Microgel culture and spatial identity mapping elucidate the signalling requirements for primate epiblast and amnion formation

Mammalian life begins with concurrent establishment of embryonic and extraembryonic lineages. The zygote develops into a blastocyst consisting of the pluripotent epiblast, which gives rise to the embryo, and extraembryonic trophoblast and hypoblast, which form placenta and yolk sac, respectively. Upon implantation, the embryo undergoes major transformations (Bedzhov and Zernicka-Goetz, 2014). In human and non-human primate embryos, the epiblast polarises into a rosette and subsequently undergoes lumen expansion, thereby giving rise to the prospective amniotic cavity (Boroviak and Nichols, 2017; Ross and Boroviak, 2020; Rossant and Tam, 2018). Concomitantly, extraembryonic trophoblast cells mediate embryo attachment to the maternal endometrium. Inside the implanting conceptus, the hypoblast expands and diversifies into visceral endoderm, the primary and secondary yolk sacs, and, in primates, is thought to generate extraembryonic mesoderm (Bianchi et al., 1993; Nakamura et al., 2016; Ross and Boroviak, 2020). The distal part of the epiblast rosette, in contact with the visceral endoderm, remains pluripotent, while the proximal part of the rosette, adjacent to trophoblast, differentiates into amnion (Enders et al., 1986; Rock and Hertig, 1948). Amnion cells downregulate the pluripotency factor SOX2 and form a squamous epithelium, in contrast to the taller columnar cells of the post-implantation epiblast (also embryonic disc) expressing the core pluripotency genes NANOG, SOX2 and POU5F1 (also known as OCT4) (Hertig et al., 1956; Luckett, 1978). Nascent amnion and post-implantation epiblast present exceedingly similar global transcriptomes (Ma et al., 2019; Niu et al., 2019), which highlights the close kinship between embryonic and extraembryonic lineages in the primate embryo.

Lineage specification is orchestrated by a small suite of signalling pathways. In the early post-implantation embryo, the pluripotent epiblast compartment is exposed to BMP, FGF, WNT and activin/nodal ligands from surrounding extraembryonic tissues (Bergmann et al., 2022; Nakamura et al., 2016; Sasaki et al., 2016; Xiang et al., 2020). BMP signalling has been implicated in amnion formation in post-implantation amniotic sac embryoids (Shao et al., 2017a,b) and microfluidic-based models for amniogenesis (Zheng et al., 2019), but is also required for spatially organised germ layer and extraembryonic differentiation in human gastruloids (Chhabra et al., 2019; Deglincerti et al., 2016a; Tewary et al., 2017; Warmflash et al., 2014). Recent reports have shown that human epiblast cells readily differentiate into extraembryonic lineages in vitro (Dong et al., 2020; Linneberg-Agerholm et al., 2019; Rostovskaya et al., 2022; Xu et al., 2002; Zheng et al., 2019) and in vivo (Guo et al., 2021), suggesting greater lineage plasticity compared with rodent models (Chazaud and Yamanaka, 2016; Grabarek et al., 2012; Nichols and Gardner, 1984; Rossant, 2018). However, the signalling code required to protect epiblast identity in the primate embryo from precocious differentiation has remained elusive.

Human post-implantation development remains poorly understood because the relevant stages of in vivo-developed embryos are difficult to access. Recent protocols for human blastocyst culture to post-implantation stages have opened exciting avenues to model embryo implantation in a dish (Deglincerti et al., 2016b; Shahbazi et al., 2016; Xiang et al., 2020), but human embryos are scarce and therefore not suitable for high-throughput functional interrogation. Stem cell-based embryo models are currently being developed to overcome these challenges (Chhabra et al., 2019; Harrison et al., 2017; Rivron et al., 2018; Shahbazi et al., 2017; Shao et al., 2017a,b; Warmflash et al., 2014; Yu et al., 2021) and provide powerful alternatives for functional analysis of human embryogenesis (Moris et al., 2020; Shao et al., 2017a,b; van den Brink et al., 2020; Zheng et al., 2019). Nevertheless, human models lack access to in vivo controls and, to date, no non-human primate stem cell-based embryo models have been developed to enable direct comparison with the post-implantation embryo by single-cell profiling within the same species.

To delineate the acquisition of regional lineage identities in primate embryogenesis in vivo, we recently conducted spatial embryo profiling in the common marmoset (Bergmann et al., 2022). Here, we encapsulate small populations of common marmoset pluripotent stem cells (cmPSCs) into agarose microgels. The scaffold provided by the soft gel creates a biomimetic 3D environment for the generation of epiblast (Epi) and amnion (Am) spheroids as a binary model for embryonic disc versus amnion lineage specification. Importantly, working with the marmoset enabled us to authenticate spheroid cultures by direct comparison with the embryo. We functionally assessed the complex signalling environment of the post-implantation primate embryo and show that the activin/nodal axis safeguards embryonic lineage identity, in contrast to FGF/MAPK and BMP signalling. Our work provides a paradigm to generate, interrogate and authenticate regionalised tissues of the primate embryo.

We set out to establish a stem cell-based model in microgels to functionally interrogate marmoset embryogenesis. Naïve (Bergmann et al., 2022) and primed (Sasaki et al., 2005; Thomson et al., 1996) cmPSCs cultured in self-renewing conditions were dissociated and mixed with low melting agarose solution (Fig. 1A, Fig. S1). Monodisperse water-in-oil emulsion droplets were generated in microfluidic polydimethylsiloxane (PDMS) devices from two streams of aqueous and oil phases by break-off flow focussing (Kleine-Brüggeney et al., 2019; Schindler et al., 2021) (Fig. 1A). The cmPSC-containing agarose microgels were transferred into conventional tissue culture plates containing mouse embryonic fibroblasts (MEFs) at the bottom of the dish.

We screened encapsulated cmPSCs in different experimental conditions to identify suitable culture regimes to model epiblast morphogenesis. Experimental conditions included media that promoted self-renewal for naïve (PLAXA) (Bergmann et al., 2022) and primed (KSR/FGF2 and E8) pluripotency and differentiation permissive, serum-free N2B27. Furthermore, we examined the addition of either no extracellular matrix (ECM) or 1% Matrigel for all culture conditions (Fig. S1A). cmPSCs readily formed 3D structures in agarose microgels, similar to human pluripotent stem cells (hPSCs) (Schindler et al., 2021). 3D-structure formation of microgel-encapsulated cmPSCs was more efficient in 1% Matrigel (40-80% of gels containing structures) than without ECM (15-40%) (Fig. S1B-E). To determine the cellular organisation and lineage identity of encapsulated marmoset 3D structures, we performed whole-mount confocal immunofluorescence at day 6.

Naïve cmPSCs encapsulated in agarose microgels and cultured in PLAXA retained the naïve pluripotency factors KLF17 and TFAP2C and the core pluripotency factor SOX2, in line with pre-implantation epiblast identity (Fig. S1B). Encapsulated naïve cmPSCs initiated polarity and rosette formation, but completely failed to undergo lumen expansion in PLAXA and at low frequency in primed or differentiating media (Fig. S1B,F). In differentiating media, loss of KLF17 expression in naïve cmPSCs correlated with lumen size (Fig. S1H). To assess the identity of KLF17-negative cells, we performed whole-mount immunofluorescence for pluripotency and differentiation markers after 6 days in N2B27 culture. The majority of cells retained expression of SOX2 and OCT4, and did not upregulate primitive streak (TBXT), endoderm (SOX17), mesoderm (PDGFRα), trophoblast (GATA3) or amnion (TFAP2A) markers, suggesting progression from naïve to primed pluripotency in microgel culture (Fig. S2).

Primed cmPSCs generated 3D-structures positive for SOX2, but negative for KLF17 and TFAP2C in differentiating and primed self-renewing conditions, similar to the marmoset (Bergmann et al., 2022), cynomolgus (Nakamura et al., 2016) and human (Tyser et al., 2021; Xiang et al., 2020) post-implantation epiblast (Fig. S1C). We observed upregulation of KLF17 and TFAP2C in encapsulated primed cmPSCs when cultured in PLAXA, indicating partial resetting towards naïve pluripotency (Fig. S1C). Primed cmPSCs formed polarised spheroids with a lumen in both self-renewing or differentiating conditions (Fig. 1B, Fig. S1C). 3D structures in differentiating conditions (N2B27) exhibited the highest levels of lumen expansion (Fig. S1B,C,G). Collectively, these data are in accordance with the reported increase in lumenogenesis of human PSCs during the naïve to primed transition (Schindler et al., 2021; Shahbazi et al., 2017; Shao et al., 2017b).

Microgel cultures of primed cmPSCs in differentiation-permissive N2B27 formed homogenous spheroids, consisting of tall columnar epithelial cells surrounding an expanded lumen (Fig. 1B-D, Fig. S1C) reminiscent of the embryonic disc in the early post-implantation embryo. Spheroid size followed a normal distribution in agarose microgels with diameters of 40-70 µm (Fig. 1C). To determine the developmental identity of N2B27-cultured spheroids from primed cmPSCs, we performed confocal imaging with lineage markers of the early post-implantation embryo. Spheroids expressed core pluripotency factors SOX2, OCT4 and NANOG, apical polarity markers ezrin, WGA and PAR6, low levels of TBXT (brachyury) and lacked expression of the naïve pluripotency factor KLF17, amnion/trophoblast marker TFAP2C and the endoderm regulator SOX17 (Fig. 1D, Movie 1). This profile was consistent with post-implantation epiblast identity (Bergmann et al., 2022). Consequently, we refer to the 3D structures derived from encapsulated primed cmPSCs in N2B27 on MEFs as Epi-spheroids.

Microgel suspension culture of cmPSCs in the presence of MEFs showed sustained SOX2 expression across experimental conditions (Fig. S1C). To examine whether signals from MEFs at the bottom of the dish were required for epiblast identity, we removed MEFs from the N2B27-based culture regime (Fig. 2A). Strikingly, instead of undergoing Epi-spheroid formation, encapsulated cmPSCs developed into cysts consisting of a thin, squamous epithelium (Fig. 2B). Removal of MEFs did not lead to an increase in cell death (Fig. S3A,B). The cysts were polarised as indicated by the apical polarity markers ezrin, WGA and PAR6, and expressed lineage markers OCT4 and TFAP2C in the absence of SOX2, TBXT and SOX17 (Fig. 2C,D, Fig. S3C,D, Movie 2). Loss of SOX2, together with TFAP2C and OCT4 expression is a distinguishing feature of nascent amnion in primates (Bergmann et al., 2022; Ma et al., 2019; Nakamura et al., 2016). Thus, we named the resulting structures Am-spheroids.

Am-spheroids underwent eightfold larger lumen expansion than Epi-spheroids (Fig. 2C). Owing to this expanded lumen size, Am-spheroids frequently outgrew the 100 µm agarose microgels (Fig. S3E). Escaped Am-spheroids could be collected and further grown in suspension culture (Fig. S3F). To capture differences in the cellular architecture between Epi- and Am-spheroids, we measured nuclear orientation and epithelial thickness. Nuclei in Epi-spheroids consistently pointed towards the centre of the lumen, in contrast to Am-spheroids, where nuclei were arranged in a perpendicular direction, pointing along the surface of the vesicle (Fig. 2E). Epithelial thickness was 10-fold increased in Epi- over Am-spheroids (Fig. 2E). Notably, the observed differences for nuclear orientation and epithelial thickness in the spheroid model were consistent with differences in the epithelial structure of human amnion and epiblast in the post-implantation embryo (Fig. 2F).

To assess the developmental identity of Epi- and Am-spheroids by comparison with the marmoset embryo, we isolated spheroids from the microgels through Agarase treatment (Fig. S3G) and performed full-length single-cell transcriptome profiling (Picelli et al., 2014). The embryo dataset (Bergmann et al., 2022) included preimplantation lineages from Carnegie stages (CS) 1-3 and post-implantation stages CS5, CS6 and CS7 (Fig. 3A). Global transcriptome comparison demonstrated that Epi-spheroids exhibited highest similarity with the early embryonic disc (CS5 and CS6) (Fig. S4A). Epi-spheroids expressed core pluripotency and post-implantation epiblast lineage markers, in the absence of differentiation- or preimplantation-associated genes (Fig. 3B, Fig. S4C).

Am-spheroids showed close correspondence to CS6 amnion and expressed the amnion marker genes TFAP2A, TFAP2C, VTCN1 and WNT6 (Fig. 3B, Fig. S4A). Importantly, Am-spheroids did not express the trophoblast-specific transcripts JAM2, CGA or CGB3 (Fig. 3B). We observed downregulation of embryonic disc-associated factors, including SOX2, NANOG, SFRP2 and DNMT3B, but also ISL1, a gene heterogeneously expressed in the marmoset amnion and gastrulating cells in vivo (Bergmann et al., 2022) (Fig. S4C). Integrated analysis showed that amnion and epiblast lineage separated along the first and second principal component (Fig. 3C). We computed a support-vector-machine-decision boundary based on in vivo samples to facilitate Epi- and Am-spheroid mapping (Fig. 3C, Fig. S4C). Am-spheroids robustly clustered with CS6 amnion, and Epi-spheroids grouped with SOX2-positive post-implantation epiblast samples at CS5-7 (Fig. S4C). To independently examine whether Epi- and Am-spheroids differentiate along the embryonic disc and amnion lineage, we compared differentially expressed genes between embryonic disc and amnion at CS6 (Fig. 3D), and between Epi- and Am-spheroids (Fig. 3E). We observed a broad overlap of lineage-specific transcripts (Fig. 3D,E, Fig. S4C) and gene ontology (Fig. S4B). These observations demonstrate that Epi- and Am-spheroids recapitulate the cellular architecture, polarity and transcriptome-wide signature of CS6 embryonic disc and amnion, respectively, and validate microgel cultures as a biomimetic maquette with which to investigate primate development.

Spatial transcriptome profiling in the early post-implantation marmoset embryo revealed the signalling environment in space and time, featuring multiple ligands of the BMP, FGF/MAPK, WNT and activin/nodal pathways (Bergmann et al., 2022) (Fig. 4A). We reasoned that the microgel suspension culture regime, generating Epi-spheroids in the presence of MEFs and Am-spheroids in the absence of MEFs, provides a powerful platform in which to functionally dissect the signals required for embryonic disc and amnion formation. Single-cell RNA-seq of MEFs used for microgel suspension culture demonstrated expression of ligands from FGF/MAPK, BMP, activin/nodal and WNT signalling cascades (Fig. S5A). To delineate the individual roles of BMP, FGF/MAPK, WNT and activin/nodal signalling, we activated or inhibited each signalling pathway during Epi- and Am-spheroid formation (Fig. 4B). Consistent with previous results, control conditions gave rise to Epi-spheroids in the presence of MEFs and Am-spheroids without MEFs (Fig. 4C, Fig. S5B). Encapsulated cmPSCs displayed robust cell survival throughout culture conditions, apart from BMP inhibition with LDN193189 (LDN) (Fig. S5C,D). BMP4 induced amnion both with and without MEFs, in accordance with observations in human post-implantation amniotic sac embryoids (Shao et al., 2017a,b) (Fig. 4C, Fig. S5C). Activation of FGF/MAPK signalling with FGF2 was compatible with Epi-spheroid morphology, but ablated Am-spheroid development (Fig. 4C, Fig. S5C). Interestingly, MEK inhibition with PD0325901 (PD03) did not change Epi- or Am-spheroids formation compared with controls (Fig. 4C, Fig. S5C). WNT activation with the glycogen synthase kinase (GSK3β) inhibitor CHIR99021 (CHIR) shifted development towards embryoid body-like structures without lumen (Fig. 4C, Fig. S5C). Inhibition of WNT signalling did not alter the dynamics of post-implantation epiblast and amnion formation in the microgel suspension culture system, similar to PD03 (Fig. 4C, Fig. S5C). In contrast, inhibition of the activin/nodal signalling pathway with SB431542 (SB43) resulted in Am-spheroid formation in the presence of MEFs, i.e. reversing cell fates compared with control conditions. Remarkably, activation of activin/nodal with activin A (ActA) had an equally profound effect, leading to the formation of Epi-spheroids in the absence of MEFs (Fig. 4C, Fig. S5C). This suggests that epiblast versus amnion formation is predominantly controlled by the activin/nodal axis in the microgel culture system.

To capture the dynamics of lineage allocation, we performed whole-mount confocal imaging for all experimental conditions with the early lineage markers SOX2 (embryonic disc) and TFAP2C (amnion) (Fig. 4D-G). We determined nuclear SOX2 and TFAP2C intensity levels on each individual frame and plotted SOX2/TFAP2C ratios in scatterplots as a quantitative readout for Epi- and Am-spheroid formation (Fig. 4D-G, Figs S6, S7). Untreated controls cultured in the presence of MEFs resulted in SOX2-positive Epi-spheroids with a columnar epithelium; in the absence of MEFs, encapsulated cmPSCs predominantly gave rise to TFAP2C-positive Am-spheroids (Fig. 4D). Microgel culture in the presence of FGF2 increased SOX2 levels in the absence of MEFs, but led to inconsistent columnar epithelial architecture across the individual structures (Fig. S6A). In the primate embryo, visceral endoderm expresses insulin growth factor 1 (IGF1) (Bergmann et al., 2022), which effectively supports the human pluripotency network in vitro (Wamaitha et al., 2020). We tested IGF1 in the Epi- and Am-spheroid assays, and observed slightly decreased efficiencies for amnion formation, similar to FGF2 (Fig. S6F). PD03 exerted only modest effects on epiblast or amnion specification (Fig. 4C, Fig. S6A). Activation of WNT signalling resulted in loss of both SOX2 and TFAP2C, while WNT inhibition had a limited effect on Epi- and Am-spheroid formation (Fig. S6B). BMP activation shifted almost the entire population towards TFAP2C positive amnion with and without MEFs (Fig. 4E). BMP inhibition through LDN was toxic in the absence of MEFs, but not in the presence of MEFs (Figs S5D, S7A), potentially suggesting that BMP is required for amnion lineage entry. We repeated this experiment using noggin (NOG), a soluble BMP inhibitor. In contrast to LDN, NOG was not toxic to spheroid formation in the absence of MEF and produced only a mild decrease in appearance of squamous spheroids (Fig. S7A,B). Microgels cultured with NOG in the absence of MEFs expressed slightly elevated levels of SOX2 and lower levels of TFAP2C, compared with Am-spheroid controls (Fig. S7C-E). Thus, BMP inhibition with NOG slowed the loss of epiblast identity, but did not prevent Am-spheroid formation.

Modulation of the activin/nodal signalling pathway profoundly changed cell fates in microgel suspension culture, with SB43 shifting the balance towards TFAP2C-positive Am-spheroids with and without MEFs (Fig. 4F). Notably, SB43 treatment with MEFs (Epi-spheroid condition) yielded more homogenous TFAP2C-positive Am-spheroids (Fig. 4F) than control Am-spheroids in the absence of MEFs (Fig. 4D). Positive stimulation of the activin/nodal axis with ActA robustly blocked TFAP2C expression and induced an increase in SOX2 levels in the absence of MEFs (Fig. 4G). Transforming growth factor β (TGFβ), commonly used in self-renewing human PSC culture, upregulated SOX2, but did not block TFAP2C induction in the absence of MEFs (Fig. S6C-E). We conclude that Epi-spheroid formation was compatible with FGF/MAPK or WNT inhibition, but crucially depended on activin/nodal signalling.

To examine the implications of signalling pathway modulation for embryo lineage entry, we performed single-cell transcriptome profiling of ActA-, SB43-, CHIR-, FGF2- and BMP4-treated spheroid cultures. Transcriptome-wide correlation analysis showed that ActA-treated cultures with and without MEFs and FGF2-stimulated spheroids correlated with Epi-spheroids (Fig. 5A). In contrast, BMP4-treated spheroids with and without MEFs and microgel cultures with SB43 on MEFs resembled Am-spheroids (Fig. 5A). These lineage-converted spheroids upregulated epiblast and amnion markers (Fig. 5B), in the absence of neural (SOX1)-, trophoblast (CGB3)- or hypoblast (TTR and SOX17)-associated gene expression, including extraembryonic mesoderm (HGF) (Fig. S8A). We individually assessed perturbation experiments with embryonic disc and amnion samples from the post-implantation marmoset embryo (CS5-7). BMP4 conditions with or without MEFs promoted amnion lineage identity (Fig. 5C,D, Fig. S8B). FGF stimulation resulted in a substantial shift towards embryonic disc in spheroid culture without MEFs, but localised close to the decision boundary (Fig. S8C). CHIR treatment led to upregulation of FOXF1, HAND1, CDX2 and SNAI2, consistent with advanced mesodermal fate (Tyser et al., 2021) (Fig. S8C,D). ActA treatment of spheroids with or without MEFs overlapped with embryonic disc identity (Fig. 5E,F), although ActA without MEFs upregulated TBXT and MIXL1 (Fig. S8A). Finally, SB43-induced reversal of embryonic disc to amnion fate in spheroid cultures with MEFs resulted in complete overlap with control Am-spheroids and amnion samples from the marmoset embryo (Fig. 5G). These results demonstrate that amnion specification occurs in the absence of activin/NODAL or through induction of BMP signalling, with a subordinate role for FGF/MAPK.

We assessed spheroids obtained from the single pathway perturbation assay by dynamic spatial identity mapping to delineate the influence of each signalling pathway on the regional identity of the embryo. Global transcriptome comparison indicated highest similarity of spheroids to embryonic stage CS6 (Fig. S4A), thus we focused spatial identity mapping on CS6 embryonic disc and amnion (Fig. 6A). Core pluripotency factors SOX2 and POU5F1 are expressed in the anterior embryonic disc, with TBXT demarcating the posterior compartment (Bergmann et al., 2022) (Fig. 6B). Amnion broadly expresses the lineage markers TFAP2A and TFAP2C, whereas VTCN1 and GABRP are confined towards central regions (Bergmann et al., 2022) (Fig. 6B). ISL1 is expressed in the primitive streak and at low heterogenous levels throughout the amnion (Fig. 6B). We performed spatial identity mapping of Epi- and Am-spheroids by projecting a similarity metric based on correlation onto embryo samples followed by Gaussian process regression (Bergmann et al., 2022) to interpolate values to the entire embryo surface. Epi-spheroids showed highest correlations with anterior embryonic disc regions (Fig. 6C). In contrast, Am-spheroids were most similar to mature amnion regions (Fig. 6D). For dynamic spatial identity mapping, we plotted the changes in lineage identity on the virtual embryo model in comparison with the static correlations for Epi- (Fig. 6C) and Am-spheroids (Fig. 6D). We monitored lineage markers in ternary plots next to dynamic spatial identity graphs to examine the effects of individual signalling perturbations (Fig. 6E-J). Dynamic spatial identity mapping highlights the embryonic region undergoing changes in lineage identity, hence ActA treatment of encapsulated cmPSCs with MEFs, i.e. Epi-spheroid conditions, did not cause changes (Fig. 6E). However, SB43 (Fig. 6F) and BMP4 (Fig. 6G) on MEFs caused a dramatic shift towards amnion. ActA without MEFs, i.e. Am-spheroid conditions, effectively changed regional identity towards the anterior embryonic disc; however, samples also mapped towards a transitional zone between embryonic disc and amnion to a similar degree (Fig. 6H). Activation of FGF/MAPK signalling directed lineage identity away from mature amnion towards the embryonic disc (Fig. 6I). FGF signalling effectively suppressed amnion lineage markers, but only marginally upregulated pluripotency-associated transcripts (Figs 5B, 6I, Fig. S8C). Thus, the FGF/MAPK axis counteracts acquisition of the transcriptional circuitry of the amnion, rather than inducing embryonic disc identity.

We noticed that Am-spheroids cultured in the presence of BMP4 exhibited faster growth dynamics than control Am-spheroids (Fig. 7A). BMP4 induced large squamous spheroids that escaped from the agarose microgels by day 4 and quantification of Am-spheroids at day 3 showed that BMP accelerated Am-spheroid formation (Fig. 7B). To assess whether this acceleration led to developmentally more advanced stages, we compared the BMP4-treated and control Am-spheroids with amnion in vivo. PCA analysis with CS3 epiblast and CS5-7 amnion samples revealed that BMP4-treated spheroids clustered slightly towards the more mature CS7 amnion (Fig. 7C), which was accompanied by upregulation of mature amnion marker genes HAND1, VTCN1 and GABRP (Fig. 7D). However, BMP4 treatment was not sufficient for downregulation of POU5F1, indicating only partial maturation (Fig. 7D). Considering that ActA promoted Epi-spheroid identity, we tested whether BMP activation, together with activin/nodal inhibition further promotes amnion maturation by extinguishing residual pluripotency-associated transcripts. Similar to BMP4, combined BMP4 and SB43 treatment increased the formation of squamous epithelial spheroids at day 3 (Fig. 7B). This combinatorial approach resulted in Am-spheroids with particularly thin nuclei, reflected in a higher basolateral/apical nuclear orientation than BMP4 treatment alone or control Am- or Epi-spheroids (Fig. S9A-D). Lineage marker quantification showed that the mature amnion marker VTCN1 was highly expressed after both BMP4 and BMP4+SB43 treatments (Fig. 7E,G, Fig. S9D), in agreement with transcriptional upregulation (Fig. 7D). TFAP2A was also upregulated in both BMP4 and BMP4+SB43 cultures, and correlated with VTCN1 levels (Fig. 7E,G). Both treatments induced downregulation of OCT4 and completely suppressed SOX2 expression at the protein level (Fig. 7F,H). Consistent with the role of ActA in promoting pluripotent circuitry, BMP+SB43 treatment further abolished OCT4 expression (Fig. 7F,H). Thus, BMP4 both promotes amniogenesis and supports amnion maturation, which can be accelerated through activin/nodal inhibition.

Morphological and transcriptional analysis suggest that marmoset spheroid cultures represent a bona fide 3D-culture system to emulate epiblast and amnion formation in vitro. Epi-spheroids consist of a columnar epithelium, in contrast to Am-spheroids, which comprise thin, squamous tissue surrounding an expanded lumen. Recent models for human amniogenesis include the generation of amnion-like tissue in amnion-on-a-chip (Zhu et al., 2020) microfluidic devices (Zheng et al., 2019) and conventional 2D tissue culture (Guo et al., 2021; Rostovskaya et al., 2022). Importantly, working with marmoset cells enabled direct authentication of lineage identity by comparison with our spatial embryo profiling dataset (Bergmann et al., 2022). This approach allowed us to dissect and quantitatively assess the individual roles of the major signalling cascades controlling epiblast and amnion specification (Fig. 8).

Single pathway perturbation assays in spheroid cultures revealed that activin/nodal was the only pathway capable of reversing the epiblast versus amnion lineage decision. This result highlights a predominant role of activin/nodal for embryonic lineage identity. NODAL is expressed in the primate preimplantation epiblast, in contrast to mouse (Blakeley et al., 2015; Boroviak et al., 2018; Nakamura et al., 2016; Petropoulos et al., 2016; Yan et al., 2013) and, upon implantation, visceral endoderm becomes the major source of NODAL in human, cynomolgus and marmoset embryos (Bergmann et al., 2022; Ma et al., 2019; Nakamura et al., 2016). Inhibition of activin/nodal diverts human naïve pluripotent cells towards trophoblast (Guo et al., 2021), in accordance with the signalling requirements for human trophoblast stem cells, which include the activin/nodal inhibitors SB43 and A83-01 as crucial components (Okae et al., 2018; Turco et al., 2018). Although the activin/nodal pathway is not essential for establishing pluripotency in the embryo (Boroviak et al., 2015), it is required for long-term maintenance of naïve cultures in the marmoset (Bergmann et al., 2022). This is consistent with human naïve culture conditions, which require activin/nodal signalling for maintenance of the naïve pluripotency gene regulatory network (Guo et al., 2021; Osnato et al., 2021) and frequently either use MEFs or include activin/nodal ligands (Takashima et al., 2014; Theunissen et al., 2014). In primed human PSCs corresponding to the post-implantation epiblast (Bergmann et al., 2022; Nakamura et al., 2016), activin/nodal suppresses neuronal differentiation and is essential for the maintenance of the pluripotency circuitry (Beattie et al., 2005; Vallier et al., 2009; Xu et al., 2008; Zorzan et al., 2020). However, this function is transformed by WNT ligands during gastrulation to induce germ layer formation (Yoney et al., 2018, 2022). We have identified localised expression of WNT3 and WNT8A in the posterior embryonic disc of the marmoset embryo, together with continuous, high-level expression of NODAL from the adjacent visceral endoderm (Bergmann et al., 2022). Considering the pronounced effects of ActA and SB43 in the marmoset spheroid culture assay, we propose a model where activin/nodal signalling preserves pluripotency in the anterior compartment of the embryonic disc (Bergmann et al., 2022). This safeguarding mechanism is likely required to prevent both extraembryonic and neuronal differentiation prior to gastrulation, and to sustain pluripotency until somitogenesis.

BMP signalling promoted amnion formation, consistent with previous stem cell-based embryo models in human (Shao et al., 2017a,b; Zheng et al., 2019). In the marmoset embryo, the visceral endoderm broadly expresses the BMP inhibitors CHRD and NOG, probably to confine BMP-induced amnion formation towards the periphery of the embryonic disc (Bergmann et al., 2022). Notably, previous in vitro studies interrogated the effects of exogenous signals on lineage specification exclusively in the presence of FGF ligands (Yoney et al., 2018; Zheng et al., 2019). In contrast, we examine the relevant signalling pathway individually and show that FGF/MAPK primarily blocks amnion lineage acquisition, rather than actively supporting embryonic disc identity. This may suggest a role for FGF in pacing amnion formation.

Rodent embryogenesis is orchestrated by strict lineage bifurcation events between embryonic and extraembryonic tissues (Chazaud and Yamanaka, 2016; Nichols and Smith, 2009; Rossant and Tam, 2009). Conversion of mouse naïve pluripotent stem cells into trophoblast or hypoblast requires either genetic manipulation (McDonald et al., 2014; Niwa et al., 2000; Schröter et al., 2015; Wang et al., 2011) or prolonged differentiation (Cho et al., 2012). In contrast, human and non-human primate naïve PSCs are prone to spontaneous differentiation into extraembryonic lineages, including trophoblast, hypoblast and amnion (Dong et al., 2020; Guo et al., 2021; Linneberg-Agerholm et al., 2019; Shao et al., 2017b; Theunissen and Jaenisch, 2014). This increased developmental plasticity might be a consequence of primate-specific features of early embryogenesis, in particular the rapid growth of extraembryonic membranes. Primate embryos specify amnion, produce substantial amounts of extraembryonic mesenchyme, and develop a primary and secondary yolk sac prior to gastrulation (Boroviak and Nichols, 2017; Ross and Boroviak, 2020). The requirement to generate such lineage diversity early on is likely to confer a higher regulative capacity for cells of the primate embryo. At the same time, mechanisms must be in place to protect the embryonic lineage from precocious differentiation. Our results suggest that activin/nodal signalling from the visceral endoderm protects the embryonic lineage in the epiblast and thus provides a developmental rationale for the role of activin/nodal and FGF signalling in promoting human and non-human primate PSC self-renewal.

Microgel encapsulation of human and non-human primate PSCs establishes a powerful technical foundation for disentangling the signals that control embryogenesis and for addressing comparative evolutionary questions about extraembryonic lineage specification. Encapsulated primed cmPSCs readily formed Am-spheroids, whereas human PSCs in the same conditions gave rise to embryonic disc-like structures (Schindler et al., 2021). At the same time, primed cmPSCs transcriptionally correspond to CS5/6, which is slightly earlier than primed human PSCs (Bergmann et al., 2022; Nakamura et al., 2016). It is tempting to speculate that amnion formation may be leveraged as a functional readout for CS5/CS6-equivalent PSCs in human. Alternatively, human PSCs may require higher levels of BMP signalling and further studies will be required. With a throughput exceeding 100 microgels per second, very large numbers of cell-laden biomimetic microspheres can be generated, providing easy access to defined primate embryo compartments that can be authenticated by direct comparison to the embryo. Microgel encapsulation can be readily extended to incorporate extraembryonic tissues through co-encapsulation (Cordero-Espinoza et al., 2021; Schindler et al., 2021), as exemplified with human PSC and mouse extraembryonic endoderm (XEN) cells (Schindler et al., 2021). Ultimately, this provides an important step for the establishment of scalable integrated stem cell-based embryo models to systematically interrogate lineage formation in primate embryogenesis.

Embryo-derived conventional marmoset PSC lines New2f and New4f (established at the Central Institute for Experimental Animals, Kawasaki, Japan; courtesy of Prof. Erika Sasaki) were maintained in KSR/FGF2 media, which comprises Dulbecco's Modified Eagle Medium (DMEM)/F12 (21331, Gibco) supplemented with 20% Knockout Serum Replacement (KSR) (10828028, Thermo Fisher Scientific), 1% GlutaMAX (35050061, Thermo Fisher Scientific), 1% MEM non-essential amino acids (11140035, Thermo Fisher Scientific), 100 μM β-mercaptoethanol (21985023, Thermo Fisher Scientific) and 10 ng/ml FGF2 (Cambridge Stem Cell Institute). Cells were routinely cultured on mitomycin C (M4287, Sigma) inactivated mouse embryonic fibroblast (MEF) feeder cells (Cambridge Stem Cell Institute) under 5% O₂ and 5% CO₂ at 37°C (Kishimoto et al., 2021). Medium was changed daily and cells were passaged every 2 to 4 days by dissociation with Accutase (00-4555-56, Thermo Fisher Scientific) for 5 min.

Conversion to naïve pluripotency was performed as previously described (Bergmann et al., 2022). Briefly, conventional marmoset PSCs were seeded as clumps of 2 to 5 cells 1 day before resetting at 50,000 cells per well of a 12-well plate (1.3×10⁴ cells/cm²) on MEFs. After 24 h, media was changed to PLAXA, which comprised N2B27 medium (NDiff, Y40002 Takara Bio) supplemented with 1 μM PD0325901 (Cambridge Stem Cell Institute), 10 ng/ml recombinant human LIF (Cambridge Stem Cell Institute), 50 μg/ml L-ascorbic acid (Sigma), 5 μM XAV939 (SM38-200, Cell Guidance Systems) and 20 ng/ml Activin A (Cambridge Stem Cell Institute). Throughout conversion, cells were passaged with Accutase (00-4555-56, Thermo Fisher Scientific) 1:1.5 every 3 to 4 days. Dome-shaped colonies first emerged at day 4 to 5 and naïve conversion was completed by day 9.

Cells were encapsulated as previously described (Kleine-Brüggeney et al., 2019; Schindler et al., 2021). In brief, flow focusing in a microfluidic device was used to achieve 100 µm diameter agarose droplet formation (Anna et al., 2003). Microfluidic devices were made from polydimethylsiloxane (PDMS) and had two inlets for an aqueous phase (agarose+cells) and a continuous oil phase, as well as one outlet for the emulsified droplets. Before encapsulations, cells were dissociated to single cells by treatment with Accutase. Cells were resuspended and diluted to 1.6×10⁶ cells/100 μl. Subsequently, the cell suspension was mixed 1:1 with an ultra-low melting agarose solution at 37°C (3% in PBS; SeaPrep ultra-low melting agarose by Lonza). HFE-7500 3M Novec (Fluorochem) supplemented with surfactant (0.3%; Pico-Surf, Sphere Fluidics, C022) was used as the continuous oil phase. Syringes (SGE Analytical Science) controlled by automated pumps (CETONI, neMYSIS) were used to inject the agarose-cell suspension separately from the oil-surfactant solution into the microfluidic chips for emulsification. Agarose droplets left the microfluidic chip through the outlet and were collected in a test tube on ice for polymerisation. Microgels were then de-emulsified via liquid-liquid extraction by adding culture medium (200 µl) and 1H,1H,2H,2H-Perfluoro-1-octanol (45 µl; AlfaAesar, B20156.18). Microgel suspension cultures were kept at 37°C under hypoxic conditions (5% CO₂, 5% O₂).

During the first 24 h after the encapsulation, ROCK inhibitor (10 µM, Tocris, 1254) was added to the medium. Media were changed every other day by gently centrifuging (200 g for 5 min), otherwise fresh medium was simply added to the wells. Cells were cultured with 1% Matrigel (Corning, 354234) unless stated otherwise. In some conditions, cells were cultured with MEFs. All media used are detailed in Table S1.

Microgel-encapsulated cells were cultured according to the 3D culture protocol. Fresh media were added every other day. Each signalling condition was screened pairwise in wells with MEFs or without MEFs.

For all the conditions, the base medium was N2B27 supplemented with 1% Matrigel, 2.5% Chemically Defined Lipids and 1% penicillin-streptomycin (Thermo Fisher Scientific, 15140122). Signalling inhibitors and activators were then added to this base medium. All supplements are detailed in Table S2.

Microgel-encapsulated cells were cultured according to the 3D culture protocol. Fresh media were added every other day. For all the conditions, the base medium was N2B27 supplemented with 1% growth factor reduced Matrigel (Corning, 356231), 2.5% Chemically Defined Lipids and 1% penicillin-streptomycin. Apart from the Epi-spheroid control, cells were cultured without MEFs. Signalling inhibitors and activators were then added to this base medium. All supplements are detailed in Table S3.

Microgel-suspension cultured cells were fixed in 4% paraformaldehyde in PBS for 10 min at room temperature and washed three times with PBS. Cells were then permeabilised (0.25% Triton X-100 in PBS) for 30 min and incubated in blocking solution (2% donkey serum, 0.1% bovine serum albumin and 0.01% Tween-20 in PBS) for 15 min. Samples were incubated with primary antibodies in blocking solution overnight at 4°C. All primary antibodies used in this study are listed in Table S4. Cells were then washed three times in blocking solution and incubated with secondary antibodies in blocking solution for 1 h at room temperature or overnight at 4°C. After another three washes in blocking solution, samples were transferred into a small drop of blocking buffer in a 35 mm dish (Ibidi, 81156) and overlaid with oil (Ibidi, 50051).

Images were acquired with an inverted Leica SP8 confocal microscope (Leica Microsystems). Fluorophores were excited with a 405 nm laser (DAPI/Hoechst), a 488 nm laser (Alexa Fluor 488), a 555 nm laser (Alexa Fluor 555) and a 638 nm laser (Alexa Fluor 647). Phase contrast pictures were acquired with an EVOS M5000 (Thermo Fisher Scientific). For each condition, a minimum of five 10X pictures were taken to capture 100 to 300 microgels per condition, except for conditions with particularly low structure formation efficiency, where 60-100 microgels were quantified (i.e. LDN no MEFs).

Images were analysed using the open-source software Fiji (Schindelin et al., 2012). Exact measurements are detailed in Table 1. Statistical tests for FIJI immunofluorescence analysis were performed in GraphPad Prism version 9.2.0 (332). A Shapiro-Wilk test was performed to determine the normality of data distribution to select appropriate tests and F-tests were used to compare variances. A two-tailed t-test with Welch's correction was used to compare two means with normally distributed samples. A two-tailed Mann–Whitney test was used to compare between two means for samples that were not normally distributed, and Kruskal–Wallis followed by Dunn's multiple comparison test was used to compare between more than two means.

For each 2D confocal imaging plane, CellProfiler version 4.0.6 (McQuin et al., 2018) was employed to segment the nuclei, using the interface from Python version 3.8.6. OrganoSeg (Borten et al., 2018) was used to segment large structures within the hydrogel, i.e. Am- or Epi-spheroid, running in Matlab version 2019a and Python version 3.6.9. In both the nuclei and structural segmentation masks, skimage version 0.18.1 in Python version 3.8.6 (Van Der Walt et al., 2014) was used to quantify the morphological features. By applying the segmentation masks over the fluorescence channels (SOX2 and TFAP2C), the fluorescence contents were quantified. The dataset was fitted from 0 to 255. Thresholds for positive signal were determined by manual assessment of representative images of control Epi- and Am-spheroids.

For each screen condition, a representative pool of 20-30 microgels were randomly selected for profiling by single-cell RNA sequencing. Structures were released from microgels by incubation with 1 U/ml Agarase (Thermo Fisher Scientific, EO0461) in N2B27 for 5 min at 37°C, followed by gentle pipetting up and down to free structures from digested agarose. Released structures were washed twice in PBS, then incubated for 10-15 min in 0.25% Trypsin-EDTA (Gibco, 25200056) until cell boundaries appeared rounded and distinct. Structures were subsequently washed in N2B27 and dissociated in a small drop of medium using blunted microcapillaries pulled to an inner diameter large enough to accommodate approximately two or three cells. Single cells were transferred immediately into 4 µl single cell lysis buffer and snap frozen.

Smart-seq2 library preparation was carried out in 96-well format as previously described (Picelli et al., 2014). Library quality was assessed using the High Sensitivity DNA Analysis Kit (Agilent, 5067-4626) on the 2100 Bioanalyzer system (Agilent). Pooled libraries were sequenced on an Illumina NovaSeq platform with a read length of PE 150 bp.

Reads were processed according to Bergmann et al. Specifically, reads were trimmed of adapter sequences using TrimGalore! (https://github.com/FelixKrueger/TrimGalore) and mapped to the common marmoset genome (Callithrix jacchus 3.2.1) using STAR (Dobin et al., 2013) aligner v2.5.4. Only samples with >100,000 mapped reads and mapping efficiency >40% were used for downstream analysis. Gene counts were quantified using FeatureCounts (Liao et al., 2014) v1.6.0 using a modified Ensembl gene annotation file (release 91) (see Bergmann et al., 2022).

Marmoset samples passing QC were analysed using Seurat (Dobin et al., 2013; Liao et al., 2014) v3.1.2. Feature counts were normalised and standardised using the NormalizeData and ScaleData function.

To summarise expression profiles for key marker genes, average expression (pseudobulk) of individual subgroups were calculated using the AverageExpression function and visualised as a heatmap using pheatmap 1.0.12. Heatmap plots were row scaled.

For visualisation, data from all in vitro models were jointly integrated within the marmoset in vivo reference datasets (Bergmann et al., 2022) using Seurat, based on Canonical Correlation Analysis (CCA) and mutual nearest neighbour (MNN) approaches. Specifically, FindIntegrationAnchors was run using 4000 features and IntegrateData (with 20 dimensions) was used to calculate corrected gene expression matrices for the three datasets. Datasets were visualised using PCA on the corrected gene expression matrix, with in vivo and in vitro datasets split to aid interpretation.

For interpretation, decision boundaries corresponding to the probability (of belonging to the embryo class) of 0.1, 0.5 and 0.9 were projected onto the PCA space in some instances. Here, decision boundaries were calculated between in vivo embryonic disc (EmDisc CS5, CS6 and CS7) and in vivo amnion (Am CS5, CS6 and CS7) based on the 2D coordinates of the first two components of the PC space using support vector machines (e1071) with a polynomial kernel (degree 3) and cost 10. Decision boundaries could be added to PC plots using the contour function of ggplot2 (Wickham, 2016).

The Pearson correlation between subgroups of cells were calculated based on corrected gene expression values following integration with Seurat using the inbuilt R function, cor. Cross correlation matrices were visualised as heatmaps using pheatmap 1.0.12.

Spatially varying genes in the embryonic disc were related to biological processes using gene ontology enrichment analysis with CytoScape (version 3.7.2) plug-ins BiNGO (Maere et al., 2005) and EnrichmentMap (Merico et al., 2010). Gene sets were derived from Gene Ontology database (Ashburner et al., 2000; Carbon et al., 2019). Enrichment maps were generated with gene sets that passed the significance threshold of P<0.005 and similarity a cutoff value of 0.7. All single, general or uninformative nodes were removed to create the final gene set interaction network.

Ternary plots between embryonic disc (CS6), amnion (CS6) and in vitro models were calculated using the ggtern package (Wickham, 2016). Additionally, ternary plots were used to visualise the change in gene expression of key markers following perturbations and visualised as projected vectors, where the base of the arrow corresponds to the expression of that gene in the control condition and the tip of the arrow the expression in the perturbation indicated.

A spatial identity score for a cell could thus be assigned to each position in the CS6 embryo for which there was a corresponding in vivo sample. Similarity scores could be interpolated to arbitrary positions within the embryo using Gaussian process regression, similarly to the interpolation of gene expression values of Bergmann et al. (2022).

Finally, to visualise the impact of a given perturbation, we could project the difference in the identity function between a treatment and its control condition, e.g. Epi-spheroids (+BMP4) versus Epi-spheroids.

We are grateful to Prof. Erika Sasaki for cmPSCs and to Charles Bradshaw for help with high performance computing. We thank the members of the Hollfelder and Boroviak labs for their enthusiasm and critical discussion of the manuscript, in particular Dylan Siriwardena.

The peer review history is available online at https://journals.biologists.com/dev/article-lookup/doi/10.1242/dev.200263.