Rapid and robust directed differentiation of mouse epiblast stem cells into definitive endoderm and forebrain organoids

Over the last two decades, methods for the directed differentiation of human PSCs (hPSCs) into numerous lineages have been developed and refined. As a result, it is now possible to generate nearly pure populations of multiple specific cell types or 3D multicellular organoids that resemble developing human tissues (Huch et al., 2017; Loh et al., 2016; Tchieu et al., 2017; Yiangou et al., 2018). However, although hPSC-based models for studying development advanced, analogous protocols for mouse PSCs have not yet emerged. Current state-of-the-art protocols for directed differentiation of mouse PSCs tend to be more complex than equivalent hPSC protocols, and they often fail to generate target cell types with high purity. Robust and efficient protocols for directed differentiation of mouse PSCs into defined lineages would be synergistic with in vivo studies in mouse and with studies of hPSC-based models of human embryonic development.

A crucial difference between most mouse and human directed differentiation protocols is the starting cell state. hPSCs are generally cultured in the primed pluripotent state, which is roughly equivalent to cells of the pluripotent epiblast around the start of gastrulation [approximately embryonic day (E) 6.5 in mouse] (Rossant and Tam, 2017). In contrast, mouse PSCs are predominantly grown in the naïve pluripotent state [mouse embryonic stem cells (ESCs)], which is equivalent to cells in the inner cell mass of the pre-implantation embryo at ∼E3.5-4.0 (Nichols and Smith, 2012). Like their embryonic equivalents, naïve PSCs must first exit the naïve pluripotent state before they can properly respond to differentiation cues (Morgani et al., 2017; Mulas et al., 2017; Smith, 2017). As a result, mouse naïve PSC directed differentiation protocols must include an additional first step of ∼48 h, during which cells exit the naïve pluripotent state and transition into an early formative and/or later primed pluripotent state that is competent to respond to signals that induce differentiation into specific somatic lineages. The requirement for this additional initial step is problematic because the naïve-to-primed transition is inherently asynchronous, leading to a heterogeneous starting population of formative/primed cells before directed differentiation has commenced (Kalkan et al., 2017; Strawbridge et al., 2020 preprint). The initial naïve-primed conversion step thus represents an impediment to the development of rapid and efficient protocols for directed differentiation that use mouse naïve PSCs as a starting population.

Mouse PSCs can be cultured in a primed pluripotent state, referred to as epiblast stem cells (EpiSCs). EpiSCs share many similarities with hPSCs, including the ability to respond rapidly to differentiation stimuli (Brons et al., 2007; Tesar et al., 2007; Vallier et al., 2009). Protocols that use EpiSCs as a starting point for directed differentiation have been developed, including (but not limited to) robust protocols for generation of oligodendrocyte progenitor cells (OPCs) and neuromesodermal progenitors (Edri et al., 2019; Najm et al., 2011). However, EpiSC-based protocols are not widely used for several reasons: (1) EpiSCs are prone to spontaneous differentiation and thus are more difficult to culture than mouse naïve PSCs or hPSCs; (2) EpiSCs exhibit significant line-to-line variability in their differentiation propensity; and (3) EpiSCs express markers of lineage-primed epiblast cells transiting through the anterior primitive streak (T, FOXA2), which may limit their potential to differentiate into some lineages (Bernemann et al., 2011; Jouneau, 2019; Kojima et al., 2014; Kurek et al., 2015; Rossant and Tam, 2017; Song et al., 2016).

More recently, several groups have demonstrated that inhibition of the canonical Wnt signaling pathway can limit spontaneous differentiation in EpiSC cultures, enabling consistent, long-term maintenance of pluripotent EpiSCs (Kurek et al., 2015; Sumi et al., 2013; Tsakiridis et al., 2014; Wu et al., 2015). Despite this progress, EpiSCs cultured in these conditions (i.e. in the presence of Wnt inhibitors; hereafter referred to as ‘+WI’) are still not commonly used for directed differentiation experiments. Therefore, whether EpiSCs (+WI) have the potential to improve directed differentiation towards multiple lineages in vitro remains a significant gap in the field.

Here, we examine the potential of EpiSCs (+WI) as a starting point for in vitro directed differentiation studies. We focused on two specific cell populations that have proven difficult to generate reproducibly and efficiently from mouse naïve PSCs: definitive endoderm (DE) and neural organoids. First, we systematically optimized conditions to differentiate EpiSCs into DE, developing a protocol that converts EpiSCs into nearly pure DE in only 40 h. Second, we show that EpiSCs can robustly and reproducibly generate forebrain-patterned neural organoids. In-depth characterization of forebrain-patterned organoids revealed that they undergo polarization into multiple distinct domains of progenitors, including the prethalamic (P3) region of the caudal forebrain (diencephalon), hippocampus and cortical hem, as well as OPCs. The prethalamic progenitors, which give rise primarily to inhibitory (GABAergic) neurons of the thalamic reticular nucleus and the zona incerta, are notable because they have not previously been efficiently generated in vitro (Shiraishi et al., 2017; Xiang et al., 2019). Our data suggest that mouse EpiSCs (+WI) could be a powerful platform for directed differentiation studies, enabling direct comparative studies with mouse embryonic development in vivo and with hPSC directed differentiation models.

Given the issues associated with using mouse naïve PSCs for directed differentiation, we sought to assess the potential of mouse EpiSCs as a starting point for directed differentiation experiments. We chose to use mouse EpiSCs cultured with Wnt inhibitors (EpiSCs +WI), which had been previously shown to significantly reduce spontaneous differentiation and allow for maintenance of pluripotency over many passages in culture (Kurek et al., 2015; Sumi et al., 2013; Wu et al., 2015). These culture conditions have several features that make them potentially useful for directed differentiation experiments. First, EpiSCs (+WI) are easy for labs that currently work with mouse naïve PSCs to acquire, as naïve PSCs can be converted into stable EpiSC (+WI) lines in vitro. Second, there is suggestive data from previous studies indicating that EpiSC culture conditions work well across a variety of genetic backgrounds (including distinct inbred mouse strains) and with both male and female cells (Brons et al., 2007; Tesar et al., 2007). In contrast, culture of naïve PSCs from most inbred mouse strains requires addition of inhibitors of FGF/ERK signaling and GSK3 (2i culture conditions), which impedes de novo DNA methyltransferase activity, leading to erosion of parent-of-origin-specific gene regulation (Choi et al., 2017; Czechanski et al., 2014; Yagi et al., 2017; Ying et al., 2008). Finally, given the similarities between EpiSCs (+WI) and hPSCs, we reasoned that it should be possible to adapt strategies developed for hPSC differentiation to EpiSCs (Vallier et al., 2009; Greber et al., 2010).

We derived EpiSCs (+WI) from 12 distinct naïve PSC lines, including lines from several distinct inbred mouse strains: C57Bl/6J (n=5), B6129SF1/J (n=3), DBA/2J (n=1) and PWK/PhJ (n=3) (Czechanski et al., 2014) (Table S4). Briefly, naïve PSCs were converted into epiblast like-cells (EpiLCs) using previously established protocols (Hayashi et al., 2011). Following EpiLC conversion, cells were split onto irradiated mouse embryonic fibroblast feeders and maintained EpiSC media containing a tankyrase inhibitor (NVP-TNKS656) to block canonical Wnt signaling (Shultz et al., 2013). Consistent with previous studies, EpiSCs (+WI) express markers of post-implantation epiblast [OCT4 (POU5F1), SOX2, OTX2, NANOG], do not express markers specifically expressed in naïve pluripotent stem cells (KLF4), and exhibit sparse expression of markers of primed epiblast (FOXA2) and primitive streak (T) (Fig. 1A,B).

To characterize these EpiSC (+WI) lines in greater depth, we measured gene expression by RNA sequencing (RNA-seq; n=12 lines) and mapped accessible chromatin regions across the genome by assay for transposase-accessible chromatin with high-throughput sequencing (ATAC-seq; n=9 lines) (Buenrostro et al., 2015). We observed high expression of primed [Otx2, Oct6 (Pou3f1), Dnmt3b] and general (Oct4, Sox2, Nanog) pluripotency markers, and low expression of naïve pluripotency markers (Klf4, Tbx3, Prdm14, Nr5a2) across all lines, in agreement with results from immunofluorescence staining (Fig. 1B,C). Similarly, genes associated with lineage commitment/lineage priming (T, Foxa2, Mixl1, Sox17, Gata6, Cdx2, Sox1) exhibited low to moderate expression across all lines (Fig. 1C). Principal component analysis (PCA) of RNA-seq and ATAC-seq data revealed that pluripotency state (naïve PSCs versus EpiSCs) is the primary driver of gene expression differences between lines, followed by genetic background (PC1; ∼56% and ∼71% of total variance, respectively, Fig. 1D,E). EpiSC lines from the same genetic background had highly correlated profiles of both gene expression and chromatin accessibility.

Next, we compared chromatin accessibility across the genome in our new EpiSC lines to cell populations in the embryo at E6.5 (epiblast, visceral endoderm, primitive streak, and germ layer progenitors) as well as EpiSC lines cultured in the presence (EpiSC-AFX) or absence (EpiSC-AF) of the tankyrase inhibitor XAV939 (Kinoshita et al., 2021; Xiang et al., 2020). PCA of chromatin accessibility profiles revealed that EpiSC lines most closely resemble EpiSC-AF and EpiSC-AFX, followed by in vivo epiblast tissue, and are least similar to primitive streak and endoderm/mesoderm (Fig. 1F). Taken together, these data suggest that EpiSCs (+WI) exhibit expected features of primed PSCs and can be stably maintained in an undifferentiated state in culture.

We next sought to test whether EpiSCs (+WI) can be directed to differentiate efficiently into DE. Current protocols for directed differentiation of hPSCs into DE are rapid and efficient, and serve as the foundation for robust protocols for generating progenitors of various endodermal organs (Yiangou et al., 2018). In contrast, protocols for DE differentiation of mouse PSCs require additional steps and give rise to heterogeneous populations of DE and mesodermal cell types (Fig. S1A, Table S1).

hPSC DE differentiation protocols consist of two steps that mimic the signals that sequentially specify DE during gastrulation (Gadue et al., 2006; Loh et al., 2014; Yiangou et al., 2018; Zorn and Wells, 2009). In the first step, Wnt and TGFβ/Nodal signaling promote exit from pluripotency and differentiation into anterior PS-like (aPS) progenitors. In the second step, high TGFβ/Nodal signaling promotes the commitment of aPS progenitors into DE. This general approach has been applied to a variety of hPSC lines and can consistently generate >85% pure DE (Loh et al., 2014). We hypothesized that signaling requirements for mouse DE differentiation from EpiSCs (+WI) would be similar to those of hPSCs, but that it was likely that the timing of each stage and the appropriate concentration of signaling factors would be different for mouse EpiSCs compared with hPSCs (Greber et al., 2010; Vallier et al., 2009)

To determine the optimal timing and concentration of each signal for aPS induction, we applied a gradient of increasing concentrations of CHIR99201 (to activate Wnt signaling via GSK3 inhibition) and activin A (to activate TGFβ/nodal signaling) to EpiSCs (DBA/2J) for either 16, 20 or 24 h (see Materials and Methods for further details). After aPS induction, all conditions were shifted to the same DE commitment conditions for an additional 24 h, and DE induction purity (fraction of the total population co-expressing FOXA2 and SOX17) was determined by immunostaining (Fig. 2A, Fig. S1B). Across all conditions, we observed DE purity ranging from 7% to 82%. The highest purity was achieved in conditions with higher activin A and the highest levels of Wnt activation. Exposure to these aPS conditions for 16 h was sufficient to enable DE commitment in the second stage of the protocol, and no further increase in purity was observed in the 20 or 24 h aPS conditions.

Having defined these improved conditions for aPS induction, we next sought to define the optimal conditions for DE commitment of aPS-like cells (Stage 2) (Fig. 2B, Fig. S1C). To promote commitment of aPS-like cells into DE, DE differentiation protocols for hPSCs add a high concentration of activin A for 48 h. Concurrent inhibition of BMP and/or Wnt signaling has also been reported to reduce off-target differentiation of aPS cells into mesodermal lineages (Loh et al., 2014). Based on these data, we tested the effects of activin A concentration, BMP inhibition (LDN-193189), Wnt inhibition (NVP-TNKS656, LGK-974), and the total duration of Stage 2 on DE commitment.

We induced aPS using the optimized conditions from Fig. 2A and examined the effects of different signaling conditions and timing (24, 36, 48 h) on DE commitment. We observed the highest purity of DE differentiation in conditions with the highest levels of activin A (∼75% of SOX17⁺ cells with 100 ng/ml versus ∼65% SOX17⁺ cells with 40 ng/ml; Fig. 2B). Inhibition of BMP signaling markedly increased DE purity across all conditions, whereas Wnt inhibition had minimal effects. DE purity was highest in the 24 h Stage 2 condition, with longer durations leading to a substantial reduction of FOXA2/SOX17-positive DE cells. In the 24 h condition, rare SOX17-negative cells were observed in small clusters of densely packed cells. Most of these off-target cells co-expressed OCT4 and T, were FOXA2 negative and lacked expression of CDX2, which is expressed by paraxial mesoderm cells that are also generated from aPS-like progenitors. These data suggest that the OCT4⁺/T⁺/FOXA2⁻ cells are most likely to be lagging aPS-like cells or immature mesodermal cells. However, in the 36 and 48 h conditions, SOX17⁻ cells could also result from properly specified DE cells that have started to adopt distinct anterior-posterior (A-P) identities that no longer express SOX17, such as anterior or posterior foregut (Li et al., 2018).

These systematic optimization experiments demonstrate that the combination of Wnt activation in the presence of moderate TGFβ/Nodal signaling, followed by TGFβ/Nodal activation and BMP inhibition can generate high purity DE from mouse EpiSCs in ∼40 h (Fig. 2C, supplementary Materials and Methods). Application of this protocol to EpiSCs (+WI) cultured in feeder-free conditions gave similar results (Fig. S3B).

To determine the purity of DE generated by our protocol more quantitatively, we used flow cytometry to measure the levels of the surface marker CXCR4 (expressed in endoderm and mesoderm) and the transcription factor (TF) SOX17 (expressed exclusively in DE) (Fig. 2D). This analysis indicated that the DE purity was ∼95% using the optimized conditions. Next, we used flow cytometry to assess the performance of our DE protocol on two additional EpiSC lines derived from distinct inbred strains of mice. The average DE purity achieved with a C57Bl/6J EpiSC line was similar to the DBA/2J line used for optimization (∼95%), whereas the purity achieved with a PWK/PhJ EpiSC line was slightly lower (∼80%) (Fig. 2E). However, for each line the results were similar between technical replicates from independent experiments (n=3-4 technical replicates/line).

To further characterize DE generated by our new protocol, we performed RNA-seq and ATAC-seq on DE generated from four independently derived C57Bl/6J EpiSC lines (characterized in Fig. 1C-E). CXCR4⁺ DE was purified with magnetic assisted cell sorting (MACS). EpiSC-derived DE exhibited large-scale transcriptional changes compared with EpiSCs, including upregulation of canonical DE marker genes (Sox17, Foxa2, Cxcr4, Gata4/6, Cer1) and downregulation of pluripotency-associated genes, including Oct4 and Nanog (Fig. 3A, Fig. S2A-D), consistent with results obtained using immunofluorescence staining (Fig. S2E) (Genga et al., 2019; Nowotschin et al., 2019). There were also widespread changes in chromatin accessibility in DE compared with EpiSCs. PCA of ATAC-seq data indicated that cell type (EpiSCs versus DE) accounted for the majority of variation observed across samples (PC1; 92.5%) (Fig. S2B). Accordingly, we observed significant increases in chromatin accessibility at numerous putative enhancer elements near genes known to be essential for DE differentiation (Foxa2, Gsc, Otx2, Sox17) (Fig. 3B). Across the genome, regions of hyper-accessible chromatin (peaks identified by ATAC-seq) in DE compared with EpiSCs were enriched for binding motifs for TFs known to play key roles in DE specification (GATA4/6, OTX2, SOX17, GSC). In contrast, ATAC-seq peaks in EpiSCs compared with DE exhibited enrichment for OCT4 and AP-1 (FOS/JUN) binding motifs (Fig. 3C-E). In summary, these genomic data demonstrate that DE exhibits global changes in gene expression and TF binding compared with EpiSCs and has the expected genomic features of bona fide PSC-derived DE.

Finally, we examined whether the DE generated by this protocol can be further patterned into specific populations of organ progenitors. We treated EpiSC-derived DE with conditions that promote posterior foregut patterning and subsequent differentiation into antral gastric progenitors (modified from protocols developed for hPSCs by Broda et al., 2019; McCracken et al., 2014) (Fig. 3F). After 72 h of additional differentiation, we observed a high density of cell clusters expressing markers of antral gastric epithelial progenitors (SOX2, PDX1, GATA4) and lacking expression of markers of midgut/hindgut (CDX2) and DE (FOXA2 and SOX17) (Fig. 3G). These data indicate that EpiSC-derived DE can be further patterned into organ progenitors and demonstrate the value of generating a high-purity population of DE for subsequent stages of differentiation.

The first neural organoid protocols were developed using embryoid bodies (EBs) formed from naïve mouse PSCs (Eiraku et al., 2008; Wataya et al., 2008). Given the success of our efforts to generate DE from EpiSCs (+WI), we next sought to determine whether EpiSCs (+WI) would also be a better starting cell type for generating neural organoids. We chose to focus our attempts on generating dorsal forebrain-patterned organoids, which have the potential to give rise to progenitors of multiple brain regions, including the cerebral cortex, hippocampus and thalamus (Montiel and Aboitiz, 2015).

Following extensive testing, we were able to identify conditions for robust EB formation, neural induction and forebrain patterning of EpiSC-derived EBs (summarized in Fig. 4A,B). Briefly, EpiSCs (+WI) were dissociated into single cells and EBs formed from ∼1000 EpiSCs in neural induction media in AggreWell plates. We found that addition of chroman 1 (ROCK inhibitor) and emricasan (pan-caspase inhibitor), components of the recently reported CEPT cocktail (Chen et al., 2021), help to dramatically reduce cell death during EB formation compared with commonly used ROCK inhibitors (Y-27632 or thiazovivin; Fig. S4A). To promote the generation of forebrain fates (telencephalon, diencephalon), we inhibited Wnt signaling by addition of the porcupine (PORCN) inhibitor LGK-974 for the first 2 days of differentiation (Rifes et al., 2020; Tchieu et al., 2017). Inhibition of FGF signaling (by the FGFR inhibitor PD173074) during this period can promote forebrain patterning, but it is not essential for forebrain specification in this system. After the first 24 h, EBs were removed from AggreWell plates and embedded in Matrigel (based on Qian et al., 2018). Under these conditions, many EpiSCs expressed SOX1 after the first 24 h of EB formation, and nearly all cells in EBs expressed SOX1 after 48 h (Fig. 4C). In contrast, mouse neural organoid protocols that start from naïve PSCs have few SOX1⁺ cells after 72 h in culture (Eiraku et al., 2008).

During early A-P patterning of the neural plate, FGF signaling promotes acquisition of more posterior neural fates, but shortly thereafter promotes telencephalic identity in the anterior-most region of the neural epithelium (∼E8.25) (Garel et al., 2003; Shimamura and Rubenstein, 1997). Therefore, we hypothesized that addition of FGF8 (FGF8b) after the first 48 h of differentiation would facilitate induction of anterior telencephalic fates and could subsequently promote the proliferation and survival of telencephalic neuroepithelial progenitors (Storm et al., 2006). By 96 h [day (d) 4], EBs formed continuous neuroepithelia, with consistent apicobasal polarity (Fig. 4D,E, Fig. S4B) (Movie 1). Characterization of the complete 3D structure of d4 organoids using tissue clearing revealed mitotic neuroepithelial progenitors at the apical membrane, a central lumen, nestin-expressing cells with characteristic radial-glial morphologies spanning the length of the apical-basal axis of the developing tissue, as well as intermediate/basal progenitors and early-born postmitotic neurons migrating outward towards the basal lamina (Fig. 4D-G). Live-imaging of sparsely labeled radial glial cells in organoids over a 24 h period (d3-d4) revealed interkinetic nuclear migration and cell division behaviors characteristic of radial glial cells in the embryonic brain (Arai and Taverna, 2017; Noctor et al., 2004) (Movie 2). After d4, organoids were removed from Matrigel and switched into maintenance media (N2B27 media with vitamin A, and the neurotrophic factors BDNF and GDNF) in low-adhesion cell culture dishes on an orbital shaker.

Next, we sought to characterize the specific regional identity of progenitors and neuronal populations generated in these organoids. At d4, all organoids expressed PAX6, which is expressed by progenitors in the caudal forebrain (diencephalon) and/or dorsal telencephalon, depending on the exact stage (Puelles et al., 2000). However, PAX6 expression was not homogeneous throughout the organoids, suggesting that not all neuroepithelial progenitors acquire the same regional identity under these conditions (Fig. 5A). By d8, most organoids spontaneously pattern into two opposing domains, which mirror the Pax6⁺ and Pax6⁻ regions at d4 (Renner et al., 2017; Takata et al., 2017) (Fig. 5B-G). Surprisingly, one domain contained large numbers of PAX6⁺ post-mitotic neurons, which occur almost exclusively in the developing prethalamus, generated from the anterior-most domain of the diencephalon (prosomere 3, p3) (Caballero et al., 2014; Manuel et al., 2015) (Fig. 5C). In the other domain, we observed neurons expressing markers of the cerebral cortex, such as TBR1, BCL11B (also known as CTIP2) and BRN2 (POU3F2), as well as RELN, a marker of Cajal–Retzius cells (Fig. 5D-G, Fig. S5). Taken together, these data suggest that organoids generated in these conditions polarize early into distinct domains that represent different A-P identities: the dorsal telencephalon and the anterior diencephalon (prosomere 3). Importantly, these results are reproducible across technical replicates and when starting with distinct EpiSC (+WI) lines (three additional EpiSC lines) (Fig. S5C,D, Table S5).

To gain further insight into the precise cell populations present within forebrain organoids, we generated single-cell (sc)RNA-seq data from d12 organoids (5652 total cells). Reduced-dimensionality visualization using uniform manifold approximation and projection (UMAP) revealed multiple distinct clusters, which we tentatively classified as specific cell types based on their expression of known marker genes. Importantly, nearly all cells within the organoid appeared to be neuroectodermal derivatives, suggesting that neural induction in EBs is highly efficient (Fig. 6, Fig. S6A, Fig. S7).

Consistent with our immunostaining results, we observed clusters that appeared to be immature (cluster 7) and more mature (cluster 8) prethalamic neurons. In the embryo, the prethalamic domain generates multiple distinct types of GABAergic neurons that form the thalamic reticular nucleus and the zona incerta (Puelles et al., 2013). Accordingly, the early prethalamic neuron cluster (cluster 7) was characterized by expression of markers associated with early development of GABAergic neurons (Ascl1, Dlx5, Dlx6, Gad1, Gad2). The more mature prethalamic neuron cluster (Cluster 8) was characterized by expression of the prethalamic neuron markers (Pax6, Isl1, Meis2, Mmp16) (Mandai et al., 2014; Nagalski et al., 2016; Ono et al., 2014; Virolainen et al., 2012) (Fig. 6E). Within cells in this cluster, there was further heterogeneity based on the relative expression levels of Ptprd and Pax6, which distinguishes the lateral or the medial portions of the developing prethalamus in vivo (Guo and Li, 2019; Kim et al., 2020; Li et al., 2020; Sommer et al., 1997; Tuttle et al., 1999). In addition, more mature prethalamic neurons express higher levels of L1cam and Nrcam, which play a role in the pathfinding of axons from prethalamic neurons that project into the thalamus (Molnar et al., 2012; Quintana-Urzainqui et al., 2020) (Fig. S6B).

We identified one cluster (Cluster 6) that appeared to correspond to the glutamatergic neurons expressing TBR1 and/or BCL11B. However, although expression of TBR1 and BCL11B is often associated with cerebral cortex development, neurons in this cluster instead expressed several markers specific to hippocampal neurons (Zbtb20, Neurod1, Prox1), and lack expression of the telencephalic marker Foxg1, which is expressed broadly in the cerebral cortex, but has low expression, or is even absent, in the hippocampus (Hatami et al., 2018) (Fig. 6F). These data suggest that these neurons more closely resemble developing hippocampal neurons than deep layer cortical neurons. The observed similarity in marker profiles between these two cell types is consistent with in vivo data showing that the developing neocortex (cerebral cortex) and allocortex (hippocampus and olfactory bulb) have similar expression profiles during embryonic development (Cadwell et al., 2019).

We also observed a cluster of cells (Cluster 3) that resemble cortical hem, and are characterized by expression of RSPO genes, Wnt8b, Wnt9a and Bmp7 (Grove et al., 1998; Hasenpusch-Theil et al., 2012) (Fig. S7). The cortical hem is a secondary organizer located next to the developing hippocampus in the dorsomedial region of the developing telencephalon (Vieira et al., 2010). In addition to its role in cortical patterning via secretion of Wnt and BMP, neural progenitors in the cortical hem also generate Cajal–Retzius cells during early stages of telencephalon development (Takiguchi-Hayashi et al., 2004). Accordingly, we observed cells bridging clusters 3 and 6 that co-expressed the Cajal–Retzius cell markers Trp73 (also known as p73) and Reln (reelin) (Hanashima et al., 2007; Simon et al., 2012) (Fig. 6F).

We also used VoxHunt (Fleck et al., 2021) to comprehensively compare our single-cell dataset to gene expression data from the developing embryonic brain (in situ hybridization data from the E18 Allen Developing Mouse Brain Atlas; Thompson et al., 2014). This analysis confirmed that the forebrain organoids most closely resemble medial pallium (hippocampus/cortical hem) and prethalamus/p3 (Fig. 6H).

In addition to these neuronal clusters, we also found clusters (clusters 4 and 5) that appear to represent OPCs, characterized by high expression of Olig1, Olig2, Pdgfra and Dll1 (Fig. 6G, Fig. S7). Among all of the clusters, OPCs appeared to be the only actively proliferating cells within organoids at d12 (Fig. 6D). Immunofluorescence staining for OLIG2 at d12 confirmed the presence of these cells, which were intermingled with BCL11B⁺ hippocampal neurons, suggesting that the same progenitors that give rise to the hippocampal glutamatergic neurons progress to produce OPCs (Fig. S5C). This is consistent with prior studies showing that mouse radial glial cells from the dorsal telencephalon give rise to OPCs after birth, and explant studies demonstrating that dorsal telencephalic progenitors from the embryonic brain often generate OPCs when explanted or cultured in vitro (Kessaris et al., 2006; Qian et al., 1997; Tekki-Kessaris et al., 2001).

Taken together, these data reveal that our EpiSC-based forebrain organoid protocol can robustly generate organoids containing multiple distinct regions of the developing forebrain, including dorsal, caudo-medial telencephalon (hippocampus and cortical hem) and dorsal anterior diencephalon (prosomere 3, prethalamus). These two domains are adjacent to one another, each bordering the boundary between the telencephalon and diencephalon. Prethalamic domains and hippocampal/cortical hem domains generally formed on opposite poles of organoids, suggesting that organoids undergo spontaneous polarization during early stages of development and develop into distinct domains representing different A-P identities (Takata et al., 2017).

Directed differentiation models enable a range of experimental techniques that would be difficult or impossible to perform in vivo, including genetic and chemical screens, live imaging, and experimental embryology approaches such as in vitro reconstitution (Hadjantonakis et al., 2020; Li et al., 2019; Moris et al., 2020; Schlissel and Li, 2020). Using mouse PSCs for directed differentiation has multiple advantages compared with hPSCs. For example, using mouse PSCs makes it possible to use the extensive genetic tools and resources that have been generated for mice. This includes not only knockout lines, reporter lines and human disease models, but also numerous distinct inbred strains and outbred heterogeneous stocks such as Diversity Outbred mice that can be used for quantitative genetic approaches (Churchill et al., 2012; Skelly et al., 2020). In addition, results from mouse PSC differentiation models can be directly compared with genetically matched tissues from mouse embryos, which allows for rigorous assessments of model validity (Voelkl et al., 2020). Such data represent an important proof of concept for hPSC-based cellular models of development and disease processes, as it will never be possible to perform similar cross-validation experiments in developing human embryos (Kleiman and Engle, 2021).

Despite the advantages of mouse directed differentiation, mouse PSC models are much less frequently used than hPSC-based models. This is likely due in part to the limited availability of mouse PSC directed differentiation protocols that are robust, efficient, generalizable and user-friendly. We found that by using EpiSCs (+WI) it is possible to develop protocols for directed differentiation that are as robust as state-of-the-art hPSC protocols. As naïve PSCs can be converted into EpiSCs (+WI) in vitro, our new protocols can be applied to generate DE and neural organoids from pre-existing naïve PSC lines that carry specific genetic modifications, including gene knockouts or fluorescent reporters. These new optimized methods for directed differentiation represent a valuable addition to the experimental toolkit available for studying mouse development and thus should be of substantial interest to the developmental and stem cell biology research communities.

By adapting protocols developed for hPSCs, we demonstrate that EpiSCs (+WI) can be differentiated into DE with high purity (>90%) in only 40 h. Our findings are consistent with previous work showing that EpiSCs and hPSCs respond similarly to most differentiation cues (Greber et al., 2010; Vallier et al., 2009). The most significant difference we observe between hPSC and EpiSC differentiation is timing. The optimal timing for EpiSCs is 16 h for aPS induction and 24 h to DE commitment, whereas hPSC protocols use 24 h for aPS induction and 48 h to DE commitment (Loh et al., 2014). These differences are consistent with the faster pace of mouse development compared with human, which has been previously observed when directly comparing mouse and human differentiation protocols (Ebisuya and Briscoe, 2018).

The high-purity DE generated by our protocol will be an ideal starting point for subsequent differentiation into progenitors of a variety of endodermal organs, such as lung, liver, intestinal tissues or organoids (Fowler et al., 2020; Yiangou et al., 2018). As a proof of concept, we demonstrated that EpiSC-derived DE can be further patterned to posterior foregut progenitors and antral gastric progenitors by adapting previously developed protocols for hPSC differentiation (Broda et al., 2019; McCracken et al., 2014). The high purity of DE achieved by this protocol should also facilitate large-scale genetic screens to identify genes that control endoderm commitment (Li et al., 2019).

The first neural organoid protocols were developed using naïve mouse PSCs (Eiraku et al., 2008; Wataya et al., 2008). Although these pioneering studies provided the foundation for the subsequent development of protocols to generate neural organoids from hPSCs, the original mouse protocols are now rarely used. This is due, at least in part, to the fact that these protocols were less robust and user-friendly than current hPSC organoid protocols. In addition, most of these early mouse neural organoid protocols pre-date the development of improved methods for consistently generating polarized neuroepithelia in neural organoids (e.g. Matrigel embedding or addition of laminin/Matrigel to the culture media) (Lancaster et al., 2013; Nasu et al., 2012). Therefore, we reasoned that by starting with EpiSCs instead of naïve PSCs, and by implementing improved techniques for organoid generation developed for hPSCs, we would be able to develop an improved protocol for the generation of mouse neural organoids.

Although our initial goal was to generate a mouse organoid model of cerebral cortex development, we instead observed domains of hippocampal and prethalamic progenitors. This suggests that our conditions for A-P patterning must be further modified to generate cerebral cortex, which comes from the most anterior region of the developing neuroectoderm. Previous work with mouse neural organoids found that insulin exerts posteriorizing effects on developing neuroectoderm via activation of FGF signaling, and it is necessary to remove insulin from the media at early stages of neural induction to maintain anterior neuroectoderm patterning (Takata et al., 2017; Wataya et al., 2008). However, this strategy is not readily adaptable to primed PSC organoid models because insulin is required for primed PSC growth and survival, making it difficult to form EBs and perform neural induction in insulin-free conditions. Based on these studies, we added FGFR inhibitors in addition to Wnt inhibitors for the first 2 days to block the posteriorizing influence of FGF signaling. However, FGFR inhibition was not strictly required for forebrain specification if Wnt signaling is sufficiently inhibited. Within forebrain organoids, we observed spontaneous polarization into telencephalic and diencephalic regions (Renner et al., 2017; Takata et al., 2017). Interestingly, for hPSC-based cerebral cortex organoids, inhibition of Wnt signaling (in addition to dual SMAD inhibition for neural induction) is generally sufficient to maintain anterior neuroectodermal identity and specification of cortical fates, even in the presence of high levels of insulin. It will be interesting to explore the mechanisms underlying this apparent difference in A-P patterning between mouse and human.

Methods for the generation of organoids containing diencephalic progenitors have previously been reported (Shiraishi et al., 2017; Xiang et al., 2019). These protocols use a combination of moderate insulin, inhibition of FGF/ERK and activation of BMP signaling (BMP7) to specify diencephalic progenitors of the thalamus and epithalamus (prosomere 2) and pretectum (prosomere 1). In contrast, we find that addition of high levels of FGF8b without exogenous BMP7 induces prethalamic (prosomere 3) fates. This is consistent with the high expression of FGF8b in the dorsal midline of p3 and prethalamic eminence at early stages of diencephalon development (E10.5-E12.5) (Martinez-Ferre and Martinez, 2009). In vivo, prethalamic progenitors generate numerous distinct types of inhibitory (GABAergic) neurons that form the thalamic reticular nucleus and the zona incerta (Li et al., 2020). Prethalamic neurons have not been generated efficiently in previous mouse and human thalamic organoids, and thus our forebrain organoid protocol should be useful for studying the development of this complex and understudied brain region.

Mouse forebrain-patterned organoids can complement in vivo mouse models of forebrain development as well as neurodevelopmental disorders (Sestan and State, 2018). Our neural organoid model should facilitate direct comparisons between the phenotypes observed in vivo in mice with mutations associated with neurodevelopmental disorders, and those observed in analogous mouse organoid models in vitro. In contrast, hPSC-based models of neurodevelopmental disorders cannot be conclusively validated in vivo, and thus it is difficult to evaluate the relevance of observations made in neural organoid models in vitro for developmental pathophysiology in vivo. More rigorous assessment of the strengths and limitations of in vitro models in mice would help to inform experimental design for hPSC-based modeling of neurodevelopmental disorders. In future studies, it will be interesting to perform in-depth comparisons of developing neural organoids and their analogous cell types in vivo, and to compare mouse and human organoid development.

Although we were able to develop robust protocols for differentiation of EpiSCs (+WI) into DE and neural organoids, it is possible that EpiSCs (+WI) will not be an ideal starting point for differentiation into all cell types.

For some protocols, EpiLCs or more recently described stable formative stage stem cells might be superior to EpiSCs (+WI) (Hayashi et al., 2011; Kinoshita et al., 2021; Morgani et al., 2018; Yu et al., 2021). Even when cultured with Wnt inhibition, we observe some heterogeneity in EpiSC cultures, marked by expression of anterior primitive streak markers Foxa2 and T (Bernemann et al., 2011; Blauwkamp et al., 2012; Song et al., 2016; Tsakiridis et al., 2014). Additional optimization of EpiSC (+WI) conditions could help to develop conditions that are as robust as hPSCs and further mitigate issues with current culture conditions, such as residual heterogeneity and the need for fibroblast feeder cells. In this study, we exclusively used EpiSCs (+WI) that we derived in vitro via conversion of naïve mouse PSCs. Previous work suggests that EpiSCs derived in vitro are highly similar to EpiSC lines derived directly from post-implantation epiblast (Kojima et al., 2014), so we would expect cells from both sources to behave similarly, but we have not demonstrated that this is the case. There is an emerging consensus in the hPSC directed differentiation field that the initial culture conditions of hPSCs have a significant impact on their ability to differentiate into specific lineages (Cornacchia et al., 2019; Watanabe et al., 2019 preprint). Thus, it is likely that further improvements of EpiSC (+WI) culture will be the best way to generally improve their performance in directed differentiation experiments. The impact of long-term culture on genetic and epigenetic stability of EpiSCs (+WI) has not been characterized in depth (Halliwell et al., 2020). Similarly, the long-term effects of tankyrase inhibition on the genetic and epigenetic stability of EpiSCs remain to be characterized. Several types of Wnt inhibitors have been used for EpiSC culture, and it remains uncertain how different means of inhibiting Wnt activity might change the properties of the resulting EpiSCs (Kurek et al., 2015; Sumi et al., 2013; Wu et al., 2015).

Step-by-step protocols for directed differentiation experiments are provided as supplementary Materials and Methods. Detailed information about the reagents used can be found in Table S3.

All mouse ESC lines were maintained on gelatin-coated dishes with irradiated mouse embryonic fibroblast feeder cells using serum/LIF media composed of DMEM (high glucose, GlutaMAX, HEPES), 1% nonessential amino acids, 1% sodium pyruvate, 1% penicillin-streptomycin, 0.1% 2-mercaptoethanol, 10% fetal bovine serum and 1000 units/ml ESGRO LIF. All ESC lines (except lines from B6129SF1/J background) were cultured in serum/LIF media with 2i containing 3 µM CHIR99201 and 1 µM PD0325901. Media was changed daily, and ESCs were passaged upon 70% confluence at a 1:6 ratio using TrypLE. ESC-to-EpiLC conversion was performed as described by Morgani et al. (2018). Briefly, ESCs were lifted using 0.5 µ/µl collagenase IV, centrifuged (200 g for 3 min at room temperature) and dissociated into a single-cell solution using Accutase. The ESC suspension was plated on fibronectin-coated plates (16.7 µg/ml) at a density of 17,500 cells/cm² in N2B27 media supplemented with 12.5 ng/ml heat-stable recombinant human bFGF, 20 ng/ml activin A, and 1% Knockout Serum Replacement (Gibco, 1082028). N2B27 media consists of 50% DMEM-F12, 50% Neurobasal medium, 0.5% N2 supplement, 1% B27 supplement without vitamin A, 2 mM GlutaMAX, 1% penicillin-streptomycin and 0.1% 2-mercaptoethanol. The media was changed after 24 h, and cells were converted to EpiSCs after 48 h. For EpiLC-to-EpiSC conversion, EpiLCs were dissociated into small clumps (approximately three to five cells) with Accutase and plated at a density of ∼50,000-100,000 cells per cm² on irradiated mouse fibroblast feeder cells in N2B27 supplemented with 20 ng/ml activin A, 12.5 ng/ml heat-stable bFGF and Wnt inhibitor (175 nM NVP-TNKS656). EpiSC media was changed daily, and cells were passaged every ∼48 h at a 1:6 ratio using 0.5 µ/µl collagenase IV followed by dissociation with Accutase into small clumps of three to five cells. For feeder-free EpiSC culture, EpiSCs growing on feeders were isolated using 0.5 μ/μl collagenase IV and seeded on fibronectin-coated plates (16.7 µg/ml). After the first feeder-free passage, cells were passaged using Accutase only. The EpiSCs were considered as feeder-free after two passages with no visible feeder cell remaining.

For DE differentiation, EpiSCs were generally used after approximately four to eight passages in EpiSC conditions. For differentiations, high-quality cultures of EpiSCs were detached using collagenase IV (0.5 µ/μl), washed once with PBS, and then dissociated into a single-cell suspension using Accutase. EpiSCs were plated at a density of 110,000 cells/cm² in chemically defined media (CDM) containing 50% IMDM, 50% Ham's F12 Nutrient Mix with GlutaMAX, 1% chemically defined lipid concentrate, 450 µM monothioglycerol, 1% polyvinyl alcohol (w/v), 15 µg/ml Apo-transferrin, 0.5% GlutaMAX, 0.7 µg/ml insulin and supplemented with 20 ng/ml activin A, 12.5 ng/ml heat-stable bFGF, 175 nM NVP-TNKS656, 1% Knockout Serum Replacement, and 2 µM thiazovivin. Plating media was removed 6 h after seeding, cells were washed gently with PBS without calcium or magnesium (PBS-/-) and the media was changed to CDM supplemented with 3 µM CHIR99201 and 40 ng/ml activin A. Sixteen hours after the first media change, cells were washed with PBS-/- and the media was replaced with CDM supplemented with 100 ng/ml activin A and 100 nM LDN-193189. After 24 h (46 total hours after cells were initially seeded), cells were fixed for immunostaining or collected for further analysis. For antral gastric progenitor differentiation, DE cells were kept in CDM with 100 nM LDN-193189 and 2% fetal bovine serum, with a renewal of the media after 24 h. After 48 h, the media was changed to CDM with 100 nM LDN-193189, 2 µM retinoic acid and 2% fetal bovine serum for a further 24 h. After 72 h, cells were fixed for immunostaining.

For immunostaining of adherent cells (EpiSCs and DE), cells were rinsed twice with PBS-/- and fixed with 4% paraformaldehyde solution for 30 min at room temperature. After fixing, cells were washed twice with PBS-/- and permeabilized with PBS-/- containing 1% Triton X-100 (PBST) for 10 min at room temperature and blocked for 30 min with PBS-/- containing 0.3% Triton X-100 and 5% fetal bovine serum. Primary antibodies were diluted in PBST with 1% fetal bovine serum and incubated overnight at 4°C. Detailed information about the antibodies used can be found in Table S2. Following overnight incubation and three 10 min washes with PBST, cells were incubated with secondary antibodies for 2 h at room temperature. Cells were rinsed with two 10 min washes of PBST before DAPI staining (1 µM) for 15 min at room temperature. Cells were rinsed with PBST and imaged using a Leica Dmi8 microscope.

All image quantification was performed using Cell Profiler version 4.2.1 to calculate efficiency of DE differentiations based on FOXA2 and SOX17 expression. DAPI-stained nuclei were identified using the ‘Identify Objects’ module. To identify nuclei stained with each antibody, segmented nuclei were further analyzed using the ‘Identify Objects’ module and the ‘Relate Objects’ Module. These data were then classified as either positive or negative for each marker using the ‘Filter Objects’ module and overall efficiency was calculated based on percentage of expression for each marker.

Following DE differentiation, cells were dissociated in Accutase and washed in FACS buffer consisting of PBS-/-, 20% fetal bovine serum and 0.1% 0.5 M EDTA. CXCR4 antibody conjugated to Alexa Fluor 647 and live/dead zombie UV staining were diluted in FACS buffer at the manufacturer's recommended concentration (Thermo Fisher) and incubated for 15 min at room temperature. Cells were washed and resuspended in FACS buffer with fixation/permeabilization diluent and fixation/permeabilization concentrate for 30 min at room temperature. The cells were then resuspended in permeabilization buffer with SOX17 antibody conjugated to Alexa Fluor 488 at the manufacturer's recommended concentration for an additional 30 min at room temperature. Fixed cells were then analyzed using a Cytek Aurora cell sorter.

Flow cytometry results were analyzed using FlowJo software v10.8 and single-color stainings were used as a negative control for gating purposes.

DE cells were isolated for ATAC-seq and RNA-seq analyses using MACS following the manufacturer's instructions, with slight modifications. Cells were dissociated in Accutase, washed in FACS buffer, counted, and resuspended with rat anti-mouse CD184/CXCR4 Alexa Fluor 647-conjugated antibody at a concentration of 1 µl/10⁶ cells and incubated for 1 h at 4°C. Cells were then washed with FACS buffer, centrifuged (200 g for 3 min at 4°C) and resuspended in a solution composed by 80% FACS buffer and 20% anti-rat IgG MicroBeads (Miltenyi Biotec, 130-048-501). Cells were incubated in the mixture for 15 min at 4°C, then washed and resuspended in 500 µl buffer. The MS column was placed in an OctoMACS separator (Miltenyi Biotec, 130-048-501) on the MACS multistand. The cell suspension was added to the column and rinsed three times with buffer. Finally, the column was removed from the separator and the magnetically labeled cells were collected.

Phase separation in cells lysed in TRIzol was induced with 50% isopropanol and 0.5% 2-mercaptoethanol and RNA was extracted from the aqueous phase using the MagMAX mirVana Total RNA Isolation Kit (Thermo Fisher, A27828) on a KingFisher Flex Magnetic Particle Processor according to the manufacturer's protocol with 1-2 million cells input. Samples were eluted in 38 µl elution buffer.

After RiboGreen quantification and quality control using an Agilent BioAnalyzer, 100-500 ng of total RNA with RIN values of 9.5-10 underwent polyA selection and TruSeq library preparation according to instructions provided by Illumina (TruSeq Stranded mRNA LT Kit), with eight cycles of PCR. Samples were barcoded and run on a NovaSeq 6000 in a PE100 run, using the NovaSeq 6000 S4 Reagent Kit (200 Cycles) (Illumina). An average of 36 million paired reads were generated per sample and the percentage of mRNA bases per sample ranged from 84% to 89%.

After RiboGreen quantification and quality control using an Agilent BioAnalyzer, 2 ng total RNA with RNA integrity numbers ranging from 7.7 to 10 underwent amplification using the SMART-Seq v4 Ultra Low Input RNA Kit, with 12 cycles of amplification. Subsequently, 10 ng of amplified cDNA was used to prepare libraries with the KAPA Hyper Prep Kit using eight cycles of PCR. Samples were barcoded and run on a NovaSeq 6000 in a PE100 run, using the NovaSeq 6000 S4 Reagent Kit (200 Cycles) (Illumina). An average of 39 million paired reads were generated per sample and the percentage of mRNA bases per sample ranged from 83% to 87%.

ATAC-seq was performed as previously described (Buenrostro et al., 2015; see also dx.doi.org/10.17504/protocols.io.bv9mn946 at https://www.protocols.io/view/atac-sequencing-protocol-dm6gpwmpplzp/v1) with minor modifications. Briefly, cells were dissociated into single cells, filtered through a Flowmi cell strainer and the nuclei were isolated by incubation with lysis buffer [10 mM Tris-HCl pH 7.4, 10 mM NaCl, 3 mM MgCl₂, 0.1% Tween-20, 0.1% NP-40, 0.01% digitonin and 1% bovine serum albumin (BSA)] for 5 min at 4°C; 20,000 and 40,000 nuclei were isolated, and the DNA was tagmented with Tn5 (Illumina) and amplified using NEBNext® High-Fidelity 2X PCR Master Mix (NEB). The number of cycles was estimated by qPCR. DNA tagmentation efficacy was evaluated with a Bioanalyzer 2100 (Agilent Technologies) and the DNA amounts calculated with Qubit. i5 and i7 primer sequences were obtained from Mezger et al. (2018). The resulting DNA libraries were sequenced using the NextSeq550 system (Illumina) and about 25 million reads were obtained per sample in duplicate.

Resulting fastq files were mapped to their transcripts and quantified using Salmon (Patro et al., 2017) against the genome version mm10. For downstream analyses, Rstudio was used with the packages DESeq2, pcaExplorer and pheatmap (Love et al., 2014; Marini and Binder, 2019). Statistical significance was calculated using a Wald test and corrected using the Benjamini–Hochberg method.

The raw fast data were analyzed using BasePair software (https://www.basepairtech.com/) with a pipeline that included the following steps. The raw reads were trimmed using fastp to remove low-quality bases from reads (quality<20) and adapter sequences. The trimmed reads were aligned using Bowtie2 (Langmead and Salzberg, 2012) to UCSC genome assembly mm10. Duplicate reads were removed using Sambamba. Peaks were identified with MACS2 (Gaspar, 2018 preprint) and those overlapping with satellite repeat regions were discarded. For further analyses, a union peak atlas was created from the MACS2 files. Peak intensity for each sample was counted using featureCounts (Liao et al., 2014). HOMER (v4.11) (Heinz et al., 2010) was used for motif analyses. PCA was carried out using the R packages DESeq2 (Love et al., 2014) and pcaExplorer (Marini and Binder, 2019).

For motif enrichment analysis, ATAC-seq peaks in the atlas were associated with TF motifs in the CIS-BP database (Weirauch et al., 2014) using FIMO (Grant et al., 2011) of MEME suite (Bailey et al., 2009), under the P-value cutoff of 1e−4. We limited the motif analyses to motifs for 236 TFs that are expressed in EpiSCs or DE (RPKM>5). Normalized counts were obtained using variance-stabilizing transformation in DESeq2. The shift in the cumulative distribution of normalized counts was compared between the subset of the atlas containing the TF motif and the total atlas using a one-sided Kolmogorov–Smirnov test.

For forebrain organoid generation, B6129SF1/J EpiSC colonies were detached using collagenase IV (0.5 µ/μl), washed once with PBS-/-, and then dissociated into a single-cell suspension using Accutase. EpiSCs were then seeded at 1000 cells/microwell in an AggreWell plate in EB formation media containing 50 nM chroman-1, 5 μM emricasan, 400 nM LDN-193189, 10 μM SB-431542, 100 nM PD173074 and 4 nM LGK-974 in N2B27 (B27 without vitamin A) media. After 24 h, the EBs were recovered and transferred to the same media but without chroman-1 and emricasan and embedded in Matrigel as previously described (Qian et al., 2018). Briefly, 66.7 μl of EBs and media were mixed with 100 µl of Matrigel using wide-bore tips. The mix was added to a non-adherent 6-well plate without touching the edges of the well and incubated at 37°C for 30 min. After 30 min, 3 ml of warm media were added on top. At day 2, the media was changed to plain N2B27 (B27 without vitamin A) with 100 ng/ml Fgf8b and kept for 48 h more. Day 4 EBs were recovered from Matrigel using cell recovery solution (Corning) and incubating for 30 min at 4°C. After day 4, the EBs were transferred to a Petri dish and kept in N2B27 (B27 supplement with vitamin A) with BDNF and GDNF while shaking at 65 rpm (Infors HT, Celltron benchtop shaker). For sparse labeling of neuronal progenitor cells with CytoTune EmGFP Sendai fluorescence reporter, 7.5 μl were added to the 2 ml media in the AggreWell plate at d0 and kept until d1. The organoids were then imaged for ∼24 h from d3 to d4.

For organoid staining, the organoids were collected in tubes pre-coated with 1% BSA. The samples were washed twice with PBS-/- and fixed with 4% paraformaldehyde for 45 min at 4°C. Afterwards, they were washed twice with PBS-/- and permeabilized with 0.5% Triton X-100 for 15 min at 4°C. Blocking was performed for 15 min in organoid washing solution (OWS; adapted from Dekkers et al., 2019), consisting of 0.2% Triton X-100, 0.02% SDS and 0.2% BSA in PBS-/-) at 4°C. Next, the samples were incubated in OWS with primary antibodies overnight at 4°C. On the following day, they were washed three times for 2 h each with OWB while shaking, then incubated overnight at 4°C with secondary antibodies and DAPI (1 µM) in OWB. The next day, the organoids were washed three times with OWB for 6 h total while shaking. Finally, the organoids were mounted in DeepClear solution (CelExplorer Labs). The samples were imaged using a confocal Nikon A1RHD25 or Leica SP8 confocal microscope and analyzed with Imaris software. For live imaging, the organoids were embedded in phenol-free Matrigel and imaged using the TrueLive3D Imager system (Luxendo).

Multiple organoids were collected and mixed at d12, washed with PBS-/- and dissociated with Accutase until a single-cell suspension was achieved. Then, the sample was washed once again with PBS-/-. The resulting cell suspension was used for 10x scRNA-seq (10x Genomics, Single Cell 3′ Kit v3.1, dual index) following the manufacturer's directions; 326 M reads were obtained. The resulting fastq files were processed with Cell Ranger (10x genomics cloud). CellBender was used to eliminate background reads and other artifacts (Fleming et al., 2019 preprint), and any cell with more than 10% mitochondrial reads was excluded. The Seurat package (v4) was used for downstream analyses (Hao et al., 2021). Cluster annotation was performed using the cited literature and the Allen Developing Mouse Brain Atlas, stages E15.5 and E18.5. VoxHunt analyses were performed as described by Fleck et al. (2021).

This study would not have been possible without the core facilities at MSKCC. We would like to specifically thank Dr Yas Furuta (Mouse Genetics Core facility, MSKCC), Murray Tipping and Vitaly Boyko (Molecular Cytology Core, MSKCC), Dr Richard Koche (Epigenetics Innovation Lab, MSKCC), and members of the Integrated Genomics Operation Core (MSKCC) for their advice and assistance with experiments and data analysis throughout the course of this project. In addition, we would like to thank Dr James Muller (Director of Developmental Biology Program's Microscope Cluster) for assistance with light-sheet microscopy and advance about imaging organoids and Dr Ronan Chaligne (Single Cell Research Initiative, MSKCC) for advice and assistance with single cell RNA-sequencing analysis. We would like to thank Dr Laura Reinholdt and Anne Czechanski (The Jackson Laboratory) for sharing PWK/PhJ ESC lines, Dr Christopher Baker and Dr Candice Byers (The Jackson Laboratory) for sharing C57Bl/6J ESC lines, and Dr Matthias Stadtfeld (Weill Cornell Medical College) for sharing B6129SF1/J ESC lines. We would like to thank our colleagues Dr Anna Katerina Hadjantonakis, Dr Danwei Huangfu, Dr Lorenz Studer, Dr Nan Yang, Dr Matthias Stadtfeld, Dr Eftychia Apostolou, Dr Sophie Morgani, Dr Clayton Schwarz, Dr Gabriele Ciceri, Dr Ryan Walsh, Bess Rosen (Huangfu Lab), Renhe Luo (Huangfu lab), Dapeng Yang (Huangfu Lab) for helpful discussions and advice throughout the course of the study. Figure panels 2C, 3F and 4A were created with BioRender.com. We would like to thank Dr Nan Yang (Mount Sinai), Dr Matthias Stadtfeld, Dr Timothy Cherry for helpful comments on the manuscript.

The peer review history is available online at https://journals.biologists.com/dev/article-lookup/doi/10.1242/dev.200561.