Embryo Models

A stem-cell-based embryo model is a structure assembled from pluripotent stem cells that reproduces some of the organisation of an early embryo without being one. The field exists because human development between implantation and the end of gastrulation — roughly days 7 to 21 — is the least observed period in human biology. Embryos cannot be obtained from a pregnancy, donated embryos are scarce, and culture beyond fourteen days is prohibited or restricted almost everywhere.

Established The 2023 human models were a real advance and a badly reported one. Three groups published structures containing embryonic and extra-embryonic compartments — epiblast-like disc, amniotic cavity, yolk sac, extra-embryonic mesoderm — at an organisation resembling a human embryo around days 13 to 14. They were formed from stem cells without eggs, sperm or fertilisation. They were also inefficient, variable, and in the published transcriptomic comparisons only partly matched their reference stages.

Established No human embryo model has been shown to have developmental potential, and none may lawfully be tested for it. The only direct test in a primate was done in macaques: models transferred into recipient uteri produced transient hormonal and sac-like signs of pregnancy in a minority of animals and no fetus. That result is the single most informative datum in the whole debate about what these things are.

This brief treats scientific fidelity, endpoints, implantation potential, limits, oversight and misconceptions. Gestation devices are a different technology and belong to Artificial Wombs; editing of embryos and germline questions belong to Genetic Engineering. Reproductive use of embryo models is out of scope here because no responsible framework permits it and no evidence supports it.

Established A note on sourcing. This brief was commissioned in September 2026 from the Institute’s research base. Reading-list entries without links are cited from the bibliographic record rather than re-fetched, and claims are dated no later than early 2026 unless carried by a linked source.

Established The field has a taxonomy and it does real work. Guidance distinguishes non-integrated models, which reproduce one aspect of development — gastruloids, somitoids, cardioids — from integrated models, which combine embryonic and extra-embryonic lineages and could in principle organise like a whole conceptus. The distinction is the hinge of every oversight regime: non-integrated models are treated much like organoids, integrated models attract specialised review.

Established The mouse work came first and went furthest. Blastocyst-like structures from mouse stem cells were reported in 2018. In 2022 two groups grew post-implantation mouse models in rotating-bottle culture to roughly embryonic day 8.5, with a beating heart-like structure, neural folds and somites, starting from stem cells alone. Efficiencies were low, typically a few per cent of aggregates forming a well-organised structure, and the models were not transferred to produce animals.

Established Human blastoids arrived in 2021. Two groups reported blastocyst-like structures from human pluripotent stem cells and from directly reprogrammed fibroblasts; a third reported blastoids that attach to hormonally stimulated endometrial cells in culture, which turned implantation into something that can be screened. Reported formation efficiencies range from single-digit to low-tens of per cent depending on protocol, and cell-type composition varies between protocols.

Established The 2023 human post-implantation models are the current state of the art. Independent groups reported integrated models with an epiblast-like disc, amniotic and yolk-sac cavities and extra-embryonic mesoderm, corresponding roughly to days 13 to 14. The best-characterised protocol reported that of the order of one per cent of starting aggregates formed a properly organised model. Nothing published has carried a human model reliably past the point where a real embryo would begin organogenesis.

Frontier Fidelity is the contested question and the field has no agreed test. Single-cell reference atlases of human embryos now allow model cells to be scored against staged references. Published comparisons show good correspondence for some lineages, poor correspondence for others, off-target cell states with no counterpart in embryos, and asynchrony within a single model. There is no agreed pass mark. In practice a model is called successful when it looks right to the authors and reviewers, which is the weakest link in the entire literature.

Established What these models demonstrably cannot do. No human model has produced a functional placenta or a maternal-fetal interface; trophoblast identity and invasion are the least faithful compartments in most protocols. None recapitulates the mechanical environment of the uterus. None has produced a complete body plan. And none has been through an independent, blinded replication of the kind that would settle whether inter-laboratory variation or biology explains the differences between protocols.

Established The primate transfer experiment is the one direct test of potential. In 2023 a group generated cynomolgus monkey embryo models and transferred them into the uteri of recipient females. A minority of recipients showed early pregnancy signals — a rise in chorionic gonadotropin and gestational-sac-like structures on ultrasound — which resolved. No fetus developed. The result is genuinely two-sided: models can engage the uterine machinery enough to produce a hormonal pregnancy signal, and they did not develop.

Established The natural comparison is human embryo culture, which stops at fourteen days by rule, not by biology. Two groups showed in 2016 that human embryos self-organise in vitro past implantation stages, and both terminated their cultures at day 13 to 14 in compliance with the rule. That is why the window this field models is empty of primary data, and why a faithful model would be so valuable.

Established The 14-day rule’s status changed in 2021 and the change is under-appreciated. The rule originated with a United States advisory board in 1979 and the United Kingdom’s Warnock committee in 1984, tied to the appearance of the primitive streak. In the United Kingdom it is statute and breach is a criminal offence. In 2021 the international society’s guidelines moved culture beyond fourteen days out of the prohibited category into a category requiring specialised review where national law permits, and separately assigned integrated embryo models to specialised oversight.

Frontier Embryo models fall outside most statutory definitions of an embryo, and that is a legal fact with no settled consequence. The United Kingdom statute defines an embryo by reference to fertilisation, which models do not involve, so models are not regulated as embryos there. A voluntary national code of practice published in 2024 filled the gap: a national oversight committee, project-by-project review, no transfer of a model into any uterus, human or animal, and justification of culture duration case by case. It has no statutory force.

Established In the United States the constraint is funding, not prohibition. A long-standing appropriations rider bars federal funding for research in which human embryos are destroyed. Because models are not embryos under that language, the rider arguably does not reach them, which leaves the most consequential category of this research governed by institutional review boards, state law and journal policy rather than by a statute written for it.

Established The research uses are concrete and already being pursued. Implantation failure and very early loss account for a large share of failed pregnancies and are almost unstudiable in humans; blastoid-endometrium co-culture turns that into an assay. Other uses: contraceptive and fertility drug screening, teratogen testing outside animal models, and genotype-to-phenotype work on congenital disorders that manifest in the first three weeks. These are the justifications the field offers, and unlike many such lists they describe experiments people are running.

Frontier Models do not remove the need for embryos. Every fidelity claim is scored against reference data from real human embryos. A model validated against an atlas built from a handful of donated embryos inherits that atlas’s limits, and a field that stopped studying embryos entirely would lose its only ground truth.

Frontier How far can an integrated human model be taken before it stops being a model? Groups are pushing past the day-14-equivalent stage toward early organogenesis. The scientific value rises with stage, and so does the argument that the structure now warrants embryo-like protection. Nobody has proposed a criterion that does not reduce to appearance.

Frontier Would a human model implant? The macaque result says a primate model can trigger a hormonal pregnancy response and then fail. Whether that reflects a ceiling on potential or a technical limitation of the protocol is unresolved, and the human version of the experiment is prohibited by every framework in force. This is the field’s central epistemic problem: the decisive fact about what these things are cannot be determined in the species that matters.

Frontier Can efficiency and reproducibility be raised enough for screening? A one-per-cent success rate is workable for a landmark paper and useless for a dose-response assay. Automation, defined media and standardised cell lines are the obvious routes, and the field has not yet published an integrated-model protocol with the reproducibility a contract laboratory would accept.

Frontier What counts as a faithful trophoblast? The extra-embryonic compartments are where models diverge most from embryos, and they are also the compartments that matter for implantation, pre-eclampsia and pregnancy loss — the clinical problems the field most wants to address. Solving fidelity here would do more for the research case than another week of culture.

Speculative Could models replace animal developmental toxicology? Human-specific developmental toxicity is exactly what rodent models miss. The argument is coherent and the evidence is absent: no embryo model has been formally qualified as a regulatory test method for anything.

Established Reference data are the binding constraint on validation. Human embryo atlases covering implantation and gastrulation rest on very small numbers of donated specimens, many decades old in the case of histological collections. Every fidelity claim is therefore measured against a sparse ruler.

Frontier Efficiency and variability limit everything downstream. Low and protocol-dependent formation rates mean that published results describe selected structures. Without reporting standards that state how many aggregates entered and how many were scored, the literature systematically over-represents the best specimens.

Established Culture systems cannot supply a uterus. Beyond the early post-implantation window, real development depends on maternal blood supply and mechanical constraint. Extended culture of mouse embryos in rotating-bottle systems reaches only a limited window before failing for want of perfusion, and no human equivalent exists.

Frontier The oversight bottleneck is jurisdictional, not ethical. Committees exist and are willing to review; what is missing is agreement on which body has jurisdiction over an entity that is not an embryo, not an organoid, and not an animal. Projects therefore face different rules in different countries for identical experiments.

Established It depends on naive pluripotency culture. Integrated models require stem cells in a developmentally early state, and the protocols that produce and stabilise naive human cells are recent, various and not fully interchangeable. Differences between starting cell states explain part of the between-laboratory variation.

Established It depends on single-cell and spatial transcriptomics for its only quantitative claims. Without reference atlases and cell-type classifiers, fidelity would be an entirely morphological argument. The field’s evidentiary standard is therefore inherited from a measurement technology that is itself still improving.

Frontier It depends on donated human embryos it is sometimes said to replace. The ground truth comes from the same scarce, ethically constrained material, which is why a policy that ended embryo research would damage embryo-model research rather than liberating it.

Established It depends on organoid and microphysiological engineering. Media formulation, matrix chemistry, microfluidics and imaging all come from the wider tissue-engineering programme covered in Microphysiological Systems and Lab-Grown Organs.

Frontier The decisive experiment is the primate transfer test, extended and properly powered: transfer integrated monkey embryo models into a sufficient number of recipient females, with pre-specified endpoints for hormonal signalling, sac formation and fetal development, and report the negatives. It is decisive because every argument about the moral status and legal category of embryo models turns on whether they have developmental potential, and because it is the only version of that test that can be run at all. The first version has been performed once: transfer produced transient pregnancy signals in a minority of recipients and no fetus. A single small experiment carries the entire weight of the potentiality question, and nobody has scheduled the definitive one.

Frontier Second: a blinded, pre-registered benchmarking exercise. Send coded single-cell datasets from integrated models made by several laboratories to independent analysts, scored against staged human reference atlases on criteria fixed in advance. It would establish whether protocol differences are real, and it would give the field its first shared definition of a faithful model.

Frontier Third: an inter-laboratory reproducibility study with full denominators. One protocol, five laboratories, identical cell lines, reporting aggregates in and models out. This is cheap, unglamorous and would change how the published efficiency figures are read.

Frontier Fourth: a prospective test of clinical relevance. Take blastoid-endometrium co-culture, score implantation competence, and compare against outcomes in a matched clinical dataset of implantation failure. Until a model predicts something about a patient, the translational case rests on plausibility.

Speculative Fifth: a natural experiment in governance. Jurisdictions have adopted different rules for identical work — statutory silence with a voluntary code in one, funding restriction in another, specialised review in a third. Comparing where the work actually happens, and what is published from each, would tell us which regime produces both science and oversight rather than one at the expense of the other.

Established The engineering is culture engineering. What separates a pile of cells from an organised model is the spatial and temporal control of signalling: geometric confinement on micropatterns, defined media with staged factor addition, matrix stiffness, and in the mouse case rotating-bottle culture with controlled gas mixtures. None of it requires exotic hardware.

Frontier Scale-up means automation and imaging, not bigger flasks. An assay needs thousands of structures, each tracked from aggregate to endpoint. That is a robotics, microscopy and image-analysis problem, and the laboratories leading on biology are not the ones leading on automation.

Frontier Perfusion is the unsolved requirement for later stages. Any attempt to carry a model past the stages where diffusion suffices runs into the same wall as every organoid programme: no vasculature. Lab-Grown Organs owns that constraint, and it caps this field at roughly the stage it has already reached.

Established Reporting standards are an engineering deliverable, not an ethical flourish. Denominators, cell-line provenance, passage number, media lot and scoring criteria are what would let another laboratory reproduce a result. Their inconsistent reporting is why the fidelity literature is hard to compare.

Established Gestation devices are the adjacent technology most often confused with this one. Artificial Wombs covers extracorporeal support of a fetus that already exists, with large-animal data and no cleared human trial. Embryo models start from stem cells and are not gestated. Joining them — culturing a model in a perfusion device toward viability — is prohibited by every code in force and is not what either field is doing.

Established Embryo editing and selection are adjacent through the same regulatory machinery. Genetic Engineering covers what editing does to real human embryos, including the evidence on unintended rearrangements and loss of heterozygosity, and the international response to heritable modification. Models are an obvious platform for studying those effects without using embryos.

Frontier Embryo selection is adjacent through the ethics rather than the biology. Polygenic embryo screening, covered in Precision Medicine and Biological Enhancement, shares this field’s central institutional weakness: commercial services operating in a regulatory space designed for something else.

Established Organoids and microphysiological systems are the direct methodological neighbours. They supply the culture platforms, and they also supply the cautionary precedent: organoid claims outran organoid validation for a decade, and the qualification standards that eventually arrived came from regulators, not from the field.

Established The governing document is a professional guideline, not a law. The international society’s guidelines set the categories used worldwide: non-integrated models under standard review, integrated models under specialised oversight, no transfer to a uterus, and case-by-case justification for extended culture. Compliance is a condition of publication in most serious journals, which is the real enforcement mechanism.

Established The United Kingdom built a voluntary code because its statute does not apply. Published in 2024 by an academic-led group with regulator participation, it establishes a national oversight committee, requires registration of projects, bans transfer into any uterus and requires scientists to justify how far they culture. It is a serious instrument with no statutory backing, and its authority depends on funders and journals insisting on it.

Frontier Statutory reform has been discussed everywhere and enacted nowhere. Proposals include extending embryo law to cover entities with developmental potential, creating a new category defined by capability rather than origin, or licensing model work under fertility regulators. Each founders on the same problem: writing a definition that captures what we want to control without either exempting everything or sweeping in organoids.

Established Funding rules shape the geography of the work. Where public funds cannot support embryo research, model work migrates to private and philanthropic funding, which is less transparent and less bound by the conditions that come with public grants. This is the same displacement pattern that governs every restricted area of reproductive science.

Frontier The commercial layer is arriving before the rules. Companies working on blastoid-based fertility and contraceptive screening are building on this science, and the precedent from unregulated stem-cell clinics is that commercialisation of an unvalidated cell product runs ahead of oversight unless a regulator moves first.

Frontier Moral status here is argued from potential, and potential is exactly what has not been measured. The traditional case for protecting embryos rests on what they can become. Models formed from stem cells have no demonstrated capacity to become anything, and the only primate test produced a hormonal signal and no fetus. That leaves the argument suspended between two defensible positions rather than settled for either.

Established Consent is a real and tractable issue. Models are built from stem cell lines derived from donated embryos, tissue or reprogrammed adult cells. Many donors consented years ago to research framed quite differently, and the code-of-practice approach of requiring specific consent for embryo-model use is the correct response and is not universal.

Frontier Public language is doing ethical work that scientists did not authorise. The phrase “synthetic embryo” entered headlines in 2022 and 2023 over the objections of many of the researchers involved, who argued it implies both artificiality and embryo-hood. Terminology is not cosmetic here: legal categories, funder rules and public consent all follow the words people use.

Frontier The strongest ethical argument for the field is that the alternative is using embryos. If models can answer questions currently answerable only with scarce donated embryos, the ethical cost of the research falls. The argument only works if fidelity is demonstrated, which returns the ethics to the unresolved measurement problem.

Speculative The reproductive-use question should be treated as closed and monitored anyway. No framework permits transfer of a model to a uterus, there is no evidence any human model could develop, and there is no clinical reason to try when established reproductive technologies exist. The risk is not a research programme; it is a clinic in a permissive jurisdiction, which is how every previous reproductive boundary was actually crossed.

Frontier The most plausible large effect is on early pregnancy loss. A substantial fraction of conceptions fail at or shortly after implantation, largely invisibly and almost entirely unstudied. A faithful implantation assay would be the first tool for investigating it, and the clinical population is enormous compared with anything else in this brief.

Speculative A second effect is on how societies argue about embryos. Embryo law was built on fertilisation as the bright line. An entity with partial embryonic organisation and no fertilisation event puts pressure on that line from a direction nobody anticipated, and the resulting definitions will outlive the technology that prompted them.

Speculative A third is methodological. If developmental toxicology can be done in human cells, the animal-testing baseline for an entire regulatory domain changes. This is a decade-scale institutional shift and requires qualification standards that do not exist.

Handwave The claim that embryo models lead to babies grown outside the body is where the argument stops being evidence-led. It requires developmental potential nobody has demonstrated, gestational support nobody has built, and a legal permission nobody has sought. Each of the three is separately unsolved, and the argument usually treats the sequence as inevitable.

These horizons track what would have to be demonstrated or decided, not what could be announced. The field’s rate limit is validation and oversight, not cell culture.

• 10 yr: Frontier An agreed benchmarking standard for integrated models, with blinded scoring against reference atlases; reproducible protocols at efficiencies workable for screening; a first regulatory qualification of a non-integrated model as a developmental toxicity test method. Speculative Statutory categories for embryo models in at least one major jurisdiction.
• 25 yr: Speculative Implantation and early-loss biology studied primarily in models rather than in embryos, with a first clinical intervention for recurrent implantation failure derived from model work; the primate potentiality question answered one way or the other and the law written accordingly.
• 50 yr: Speculative Human developmental biology taught from model-derived data as a matter of course, with donated embryos reserved for calibration. Whether that counts as replacing embryo research or entrenching dependence on it is a question of how the reference atlases are maintained.
• 100 / 250+ yr: Handwave Claims at this range — that embryo models become a reproductive route, or that human development is fully reconstructible from stem cells — assume a developmental potential that has never been demonstrated and a gestational technology that does not exist.

• Depends on Nothing on this map blocks the basic science, which is the unusual feature of this field: the techniques are stem-cell culture and imaging that already exist. What it depends on is adjacent evidence and rules. Genetic Engineering supplies the human-embryo evidence base — including what editing does to real embryos — against which model work is justified as an alternative, and the governance climate that every oversight body in this area now writes against. Lab-Grown Organs and Microphysiological Systems supply the culture platforms and the vascularisation ceiling that caps how far any model can be taken.
• Requires (not on this map) A shared, blinded standard for scoring fidelity against staged human references, without which every claim in the literature is an argument from appearance; a properly powered primate transfer experiment, because one small study currently carries the entire potentiality question; a legal category for entities that have partial embryonic organisation and no fertilisation event, since existing statutes define embryos by a process these structures skip; production reproducible enough that an assay can be run rather than a landmark published; an oversight body that actually has jurisdiction, rather than a voluntary code enforced by journals; and a funding route that does not inherit restrictions written for embryo research, or the work migrates to private money and out of public view.
• Enables A validated integrated model enables the first systematic study of implantation and early pregnancy loss, human-specific developmental toxicology without animal extrapolation, mechanistic work on congenital disorders that arise before a pregnancy is detected, and a substitute for scarce donated embryos in a large fraction of developmental research. It would also enable the moral-status debate to be conducted about a measured capability rather than an inferred one.
• Adjacent Artificial Wombs is the technology this one is most often confused with and shares no method with; Precision Medicine and Biological Enhancement carry the embryo-selection debates that share this field’s regulatory gap; Mitochondrial Medicine is adjacent as the one area where a heritable reproductive intervention has already been licensed, reported and followed up.

Established “Scientists have created synthetic human embryos.” They have created structures that reproduce part of the organisation of an embryo at about two weeks, inefficiently, with off-target cell states and imperfect extra-embryonic compartments, and with no demonstrated capacity to develop. Many of the researchers involved objected publicly to the word synthetic and to the word embryo, in that order.

Established “These models could be implanted to make a baby.” The only transfer experiment in a primate produced transient pregnancy hormone signals and no fetus. Transfer into a human uterus is prohibited by every framework in force, and no evidence suggests it would work. Treating the possibility as imminent misstates both the biology and the rules.

Established “The 14-day rule blocks this research.” Since 2021 the international guidelines have moved culture beyond fourteen days from prohibited to specialised review where national law allows, and the rule applies to embryos, which models legally are not in the jurisdictions that define embryos by fertilisation. The binding constraints on model research are validation and jurisdiction, not the fourteen-day limit.

Frontier “Embryo models make embryo research unnecessary.” Fidelity is measured against reference data from donated human embryos. Without continued embryo research the atlases stop improving and the models lose the only standard by which they are judged.

Established “A model that looks like an embryo is an embryo.” Morphological resemblance is exactly the evidence the field is trying to replace with quantitative scoring, precisely because published models with convincing appearance have shown poor transcriptomic correspondence in specific lineages. Appearance has been the least reliable indicator in this literature.

Frontier “This is the same thing as an artificial womb.” One grows a structure from stem cells in a dish; the other supports an already-existing fetus in a perfusion circuit. The confusion matters because it imports the extreme-prematurity ethics of one field into the cell-culture ethics of the other, and vice versa.

Speculative “Embryo models have no moral status because they are just cells.” This is a conclusion, not an argument, and it is doing the same work in the permissive direction that “synthetic embryo” does in the restrictive one. The honest position is that status depends on capability, capability is unmeasured, and the only measurement available — in another species — returned a partial result.

Handwave “Complete human development in a dish is a matter of time.” It requires solving vascular perfusion, maternal signalling, mechanical constraint and a fidelity problem the field cannot currently score, in a legal environment that prohibits the experiments that would test progress. The assertion treats four separate unsolved problems as a schedule.