What makes this study different is that it treats glioblastoma relapse like leftovers, not dinner: instead of asking how the tumor behaves on day one, it asks how to recreate the miserable moment when the dish comes back after treatment, somehow meaner than before.
That matters because glioblastoma, the roughest customer in the brain-tumor kitchen, almost never stays gone for long. Surgeons remove what they can, radiation turns up the heat, temozolomide adds the chemo seasoning, and still the tumor often returns. The new review by Hernandez Tellez and colleagues is basically a serious attempt to answer a very practical question: if relapse is the part that keeps hurting patients, why are so many lab models still only cooking the first round? (Hernandez Tellez et al., 2026)
The old recipe was a little too tidy
For years, a lot of glioblastoma research leaned on 2D cell cultures. That means tumor cells spread out flat in a dish like spilled batter. Those systems are cheap, fast, and handy for testing lots of drugs. No shame there. Every kitchen needs a sheet pan.
But flat cultures miss a lot of the messy business that makes glioblastoma such a menace in real life. Tumors in patients are not polite single-layer pancakes. They are dense, mixed, chaotic casseroles with cancer cells, support cells, immune cells, blood vessel signals, oxygen shortages, and a blood-brain barrier acting like the world’s pickiest maître d'. Reviews over the past few years have hammered this point: simpler models help with speed, but they often fail to mimic the conditions that shape resistance and recurrence (Boccellato and Rehm, 2022; Stankovic et al., 2021).
Why relapse is the burnt part nobody can ignore
Here is the key move in this paper: relapse is not just “the same tumor, again.” It is the tumor after surgery, after radiation, after chemo, after a survival contest that selected the hardest, weirdest, most stubborn cells still standing. In cooking terms, you are no longer tasting the fresh sauce. You are tasting what survived the boil, the chill, the reheat, and the bad decisions.
The review argues that a useful relapse model has to copy that timeline. You need treatment phases that resemble real clinical care, then a break in treatment, then time for regrowth. That pause matters. If you never let the cells recover and come back, you are not really modeling relapse. You are just repeatedly poking the original batch and hoping it confesses.
That sounds obvious once stated, which is often how science works. Somebody spends years proving the thing that makes everyone else say, "Well yes, of course," while quietly realizing they had not done it properly.
From meatloaf to layered lasagna to tiny chip kitchens
The field has been moving from 2D cultures to 3D spheroids, organoids, microfluidic systems, and organ-on-chip platforms. Each step adds more of the real recipe back in.
Spheroids and organoids let tumor cells grow in clumps and structures that better preserve heterogeneity, gradients of oxygen and nutrients, and some features of the tumor microenvironment. Patient-derived explant organoids are especially appealing because they can keep more of the original tumor’s architecture and cell diversity, at least for a while. That makes them attractive for personalized therapy testing, which is basically precision oncology’s version of asking, "Can we taste this before serving it to the whole table?" (Skarne et al., 2025)
Then come organ-on-chip systems, which sound sci-fi because they are. These platforms use tiny channels and controlled fluid flow to mimic blood supply, tissue boundaries, and treatment exposure more realistically than a static dish can. Recent reviews suggest glioblastoma-on-a-chip models may help researchers study drug delivery, the blood-brain barrier, and therapy resistance in a setting that behaves more like an actual human tissue environment (Maity et al., 2025).
Why this paper is worth your time
Hernandez Tellez and colleagues are not claiming they built the final perfect relapse model. They are offering a roadmap. That may sound less flashy than a miracle therapy, but honestly, the field needs better recipes before it can expect better meals.
If these models improve, they could help researchers do three useful things. First, identify which treatments still work after the tumor has adapted. Second, test combinations in a setup that resembles the patient journey instead of a lab fantasy camp. Third, inch closer to patient-specific treatment decisions, where a person’s own tumor sample helps guide what gets tried next.
That hope comes with caveats, because glioblastoma loves caveats. Even the fanciest organoid or chip cannot fully capture a living human brain, immune system, and treatment history. These models can be expensive, finicky, and hard to standardize across labs. And no model, however artisanal, can guarantee clinical success. A recent review of clinical trials in newly diagnosed and recurrent glioblastoma makes plain that progress for patients remains slow and uneven (Fukushima and de Groot, 2024).
Still, this paper lands on an idea that feels both humble and sharp: if relapse is the meal that keeps sending patients back to the hospital, then research should stop practicing only on the appetizer.
References
Hernandez Tellez C, Camenisch S, Wick P, Sanchez-Arreola E, Ayala-Nunez V. Modeling glioblastoma relapse in vitro: a critical journey from 2D models to organ-on-chip alternatives. NPJ Precision Oncology. 2026 Apr 27. doi: 10.1038/s41698-026-01403-3. PubMed: https://pubmed.ncbi.nlm.nih.gov/42045324/
Boccellato C, Rehm M. Glioblastoma, from disease understanding towards optimal cell-based in vitro models. Cell Oncol (Dordr). 2022;45(4):527-541. doi: 10.1007/s13402-022-00684-7. PMCID: PMC9424171
Stankovic T, Ranđelović T, Dragoj M, et al. In vitro biomimetic models for glioblastoma-a promising tool for drug response studies. Drug Resist Updat. 2021;55:100753. doi: 10.1016/j.drup.2021.100753. PubMed: https://pubmed.ncbi.nlm.nih.gov/33667959/
Maity S, Bhuyan T, Jewell C, et al. Recent Developments in Glioblastoma-On-A-Chip for Advanced Drug Screening Applications. Small. 2025;21(1):e2405511. doi: 10.1002/smll.202405511. PMCID: PMC11719323
Skarne N, D'Souza RCJ, Palethorpe HM, et al. Personalising glioblastoma medicine: explant organoid applications, challenges and future perspectives. Acta Neuropathol Commun. 2025;13:6. doi: 10.1186/s40478-025-01928-x. PMCID: PMC11724554
Fukushima CM, de Groot J. Updates for newly diagnosed and recurrent glioblastoma: a review of recent clinical trials. Curr Opin Neurol. 2024;37(6):666-671. doi: 10.1097/WCO.0000000000001320. PMCID: PMC11540275
Disclaimer: The image accompanying this article is for illustrative purposes only and does not depict actual experimental results, data, or biological mechanisms.