A KAIST Barrier Chip Separated Three Glioblastoma Patients That MGMT Testing Called Identical
The argument is that the delivery route varies as much as the tumour does, and the sample size is three. A KAIST-led team built a blood-brain tumour barrier chip from patient-derived glioblastoma cells, applied temozolomide and bevacizumab, and reports that both barrier characteristics and drug response differed across three patients whose MGMT promoter methylation status was the same.
Patient-derived glioblastoma cells were co-cultured with brain vascular endothelial cells and astrocytes inside a microfluidic chip, reconstructing the boundary where tumour tissue meets normal brain and the vascular barrier a drug must cross to reach it. Chips were built from three patients. Temozolomide and bevacizumab, the standard agents, were applied and responses compared against each patient’s subsequent clinical course. The architecture also accommodates perivascular and immune cells.
Findings
- All three patients returned the same result on MGMT promoter methylation testing, the standard genomic biomarker in glioblastoma, and would conventionally have been expected to respond similarly.
- On chip, both the vascular barrier characteristics and the drug responses differed patient to patient. The team reports a high level of agreement between chip response and actual clinical course, without publishing a concordance statistic in the announcement.
- The claimed advance is what the platform measures. Rather than asking only whether tumour cells are sensitive to a drug, it asks whether the drug arrives, which existing genomic prediction does not address.
- The proposed use is sequential rather than diagnostic: build chips from a patient’s own tumour, test several agents, and select before treating. The same construct is offered as a screening environment for new candidates.
Three patients cannot establish predictive performance, and the team says so. What survives the small number is the framing. Glioblastoma trials stratify on MGMT status and little else, which means a failed trial cannot distinguish patients whose drug never crossed the barrier from patients whose drug crossed and did not work. Those are different failures requiring different responses, one a formulation and delivery problem and the other a target problem, and the field currently has no routine way to tell them apart. If barrier state proves to be a genuine axis of variance, it is a stratification variable rather than a new drug, which is a cheaper thing to add to a protocol and a harder thing to commercialise.
Three patients, with no reported concordance statistic, no confidence interval and no prospective test. Chip responses were compared against clinical courses already known, so this is retrospective agreement rather than demonstrated prediction. Bevacizumab acts on the vasculature itself, which makes a vascular chip an unusually favourable place to observe between-patient differences in its effect. The immune compartment is described as accommodated rather than included in the reported work.
No competing interests, funding sources or commercial vehicle were stated in the institutional announcement, and the journal article was not accessible for verification. No company is named against the platform.
Ryoo M, et al. Human Blood-Brain Tumor Barrier on a Chip to Investigate Personalized Treatment for Glioblastoma Patients. Small, DOI 10.1002/smll.202506712. Announced by KAIST on 18 August 2026 with Sungkyunkwan University, CHA Bundang Medical Center and CHA University. The publisher blocked automated access to the full text, so the figures above rest on the institutional announcement rather than the paper.