Talazoparib plus temozolomide in DNA-repair wild-type prostate cancer: toxicity scaled, response did not

What was tried

Memorial Sloan Kettering Cancer Center opened NCT04019327 in July 2019 as a single-group, open-label Phase 1b/2 study of intermittent talazoparib plus temozolomide in men with metastatic castration-resistant prostate cancer and no mutations in DNA damage repair genes. Talazoparib ran on days 1 to 6 of each cycle and temozolomide on days 2 to 8, placing the alkylating agent inside a window of sustained PARP inhibition.

Four dose levels opened: talazoparib 1.0 mg with temozolomide 37.5 mg/m2 (3 participants), 0.75 mg with 50 (3), 1.0 mg with 50 (3), and 1.0 mg with 75 (7). Sixteen men enrolled, and protocol levels running to 1.5 mg and 125 mg/m2 were never reached. Phase 2 primary endpoints were RECIST v1.1 response, PSA decline of 50% or more, and circulating tumour cell conversion to zero per 7.5 mL. The record moved to Terminated on 24 August 2026 with results posted.

The biological hypothesis

PARP inhibitors are licensed in prostate cancer for tumours with homologous recombination repair defects, and TALAPRO-2 concentrated its talazoparib benefit in the HRR-deficient population. This trial aimed at the HRR-proficient majority, which has no established route to that benefit.

The hypothesis was that the required DNA lesion could be supplied pharmacologically rather than inherited. Murai and colleagues showed that BMN 673, the development code for talazoparib, is roughly 100-fold more potent than olaparib or rucaparib at trapping PARP1 and PARP2 on DNA, and roughly 100-fold more cytotoxic combined with alkylating agents including temozolomide. Temozolomide methylates DNA at N7-guanine and N3-adenine, lesions cleared by base excision repair, which recruits PARP1 to the strand-break intermediate. Trapping PARP1 there turns a routine base lesion into a replication-blocking adduct and then a double-strand break, so sensitivity should track delivered alkylating dose rather than repair genotype. Veliparib plus temozolomide had produced 55% to 100% tumour growth inhibition across xenografts including prostate, uncorrelated with MGMT or mismatch repair status. Open Targets scores PARP1 against prostate carcinoma at 0.385, from literature and clinical evidence with no genetic component.

What actually happened

Toxicity scaled with dose. Posted counts of treatment-emergent toxicities were 10, 16, 25 and 95 across the four levels, or 3.3, 5.3, 8.3 and 13.6 per participant. Serious adverse events appeared only at the top level, in 4 of 7 participants there against 0 of 9 below it, led by platelet count decreased in 4 of 7 and febrile neutropenia in 2 of 7.

Efficacy did not follow. One participant of 16 achieved a confirmed response, at the lowest dose level. None reached a PSA decline of 50% or more. Two converted circulating tumour cells to zero. Accrual ran at about 2.5 participants per year over 6.4 years.

Failure mechanism, best guess

Both arms of the therapeutic index failed at once. This was not a drug that worked but could not be dosed: zero of 16 men reached the PSA threshold at any level, including the three where no serious adverse events occurred.

The toxicity ceiling has a clean explanation. Base excision repair and PARP1 recruitment are not tumour-specific, so haematopoietic progenitors run the pathway that makes trapped PARP1 lethal to a methylated tumour cell. Of 829 openFDA FAERS reports naming talazoparib as a suspect product, the leading reactions are anaemia (163), thrombocytopenia (66), febrile neutropenia (56) and pancytopenia (45). Temozolomide's profile is led by thrombocytopenia (1,419 of 18,252) and neutropenia (934). Two agents whose dominant toxicity is the same cytopenia do not combine into a wide window.

The efficacy floor points elsewhere. The population was defined by the absence of a marker rather than the presence of one, and "no mutations in DNA damage repair" enriches for nothing. Later work sharpened the point: temozolomide plus a PARP inhibitor is potently synthetic lethal in ARID1A-mutated cancers, which accumulate base lesions and abasic sites and show delayed recruitment of long-patch base excision repair effectors, while temozolomide alone does nothing there. The combination is genotype-dependent after all, just on a different genotype from the one being screened.

We classify this as translational_mismatch. The preclinical models measured potentiation in tumour cells with no marrow compartment, and the trial then defined its population by a negative.

How to prevent this next time

The data cannot support a Bayesian update or a power calculation. There is no control arm, no time-to-event data and no biomarker-stratified subgroup, so we use only what the counts permit. With 0 of 16 participants meeting the PSA endpoint, the one-sided exact binomial 95% upper bound on the true PSA50 rate is 1 minus 0.05 raised to the power 1/16, or 17.1%. The trial excludes a rate above roughly 17% and says nothing below it.

Three qualitative levers apply. Biomarker enrichment: a combination running through base excision repair should enrol on a base excision repair or replication-stress lesion, ARID1A status being one candidate, not on the absence of a homologous recombination lesion. Base-rate adjustment: the governing prior was the 2013 melanoma Phase 2, not the xenograft data. Rucaparib plus temozolomide there gave a 17.4% response rate in a far more alkylator-sensitive tumour type, with 25 of 46 patients requiring reduction for myelosuppression. Red-team review: which normal tissue shares the potentiating mechanism, and was a therapeutic index ever measured in a system containing it? Colony-forming assays on human CD34-positive progenitors would have produced an index rather than a potency.

The single highest leverage change would have been to enrol on a positive base excision repair or replication-stress biomarker instead of defining the population by the absence of a DNA damage repair mutation.

What this means for similar programs

PARP1 carries a Claidex mechanistic risk score of 40 of 100, yellow band. Saturation is the largest component at 13.0 of 15, reflecting 10 distinct PARP1-directed clinical-stage molecules in Open Targets, six of them approved. The genetic component adds 9.2 of 15, since the prostate carcinoma association carries no genetic anchor.

The read-across is narrow. Nothing here counts against PARP inhibition in HRR-deficient prostate cancer, which rests on randomised evidence. What it counts against is chemically manufacturing synthetic lethality in a repair-proficient tumour. Any programme pairing a high-trapping PARP inhibitor with an alkylator or platinum in an unselected population inherits the same overlap, because trapping potency and marrow toxicity are one property measured twice.

Open questions

Was a maximum tolerated dose declared, or did escalation stop with higher levels unopened? The posted results give toxicity counts by level but no dose-limiting toxicity determination.

Was archival tissue sequenced beyond the eligibility panel? An ARID1A or base excision repair analysis on these 16 tumours would test the enrichment hypothesis cheaply.

Sources

1. ClinicalTrials.gov NCT04019327, protocol and posted results, 29 August 2026. https://clinicaltrials.gov/study/NCT04019327 Murai J, et al. Mol Cancer Ther 2014.https://Palma JP, et al. Clin Cancer Res 2009.https://Plummer R, et al. Cancer Chemother Pharmacol 2013.https://Yu ZC, et al. Cancer Res 2023.https://Agarwal N, et al. Lancet 2023.https://Fizazi K, et al. Nat Med 2024.https://Open Targets API v4, ENSG00000143799, 29 August 2026. https://platform.opentargets.org/target/ENSG00000143799 ChEMBL CHEMBL3137320 and CHEMBL810, 29 August 2026. https://www.ebi.ac.uk/chembl/compound_report_card/CHEMBL3137320/ openFDA drug event API, talazoparib and temozolomide, 29 August 2026. https://api.fda.gov/drug/event.json.