Same Biomarker, Different Drug Responses: Brain Tumor Barrier Chip Captures Patient Differences

The challenge of treating glioblastoma lies not only in whether a drug can kill cancer cells, but also in whether it can cross the still partially intact vascular barrier surrounding the tumor. A Korean research team has now incorporated both issues into a single microfluidic chip, seeking to reproduce the barrier characteristics and drug responses at the edge of a patient’s tumor before treatment begins.

The study, conducted jointly by the Korea Advanced Institute of Science and Technology, Sungkyunkwan University, CHA Bundang Medical Center, and CHA University School of Medicine, was published in *Small* on June 27. On the chip, human brain microvascular endothelial cells form a vascular channel in the upper layer, while normal astrocytes and patient-derived glioblastoma cells are co-cultured in a three-dimensional matrix in the lower layer. An expanded model also incorporates pericytes and microglia to simulate interactions among the tumor, glial cells, vasculature, and immune environment.

The study focused on the edge where the tumor infiltrates normal brain tissue, rather than the tumor core. The team observed that interactions between the tumor and surrounding cells altered vascular permeability, intercellular tight junctions, genes associated with drug efflux, and astrocyte states. Chips from different patients also exhibited different degrees of barrier integrity.

The main comparison included three newly diagnosed patients. All three tumors were IDH-wildtype and had MGMT promoter methylation—the latter is generally regarded as a biomarker indicating a greater likelihood of responding to the alkylating agent temozolomide. Nevertheless, the chips showed clear differences in temozolomide sensitivity: the patient B model had the strongest response, patient C’s was intermediate, and patient A’s was weaker. When the anti-angiogenic antibody bevacizumab was used, the B model showed a significant response, the A model had no detectable effect, and the C model showed only a slight trend that did not reach statistical significance.

These differences broadly matched the patients’ subsequent clinical courses. The A model responded least to both drugs, and the corresponding patient had the shortest progression-free survival during temozolomide treatment and the shortest survival after receiving bevacizumab following disease progression. The B model responded best, and its patient had the longest duration in both phases of the clinical course, while C fell between the two. Another model without MGMT methylation showed no clear response to temozolomide, also consistent with the general clinical trend.

The study also showed that a more “leaky” barrier does not necessarily mean treatment will be more effective. Although the A model had greater permeability, its tumor cells remained insensitive to both drugs. The B model had a relatively intact barrier, yet showed the strongest drug response. This indicates that efficacy may depend simultaneously on drug penetration, efflux transport, and the intrinsic sensitivity of cancer cells, and that tumor markers such as MGMT or imaging findings alone may not be sufficient to capture these differences.

However, the ability to “predict” remains a potential that requires validation, rather than an established clinical capability. The core comparison included only three patients, and the study examined whether the chip results aligned with clinical courses that were already known. The fixed drug concentrations and exposure times used in vitro also cannot fully reproduce human metabolism, immune responses, and dose variations. Before the chip can enter the drug-selection process, larger, independent, prospective patient cohorts will be needed to confirm the manufacturing success rate, consistency of results, turnaround time, and whether chip-guided decisions can actually improve treatment outcomes.