Efficient and scalable generation of primordial germ cells in 2D culture using basement membrane extract overlay

Summary Current methods to generate human primordial germ cell-like cells (hPGCLCs) from human pluripotent stem cells (hPSCs) can be inefficient, and it is challenging to generate sufficient hPGCLCs to optimize in vitro gametogenesis. We present a differentiation method that uses diluted basement membrane extract (BMEx) and low BMP4 concentration to efficiently induce hPGCLC differentiation in scalable 2D cell culture. We show that BMEx overlay potentiated BMP/SMAD signaling, induced lumenogenesis, and increased expression of key hPGCLC-progenitor markers such as TFAP2A and EOMES. hPGCLCs that were generated using the BMEx overlay method were able to upregulate more mature germ cell markers, such as DAZL and DDX4, in human fetal ovary reconstitution culture. These findings highlight the importance of BMEx during hPGCLC differentiation and demonstrate the potential of the BMEx overlay method to interrogate the formation of PGCs and amnion in humans, as well as to investigate the next steps to achieve in vitro gametogenesis.


Correspondence lopes@lumc.nl
In brief Overeem, Chang, et al. demonstrate an important role for extracellular matrix in human primordial germ cell-like cell (hPGCLC) differentiation from pluripotent stem cells. With combined treatment with basement membrane extract (BMEx) and BMP4, hPGCLCs form in standard 2D culture, yielding an efficient differentiation method that greatly facilitates in vitro gametogenesis research.

INTRODUCTION
In mammals, gametogenesis is a complex and long process that is initiated by the specification and lineage restriction of primordial germ cells (PGCs), the founding population of the gametes. 1 Recapitulating (female and male) gametogenesis in vitro would enable modeling of infertility-causing diseases and may ultimately lead to new assisted-reproduction techniques.
In mice, Bmp4 was identified as a crucial morphogen, inducing PGC specification in the posterior-proximal epiblast. 2 Acting through the intracellular factors Smad1/5/9, Bmp4 is able to upregulate Tbxt (Brachyury or T) as well as a specific gene regula-tory network that includes Prdm1, Prdm14, and Tfap2c. 3 This knowledge has led to the recapitulation of mouse PGC-like cell (PGCLC) formation in vitro by exposure of mouse pluripotent stem cells (mPSCs) grown as embryoid bodies (EBs) to BMP4. 3 Subsequently, human PGCLCs (hPGCLCs) have been generated from human PSCs (hPSCs) using a similar approach, 4,5 highlighting the high degree of conservation between human and mouse regarding PGC specification but also uncovering differences in the specification mechanisms 6 as well as differences in their molecular signatures. 7 The origin of PGCs in mice and humans in vivo may also differ. In mice, specified PGCs are located in the posterior-proximal epiblast at the base of the allantois shortly after the onset of gastrulation. Although it remains unknown when and where exactly PGC specification takes place in humans, in cynomolgus monkey embryos, PGCs were first observed in the amnion prior to gastrulation. 8 In contrast to mice and pigs, which undergo amniogenesis by folding, humans and non-human primates undergo amniogenesis by cavitation instead 9,10 ; therefore, amnion and PGCs may share a similar origin in primates. In agreement, hPGCLCs share a common TFAP2A+ progenitor with amnion ectoderm-like cells in EB differentiation assays, 11 and hPGCLC formation has been demonstrated in an amniotic sac embryoid model. 12,13 The most widely used directed differentiation protocols to derive hPGCLCs from hPSCs include EB aggregation and treatment with high concentrations of BMP4. However, while these EB-based methods were instrumental in understanding hPGCLC formation, 4,5 they can be characterized by low efficiency and high variability on a per-hPSC line basis. 14- 16 Reported hPGCLC yields ranged from 5% to 60%, but for the majority of hPSC lines, hPGCLC differentiation efficiencies are below 10%. In addition, EB differentiation is low throughput, laborious, and requires harsh and stressful cell dissociation. As a result, efficient hPGCLC generation for high-throughput downstream experiments aiming at optimizing human gametogenesis in vitro remains challenging.
In this study, we have uncovered a critical role of the extracellular matrix (ECM) during hPGCLC differentiation from hPSCs in 2D culture. We show that the addition of basement membrane extract (BMEx) and BMP4 at a concentration as low as 10 ng/ mL to an ordinary 2D cell culture format is sufficient to consistently generate hPGCLCs at high yields, ranging between 30% and 50%, within 5 days of differentiation. The hPGCLCs in this 2D system originated from a TFAP2A+CDX2+GATA3+ EOMES+ progenitor population that also gave rise to amniotic ectoderm-like and presumably amniotic mesoderm-like cells. Importantly, the presented hPGCLC differentiation method is highly scalable and cost effective, which will greatly facilitate progress achieving human in vitro gametogenesis (IVG).

Robust generation of hPGCLCs in 2D culture with BMEx overlay
The application of diluted BMEx on 2D plated hPSCs (BMEx overlay) has been shown to be important to induce lumen formation (lumenogenesis), 17 enabling that system to model aspects of early human embryogenesis. 18 As hPGCLCs formed readily in the amniotic sac embryoid model, which consisted of BMP4-treated hPSC spheres cultured on a microfluidic device, 12 we hypothesized that BMEx-supplemented culture may facilitate the generation of hPGCLCs in regular 2D culture formats. To test this, single-cell-passaged human induced PSCs (hiPSCs) were plated in mTeSR-plus medium supplemented with 2% BMEx (day 0) ( Figure 1A). One day later (day 1), the medium was replaced with previously described hPGCLC induction medium 19 containing 200 ng/mL BMP4 as well as stem cell factor (SCF), leukemia inhibitory factor (LIF), and epidermal growth factor (EGF). The differentiation was carried out in hPGCLC-induction medium for 4 days (days 1-5), with the medium in the initial 2 days (days 1-3) supplemented with 2% BMEx ( Figure 1A).
Pronounced morphological changes were observed when hiPSCs were differentiated with BMEx overlay ( Figure 1B). In contrast to the flat colonies observed in the absence of BMEx overlay, tightly packed colonies were present with BMEx overlay. Immunofluorescence on day 5 of differentiation revealed a large number of ITGA6+POU5F1+SOX17+ hPGCLCs only in the BMEx overlay condition across three hPSC lines, M54, F99, and H1 ( Figure 1C), in addition to the expression of other known PGC markers, such as TFAP2C, PDPN, PRDM1, and ALPL ( Figure 1D). In agreement, flow cytometry analysis using PGC markers ITGA6 and EPCAM 5,20 revealed that the hPGCLC generation efficiency was about 50% in line M54, whereas basically no hPGCLCs were detected in the absence of BMEx overlay ( Figure 1E), revealing a critical role for the cell-ECM interaction during hPGCLC differentiation.

Optimization of BMEx overlay differentiation method
The response of hPSCs to BMP4 signaling is highly dependent on both culture format and cell density. 21,22 The hPGCLC induction medium contained a high dose of BMP4 (200 ng/mL), which was optimized for EB-based methods. To establish the optimal BMP4 dosage in our 2D system, we tested different concentrations of BMP4 while removing SCF, LIF, and EGF from day 1 to 3 and reducing the concentration of BMP4 to 10 ng/mL from day 3 to 5 ( Figure 2A). Strikingly, we observed comparable efficiencies to induce (ITGA6+EPCAM+) hPGCLCs with vastly reduced BMP4 concentrations in four independent hPSC lines (Figures 2B, S1A). Using immunofluorescence, we further confirmed an associated increase in POU5F1+SOX17+ hPGCLCs ( Figure 2C).
In previous work using EB-based differentiation, we identified the lines F20 and M72 as inefficient hPGCLC-generating lines. 14 We observed the same in BMEx overlay differentiation, with F20 yielding 15% and M72 0.3%, respectively ( Figure S1A). This suggested that variance in hPGCLC generation efficiency could be an inherent cell line property independent of the differentiation method used. 16 It was previously demonstrated that activin A (ActA)/NODAL induced hPGCLC differentiation competency in hPSCs. 19 Moreover, the addition of a low dose of ActA together with BMP4 improved the specification of hPGCLCs in micropatterned colonies. 23 Hence, we tested whether exogenous ActA could increase induction of hPGCLCs in our system ( Figure 2D). We observed that simultaneous treatment with BMP4 and ActA from day 1 to 3 lowered the hPGCLCs' yield in all tested concentrations compared with treatment with BMP4 alone ( Figure 2D); shortening the ActA treatment to 1-2 days gave a similarly poor outcome ( Figure S1B). Interestingly, inhibiting endogenous transforming growth factor b (TGF-b)/ActA signaling by blocking the receptor type I (ALK4/ACVR1B, ALK5/TGFBR1, ALK7/ACVR1C) using SB431542 reduced hPGCLC formation ( Figures 2D, S1B).
BMEx is a biologically complex product of animal origin that is highly prone to variabilities between batches and manufacturers.
To determine the robustness of using BMEx for hPGCLC differentiation, we tested multiple lots of three commercially available stem cell-grade BMEx products. Geltrex and Cultrex consistently induced hPGCLC formation with about 50% efficiency  Figures 2E and 2F). Surprisingly, Matrigel only had a minor hPGCLC-inducing effect ( Figures 2E and 2F). This outcome was not due to differences in total protein concentrations (Figure S1C). Moreover, increasing the percentage of Matrigel during differentiation to up to 3.5% had no effect on differentiation efficiency ( Figure S1D).
Efficient PGCLC differentiation is accompanied by parallel induction of amniotic ectoderm-like and mesoderm-like cells To identify the cell types present in our BMEx overlay model during differentiation (2% BMEx overlay from day 0 to 3 and 10 ng/ mL BMP4 from day 1 to 5), we performed single-cell transcrip-tomics of two PGCLC-efficient lines (M54 and F99) and two PGCLC-inefficient lines (F20 and M72) at days 0, 2, and 5 ( Figures 3A and S2A).
As expected, Cl3 (PGCLCs) was comprised almost exclusively of cells derived from the efficient PGCLC-generating PSC lines  M54 and F99 and a very small fraction from F20 ( Figure S2A). Intriguingly, at day 0, the hiPSCs formed two separate clusters: Cl0 included cells from M54, F99, and F20, whereas Cl1 included cells from M72 and F20 ( Figure S2A). This suggested that F20 consisted of two subpopulations with distinct transcriptomes, one similar to M72 and the other to M54 and F99. This is consistent with the observation that, unlike M72, F20 could generate 15% PGCLCs ( Figure S1A). We performed differential expression analysis between Cl0 and Cl1 and observed that Cl1 showed higher levels of UTF1, NODAL, PRAC1, and SIX3, whereas Cl0 cells expressed SFRP1, PCLAF, RAB17, and TAGLN ( Figure S2D). We speculate that these genes, in particular NODAL, may be used as potential markers to distinguish efficient from inefficient hPGCLC-generating hPSCs, although characterization of additional hPSC lines will be required to verify this.
To compare the developmental timeline and cell types generated using our BMEx overlay method with the ''conventional'' EB differentiation method, we merged our single-cell dataset with the single-cell dataset generated by Chen and colleagues 11 using the ''conventional'' EB differentiation method 5 (Figures 3D, S2E, and S2F). The molecular signatures were largely similar, and the three endpoint cell types PGCLCs, AELCs, and AMLCs from the BMEx overlay method mapped on to the PGCLC, amnion-like cells and extraembryonic mesenchyme (EXMC) from the Chen dataset. A small population of human endoderm-like cells (part of Cl2) now formed an independent cluster together with cells previously identified as primitive endoderm-like cells ( Figures 3D and S2F). Hence, we demonstrated that despite the different culture formats and pre-treatment step, both differentiation methods generate the same cell types, and in both methods, PGCLCs are formed alongside amnion-like cells.
The observation that PGCLCs arise alongside amniotic cells has also been made when using the amniotic sac embryoid system (mPASE). 13 To compare the cell types generated by BMEx overlay differentiation and the mPASE method, we merged our data with single-cell transcriptomics data generated by Zheng and colleagues 13 (Figures S2G and S2H). As expected, our PGCLCs clustered together with the PGCLCs in the mPASE, expressing NANOS3 and PRDM1 ( Figure S2H). Our AELCs clustered together with amniotic ectoderm-like cells named AMLCs (AMLC2s)in the mPASE, expressing TFAP2A, ISL1, and GABRP ( Figure S2H). Finally, our AMLCs clustered together with mesoderm-like cells (MeLC1s) in the mPASE, expressing GATA6, PDGFRA, FOXF1, and SNAI2 ( Figure S2H).
Finally, using human amnion from 9 weeks of gestation (WG9), we confirmed by whole-mount immunofluorescence the expression of TFAP2A in the amniotic ectoderm and GATA6 and PDGFRA in amniotic mesoderm ( Figure 3G).
The transcriptome of PGCLCs is similar to that of Carnegie stage 7 human PGCs To compare the transcriptome of our PGCLCs and amnion-like cells with that of their in vivo counterparts, we merged our single-cell RNA sequencing (RNA-seq) dataset with an available single-cell RNA-seq dataset from a Carnegie stage 7 (CS7) human embryo. 24 UMAP visualization of the merged datasets showed that the cell types in vitro clustered with the corresponding in vivo counterparts ( Figure 4A). The day 0 PSCs clustered with epiblast cells, whilst PGCLCs clustered with PGCs marked by NANOS3 and POU5F1 ( Figure 4B). Moreover, AELCs clustered with amniotic ectoderm, both expressing TFAP2A, ISL1, KRT7, and GATA3 ( Figures 4A and 4B), whereas AMLCs clustered with advanced mesoderm, expressing GATA6 and PDFGRA ( Figures 4A and 4B). Since amniotic ectoderm cells are present in the dataset of the human embryo, but the extraembryonic mesoderm cells covering the amniotic ectoderm are missing from the annotation, it is likely that the authors annotated those (and perhaps the related mesodermal cells forming the connecting stalk) as advanced mesoderm.
Next, we compared the PGCLCs generated with the BMEx overlay method with more mature human fetal germ cells (FGCs) by merging our dataset with available single-cell RNAseq data from first and second trimester human fetal gonads. 25 Visualization by UMAP showed that PGCLCs clustered with migratory and mitotic FGCs, which expressed PGC markers POU5F1 and NANOS3 ( Figures 4C and 4D). In contrast to FGCs, PGCLCs did not express more mature FGC markers such as DDX4, DAZL, and SYCP3 ( Figure 4D), confirming that PGCLCs were similar to pre-migratory PGCs. Comparing the expression of a set of known PGC and germ cell markers between in-vitro-generated PGCLCs, CS7 PGCs, and FGCs indicated that PGCLCs, regardless of the differentiation method used, were most similar to CS7 PGCs ( Figure 4E). Strikingly, PGCLCs showed lower expression of KIT, DMRT1, and DPPA3 than CS7 PGCs ( Figure 4E), suggesting that PGCLCs may be less mature than CS7 PGCs.

Lumenogenesis and PGCLC differentiation are independent events
A particular feature of BMEx overlay culture is the formation of lumen-containing structures. 17 Since we observed distinctive morphology resembling tube/lumen structures in the BMEx overlay method, we investigated whether lumenogenesis is linked to the differentiation of PGCLCs. We were able to detect laminin deposition on top of formed luminal structures at day 2 of differentiation with BMEx overlay ( Figure 5A). By contrast, in the absence of the BMEx overlay, cells remained as a single-cell  Figure 5A). Moreover, we observed the expression of basal-lateral markers ITGB1 and CTNNB1, apical marker PODXL, and tight-junction marker TJP1 (Figures 5A, 5B, and S3A), confirming lumenogenesis at days 1-2. At day 3, the lumen expanded, and SOX17+/PRDM1+/PDPN+ PGCLCs were visible adjacent to the lumen (Figures 5C and 5D). At day 5, the lumens lost structural integrity, and a large number of SOX17+ PGCLCs could be observed (Figures 5D and S3B). In contrast to TPAP2A+ AELCs that expressed a clear rim of TJP1, SOX17+ PGCLCs only showed a focal accumulation of TJP1 ( Figure S3B).
Next, we varied the period of the initial plating step (mTesRplus +2% BMEx) from 24 to 0 h (cells plated directly in RB27 + 10 ng/mL BMP4 + 2% BMEx) (Figures 5E-5G) to investigate whether PGCLC differentiation depended on the timing of lumen formation. Although the initial plating step was necessary to obtain PGCLC differentiation, a 3 h plating step was sufficient to obtain robust differentiation to PGCLCs using two different lines, M54 ( Figure 5F) and F99 ( Figure S3C). Interestingly, independently of the duration of the initial plating step (between 0 and 24 h), small lumens marked by PODXL+TJP1+ apical membrane domains were observed by day 2 ( Figure 5G). In the absence of BMEx overlay, we observed cellular polarization with the formation of a clear TJP1+ apical rim but no lumen formation ( Figure S3D).
To further test whether lumen maintenance at day 2 was necessary for PGCLC differentiation, we disrupted the lumens at day 2 by dissociating and replating the cells, followed by analysis at day 5 ( Figure 5H). Despite the disruption of lumens at day 2, immunofluorescence revealed the formation of TFAP2C+/ SOX17+/POU5F1+ PGCLCs ( Figure 5H). In conclusion, lumenogenesis and PGCLC differentiation appeared to be two independent events, and only exposure to BMEx (for a period as short as 3 h) prior to BMP4+BMEx treatment (for 2 days) was essential for PGCLC differentiation.
BMEx overlay potentiates BMP4 signaling in PGCLC progenitors at day 2 We performed differential expression analysis between the PSCs at day 0 (Cl0 and Cl1) and the progenitor population at day 2 (Cl5) and observed that from day 0 to 2, DPPA4 and SOX2 were downregulated, whereas many BMP responsive genes such as ID1, ID3, GATA3, TFAP2A, and MSX2 were upregulated ( Figures 6A and S4A). In addition, TFAP2A was expressed in the progenitor cells at days 2-3 but not in PRDM1+/ SOX17+ PGCLCs at day 5 ( Figure 6B). Interestingly, in the absence of BMEx overlay, TFAP2A was basically absent at day 2 ( Figure S4B), whereas GATA3 showed comparable levels with or without BMEx overlay at days 2-3 ( Figures 6C and S4C).
To test whether BMP4 signaling was influenced by BMEx overlay, we examined the levels of phosphorylated (p)SMAD1/5/9 ( Figures 6C, S4C). Even though both culture conditions (with and without BMEx overlay) contained 10 ng/mL BMP4, the fluorescence intensity of nuclear pSMAD1/5/9 was higher in the presence of BMEx overlay, in particular at day 2 ( Figures 6C and S4C).
In addition to TFAP2A and GATA3, CDX2 and EOMES were also identified as markers of PGCLC progenitors in EB differentiation 11 and the BMEx overlay method ( Figure S4D). In agreement, similarly to TFAP2A, both CDX2 and EOMES were upregulated at days 2-3 only in the presence of BMEx overlay ( Figures 6D, S4E, and S4F). EOMES was previously shown to be activated by ActA/NODAL signaling during PGCLC differentiation and to be essential for initiating the PGCLC transcriptional network. 6,19 However, we observed that the addition of exogenous ActA was detrimental for PGCLC differentiation in the BMEx overlay method ( Figure 2D). To understand this discrepancy, we quantified the expression of EOMES and TFAP2A in the common progenitor population at day 2 in the presence or absence of BMEx overlay and after treatment with 10 ng/mL ActA or inhibition of endogenous TGF-b/ActA signaling using 10 mM SB431452 (Figures 6D and 6E).
Compared with the absence of BMEx overlay, the day 2 progenitors cultured with BMEx overlay upregulated both TFAP2A and EOMES and downregulated SOX2 in line F99 ( Figure 6D) and in lines F20, M72, and F31 ( Figure S4F). When treated with 10 ng/ mL BMP4 and 10 ng/mL ActA in the presence of BMEx overlay, day 2 progenitors upregulated EOMES considerably, whereas inhibition of endogenous TGF-b/ActA signaling blocked EOMES expression ( Figures 6D and 6E), indicating that EOMES is strongly regulated by ActA signaling in our culture system. Interestingly, treatment with a combination of BMP4 and ActA from day 1 to 3 with BMEx overlay resulted at day 5 in the induction of SOX17+-FOXA2+ cells ( Figure S4G), presumably endoderm, which is consistent with the role of ActA and its target EOMES in endoderm differentiation. 26,27 Blocking endogenous TGF-b/ActA signaling from day 1 to 3, on the other hand, resulted in generation of mainly amniotic ectoderm cells at day 5, expressing markers such as TFAP2A, KRT7, HAND1, and GATA3 ( Figure S4H).
In conclusion, in our optimized PGCLC differentiation method, the addition of BMEx overlay between days 0 and 3 resulted in faster downregulation of SOX2, increased pSMAD1/5/9 signaling, and increased expression of TFAP2A, CDX2, and EOMES ( Figure 6F). This led to the formation of nascent PGCLCs at day 3, with downregulation of TFAP2A and CDX2 and upregulation of NANOS3 by day 5, which make up about 50% of the cells in culture, alongside amniotic ectoderm-and mesoderm-like cells ( Figure 6F).
In vitro maturation of hPGCLCs by co-culturing with human fetal ovary cells The transcriptome of hPGCLCs resembles pre-migratory PGCs that still lack expression of DDX4 and DAZL ( Figure 4E). Previous studies have shown that hPGCLCs co-cultured with mouse fetal Article ll OPEN ACCESS ovary or testis cells resulted in DDX4+SYCP3+ oogonia-like cells or DDX4+MAGEA3+ prospermatogonia-like cells, respectively, after 120 days of culture. 28,29 To investigate whether hPGCLCs generated with the BMEx overlay method have the potential to mature further, we co-cultured hPGCLCs with cells isolated from human fetal ovaries ( Figure 7A). To track the hPGCLCs in the co-culture, we generated a female POU5F1::EGFP reporter hiPSC line by fusing EGFP to the C terminus of POU5F1 using  30 This method is based on CRISPR-Cas9 nickases instead of nucleases and allows for the isolation of POU5F1::EGFP-tagged hiPSCs while minimizing endogenous POU5F1 disruption 31 and off-target EGFP tag insertions. 32 Next, we differentiated the POU5F1::EGFP reporter line into hPGCLCs using the BMEx overlay method and used fluorescence-activated cell sorting (FACS) to isolate POU5F1EGFP+/ ITGA6+/EPCAM+ hPGCLCs from day 5 culture ( Figure 7B). Subsequently, we aggregated the POU5F1::EGFP+/ITGA6+/ EPCAM+ hPGCLCs with dissociated WG19 fetal ovarian cells and cultured them for 3 days in ultra-low attachment 96-well V bottom plates before embedding in agarose droplets ( Figure 7A). Live imaging of the aggregates at day 3 showed that EGFP+ hPGCLCs spread evenly in the aggregates (Figure 7C).
At day 25 of culture, as the aggregate increased in size, EGFP+ hPGCLCs were still present in the aggregate but concentrated in specific regions ( Figure 7C) and expressed hPGC markers such as POU5F1 and SOX17 ( Figure 7D). Moreover, some EGFP+ cells started expressing more mature germ cell markers, such as DDX4 and DAZL ( Figure 7D), similar to WG19 FGCs in vivo ( Figure 7E). In conclusion, we have shown that hPGCLCs generated with the BMEx overlay method are capable of maturing to DDX4+/DAZL+ germ cells when co-cultured with human fetal ovarian cells.

DISCUSSION
The current methods to generate hPGCLCs in vitro have drawbacks regarding efficiency and scalability. As a consequence, progress regarding the differentiation of hPGCLCs into more mature germ cells in vitro has been hampered. We report a new hPGCLC differentiation method that is efficient, simple, and cost effective in a highly scalable 2D format. This new method will contribute to accelerate the progression of human IVG research, and as such, we were able to demonstrate that hPGCLCs generated using the BMEx overlay method matured into DDX4+/DAZL+ germ cells when co-cultured with fetal ovary cells. Compared with a previous study that used reconstitution with mouse somatic niche to mature hPGCLCs in 77 days, 29 we observed upregulation of DDX4 in hPGCLCs by day 25 after reconstitution with human fetal ovarian cells.
There have been several previous reports on the generation of hPGCLCs in 2D culture. Differentiation of hiPSCs grown as small micropatterned colonies enabled generation of hPGCLCs with up to 70% efficiency. 23 Interestingly, substantial changes in cell-cell interaction and cell morphology took place in both the micropatterned and our BMEx overlay cultures compared with regular 2D culture. As such, these changes might be linked to a shared mechanism that enabled hPGCLC differentiation in both systems. While both methods showed high hPGCLC differentiation efficiency, the BMEx overlay method does not require manufacture of a specialized cell culture surface and is therefore less technically demanding. A second 2D hPGCLC differentiation method relies on WNT inhibition to improve hPGCLC specification, achieving around 20%-30% differentiation efficiency. 33 The recent single-cell transcriptomics dataset of a single gastrulating CS7 human embryo, containing both amnion and hPGCs, is a tremendous resource for comparing in-vitro-differentiated cells with their in vivo counterparts. 10,24 We were able to verify that PGCLCs and AELCs showed similar transcriptomes to the PGCs and amniotic ectoderm in the human embryo, respectively. Interestingly, the AMLCs clustered together with PDGFRA+/GATA6+ mesodermal cells, annotated as advanced mesoderm in the Tyser dataset. Since the amniotic mesoderm annotation is missing in the Tyser dataset, but considering that that cell population must be present, as those cells are in close contact with the (annotated) amniotic ectoderm, we suggest that cell type may have been labeled as advanced mesoderm. In support of this, we showed by immunofluorescence that human amnion at WG9 consists of TFAP2A+ amniotic ectoderm and PDGFRA+/GATA6+ amniotic mesoderm. To univocally reveal the molecular signature of the extraembryonic mesoderm covering the amniotic ectoderm will require the generation of new single-cell RNA-seq (scRNA-seq) datasets from additional human embryos, including the annotation and further validation of the cell types that form the amnion as well as other extraembryonic structures.
The application of the BMEx overlay primes hPSCs to gain competency to efficiently differentiate to hPGCLCs. Priming hPSCs for 3 h was sufficient, and we report that (some component[s] in) BMEx acted directly and quickly to potentiate BMP signaling via pSMAD1/5/9. The ECM components of BMEx may directly interact with BMP4 such as in Drosophila, where BMP4 homolog Dpp binds to collagen type IV, mediating BMP signaling. 34 Alternatively, cell-ECM interactions may change the availability and activity of the BMP receptors. For example, ECM-integrin interactions reorder membrane into caveolaerich lipid raft domains, 35 which is where BMP reeceptor type I (BMPRI) receptors are typically localized, affecting their activity. [36][37][38][39] Finally, integrins activate a multitude of downstream pathways that could result in crosstalk with BMP/SMAD signaling.
Downstream of BMP4, BMEx-treated hPSCs showed increased pSMAD1/5/9, leading to upregulation of GATA3, TFAP2A, CDX2, and indirectly of EOMES. EOMES is essential for hPGCLC formation, 6,19 but its continuous and high expression has been shown to promote differentiation to endoderm. 26,27 Consistent with this, EOMES is moderately expressed in day 2 progenitor cells exposed to the BMEx overlay. Moreover, in agreement with EOMES being a direct target of TGF-b/ ActA signaling, the addition of ActA increases EOMES expression, resulting in a reduction in hPGCLC yield and a shift toward differentiation to SOX17+/FOXA2+ endoderm-like cells. Surprisingly, in the absence of BMEx, the day 2 progenitors fail to upregulate EOMES. This may explain why 3D differentiation methods have relied on ActA/CHIR99021 pre-induction, which results in EOMES expression.
Using the BMEx overlay method, we were able to differentiate hPGCLCs robustly from most hiPSC lines tested, but two lines showed consistently lower differentiation efficiency. A previous analysis of hiPSC lines from 317 individuals showed that hiPSCs display different levels of expression of genes, such as GATA4, GATA6, EOMES, CER1, and NODAL. 40 While endogenous NODAL signaling is required for hPGCLC differentiation, increased NODAL signaling (by adding ActA) severely hampered hPGCLC differentiation. In agreement, we observed that the two inefficient hiPSC lines F20 and M72 have higher levels of NODAL than the efficient lines F99 and M54. Although additional iPSC lines need to be investigated, our observations suggest that hPSCs lines with low levels of endogenous NODAL may have higher capacity to undergo hPGCLC differentiation.
The observation that BMEx potentiates BMP/SMAD signaling has significance beyond the field of IVG. BMP4 is widely used in various differentiation protocols and models of early embryogenesis. 12,13,41,42 Moreover, the presence of BMEx has proven to be beneficial for the development of somite-like structures in a mouse 3D gastruloid stem cell model 43 as well as for developing a non-human primate peri-implantation assay. 44 Hence, the combination of BMEx and treatment with BMP4 may prove beneficial to mimic in vivo processes more accurately in human models of early embryogenesis.

Limitations of the study
We established that a BMEx overlay method resulted in robust differentiation of hPGCLCs for the majority of the tested hPSC lines, including the commonly used ESC line H1. However, hPSC lines F20 and M72 were characterized by low hPGCLC yields. The mechanism for this line-dependent variability remains unclear, and users of the presented method will have to test hPSC lines for compatibility. In addition, the presented method is reliant on BMEx isolated from murine Engelbreth-Holm-Swarm (EHS) tumor, which is a complex mix of biologically active compounds that may influence differentiation outcome, including trace amounts of growth factors. Hence, the method presented is neither chemically defined nor clinical grade. For this purpose, the BMEx components that play a role in hPGCLC induction need to be determined.

STAR+METHODS
Detailed methods are provided in the online version of this paper and include the following:

ACKNOWLEDGMENTS
We would like to thank the patients who donated the human fetal material used in this study as well as the staff of the Gynaikon Clinic in Rotterdam and het Vrelingshuis in Utrecht. We thank X. Fan for technical assistance; members of the Chuva de Sousa Lopes group, T. den Hamer, and V. Ramovs for fruitful discussions; and M. Bellin for providing the hiPSC line F20. This work was sup-

DECLARATION OF INTERESTS
The authors declare no competing interests.

RESOURCE AVAILABILITY
Lead contact Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact Susana M. Chuva de Sousa Lopes (lopes@lumc.nl).

Materials availability
This study did not generate new unique materials or reagents.
Data and code availability d Single-cell RNA-seq data have been deposited at GEO and are publicly available as of the date of publication. Accession numbers are listed in the key resources table. d The code used here is available at https://github.com/johnmous/single_cellhPGCLCs. All original code has also been deposited at Zenodo and is publicly available as of the date of publication. DOI is listed in the key resources table. d Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Human samples and ethics statement
All experiments performed in this study were carried out strictly under the guidelines specified in the Declaration of Helsinki for Medical Research involving Human Subjects. For ethics approval, a letter of no objection was issued by the Medical Ethical Committee of Leiden University Medical Center (B21.054).
The human amnion and fetal ovary samples used were collected from elective abortions without medical indication, after obtaining informed consent from the donors. The amnion (2 cm 3 2 cm fragment) was dissected in 0.9% NaCl solution (Fresenius Kabi), fixed in 4% paraformaldehyde (PFA) (Sigma) overnight (o/n) at 4 C washed three time in PBS, and transferred to 70% ethanol for storage at 4 C until further use.

METHOD DETAILS
2D hPGCLC differentiation High quality hiPSCs of 60-80% confluency with minimal differentiation were used for hPGCLC differentiation. Briefly, cells were dissociated with TryPLE (Thermo Fisher Scientific) at 37 C for 5 min (min), diluted in DMEM/F12 (Thermo Fisher Scientific) to stop digestion, and spun down. Single cells were resuspended in cold mTeSR-Plus media containing RevitaCell supplement (Thermo Fisher Scientific) and 2% Geltrex (LDEV-free, hESC-Qualified, Reduced Growth Factor) or Cultrex (Stem Cell Qualified Reduced