Ethics declaration

The use of human embryonic tissues described in this study is compliant with the International Society for Stem Cell Research guidelines. All human embryonic tissue samples used in this study were discarded material from elective terminations that were obtained once informed written consent to the use of samples in research was obtained from patients. The donated human embryonic tissues were anonymized and did not carry any personal identifiers. In all cases, the decision to terminate the pregnancy occurred before the decision to donate tissue. No payments were made to donors, and the donors knowingly and willingly consented to provide research materials without restrictions for research and for use without identifiers. Human embryonic tissues employed for RNA-seq, immunohistochemistry and immunofluorescence were obtained from voluntary abortions performed according to the guidelines and with the approval of the French National Ethics Committee. The study was approved by Ospedale San Raffaele Ethical Committee (TIGET-HPCT protocol) and by the Institutional Review Board of the French Institute of Medical Research and Health (number 21-854). Human embryonic tissues employed for ex vivo haematopoietic cultures were collected by the Human Developmental Biology Resource (HDBR; HDBR project number 200430), Newcastle University, Newcastle, United Kingdom, with approval from the Newcastle and North Tyneside NHS Health Authority Joint Ethics Committee (08/H0906/21 + 5). The HDBR is regulated by the UK Human Tissue Authority (HTA; https://www.hta.gov.uk/) and operates in accordance with the relevant HTA Codes of Practice. This use was also approved by Ospedale San Raffaele Ethical Committee (TIGET-HPCT protocol). No embryos were created nor cultured for research purposes. The use of hESCs was approved by the Ospedale San Raffaele Ethical Committee, included in the TIGET-HPCT protocol.

Human embryonic tissues

Human embryos were staged using anatomic criteria and the Carnegie classification. Samples employed for RNA-seq, immunohistochemistry and immunofluorescence were either used immediately as fresh tissues (ex vivo experiments and RNA-seq analysis) or fixed in phosphate-buffered saline (PBS) supplemented with 4% paraformaldehyde (Sigma-Aldrich), embedded in gelatin and stored at −80 °C (immunohistochemistry and immunofluorescence). Human embryonic tissues (CS12–CS13) analysed by RNA-seq were incubated in medium containing 0.23% w/v collagenase Type I (Worthington Biochemical Corporation, NC9482366) for 30 min at 37 °C, and the single-cell suspensions were filtered through a 70 µm cell strainer (BD Biosciences).

Human embryonic tissues (CS13) employed for ex vivo haematopoietic cultures were collected by HDBR (HDBR project number 200430), Newcastle University, Newcastle, United Kingdom, with written informed consent and approval from the Newcastle and North Tyneside NHS Health Authority Joint Ethics Committee (08/H0906/21 + 5). The human embryonic tissues were dissociated for 50 min at 37 °C with 10 mg ml−1 collagenase/dispase (Sigma-Aldrich, 10269638001) in PBS with Ca2+ and Mg2+ (Sigma-Aldrich, D8662), supplemented with 7% heat-inactivated foetal bovine serum (FBS, Hyclone, 12389802), 1% penicillin–streptomycin (Lonza, DE17-603E) and 10 µg ml−1 DNAse I (Calbiochem, 260913) and filtered through a 70 µm cell strainer (Falcon, 352235), similarly to what was previously described50.

Immunohistochemistry and immunofluorescence

The techniques employed have been previously described19. Briefly, 5-μm sections were incubated first with primary antibodies overnight at 4 °C, then for 1 h at room temperature (RT) with biotinylated secondary antibodies and finally with fluorochrome-labelled (BioLegend) or peroxidase-labelled streptavidin (Beckman Coulter). Peroxidase activity was revealed with 0.025% 3,3-diaminobenzidine (Sigma-Aldrich) in PBS containing 0.03% hydrogen peroxide. Low amounts of antigens (CD32 and ACE) were revealed by Tyramide signal amplification biotin or fluorescence amplification systems (Akoya, Biosciences). An isotype-matched negative control was performed for each immunostaining. When 3,3-diaminobenzidine was used on slides, they were counterstained with Gill’s haematoxylin (Sigma-Aldrich), mounted in XAM neutral medium (BDH Laboratory Supplies), analysed and imaged using an Optiphot 2 microscope (Nikon). Immunofluorescence-stained sections were cover-slipped in Prolong Gold Antifade Mountant with DAPI (Thermo Fisher Scientific) and analysed with an Axio Imager M2 microscope coupled to a Hamamatsu’s camera Orca Flash 4v3 using the ApoTome.2 function (Zeiss) for optical sectioning. The antibodies employed are listed in Supplementary Table 19.

RNA-seq

Human embryo sorted cells were collected in 6 µl of PBS supplemented with 0.5 µl of Protector RNAse inhibitor (Roche, 3335399001) and conserved in −80 °C. Full-length coding DNA (cDNA) was generated using Clontech SMART-Seq v4 Ultra Low Input RNA kit for Sequencing (Takara Bio Europe, 634891) according to the manufacturer’s instructions with 15 cycles of PCR for cDNA amplification by Seq-Amp polymerase. A total of 600 pg of pre-amplified cDNA was then used as input for Tn5 transposon tagmentation by the Nextera XT DNA Library Preparation Kit (Illumina, FC-131-1096) followed by 12 cycles of library amplification. After purification with Agencourt AMPure XP beads (Beckman Coulter, A63882), the size and concentration of libraries were assessed by capillary electrophoresis. Libraries were then sequenced with the Illumina HiSeq 4000 sequencing platform in the single-end mode and with a read length of 50 bp. For day 8 WNTd hPSC-derived haematopoietic cultures, total RNA from sorted CD34+CD32+CD43negCD184negCD73negDLL4neg and CD34+CD184+CD73+DLL4+CD43neg was purified using the ReliaPrep RNA Cell Miniprep System and RNA-seq libraries were generated using the Smart-seq2 method. One nanogram of RNA was retrotranscribed, and cDNA was PCR amplified (15 cycles) and purified with AMPure XP beads. Sequencing was performed on an Illumina NovaSeq6000 (single-end, 100 bp read length) following the manufacturer’s instruction.

For both RNA-seq datasets, raw reads quality control was accomplished using the FastQC tool (http://www.bioinformatics.babraham.ac.uk/projects/fastqc) and read trimming was performed using the Trim Galore software (https://doi.org/10.5281/zenodo.5127899) to remove residual adapters and low-quality sequences. Trimmed reads were aligned against the human reference genome (GRCh38) using STAR51 with standard parameters. Uniquely mapped reads were then assigned to genes using the featureCounts tool from the Subread package52, considering the GENCODE primary assembly v.34 gene transfer file as reference annotation for the genomic features. Gene count matrices were then processed by using the R/Bioconductor differential gene expression analysis packages DESeq2 (ref. 53) applying the standard workflow.

For the human embryos’ dataset, a paired analysis was set up modelling gene counts using the following design formula: ~donor + condition. Gene P values were corrected for multiple testing using false discovery rate (FDR). Genes with adjusted P values <0.05 were considered differentially expressed.

ORA and a GSEA were then computed considering the GO Biological Process (BP) terms from the C5 collection of the Molecular Signatures Database (MSigDB version 7.2) using the R/Bioconductor package clusterProfiler54 (http://bioconductor.org/packages/release/bioc/html/clusterProfiler.html, v 3.8.1). ORA was applied to the significantly DEGs, while GSEA was performed by pre-ranking genes according to fold change (FC) values. P values were corrected for multiple testing using FDR and enriched terms with an adjusted P value less than 0.05 were considered statistically significant. Volcano plots were generated using the R package ggplot2 (https://ggplot2.tidyverse.org) and have been used to display RNA-seq results plotting the statistical significance (adjusted P value) versus the magnitude of change (FC). Heatmaps were generated using the R package pheatmap (https://CRAN.R-project.org/package=pheatmap). Surface genes were extracted using the surfaceome database (http://wlab.ethz.ch/surfaceome/)55.

scRNA-seq

CD34+CD43negCD73negCD184neg cells were sorted at day 8 from WNTd hPSC-derived haematopoietic cultures that were treated at day 2 with 6 µM SB-431542 (Tocris, 1614)56. Libraries were prepared following the manufacturer’s instructions using the Chromium platform (10x Genomics) with the 3′ gene expression (3′ GEX) V3 kit, using an input of ~10,000 cells. Briefly, Gel-Bead in Emulsions (GEMs) were generated on the sample chip in the Chromium controller. Barcoded cDNA was extracted from the GEMs by post GEM reverse transcription cleanup and amplified for 12 cycles. Amplified cDNA was fragmented and subjected to end-repair, poly A-tailing, adapter ligation and 10x-specific sample indexing following the manufacturer’s protocol. cDNA libraries were sequenced in paired-end mode on a NovaSeq instrument (Illumina) targeting a depth of 50,000–100,000 reads per cell.

Sequencing reads were processed into gene count matrix by Cell Ranger (https://support.10xgenomics.com/single-cell-gene-expression/software/pipelines/latest/what-is-cell-ranger, v 4.0.0) from the Chromium Single Cell Software Suite by 10x Genomics. In detail, fastq files were generated using the Cell Ranger ‘mkfastq’ command with default parameters. Gene counts for each cell were quantified with the Cell Ranger ‘count’ command with default parameters. The human genome (GRCh38.p13) was used as the reference. The resultant gene expression matrix was imported into the R statistical environment (v 4.0.3) for further analyses. Cell filtering, data normalization and clustering were carried out using the R package Seurat57 v 3.2.2. For each cell, the percentage of mitochondrial genes, number of total genes expressed and cell cycle scores (S and G1 phases) were calculated. Cells with a ratio of mitochondrial versus endogenous gene expression >0.2 were excluded as putative dying cells. Cells expressing <200 or >6,000 total genes were also discarded as putative poorly informative cells and multiplets, respectively. Cell cycle scores were calculated using the ‘CellCycleScoring’ function that assigns to each cell a score based on the expression of the S and G2/M phase markers and stores the S and G2/M scores in the metadata along with the predicted classification of the cell cycle state of each cell. Counts were normalized using Seurat function ‘NormalizeData’ with default parameters. Expression data were than scaled using the ‘ScaleData’ function, regressing on the number of unique molecular identifier, the percentage of mitochondrial gene expression and the difference between S and G2M scores. By using the most variable genes, dimensionality reduction was then performed with PCA by calculating 100 principal components (PCs) and selecting the top 55 PCs. Uniform manifold approximation and projection (UMAP) dimensionality reduction58 was performed on the calculated PCs to obtain a two-dimensional representation for data visualization. Cell clusters were identified using the Louvain algorithm at resolution r = 0.6, implemented by the ‘FindCluster’ function of Seurat. To find the differentially expressed (marker) genes from each cluster, the ‘FindAllMarkers’ function (iteratively comparing one cluster against all the others) from the Seurat package was used with the following parameters: adjusted P values <0.05, average log FC >0.25, and percentage of cells with expression >0.1. A comprehensive manual annotation of the cell types was performed using the previously obtained markers list. DEGs between cells of cluster 11 against cells of clusters 0, 1 and 2 and clusters 16 and 17 were determined by the ‘FindMarkers’ function using the following parameters adjusted P values < 0.05, |average log FC| >0 and percentage of cells with expression >0. GSEA was then performed considering GO BP terms from the C5 collection of the Molecular Signatures Database (MsigDB v 7.2) using the R/Bioconductor package clusterProfiler54 (http://bioconductor.org/packages/release/bioc/html/clusterProfiler.html, v 3.8.1). ORA was computed on the significantly DEGs considering GO BP terms from the C5 collection of the Molecular Signatures Database (MsigDB v 7.2) and the Reactome Pathways Database using the R/Bioconductor package54 (v 3.8.1). P values were corrected for multiple testing using FDR and enriched terms with an adjusted P value less than 0.05 were considered statistically significant. A barplot was constructed using the R package ggplot2 (https://ggplot2.tidyverse.org).

scRNA-seq samples from the public dataset GSE162950 were retrieved and processed as described in ref. 38. The ‘DotPlot’ and the ‘VlnPlot’ functions from the Seurat R package were used to construct a scorecard highlighting the expression pattern of selected cells having RUNX1 + CDH5 + FCGR2B+ or RUNX1 CDH5 + FCGR2B− expression patterns.

Pseudotime trajectory was constructed using Monocle3 (https://cole-trapnell-lab.github.io/monocle3/, v 0.2.3)59,60. Expression and feature data were extracted from the Seurat object, and a Monocle3 ‘cell_data_set’ object was constructed. The processed data were normalized followed by PCA analysis using the Monocle3 function ‘preprocess_cds’. Dimensionality reduction was performed using the ‘reduceDimension’ function. Trajectory graph learning and pseudo‐time measurement through reversed graph embedding were performed with ‘learn_graph’ function. Cells were ordered along the trajectory using the ‘orderCells’ method with default parameters. The ‘plot_cells’ function was used to generate the trajectory plots.

To corroborate the findings from Monocle3, a trajectory inference analysis was conducted using the dynverse workflow, a component of the R package dyno (v 0.1.2)40. The Dynbenchmark utility, which offers a comprehensive framework for selecting the most suitable trajectory inference method according to the available experimental data, was utilized via the ‘guidelines_shiny()’ function. Following these guidelines, the trajectory inference analysis was performed using the partition-based graph abstraction (PAGA)-tree algorithm61. Input data for dyno, including gene expression matrices, dimensionality reduction coordinates, clustering information and cell metadata, were derived from Seurat output and processed using the ‘wrap_expression()’ function. The cell trajectory was subsequently calculated by dyno using the ‘infer_trajectory()’ function, employing the ‘ti_paga_tree()’ method. Trajectory paths and pseudotime values were visualized on UMAP coordinates through the ‘plot_dimred()’ function provided by dyno.

To investigate the role of the CD32 gene in HEC ontogeny, genetic knockouts were simulated using the CellOracle tool (v 0.12.0)41. CellOracle integrates a gene regulatory network (GRN) with pseudotime analysis to predict shifts in cellular identities resulting from gene perturbations. This tool simulates alterations in gene expression due to perturbations and compares these changes with the cell’s developmental trajectory within the GRN. This comparison allows for the estimation of transition probabilities between different cell states along the pseudotime axis. Following this, CellOracle generates a transition trajectory graph, illustrating the potential shifts in cellular identities after perturbation. This analysis was performed in a Python (version 3.8) environment, using Jupyter notebooks. scRNA-seq data, initially processed with Seurat, were converted to AnnData format using the anndata2ri tool (https://github.com/theislab/anndata2ri), ensuring content preservation for subsequent analysis. The CellOracle object construction utilized this data. Highly variable genes, critical for downstream analysis, were identified using the scanpy.pp.filter_genes_dispersion() function from scanpy62, specifying n_top_genes = 3,000. A preliminary GRN was constructed using the oracle.get_links() function within the Oracle() class, based on ligand–receptor interactions from the CellTalkDB database63. This base GRN was further refined by incorporating the CD32 gene and its interactors, as identified in the STRING database (https://string-db.org/). Pseudotime analysis was conducted using the Pseudotime_calculator() class, employing the PAGA method from scanpy and integrating it into the CellOracle framework. This analysis culminated in the creation of a pseudotime gradient vector field with the Gradient_calculator() class from CellOracle, depicting the normal developmental trajectory. Subsequently, in silico perturbation of CD32 expression and simulation of resultant cell identity shifts were performed using the simulate_shift() and estimate_transition_prob() functions from the Oracle class. To compare the effects of CD32 perturbation with normal development, the Oracle_development_module class was used to calculate perturbation scores by computing the inner product of the respective vector fields with the calculate_inner_product() function.

Human pluripotent stem cell maintenance and differentiation

The already-established H1 (WiCell Research Institute, WA01)64 and RUNX1C–EGFP H9 (ref. 27) hESC lines were grown on irradiated mouse embryonic fibroblast feeders in hES medium defined as Dulbecco’s modified Eagle medium/F12 medium (Corning, L022046-10092CVR) supplemented with 25% of KnockOut Serum Replacement (Thermo Fisher Scientific, 10828028), 1% penicillin–streptomycin (Lonza, DE17-603E), 2 mM l-glutamine (Lonza, BE17-605E), 0.1% β-mercaptoethanol (Sigma-Aldrich, M3148) and 0.7% of MEM non-essential amino acids solution (Thermo Fisher Scientific, 11140035). Right before usage, 1 µg ml−1 ciprofloxacin HCl (Sigma-Aldrich, PHR1044-1G) and 20 ng ml−1 human recombinant basic fibroblast growth factor (bFGF, R&D, 233-FB-500/CF) were added to hES medium. Alternatively, cells were cultured using Essential 8 medium (Thermo Fisher Scientific, A1517001) on Matrigel-coated plasticware (Corning Life Sciences, 356230). Cells were maintained and expanded at 37 °C, 21% O2, 5% CO2.

For differentiation, hPSCs were processed for embryoid body (EB) generation. EB aggregates were resuspended in SFD medium defined as 75% (Iscove’s modified Dulbecco’s medium (Corning, 15343531), 25% Ham’s F12 (Corning, 10-080-CVR), 0.005% bovine serum albumin—fraction V, B27 supplement (Thermo Fisher Scientific, cat. no. 12587010), N2 supplement (Thermo Fisher Scientific, cat. no. 17502048), 1% penicillin–streptomycin and 1 µg ml−1 ciprofloxacin HCl. The differentiation medium was supplemented as previously described14,56. Briefly, the first day of differentiation, SFD medium was supplemented with 2 mM l-glutamine, 1 mM ascorbic acid (Sigma-Aldrich, A4544), 400 µM 1-thioglycerol solution (Sigma-Aldrich, M6145), 150 μg ml−1 transferrin (R&D, 2914-HT) and 10 ng ml−1 BMP4 (R&D, 314-BP-MTO). Twenty-four hours later, 5 ng ml−1 bFGF (R&D, 233-FB-500/CF) was added. At the second day of differentiation, 3 μM (or 5 μM for feeder-free cultures) CHIR99021 (Cayman Chemical Company, CT99201) was added, as indicated. On the third day, EBs were changed to StemPro-34 medium (Thermo Fisher Scientific, 10639011) supplemented with penicillin–streptomycin, l-glutamine, ascorbic acid, 1-thioglycerol and transferrin, as above, with additional 5 ng ml−1 bFGF and 15 ng ml−1 VEGF (R&D, MAB3572). On day 6, 10 ng ml−1 interleukin (IL)6 (130-093-934), 25 ng ml−1 insulin-like growth factor 1 (IGF1, 130-093-887), 5 ng ml−1 IL11 (130-103-439), 50 ng ml−1 stem cell factor (SCF) (130-096-696) and 2 U ml−1 erythropoietin (EPO) (Peprotech, 100-64) were added. Where indicated, 10 ng ml−1 of BMP4 or 250 nM of LDN 193189 dihydrochloride (Tocris, 6053) were added from day 3 to day 8. To assess the emergence of CD32+ cells from CD32neg, CD32neg cells were isolated at day 8 of WNTd haematopoietic cultures and cultured in day 6 medium for 48 h. All cytokines were purchased from Miltenyi Biotec, unless indicated differently. All differentiation cultures were maintained at 37 °C. All EBs and mesodermal aggregates were cultured in 5% CO2, 5% O2, 90% N2.

Generation of CD32 KD hESC line

An AAVS1-CA-GFP-Puro donor plasmid comprising homology arms for the integration into the AAVS1 locus, puromycin resistance gene (Puro) and GFP sequence fused to a polylinker site was used43. In this donor plasmid, we synthetized three different short hairpin RNAs (shRNAs) against the 3′ untranslated region of FCGR2B (shRNA1: 3′-GGTTGGAGTGTAGACTGAACTGCCT-5′, shRNA2: 3′-TCAAGGCTGTATTGGTTGGAGTGTA-5′, shRNA3: 3′-CAAGGCTGTATTGGTTGGAGTGTAG-5′). H1 cells cultured under feeder-free conditions (Essential 8) were nucleofected using a Lonza 4D nucleofector and P3 Primary Cell reagents. Cells were nucleofected with 16 µg total of plasmid DNA including 7 µg of eCas9 + gRNA plasmid (Addgene #71814, modified to contain T2 gRNA: 5′-GGGGGCCACTAGGGACAGGA-3′), 7 µg of donor plasmid containing the 3×-CD32 miRNA insert, and 2 µg of p53DD (Addgene #4156). The modified cells were then enriched by puromycin selection and purified by FACS sorting based on GFP expression.

T cell and NK cell differentiation

To test the T cell potential, candidate cells isolated by FAC sorting as indicated were seeded on OP9DLL4-coated 24-well plates. OP9DLL4 were a kind gift from Juan-Carlos Zúñiga-Pflücker and described previously65. The cells were cultured in Alpha MEM (Thermo Fisher, 12000063) supplemented with 2.2 g l−1 sodium bicarbonate (Corning, 61-065-RO), 20% FBS (HyClone), 1% penicillin–streptomycin, 2 mM glutamine (Thermo Fisher Scientific) and 400 µM 1-thioglycerol solution. Cells were supplemented with 5 ng ml−1 IL7, 5 ng ml−1 FLT3L and, for the first 5 days of differentiation, 50 ng ml−1 SCF. The cells were split every 4–5 days by vigorous pipetting and passaging through a 40-μm cell strainer and plated on freshly seeded stromal cells. T lymphoid output was assayed by FACS analysis after 21–24 days of differentiation. For the analysis of the T cell maturation and activation markers, the three-dimensional artificial thymic organoid (ATO)66 system was adapted to our protocol. CD32+ cells were mixed with MS5DLL4 cells (kind gift of Tom Taghon, University of Ghent)67 in a 1:50 ratio. Haematopoietic progenitors and stromal cells were centrifuged together at 1,500 rpm for 5 min and resuspended in the T cell differentiation medium, consisting of RPMI 1640 (Corning 10-040-CV), 4% B27 (Thermo Fisher Scientific 17504044), 30 µM ascorbic acid and 1% penicillin/streptomycin and GlutaMAX (Thermo Fisher #35050061). Each single aggregate was seeded in a volume of 5 µl onto the cell insert membrane for the air–liquid interphase culture. Cells were analysed after 6 weeks of culture. For NK-specific differentiation, CD32+ cells were cultured in Alpha MEM (Thermo Fisher, 12000063) supplemented with 2.2 g l−1 sodium bicarbonate (Corning, 61-065-RO), 20% FBS (HyClone), 1% penicillin–streptomycin, 2 mM glutamine (Thermo Fisher Scientific) and 400 µM 1-thioglycerol solution onto OP9DLL4 stromal cells. Cultures were supplemented with 5 ng ml−1 IL7 (130-095-362), 5 ng ml−1 FLT3L (130-096-479), 10 ng ml−1 IL15 (130-95-765) and, for the first 7 days of differentiation, 30 ng ml−1 IL3 (130-095-070). All cytokines were purchased from Miltenyi Biotec. The cells were maintained for 14 days on the same stromal cells before performing FACS analysis. Every 7 days, the culture was supplemented with fresh medium.

CFC generation assay

The colony-forming cell (CFC) generation assay was performed as previously described68. Briefly, sorted cells were cultured on irradiated OP9DLL1 (a kind gift from Juan-Carlos Zúñiga-Pflücker) monolayers in Alpha MEM (ThermoFisher, 12000063) supplemented with 20% FBS (HyClone), 1% penicillin–streptomycin 2 mM l-glutamine, 30 ng ml−1 thrombopoietin (TPO) (Miltenyi Biotec, 130-095-747), 10 ng ml−1 BMP4, 50 ng ml−1, 25 ng ml−1 IGF1, 10 ng ml−1 IL11, 10 ng ml−1 FLT3L and 4 U ml−1 EPO. After 5 days, cells were collected using 0.25% trypsin–EDTA (Thermo Fisher Scientific, 25-200-056) for 3 min at 37 °C. Cells were then filtered through a 40 μM filter and seeded on methylcellulose medium (STEMCELL Technologies, H4034). Cells were seeded on methylcellulose supplemented with 150 µg ml−1 transferrin, 50 ng ml−1 TPO, 10 ng ml−1 VEGF, 10 ng ml−1 IL6, 50 ng ml−1 IGF1, 5 ng ml−1 IL11 and 4 U ml−1 EPO. Colonies’ number and morphology were evaluated after 15 days by light microscopy.

HEC culture

CD34+CD43negCD184negCD73negDLL4negCD32+/neg or CD34+CD43negCD184negCD73negDLL4negCD32negCD44+ were isolated at day 8 of WNTd haematopoietic culture and re-aggregated overnight at 3 × 105 cells ml−1 as previously described14. The cells were seeded in Stempro medium, supplemented with 1% glutamine, 50 μg ml−1 ascorbic acid, 150 μg ml−1 transferrin, 400 μM 1-thioglycerol solution, 30 ng ml−1 TPO, 10 ng ml−1 VEGF, 5 ng ml−1 bFGF, 30 ng ml−1 IL3, SCF, 50 ng ml−1 IGF1, 10 ng ml−1 IL6, 5 ng ml−1 IL11 and 4 U ml−1 EPO. Aggregates were then transferred onto thin-layer Matrigel-coated plasticware where they were cultured for an additional 1–7 days in the same media. Where indicated 10 μM γ-secretase inhibitor (γSi) L-685,458 (Tocris, 2627) or equal volume of dimethyl sulfoxide (DMSO, Sigma-Aldrich, D4540), as control, were added to the HEC culture test the effect of NOTCH signalling inhibition. Where indicated, 10 μM of the Rho kinase inhibitor (ROCKi) Y-27632 dihydrochloride (Cayman Chemical, TB1254-GMP) or equal volume of DMSO, as control, were added to the HEC culture to test the effect of ROCK signalling inhibition.

Single CD34+CD43negCD184negCD73negDLL4negCD32+/CD44+ cells were FACS-sorted directly onto a Matrigel-coated well of 96-well plate at day 8 of WNTd haematopoietic cultures. Cells were cultured as above. Haematopoietic and non-haematopoietic clones were evaluated by light microscopy and FACS analysis after 10–14 days of culture.

Cell staining, flow cytometry and cell sorting

Samples for FACS analysis or cell sorting were incubated with antibody mixes for 15–30 min at 4 °C. Dead cells were excluded using 7-aminoactinomycin D (7AAD) during staining. For the analysis of the T cell maturation, ATO aggregates were stained with Maleimide PromoFluor840 for dead cell exclusion. Cells were then incubated with antibody mixes diluted in Brilliant stain buffer (BD Biosciences, 563794) + PBS (Corning, 21-040-CM) + 2% FBS (HyClone, SH30066.03) + 4% FcR blocking for 15 min at RT. After washing, cells were fixed in 1% paraformaldehyde and analysed. The antibodies employed are listed in Supplementary Table 19. Cells were sorted with FACSAria II with the FACSDiva software (BD Biosciences). Sorting gates were set using appropriate fluorescence minus one and single staining controls. FACS analysis was performed using FACS Canto with the FACSDiva software (BD Biosciences) or Cytoflex S or Cytoflex LX with Coulter CytExpert software (Beckman Coulter) for the acquisition and the FlowJo software (BD Biosciences) for the analysis. Where indicated, WNTd day 8 CD144+ cells were selected using magnetic bead-based separation with CD144 MicroBeads (Miltenyi Biotec, 130-097-857) following the manufacturer’s instructions.

Statistics and reproducibility

For all multivariate statistical analyses, analyses of variance (ANOVAs) were performed with the appropriate corrections for multiple comparisons. One-way ANOVA with Tukey’s multiple comparison test was chosen for single metrics with more than two populations. The data distribution was not formally tested but was assumed to be normally distributed with equal variance. Sample size and replication were determined by historical controls14. For bivariate statistical analyses, Student’s t-test was performed with the appropriate corrections (one tail, non-parametric). In general, biological replicates were excluded only if internal controls failed and technical replicates were not excluded. Experimental conditions were not randomized, but covariates were controlled by an equal distribution of sorted cells across controls and experimental conditions. Blinding of experimental conditions was not relevant as our studies do not require grading of the results.

Reporting summary

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