Gastric precancer mechanics and the organoid matrix problem

What the work claims

This is a review and analytical framework, not a primary experiment. Hu, Huang, Zhang and colleagues integrate human tissue studies, gastric organoid work and animal models of the Correa cascade, the atrophy-to-metaplasia-to-dysplasia sequence by which chronic gastritis becomes gastric adenocarcinoma. Their central claim has two halves. First, descriptive: current evidence does not support representing gastric precancer as one monotonic stiffening trajectory; direction, magnitude, tissue layer, anatomical region and disease stage have not been established as a general sequence. Second, methodological: mechanotransduction claims in this literature repeatedly rest on experiments where the physical variable was never measured or was confounded, and standard organoid matrices have poorly controlled, batch-variable mechanical properties that make them unfit to settle the question.¹

To make those claims testable, the authors introduce GPMES, a five-tier evidence grading that ranks studies by proximity to causal gastric-precancer mechanobiology: Tier 1, direct human mechanical measurements; Tier 2, human structural or spatial proxies such as second-harmonic-generation imaging of collagen; Tier 3, functional perturbation in gastric organoids or animal models; Tier 4, pathway manipulation in established gastric cancer; Tier 5, cross-tissue analogy. Their survey finds the field's mechanistic self-image runs several tiers ahead of its actual evidence.¹

How it works

The review's logic is a causal chain that most published work fails to complete: a specified physical exposure, a molecular sensor, demonstrated force transmission, and a stage-relevant epithelial phenotype. Take the best-studied candidate, the Hippo pathway effector YAP. Human and mouse studies link YAP activity to regeneration, metaplasia and malignancy, and H. pylori-associated experiments define a regulated LATS2-YAP sequence in which early nuclear YAP is followed by YAP phosphorylation and restraint of epithelial-mesenchymal transition. But as the authors note, none of these studies defines a force threshold, and YAP localization cannot be read as a mechanical measurement: receptor abundance can change without force, ligand availability can activate integrin signaling biochemically, and oncogenic signaling can raise contractility downstream of adhesion. "YAP on equals cancer" is explicitly rejected as a simplification.¹

The same discipline is applied to the extracellular matrix itself. Collagen accumulation, protease activity and edema each change tissue mechanics differently; a protease can interrupt a basement-membrane boundary without raising bulk stiffness, and collagen density changes ligand availability, porosity and transport at the same time as stiffness, so a collagen-density experiment is not a stiffness experiment. The organoid passage is the practical consequence: tunable synthetic or semi-synthetic hydrogels are needed that control stiffness, viscoelasticity, ligand density, degradability and fiber architecture independently, ideally with fibroblast, immune and endothelial co-culture or organ-on-chip flow to restore the multicellular context that Matrigel removes.¹

Where a skeptic should push

The most load-bearing assumption is that the review's own survey is complete and unbiased. It is a narrative review: the search strategy, inclusion criteria and inter-rater process behind the GPMES tier assignments are not described with systematic-review rigor, and tiering the entire literature is unavoidably a judgment call. Authors affiliated with traditional Chinese medicine hospital systems are reviewing a field largely built elsewhere, and while the mechanistic reasoning here is mainstream, the tier assignments should be read as an expert argument, not a reproducible census.¹

Second, the review's central negative claim, no established stiffness trajectory, is partly a claim about absent measurement, and absence of evidence rarely makes a headline. There may be real regional stiffening that current instruments miss; the honest formulation is that the field cannot currently say, which is what the authors in fact write. Third, the YAP and integrin literatures they critique include genuinely strong perturbation experiments; downgrading them to Tier 3 because force was not measured is defensible for the causal question but should not be misread as dismissing the biology. Finally, the gastric precancer focus matters: how much of this transfers to other organoid tissues is exactly what remains to be shown, and the review wisely restricts its claims to the stomach.¹

Organoid drug assays inherit matrix unknowns

For organoid models of human organs and the drug discovery built on them, the non-obvious implication is that a huge fraction of published organoid pharmacology runs on a substrate whose mechanical properties are an unmeasured, batch-variable confound. Matrigel is a tumor extract that varies lot to lot; if drug response depends on stiffness-driven YAP localization or integrin signaling, as the cancer literature broadly suggests, then two aliquots of the same assay can differ by a variable nobody recorded. This review states the consequence plainly for gastric precancer: common matrices cannot test mechanical hypotheses, so any organoid result presented as evidence about mechanosensitive drugs inherits that gap.¹

The opportunity is a concrete engineering specification the organoid field has needed: define hydrogels in which stiffness, viscoelasticity, ligand density, degradability and fiber architecture are controlled independently, and report the matrix variables alongside the biology. That turns organoid matrix choice from a procurement decision into an experimental variable, which is a prerequisite for reproducible screening of mechanosensitive or mechano-activating compounds, from antifibrotics to YAP-TEAD pathway drugs. The review's deeper opportunity for drug discovery is its insistence on separating repair from progression: a compound that activates regenerative YAP signaling could look protective in a short assay and be dangerous in a chronically inflamed gland, which is precisely the kind of false signal a 48-hour viability readout cannot catch.¹

The threat is to the credibility of the installed base. Thousands of published organoid drug-response datasets, including several that underpin organoid-guided clinical trial strategies, used undefined matrices with unknown mechanics. If matrix stiffness measurably shifts drug sensitivity for mechanosensitive targets, effect sizes in that literature are overdispersed by an unrecorded variable, and the fix, re-running assays in characterized hydrogels, is expensive and unglamorous. The governance consequence the review draws for human tissue, that mechanical measurements must be spatially registered with region, stage and preparation and reported with enough metadata for cross-study comparison, is the same metadata discipline organoid screening platforms will need if their predictive claims are to survive external audit.¹

The bottom line

Established in this review: the gastric precancer field lacks the spatially registered mechanical measurements needed to describe progression as any stiffness trajectory; mechanotransduction claims mostly rest on experiments where force was not measured or was confounded with ligand density; and standard organoid matrices cannot test these hypotheses. Hypothesis, not established: that mechanical state is a driver, rather than a consequence or correlate, of epithelial fate in precancer. What would confirm the driver model: hydrogel experiments that vary stiffness independently of ligand density, quantify traction, demonstrate integrin or FAK dependence, and rescue the lineage outcome when the mechanical exposure is withdrawn. What would break it: measurements showing matrix properties track epithelial reprogramming without influencing it. For organoid drug discovery the takeaway is immediate and cheap relative to its value: characterize and report the matrix, or accept that mechanosensitive assay results are uninterpretable.