A CK2 inhibitor reached medulloblastoma tissue at 264 nM and produced no responses in 11 patients

What was tried

The Pediatric Brain Tumor Consortium opened PBTC-053 (NCT03904862) in July 2019 to test CX-4945, also called silmitasertib, in children and adults whose sonic hedgehog subgroup medulloblastoma had recurred after standard therapy. CX-4945 is an orally bioavailable small molecule inhibitor of protein kinase CK2, registered in ChEMBL as CHEMBL1230165 with a molecular weight of 349.78 g/mol. Senhwa Biosciences supplied the compound, with the NCI and St. Jude Children's Research Hospital as collaborators.

The protocol carried three arms, all non-randomised and open label. A rolling-six dose-finding arm enrolled skeletally immature patients at 600 and 800 mg/m2 twice daily. A phase 2 arm enrolled skeletally mature patients at 600 mg/m2 or 1000 mg twice daily. A surgical arm dosed patients before planned resection so tumour tissue could be assayed directly for drug concentration.

Actual enrolment reached 21 of a larger target. The registry gives an administrative reason for stopping: closure followed word from the NCI that the PBTC grant would not extend beyond 31 March 2026. Accrual closed on 23 October 2025.

The biological hypothesis

CK2 is a constitutively active serine and threonine kinase that supports prosurvival signalling rather than driving transformation itself. The compound's founding paper called it a prototypical non-oncogene and showed that CX-4945 attenuates AKT signalling through the CK2-specific S129 site, blocks HIF-1 alpha transcription, and shrinks xenografts after oral dosing.

The medulloblastoma rationale added a second layer. CK2 phosphorylates GLI transcription factors downstream of sonic hedgehog signalling, so inhibiting it was expected to shut down pathway output in tumours whose upstream lesions in PTCH1 and SMO sit beyond the reach of smoothened antagonists.

One asymmetry sat under that logic from the start. Open Targets records 3,855 target-disease associations for medulloblastoma, led by SUFU at 0.805, CTNNB1 at 0.753, ELP1 at 0.721 and PTCH1 at 0.673. An explicit pair query for CSNK2A1 against medulloblastoma, childhood medulloblastoma and SHH-activated medulloblastoma returns no association row. The genetics point upstream, to components that mutate, not to the kinase that was drugged.

What actually happened

PBTC-053 posted results, making it one of the few terminated paediatric brain tumour studies whose numbers can be read directly.

Dose-limiting toxicity was uncommon. No DLTs occurred in either phase 1 cohort of three, and one occurred among six phase 2 patients. The maximum tolerated dose in skeletally immature patients is reported as not estimable. Safety was dominated by gastrointestinal events, with diarrhoea in 9 of 12 phase 2 patients and transaminase elevation in 4 of 12. One death is recorded there.

Efficacy did not materialise. The sustained objective response rate in skeletally mature patients was 0 percent across 11 evaluable participants, 95 percent confidence interval 0 to 28.5. Median progression-free survival was 1.8 months in every reported cohort.

Two correlative endpoints returned no data. The outcome measuring reduction in CK2-mediated signalling in surgical tissue has a denominator of zero participants, as does the endpoint correlating response with genomic alterations. The pharmacodynamic question the surgical arm existed to answer went unanswered.

The surgical arm did return the number that matters. Mean intratumoral CX-4945 was 92.5 ng/mL across three patients, range 54 to 131 ng/mL.

Failure mechanism, best guess

Converting that using the ChEMBL molecular weight of 349.78 g/mol gives 264 nM, range 154 to 374 nM. Set it against the compound's own potency data. The enzyme numbers are a CK2 alpha Ki of 0.38 nM and a recombinant CK2 IC50 of 1.0 nM, so tumour concentration sat roughly 700-fold above the biochemical binding constant. The cellular numbers invert the picture. Across 12 human cancer lines in a four-day antiproliferative assay, CX-4945 IC50 values ran from 1,100 to 8,900 nM, median 3,450 nM. The concentration achieved inside tumour tissue was about 13-fold below that median.

That gap is the most economical explanation for a 0 percent response rate in a drug that demonstrably crossed into tumour. A 2025 comparison with the more selective probe SGC-CK2-2 found that partial CK2 inhibition, enough to suppress the most sensitive phosphosites including AKT S129, does not measurably affect cell viability, while near-complete suppression does. Occupying the enzyme is not the same as collapsing the pathway.

Two complications deserve naming. CX-4945 is not a clean CK2 inhibitor, and structural work shows meaningful affinity for DYRK1A and GSK3 beta. The trial also closed for funding reasons before full accrual, so the efficacy estimate is imprecise even though it points at the null. The archetype recorded here is translational mismatch rather than sponsor decision, because the efficacy data were already null at closure and the pharmacodynamic endpoint that might have rescued it was never collected.

How to prevent this next time

Endpoint-level data exist here, so the quantitative lever is real rather than hypothetical. The exposure comparison above is a go and no-go criterion evaluable after three surgical patients. Set the bar in advance at the median cellular antiproliferative IC50 from the compound's own published panel, require measured intratumoral concentration to reach it, and treat failure as a dosing problem rather than a biology problem. In PBTC-053 that test fails by more than an order of magnitude on the first three patients assayed.

The second lever is sequencing. The surgical arm was the decisive one and ran parallel to the efficacy arms rather than ahead of them. Running it first, with CK2 substrate phosphorylation measured in paired tissue, turns a 21-patient null study into a 3-patient answer about whether the mechanism engages.

The single highest leverage change would have been gating phase 2 expansion on a prespecified intratumoral concentration threshold taken from the compound's own cellular potency panel and measured in the first three surgical patients, rather than running the surgical and efficacy arms in parallel.

What this means for similar programs

CK2 remains a thin clinical field. ClinicalTrials.gov returns 11 interventional studies matching CK2 inhibitor terms, and the drug arms are almost entirely CX-4945 under its various names, so target saturation is low.

The lesson generalises past CK2. Any programme built on suppressing a node that is active but not mutated carries two burdens at once. It must show that partial inhibition produces a phenotype, and that achievable tissue exposure clears the cellular rather than the biochemical potency threshold. Non-oncogene targets tend to have shallow cellular dose-response curves, so the margin between occupancy and effect is wide and has to be bought with exposure.

Open questions

Would higher intratumoral exposure have changed the outcome, or does near-complete CK2 suppression carry toxicity that puts the required concentration out of reach? The missing maximum tolerated dose leaves this unresolved.

How much of the compound's preclinical activity depended on DYRK1A and GSK3 beta inhibition rather than CK2, and would paired tissue have shown substrate dephosphorylation at 264 nM? No clinical data answer either question.

Sources

1. • ClinicalTrials.gov, NCT03904862 (PBTC-053), record and posted results. https://clinicaltrials.gov/study/NCT03904862 - ChEMBL CHEMBL1230165 (silmitasertib): weight, Ki, IC50, 12-line panel. https://www.ebi.ac.uk/chembl/compound_report_card/CHEMBL1230165/ - Open Targets Platform, CSNK2A1 and medulloblastoma (MONDO_0850197), queried 26 October 2026. https://platform.opentargets.org/ - Siddiqui-Jain A et al. Cancer Res 2010, 70:10288-98. https://- Noventa F et al. Int J Mol Sci 2025, 26:10006. https://- Grygier P et al. J Med Chem 2023, 66:4009-24. https://- Zhong B et al. J Chromatogr B 2020, 1152:122254. https://Article metadata from PubMed.