Tumor-on-a-Chip: Rebuilding the Microenvironment, Not Just the Cells

Team Dynamic Matrices | 10 October 2025

Early tumour models optimised for one thing: keeping cancer cells alive outside the body. Tumour-on-a-chip platforms optimise for something harder: capturing how a tumour behaves inside a specific mechanical and cellular context. The reason is that context, not just cell identity, drives invasion, immune evasion and drug resistance. Here is what we think the evidence supports, and what it means for the matrix you put in the chip.

The microenvironment was the missing variable

A tumour in vivo is not a homogeneous mass. It sits in a stiffening, remodelling ECM, threaded with leaky, chaotic vasculature, surrounded by fibroblasts and immune cells in constant crosstalk with the tumour. Spheroids and organoids capture some of the cellular complexity but strip away mechanics and vascular architecture. Tumour-on-a-chip exists to put those back: microfluidics, a biomimetic matrix and controlled perfusion in one device (reviewed by Sontheimer-Phelps, Hassell and Ingber, Nature Reviews Cancer, 2019).

Vasculature changes the question

Drug delivery in vivo is a transport problem before it is a pharmacology problem. Sobrino et al. (Scientific Reports, 2016) built perfusable, self-assembled vascular networks around tumour cells and showed drugs delivered through the vessels produced different responses from direct exposure. Nashimoto et al. (Biomaterials, 2020) vascularised tumour spheroids on chip and demonstrated that perfusion through the network changed proliferation and drug sensitivity. A model without vasculature can only ask "does this drug kill these cells". A model with perfusable vessels asks "does this drug reach and kill these cells under realistic transport constraints", which is the question that predicts clinical performance.

Matrix stiffness is not a passive backdrop

Tumours are stiffer than their tissue of origin, and that stiffness is causal. Levental et al. (Cell, 2009) showed that collagen crosslinking and ECM stiffening drive integrin clustering, focal adhesions and malignant progression in mammary tissue. Northey, Przybyla and Weaver (Cancer Discovery, 2017) reviewed how tissue force programs aggression and therapy resistance. On the immune side, Nicolas-Boluda et al. (eLife, 2021) showed that softening the tumour ECM with a LOX inhibitor improved T cell migration and the efficacy of anti-PD-1 therapy in mice. That result reframes what a tumour matrix is for in a chip: it is an active input that shapes how immune cells behave once they arrive, which is directly relevant to why some immunotherapies fail in stiff, fibrotic tumours and succeed in softer ones.

A chip that cannot tune matrix stiffness independently of composition cannot ask this question, because stiffness and cell identity are confounded in any model where the matrix is fixed or undefined.

Immune-competent chips are the frontier

The most recent wave is explicitly immune-focused. Aung et al. (Cancer Research, 2020) monitored T cell infiltration into a 3D tumour microenvironment on chip in real time. Pavesi et al. (JCI Insight, 2017) used a microfluidic model to quantify TCR-engineered T cell killing of hepatocellular carcinoma under hypoxia. Chips profiling CAR T cell trafficking, activation and exhaustion inside a defined matrix are now a recognised category. This matters commercially as well as scientifically: immuno-oncology holds a large share of current development spend, and a preclinical model that cannot represent immune cell trafficking in a realistic matrix is a poor predictor for exactly the therapies the field most needs to de-risk.

Why the matrix has to be tunable, not just present

The thread across all of this work: including a matrix is not enough. It has to be independently tunable in stiffness, stress relaxation, degradability and ligand composition, so you can isolate which property drives a given behaviour instead of accepting whatever an off-the-shelf gel provides. Collagen I gels, the default in most vascularised chips, couple stiffness to fibre density and cannot be stiffened without changing everything else. Basement membrane extracts add lot variation and undefined growth factors on top.

This is where chemically defined, programmable matrices have a structural advantage. In DyNAtrix, stiffness is set by polymer content and stress relaxation by DNA crosslinker design, so each can be varied on its own, and the gel can be stiffened in situ by adding a DNA strand to model progressive fibrosis without disturbing the cells. We designed it to load at room temperature and gel thermoresponsively in a channel, and to dissolve on demand with a nuclease so tumour and immune cells can be recovered intact for phenotyping. Our organ-on-chip application note covers loading.

Building the tumour's context, deliberately

Tumour-on-a-chip has matured past the point where the interesting question is whether cancer cells survive on a device. The question now is whether the device recreates enough real mechanical and vascular context to give trustworthy answers about invasion, immune interaction and delivery. That means treating the matrix as a first-class design element rather than a support material chosen for convenience. The chips that get this right will be the ones that predict how a therapy performs in the stiffening, imperfectly perfused tissue it will eventually have to work in.