The parts of a tumour a screen throws away

What the work claims

The central argument is that therapeutic resistance in solid tumours is not only a property of mutations inside cancer cells but a property of the tissue those cells build around themselves.¹ Therapy itself, the review argues, remodels that tissue: it changes which stromal and immune cells are present, stiffens and rearranges the extracellular matrix, reshapes the vasculature, rewires metabolism and alters the signals cells exchange, and these changes together create niches where tumours evade immunity, resist drug delivery, protect stem-like populations and survive. It is a position piece with a clear thesis, that resistance is an adaptive, environmental process, and it marshals mechanistic examples to support it.

As a claim about biology this is mainstream rather than bold; few oncologists would dispute that the microenvironment matters. What makes it worth an analysis is not novelty. It is that the review, almost incidentally, itemises the machinery of resistance in enough detail to serve as a checklist, and that checklist has sharp consequences for anyone who builds tumour models to discover drugs.

How it works

The review organises the microenvironment into cellular and acellular parts. On the cellular side sit cancer-associated fibroblasts, tumour-associated macrophages, myeloid-derived suppressor cells, endothelial cells and pericytes; on the acellular side, the extracellular matrix, hypoxic regions, metabolic products and cytokine networks. Resistance, in this telling, emerges from the dynamic interplay of these parts under treatment pressure.

Several concrete mechanisms carry the argument, and they are worth naming because they show how specific the pathways are. The review describes temozolomide, a standard glioblastoma chemotherapy, provoking release of the metabolic enzyme ENO1 through autophagy, which then activates a TLR4 to PI3K-Akt and ERK-SPHK1 signalling cascade that polarises macrophages toward an immune-suppressing state. In ALK-rearranged lung adenocarcinoma it describes fibroblast-derived growth factors, hepatocyte growth factor and neuregulin-1, reprogramming tumour lipid metabolism to blunt targeted therapy. In colorectal cancer it describes fibroblast exosomes lowering sensitivity to the EGFR inhibitor gefitinib by driving an epithelial-to-mesenchymal transition. The common shape is that a non-tumour cell, or a physical feature of the tissue, changes how the cancer cell responds to a drug.

Underpinning these are the general processes the review foregrounds: matrix mechanotransduction, in which a stiffer matrix is itself a survival signal; hypoxia-driven signalling; epigenetic and metabolic reprogramming; and communication carried by extracellular vesicles. Its translational half then surveys strategies aimed at the environment rather than the tumour cell, namely normalising the stroma, reprogramming macrophages, modulating metabolism, normalising the vasculature and using nanotechnology to force drug delivery. Toward the end the review turns to tools, and explicitly names patient-derived organoids co-cultured with immune and stromal components, tumour-on-chip systems and spatial omics as the way to capture what it calls the intrinsic limitation of static models in seeing the true, evolving state of the tumour ecosystem.

Where a skeptic should push

A review is only as strong as the selection behind it, and the load-bearing assumption here is that the collected examples are representative of clinical resistance rather than a curated set that fits the thesis. That assumption is never tested. The mechanistic vignettes are each drawn from individual studies, several of them preclinical or in a single tumour type, and the connective claim that they converge to drive multimodal resistance is a narrative one. Nowhere does the review offer an effect-size budget, meaning an estimate of how much of clinical resistance is environmental versus cell-intrinsic. Without that, the reader is asked to accept importance without magnitude.

The translational optimism deserves the hardest push. Targeting the microenvironment reads well on paper and has a genuinely mixed record in patients. The most cautionary lesson in the field is that stromal depletion, pursued on exactly the reasoning this review advances, has in pancreatic cancer accelerated rather than restrained disease: genetically removing the hedgehog signal, or depleting the fibroblasts, produced more aggressive tumours and shorter survival, because the stroma was partly restraining the tumour, not only shielding it. I am stating that as well-established background from the pancreatic literature rather than from anything inside this review, and the reader should treat it as a directional caution, but it is the right corrective to a section that lists environmental strategies mostly by their promise. Normalisation is not obviously safer than destruction; a matrix or a vessel network you perturb can help the tumour as easily as hurt it.

What is solid should still be credited. The individual mechanisms cited are real and mechanistically specific, matrix stiffness as a mechanotransduced survival signal is well founded, and the review's core claim, that resistance is dynamic and environmental as well as genetic, is correct even if under-quantified. The discipline is to read it as a well-organised map of possibilities rather than as an argument that the environment dominates, which it asserts more than it shows.

Where tumour resistance hides from the model

Read this review beside a bench and it stops being about oncology and becomes about instrumentation. Every mechanism it catalogues is a mechanism that requires a specific non-tumour component to exist. Cancer-associated fibroblasts, macrophages, endothelium, a stiff matrix, a hypoxic gradient, a flux of extracellular vesicles: each is a part that a tumour-cell-only organoid simply does not have. The review is, in effect, a specification for what a drug-discovery model must contain to be valid, written from the resistance side.

The direct consequence is a directional bias, and it runs one way. A model missing the compartment that confers resistance cannot register that resistance, so it will tend to call a drug more effective than it will prove in a patient. A tumour-cell-only screen therefore tends toward false-sensitive results for any agent whose real-world failure is environmental, and the review names the categories precisely: immune-excluded tumours where macrophages and suppressor cells do the shielding, desmoplastic cancers where a dense matrix blocks delivery, and settings where the vasculature governs whether a drug arrives at all. For those, a clean kill in a well is not reassuring; it is uninformative.

This is the mirror image of the lesson from functional organoid studies, where resistance to cytotoxic chemotherapy often does travel with the tumour cell and a simple model reads it correctly. Both facts are true and they define an envelope. A stroma-free organoid is a competent instrument for cell-autonomous cytotoxic and targeted agents and an incompetent one for drugs whose mechanism of failure lives in the surroundings. The failure mode to fear is not using a simple model; it is using a simple model outside that envelope and trusting the answer. Running an immune-directed or anti-stromal drug through a compartment-free organoid does not produce a cautious result, it produces a confident and mechanistically empty one, and automation only manufactures those faster.

The constructive path is the one the review itself points at, and it is more disciplined than adding everything. If resistance is per-mechanism, then validity is per-mechanism: build the specific compartment the drug's mechanism requires and no more, and state the drug classes the resulting model is competent to judge. Matrix stiffness is the encouraging case, because it is a physically tunable, reproducible variable, so an organoid can set matrix mechanics as a designed factor and convert what is otherwise an uncontrolled confound into a readout. The opposite temptation, stacking fibroblasts, immune cells and vasculature into one system to be safe, buys fidelity with reproducibility, because every added lineage is another differentiation step whose variance can quietly degrade the screen. A more complex model that fails silently is worse than a simple one whose limits are declared.

The threat worth naming plainly is a governance one. As regulators and funders lean toward replacing animal studies with human microphysiological systems, the phrase validate in an organoid acquires weight it has not earned uniformly across drug classes. This review is the argument for why that phrase must come with a scope. An organoid validation is only as good as the match between the model's compartments and the drug's mechanism, and the burden belongs on the platform to show that match, not on the reader to assume it.

The bottom line

Established: the tumour microenvironment contributes to therapeutic resistance through specific, mechanistically characterised routes involving stroma, immune cells, matrix, vasculature and vesicle signalling. Under-supported: the review's implied weighting that environmental remodeling dominates resistance, and its optimistic framing of microenvironment-targeting strategies, both of which are asserted more firmly than the cited evidence and the clinical record justify. For organoid drug discovery the durable takeaway is a rule, not a result: a model earns trust only for drugs whose resistance mechanism it actually contains, and this review is a usable inventory of the mechanisms a tumour-cell-only model omits. What would sharpen it is quantitative, namely head-to-head studies estimating how often environmental resistance overturns a monoculture drug call, per tumour type and per drug class. That number, not another mechanism, is what the field is missing.