A bioengineering approach to guide non-integrated stem cell-derived embryo model towards yolk sac development and hematopoiesis Authors: Vinidhra Shankar, Ensieh Zahmatkesh, Laury Roa Fuentes, Fabiola Marques Trujillo, Willine van de Wetering, Carmen López-Iglesias, Florian Caiment, Clemens van Blitterswijk, Erik Vrij, Stefan Giselbrecht Published: 2026 ABSTRACT Combining stem cell-derived multi-tissue constructs, such as non-integrated embryo models, with bioengineering technologies provide scalable tools for tissue engineering and drug testing. Here, we show the controlled initiation of organogenesis in XEn/EpiCs, a partial chemically-controllable mouse embryo model emulating aspects of E5.5 stage embryo, within milliwells. XEn/EpiCs spontaneously progress to develop a fluid-filled cavity, resembling a yolk sac encasing an embryonic compartment. Locally restricted inner regions of yolk sac-like (YS+) structures undergo cardiogenesis, displaying spontaneous contractions, and outer regions initiate vasculogenesis. Some vascularized YS+ structures develop blood islands composed of primitive erythrocytes, endothelial cells and mesenchyme. Further, we developed a continuous single-step culture using thermoformed milliwell platforms, enabling live tracking, pathway modulation, and in situ visualization of XEn/EpiCs towards organogenesis. Collectively, we demonstrate a multi-tissue vascularized embryo model devoid of trophoblast cells and external stimuli, capable of undergoing post-implantation morphogenesis and amenable to modulation within micro-engineered platforms for diverse applications. Introduction The convergence of tissue engineering and stem cell biology represents a crucial advancement in the development of multicellular, functional constructs with sophisticated architecture. Such constructs are urgently needed for a range of in vitro and in vivo applications, including regenerative medicine, developmental studies, and disease modeling (Kwon et al., 2018; Gholobova et al., 2020). By combining stem cell-based constructs that self-organize into functional organotypic structures in response to biochemical and biomechanical cues (Evangelopoulos et al., 2016; Murphy et al., 2013), researchers have derived tissue analogues such as skin (Sierra-Sanchez et al., 2021), muscle (van der Wal et al., 2023), bladder (Adamowicz et al., 2020), cartilage (Nordberg et al., 2024) and others, for therapies. However, these analogues often lack the morphological and functional complexity of their in vivo counterparts, posing challenges for direct and functional integration into host tissues. The advent of stem cell-derived embryo models represents a new pioneering avenue in tissue engineering, addressing longstanding limitations of traditional methods (Shankar et al., 2021; Rossant and Tam, 2021; Luijkx et al., 2022; Bao et al., 2022). These models offer the potential for integrated organogenesis achieving higher-order tissue complexity and functionality, paving the way for more effective therapies and research applications (Rivron et al., 2018; Sozen et al., 2018; Oldak et al., 2023). Yet, fully integrated embryo models are often too complex for high-throughput culture and accessible readouts, and they may raise ethical concerns similar to those associated with natural embryos. These challenges may be addressed by instead directing the development of specific subsets of organ anlagen while exploiting the principles of reciprocal tissue co-inductions during embryonic development (Rivera-Perez and Hadjantonakis, 2014; Christodoulou et al., 2019; Pijuan-Sala et al., 2019). Such sophisticated yet less ethically charged non-integrated embryo-like structures (Vrij et al., 2022; van den Brink et al., 2020) will become essential and powerful tools when translating this newly gained and so far, inaccessible knowledge into future bioengineering solutions. These models may render (multi-) organ development accessible for applications in drug and toxicity screening, disease modelling, developmental biology and even regenerative medicine. During mouse embryogenesis, perturbation through extrinsic factors—such as geometric confinements (Blin et al., 2018), mechanical forces (Navis and Bagnat, 2015), chemical induction (Vrij et al., 2022; Li et al., 2024), or genetic perturbation (Renzis, 2020; Hislop et al., 2024) — in controlled microenvironments, allows better access and easier study of developmental processes. The mouse yolk sac, derived from both visceral endoderm (derived from primitive endoderm) and extraembryonic mesoderm (derived from the ingression of epiblast cells) (Yamane, 2018), is essential for embryonic growth, functioning as a selective barrier for nutrient, oxygen, and waste transfer. It is the primary site of hematopoiesis, with blood cell production observed as early as E7.5 (Yamane, 2018; Ross and Boroviak, 2020; Chuva de Sousa Lopes et al., 2022). Primitive hematopoiesis is characterized by the differentiation of hematopoietic cells into primitive erythrocytes that cluster along with other cells such as hematoendothelial progenitors and macrophages, to form blood islands (Yamane, 2018). Concurrently, around embryonic day 7.0 (E7.0), cardiac precursors emerge in proximity of the yolk sac, culminating in the first heart tube by E8.5. Both cardiac precursors and hematopoietic cells originate from mesodermal tissues with the yolk sac providing the environment for their early development, encompassing the cardiovascular/hematopoietic niche (Yamane, 2018). The intricate interplay between yolk sac formation, early heart development, and primitive hematopoiesis highlights the complexities of early embryogenesis and the importance of each component in supporting further development of complex organs. In our previous work, we demonstrated the formation of a chemically-induced embryo model within hydrogel microwells, recapitulating epiblast (Epi) and extraembryonic endoderm (XEn) co-development (Vrij et al., 2022). This model, termed XEn/EpiCs, progressed to form a pro-amniotic cavity (PAC), resembling a partial E5.5 embryo. These structures develop from mouse embryonic stem (ES) cells, in the absence of trophoblast stem (TS) cells, within microwells. In this study, we illustrate the continuous developmental progression of this non-integrated XEn/EpiC model in basal serum-free medium towards an embryonic yolk sac-like (YS+) structure capable of self-organizing, giving rise to various embryonic tissues, and demonstrating restrictive diffusion of factors into the lumenal structure. A subset of YS+ structures develop clusters of cardiomyocytes (CM) exhibiting spontaneous oscillatory contractions and expressing early heart field markers. They further progress to undergo in vitro vasculogenesis, forming blood islands (Ratajska et al., 2006; Ferkowicz and Yoder, 2005), comprising blood cells, endothelial-associated cell populations. Leveraging this system, we steer developmental trajectories using known modulators of Wingless and Int-1 (Wnt) and bone morphogenetic protein (BMP) pathways to achieve targeted more efficient yolk sac tissue morphogenesis. Furthermore, the enhancement of our platform using customized milliwells not only facilitated the formation of complex multi-tissue YS+ structures but also expedited real-time, spatial visualization of development and analysis to track morphogenesis over time when perturbed externally using pathway modulators. Overall, we present a novel platform for generating and analyzing multi-tissue, functional constructs mimicking the embryonic yolk sac in parallel which recapitulates the developmental complexity of early organogenesis in the absence of trophoblast signals and serum. This study demonstrates, for the first time, a non-integrated embryo model that continuously progresses through sequential morphogenetic steps from pre-implantation lineage specification to organogenesis. Translation of embryo models to a scalable, tractable and robust platform such as micro-/milliwell systems opens the possibility of large-scale screens in pre-clinical applications and drug development. When coupled with advanced microscopy, this approach allows for in situ live tracking and manipulation of morphogenetic steps in post-implantation and organotypic development. Results Characterization of the post-implantation progression of XEn/EpiCs towards organogenesis It has been shown that a 24-h exposure of mouse naive embryonic stem (ES) cell aggregates to a combination of RA, mFgf4, Chir99021 and 8Br-cAMP, named as RF2C induction, drives the co-formation of epiblast (Epi) and extraembryonic endoderm (XEn) (Rivera-Perez and Hadjantonakis, 2014). In basal serum-free medium these structures autonomously progress towards early post-implantation-like XEn/EpiC structures, defined by a polarized Epi with pro-amniotic cavity (PAC) that is enveloped by an epithelialized XEn layer (Shankar et al., 2024). When XEn/EpiCs were cultured beyond the 5-day culture window in minimal, N2B27 medium in standard culture plates (6-well plates), we observed the formation of delaminated and expanded structures with increased complexity (Figures 1a,b). The XEn/EpiCs, devoid of trophectoderm (TE) cells and associated induction cues like BMP4, further self-organized into an expanded fluid-filled cavity surrounded by an epithelium, reminiscent of a mouse embryonic yolk sac (Figure 1a). Maintaining the culture conditions with periodic refreshment allowed the spontaneous differentiation of a group of cells in the yolk sac-like structures towards cardiac lineages leading to oscillatory contracting regions within these structures (Figure 1a) (SV1). Characterization of the different morphologies during the post-implantation progression of XEn/EpiCs towards organogenesis: (a) Protocol for the formation of XEn/EpiCs following an initial 24 h (devoid of trophectoderm) induction using soluble factors, namely, retinoic acid (RA), mFgf4, Chir99021, and 8Br-cAMP, called the RF2C factors, within microwells. The schematic shows the developmental progression of XEn/EpiCs developing into fluid-filled sacs reminiscent of a yolk sac, with or without rhythmically contracting cell clusters on day 10. The top row shows the congruous developmental timeline of the XEn/EpiCs in parallel to natural mouse embryonic development. The developmental progression was aligned to approximate corresponding stages of mouse embryogenesis (E5.5–E8.5) based on morphological and lineage characteristics. Created in https://BioRender.com. (b) The yield and classification of different morphologies of XEn/EpiCs over time from n = 3 wells each containing ∼300 structures; yolk sac negative (YS-), yolk sac positive structures with partial engulfment (partial YS+), and structures fully engulfed with the yolk sac (full YS+). (c) Graph showing the yield of only yolk sac (YS+) morphologies over time (further bifurcation of the green bar in b)) namely, yolk sac-like structures with contracting/cardiac clusters (YS+CM+) and yolk sac-like structures without cardiac clusters (YS+CM-). (d) Lineage ratios as identified through CIBERSORT analysis of bulk RNA sequencing data from YS+CM-, YS+CM+, and all structures cultured for 7 days, (e) 10 days, and (f) 14 days. (g) relative fold gene expression of germ layer markers Brachyury, (h) Otx2, and (i) Sox17 by RT-qPCR analysis on pooled structures from days 3–8. In all qPCR experiments n = 2 biological replicates (with each n being the average of two technical replicates, RNA from pooled 10+ structures). Data are mean ± SD; Ordinary two-way ANOVA with Tukey multiple comparisons test with p < 0.0001 (****), p = 0.002 (***), p = 0.0021 (**), p = 0.0332 (*), and p = 0.1234 (ns). Legend: Epi: epiblast, XEn: extraembryonic endoderm/primitive endoderm, VE: visceral endoderm, PaE: parietal endoderm, PAC: pro-amniotic cavity, PS: primitive streak, Ex. coe: exocoelomic cavity, EC: ectoplacental cone, ExE: extraembryonic ectoderm, vYS: visceral yolk sac, All: allantois, PGC: primordial germ cells, Meso: mesoderm, Ecto: ectoderm, Chor: chorion, Am: amnion, pYS: parietal yolk sac, CM: cardiac mesoderm, Som: somites. At 7 days of culture, we observed two predominant morphologies, classified as XEn/EpiCs with a partially emerging yolk sac-like structure (partial YS+, 22% of total), and those without a yolk sac-like structure (YS-) (Figure 1b). From day 9 on, 3% of structures had formed a fully engulfed yolk sac-like structure (YS+) encompassing a dense embryonic-like cell cluster (Figure 1b), which increased to 17% on day 10. From 10 days onwards, the remaining subset of YS-, that did not grow further appeared to be halted in size at the XEn/EpiCs stage. From day 10 onwards, a proportion (5%) of YS+ structures showed a cluster of cells within the embryonic-like compartment spontaneously contracting rhythmically, reminiscent of a primitive cardiomyocytes (CM) population. Quantification of YS+ structures with a CM region (YS+CM+) and those without (YS+CM-) (Figure 1c, SV1), showed an increase of YS+CM-from 5% on day 9%–12% on day 13 and YS+CM+ from 1% on day 9% to 6% on day 13 (Figure 1c). To validate the different cell types observed in these morphologies, XEn/EpiCs from day 4 (Supplementary Figure S1), 7, 10 and 14 were collected for bulk RNA sequencing and deconvoluted using CIBERSORT analysis to estimate the cell type abundances (annotated using the gene lists from the mouse gastrulation database of Marioni, et al. (van den Brink et al., 2020)). Because these analyses were performed on pooled and morphologically heterogeneous structures, the inferred cellular compositions should be interpreted as qualitative estimates of lineage representation rather than precise quantitative cell-type abundances. kThe day 4 structures were overlaid with E6.25-E6.75 embryo, day 7 structures with E7.0-E7.75 embryo, and days 10 and 14 with E8.0-8.5 embryo. While day 4 Epi and non-Epi rosettes were majorly composed of epiblast (Epi), extraembryonic endoderm, and visceral endoderm (VE), they contained signatures consistent with blood progenitor and hematoendothelial progenitor populations (or hematopoietic progenitor cells that can give rise to all blood cell lineages) when overlaid with E6.25-E6.75 embryo (Supplementary Figure S1). From the analysis of day 7 structures, it was observed that the ratio of different cell types present in the YS- and YS+ structures were similar, consisting primarily of epiblast, visceral endoderm and surface ectoderm (Figure 1d). However, interestingly, the YS+ structures also contained signatures consistent with hematoendothelial progenitor and endothelial populations according to CIBERSORT deconvolution analysis, which were absent in the YS- structures (Figure 1d), the presence of which draws similarities to the early yolk sac formation in E7.5-8.0 stage mouse embryo (Yamane, 2018). In addition, YS+ structures showed enrichment of mesenchymal and pharyngeal mesoderm-associated signatures compared with YS- structures. Pharyngeal mesoderm has been shown to play a role during cardiogenesis through direct involvement of these cells in the heart and pharyngeal muscle development (Tzahor and Evans, 2011). In contrast to the YS+ structures, YS- structures displayed enrichment of forebrain/midbrain/hindbrain and epiblast-associated signatures., which may suggest that the presence of YS in these structures influence cell fate decisions, potentially promoting mesendodermal differentiation at the expense of neural and epiblast fates. (Figure 1d). This observation resembles aspects of early in vivo yolk sac development around E7.5–E8.0, where extraembryonic mesoderm and visceral endoderm cooperatively support hematovascular specification while restricting alternative lineage trajectories. The bulk RNA sequencing of day 10 structures included four morphological classes based on the presence or absence of yolk sac and cardiac mesoderm, including YS+CM-, YS+CM+, and all structures (mixed population of all morphologies from a well) (Supplementary Figure S2) (Figure 1e). We found that the YS+CM-structures showed enrichment of visceral endoderm, mesenchymal, and nascent mesoderm-associated signatures. (Figure 1e). This is in line with the cell types commonly found in the early-stage embryonic yolk sac, which are composed of an endodermal layer derived from primitive endoderm and extraembryonic mesodermal cells from epiblast (Yamane, 2018). In line with the observed morphology, the YS+CM+ structures showed enrichment of cardiomyocyte and endothelial signatures (Figure 1e). coinciding with the emergence of cardiac cells in E8.0 stage mouse embryos (Liu et al., 2006). The ratio of cell types also revealed that the YS+CM-structures had increased representation of somitic mesoderm-associated signatures. (Figure 1e). These cell clusters expressed Tbx6 (Supplementary Figure S5), which is an early indicator of somitogenesis. The gut-associated signatures were detected across all sample groups (Nowotschin et al., 2019). Interestingly, the sample ‘all structures’, which included a combination of all the morphologies predominantly consisting of YS-CM-structures, had a higher proportion of cells belonging to forebrain/midbrain/hindbrain and gut-associated signatures compared to the other classes (Figure 1e). Additionally, this class also had a increased representation of neural crest-associated signatures compared to the other samples that had less than 3% (Figure 1e). Day 14 structures showed a similar high proportion of forebrain/midbrain/hindbrain cells (45%, Figure 1f). The increased neural specification was further supported by the quantification of relative fold gene expression of neural markers, such as Pax6, over time. In day 7 structures, Pax6 expression showed a gradual upregulation, reaching a 1000-fold increase compared to the control (Supplementary Figure S3). On both days 10 and 14, the YS+CM- and YS+CM+ structures showed different proportions of mesenchymal cells and cardiomyocytes. YS+CM-structures showed stronger mesenchymal signatures, whereas YS+CM+ structures showed stronger cardiomyocyte-associated signatures. (Figure 1f). The ‘all structures’ sample showed a significantly higher ratio of forebrain/midbrain/hindbrain (40%) in comparison to the other two classes (Figure 1f). Taken together, the deconvolution analysis suggests that YS+CM-structures are enriched for yolk sac-associated cell signatures, whereas YS+CM+ structures are enriched for cardiomyocyte-associated signatures. However, these findings should be interpreted cautiously, as the analyses were performed on pooled heterogeneous structures and therefore provide an overview of lineage representation rather than precise cell-type quantification. Specification of mesendodermal tissues in day 7 yolk sac-like structures To assess the formation of germ layers, relative fold gene expression of key markers was measured over time using qPCR on the pooled ‘all-structures’ samples. Brachyury, a mesoderm marker, displayed an increase starting from day 5, peaking around day 8, indicating the initiation of primitive streak formation and mesoderm specification in the XEn/EpiCs (Figure 1g). Sox17, an endoderm marker, remained consistently elevated (up to 100-fold) from day 3 to day 8, reflecting persistent endoderm differentiation (Figure 1i). Otx2, an early neuroectoderm marker, showed an apparent increase in expression starting on day 3, peaking on day 6 (near 1000-fold), before slightly decreasing, suggesting a transient neuroectoderm specification (Figure 1h). These results were corroborated by immunostainings for Brachyury, Sox2 and Sox17 in day 7 YS+ structures (Supplementary Figure S3), where 38%, 54%, and 34% of structures were Brachyury+, Sox2+, and Sox17+, respectively (Supplementary Figure S4). However, only a small percentage of structures were positive for all three germ layers at the same time. This, together with the bulk sequencing data where structures without YS and/or CM showed a pronounced neural/ectodermal identity, suggests that most structures specify predominantly either both mesoderm and endoderm, or neuro-ectoderm. In conclusion, these results suggest that prolonged cultures of XEn/EpiCs may permit germ layer specification and (extra)embryonic organogenesis. Yolk sac-like structures with contracting cardiomyocyte-like clusters To further characterize the cell types present in the contracting structures (YS+CM+) (SV1, 2), we performed single cell RNA sequencing on day 10 YS+CM+ structures (four pooled structures, sorted for Gata6:mVenus) with UMAP clustering and annotated the 17 different clusters based on E8.0-8.5 mouse embryos derived from the Marioni database (van den Brink et al., 2020) of (Figure 2a). As the analysis was performed on pooled structures, inter-structure variability could not be assessed and cluster annotations should therefore be interpreted with caution. From the clustering and annotations, we observed that the biggest proportion of cells belonged to mesenchyme, endothelium, cardiomyocytes, and visceral endoderm/extraembryonic endoderm/gut (Figure 2a), which coincided with findings from the bulk sequencing analysis (Figure 1e). Expectedly, the proportion of the cells expressing makers for amniotic ectoderm, surface ectoderm and visceral endoderm, corresponded to the endodermal reporter expressing Gata6-mVenus+ (red) cells (Figure 2b). The Gata6-mVenus+ cells also contributed to the endothelium and mesenchyme and cardiomyocytes, which are of mesendodermal origin. Meanwhile, the Gata6-mVenus- (blue) cells appeared to majorly express markers for paraxial mesoderm, pharyngeal mesoderm, caudal neuroectoderm and cardiomyocytes (Figure 2b). Formation and characterization of YS+CM+ structures from XEn/EpiCs: (a) Single cell RNA sequencing on day 10 structures. (b) UMAP plot showing the different clusters based on sample plate containing Gata6:mVenus positive cells (red) and Gata6:mVenus negative cells (blue). (Right) UMAP plot showing the different cell types predominant in these clusters annotated based on genes they expressed. (c) Heat map of the top 10 differentially expressed genes of each cell type plotted with the yellow being the highest expression of that gene and the size of the column represents the number of cells expressing those genes. (d) Ultrastructure of the contracting region in YS+CM+ structures; N: nucleus, red arrow: muscle fibers showing intercalated discs inside the cytoplasm of cells, yellow arrow: large group of mitochondria, blue arrow: mass of glycogen. (e) qPCR analysis of the relative fold gene expression of early cardiac and cardiac-associated markers, namely, Tnnt2, Nkx2, Hcn4, and Isl1 on structures from day 7 to day 14 normalized to day 0 control. (f) Immunofluorescence staining for early cardiac markers CTNT and Mef2ac in paraffin-embedded tissue sections of in contracting structures. (f) Protocol for the administration of adrenaline to contracting structures. (Right) contraction frequencies of YS+CM+ structures before (baseline) and after adrenaline treatment measured using ImageJ. (g) Graph showing the contraction frequencies before and after adrenaline treatment. (h) Graph showing the contraction frequencies of different YS+CM+ structures before and after treatment. Each dot represents one YS+CM+ structure. Scale bar: 100 µm. Data are mean ± SD. n = 2 biological replicates (with each n being the average of two technical replicates, RNA from pooled 10+ structures). Statistical analysis in the qPCR experiments were performed using ordinary two-way ANOVA with Tukey multiple comparisons test with p < 0.0001 (****), p = 0.002 (***), p = 0.0021 (**), p = 0.0332 (*), and p = 0.1234 (ns). The above findings show that the in vitro contracting structures are mesendoderm dominant compared to tissues present in the E7.0-E7.75 stage mouse embryos that are also composed of anterior lineage neural tissues, primordial germ cells, neuro-mesodermal progenitors, and spinal cord. Of note, these structures are developed in the absence of trophoblast cells and thereby do not form the extraembryonic ectoderm, ectoplacental cone, Reichert’s membrane and parietal endoderm. The hematopoietic/cardiac cluster expressed endothelial markers such as Cd93, Gpihbp1 (capillary endothelial cells), Robo4 (endothelial-specific transmembrane receptor), and Arhgef15 (an early blood vessel and vasculogenesis marker), mesenchymal markers such as Col613, Postn, and Sp5, and cardiomyocyte markers such as Csrp3 (skeletal muscle marker), Hspb7 (cardioprotective marker), Tnni3, and Myl17 (Figure 2c). The commonly observed markers of extraembryonic endoderm (Apob, Serpina10), gut/definitive endoderm (Serpinb5, Lapk10, Esr1, Pax1), caudal neuroectoderm (Pax7, Myod1, Cdh4, Cdh15) were also expressed in the respective clusters. However, there was an overlap of some genes, such as Kif23, Cenpf, and Fam83d, in closely identified clusters such as pharyngeal mesoderm and visceral endoderm (Figure 2c) (Supplementary Figure S9). The ultrastructure YS+CM+ structures were observed using transmission electron microscopy (TEM), which identified a large group of mitochondria and muscle fiber in the cardiac region (Figure 2d). Interestingly, large masses of glycogen were observed in the sections indicating an active energy-driven contraction in these regions. Immunohistochemistry was performed for cardiac and early heart field markers on the sections of day 10 YS+CM+ structures to observe the localization of these proteins (Figure 2d). It was found that the expression of cardiac troponin or CTNT (from the gene Tnnt2), which confers calcium sensitivity to the striated muscle, was localized to the contracting region prior to fixation (Figure 2f), similarly for the cardiac skeletal muscle marker Mef2ac. To identify the onset of cardiac lineage specification and marker expression in the YS+CM+, qPCR was performed on the structures from day 7 to day 14 (Figure 2f). They were tested for the expression of the commonly used cardiac markers such as Tnnt2, Nkx2 (an early stage cardiac marker expressed in the mesodermal cells specified to form the cardiac tissues) (Figure 2e), Hcn4 (cardiac pacemaker marker), and Isl1 (cardiac progenitor marker) (Supplementary Figure S5a). From the relative fold gene expression (log10) in comparison to day 0 cells, all the cardiac genes were upregulated, with the expression of Tnnt2 significantly increasing to 1000-fold from day 10 to day 14 (Figure 2e). Together, the characterization of YS+CM+ structures provided spatial and temporal information on the cardiac specification of the mesodermal cells and the onset of contraction. Finally, to test whether the YS+CM+ structures with the actively contracting population of cells respond functionally similar to cardiomyocytes, they were exposed to 20 μM DL-adrenaline for 1 h, a hormone that causes an increase in the contraction rate in cardiomyocytes (Figure 2g) (SV3,4). The YS+CM+ structures were imaged prior to the treatment, and the contraction frequencies were measured and plotted using ImageJ to obtain the Z-axis profile (Figures 2g,h). It was observed that upon exposure to adrenaline, the contraction frequencies of the structures increased significantly in comparison to the baseline control (SV4) (Figures 2g,h) This variability likely reflects differences in developmental maturity, structural organization, and baseline contraction frequencies among individual structures. (Figure 2g). In conclusion, XEn/EpiCs can autonomously progress to form yolk sac-like structures encompassing functional cardiac clusters (YS+CM+) that respond to adrenaline. Vascularized yolk sacs The structures exhibiting a fully engulfed yolk sac-like morphology, with or without an encompassing cardiac niche, collectively referred to as YS+ structures, expanded further around day 10. These YS+ structures began to form capillary-like strands around their periphery, progressively developing into a network (SV1) (Figure 3a). Under bright-field microscopy, these vascular structures appeared as high-contrast and extended from the outer yolk sac to the inner layers, reminiscent of the yolk sac capillary plexus (Ross and Boroviak, 2020) (Figure 3a) (SV1). Immunohistochemistry revealed that the hollow YS+ structures had an Apoa4 positive epithelial outer layer, a key yolk sac marker (Figure 3b). Ultrastructure analysis using TEM highlighted the YS epithelial layer including presence of cytoplasmic vesicles, vacuoles, and the microvilli protruding from the YS epithelium, similar to what is observed in natural mouse yolk sacs (Hsu et al., 1974; Carter, 2016) (Figure 3c). YS+ vascularization markers: (a) Brightfield image of a day 10 yolk sac-like (YS+) structure undergoing vasculogenesis. (b) Immunofluorescence staining for yolk sac marker Apoa4 in paraffin-embedded tissue sections of YS+ structures. Inset: zoomed in image of the two boxed regions. (c) Ultrastructure of yolk sac epithelium using TEM. (Left-top) inset showing the structure before sectioning for TEM. (left bottom) inset showing the image of the epithelial layer showing the thin layer of membrane containing typical storage sacks (asterisks). (Right) zoom in on the yolk sac epithelium showing the microvilli projecting from the side of the membrane similar to the natural mouse yolk sac. (d) qPCR analysis of YS+ structures from day 7 to day 14 showing the fold gene expression (log 10) normalized to day 0 control for yolk sac and vascularization genes, Apoa1 and VEGFa, respectively. (e) Collapsed fluorescence image stack of wholemount immunostained YS+ structures showing the network of CD31+ cells. (f) VEGFa staining showing vasculogenesis in YS+ structures. (g) Dextran diffusion assay to observe membrane integrity using FITC-labeled Dextran in integer YS+ structures and in (h) EDTA-pre-treated YS+ structures. Scale bar: 100 µm. Data are mean ± SD; n = 2 biological replicates (with each n being the average of two technical replicates, RNA from pooled 10+ structures). Statistical analysis in the qPCR experiments were performed using ordinary two-way ANOVA with Tukey multiple comparisons test with p < 0.0001 (****), p = 0.002 (***), p = 0.0021 (**), p = 0.0332 (*), and p = 0.1234 (ns). RNA analysis confirmed that the YS+ structures contain endothelial cells, mesenchyme and blood progenitors (Figure 2b). qPCR analysis showed an upregulation of genes involved in yolk sac differentiation (Apoa4 and Lrp2) and vasculogenesis (VEGFa and CD31). Apoa4 was found upregulated from day 7 (50,000-fold) until day 14 (100,000-fold) when normalized to the day 0 control, coinciding with the occurrence of a bloated, fluid-filled YS+ cavities at day 7 (Figure 3d). VEGFa expression progressively increased over time, reaching an 8-fold rise by day 9, which was followed by a decrease on days 11 and 12, before rising sharply again, achieving a 30-fold increase by day 13 (Figure 3d). To visualize vasculogenesis, YS+ structures were stained for CD31 and VEGFa. Wholemount staining revealed CD31+ endothelial cells enveloping the entire structure (Figure 3e). Similarly, VEGFa+ capillaries were observed throughout the YS+ structure (Figure 3f). The yolk sac membrane acts as a selectively permeable barrier that allows diffusion of certain substances from the maternal environment into the yolk sac cavity to nourish the embryo. To assess the barrier function of YS+ structures, we performed a diffusion assay by exposing them to FITC-labeled Dextran for 1.5 h. After washing and imaging, it was observed that the Dextran did not penetrate the yolk sac epithelium into the YS-like lumen (Figure 3g), indicating the integrity of a tight epithelial barrier similar to the in vivo yolk sac epithelium. Since the yolk sac epithelial barrier is composed of calcium-dependent cell-cell junctions, we next exposed the YS+ structures to EDTA to disrupt membrane integrity (Figure 3h). Indeed, EDTA-treated YS+ structures were rendered permeable for FITC-labeled Dextran as it easily passed through the membrane and filled the fluid-filled sac (Figure 3h). These findings provide preliminary evidence that the in vitro YS+ structures form a functional epithelial barrier with selective permeability properties that resemble the in vivo yolk sac membrane. The barrier function depends on intact calcium-dependent cell-cell junctions, consistent with the properties of epithelial tight junctions. Overall, the results support the interpretation that YS+ structures possess barrier-like properties as a physiologically relevant in vitro model of the yolk sac epithelium. The assay was performed on representative structures and future studies will be required to quantify variability across structures. Yolk sac XEn/EpiCs as a site for primitive hematopoiesis and blood island formation Blood island clusters have been recognized as the primary source of erythropoiesis in the mouse embryo (Yamane, 2018; Ferkowicz and Yoder, 2005). Blood islands arise from extraembryonic mesoderm layer of the yolk sac as early as E7.0-E7.5 (Ferkowicz and Yoder, 2005; Yamane, 2018). Hematopoietic cells aggregate into blood islands, which consist of clusters of primitive erythroid cells, associated with endothelial and other supporting cell populations (Canu and Ruhrberg, 2021). We found that the prolonged culture of YS+ structures from day 10 to day 14 led to the spontaneous induction of a specific cluster of cells that are reminiscent of blood islands (Figure 4a) (SV1). Cells within these blood islands showed red domains in their apparent center regions when viewed under a bright-field microscope (Figure 4b), indicative of hemoglobin. Transmission electron microscopy (TEM) of the blood island regions revealed a high density of cells resembling primitive erythrocytes, erythroblast-like cells, pyrenocytes (nuclei expelled from erythrocytes), and reticulocytes. (Figure 4c). Immunofluorescence staining for RunX1, a common maker for blood island formation, revealed positive staining along the yolk sac-like lining in YS+ structures (Figure 4d), suggesting the onset of blood formation. qPCR validation revealed a 100-fold upregulation of Runx1 from day 7 to day 9 (Figure 4e), consistent with its expression in the mesodermal mass of the yolk sac, which gives rise to primitive erythrocytes and endothelial cells (Yzaguirre et al., 2017). Runx1 expression in hematoendothelial progenitors was further confirmed in day 7 YS+ structures via bulk RNA sequencing (Figure 1d). Similarly, other hematopoietic markers, Flk1 (hemato-cardiovascular marker) and Hoxb4 (hemogenic endothelium formation marker) showed a 100-fold increase from day 7 to day 14 (Figure 4e; Supplementary Figure S5). Interestingly, the expression of Hoxb4 showed an alternating expression profile with a period of 2 days (Supplementary Figure S6). Characterization of day 14 blood island structures: (a) Schematic representation of the timeline for blood island formation in YS-like structures; Created in https://BioRender.com. (b) bright-field images of the blood island structures showing a bright red region (highlighted with yellow circles). (c) TEM on sections of day 14 YS+ blood island structures; The red arrows indicate the primitive erythrocytes, the violet arrows indicate cells with macrophage-like morphology, and green arrow indicates a pyrenocyte (enucleated erythrocytes). (d) Immunostaining day 14 structures for Runx1. Scale bar (when not specified): 100 μm. (e) qPCR analysis of the relative fold gene expression (Log 10) of primitive hematopoiesis and blood formation markers, namely, Runx1 and Flk1 from day 7 to day 14. Data are mean ± SD; n = 2 biological replicates (with each n being the average of two technical replicates, RNA from pooled 10+ structures); ordinary two-way ANOVA with Tukey multiple comparisons test with p < 0.0001 (****), p = 0.002 (***), p = 0.0021 (**), p = 0.0332 (*), and p = 0.1234 (ns). (f) Flow cytometry analysis for mouse blood progenitor markers found in YS+ blood island structures (n = 5 structures pooled). Live cells were divided into either Gata6:mVenus + (V+) or Gata6:mVenus - (V-) and gated for Lin (lineage) signal to identify the undifferentiated blood cell population. The Lin-cells were further gated for Sca-1 and c-Kit expression. The Lin-c- Kit+ population (LK cells) were finally gated into Erythroid-mega karyocyte progenitor (EMP), common myeloid progenitor (CMP) and granulocyte monocyte progenitor (GMP). (g) Lin+ vs. CD45+ cells from the V- population. (h) Legend and ratios of different gates and cell population. Characterization of primitive erythropoiesis in YS+ structures To identify the different cell types observed in the blood islands, single cell-dissociated YS+ structures were stained for a panel of blood progenitor markers (Table 1) and analyzed using flow cytometry. A total of 58,349 (live, singlet) cells were analyzed through flow cytometry (Figure 4h). The cells had a Gata6: mVenus reporter, which were initially gated to separate the mVenus+ (V+) and mVenus-- (V-) cells, with 86.5% of cells being V+. This could be due to the large cell population of yolk sac-origin in these structures. The rationale behind separating the Gata6 positive and negative population, which identified the endodermal cell type, was to visualize the contribution of different cell lineages towards yolk sac blood emergence. We first determined the cells lacking mature blood cell lineage markers (Lin-) to identify the immature, undifferentiated hematopoietic cells. These cells were then classified according to their Sca-1 (early hematopoietic marker) and c-Kit (hematopoietic stem cell self-renewal marker) expression, into three groups, namely, LSK (Lin− Sca-1+ c-Kit+), LK (Lin− Sca-1- c-Kit +), and LS (Lin− Sca-1+ c-Kit-) (Figure 4f). The LK cells were further gated to isolate the hematopoietic population expressing CD16/32 (marker for B cells, monocytes/macrophage like cells, NK cells, granulocytes, mast cells, and dendritic cells) and CD34 (marker for hematopoietic stem cell) expression. TABLE 1Primers and their sequences.PrimersSequenceMD_Brachyury_F_1CAT​GCT​GCC​TGT​GAG​TCA​TAAMD_Brachyury_R_1TGT​CTG​GGA​GCC​TGG​GGT​GATPV_VEGFa_F_1TTG​TTC​AGA​GCG​GAG​AAA​GCPV_VEGFa_R_1GAG​AGG​TCT​GGT​TCC​CGA​AAPV_CD31_F_1GAG​CCC​AAT​CAC​GTT​TCA​GTTPV_CD31_R_1TCC​TTC​CTG​CTT​CTT​GCT​AGCEnD_Gata6_F_1TTT​TTA​TTC​ACC​AGC​AGC​GAC​TAEnD_Gata6_R_1GCGCTACTCCAACCTGACEnD_Sox17_F_1CTC​AGC​ATG​TCA​CCT​CAT​GGAEnD_Sox17_R_1CTT​GGG​GAA​AAC​TGG​CTG​GAEcD_Pax6_F_1GAC​GGC​ATG​TAT​GAT​AAA​CTA​AGG​AEcD_Pax6_R_1GAG​TTG​GTG​TTC​TCT​CCC​CCEcD_Otx2_F_1AAC​TTG​CCA​GAA​TCC​AGG​GTEcD_Otx2_R_1TGA​ACT​CAC​TTC​CCG​AGC​TGS_Tbx6_F_1GGA​GAA​CCA​GGA​ACT​GTG​GAAS_Tbx6_R_1TCC​AGG​CCA​GTG​ACT​GAT​ACTPV_Vegfa_F_2TTG​TTC​AGA​GCG​GAG​AAA​GCPV_Vegfa_R_2CACCGCCTTGGCTTGTCAPV_Flk1_F_1GTG​CAG​GAT​GGA​GAG​CAA​GGPV_Flk1_R_1TTC​TGT​GTG​CTG​AGC​TTG​GGVYS_Apoa4_F_1AAG​ACG​GAT​GTC​ACT​CAG​CAVYS_Apoa4_R_1CAG​CTG​TAC​GAC​AAA​GGG​CAVYS_Lrp2_F_2ACC​TTC​TCT​ACT​GGG​CGG​ATVYS_Lrp2_R_2AGC​CAA​AGG​AAT​GGA​AGG​CAPYS_Hhex_F_1ACG​GTG​AAC​GAC​TAC​ACG​CAPYS_Hhex_R_1CTT​TCC​TTT​TGT​GCA​GAG​GTC​GHDM_Tnnt2_F_1CCT​GAT​GAA​GAA​GCC​AAA​GAT​GCHDM_Tnnt2_R_1CCA​CTC​TCT​CTC​CAT​CGG​GGHDM_Mef2c_F_1GGA​GAT​CTG​ACA​TCC​GGT​GCHDM_Mef2c_R_1GGG​GAG​GAG​ATT​TGG​CTT​GTHDM_Nkx2-5_F_1GAG​CCT​ACG​GTG​ACC​CTG​ACHDM_Nkx2-5_R_1TCCGTCTCGGCTTTGTCCHDM_Hcn4_F_1TGC​CCT​CAT​CCA​GTC​GCT​AHDM_Hcn4_R_1CTT​GCC​TTG​GTA​GCG​GTG​TTHDM_Mesp1_F_1CGCCTGCCTACCCTAGAHDM_Mesp1_R_1CTA​GAA​GAG​CCA​GCA​TGT​CGHDM_Isl1_F_2AGT​CCC​AGA​GTC​ATC​CGA​GTHDM_Isl1_R_2GGA​GTT​CCT​GTC​ATC​CCC​TG Primers and their sequences. In the case of the V- population, it was observed that the LK cluster (n = 386 = 100%) was composed of different cell types associated with blood progenitors, with the majority of the cells falling into EMP cluster (erythrocyte-megakaryocyte progenitors) (67.1%) (Figure 4f, h). A few cells belonged to the common myeloid progenitors (CMP) (7%) and granulocyte-monocyte progenitors (GMP) (11.4%) (Figure 4f, h). Interestingly, 10.9% of cells were CD34+, which marks the pre-endothelial and hematopoietic progenitor populations (Yzaguirre et al., 2017). Meanwhile, the V+ population of LK cells (n = 1998 = 100%) showed around 95.6% in the EMP gate while CMP and GMP clusters were present below 2%. The cells found in YS+CM+ and YS+CM-structures were also compared specifically to YS+ blood island structures using flow cytometry. YS+CM-showed a significantly lower number of LK cells, with cells in the EMP gate for both V- (n = 62 cells) and V+ (n = 1,574 cells) and negligent cell numbers in GMP and CMP gates (Supplementary Figure S8). The trend was similar in YS+CM+ structures. A larger population of CD34+ cells was found in YS+CM+ structures in comparison to YS+CM- and YS+ blood island structures. This correlates with the presence of a cardiac niche, where there is an increased incidence of endothelial and vascular cells expressing CD34, including the para-aortic mesenchyme (Yzaguirre et al., 2017). Isolation of Lin+ cells from the three samples revealed a distinct cluster of CD45+ cells, a lymphocyte marker, in all samples. Among these, the YS+ blood island structures contained the highest number of CD45+ cells, with a count of 104. (Figure 4g). This could indicate the onset of hematopoietic progenitor populations proliferation in the YS+ structures, consistent with the hematopoietic progression of in vivo YS hematopoiesis, which is characterized by the presence of EMP and lymphoid progenitors around E8.25 38. Although there needs to be more detailed characterization with extensive and validated marker panels to identify the observed cell types conclusively in our defined but basic conditions, our results underline the presence of the key cell types commonly observed in blood islands and suggest consistency with known cellular signatures. Further work will focus on the development of more specific directed differentiation protocols to better coax cell populations in our constructs towards certain lineages. Together, these results validate the formation of blood island clusters on day 14 YS+ structures composed of primitive erythrocytes, primitive macrophages and endothelial cells. Modulation and tracking of in vitro organogenesis using XEn/EpiCs in a customized milliwell platform We have demonstrated that XEn/EpiCs spontaneously differentiate into YS+ structures, of which a subset of them later developed a cardiac niche inside (YS+CM+) and further displayed vascularization and blood island formation. Studies have shown that extrinsic factors can be used to manipulate differentiation processes during development leading to outcomes diverging from canonical trajectories (Rivera-Perez and Hadjantonakis, 2014; Shankar et al., 2024; Wood et al., 1997). During early mouse embryogenesis, Wnt and BMP pathways play crucial and interconnected roles in regulating different morphogenetic events. BMP signaling, especially through BMPR1A, is particularly important for the formation of extraembryonic lineages, prevents pre-mature neural differentiation and establishes the left-right asymmetry (Girgin et al., 2021; Arkell et al., 2013). Wnt agonists are primarily expressed in the posterior tissues of mesendoderm origin, later allowing the development of somites (Martin and Kimelman, 2009). Wnt antagonists have been shown to induce anterior tissues, namely, the neural development (Girgin et al., 2021). BMP signals are released by the extraembryonic ectoderm to induce anterior-posterior patterning in the epiblast (Sozen et al., 2021; Di-Gregorio et al., 2007). To induce these signals in the XEn/EpiC development, the structures were exposed to Wnt and BMP pathway modulators from 96 to 120 h, which is the time of specification of post-implantation tissues in the XEn/EpiCs, undergoing processes such as Epi polarization, XEn epithelialization, lumenogenesis, and basement membrane deposition. This is the stage when the in vivo counterparts secrete BMP, Wnt and Nodal pathway modulators to guide development of complex tissues (Figure 5a). We aimed to guide the development of YS+ structures, preferentially enriching them towards the specification of a fully vascularized multi-tissue structure with cardiac, blood island and/or neural clusters. We found that a 24 h pulse from 96 to 120 h of culture with BMP4, a BMP pathway activator, increased the proportion of empty YS+CM-structures to 25% on day 11 compared to 13% in control (Figure 5b). A pulse of Wnt/b-catenin activator Chir resulted in more YS+CM+ structures within the day 9–11 window with 20% compared to12% in controls (Figure 5b). We found that Chir-treated YS+CM+ structures progressed to form YS+ structures with a periodically contracting cardiac niche while expanding to near 2 mm in size. Meanwhile, 96–120 h exposure to the PARP/Wnt antagonist, XAV939, led to a relatively higher number of vascularized YS+ structures, with a visible vascular-like network, of 6% compared to 2% in control (Figure 5b). These results demonstrate that developmental outcomes can be guided by targeted by timely modulation of key developmental pathways through external soluble molecules. In situ modulation and tracking of in vitro organogenesis using XEn/EpiCs in a customized milliwell platform: (a) Protocol for the treatment of signaling compounds to XEn/EpiCs as a 24 h pulse between 96 h and 120 h and the continued culture of these structures until day 10 forming YS+CM+, YS+CM- and vascularized YS+ structures; Created in https://BioRender.com. (b) Yield of the different morphologies observed on days 9, 10, and 11, namely, YS+CM+ (blue), YS+CM- (red), and vascularized YS+ (green) structures; n = 4 replicates (c) Continuous culture of XEn/EpiCs within 800 µm thermoformed milliwells. Brightfield images showing development in milliwells over time; (d) Stencil of the milliwells for (back)tracking the structures; (e) Tracking of YS+CM+ structures from day 1 to day 9, with contraction initiation from day 8. (f) Different morphological structures, namely, YS+CM+, YS+CM-, vascularized YS+, and YS+ blood island structures, formed within milliwells. (g) Efficiency of formation of YS+CM+ structures in flushed-out 6-well plate versus milliwell culture shows the earlier formation and higher proportion of YS+CM+ structures within milliwells. Data are mean ± SD; in all yield measurements, n = 3 (biological replicates). To enable live tracking of the entire development of XEn/EpiCs in vitro and to enable screening and chemical/genetic manipulation experiments, we aimed to create a robust set-up to culture the entire development of these structures within our milliwell culture platforms. For this purpose, we fabricated thermoformed arrays with each containing 63 milliwells of 800 µm diameter. From cell seeding on the development of XEn/EpiCs structures was tracked Figure 5c) using a numbered grid stencil for the milliwell array (Figure 5d). We found that XEn/Epics formed efficiently at 120 h (90% efficiency) and later progressed to form complex tissues resembling YS+CM+ and YS+CM-structures (Figure 5e) (SV5). Further observation of these structures revealed the formation of vascularized YS+ structures as well as YS+ blood island structures at day 11 (Figure 5e). Quantification of contracting cluster specification over time showed that the continuous culture in thermoformed milliwells initiated the formation of YS+CM+ as early as day 7, significantly earlier than in free-floating suspension culture, which began after the transfer from hydrogel microwells to a 6-well plate at 72 h (Figure 5f). The former also formed a higher ratio of YS+CM+ structures, with an efficiency of 20% in milliwells compared to 7% in the 6-well plate (Figure 5g). This difference may be attributed to the spatial confinement and reduced diffusion or dilution of secreted factors, such as cytokines, which likely provide cues for lineage specification. However, vascularized YS+ and YS+ blood island structures occurred less frequently in milliwells, which may be due to geometric confinement of milliwells, potentially restricting the yolk sac expansion. Future directions could involve modulating signaling pathways within a milliwell setup while simultaneously tracking development and increasing the size of milliwells to promote the formation of more advanced embryo-like structures. Together, we show the development of a novel, scalable milliwell-based system capable of tracking and monitoring the development of embryo-like structures as they progress through various morphogenetic stages up to organogenesis. Discussion We developed a chemically induced XEn/EpiC model in micro-/milliwells, which progresses without a trophoblast compartment from a post-implantation morphology to yolk sac (YS+) formation and early organogenesis in a serum-free medium. Upon transfer to suspension culture, these non-integrated structures mimicked in vivo embryogenesis by specifying mesoderm, undergoing vasculogenesis, forming a contracting cardiac niche, and generating blood islands with primitive erythrocytes, cells exhibiting macrophage-like ultrastructural features, and endothelial cells. Other than the YS+ structures, the remainder of phenotypes resembled XEn/EpiCs that expressed markers of neuroectoderm fate (Figures 1e, f; Supplementary Figure S4c), but appeared halted in further specification. Structures without a yolk sac-like cavity predominantly adopted anterior epiblast fates, forming neural cell types, which may result from spontaneous Epi differentiation in the absence of extrinsic cues. This has also been observed in conventional PSC cultures, where cells spontaneously differentiated into neural fate after blocking BMP and Wnt pathways (Sozen et al., 2021). Future work will focus on characterizing this anterior fate and the associated neural morphogenesis in XEn/EpiCs, exploring its co-development with the yolk sac lineage, and directing these processes through chemical modulation, such as Wnt and/or Nodal inhibition. Of note, the 17%–25% of gut cells across all phenotypes from day 10 may result from the transcriptional similarity and cell fate plasticity between visceral endoderm, definitive endoderm, and gut endoderm which co-develop with yolk sac tissue around E8.0 of mouse embryo development (Nowotschin et al., 2019). Importantly, while the yolk sac-like compartment developed robustly and reproducibly, the embryonic/epiblast-like compartment appeared less spatially coordinated and displayed higher heterogeneity between structures. This is likely linked to the intentionally non-integrated nature of the XEn/EpiC model, which lacks trophoblast-derived signaling centers and associated extraembryonic ectodermal tissues that normally contribute to embryonic-extraembryonic coordination and axis patterning during natural embryogenesis. Consequently, the current system more efficiently supports yolk sac-associated mesendodermal morphogenesis than synchronized embryonic patterning. Future studies integrating controlled morphogen delivery, engineered signaling gradients, or additional extraembryonic compartments may further improve coordinated co-development between embryonic and extraembryonic lineages. Our study demonstrates that the formation of the multi-tissue structures from XEn/EpiC model primarily relies on self-organization following initial lineage bifurcation and sorting of XEn and Epi cells, requiring minimal external signaling inputs. While this approach introduces some variability, it offers a robust foundation for more directed strategies. By leveraging our microengineered culture platforms, which enable precise spatiotemporal control over external cues, we can improve the consistency of in vitro morphogenesis, resulting in more uniform multi-tissue structures. We implemented this strategy by administering a BMP or Wnt pathway modulator pulse from 96–120 h, coinciding with Epi polarization, XEn epithelialization, and PAC formation. These pathways, tightly regulated in space and time, are key to anterior-posterior patterning and symmetry breaking through morphogen gradient establishment (Di-Gregorio et al., 2007; Miyagi et al., 2012). We observed that BMP4 treatment increased the formation of YS+ structures from day 9 onwards, with a higher proportion of empty yolk sac-like structures. Notably, BMP4 treatment yielded a higher proportion of YS+CM-structures compared to the control (Figure 5b). During natural embryonic development, BMP4 secretion by the extraembryonic ectoderm around E6.5 facilitates symmetry breaking and germ layer formation (Moody et al., 2013). Exposing XEn/EpiCs to BMP4 at a similar developmental timeline in vitro promoted a preferential transition towards yolk sac specification and hematopoiesis (Figure 5b). Interestingly, exposure to Wnt pathway modulators from 96–120 h yielded distinct outcomes. Chir99021 (a small molecule Wnt agonist) significantly enhanced the development of YS+ structures, particularly increasing the efficiency of YS+ structures encasing a periodically contracting cardiac niche formation over time. Wnt activation was associated with increased frequency of structures displaying mesendoderm-associated morphologies Conversely, XAV (a Wnt antagonist) treatment was associated with an increased occurrence of vascularized structures (Figure 5b). These observations underscore the versatility of our system in enabling guided morphogenesis while promoting the development of biologically relevant tissue types derived from non-integrated embryo models. Further studies should be performed to evaluate the roles of different combinatorial pathway modulation towards specific cell fates during yolk sac hematopoiesis. To contextualize our findings, we compared our model to recent human yolk sac studies that have explored early hematopoietic emergence and extraembryonic development. Recent studies have established several human pluripotent stem cell–derived models to recapitulate yolk sac development, each contributing distinct insights into early hematopoiesis and extraembryonic morphogenesis. Notably, Atkins et al. (2022) and Tamaoki et al. (2023) generated yolk sac–like organoids by inducing mesodermal progenitors or through stromal co-culture, respectively. These models exhibit erythroid and myeloid hematopoietic outputs and represent scalable systems for studying primitive hematopoiesis. More integrated models have since emerged. Hislop et al. (2024) developed a post-implantation human embryoid platform capable of self-organizing extraembryonic and mesodermal lineages, forming yolk sac-like cavities and supporting multilineage hematopoiesis, including endothelial, erythroid, myeloid, and lymphoid cells. Similarly, the heX-embryoid model by Weatherbee et al. (2023) demonstrates yolk sac morphogenesis and hematopoietic blood island formation in the absence of trophoblast, further emphasizing the developmental potential of non-integrated systems. Although our transcriptomic analyses were benchmarked against published mouse gastrulation and embryonic reference datasets, future studies incorporating direct side-by-side comparisons with in vivo mouse embryos, including comparative spatial immunostaining and lineage distribution analyses, will further strengthen biological benchmarking and validation of the system. Compared to these human models, our mouse XEn/EpiC system uniquely combines chemically induced lineage specification, microengineered spatial organization, and serum- and trophoblast-free conditions to drive yolk sac morphogenesis and cardiogenesis via spontaneous self-organization. Unlike most human models that require co-culture with stromal cells, matrix support, or complex cytokine cocktails, our platform leverages intrinsic developmental programs and minimal exogenous signals to reproducibly generate pre-vascularized yolk sac-like vesicles, contracting cardiac niches, and blood islands. While human models are essential for capturing species-specific aspects of hematopoietic emergence and are better suited for translational applications, our system offers key advantages in mechanistic accessibility, spatiotemporal control, and scalability for high-content developmental analysis. Together, these approaches are complementary: human models provide clinical relevance, while bioengineered mouse platforms like XEn/EpiCs offer precise control to interrogate fundamental principles of extraembryonic and mesodermal morphogenesis. Guiding morphogenesis towards specific cell fates necessitates real-time monitoring of development to comprehensively understand cellular trajectories. Our previous method of generating structures in agarose hydrogels and transferring them to suspension culture in 6-well plates resulted in random formation of YS+ and YS+CM+ structures. To address this, we transitioned the entire XEn/EpiC developmental process to custom thermoformed milliwell arrays (Figure 5c), enabling controlled spatio-temporal formation of structures and in situ tracking of rudimentary organ formation. By tracing the developmental trajectory of YS+CM+ structures, we could retrospectively identify the specific morphological characteristics of XEn/EpiCs that led to their formation (Figure 5e,f). The continuous culture of YS+ structures from day 0 to day 14, progressing through various developmental checkpoints would enable high-throughput screening, for example, for assessing developmental and reproductive toxicity. In conclusion, the non-integrated (no trophoblast cells) XEn/EpiC model shows spontaneous multi-tissue development of extraembryonic endoderm and mesoderm type, yielding the formation of pre-vascularized yolk sac-like vesicles and cardiogenesis in the absence of trophoblast signals or serum. These structures exhibit remarkable plasticity and can be further guided towards diverse cell fates through precisely timed external cues. To address gaps in biologically-driven systems, we propose that bioengineered platforms will be essential for better organizing and guiding cellular processes by providing extrinsic coordinate systems that help synchronize morphogenetic events. In line with this reasoning, we developed a bioengineered platform that efficiently guided our non-integrated embryo model toward increased yolk sac hematopoiesis, showcasing the potential of combining biological insights with engineering principles to advance control of complex developmental processes. This versatile platform offers exciting possibilities for in-depth developmental studies and efficient drug screening, potentially transforming our understanding of early embryonic development and related disorders. Applications of multi-tissue structures formed from XEn/EpiCs The XEn/EpiC model offers several advantages, including a streamlined experimental procedure that requires no transgene-directed differentiation, utilizes a single starting cell type in defined culture conditions, and exhibits progressive growth with minimal intervention. Unlike models that focus solely on expanding the embryonic compartment, this approach incorporates XEn cells, which serve as crucial signaling centers for epiblast patterning. This integration significantly enhances the organizational complexity of the resulting structures by providing extraembryonic cues that mirror those present in normal development. Consequently, the XEn/EpiC model more faithfully recapitulates the intricate interplay between embryonic and extraembryonic tissues, offering a more comprehensive platform for studying early developmental processes (Vrij et al., 2022). The translation of the formation of the multi-tissue structures from XEn/EpiCs to a bioengineered milliwell-based cell culture platform opens applications for live imaging, single cell tracking (4D) (Soares et al., 2008), longitudinal morphometric studies, and screening (Shankar et al., 2024; Soares et al., 2008). Using porous milliwell arrays (Vrij et al., 2016; Giselbrecht et al., 2006; Truckenmuller et al., 2008; Truckenmuller et al., 2011; Samal et al., 2019) will enable their integration into dynamic flow culture systems that can be combined with the development of XEn/EpiCs to create a closer-to-natural and more spatiotemporally controlled microenvironment. This approach opens new avenues for studying embryonic development and tissue engineering with improved control and precision. Limitations of this model Several limitations of the present study should be considered. First, gene expression analyses were performed using two biological replicates and therefore should primarily be interpreted as indicative of developmental trends. Second, computational deconvolution and single-cell analyses were performed on pooled structures, limiting assessment of inter-structure variability and reducing the precision of inferred cellular compositions. Finally, the barrier-function and hematopoietic analyses provide proof-of-principle evidence and should be interpreted within the context of the current experimental design. While our research has yielded promising results, there remains room for improvement in culture conditions to enhance blood island formation in YS+ structures. We propose that controlled perfusion of soluble factors during XEn/EpiC development, such as by integrating milliwells into a perfused microfluidic chip, could significantly advance this aspect. Our medium-throughput milliwell-based tracking platform provides an essential tool to efficiently optimize protocols, increase the yield of different phenotypes, and offer real-time, in situ information on developing structures. An important consideration is the inter-embryoid heterogeneity observed in XEn/EpiC morphologies, with YS+CM+ structures displaying varying contraction frequencies and developmental rates. This phenomenon parallels gastruloid studies, which have revealed variable differentiation trajectories towards mesodermal or ectodermal lineages (Carvalho et al., 2024). Additionally, although the model recapitulates several key aspects of yolk sac morphogenesis and primitive hematopoiesis, it does not fully reproduce the coordinated embryonic-extraembryonic integration observed in vivo, likely due to the absence of trophoblast-derived signaling centers and incomplete anterior-posterior axis organization. Materials and methods Cell culture mES cell culture and reagents Mouse embryonic stem cells (mES cells) were expanded on a feeder layer of mouse embryonic fibroblasts (MEFs) on 0.15% gelatin in ES medium containing, Dulbecco’s Modified Eagle’s Medium High Glucose (Life Technologies) supplemented with 10% fetal bovine serum (FBS, Greiner), 4 mM Glutamax (Life Technologies), 100 U mL−1 penicillin (Life Technologies), 100 mg mL−1 streptomycin (Life Technologies), 10 mM non-essential amino acids (Life Technologies), and freshly supplemented with 0.1 mM 2-mercaptoethanol (Life Technologies), 1000 U mL−1 leukemia inhibitory factor (LIF, Life Technologies), 3 μM CHIR99021 (GSK3β inhibitor, Axon Medchem) and 1 μM PD0325901 (MEK/ERK inhibitor, Sigma Aldrich). The cells were refreshed every 2 days, passaged on the third day with 0.5 mL Accutase for 3 min, and seeded at a density of 10,000 cells/cm2 along with 0.5 µM Y27632. For all the experiments in this article, mES Gata6: H2B: Venus reporter line was used (mESCs comprising an H2B-Venus reporter under the regulatory elements of Gata6), a kind gift from Schröters’ laboratory (Max Planck Institute of Molecular Physiology, Dortmund, Germany). Fabrication of agarose hydrogel microwells The agarose hydrogel stamps were developed and used for cell culture as described in (Rivera-Perez and Hadjantonakis, 2014). Elastomeric stamps for imprinting the agarose microwell arrays were fabricated using PDMS Sylgard 184 kit. Microwell arrays were molded as described previously (Rosen et al., 2022) using a 2.2% w/v solution of Ultrapure agarose (ThermoFisher, 11560166). Each well of a 96-well plate contained 430 microwells of 200 µm that could hold a calculated liquid volume of 250 µL split between 225 µL medium and 25 µL hydrogel buffer. Protocol to culture of yolk sac-like structures ES cells were expanded for a minimum of two passages before aggregating into EBs (embryoid bodies) within the agarose microwells. The microwells were washed 3x with PBS to remove any remaining gel bits and occasionally pre-wetted in ES medium to saturate the pores in the hydrogel. mES cells were washed 2x with PBS to remove the dead cells, treated with 0.5 mL Accutase for 3 min, centrifuged at 200 g for 5 min, and the pellet was suspended in 2 mL MEF medium containing DMEM (high glucose, Sodium pyruvate, and Glutamax) with 15% FBS. MEFs depletion involved seeding the cell suspension first onto a non-coated T75 flask, to allow MEFs to adhere to the plate, for 20–30 min. The cell suspension was then collected from the flask, centrifuged, re-suspended in advanced. N2B27 medium and counted. The EBs were formed by seeding 18–24 ES cells per microwell in XEn-induction medium containing 46.3% advanced DMEM/F12 (Invitrogen) (46.3% Neurobasal (Invitrogen), 1% N2 supplement (Invitrogen), 2% B27 supplement (Invitrogen), 1% Glutamax, 1% Non-Essential Amino Acids, 1.5% BSA (Sigma), 0.5% HEPES, 0.4% Sodium Pyruvate)) + induction factors consisting 3 μM CHIR99021, 0.1 mM β-mercaptoethanol, 1 mM 8Br-cAMP, 25ug/mL Fgf4, one ug/mL Heparin, 10 nM Retinoic acid, 1 µM Y27632, 5% FBS, and 1% Pen/Strep. The induction medium was washed after 24 h and replaced with plain advanced N2B27 medium with 100 µM β-mercaptoethanol. At 72 h, EBs were flushed out and transferred into non-TCPS six-well plates with 2 mL advanced N2B27 medium supplemented with 100 µM β -mercaptoethanol. The EB clusters in suspension were then refreshed at regular intervals until day 14. Molecular biology techniques qPCR Structures collected at various time points were lysed utilizing RLT Lysis buffer (Qiagen, Hilden, Germany), followed by total RNA extraction using the RNeasy Micro Kit (Qiagen) as per the manufacturer’s protocols. Subsequently, RNA samples were reverse transcribed to synthesize cDNA, employing the iScript cDNA Synthesis Kit (BioRad, Hercules, California, United States). Primer sequences are comprehensively outlined in Table 1 Quantitative PCR analysis was executed on a BioRad CFX96 system utilizing SYBR Green (BioRad), applying a three-step program comprising 40 cycles. Data were normalized to the housekeeping gene (GAPDH) and presented as relative gene expression. Whole mount staining Samples were fixed with 4% v/v PFA (30 min, room temperature). After fixation, samples were washed three times with PBS, permeabilized in a 1% Triton-X solution in PBS, and blocked in blocking buffer (2% BSA, 5% serum of host secondary antibody species, 0.5% glycine, 0.1% Triton-X, 0.2% Tween-20) for 60 min. Samples were incubated in antibody solution (25% blocking buffer with 50% PBS and 25% 0.1% Triton-X in PBS) with primary antibodies for overnight at 4 °C, washed three times for 10 min with washing buffer (0.1% Triton-X with 2% BSA in PBS) followed by incubation with secondary antibodies in antibody solution for 3 h at 37 °C. Some samples were also stained with Hoechst (0.2 μg/mL) and phalloidin (ThermoFisher, A22287 or A12380, 1/100 dilution). A list of used antibodies can be found in Table 2. TABLE 2Primary and secondary antibodies.Primary antibodyTypeCompanyCat.NoDilutionSOX2GoatR&D systemAF20181:100Sox17GoatR&D systemAF1924-SP1:200BrachyuryRabbitAbcamab2096651:200Apoa4RabbitProteintech17996-1-AP1:100NKX2.5RabbitInvitrogenPA5494311:100CTNTRabbitProteintech15513-1-AP1:100CD31RabbitAbcamab283641:100Runx1 & 2RabbitAbcamab923361:200VEGF-ARabbitAbcamab529171:100MEF2CRabbitAbcamab1970701:100Secondary antibodyAnti-Rabbit- Alexa fluor 647DonkeyInvitrogenA-315731:500Anti-Rabbit- Alexa Fluor-568DonkeyInvitrogenA100421:500Anti-goat Alexa fluor 647DonkeyInvitrogenA-214471:500Anti-goat Alexa fluor 568DonkeyInvitrogenA-110571:500 Anti-Rabbit IgG (H + L)-HRP conjugateGoatInvitrogen31,4601:500 Primary and secondary antibodies. Sectioning: Immunofluorescence and immunohistochemistry For immunohistochemistry (IHC), samples were meticulously collected, subjected to triple washes with PBS, and then fixed with 4% v/v PFA for 30 min at room temperature. Agar-embedded structures underwent dehydration via an ethanol series, followed by clearing in xylene, and subsequent embedding in paraffin wax. These paraffinized samples were precisely sectioned into 6-μm slices using a microtome. After deparaffinization and rehydration, the sections underwent antigen retrieval for 2–3 min at 95 °C. Following this step, sections were incubated with blocking buffer for 60 min at room temperature, followed by overnight incubation at 4 °C with primary antibodies, and subsequent incubation with secondary antibodies at 37 °C for 2 h. The detailed list of antibodies utilized in this study is provided in Table 2. Lastly, the nuclei were counterstained with Hoechst dye. Additionally, a standard IHC protocol was adhered to after deparaffinization, rehydration, and antigen retrieval. Endogenous peroxidase activity was quenched with 3% hydrogen peroxide. Subsequently, sections were incubated overnight at 4 °C with primary antibodies, followed by a 60-min incubation at room temperature with secondary antibodies. Signal development was achieved using a two-component high-sensitivity diaminobenzidine (DAB) chromogenic substrate (DakoCytomation) for 10 min, and counterstaining was performed with hematoxylin. Flow cytometry The YS+ blood island structures alone with YS-CM, YS+CM structures were collected and washed once with 1X PBS to remove the residual medium. The structures were incubated in Triple-E (Fisher Scientific #12563011) for at least 30 min with an intermediate vortex step to break the membrane and dissociate into single cells. Sometimes, the incubation time was extended depending on the size and compactness of the structure. After dissociation, the cell pellets were washed and split into three tubes. One tube was incubated with 1 mL of flow buffer or PBS containing the entire panel of blood progenitor makers (Table 3) (Sca-1, cKit, CD34, B220, CD3, Ly6G, NK 1.1, Ter-119), tube 2 with only CD45 and tube 3 with CD11b and F40/80 (to identify the early macrophage cells). The tubes were incubated in ice for at least 20 min and then washed and imaged using BD FACSCanto II. The gating was performed using the pre-existing knowledge of mouse blood samples with the same marker panel. The post-processing and analysis were performed using the BD FACSDiva software. TABLE 3Flow cytometry marker panel.Stem cell markerCABDilutionCatalog numberCompanySca-1 PerCp Cy5.55241:00045–5981-80eBiosciencecKit APC-H76151:10047–171-81eBioscienceCD34 eFluor 4506261:5048–0341-82eBioscienceCD16/32 PE Cy76281:00025–0161-82eBioscienceB220 PE911:100553089BDCD3 PE4501:80012–0031-83eBioscienceLy6G- PE3871:100551461BDNK1.1- PE3851:100557391BDTer-119 PE5881:40012–5921-82eBioscienceCD45 APC8091:10017–0451-82eBioscienceCD11b PE-Cy77811:800561098BDF40/80 APC6351: 800123115Biolegend Flow cytometry marker panel. Functional assays Dextran diffusion assay The integrity of the epithelial barrier was evaluated by assessing the permeability of the fluorescent marker 4 kDa fluorescein isothiocyanate (FITC)-labelled dextran (Sigma-Aldrich). Intact XEn/EpiC models were carefully collected and incubated in a solution containing 2 mg/mL of 4 kDa FITC-dextran for 1 h and 30 min in the controlled environment of the incubator. To disrupt the barrier integrity for comparative analysis, XEn/EpiC models were subjected to treatment with 2 mM ethylenediamine tetraacetic acid (EDTA; VWR) in Hanks’ balanced salt solution (without calcium and magnesium; Thermofisher) while maintained on ice for 1 h and 30 min. Following this treatment, they were promptly resuspended in the same FITC-dextran solution as intact XEn/EpiC models. Subsequently, the XEn/EpiC models were carefully mounted and immediately subjected to imaging using a live cell microscope to visualize and quantify any changes in permeability. This method allowed for a direct comparison of barrier integrity between intact and disrupted XEn/EpiC models, providing valuable insights into the functional characteristics of the XEn/EpiC model. Adrenaline treatment The contracting YS+CM structures were picked and transferred to a glass-bottom 18-well dish (IBIDI) with 100 µL volume of medium. The slides were then imaged inside a temperature and CO2-controlled chamber in a FEI’s CorrSight microscope. The CorrSight was combined with an Andromeda spinning disk, offering a high-speed imaging of the live specimen. The YS+CM+ structures were imaged first to extract the real-time videos of individual structures. The structures were then exposed to 20 µM DL-adrenaline for 1 h at 37° and then imaged again to generate the videos of the contractions. The resulting videos were processed using ImageJ (Fiji) using the Z-stack profile module to extract the contraction frequencies of specific regions. Comparing the contraction profiles of the same structures before and after adrenaline treatment (in the same region of interest) provided the difference in contraction frequencies. Imaging Live cell microscope The structures were imaged using an inverted Nikon Ti-E live cell spinning disk confocal imaging microscope, equipped with environmental control and a CrestOptics X-Light V2 spinning disk unit. For all automated measurements, a ×20 air objective was used (CLWD (O.D. = 2.10 mm, NA = 0.5)). To perform manual measurements of the area of individual compartments (Supplementary Figure S12) a ×40 air-immersion objective (O.D. = 0.66 mm, NA = 0.75) was used. The ‘large image’ module was used to achieve a montage of the entire well with all the milliwells stitched together. Manual image analysis was performed using NIS software and ImageJ (yield and area measurement). All the images were taken using epi-fluorescence imaging in a single focal plane, roughly at the plane where the size of the cavity is at maximum. This assumes that most structures are spherical and the cavity forms in the center plane of the structures. TEM (transmission electron microscopy) YS+CM+, YS+ vascularized network, and YS+ blood island structures were chemically fixed with 2.5% glutaraldehyde in 0.1M phosphate buffer buffered to pH 7.4. Then they were washed with 0.1M cacodylate buffer, embedded in agarose, and postfixed with 1% osmium tetroxide in the same buffer containing 1.5% potassium ferricyanide for 1 h in the dark at 4 °C. After rinsing with MQ, the structures were dehydrated at RT in ethanol, infiltrated with Epon, embedded in the same resin and polymerized for 48 h at 60 °C. Ultrathin sections of 60 nm were obtained using a Leica Ultracut UCT ultramicrotome (Leica Microsystems, Vienna) and transferred onto formvar-coated copper grids. Sections were stained with 2% uranyl acetate in 50% ethanol and lead citrate. Then, sections were observed in a Tecnai T12 Electron Microscope equipped with an Eagle 4kx4k CCD camera (Thermo Fisher Scientific, Netherlands). Milliwell culture Thermoformed milliwell culture Polycarbonate custom thermoformed milliwell sheets were created in house with 800 µm-sized wells, which were cut and placed in a 12-well plate with an O-ring (ERIKS) to secure them in position (Samal et al., 2020). Before usage, the wells were washed in a serial dilution of isopropanol and then, 2x with water. The wells were then incubated with an anti-adherence solution (StemCell Technologies) for ∼20 min at RT inside the laminar flow hood and then washed 3x with PBS and then incubated with fresh PBS until further use. The mES cells were seeded directly into the wells in a droplet pattern such that there is a (∼) homogenous distribution of 18–24 cells per milliwell. The cells were cultured for 24 h in XEn-induction medium (as described above) and then, refreshed with plain advanced N2B27 medium with 100 µM β-mercaptoethanol until day 14. The different phenotypes of structures start to develop sequentially in the milliwells that can be tracked simultaneously. Data analysis Statistical analysis for qPCR The qPCR Ct values were extracted and the 2−ΔΔCT measurements were calculated to obtain the relative fold change values, which were imported into GraphPad Prism (v10.0.2.3) for further analysis. All data were presented as mean ± SD. All experiments were conducted in triplicates with each treatment condition containing pooled sampled from each day. All statistical hypothesis testing was done using Dunnett’s test; ∗ represents P ≤ 0.05, ∗∗ represents P < 0.01, ∗∗∗ represents P < 0.001, ∗∗∗∗ represents P < 0.0001 (One-way ANOVA with Dunnet’s post-test). Bulk sequencing Protocol RNA was extracted from beating structures as part of the qPCR analysis process. Bulk mRNA sequencing was carried out in FHML, Maastricht University, Netherlands. Total RNA was isolated from various stages of the beating structures using the RNeasy Micro Kit (Qiagen), following the manufacturer’s instructions. This RNA was then processed for library preparation using a Takara Bio kit named “SMART-Seq mRNA Low Input” kit). The libraries were then paired-end sequenced on an Illumina NextSeq™ 6000 system. The raw data was then analysed using Takara Cogent™ NGS Analysis Pipeline Software (CogentAP), using the mouse genome GRCm39 (gencode vM27). Deconvolution of embryoid cell types: Pre-processing of external scRNA-Seq data All code necessary for the methods are available upon request. For an overview of the steps, see Supplementary Figure S1. 10X Genomics datasets consisting of raw scRNA-Seq data (FASTQ file format) from mouse embryos at developmental time points E7.5 and E8.5, produced by the Marioni lab, were obtained from the publicly available database ArrayExpress (van den Brink et al., 2020). Samples were filtered to only include wild type controls. Cell Ranger (v7.1.0) was used on a Linux operating system for sample demultiplexing, barcode processing, gene counting, and annotation of the data. The data was aligned to the mouse reference transcriptome (mm10–2020 -A, version M23), which is publicly available at 10X Genomics and is specific for use in Cell Ranger. The Cell Ranger script was utilized with the following parameters: Include introns = true, Expect_cells = 5000, No_bam. This generated a data file for each sample used in the downstream single cell transcriptomics analysis. Generation of single cell signature matrix Using Rstudio 2023.09.1 + 494, R 4.3.2, the pre-processed data for each sample was merged to create a Seurat object containing all samples. Two parameters were used to filter the data. nCount, the total number of reads detected in the sample, and nFeature, the total number of genes detected after alignment of the reads. Exploratory analysis using violin plots was done to determine the filtering thresholds for both these parameters. Thresholds of nCount>1,600 and nFeature>500 were used. No filtering based on mitochondrial or ribosomal RNA was done due to the possibility that developing embryos could have variable content of these forms of RNA. After filtering, Seurat (v4.3.0) was used for data processing steps. Firstly, normalization was carried out using the log normalize method (scale factor = 10,000). Second, dispersion calculation and variable feature selection were performed using the FindVariableFeatures () function (selection method = vst, nFeature = 2000). Dispersion calculation is used to determine how much the expression of a gene varies between cells compared to the mean expression of the gene. In the same function, variable feature selection identifies the most highly variable genes based on dispersion values, using an automatically determined statistical criteria. Finally, the data was scaled using all identified features from the previous step. Principal component analysis (PCA) was done to determine the principal components (PCs) using the variable features found in the previous step. Genes influencing the top two PCs were found, using the VizDimLoadings () function. An elbow plot was used to determine the number of relevant PCs. Computation of the K nearest neighbors (KNN) was done using FindNeighbors () (dims = 20), followed by data clustering with FindClusters () (resolution = 2.5). t-distributed Stochastic Neighbor Embedding (tSNE), a non-linear dimensionality reduction algorithm, was used for data visualization. Cell type annotation on the tSNE plot clusters was done using the scType algorithm created by Aleksandr lIanevski (Ianevski et al., 2022). The annotation process works by a thresholding method (Ianevski et al., 2022). For each cluster, cell types are ranked based on the degree to which the gene markers from the gene lists correlate to the genes expressed in the cluster (Ianevski et al., 2022). The highest-ranking annotation that passes the threshold is used. Additionally, if no cell types pass the threshold, the cluster is labelled as unknown. Several clusters can receive the same annotation, leading to clusters becoming merged. Annotation requires a data file containing a list of expected cell types along with their corresponding positive marker genes (genes overexpressed in the cell type). As a range of developmental times were explored using the embryoids, several gene lists were necessary to explore the corresponding developmental times in the cultured. This resulted in gene lists for the following embryonic days: E3.5, E4.5, E6.5-E7, E7.25-E7.75, E8-E8.5, based on (van den Brink et al., 2020). Cell type annotation of the scRNA-Seq data was carried out individually for each gene list involved. The annotation produces a labelling of the cell clusters. Besides annotation of cells, a reference matrix is also generated in this step. This matrix contains the gene count for each annotated cell type, which has been labelled by the cell type annotation process. This reference matrix requires column name editing, to ensure consistency with bulk RNA-Seq data. This is required for the construction of the signature matrix with CIBERSORTx, necessary for cell type deconvolution. Cell fraction imputation The reference matrix, consisting of single cell gene expression counts for each annotated cell, is used in the “Generate signature matrix” setting of CIBERSORTx in the scRNA-Seq mode. This generated a signature matrix containing the characteristic expression for each cell type. This acts as the data by which the bulk RNA-Seq data was compared to and deconvoluted with. The signature matrix was used in the “impute cell fractions” setting of CIBERSORTx. Additionally, the bulk RNA-Seq mixture matrix to be deconvoluted is required. CIBERSORTx uses a non-linear method for deconvolution. The types of cells present, as well as their quantification, are produced per sample. Single cell RNA sequencing Protocol On day 10, beating structures were collected and treated with TrypLE enzyme for 30 min at 37 °C to dissociate them into single cells. These single cells were then resuspended in FACS buffer, and viable cells (DAPI negative) were sorted into GFP+ and GFP- populations using a FACS sorter (FacsCalibur). The cells were then seeded into cell capture 384-well plates for further processing. Each well in the cell capture plates contained a small 50 nL droplet of barcoded primers and 10 μL of mineral oil (Sigma M8410). Post-sorting, the plates were immediately centrifuged, placed on dry ice, and stored at −80 °C. The plates were then shipped on dry ice to Single Cell Discoveries (Utrecht, Netherlands), where single-cell RNA sequencing was performed using a modified version of the SORT-seq protocol. The cells were heat-lysed at 65 °C, followed by cDNA synthesis. After second-strand cDNA synthesis, all barcoded materials from a single plate were pooled and amplified using in vitro transcription (IVT). The amplified material was then prepared into a cDNA library following the CEL-Seq2 protocol, using TruSeq small RNA primers (Illumina). The DNA library was sequenced on an Illumina NextSeq™ 500 system with a high-output, paired-end sequencing protocol, using a 1 × 75 bp Illumina kit (read 1: 26 cycles, index read: six cycles, read 2: 60 cycles). Validation with embryoid scRNA-seq data scRNA-Seq resulted in a Seurat object containing single cell gene expression counts. To analyze this, a similar approach was taken as with the data obtained from the online database. An additional step of predicted gene removal was done. This removes genes which have been previously found but not yet given a canonical name and have an uncharacterized function. This was done so that downstream analysis results in interpretable findings. The remainder of steps taken follow the same pipeline as for the scRNA-Seq data obtained from the online database. This consisted of filtering using nCount and nFeature, then dispersion calculation and feature selection, and finally data scaling. Again, data exploration with PCA and an elbow plot was carried out. KNN calculations were done, then clustering and non-linear dimensionality reduction tSNE. These clusters were then annotated using scType using the corresponding gene list (E8-E8.5). The reference matrix was created, to be used in CIBERSORTx for generation of the signature matrix. This was then applied to embryoid bulk RNA-Seq data deconvolution. Additionally, DGE analysis was carried out, to find the top 10 DEGs for each identified cell type cluster. This allowed for further analyses using heatmaps and dot plots.