical at 0.684. It records three clinical programmes against MMP2, marimastat and rebimastat at Phase 3 and CTS-1027 at Phase 2, none of which reached approval.

The imaging hypothesis is different from the drug hypothesis. It does not require MMP2 to drive the disease. It requires the agent to accumulate in tumour at a high enough ratio over adjacent normal tissue for a camera to resolve the boundary, and the clinical track record supporting that came almost entirely from the brain. Blaze ran Phase 1 studies in adult glioma and paediatric CNS tumours, received FDA Fast Track designation in April 2020, and completed a pivotal paediatric CNS Phase 2/3 with 118 enrolled in June 2022 (NCT03579602) and an adult CNS Phase 2 with 50 enrolled in March 2025 (NCT04743310).

What actually happened

All 8 participants were dosed and imaged. In 4 of 8, the tumour showed no discernible fluorescence compared with background. Negative margins were achieved in 8 of 8, which reflects standard surgical technique rather than any contribution from the imaging, since half the tumours produced no usable signal.

The posted adverse events at 7 to 21 days were fatigue at grade 3 with 2 events, one event each of acute hypoxemia, aspiration pneumonia, back wound, COVID-19, ileus and pulmonary embolism at grade 3, and one each of constipation, neck infection and seroma at grade 2. These are the expected complications of major head and neck resection. No event was attributed to tozuleristide, and no openFDA FAERS report names the agent. The investigators stopped at 8 participants and posted both the reason and the results, which is an unusually clean termination.

Failure mechanism, best guess

The archetype is translational mismatch, and the mismatch is about compartment rather than about binding.

The brain is an exceptionally favourable setting for a circulating fluorescent tracer. An intact blood-brain barrier keeps the agent out of normal parenchyma, so even modest tumour uptake yields high contrast, and the barrier is disrupted preferentially at tumour. Oral mucosa offers none of that. It is richly perfused, has no barrier, is chronically inflamed in the population that develops these cancers, and sits within field-cancerized epithelium.

Both of chlorotoxin's characterized binding partners are present in that normal tissue. MMP2 is a stromal remodelling protease induced during wound healing and inflammation rather than a tumour-exclusive antigen, and neuropilin 1 is broadly expressed on endothelium and many normal cell types. Neither is a clean on-off switch in a perfused mucosal surface.

The most economical explanation is that tumour paint's selectivity in the brain was substantially compartment selectivity rather than pure molecular selectivity, and that moving to a barrier-free compartment collapsed the contrast. Four of eight tumours producing no discernible fluorescence is the signature of a signal-to-background problem, not a dosing or device problem. One alternative cannot be excluded: this protocol allowed imaging anywhere from 1 to 30 hours after injection, so timing may have contributed to the variance even if it does not explain the floor.

How to prevent this next time

No tumour-to-background ratios were posted, so a quantitative contrast model cannot be built from public data. The available levers are qualitative.

Measure the background before dosing surgical patients. A paired ex vivo study on resected tumour and adjacent normal mucosa quantifies the contrast ratio at a fraction of the cost of an in-human surgical protocol. Blaze ran precisely this design once, an exploratory ex vivo fluorescence study with 30 adult subjects (NCT02496065), and repeating it per new tissue compartment is cheap insurance.

Treat compartment as a covariate in the target product profile. A contrast claim earned behind the blood-brain barrier does not transfer to mucosa, skin or peritoneum, and that assumption deserves the scrutiny a new indication would receive.

Use biomarker enrichment to make a negative result interpretable. Measuring MMP2 and neuropilin 1 in each patient's tumour and adjacent normal tissue separates a target-expression explanation from a perfusion and clearance one.

Pre-specify a stopping rule on contrast rather than on safety alone. This trial got that right, and stopping after 8 spared later patients an uninformative exposure.

The single highest leverage change would have been requiring a paired ex vivo tumour-versus-normal-mucosa contrast measurement in oral cavity tissue before opening any in-human surgical trial in that compartment.

What this means for similar programs

This is the first MMP2 entry in the Claidex graph, and the recomputed mechanism risk score is 31, in the yellow band. It is held down by the single Phase 1/2 failure and the strong Open Targets association, and pushed up by archetype severity and saturation, since three prior MMP2 clinical programmes reached Phase 2 or Phase 3 without approval.

The read-across is not about MMP2 as a drug target. It is about the compartment claim every tumour-targeted fluorophore makes implicitly. Agents built on differential accumulation should be assumed to have indication-specific contrast until measured otherwise, and the published head and neck experience with 5-aminolevulinic acid shows that oral cavity surgery is a demanding setting for all of them.

This trial is a useful negative control for the field. The agent was safe, the binding biology is real, the surgical endpoint was met by conventional technique, and the imaging still did not work.

Open questions

What were the actual tumour-to-background ratios in the four tumours that did fluoresce, and did they cluster by injection-to-imaging interval?

Did MMP2 or neuropilin 1 expression in tumour versus adjacent mucosa differ between the fluorescent and non-fluorescent cases? No tissue correlative data were posted.

What is the development status of tozuleristide now? No regulatory outcome has been announced since the pivotal paediatric trial completed in 2022.

Sources

1. • ClinicalTrials.gov, NCT05316688, protocol and posted results retrieved 23 September 2026. https://clinicaltrials.gov/study/NCT05316688 - ClinicalTrials.gov, NCT03579602 (paediatric CNS pivotal), NCT04743310 (adult CNS Phase 2), NCT02496065 (ex vivo fluorescence), NCT02462629 and NCT02234297 (Phase 1), retrieved 23 September 2026. https://clinicaltrials.gov/ - Parrish-Novak J, et al. Nonclinical Profile of BLZ-100, a Tumor-Targeting Fluorescent Imaging Agent. Int J Toxicol.

2. - Farkas S, et al. Chlorotoxin binds to both matrix metalloproteinase 2 and neuropilin 1. J Biol Chem.

3. - McGonigle S, et al. Neuropilin-1 drives tumor-specific uptake of chlorotoxin. Cell Commun Signal.

4. - Patil CG, et al. Phase 1 Safety, Pharmacokinetics, and Fluorescence Imaging Study of Tozuleristide (BLZ-100) in Adults With Newly Diagnosed or Recurrent Gliomas. Neurosurgery.

5. - Yamada M, et al. A first-in-human study of BLZ-100 (tozuleristide) demonstrates tolerability and safety in skin cancer patients. Contemp Clin Trials Commun.

6. - Kobets AJ, et al. Unexpected Binding of Tozuleristide "Tumor Paint" to Cerebral Vascular Malformations. Neurosurgery.

7. - Filip P, et al. 5-Aminolevulinic Acid Fluorescence-Guided Surgery in Head and Neck Squamous Cell Carcinoma. Laryngoscope.

8. - Open Targets Platform, MMP2 (ENSG00000087245) association with oral cavity squamous cell carcinoma (MONDO_0004958), retrieved 23 September 2026. https://platform.opentargets.org/target/ENSG00000087245 - ChEMBL, CHEMBL3990023 (tozuleristide), retrieved 23 September 2026. https://www.ebi.ac.uk/chembl/compound_report_card/CHEMBL3990023/ - Blaze Bioscience, FDA Grants Fast Track Designation to BLZ-100 (tozuleristide) for Pediatric Central Nervous System Tumors, 7 April 2020. https://www.businesswire.com/news/home/20200407005238/en/.