EMT in breast cancer does not converge on one metastatic phenotype

What the work claims

This is a primary results paper that asks when and how metastatic-competent cell states emerge during primary tumor progression. The central claim is that epithelial-mesenchymal transition (EMT), the process in which epithelial tumor cells acquire mesenchymal traits that favor invasion and spread, does not funnel breast cancer cells into a single metastatic phenotype. Instead, early malignant cells first lose mammary lineage identity and produce lineage-altered epithelial states with high intrinsic plasticity. Those plastic states then diversify through two parallel EMT-associated programs distinguished by high and low ERK1/2 signaling. Both programs generate cell populations that can initiate metastases, but the resulting states differ: early tumors contain heterogeneous hybrid epithelial-mesenchymal cells, while later tumors contain more uniform mesenchymal-like cells with canonical EMT features and diminished plasticity.¹

How it works

The authors use a genetically engineered mouse model of triple-negative breast cancer (TNBC), a subtype with few targeted therapies and a strong tendency to metastasize. They profile cells across tumor progression by single-cell RNA sequencing and then use tumor organoids to test whether the sequence of states they observe is functionally meaningful. The lineage-loss step is the earliest detectable change: malignant cells shed mammary epithelial identity before engaging classic EMT transcriptional programs. That loss creates a permissive, plastic state from which cells can enter either an ERK1/2-low or an ERK1/2-high EMT program.

These two programs do not represent minor transcriptional variants. They produce different cellular phenotypes. The ERK programs generate distinct hybrid epithelial-mesenchymal states in early lesions and more uniformly mesenchymal-like populations at later stages. Critically, metastatic competence is not restricted to either the hybrid or the mesenchymal-like form. Both initiated metastases, and the metastatic lesions retained features of the initiating population. The implication is that metastatic potential is distributed across multiple EMT-associated states rather than concentrated in one canonical mesenchymal endpoint.

The strongest version of the result

The steelman is that the paper resolves a long-standing ambiguity in the EMT field. EMT has often been treated as a binary switch or a linear gradient from epithelial to mesenchymal, with the assumption that the most mesenchymal cells are the most metastatic. By combining single-cell trajectory analysis with organoid functional readouts, the authors make a more nuanced picture: metastatic competence can arise through several distinct transcriptional routes, and the hybrid states are not failed intermediates but genuine metastatic precursors. For TNBC, which is defined partly by its heterogeneity, this pluralistic model fits the clinical reality better than a single EMT endpoint. The use of organoids to test state-to-function relationships is also methodologically sound, because organoids preserve 3D architecture and cell-cell interactions that flat cultures lose.

Where a skeptic should push

The most load-bearing assumption is that the transcriptional states identified by single-cell sequencing correspond to discrete, stable, and functionally distinct cell populations rather than to transient or context-dependent expression fluctuations. Single-cell snapshots can make dynamic plasticity look like fixed cell types, and the ERK-low versus ERK-high distinction, while mechanistically plausible, needs to be tested with lineage tracing and perturbation to know whether cells switch between the two programs or commit irreversibly.

Because I could not access the full preprint, I cannot verify sample sizes, the number of mice or organoids per state, the exact statistical tests, or whether the metastasis-initiation experiments were performed once or replicated. That matters for a claim this central. A second concern is generalization: the model is one GEMM of one breast cancer subtype. Whether the same programs appear in human TNBC organoids, in other breast cancer subtypes, or in carcinomas of other tissues is unknown. Finally, the organoid experiments are ex vivo; proving that the same state-to-metastasis relationships hold in spontaneous metastasis in vivo would require orthotopic or autochthonous validation beyond what I can confirm from the abstract.

EMT heterogeneity as an organoid-screen confound

For organoid models of human organs and the drug discovery built on them, the most important implication is that a tumor organoid line is not one disease model; it is potentially several coexisting ones. If metastatic competence can be carried by distinct EMT-associated states, then a drug screen that reads out only one of those states, or that accidentally enriches for one over another during passaging, will miss therapies active against the others. This is a concrete, mechanistic source of irreproducibility. A compound that kills the mesenchymal-like cells in a late-passage organoid might leave the hybrid population untouched, and a protocol that selects for the hybrid state could make an otherwise effective anti-metastatic agent look useless.

The opportunity is to build organoid-based screens that deliberately preserve and report EMT state diversity. That means adding single-cell or imaging readouts to conventional viability assays, tracking lineage identity loss as an early marker of plasticity, and using ERK pathway reporters or EMT signature panels to determine which state a given organoid culture is in before treating it. For patient-derived tumor organoids, the threat is the opposite: if a biopsy captures only one EMT state, or if culture conditions collapse the natural diversity into a single dominant state, the model will misrepresent the patient's tumor and the screen will recommend the wrong drug. The non-obvious lesson is that EMT heterogeneity is not a nuisance to average away; it is the variable that determines whether an organoid predicts metastasis accurately.

The bottom line

Established from the abstract and metadata: in this TNBC GEMM, early lineage-state disruption is followed by parallel ERK1/2-low and ERK1/2-high EMT programs that generate distinct hybrid and mesenchymal-like cell states, and both states can initiate metastases. This is a mechanistically grounded challenge to the idea that EMT converges on a single metastatic phenotype. What remains to be verified, because the full text was not accessible, is the statistical strength, replication, and in vivo generality of those state-to-metastasis mappings. For drug discovery, the actionable takeaway is that tumor organoid assays need state-resolved readouts; a single bulk viability endpoint is no longer enough when the biology says the metastatic phenotype is plural.