Korean researchers develop microfluidic chip to predict personalized glioblastoma treatment responses

Two patients walk into a clinic with the same diagnosis, the same genetic markers, the same prescribed drug. One improves. One does not. For years, doctors have had no way to predict which outcome belongs to whom.

Glioblastoma, one of the brain's most aggressive cancers, spreads rapidly through normal tissue and resists treatment in ways that vary wildly from person to person. The disease is lethal partly because of its speed, partly because the brain itself is defended by a barrier—the blood-brain barrier—that evolved to keep toxins out. The problem is that it keeps medicine out too. When a tumor forms, this barrier changes shape around it, but how much it changes, and how that affects drug delivery, remains invisible to current diagnostic tools.

A research team at KAIST, led by Professor Song Ih Ahn of the Department of Mechanical Engineering, working with collaborators at Sungkyunkwan University, CHA Bundang Medical Center, and CHA University, has built something that makes that invisible landscape visible: a microfluidic chip that recreates a patient's own tumor cells alongside the vascular environment that surrounds them. The chip is small enough to hold in your hand. Inside it, patient-derived glioblastoma cells are grown together with brain blood vessel cells and astrocytes—the supporting cells that form part of the barrier. The design also incorporates immune cells and perivascular cells, allowing the tumor's actual microenvironment to take shape in miniature.

The team tested the chip using tumor samples from three glioblastoma patients. All three had identical results on the standard genetic test used to predict treatment response: the MGMT promoter methylation biomarker. Conventional medicine would have predicted they would all respond similarly to treatment. But when the researchers applied the two standard glioblastoma drugs—temozolomide and bevacizumab—to the chips, the responses diverged. The vascular barrier characteristics differed from patient to patient. The drug responses differed. And when the researchers compared the chip results to what actually happened in those patients' clinical courses, the predictions aligned closely with reality.

This is the core insight: two patients with identical genetic profiles can have entirely different treatment outcomes because the physical structure of the barrier around their tumor is different. Conventional biomarkers cannot see this. They measure the tumor's genetic code but miss the architecture of the space the drug must travel through to reach it. The chip captures both.

The implications are substantial. If the platform proves reliable across larger patient populations, it could transform how glioblastoma treatment is chosen. Instead of prescribing a standard drug and hoping it works, doctors could grow a chip from a patient's own tumor cells, test multiple drugs on that chip, and select the one most likely to succeed before administering anything. The same platform could accelerate the discovery of new drugs by allowing researchers to screen candidates against patient-derived tumor environments rather than generic cell lines.

Professor Song Ih Ahn described the work as meaningful because it presents a way to evaluate treatment response that reflects the actual complexity of a patient's disease. The next step is validation in a larger cohort of patients. If that succeeds, the chip could become a standard tool in personalized cancer medicine—a small device that answers the question doctors cannot currently answer: which patient gets which drug, and why.