Onvansertib in second-line pancreatic cancer: a Phase 2 closed to move PLK1 to first line

What was tried

Cardiff Oncology opened NCT04752696 in May 2021 as an open-label, non-randomized, sequential Phase 2 study of onvansertib added to nanoliposomal irinotecan, leucovorin and fluorouracil in second-line metastatic pancreatic ductal adenocarcinoma. Eligible patients had histologically or cytologically confirmed metastatic disease, measurable lesions by RECIST v1.1, and exactly one prior gemcitabine-based line. A safety lead-in dosed onvansertib orally at 12 mg/m2 on days 1 to 10 of two-week cycles before a treatment period opened at the selected schedule alongside nal-IRI at 70 mg/m2. Actual enrollment reached 41 participants. The primary endpoint was investigator-assessed overall response rate. Primary completion was recorded on 24 December 2024, and the record was switched to terminated with the reason given as a sponsor decision based on meeting essential enrollment and obtaining substantial data to support future PDAC trials in a first-line setting.

The biological hypothesis

Onvansertib (ChEMBL1738758, also NMS-1286937 and PCM-075) is an oral small-molecule inhibitor of polo-like kinase 1, the mitotic kinase that governs centrosome maturation, bipolar spindle assembly and cytokinesis. The rationale for pancreatic cancer rests on two preclinical lines. PLK1 inhibition potentiates gemcitabine-induced killing of pancreatic cancer cells (doi:10.1080/15384101.2016.1148838), and PLK1 sits inside a synthetic-lethal relationship with mutant KRAS, which is mutated in the large majority of PDAC (doi:10.1038/ncomms11363). Later work reported that Plk1 blockade sensitizes pancreatic tumors to checkpoint inhibition (doi:10.1158/0008-5472.CAN-22-0018).

The human evidence is thinner than the preclinical narrative. Open Targets scores the PLK1 to PDAC association at 0.123, and the only contributing datatypes are literature at 0.818 and clinical at 0.155. There is no genetic association evidence linking PLK1 to pancreatic cancer. The same platform flags PLK1 as an essential gene with a loss-of-function constraint in the most constrained bin, which is the signature of a target whose inhibition is hard to confine to tumor tissue.

What actually happened

The sponsor posted results. In the treatment-period cohort, 28 participants were evaluable. Overall response rate was 10.7 percent (95% CI 2.3 to 28.2) with three confirmed responders, disease control rate was 53.6 percent (95% CI 33.9 to 72.5), median duration of response among responders was 5.0 months (95% CI 4.9 to not reached), and median overall survival was 7.0 months (95% CI 4.3 to 10.8). The two small safety lead-in cohorts, of four and two evaluable participants, recorded no responses.

Toxicity was heavy and gastrointestinal. All 33 participants dosed in the treatment period had a treatment-emergent adverse event, 23 of 33 had a serious adverse event, and 24 of 33 died during the post-treatment follow-up period in a disease with a short natural history. The most frequent non-serious events were nausea in 25 of 33, diarrhoea in 22 of 33, constipation in 16 of 33 and vomiting in 13 of 33. Serious gastrointestinal events included diarrhoea, colitis and ileus.

For scale, the registrational trial that established this backbone, NAPOLI-1, reported median overall survival of 6.1 months (95% CI 4.8 to 8.9) for nal-IRI plus fluorouracil and folinic acid against 4.2 months for fluorouracil and folinic acid alone (doi:10.1016/S0140-6736(15)00986-1). The 95% confidence interval observed here contains that benchmark across its full width.

Failure mechanism, best guess

This was not a safety stop and not a declared futility stop. It reads as a setting problem that the design could not detect.

The clearest clue comes from the same drug in colorectal cancer. In second-line KRAS-mutant mCRC, onvansertib with FOLFIRI and bevacizumab produced a confirmed ORR of 26.4 percent (95% CI 15.3 to 40.3), but a post hoc split showed 76.9 percent in bevacizumab-naive patients against 10.0 percent in bevacizumab-exposed patients, with median progression-free survival of 14.9 against 6.6 months. The authors reported translational and preclinical data indicating that onvansertib inhibits the hypoxia pathway and that its antitumor effect depends substantially on combination with an anti-angiogenic agent (doi:10.1200/JCO-24-01266). Cardiff discontinued its randomized second-line colorectal study in February 2024 and redirected the asset to first line, citing the same hypothesis being tested twice and regulatory guidance to focus on the earlier setting.

The pancreatic study carried neither of the conditions associated with activity in that model. There was no anti-angiogenic partner in the regimen, and every participant was pretreated by design. A pan-essential mitotic target was asked to produce single-agent-like increment on top of an already myelosuppressive and gastrointestinal backbone, in a population with limited marrow and performance reserve. The result was a response rate indistinguishable from the backbone and dose intensity competing with toxicity.

How to prevent this next time

Endpoint-level data here are single-arm, so the usable levers are structural rather than statistical, and the arithmetic that does apply is simple interval reasoning.

First, base-rate adjustment. A 28-patient single-arm cohort yields a 95% confidence interval on ORR of 2.3 to 28.2 and on median OS of 4.3 to 10.8. The historical control sits at 6.1 months. An interval that wide cannot rule a clinically meaningful increment in or out, which means the study could not have changed the development decision regardless of outcome. Deciding the width of the answer before enrolling would have shown that.

Second, biomarker and mechanism enrichment. The colorectal program had already generated a usable enrichment hypothesis, prior anti-angiogenic exposure, and a mechanistic account tied to hypoxia signaling. No equivalent stratifier was prespecified in the pancreatic study, and no anti-angiogenic partner was included.

Third, red-team the setting rather than the molecule. The sponsor eventually relocated both second-line programs to first line. That conclusion was available from the colorectal data in early 2024, before the pancreatic study completed.

The single highest leverage change would have been to make the pancreatic study randomized against the nal-IRI backbone alone, or to not run it in the second-line setting at all until the anti-angiogenic combination hypothesis had been tested.

What this means for similar programs

PLK1 is a crowded and repeatedly disappointing target. Open Targets lists eight clinical candidates against it, including volasertib at Phase 3 and BI-2536, GSK-461364, TAK-960, HMN-214, MK-1496 and cafusertib at earlier stages, with no approved agent. The Claidex mechanism risk score for PLK1 after this entry is 39 of 100, band yellow, built from a phase-burden component of 8.05, an archetype component of 2.0, a recency component of 4.25, a genetic-deficit component of 13.15 and a saturation component of 11.97. The profile is characteristic: low archetype severity because the program was reprioritized rather than refuted, high genetic deficit because no human genetics connect PLK1 to this disease, and high saturation because many sponsors have already tried.

Programs built on pan-essential mitotic kinases should treat the absence of human genetic support as a design constraint, not a gap to be filled with preclinical synergy data. The combination partner, not the target, is often what carries the effect.

Open questions

Would the same regimen with an anti-angiogenic agent reproduce the colorectal pattern in PDAC, and is there a measurable hypoxia-pathway readout that would identify responders prospectively? Were the three responders enriched for any molecular feature, and were KRAS variant allele frequency or other circulating measures collected in this cohort? Does the first-line pancreatic strategy that this study was stopped to support carry a prespecified biomarker, or does it repeat the all-comers design at an earlier line?

Sources

1. • ClinicalTrials.gov record and posted results for NCT04752696, Cardiff Oncology, accessed 2026-11-08. https://clinicaltrials.gov/study/NCT04752696 - Open Targets Platform GraphQL API v4, target ENSG00000166851 (PLK1), association with MONDO_0005184, essentiality, constraint and clinical-candidate list, accessed 2026-11-08. - ChEMBL molecule record CHEMBL1738758 (onvansertib), accessed 2026-11-08. - Wang-Gillam A, et al. Nanoliposomal irinotecan with fluorouracil and folinic acid in metastatic pancreatic cancer after previous gemcitabine-based therapy (NAPOLI-1). Lancet 2016;387:545-557.- Onvansertib in combination with chemotherapy and bevacizumab in second-line treatment of KRAS-mutant metastatic colorectal cancer. J Clin Oncol 2025.- Onvansertib in combination with FOLFIRI and bevacizumab in second-line treatment of KRAS-mutant metastatic colorectal cancer. Clin Cancer Res 2024.- Suppression of KRas-mutant cancer through the combined inhibition of KRAS with PLK1 and ROCK. Nat Commun 2016.- Plk1 inhibition enhances the efficacy of gemcitabine in human pancreatic cancer. Cell Cycle 2016.- Targeting Plk1 sensitizes pancreatic cancer to immune checkpoint therapy. Cancer Res 2022.- Cardiff Oncology clinical update on the randomized second-line ONSEMBLE trial in RAS-mutated mCRC, 29 February 2024. https://investors.cardiffoncology.com/news-releases/news-release-details/cardiff-oncology-provides-clinical-update-phase-2-randomized - openFDA FAERS queried for onvansertib on 2026-11-08: no matching reports returned.