Failure museum
Oncology learns more from what did not work than from what did, but failures vanish from pipelines and press releases. This room keeps them: drugs stopped, targets abandoned, approvals withdrawn, each with the lesson it left. Nothing here is a verdict on the idea; several of these targets later worked with a different weapon.
The phase 2 mirage
Encouraging early data, often single-arm response rates on a checkpoint-inhibitor backbone, that phase 3 could not reproduce.
Bempegaldesleukin was a re-engineered interleukin-2 meant to be a safer version of a famous old immunotherapy. It added nothing to nivolumab in three phase 3 trials.
Lesson: single-arm combination response rates in melanoma are unreliable because nivolumab alone already produces them; biological rationale (Treg sparing) did not translate.
An enzyme blocker meant to stop tumours starving T cells of tryptophan. Its 2018 phase 3 failure ended an entire class overnight.
Lesson: pharmacodynamic proof of target engagement in the tumour should precede phase 3; a large uncontrolled response rate on a pembrolizumab backbone in melanoma is not evidence of added benefit.
An immune-brake blocker that looked excellent in a phase 2 lung cancer trial and then failed every phase 3.
Lesson: a small randomised phase 2 with a surrogate endpoint in a selected population can mislead; redundancy between checkpoints means blocking a second one does not necessarily add to PD-1/PD-L1 blockade.
Right idea, wrong molecule
The target or concept survived; the specific drug did not do what was claimed, or carried the wrong payload.
ADU-S100 was the first STING agonist in the clinic. Injected directly into tumours, it produced almost no responses, alone or with checkpoint blockade.
Lesson: intratumoural delivery to one lesion rarely produces systemic immunity in humans as it does in mice; pharmacology (rapid clearance, dosing) matters as much as the target.
Billed as the first PARP inhibitor for triple-negative breast cancer, it failed its phase 3 in 2011. It turned out not to inhibit PARP at all.
Lesson: mechanism claims should be verified before a class label is attached; the true PARP inhibitors (olaparib, talazoparib) later succeeded in BRCA-mutant breast cancer.
Rovalpituzumab tesirine (Rova-T) was the first drug against DLL3 in small-cell lung cancer. AbbVie bought it for $5.8B and abandoned it after two failed phase 3 trials. The target later worked with a different weapon.
Lesson: target validation and modality are separable. DLL3 was the right address; a PBD payload with a narrow window in a frail population was the wrong weapon.
Tusamitamab ravtansine was Sanofi's ADC against the classic CEA tumour marker, stopped for futility in lung cancer in 2023.
Lesson: a tubulin-inhibitor payload at DAR ~4 with a non-permeable release mechanism could not match the TOP1-payload ADC bar set in lung cancer; CEACAM5 itself remains under study with T-cell engagers.
The biology did not hold
The mechanism looked sound in cells and mice but did not translate to patients.
Efficacy without a window
Activity was real but toxicity, or a safer competitor, made it unusable.
The first 'don't eat me' signal blocker. Gilead paid $4.9B for it; it was stopped in 2024 after trials showed more deaths, not fewer.
Lesson: a ubiquitous target (CD47 is on every red cell) plus an immunosuppressed population is a narrow window; single-arm response rates in TP53-mutant disease were not predictive of survival.
A HER2 ADC with a DNA-alkylating payload that beat chemotherapy in a phase 3 trial yet never reached the market, because eye and lung toxicity and a stronger rival arrived first.
Lesson: statistical significance is not enough when a competitor redefines the standard; payload toxicity profile decides whether an ADC survives.
Partner and assay decided it
The chemotherapy partner or the companion diagnostic, not the drug, determined the result.
Approved, then withdrawn
Accelerated approvals whose confirmatory trials disappointed or showed harm.
Other setbacks
Negative or withdrawn items without a single dominant lesson.
A double-blind test of a promising glioblastoma vaccine that showed no benefit, and taught the field how misleading historical-control comparisons can be.
The randomised test of an off-the-shelf KRAS vaccine after pancreatic surgery. It missed its primary goal in June 2026.
In healthy older people, daily low-dose aspirin did not extend disability-free life and, unexpectedly, was linked with more cancer deaths. It changed guidelines against routine aspirin in the elderly.
The 'don't eat me' signal cancer cells show to macrophages. Blocking it looked promising but the lead drug failed.
Biotech that ran one of the early GPNMB ADCs in triple-negative breast cancer, which failed, and has since pivoted to mast-cell (KIT) antibodies for allergic disease.
CheckMate 548, 143 and 498 were three large trials, all negative: the immunotherapy that transformed melanoma and lung cancer did nothing in glioblastoma.
Copanlisib is an intravenous PI3K inhibitor for relapsed follicular lymphoma, approved in 2017 and withdrawn in 2023 when its confirmatory trial failed.
The only randomised trial of a ketogenic diet in brain tumours found no benefit. Patients could follow the diet and their ketone levels rose, but progression came just as quickly.
ESSA Pharma developed masofaniten, an AR N-terminal-domain inhibitor whose phase 2 was stopped for futility in 2024-25.
Fianlimab is Regeneron's LAG-3 blocker. Promising early data with cemiplimab did not hold up against pembrolizumab in a first-line phase 3 in 2026.
Neoantigen-vaccine pioneer (GRANITE, SLATE) that ran out of money and filed for bankruptcy in October 2024; a cautionary tale for the personalised-vaccine field.
A drug designed to block the part of the androgen receptor that resistant variants keep; its phase 2 was stopped for futility and development ended.
Mobocertinib was the first oral drug for EGFR exon 20 insertion lung cancer, approved in 2021 and withdrawn in 2023-24 after its confirmatory trial failed.
An immunotoxin for hairy cell leukaemia that produced lasting remissions in relapsed patients but was withdrawn from sale in 2023 for commercial reasons.
CAR-NK company that shelved its oncology programmes after falling response rates and pivoted to autoimmune disease.
An HDAC inhibitor for relapsed myeloma approved in 2015 and withdrawn in 2021 after its confirmatory trial was never completed.
A vaccine against a glioblastoma-specific mutant protein that looked promising for years and then failed its phase 3 trial in 2016. It is a landmark failure.
Tazemetostat was the first EZH2 inhibitor, approved for epithelioid sarcoma and follicular lymphoma in 2020 and withdrawn worldwide in 2026 after secondary blood cancers.
TIGIT is an immune brake that looked promising, then failed in several big lung cancer trials. It is a cautionary tale.
The largest screening trial ever run for ovarian cancer found earlier detection but no reduction in deaths, so there is still no recommended screening.
A PI3K inhibitor for marginal zone and follicular lymphoma approved in 2021 and withdrawn in 2022 after the UNITY-CLL trial suggested more deaths.
The definitive test of whether vitamin D or fish-oil pills prevent cancer or heart disease in healthy adults. They did not.
Three thousand breast cancer survivors were coached for years to eat far more vegetables, fruit and fibre. They did, and it made no difference to recurrence or survival.