Ethics statement

Human embryo work was regulated by the Human Fertility and Embryology Authority under licence R0193. Approval was obtained from the Human Biology Research Ethics Committee at the University of Cambridge (reference HBREC.2021.26). All work is compliant with the 2021 International Society for Stem Cell Research (ISSCR) guidelines. Patients undergoing IVF at CARE Fertility, Bourn Hall Fertility Clinic, Herts & Essex Fertility Clinic, and King’s Fertility were given the option of continued storage, disposal or donation of embryos to research (including research project specific information) or training at the end of their treatment. Patients were offered counselling, received no financial benefit and could withdraw their participation at any time until the embryo had been used for research. Research consent for donated embryos was obtained from both gamete providers. Embryos were not cultured beyond day 14 post-fertilization or the appearance of the primitive streak. Human stem cell work was approved by the UK Stem Cell Bank Steering Committee (under approval SCSC21-38) and adheres to the regulations of the UK Code of Practice for the Use of Human Stem Cell Lines. Mice were kept in an animal house in individually ventilated housing on 12:12 h light–dark cycle with ad libitum access to food and water. Ambient temperature was maintained at 21–22 °C and humidity at 50%. Experiments with mice are regulated by the Animals (Scientific Procedures) Act 1986 Amendment Regulations 2012 and carried out following ethical review by the University of Cambridge Animal Welfare and Ethical Review Body. Experiments were approved by the Home Office under licences 70/8864 and PP3370287. CD1 wild-type males aged 6–45 weeks and CD1 wild-type females aged 6–18 weeks were used for this study. Animals were inspected daily, and those showing health concerns were culled by cervical dislocation.

Sequencing analysis and code availability

Raw fastq files from human datasets26,27,36,45, cynomolgus monkey datasets28,35 and mouse datasets68,69,70,71 were obtained from public repositories with wget. All human datasets were aligned to the GRCh38 reference using kb-python’s kb ref function to generate a reference. For cynomolgus monkey, National Center for Biotechnology Information (NCBI) genome build 5.0 transcriptome fasta files were adjusted to Ensembl style and used in kb ref to generate a custom index. For the mouse, GRCm39 reference was used with kb ref to generate a custom index. All datasets were re-aligned using either kb-python or kallisto72,73, after data handling as below. Human datasets: 10x v2 data from Molè et al. were processed as previously described10. For Zhou et al.27, read1 files were trimmed using cutadapt74 for the reported adapter sequence. Trimmed reads were then aligned using the kb-python kb count function with custom specifications (-x 1,0,8:1,8,16:0,0,0) and the custom barcode whitelist available. Each pair of fastqs was processed individually into barcode–gene matrices and concatenated. For Xiang et al.26, a batch file was generated with cell ID, read1 and read2 for each fastq pair listed. Kallisto’s pseudo –quant command was then used to generate a cell ID–gene matrix. For Blakely et al.36, reads were aligned using kallisto pseudo –quant. For Petropoulos et al.45, single-end reads were processed with kallisto pseudo –quant with a pre-made batch file as above with 43 base pair read length specified. Cynomolgus datasets: for Ma et al.28, read1 fastqs were trimmed using cutadapt for TSO and polyA tail as described in the original publication. Next, kb python’s kb count function was used with custom specifications (–x 1,0,8:1,8,16:0,0,0). For Yang et al.35, reads were aligned using kb python’s kb count command with 10xv3 technology specified. For Nakamura et al.21, available count tables were used given the use of SOLiD sequencer limiting re-alignment program options. Mouse datasets: for Mohammed et al.70, kallisto pseudo –quant with a generated batch file was used to generate a cell ID–gene matrix. For Deng et al.69 and Cheng et al.68, single-end reads were aligned with kallisto pseudo –quant. Finally, for Pijuan-Sala et al.71, each sample set of 33 fastq files was aligned with kb count, with 10xv1 technology specified. The resulting set of barcode–gene matrices was then concatenated for downstream analysis.

Following re-alignment, any datasets not generated using unique molecular identifier counts were normalized using quminorm75. First, matrices were converted to transcripts per kilobase million (TPM), and then the TPM matrix ran through quminorm with a shape parameter up to a maximum of 2 that did not create not available/applicable (NA) values in the matrix. Then, each individual dataset was made into a Seurat object76. Each individual dataset was then merged into a species-specific Seurat object, with SCT batch correction applied across datasets. Clusters were identified on the basis of canonical marker expression. To perform module scoring, gene lists were obtained from rWikiPathways77. For the monkey and mouse, gene symbols were converted to human homologues using bioMart78. Seurat’s AddModuleScore function was used with WikiPathway gene lists of interests as input. For CellPhoneDB analysis41, human data were split on the basis of stage, and subset matrix and metadata for cell type were output as txt files. CellPhoneDB was then run with respective files and –counts-data set to gene_name. Data visualization was performed using Seurat’s DimPlot, FeaturePlot and VlnPlot functions, Scillus’ (https://scillus.netlify.app) Plot_Measure function, pheatmap and CellPhoneDB’s dotplot function.

The scripts used for analyses are available at ref. 79.

Thawing and in vitro culture of human embryos

Human embryos were thawed and cultured as described previously10,24. Briefly, cryopreserved human blastocysts (day 5 or 6) were thawed using the Kitazato thaw kit (VT8202-2, Hunter Scientific) according to the manufacturer’s instructions. The day before thawing, TS solution was placed at 37 °C overnight. The next day, IVF straws were submerged in 1 ml pre-warmed TS for 1 min. Embryos were then transferred to DS for 3 min, WS1 for 5 min and WS2 for 1 min. These steps were performed in reproplates (REPROPLATE, Hunter Scientific) using a STRIPPER micropipette (Origio). Embryos were incubated at 37 °C and 5% CO2 in normoxia and in pre-equilibrated human IVC1 supplemented with 50 ng ml−1 insulin growth factor-1 (IGF1) (78078, STEMCELL Technologies) under mineral oil for 1–4 h to allow for recovery. Following thaw, blastocysts were briefly treated with acidic Tyrode’s solution (T1788, Sigma) to remove the zona pellucida and placed in pre-equilibrated human IVC1 in eight-well µ-slide tissue culture plates (80826, Ibidi) in approximately 400 µl volume per embryo per well. Half medium changes were done every 24 h. For small-molecule experiments, human IVC1 was supplemented with either 2 µM A83-01 (72022, STEMCELL Technologies)80,81, 25 ng ml−1 Activin-A (Qk001, QKINE)82,83,84, 200 nM LDN (S2618, SelleckChem)85,86, 50 ng ml−1 BMP6 (SRP3017, Sigma Aldrich)85,86, 20 µM DAPT (72082, STEMCELL Technologies)87,88,89,90, 10 µM Compound-E (ab142164, Abcam)91,92,93, 20 µM MK-0752 (S2660, Selleck Chemicals)94,95,96 or dimethyl sulfoxide (DMSO) for 48 h. In all cases, these concentrations fall within a range of those used for either vertebrate embryos or complex human ES cell-derived models of the embryo. Within these ranges, a low-to-intermediate concentration was selected to avoid non-specific cytotoxic effects while still considering the higher concentration needed for embryo permeation compared with minimal 2D cell culture to achieve inhibitor action. Further, all small molecules and proteins were tested on human ES cells to validate the efficacy and test for cytotoxicity. For analysis, embryos were fixed in 4% paraformaldehyde for 20 min at room temperature for downstream analysis.

Recovery of mouse embryos and in vitro culture

Pregnant, time-staged mice were culled by cervical dislocation, and uteri were dissected and placed in M2 medium (pre-warmed if embryos were for in vitro culture, ice cold if for fixing). E3.5 blastocysts were flushed out of uteri of pregnant females and either fixed for immunofluorescence analysis or transferred to acidic Tyrode’s solution for zona pellucida removal. Embryos were cultured for 48 h in CMRL (11530037, Thermo Fisher Scientific) supplemented with 1× B27 (17504001, Thermo Fisher Scientific), 1× N2 (made in-house), 1× penicillin–streptomycin (15140122, Thermo Fisher Scientific), 1× GlutaMAX (35050-038, Thermo Fisher Scientific), 1× sodium pyruvate (11360039, Thermo Fisher Scientific), 1× essential amino acids (11130-036, Thermo Fisher Scientific), 1× non-essential amino acids (11140-035, Thermo Fisher Scientific) and 1.8 mM glucose (G8644, Sigma) supplemented with 20% foetal bovine serum5,28. Embryos were incubated with 25 ng ml−1 Activin-A, 200 nM LDN, 50 ng ml−1 BMP6, 20 µM DAPT or DMSO for 48 h. For E4.5, E5.5 and E5.75 collections, embryos were dissected directly from the uteri and fixed for analysis. For E5.0 collection, embryos were dissected from the uteri, and Reichert’s membrane was removed before culturing or 36 h with relevant small molecules as described above.

Human ES cell culture

Shef6 human ES cells (R-05-031, UK Stem Cell Bank) were routinely cultured on 1.6% v/v Matrigel (354230, Corning) in mTeSR1 medium (85850, STEMCELL Technologies) at 37 °C and 5% CO2. Cells were passaged every 3–5 days with TrypLE Express Enzyme (12604-021, Thermo Fisher Scientific). The ROCK inhibitor Y-27632 (72304, STEMCELL Technologies) was added for 24 h after passaging. Cells were routinely tested for mycoplasma contamination by polymerase chain reaction. To convert primed human ES cells to RSeT or PXGL naive conditions, cells were passaged onto mitomycin-C inactivated CF-1 MEFs (3 × 103 cells cm−2; GSC-6101G, Amsbio) in human ES cell medium containing Dulbecco’s modified Eagle medium (DMEM)/F12 supplemented with 20% Knockout Serum Replacement (10828010, Thermo Fisher Scientific), 100 µM β-mercaptoethanol (31350-010, Thermo Fisher Scientific), 1× GlutaMAX (35050061, Thermo Fisher Scientific), 1× non-essential amino acids, 1× penicillin–streptomycin and 10 ng ml−1 FGF2 (University of Cambridge, Department of Biochemistry) and 10 µM ROCK inhibitor Y-27632 (72304, STEMCELL Technologies). For RSeT conversion, cells were switched to RSeT medium (05978, STEMCELL Technologies). Cells were maintained in RSeT and passaged as above every 4–6 days. For PXGL conversion, previously described protocols were used97. Briefly, cells were cultured in hypoxia and medium was switched to chemically Resetting Media 1 (cRM-1), which consists of N2B27 supplemented with 1 µM PD0325901 (University of Cambridge, Stem Cell Institute), 10 ng ml−1 human recombinant LIF (300-05, PeproTech) and 1 mM valproic acid. N2B27 contains 1:1 DMEM/F12 and Neurobasal A (10888-0222, Thermo Fisher Scientific) supplemented with 0.5× B27 (10889-038, Thermo Fisher Scientific) and 0.5× N2 (made in-house), 100 µM β-mercaptoethanol, 1× GlutaMAX and 1× penicillin–streptomycin. cRM-1 was changed every 48 h for 4 days. Subsequently, medium was changed to PXGL–N2B27 supplemented with 1 µM PD0325901, 10 ng ml−1 human recombinant LIF, 2 µM Gö6983 (2285, Tocris) and 2 µM XAV939 (X3004, Merck). PXGL cells were passaged every 4–6 days using TrypLE (12604013, Thermo Fisher Scientific) for 3 min, and 10 µM ROCK inhibitor Y-27632 and 1 µl cm−2 Geltrex (A1413201, Thermo Fisher Scientific) were added at passage for 24 h.

For small-molecule experiments, primed or PXGL human ES cells were plated into ibiTreat dishes at normal passage densities. Forty-eight hours after passage, medium was changed to N2B27 supplemented with 25 ng ml−1 Activin-A, 2 µM A83-01, 50 ng ml−1 BMP6, 200 nM LDN or 20 µM DAPT. Plates were then fixed for 20 min in 4% paraformaldehyde for downstream analysis. For 3D culture of primed human ES cells, 30,000 cells were resuspended in 200 µl of ice-cold Geltrex and the resulting mix was plated into a single well of an 8 µ-well ibiTreat dish. Geltrex was polymerized by placement at 37 °C for 10 min. Two-hundred microlitres of mTeSR1 with ROCK inhibitor Y-27632 was added after polymerization. Twenty-four hours later, the medium was changed to N2B27 (±10 µM DAPT). Medium was refreshed 24 h later, and the plate was fixed in 4% paraformaldehyde for 30 min after a total of 48 h in experimental conditions. Conditioned medium experiments were performed as described previously48. Briefly, 80 µl of ice-cold Geltrex was added to an 8 µ-well ibiTreat dish to create a 100% Geltrex bed. This was polymerized at 37 °C for 4 min. A total of 1 × 103 cells cm−2 primed human ES cells were then added onto this bed in DMEM/F12 and allowed to settle for 15 min. After this, medium was carefully switched to conditioned medium (described below) with 5% Geltrex (v/v) and 10 µM ROCK inhibitor Y-27632. Conditioned medium with 5% Geltrex was refreshed daily for the next 2 days, and the resulting spheroids were fixed after a total of 72 h.

Human YSLC culture

YSLC differentiation was carried out as published48. Briefly, Shef6 human ES cells cultured in RSeT medium for at least 2 weeks were plated onto ibiTreat dishes at 1 × 103 cells cm−2 in RSeT medium with 10 µM Y-27632. Medium was changed the next day to ACL differentiation medium consisting of N2B27 supplemented with 5% v/v Knockout Serum Replacement, 100 ng ml−1 Activin-A, 3 µM CHIR99021 (University of Cambridge Stem Cell Institute) and 10 ng ml−1 human recombinant LIF. Medium was refreshed every 48 h, and 2 µM A83-01, 200 nM LDN or 20 µM DAPT was added to ACL medium for 48 h from either day 2 to day 4, followed by fixation, or day 4 to day 6 followed by fixation. For conditioned medium experiments, at day 6 cells were washed three times with phosphate-buffered saline and then mTeSR Plus medium (100-0276; STEMCELL Technologies) was added for 24 h. Medium was collected from YSLCs and passed through a 0.45-µm filter (16555, Sartorious), and stored for up to 1 week at 4 °C.

Mouse ES cell culture

CD1 mouse ES cells (generous gift from Prof. Jennifer Nichols (Stem Cell Institute, University of Cambridge, UK)) were routinely cultured on gelatin-coated (G7765, Sigma Aldrich) dishes in N2B27 supplemented with 1 µm PD0325901, 3 µm CHIR99021 and 10 ng ml−1 mouse Lif (University of Cambridge, Stem Cell Institute). Medium was changed every 48 h, and cells were passaged every 3–5 days using trypsin–ethylenediaminetetraacetic acid (25300062; Life Technologies). For experiments, cells were passaged as normal into ibiTreat dishes. The following day, medium was switched to either N2B27 + 2iLif, N2B27, or N2B27 + 200 nM LDN. Medium was refreshed after 24 h, and cells were fixed after 48 h.

Immunostaining

Embryos were fixed in 4% paraformaldehyde, permeabilized in 0.1 M glycine with 0.3% Triton X-100 and placed in blocking buffer containing 1% bovine serum albumin and 10% foetal bovine serum. Primary antibodies were diluted in blocking buffer and added overnight at 4 °C. Fluorescently tagged secondary antibodies were added for 2 h at room temperature. Primary antibodies used in this study are as follows: mouse monoclonal anti OCT3/4 (sc5279, Santa Cruz; 1:200 dilution), rat monoclonal anti SOX2 (14-19811-82, Thermo Fisher Scientific; 1:500 dilution), goat polyclonal anti NANOG (AF1997 R&D Systems; 1:500 dilution), rabbit monoclonal anti GATA6 (5851, Cell Signaling Technology; 1:2,000 dilution), goat polyclonal anti GATA6 (AF1700, R&D Systems; 1:200 dilution), mouse anti monoclonal Cdx2 (MU392-UC, Biogenex; 1:200 dilution), goat polyclonal anti CER1 (AF1075, R&D Systems; 1:250 dilution), rat monoclonal anti Cerebus1 (MAB1986, R&D Systems; 1:200 dilution), rabbit monoclonal anti Phospho-Smad1(Ser463/465)/Smad5(Ser463/465)/Smad9(Ser465/467) (13820T, Cell Signaling Technology; 1:200 dilution), rabbit monoclonal anti Smad2.3 (8685T, Cell Signaling Technology; 1:200 dilution), rabbit monoclonal anti-cleaved caspase 3 (9664, Cell Signaling Technology; 1:200 dilution), mouse monoclonal anti Podocalyxin (MAB1658, R&D Systems; 1:500 dilution), goat polyclonal anti Brachyury (AF2085, R&D Systems; 1:500 dilution), rat monoclonal anti GATA4 (14-9980-82, Thermo Fisher Scientific; 1:500 dilution), goat polyclonal anti AP2-gamma (AF5059, R&D Systems; 1:500 dilution), goat polyclonal anti Otx2 (AF1979, R&D Systems; 1:1,000 dilution) and Alexa Flour 594 Phalloidin (A12381, Thermo Fisher Scientific; 1:500 dilution).

Quantifications

Immunofluorescence images were captured on a Leica SP8 confocal and processed and analysed using Fiji (http://fiji.sc). Epiblast, hypoblast and CER1-positive cell numbers were manually counted using the multi-point cell counter plugin. Quantification of trophectoderm was performed using Imaris software (version 9.1.2) using the spots tool with manual curation. To quantify n/c SMAD2.3 in human and mouse embryos, the central three planes of individual cells were used to generate a three-plane z-stack. Individual 4′,6-diamidino-2-phenylindole (DAPI)-positive nuclei were used to generate a nuclear mask using the ‘Analyze Particles’ function on either the DAPI or lineage-associated transcription factor channel. The adjacent cytoplasmic area was drawn individually for each nucleus and the mean fluorescence of each region was measured, and the ratio computed. When embryos were stained with E-Cadherin, the membrane was delineated to allow for cytoplasmic region of interest determination. When embryos were stained with podocalyxin, the cytoplasmic region of interest was drawn to ensure delineation of a region captures suitable intra-cellular variation allowing for valid normalization. Measurements were computed on raw SMAD2.3 signal. To quantify pSMAD1.5.9 nuclear intensity, a nuclear mask generated on a central three-plane z-stack for each nucleus, and mean fluorescence values were measured. Within each three-plane z-stack, a background fluorescence taken adjacent to or within a cavity of the embryo was used for background normalization (\(\frac}}\)). Background was normalized to (that is, to provide a comparable signal-to-noise ratio) rather than subtracted to account for the variability in laser penetration between experiments and z-planes. In stem cell experiments, nuclear masks were generated, mean fluorescence was measured, and all values were normalized to a control (DMSO) value of 1. To calculate the percentage of cleaved caspase-3-positive or CER1-positive cells, individual cells were manually counted using the cell counter plugin and presented as a percentage of all DAPI-negative or GATA6-positive cells. For 3D spheroid classification, the total number of structures was counted manually using the Cell Counter plugin, and each was assigned to a class of spheroid. For conditioned medium 3D spheroid quantifications, the central three planes of individual spheroids were used to generate a nuclear mask on the DAPI channel, and the mean nuclear pSMAD1.5.9 signal was quantified along with the signal of an acellular region for background normalization. To generate figures, images were processed by generating z-stacks of approximately five to ten planes to allow for visualization of embryo topology with cells on disparate planes followed by consistent adjustment of brightness and contrast.

Statistics and reproducibility

No statistical method was used to pre-determine sample sizes. Sample sizes are similar to previous publications7,10,24. For characterization of normal development, embryos lacking any of the three lineages were excluded. Multiple SMAD fluorescence intensities were taken per lineage per embryo. All embryos were included in functional experiments and each cell type count is taken from individual embryos. All stem cell experiments were performed independently at least twice. Investigators were not blinded to group allocation during the experiment or analysis, as blinding would not have been possible due to medium preparation and changing requirements. Group allocation was not performed randomly; rather, based on visual assessment of embryos, investigators attempted to ensure balanced distributions of blastocysts/implanting embryos assessed as expanded with nice inner cell masses versus embryos that appeared delayed or with visible cell death across experimental groups. Statistical tests, except the Bayesian distribution model, were performed in Prism 9 (GraphPad), and where relevant, two-sided tests were used. Normality was tested with a Shapiro–Wilk test. Bayesian distribution modelling, which is suited to the small sample sizes used in human embryo studies, was used as a supplemental tool to assess how each small-molecule treatment affected the distribution of cell number. To do this, the brms R package was used98,99, with the assumption of a Poisson distribution and the Control counts set to inform the priors and be used as reference. Brms’ default Markov chain Monte Carlo settings were used. Coefficient ± credible intervals were either below −0.33, denoting a decrease in the distribution compared to control, or above 0.33, indicating an increase in the distribution compared to control. Credible intervals that bridged this range indicate no significant difference. All coefficients and credible intervals in addition to Mann–Whitney test P values are presented in Supplementary Table 8. For data presentation, box plots encompass the 25th to 75th percentile in the box, with the median marked by the central line, and the mean marked by a cross. The minimum and maximum are marked by the whiskers. For violin plots, the dashed line marks the median value and dotted lines mark the 25th and 75th percentiles. For summary plots (for example, Fig. 2h,i), the mean ± standard error of the mean is plotted.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.