Stem cell-based human embryo models: current knowledge and open questions

Abstract

Recently, the research field revolving around the stem cell-based modeling of the human embryo gained particular momentum when the first integrated models were designed with the aim to recapitulate the development of the entire early human conceptus. The underlying driving force to reconstruct embryo-like structures is the prospect of a more comprehensive understanding of the fundamental processes controlling early human embryogenesis including their deregulation causing reproductive failures, and the endeavor to use these embryo models for drug testing and disease modeling. Although efforts will continue to create improved models with steadily increasing fidelity to human embryogenesis, the next phase focusing on the application of human embryo models as a platform to address particular scientific questions is currently being entered. In this review, we discuss the benefits, promises and limitations associated with the use of non-integrated and integrated stem cell-based human embryo models in translational research and biomedical applications.

Introduction

The pre-implantation period of human development starts with the fertilization of the oocyte by sperm, is characterized by the activation of the zygote genome, cleavage divisions and the development of the multi-cellular morula, and ends with the implantation of the blastocyst invading into the endometrium of the uterus at about day 7. Just before implantation the blastocyst consists of the inner cell mass (ICM)-derived epiblast (which develops into the embryo proper) encircled by two extra-embryonic lineages, the ICM-derived hypoblast (primitive endoderm, which forms the yolk sac) and the trophoblast (trophectoderm (TE), which develops into the cytotrophoblast and the syncytiotrophoblast, invades the endometrium of the uterus, and forms the placenta). At implantation the epiblast flattens and together with the hypoblast forms the bilaminar embryonic disc [1–5]. Further characteristics of the post-implantation period are the process of lumenogenesis causing the amniotic cavity to open up and the separation of the extra-embryonic amnion (amniotic ectoderm) from the epiblast. Additional hallmarks are the formation of the primary yolk sac from the hypoblast, the development of primordial germ cells (PGCs), and the occurrence of two additional extra-embryonic lineages, the anterior hypoblast (analogous to the anterior visceral endoderm in the mouse) derived from the hypoblast and the extra-embryonic mesoderm (ExEM), for which it is unclear whether it originates from the epiblast, the hypoblast or the TE. Gastrulation is characterized by the development of the epiblast-derived primitive streak (PS), cell emigration from the PS, and the transformation of the bilaminar disk into the trilaminar disk containing ectoderm, mesoderm and endoderm. Gastrulation is followed by neurulation (the transformation of the ectoderm-derived neural plate into the neural tube, which later differentiates to the spinal cord and brain) and the preparation for early organogenesis (Fig. 1) [1–5].

Fig. 1.

Early human embryogenesis. Canonical lineage specification in the early human embryo, and illustration of the pre-implantation and post-implantation human embryo depicting its essential components (for details see the text)

The origin of descriptive investigations of the early human embryo is going way back to the time when histological specimens were obtained through surgery and the first snapshots of rare fixed embryonic samples were made available in the Carnegie or Kyoto collections. However, due to the inaccessibility of the in vivo condition, the scarcity of tissue material, restrictive legal regulations based on ethical concerns, and the evident experimental limitations a comprehensive understanding of the molecular mechanisms underlying human embryogenesis is missing [2, 6, 7]. The first successful in vitro-fertilization (IVF) in 1978 laid the foundation for the use of in vitro-cultivated donated surplus human embryos for research [8–12]. However, in many countries legislation follows the Warnock 14-day rule. Since the onset of gastrulation with the advent of the first manifestations of the future body plan and the nervous system is roughly mapped to day 14 after fertilization, it was agreed on that no viable human embryo should be cultivated beyond this limit. However, by nature cultivation for no longer than day 14 does not allow to study post-implantation events such as gastrulation [7, 13]. In addition, it became evident that knowledge of mammalian embryogenesis gained from animal model organisms such as the mouse can only partially be assigned to humans. Still, mouse studies were of high relevance to reveal many molecular mechanisms of developmental principles which are translatable to humans. In the context of preclinical modeling as well as for the establishment of pioneering protocols for the generation of stem cell-based embryo models research on this animal model made an indispensable contribution [1, 3, 4]. However, meanwhile well-described differences regarding cell fate patterning and tissue morphogenesis undermine cross-species comparisons. For example, compared to mouse embryogenesis the activation of the zygote genome and the induction of lineage-specific gene expression is delayed in humans. During human embryogenesis the epiblast-derived amnion is formed ahead of PS development, whereas in the rodent the genesis of the amnion is a consequence of the formation of the ExEM from the PS. Accordingly, it may not come as a surprise that, although the existing knowledge is still fragmentary, significant differences between humans and mice regarding the molecular networks underlying these processes have been reported [1, 3, 5, 14].

In the last decade, a variety of different protocols have been developed which make use of human pluripotent stem cells (hPSCs), such as embryonic stem cells (hESCs) or induced pluripotent stem cells (hiPSCs), to generate self-organized, multi-lineage embryo-like structures in vitro. By now, a remarkable range of different hPSC-based human embryo models has been reported [7, 15–17].

Non-integrated embryo models mimic only specific aspects of human embryo development and usually do not contain extra-embryonic lineages associated with the TE, hypoblast, or both. In contrast, integrated embryo models are composed of the relevant embryonic as well as extra-embryonic cell types and are designed to model the integrated development of the entire early human conceptus. In addition, integrated models could harbor the potential to undergo further development if cultured for prolonged time in vitro and might one day form fetuses what would represent a call for a detailed ethical discussion [6, 7, 15–18]. Currently, embryo models are exclusively used for research applications. Since these models do not harbor the potential to develop into human beings, they are considered to be associated with less ethical concerns than research with human embryos [6, 18]. The International Society for Stem Cell Research has categorized attempts to transfer human stem cell-based embryo models to the uterus of either a human or animal host as unethical prohibited research activities [19]. Here it is important to note that a fully integrated stem cell-based embryo model consisting of the complete spectrum of embryonic and extra-embryonic tissues and possessing the ability to develop into stages of functionality equivalent to a genuine natural embryo has not been reported yet [6, 7, 15, 19]. However, several of the currently existing models already fulfil certain criteria that make them attractive for disease modeling approaches, to study the effects of drugs, or to screen for improved media conditions for IVF. Furthermore, the investigation of various aspects of early human development does not depend on the replication of all features of the natural embryo and can thus be considered using the existing embryo models. For example, we made use of an established non-integrated embryo model to investigate the assembly and disassembly of the human embryonic basement membrane (BM) and to identify the transcription factor OCT4 as a major regulator of this process [20]. It is foreseeable that comparable approaches will gain in importance since more and more researchers will seek to employ human embryo models as a platform to address specific scientific questions.

The non-integrated stem cell-based human embryo models

The reported non-integrated embryo models include the two-dimensional (2D) micropatterned colony (MP colony) reflecting the gastrulation process [21], and three-dimensional (3D) models such as the post-implantation amniotic sac embryoid (PASE) mimicking the onset of gastrulation [22, 23], the peri-gastrulation trilaminar embryonic disc (PTED) embryoid [24], the epiblast model recapitulating anterior–posterior symmetry breaking around day 10 of natural embryonic development [25], the gastruloid mimicking embryonic development beyond day 14 [26], and the neuronal gastruloid [27, 28] (Fig. 2). The MP colony has been reported to contain cells with extra-embryonic properties in the periphery [21], but it is indefinite whether these cells represent ExEM, trophoblast or amnion [5, 19, 21]. The PASE has been shown to form extra-embryonic amniotic ectoderm [22, 23, 29], and the PTED embryoid was reported to exhibit amnion- and yolk sac-like structures [24]. However, regarding the definition of integrated models, which harbor hypoblast- and/or trophoblast-associated tissues and the potential to achieve the complexity where they could undergo further integrated development, the MP colony, the PASE and the PTED embryoid are considered to be non-integrated models. The same applies to the human epiblast model even when it is co-cultured with yolk-sac-like cells [30] or with extra-embryonic cells [31]. All non-integrated models mimic aspects of the post-implantation period and are generated by inductive procedures making use of chemical and physical triggers to prompt one stem cell entity into self-organization and differentiation (Fig. 2).

Fig. 2.

The non-integrated stem cell-based human embryo models. Comparison of the generation process, the modeled spatial condition, the represented embryonic period, and the main biological and structural characteristics (determined by transcriptional signature studies, antibody-based detection of marker expression and/or morphological analyses. For details see the text) of the reported non-integrated models. + indicates certain indications for the presence of this feature;—indicates the absence of this feature; +/− indicates that a specific type of neuronal gastruloid exhibits this feature and the other type does not; ? indicates that the existence of this structure is suggested but not finally proven; ND, not determined. The stem cell-based human embryo models’ designations were adopted from the authors according to the initial descriptions: MP colony (micropatterned colony) [21], PASE (post-implantation amniotic sac embryoid) [22, 23], PTED embryoid (peri-gastrulation trilaminar embryonic disc embryoid) [24], Epiblast model [25], Gastruloid [26], Neuronal gastruloid [27, 28]

The MP colony is developed by inducing hESCs to form circular micropatterns on slides with arrays of disks where extracellular matrix (ECM) drives cell adhesion. Upon BMP4 treatment self-organized radial patterns develop consisting of an ectodermal center, encircled by a mesodermal ring, where cells undergo epithelial-mesenchymal transition (EMT) and migrate inwards from a PS-like structure mimicking gastrulation, and an endodermal cell layer. The outermost ring is composed of extra-embryonic cells of unclear origin [21, 32]. In addition, a BM-like structure (detected by collagen IV staining) which separates migrating PS-like cells from the epiblast was visualized upon application of specific modifications of the BMP4-induction-protocol [33]. The MP colony is easy to establish, highly reproducible and is composed of cells of all three germ layers. However, its two-dimensionality does not reflect the in vivo condition and it lacks a disk-like epiblast morphology, bilateral symmetry, and a central lumen that could develop into an amniotic cavity [21, 32] (Fig. 2).

To shape the 3D peri-/post-implantation PASE hPSCs placed onto a soft gel bed and covered with ECM-containing media are triggered to form an amniotic sac-like structure. During PASE development hPSCs undergo lumenogenesis causing the amniotic cavity to open up, and the emerging extra-embryonic amnion separates from the disk-like epiblast. The latter develops further to form a PS-like structure with cells undergoing EMT and disseminating into the microenvironment [22]. In the PASE, the development of human PGC-like cells (hPGCLCs) can be studied [23] and the assembly of a BM, its composition, and its disassembly upon PS-development can be analysed [20] (Fig. 2).

The PTED embryoid is a transgene-free 3D-model, which recapitulates specific aspects of human peri-gastrulation development. To initiate this model hPSCs are seeded onto circular adhesive islands of ECM and are treated with exogenous BMP4. PTED embryoids exhibit a trilaminar embryonic disc structure. In the center region of PTED embryoids a distinct layered organization of ectodermal, mesodermal and endodermal cells was suggested by staining for embryonic germ layer markers such as SOX2, OCT4, and NANOG (ectoderm), BRA (mesoderm), and FOXA2 (endoderm). The trilaminar embryonic disc structure is flanked by dorsal amnion and ventral definitive yolk sac. In the latter primary haematopoiesis and blood cell generation are supposed. The PTED embryoid lacks the trophectoderm lineage and does not form a structure mimicking the primitive streak, but gives rise to PGCLCs [24] (Fig. 2).

To produce the early post-implantation/pre-gastrulation epiblast model hESCs dispersed in hydrogel supplemented with ECM are stimulated to form sacs upon lumenogenesis. This is followed by BMP4-induced anterior–posterior symmetry breaking and polarization into opposing regions representative for ectoderm and mesoderm. Whereas the epiblast model does not show amnion development, this model recapitulates the onset of gastrulation with the occurrence of a PS-like formation. Moreover, collagen IV stainings allowed the detection of a BM-like structure [25] (Fig. 2).

In the case of the non-integrated gastruloid hESCs treated with the WNT-activator Chiron and seeded into low-adherence plates are induced to form 3D aggregates without lumen or extra-embryonic lineages. Spatial transcriptomics revealed that it is a post-PS model, with elongation along the anterior–posterior axis, mimicking the natural embryo at day 17 to 21. It consists of cell derivatives of all three germ layers and shows features of somitogenesis. Since this model recapitulates human embryogenesis beyond the 14-day legal boundary, it can be of particular benefit to obtain so far unexplored insights into the transcriptional, epigenetic, signaling, and differentiation processes in later stages of human embryonic development [26]. Recently, the emergence of nascent PGCLCs without exogenous BMP signaling were reported for the gastruloid model [34] (Fig. 2).

Two protocols were reported to generate non-integrated 3D neuronal gastruloids mimicking early neurulation characterized by neural tube development. In one model, hPSCs treated with the WNT-activator Chiron and growth factors such as FGF, are induced to form an elongated gastruloid via forced aggregation followed by suspension culture. The resulting structure consists of neuronal- and mesodermal-, but not endodermal-like cell types, and is a valuable model to dissect the cellular organization and patterning decisions controlling the early development of the human nervous system [27] (Fig. 2). The induction of the second neuronal model, designated elongating multi-lineage organized gastruloid, is achieved by incubating hiPSCs with Chiron, FGF2 and other growth factors followed by three germ layer differentiation and a specific maturation process in a shaking culture [28]. This gastruloid model could not only be used to dissect the development of the central and peripheral nervous system, but allows to study the interplay of important organ formations such as spinal cord-like, primitive gut tube-like, and heart tube-like structures [28, 35] (Fig. 2).

Moreover, powerful protocols have been established to specifically model human neural tube development. These hPSCs-derived 2D and 3D in vitro-approaches are well suited for the distinct analysis of this particular stage of embryonic development rather than allowing the investigation of the developmental dynamics leading to the formation of the complete post-implantation embryo [36–42].

Integrated stem cell-based human embryo models

In essence, human blastoids consist of an epiblast surrounded by extra-embryonic cells and predominantly replicate stages of development before implantation. In several studies, extended cultivation procedures were employed to test for a putative post-implantation/pre-gastrulation potential (Fig. 3). Although all of these structures resemble the size, shape and cell number of human blastocysts, which form about four days after fertilization, they exhibit significant molecular, cellular and functional differences. One approach activates naïve hPSCs to develop into blastoids containing epiblast- and trophoblast-like structures. Although hypoblast-like cells were present, a defined layer could not be observed. This blastoid contains some cells that are not typically found in human blastocysts and exhibits limited post-implantation development potential. Noteworthy, these structures were not grown beyond the equivalent of a 14 days old conceptus mindful of the ethical concerns and legal limitations [43]. Another blastoid has been made by using fibroblasts reprogrammed to cells that have gene expression profiles matching the three cell types found in the human blastocyst. When these cells are triggered to aggregate, blastoids comprising an epiblast-like structure and hypoblast-like cells—not forming a defined layer—could be developed with low efficiency. This model also contains non-allocatable cell types, exhibits confined post-implantation developmental potential (these structures were also not grown beyond the 14-days limit), and the gene expression profile of its TE cells is more similar to post-implantation amniotic ectoderm [44]. A simple and highly efficient protocol was developed to drive naïve hPSCs to form a blastoid recapitulating the human blastocyst at embryonic days 6 to 7, which can attach in vitro with a spatial embryonic and extra-embryonic tissue organization very comparable to natural human embryos [45]. Two studies made use of extended/expanded hPSCs, either assembled with pre-differentiated TE cells [46] or as the sole cell source [47]. Both protocols display low formation efficiency and the resulting blastoids show only confined resemblance to natural blastocysts, with the occurrence of intermediate cells and a limited potential on post-implantation development. Noteworthy, in the latter amnion markers were detected in the TE clusters [47]. A highly efficient strategy using naïve hPSCs achieved the maturation of a blastoid composed of an epiblast, hypoblast, and a trophoblast forming a polar TE. Although its transcriptome was proven to have a high match with human blastocysts it shows limited post-implantation development upon extended culture [48]. Two reports describe the in vitro genesis of blastoids from eight-cell stage-like cells. In one approach, these cells were isolated from human pre-implantation epiblast-like stem cells, and after successful expansion and maintenance induced to form blastocyst-like structures consisting of ICM and TE lineages [49]. In the second study, eight-cell stage-like cells generated from hPSCs were aggregated to form blastoids with a considerably high efficiency. This blastocyst model displays a blastocoel-like cavity and distinct inner and out layers resembling the ICM and TE. In addition, RNA sequencing revealed clustering of epiblast-like, TE-like and hypoblast-like cells [50]. Of note, in both reports blastoid generation was performed based on the published protocols by Kagawa et al. [48] and Yanagida et al. [45]. In another recent approach, originating from naïve hESCs extended blastoid cultivation on 3D matrices for up to 21 days has been reported. In this model epiblast lumenogenesis, an amniotic cavity, the differentiation of trophoblast-like lineages, the invasion of extra-villous trophoblast cells, and PS marker expression could be detected. Transcriptome analyses at day 21 further suggested the emergence of developmental attributes such as amnion, PGCs, ExEM, and Yolk sac endoderm [51]. Furthermore, primed hESCs were prompted to undergo the primed-to-naïve conversion process and the so obtained population of heterogeneous intermediate cells exhibiting epiblast, TE and hypoblast signatures was induced to form a cavity-containing blastoid. The quantitative comparison of the ratios of OCT4-positive (epiblast-like), GATA6-positive (hypoblast-like), and CDX2-positive (TE-like) cells in natural blastocysts and so derived blastoids revealed that the blastoid contains more epiblast-like cells and fewer TE-like cells than its natural counterpart. Further cultivation of these pro-amniotic cavity developing structures in a 3D culture system facilitating pre-gastrulation development triggered the expression of an anterior hypoblast marker, but not of a PS marker [52]. Finally, it was recently reported that naïve hPSCs can spontaneously give rise to blastoids upon switch to suspension culture in inductive medium. The so produced cavity-containing cysts are composed of epiblast-like cells, TE-like cells, and a few hypoblast-like cells. Further cultivation and differentiation for 14 days in total using an already established protocol [51] showed that these blastoids have the potential to progress further towards post-implantation stages as demonstrated by the appearance of descendants of the cytotrophoblast, syncytiotrophoblast, amnion, anterior hypoblast, PS, and yolk sac endoderm [53].

Fig. 3.

The integrated stem cell-based human embryo models. Comparison of the generation process, the modeled spatial condition, the represented embryonic period, and the main biological and structural characteristics (determined by transcriptional signature studies, antibody-based detection of marker expression and/or morphological analyses. F or details see the text) of the reported integrated models. + indicates certain indications for the presence of this feature; − indicates the absence of this feature; +/− indicates that specific types of blastoids exhibit this feature and other types do not; ( +) indicates that only limited evidence for the structure has been described; ? indicates that the existence of this structure is suggested but not finally proven; ND, not determined. The stem cell-based human embryo models’ designations were adopted from the authors according to the initial descriptions: Blastoid [43–53], Extra-embryoid [54], Embryoid [60], E-assembloid (embryo-like assembloid) [55], Peri-gastruloid [56], SEM (stem-cell-based embryo model) [61], Bilaminoid [58], eX-embryoid (extra-embryonic niche and yolk sac haematopoiesis embryoid) [59], Gastruloid [57]

The other integrated human embryo models are exclusively mimicking post-implantation development and can be grouped according to their generation process (induced or assembled) or the presence of extra-embryonic lineages (hypoblast or hypoblast/trophoblast). The extra-embryoid [54], the embryo-like assembloid (E-assembloid) [55], the peri-gastruloid [56], and the gastruloid [57], which have in common that they contain hypoblast but not trophoblast lineages, are generated via inductive approaches. The bilaminoid [58] and the extra-embryonic niche and yolk sac haematopoiesis embryoid (eX-embryoid) [59], both containing only hypoblast cells, as well as the embryoid [60] and the stem-cell-based embryo model (SEM) [61], containing both hypoblast and trophoblast lineages, result from procedures, where structure formation is achieved by the assembly of different specific stem cells (Fig. 3).

For the generation of the extra-embryoid hESCs held in intermediate states between ground and primed pluripotency are induced to spontaneously differentiate into a mixture of epiblast-like and hypoblast-like cells. Upon further development the final structure consists of an amniotic cavity separating the amnion from the epiblast, surrounded by a pronounced ring of hypoblast cells that includes an anterior hypoblast-like domain. Whereas this model lacks trophoblast cells, and gene expression representative for PGCs or ExEM could not be detected, it shows a variety of other characteristics of the peri-gastrulating embryo such as a PS-like structure, epiblast-derived cells breaching the BM in the course of a migration event, mesoderm development, and initial yolk sac induction albeit without formation of a fully developed yolk sac [54].

A mixed cell population composed of unmodified hESCs, GATA6-SOX17 overexpressing hESCs with induced hypoblast-associated gene programs, and GATA3-AP2γ-overexpressing hESCs with induced trophoblast-associated gene programs was prompted to self-organize into integrated embryoids. This approach resulted in a post-implantation embryo-like structure of high complexity. The epiblast-like domain undergoing lumenogenesis is encompassed by trophoblast-, hypoblast-, and ExEM-like tissues and equivalents of amnion, PGCs, and anterior hypoblast are formed. In addition, yolk sac-like cell differentiation without the formation of a regular yolk sac structure as well as laminin expression, suggesting the development of a BM, can be detected [60].

To create the E-assembloid pre-aggregated, naïve hESCs are exposed to BMP4-induced hESCs designated as signaling nest cells (SNCs). Although SNCs harbor a trophoblast-like identity, they do not form the trophoblast in E-assembloids but rather facilitate aggregation and self-organization of the structure-lending hESCs. The final embryo model contains cellular entities representing the bilaminar disk, the amnion and amniotic cavity, and the yolk sac—all encompassed by an extra-embryonic cell type, which somehow resembles ExEM. This model lacks trophoblast lineages, but contains a PS-like structure, mesoderm, PGCLCs, and an anterior hypoblast-like region [55].

For the generation of the peri-gastruloid, hPSCs with expanded potential cultivated under hypoblast-promoting conditions as well as in culture media supporting balanced hypoblast and epiblast differentiation were induced to undergo lineage segregation, cell sorting, and 3D self-organization. This complex procedure triggers the genesis of an embryo structure with amniotic- and yolk sac-like cavities, a bilaminar disc-like figure and a PS-like structure. Moreover, this model mimics gastrulation, the formation of a trilaminar disc, the specification of PGCs, and features of initial stages of neurulation and early organogenesis. Single-cell transcriptomic analyses suggested the presence of cellular entities representing neural ectoderm, non-neural ectoderm, blood and vascular progenitors, ExEM, and allantois among other developmental structures. However, this very advanced structure still lacks the formation of the trophoblast lineage [56].

To form a SEM naïve hESCs are first differentiated into extra-embryonic cell types including hypoblast-like, trophoblast-like and ExEM-like lineages, which are then prompted to aggregate with untreated naïve hESCs. This model recapitulates a variety of compartments and lineages of the post-implantation embryo such as the epiblast, the hypoblast, the anterior hypoblast, the amnion, the yolk sac, the ExEM, and PGCs. Although this model contains trophoblast tissue with syncytiotrophoblast-like characteristics, cytotrophoblast-like cells cannot be detected [61].

The bilaminoid has a 3D bilaminar structure and provides evidence for a hypoblast-mediated regulation of epiblast development. First, hypoblast-like cells were generated from naïve hPSCs via GATA6 expression. Next, aggregation of the so derived cells with untreated naïve hPSCs triggered the latter to develop epiblast-, amnion- and amniotic cavity-like formations, a PS-like structure, and cells resembling anterior hypoblast, PGC, and mesoderm. The final structure is surrounded by a ring of hypoblast-like cells derived from the GATA6-expressing hPSCs. Furthermore, RNA-sequencing data revealed the expression of several laminins, and laminin stainings demonstrated the generation of a BM-like structure between the epiblast and the surrounding hypoblast [58].

The eX-embryoid is a structure that contains equivalents of extra-embryonic niche and yolk sac haematopoiesis. When primed hiPSCs are co-cultivated with hiPSCs expressing GATA6 (to induce hypoblast fate) on a culture dish the latter encapsulate the non-treated hiPSCs and the formation of a bilaminar disk-like structure is initiated. This model further entails some cells with anterior hypoblast-like characteristics, amnion- and amniotic cavity-like strucutres, but lacks trophoblast-like cells. And whereas a real yolk sac cavity is not formed, the eX-embryoid shows features of an amnion facing the cell culture dish, an amniotic cavity, and an epiblast, all covered by a structure, which is similar to a flattened yolk sac cavity. The extra-embryonic layer of yolk sac endoderm exhibits yolk sac-tissue like morphogenesis harboring distinct stages of a haematopoiesis-like process associated with the emergence of erythroid-, megakaryocyte-, myeloid-, and lymphoid-like cells. Furthermore, staining approaches suggested that during the development of this embryoid the cell clusters derived from the unmodified hiPSCs were surrounded by a laminin-positive membrane [59].

In June 2023 a preprint was posted on bioRxiv describing another human embryo model containing hypoblast-like tissue. The authors generated a 3D gastruloid by employing a two-step induction-process initiated by the differentiation of hPSCs into hypoblast-like cells while preserving partial pluripotency. Aggregates of these cells were found to establish polarity and organized into an OCT4-positive cell layer and a yolk sac. In the second step, incubation with BMP4 and FGF additionally triggered the development of an amnion-like entity from epiblast-like cells. At its final stage, this model forms a bilaminar disc-like structure of epiblast and hypoblast-like cells and exhibits features of the amniotic cavity, the PS, a LAMC1-positive BM, the anterior hypoblast, mesodermal progenitors, blood islets, ExEM components and PGCs, but lacks TE-lineages [57]. Whereas data obtained with the integrated embryoid model provide evidence for ExEM differentiaton from the epiblast [60], this report suggests that ExEM originates from the yolk sac [57].

Human embryo models in translational research and biomedical applications

The long lasting interest in early human embryogenesis is stimulated by the endeavor to glean insights into the mystery of the very beginnings of human life. In addition, the clinical relevance of a better understanding of the peri-implantation period is highlighted by the fact that failures of embryo implantation or aberrations of embryonic development occur in up to 50% of human pregnancies with a clear majority occurring in the first weeks after fertilization [62, 63]. Still, owing to the variety of reasons discussed above the molecular understanding of this phase of human development is currently immature. This applies in particular to the post-implantation period hidden in the uterus womb, since pre-implantation development can at least be studied in countries where human embryo research is not prohibited by law [2, 6, 7, 64]. By overcoming the ethical, legal, biological, and experimental hurdles of research on human embryos, in vitro modeling using the toolbox of the different available hPSC-derived structures could shed light on specific processes of early human development. However, these expectations will only be met when the most appropriate model or models for the selected hypothesis-driven scientific question can be defined and chosen (Figs. 2 and 3). For the generation of the various embryo-like structures different kinds of hPSCs and different procedures with regard to the applied morphogens, transgenes, and instructive signals are employed. All existing stem cell-based models are only imperfect replicas and it is not always clear how closely they resemble the in vivo situation. Furthermore, each model recapitulates different intricacies of embryogenesis. However, probably with the exception of specific types of blastoids all of them can be used to study post-implantation development at varying extents (Fig. 4). On the one hand, structures modeling more advanced developmental stages could be applied to capture a more dynamic and comprehensive picture of human embryogenesis. So far, a few embryo models, including the non-integrated MP colony [21], the PTED embryoid [24], the gastruloid [26, 34] and the neuronal gastruloid [27, 28], as well as the integrated peri-gastruloid [56], have been suggested to exhibit cells of all three germ layers, neurulation-like processes, or features of early organogenesis. At present, the latter [56] represents the most advanced reported model. On the other hand, more reductionist human embryo models could allow to focus on the investigation of specific developmental processes without influencing effects of other structures or of unassignable cells [17, 65] (Fig. 4).

Fig. 4.

The putative scope of applications of the different stem cell-based human embryo models. Evaluation of the stem cell-based human embryo models with regard to their putative usability for specific investigations or applications. − indicates that this embryo model cannot be used; + , + + , and + + + indicates ascending degrees of usability; +/− indicates that only specific types of this embryo model can be used. The stem cell-based human embryo models’ designations were adopted from the authors according to the initial descriptions: MP colony (micropatterned colony) [21], PASE (post-implantation amniotic sac embryoid) [22, 23], PTED embryoid (peri-gastrulation trilaminar embryonic disc embryoid) [24], Epiblast model [25], Gastruloid [26], Neuronal gastruloid [27, 28], Blastoid [43–53], Extra-embryoid [54], Embryoid [60], E-assembloid (embryo-like assembloid) [55], Peri-gastruloid [56], SEM (stem-cell-based embryo model) [61], Bilaminoid [58], eX-embryoid (extra-embryonic niche and yolk sac haematopoiesis embryoid) [59], Gastruloid [57]

Furthermore, it is important to note that in addition specialized protocols have been established that allow the selective in vitro-investigation of very specific stages of human embryogenesis rather than studies on more comprehensive embryo models. As an example, hPSCs-based structures have been generated, which specifically mimic human somitogenesis [66–71].

Compared to the low number of donated natural embryos a strongly compelling advantage of the manipulable embryo models is their usability for experimental interventions. They are perfectly suited to investigate the effects of biochemical manipulations, or genetic modifications implemented by genome editing, siRNA-mediated knockdown, or the use of hiPSCs derived from mutation carriers. For example, blastoids are known to be more amenable to genetic modifications than natural blastocysts [43–53]. However, whereas genetic modifications are straightforward approaches in models generated by coaxing a single hPSC-type to self-organize, they are slightly more complex to achieve in structures assembled from different stem cell types (Fig. 4). On the other hand, assembled models could facilitate to selectively study the impact of genetic modifications on different contributing cell types and lineages. Finally, embryo models enable researchers to assess the effects of modifications in quantity and quality otherwise not feasible given the limited availability of natural human embryos for research. Embryo-like structures can be employed in higher-scale studies and multi-omics approaches, including genomic, transcriptomic, proteomic, and metabolomic analyses, and the dynamics of self-organization, cell differentiation, and morphogenesis can comprehensively be followed by time-lapse imaging (Fig. 4).

There is good reason to believe that in future the usage of stem cell-based human embryo models in fundamental research approaches will substantially contribute to a more comprehensive understanding of the essential events and processes of embryonic development (and their species-specific differences) such as self-organization, pattern formation and morphogenesis (this aspect has detailly been reviewed in [7, 15–17, 72]). Below we discuss future possibilities to use stem cell-based human embryo models for translational research and biomedical applications.

Understanding and modeling diseases

Embryo models could represent a paradigm shift for the modeling and investigation of a significant proportion of human diseases. Since many years, the use of organoids derived from hPSCs, especially from patient-derived hiPSCs, is an established approach to model a variety of human maladies [73, 74]. However, chromosomal aberrations, single gene disorders, multi-factorial congenital malformations, or specific infectious diseases can have their earliest manifestations in early human embryogenesis. In future, these defects could be studied using stem cell-based human embryo models (Fig. 4). Natural human embryos often harbor chromosomal aberrations, which cause embryonic arrest, impair implantation, or trigger miscarriage. However, the cellular processes affected by aneuploidy remain elusive. Recently, the MP colony model was demonstrated to serve as a valuable embryo-like structure to study the regulation and fate of aneuploid cells in early embryogenesis [75]. Huntington´s disease is a late onset single gene neurodegenerative disorder, which is initiated when the huntingtin gene exhibits a CAG repeat expansion and for which no therapy yet exists. Studies using the MP colony model revealed that huntingtin gene mutations mediate their earliest embryonic effects (impairment of germ layer patterning through polarity defects) as early as the second week of human embryogenesis [76]. These reports highlight the relevance of embryo models to gain completely new insights into the relation of genetic aberrations or disorders and developmental defects. Furthermore, they provide evidence for the applicability of studies on genetic modifications using embryo models with the aim to assess the earliest effects of single gene mutations, polymorphisms, or the multi-factorial interaction of genetic predispositions and environmental factors for embryo development and human maladies (Fig. 4). About 3% of all live born infants exhibit a congenital anomaly. Some of them, including neural tube defects, congenital heart defects, or specific skeletal abnormalities, have a multi-factorial genesis and are caused by perturbations in the early patterning stages of human development [7, 77]. All human embryo models, especially the neuronal gastruloid [28, 35] and the peri-gastruloid [56] exhibiting additional features of neurulation and early organogenesis, could be helpful to gain further insights into the development of these conditions (Fig. 4). And finally, during the pandemic a susceptibility of the first trimester natural human embryo was suggested because it expresses the SARS-CoV-2 receptor ACE2 [78]. Obviously, stem cell-based embryo-like structures could also be very useful to contribute to the still poorly explored question how infections affect early human embryogenesis (Fig. 4).

Developing new drugs and studying teratogenic effects

Embryo models in general and especially those designed for disease modeling can be useful to test the effects and side effects of new drugs, to learn about their mechanisms of action, to define the best dosage, or to clarify how they interact with other therapeutic concepts. In the past, this information was often gathered through pre-clinical pharmacokinetic and pharmacodynamic studies using monolayer cell cultures lacking the spatiotemporal context or animal models. A prominent example for the risk of misinterpretation due to species-specific differences is thalidomide which does not cause teratogenic effects in the mouse but can trigger serious, life-threatening birth defects in humans [79]. With the aim to test the suitability of the integrated gastruloid for the investigation of adverse drug reactions, it was found that exposure to thalidomide results in the developmental arrest of this human embryo model [57]. To enable the application of embryo models in high-throughput drug and toxicity screens, platforms, such as e.g. the microfluidic device for PASE formation [23], allowing the recapitulation of developmental landmarks in a highly controllable and scalable fashion, must be developed. In addition, the fluidic chambers used in this model [23] as well as e.g. the customized culture surfaces employed to grow MP colonies [21] allow the controlled channeling of compounds of all kinds. Such bioengineering tools making protocols more reproducible, efficient, and scalable together with the absence of disorganized or non-assignable tissues are good arguments for the application of non-integrated models in drug screenings [80] (Fig. 4).

Generally speaking, the relevance of such attempts is underlined by the fact that the pre- and peri-implantation period is considered one of the most sensitive windows in human development. Accordingly, pharmaceuticals intended for use in patients of childbearing potential need to be tested for embryotoxicity or teratogenicity prior to clinical application [81]. It is evident that in future hPSCs-derived embryo models could be applied to gain insights into the mechanism of teratogens, such as toxins, alcohol, lead, high levels of radiation exposure, and probably also nanoplastic. The PASE model has already successfully been applied to assess the differential teratogenic potentials of the two over-the-counter painkillers acetaminophen and ibuprofen, of which the latter is not recommended for use during pregnancy, of penicillin, regarded as safe for pregnancy, of the chemotherapeutic drug doxorubicin, known to be teratogenic, of the anti-depressant citalopram and of the anti-inflammatory dexamethasone [82]. Another study proved the non-integrated gastruloid to represent a useful tool to determine the differential teratogenic effects of valproic acid, all-trans retinoic acid, bosentan, applied for the treatment of pulmonary arterial hypertension, thalidomide, the anti-epileptic phenytoin, ibuprofen and penicillin [83].

Improving assisted reproductive technologies

Not only the development of the natural human embryo itself, but also the regulation of its interaction with the inner surface of the uterine wall remains largely elusive due to ethical and technical barriers. The implantation process is marked by the initiation of contact between the placenta-forming TE, exposed upon hatching of the blastocyst from the zona pellucida, and the receptive maternal endometrium. The sum of perturbations affecting either the development or the implantation of the human conceptus is responsible for many pregnancy failures [62, 63]. Next to a variety of maternal or paternal factors, deregulation of these processes is a relevant cause for both infertility and the low success rates of IVF. In the western world infertility has significantly increased to 8–15% infertile individuals at reproductive age and the IVF success rates are still low—on average probably about 25% with significant variations based on a woman's age [84, 85]. It is the hope of researchers, clinicians, and patients alike that stem cell-based human embryo models could once contribute to a more comprehensive understanding of embryo implantation, the design of better reproductive technologies, and a decrease in early pregnancy loss rates.

Today, selection of embryos in the blastocyst stage prior to their transfer into a uterus can be performed upon non-invasive assessment of morphological and morphokinetic parameters or upon embryonic cell biopsy-based aneuploidy screening, single gene disorder analysis, or genetic screening for polygenic traits [86]. Blastocyst-like blastoids could be used to identify non-invasive embryonic biomarkers via e.g. deep-learning approaches using time-lapse microscopy [87, 88], what could facilitate the identification of the most viable embryo and improve predictions of IVF success rates. Noteworthy, a robust video analysis pipeline that incorporates the use of machine learning methods to fully characterize the process of hPSC self-organization into a PASE has been established [89]. In addition, single-cell multi-omics analyses of TE cell biopsies in the course of pre-implantation genetic diagnosis could in future also be used to check for patterns of biomarkers, which have been suggested from blastoid studies to be predictive for blastocyst vitality, successful implantation, proper placenta formation, and accurate embryo development (Fig. 4). Comparable strategies could also be employed using blastoids to search for optimized culture conditions for IVF with the aim to establish a more physiological in vivo-like environment by avoiding embryo exposure to artificial stressors mediated e.g. by specific media formulations (Fig. 4). And importantly, blastoids could be used for implantation studies. To model peri- and early post-implantation processes blastoids were e.g. analysed in in vitro attachment assays in optical-grade tissue culture plates [44], syncytiotrophoblast marker-expressing cells were induced by blastoid attachment to ECM-coated dishes [45], or blastoids generated by an already established protocol [48] were attached to laminin-coated plastic substrates and cultivated in medium containing a ROCK inhibitor and ECM [90]. Furthermore, a blastoid with local maturation of polar TE has been demonstrated to attach to hormone-stimulated endometrial organoids in 2D (but not to unstimulated organoids) what could be impaired by the contraceptive levonorgestrel [48]. Finally, feto-maternal assembloids mimicking human embryo implantation have been generated by co-cultivation of a blastoid [43] and a hormone-responsive endometrial organoid what triggered blastoid-derived syncytial cells to infiltrate the organoid similar to the primitive syncytiotrophoblast during embryo implantation [91].

Embryo models could also contribute to the development of novel assisted reproduction approaches. Light on this multi-faceted aspect can e.g. be shed by addressing the current status of a method called in vitro-gametogenesis (IVG). It was demonstrated that in specific cellular environments or upon introduction into ovaries or testis mouse PSC-derived PGCs can be differentiated into eggs and sperms, which can then be used to generate viable offspring [92–96]. Since at its current developmental stage this approach raises several ethical concerns and safety issues, its application in human reproduction is currently categorized as a prohibited research activity [18, 19]. However, although hPSCs can so far be only developed into rudimentary human oocytes and sperms [97, 98], several companies, well endowed with venture capital funding, already bank on the translation of IVG to the clinics, probably coming to pass not later than in the next ten years [99]. On this route to translation human embryo models could be of assistance. First, knowledge on human PGC development is still limited. Embryo models, which were shown to develop PGCLCs [23, 24, 34, 51, 55–58, 60, 100] could serve as valuable tools to obtain important insights into the underlying molecular mechanisms of PGC development (Fig. 4). In this context, the detected PGCLC differentiation in models, which do not develop trophoblast structures [23, 24, 34, 55–58, 100], already suggest that TE might not be required for PGC development. Furthermore, studies with the integrated embryoid and gastruloid models provided evidence for the epiblast as the origin of PGC generation [57, 60]. And second, in order to circumvent potential complications caused by genetic or epigenetic aberrations, the use of e.g. fetal (stem) cells supposed to harbor fewer aberrations [85, 101] or the establishment of new protocols for the in vitro reconstitution of epigenetic reprogramming in the human germ line [102] have been suggested. It is reasonable to evaluate whether embryo models exhibiting PGCLC differentiation potential could once be harnessed to generate hiPSC-derived gametes to help advice-seeking patients to conceive healthy genetically related children (Fig. 4). The plausibility of such a putative clinical application is highlighted by a pioneering work showing that variants of the initial PASE protocol [22] can be employed to generate hPGCLCs [23, 100]. Esfahani et al. found that on day 3 of the protocol the cysts contain small clusters of hPGCLCs positive for the early hPGC markers TFAP2C, NANOG, SOX17, OCT4, and BLIMP1. Flow cytometry analyses demonstrated the proportion of these hPSCLCs in the culture to be around 20%, what is comparable with the efficiency of other hPGCLCs derivation approaches [97, 103]. In this study this line of action was successfully applied to generate hPGCLCs from hiPSCs derived from patients with obstructive azoospermia, a severe form of male infertility [100].

Developing cell and tissue transplants

In transplantation medicine the global shortage of donor organs is the driving factor for research on alternative solutions such as xenotransplantation, or cell-based approaches including tissue engineering and organoid technologies. As a logical consequence of the progress in the field of stem cell-based embryo models the discussion to which extent these structures could one day be used to develop human organs has gained momentum. The concept would evidently concentrate on the advanced cultivation of embryo models in bioreactors with the aim to nurture the development of specific organ progenitors into sizeable, functional human organs [104].

Meanwhile, research is already underway to lay the foundation for the use of embryo models to produce cells and tissues for transplantation purposes. On the one hand, one could argue that for this purpose the experimental value of an embryo model depends on its spectrum of the modeled embryonic features. However, independent of the developmental level and complexity one might prefer focusing on a model exhibiting the specific cell type of interest. As discussed above, for the generation of patient-specific gametes hiPSC-derived embryo models with the potential of hPGCLC differentiation [23, 24, 34, 51, 55–58, 60, 100] should be selected. In a similar context, for the creation of e.g. personalized blood stem cells, which could be used for transplantation replacing blood stem cells of donors, embryo models generated from patient-derived hiPSC could be used. The PTED embryoid show features of active haematopoiesis and blood cell generation in the yolk sac-like structure [24]. The peri-gastruloid contains blood and vascular progenitor-like cells [56], the eX-embryoid exhibits the emergence of blood progenitor-like cells during the primitive and erythro-myeloid stages of haematopoiesis [59], and the integrated gastruloid produces cellular entities representing blood islets [57], which could hold promise for research on the development of cell therapeutic applications (Fig. 4).

Replacing animal research

Since the first emergence of stem cell-based organoids the debate about how these technologies could affect the ethical justification of animal research continues. Due to the current limitations of stem cell-based models such as their lack of blood vessels and the missing interaction with the immune system it is considered that animal research will not become entirely obsolete. However, it has been argued that justification of animal experiments over stem cell-based technologies might become mandatory on a case-by-case basis [105]. And for many scientific investigations the use of in vitro- and in vivo-approaches will stay complementary, rather than in competition [105, 106]. Besides the ethical concerns regarding human embryo research, the goal to reduce, refine, and replace animal experimentation is an additional strong motive to use stem cell-based models of the human embryo in many of the here described applications (Fig. 4).

Limitations of the existing stem cell-based human embryo models

Not only the non-integrated stem cell-based human embryo models, but also all the existing integrated models are only imperfect replicas of the human embryo not even close to functional equivalence to the genuine natural human counterpart. Furthermore, they can contain off-target differentiation. Some cell types within the embryo model do not match in vivo counterparts and others do not reflect the spatiotemporal in vivo development. Several of the described models contain unclear cell type compositions and disorganized tissue structures [6, 7, 15, 19, 72]. None of the blastoids reported so far is expected to harbor the potential to develop into a complete human embryo, and fully organized embryo-like structures could not even be achieved by prolonged blastoid cultivation. Hitherto, the reported blastoids at large do not recapitulate the full complexity of natural blastocysts. They lack a zona pellucida, often exhibit ratios of the embryonic cell types, which are distinct from the naturally occurring ratios and occasionally include unidentified cells types with intermediate or undefined expression patterns [4, 107–109]. Furthermore, the other reported integrated models do also not harbor the potential to develop into a viable embryo, because they do not contain functional trophoblast tissue. Also, the embryoid [60] and SEM [61], which exhibit lineages derived from the TE, do not harbor the capacity to implant or to generate placental tissue needed for the ongoing support of the fetus, since the described trophoblast-like cells do not exhibit gene expression profiles characteristic of the trophoblast lineage or the trophoblast is not well developed, respectively [7, 60, 61]. And importantly, all these post-implantation models skip the first week of development and do not allow investigations of the pre-implantation period [43–61] (Fig. 3).

On the other hand, each of the reported non-integrated and integrated models has a distinct potential that could be used to address specific research questions (Figs. 2 and 3). Since each model also has specific limitations, it is important to strategically evaluate the suitability for the scientific inquiry of interest and the planned experimental design [7, 15–17, 65, 72]. This is particularly challenging since it is not always clear how closely a model resembles the in vivo situation. Recently, criteria for the standardization of stem cell-based embryo models have been proposed [110]. It was requested that beside the description of the starting materials (e.g. the pluripotency state and genome integrity of the used stem cell line) and the applied protocols, a detailed characterization regarding the cellular composition, spatial organization, morphological features and fidelity level of the cells and structures of each model should be provided [110]. In this context artificial intelligence-based lineage tracing might represent a helpful tool in future [89]. To be able to assess the commonalities of and differences between a specific structure of the stem cell-based embryo model and of the natural embryo meaningful benchmarking criteria must be applied [110]. Whereas transcriptome analyses, especially single-cell spatial transcriptomics, are very useful tools for characterization, one must always keep in mind that transcriptional signature studies do not allow conclusions on the functionality of the so identified structure. Epigenome, proteome and metabolome studies, the detection of marker expression via immunofluorescence microscopy, the investigation of the morphology of the different structures, time-lapse imaging of the spatiotemporal sequence of specific events or biochemical and genetic manipulations could be applied for a more comprehensive characterization of a stem cell-based model. Since such extensive investigations are hardly manageable within one study in one laboratory, the reproducibility of a specific model in different laboratories is of highest relevance. So far, most of the reported stem cell-based human embryo models have not been reproduced beyond the original report (Figs. 2 and 3). In addition, the repeatability between multiple experimental approaches using the same protocol in the same laboratory must also be improved. For applications such as biochemical manipulations, genetic modifications, drug testing, studies on teratogenic effects or the optimization of IVF culture conditions a high level of consistency and reproducibility and a low level of batch variability in embryo model generation would be essential for a meaningful interpretation of the obtained results. However, a high degree of standardization is challenging because the models often exhibit differences in shape, sizes, organization of the lineage composition and structures even within the same experiment [7, 15–17, 65, 72].

Blastoids could be promising tools for studies on the identification of biomarkers to improve predictions of IVF success rates, for the optimization of IVF culture conditions and for implantation studies (Fig. 4). Although the most efficient human blastoid protocol shows over 70% formation efficiency with 97% commitment to each of the three lineages [48], the efficiency of other blastoid formation protocols is still quite low ranging from 28 to 58% [50] down to only 10% and lower [43, 61]. Especially for studies on the regulation of the implantation process it is to be hoped that it may not be too long before improved protocols for the generation of blastoids, biocompatible polymers, and endometrial organoids [91, 111, 112] will come to the fore. In future, multi-modal organoid co-cultivation approaches will allow to provide more detailed insights into functional interactions in different contexts. With regard to biomedical application another limitation of blastoids is the lack of the very early stages of human development characterized by cleavage divisions and the development of the multi-cellular morula, which are known to be particularly susceptible to failure [72].

Finally, depending on the research question a more reductionist model could be the better choice. On the one hand, choosing a more complex model does not only allow the investigation of a larger range of embryonic characteristics and features, but also bypasses the shortcoming that isolated types of cells or tissues might not exhibit their full spectrum of in vivo-functions due to lack of key paracrine interactions with adjacent tissues. On the other hand, integrated models more often contain unassignable or disorganized cell types and tissues [43, 44, 54–56, 60, 61] which might have a distorting impact on the morphological and functional characteristics of the studied components and processes.

Conclusions

It is foreseeable that in the near future the use of stem cell-based embryo models will increase. However, for a wider application the efficiency, reproducibility and scalability of the embryo model protocols must be improved, the conditions for prolonged cultivation must be optimized, and there is an obvious need to avoid the occurrence of unassignable or disorganized cell types and tissue structures in the embryo-like structures. In addition, for many of the investigations and applications discussed above more advanced stem cell-based embryo models with a higher level of cellular complexity and organization must be generated. The attempts to create improved embryo models will also further include research on natural human embryos serving as the unique benchmark to assess the fidelity of each structure contained in the embryo model compared to the in vivo-counterpart. And finally, a comprehensive discussion of the ethical aspects of this kind of research is indispensable. Obviously, the usage of stem cell-based embryo models for reproductive purposes calls for prohibition. However, a regulatory network regarding their application in research must also be created. In this context it is arguable that integrated models should follow more stringent regulations than non-integrated models and that the least challenging stem cell-based human embryo model, that allows to address the specific question, should be chosen whenever possible [2, 6, 17, 18].

Author contributions

MR, SH, IK, MH conceptualized and wrote the manuscript.

Data availability

Not applicable.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors have declared that no competing interests exist.