The organoid shape you were told to discard

What the work claims

This is a primary result combining a single-cell and spatial atlas with functional organoid experiments, not a review.¹ The subject is adenogenesis, the formation of uterine glands, which in mice happens after birth and in humans begins late in gestation. The central claim has two parts. First, that before the epithelium commits to either a luminal fate (the cells lining the cavity) or a glandular fate (the secretory cells whose products are required for pregnancy), it passes through a transient, developmentally restricted plasticity state that lacks the estrogen receptor ESR1 and the progesterone receptor. Second, that this state has a visible correlate in culture: organoids grown from that window form multilayered structures carrying a basal cell signature, and the capacity to make them is lost as the tissue matures. The bolder move is to reframe a morphology the field usually calls an artifact as a functional readout of a developmental program.

The reason this is contested is that multilayering under culture stress is a generic behaviour of many epithelia, so calling it a specific developmental signal is a claim that has to be earned against a mundane alternative.

How it works

The backbone is a mouse atlas of 147,440 cells sampled from embryonic day 16.5 through postnatal days 1, 5, 12 and 15, spanning the transition from undifferentiated Mullerian duct epithelium to the first luminal and glandular cells. Pseudotime ordering of that data traces a common early epithelial pool splitting into two trajectories, with the glandular branch progressively acquiring the transcription factor FOXA2, retinoic-acid metabolic genes and epithelial ESR1. The early, pre-commitment cells express neither ESR1 nor the progesterone receptor and sit in a proliferative, transcriptionally unspecified state.

The functional handle is the endometrial epithelial organoid (EEO). Organoids made from the early postnatal window produce, alongside normal single-layered cysts, multilayered structures marked by the basal keratin KRT5, and the fraction that can do this falls as the donor tissue ages. Three independent perturbations then couple that morphology to gland specification. Adding retinoic acid to the cultures, or letting ESR1 be acquired, suppresses the multilayered form, matching the in-vivo timing. Treating neonatal mice with estradiol or progesterone during the first postnatal days, a manipulation known to block gland formation, both reduced FOXA2-positive glands in the animal and stripped the multilayered organoid capacity from cells taken afterward, with progesterone prematurely switching on an epithelial-to-mesenchymal Ihh-to-Ptch1 signal that normally appears only after glands commit. And deleting Foxa2 in the gland lineage reduced multilayered organoid formation: in single-cell profiling of the organoids the basal-marked cluster fell from 43.4 percent in controls to 19.5 percent, while the luminal fraction rose from 54.7 to 76.7 percent. Convergence from a hormonal and a genetic route onto the same morphological loss is the paper's strongest structural argument that the multilayered state is tied to the gland program rather than to random culture variation.

Where a skeptic should push

The single most load-bearing assumption is an identity claim: that the multilayered organoid in the dish is the same biological entity as the transient ESR1-negative plastic state in the animal, rather than a convergent basal or squamous default that many stressed epithelia reach and that merely shares a marker or two. The signature leans on KRT5 and the basal-cluster gene Trp63, and here the paper is honest about a gap the reader should hold onto: p63 protein was not detected in the postnatal uterine epithelium in vivo at any timepoint, so Trp63 is a single-cell cluster label, not a demonstrated in-vivo basal compartment. To earn the identity claim you would want a program-level transcriptional match across all four settings the paper invokes, the organoid, the neonatal window, and the two disease states, plus a test of temporal specificity showing that adult ESR1-positive epithelium cannot be pushed into an indistinguishable multilayered state by culture stress alone. The regulated behaviour, suppression by ESR1 and retinoic acid and reduction by Foxa2 loss, is stronger evidence than the morphology itself, and the argument should lean there.

Two framing points deserve correction, and I will correct my own initial reading rather than bury it. The paper's abstract says hormonal and genetic disruption abolished the organoid phenotype, but the quantification shows a reduction, not an abolition: the basal fraction fell to 19.5 percent rather than to zero, and residual multilayered organoids in the Foxa2 model still expressed FOXA2, consistent with a receptor-negative escaper population. Reduced is the accurate verb. And the leap from this state to cancer should be stated as what it is. The same basal signature does appear in adult Tgfbr2-driven hyperplasia and in human patient-derived endometrial cancer organoids, which is genuinely provocative, but a shared transcriptional signature is not proof that cancer re-runs the identical developmental program; a tumour can reach a basal state by another route. The most important limit is species. Every functional and developmental result here, the transient window, the ESR1 and retinoic-acid suppression, the Foxa2 requirement, the neonatal hormone effects, is mouse. The human leg is a cross-species transcriptomic match to fetal reproductive-tract cells and the historical precedent of diethylstilbestrol, plus the pathological cancer organoids. There is no human developmental organoid experiment. That does not sink the work, but it bounds it: the human relevance is inferred, not demonstrated.

Reclaiming a discarded organoid morphology

The constructive implication is a rare thing in organoid work: a quantitative, hormone-responsive readout with built-in controls. The basal fraction is a dial that moves in known directions, down with ESR1 and retinoic acid, down with Foxa2 loss from 43.4 to 19.5 percent, up in the diethylstilbestrol precedent, which means an assay reading multilayer frequency comes with its own positive and negative anchors. The non-obvious part is that the signal lives in a reject bin. Standard organoid pipelines are engineered to suppress exactly this morphology: healthy-organoid gating, segmentation that favours clean cysts, and media and passage choices that push cultures toward the ESR1-positive glandular default all normalise the multilayered state away. If the paper is right, those pipelines are discarding the developmental-tox signal they would most want for screening endocrine disruptors, because a disruptor's whole mode of action is to perturb this window.

The threat is the mirror image, and it is a generalisation-failure trap with a regulatory edge. Because the readout is defined by a developmental window and, on current evidence, by mouse biology, porting the assay to human adult tissue, or to organoids that have drifted ESR1-positive in culture, would extinguish the signal by the very mechanism the paper describes, and an extinguished signal reads as a confident all-clear. An endocrine disruptor screened on a substrate that can no longer show the phenotype would be scored safe for the wrong reason. Here I have to resolve a tension in my own enthusiasm: it is not coherent to warn that the readout is window and species specific and in the same breath sell a ready-made human gestational-drug-safety platform. What the work supports is a mouse-validated hypothesis worth building toward a human anchor, not a human assay in hand. There is even a direction problem: neonatal progesterone abolished the state in mice, so a naive port would flag prescribed gestational progesterone as protective, a reading that does not straightforwardly map to human risk and should not be used to reassure anyone. The honest offer is a promising phenotypic axis and an explicit warning about the substrate on which it can be measured.

The bottom line

Established in mouse: a transient, ESR1-negative and progesterone-receptor-negative epithelial state precedes luminal and glandular commitment; it has an in-vitro correlate in multilayered endometrial organoids; and that correlate is suppressed by ESR1 and retinoic acid and reduced by hormonal or Foxa2-based disruption of gland formation. Still hypothesis: that the cultured multilayered form is the same entity as the in-vivo state rather than a convergent stress default, that the developmental program is genuinely re-used in cancer rather than merely resembled, and that any of this transfers to human tissue. What would confirm it is a program-level match across the organoid, developmental and disease states with a demonstration that adult ESR1-positive epithelium cannot be forced into the same state, plus a human developmental organoid showing the window exists. What would break it is evidence that ordinary culture stress reproduces the multilayered morphology in fully committed epithelium, which would demote the readout from developmental signal to artifact after all.