Introduction

Blood formation in the embryo is a multi-stage process that develops across multiple cellular niches, reflecting a complex hierarchy of tissue interactions. It involves sequential production of distinct cell types in so-called hematopoietic waves, separated in time and space (Lacaud and Kouskoff, 2017; Costa, Kouskoff and Lacaud, 2012; Medvinsky, Rybtsov and Taoudi, 2011; Dzierzak and Bigas, 2018). The first wave of hematopoiesis produces unipotent red blood cell and macrophage precursors in the yolk sac (YS) from an angioblast precursor which can also form endothelium; it occurs at mouse embryo day (E)7-E7.5 (Fujimoto et al., 2001; McGrath and Palis, 2005; Moore and Metcalf, 1970). The subsequent wave, known as definitive, produces: first, oligopotent erythro-myeloid progenitors (EMPs) in the YS (E8-E8.5); and later myelo-lymphoid progenitors (MLPs – E9.5-E10), multipotent progenitors (MPPs – E10-E11.5), and hematopoietic stem cells (HSCs – E10.5-E11.5), in the aorta-gonad-mesonephros (AGM) region of the embryo proper (Ivanovs et al., 2011; McGrath et al., 2015; Palis et al., 1999; Medvinsky and Dzierzak, 1996; Medvinsky, Rybtsov and Taoudi, 2011). Definitive blood waves are characteristically specified from a specialised hemogenic endothelium (HE) through a process of endothelial-to-hematopoietic transition (EHT) (Zovein et al., 2008; Li, W. et al., 2005). HSC specification from HE is the last event in hematopoietic development and occurs in characteristic cell clusters found on the ventral wall of the dorsal aorta (Ivanovs et al., 2011; Boisset et al., 2010). All blood cell types enter circulation from their original locations and migrate to the fetal liver (FL) (Ciriza et al., 2013; Ghiaur et al., 2008). Most embryonic blood progenitors are transient and sustain blood production exclusively during embryonic development, through differentiation. In contrast, HSCs make little or no contribution to embryonic and fetal blood: instead, they proliferate (i.e. self-renew) to expand their numbers. At least a subset of HSC migrates to the bone marrow (BM) niche at the end of gestation and balances self-renewal and differentiation to maintain blood production throughout the entire post-natal life (Bowie et al., 2006; Kikuchi and Kondo, 2006; Ganuza et al., 2022; Yokomizo et al., 2022).

Blood progenitors and HSCs can be targeted by genetic alterations that confer clonal advantage and/or drive initiation and development of leukemia. Leukemia is the most common malignancy and the most common cause of cancer-related death in the paediatric age group (Steliarova-Foucher et al., 2017; Dong et al., 2020). Several paediatric leukemias originate in utero, and a subset of these are driven by mutations exclusive to the paediatric group (Masetti et al., 2015; Cazzola et al., 2021). Such pediatric-exclusive mutations target blood cell types restricted to development or depend on signals that are specific to embryonic hematopoietic niches, effectively defining distinct biological entities to adult malignancies (Cazzola et al., 2021; Bolouri et al., 2018; Chaudhury et al., 2018). Dissection of these forms of developmental leukemia is confounded by the inability to robustly identify and isolate their embryonic cells of origin, and/or to fully recapitulate the disease in post-natal cells, which leads to failure in generating relevant targeted therapies (Alexander and Mullighan, 2021). One such example is t(7;12)(q36;p13) acute myeloid leukemia (AML) driven by rearrangement and overexpression of the MNX1 gene locus: MNX1-rearranged (MNX1-r) AML is restricted to infants under the age of 2 and carries a high risk of relapse and poor survival (Ragusa et al., 2023). MNX1 is a developmental gene required for specification of motor neurons and pancreatic acinar cells (Thaler et al., 1999; Harrison et al., 1999). MNX1 is not known to participate in embryonic haematopoiesis, however it has been discordantly described to be expressed in some subsets of adult haematopoietic cells (Deguchi and Kehrl, 1991; Nagel et al., 2005; Wildenhain et al., 2012; Taketani et al., 2008). It is ectopically expressed in AML cells carrying one of a variable set of t(7;12) translocations which place the normally silent MNX1 gene under the control of ETV6 regulatory elements (Weichenhan et al., 2023; Bousquets-Muñoz et al., 2024; Ragusa et al., 2023; Ballabio et al., 2009). ETV6 is required for hematopoietic cell specification from HE via VEGF signalling (Ciau-Uitz et al., 2010), potentially positioning MNX1 within a critical embryonic hemogenic regulatory network.

Faithful in vitro recapitulation of developmental blood cell types and their embryonic niches has the potential to dissect the cellular and molecular mechanisms presiding at initiation of developmental leukemia, reveal prognostic biomarkers, and indicate therapeutic vulnerabilities which are otherwise challenging to test in the embryo. Despite widespread use of embryonic stem (ES) cell and induced pluripotent stem (iPS) cell-based models of blood specification, a system which recapitulates the spatial and temporal complexity of embryonic blood formation within the respective time-dependent niches is still lacking (Dijkhuis et al., 2023). In the last few years, gastruloid models have emerged as robust avatars of early development, self-organising processes of symmetry breaking, elongation, multi-axis formation, somitogenesis and early organogenesis, with remarkable similarity to embryo processes in space and time (van den Brink, Susanne C and van Oudenaarden, 2021; Beccari et al., 2018; van den Brink, Susanne C et al., 2020). Gastruloids are grown individually in a multi-well format that facilitates tracking and screening of developmental processes. Relevant to blood formation, a variant gastruloid culture system, which successful organises a heart primordium (Rossi et al., 2021), was also able to recapitulate YS blood formation with erythroid and myeloid cellularity (Rossi et al., 2021; Rossi et al., 2022).

Herein, we describe a new protocol of gastruloid formation, i.e. hemogenic gastruloids (hGx), that captures YS and AGM-like waves of definitive blood progenitor specification from HE. The AGM-like wave includes cells capable of short-term hematopoietic engraftment of the spleen and BM of immunodeficient mice, overall suggesting nearly full EMP-to-MPP/HSC recapitulation of embryonic blood development. We harnessed the embryonic haematopoietic context provided by hGx for modelling MNX1-r infant leukemia by proxy of MNX1 overexpression. MNX1-overexpressing hGx showed expansion of cells at the HE-to-EMP transition, which became susceptible to transformation in vitro and recapitulated MNX1-r AML patient signatures, placing MNX1’s leukemogenic effects at a very early stage of developmental blood formation.

Results

HGxs capture cellularity and topography of developmental blood formation

With the aim of recapitulating embryonic haematopoiesis with spatial and temporal accuracy, we modified existing gastruloid formation protocols to promote the specification of hemogenic mesoderm derivatives and extend developmental time beyond the E8.0-equivalent of early developmental gastruloids (Van den Brink, Susanne C et al., 2014) and the E9.0 timepoint of cardiogenic gastruloids (Rossi et al., 2022). We established a 216h hGx protocol (Fig. 1A) using a Flk1-GFP mouse ES cell line (Jakobsson et al., 2010), which allows the tracking of early hemato-endothelial specification by expression of the Flk1 (i.e. Kdr) locus. Activation of the Flk1 expression, visible by fluorescence of the GFP tag, necessitated induction of the TGF-β signalling pathway by Activin A, in addition to the characteristic gastruloid patterning via WNT activation by CHI99021 supplementation at 48h (Fig. S1A). We sought to promote hemato-endothelial programmes through addition of VEGF and FGF2 at 72h (Sroczynska et al., 2009) (Fig. 1A), and observed extension of a polarised Flk1-GFP-expressing cellular network (Fig. 1B), which included VE-cadherin⁺ C-Kit⁺ cells suggestive of EHT (Fig. 1C-D). This process of HE specification and EHT peaked at 120-144h; it was followed by a surge in CD41⁺ candidate hematopoietic cells at 144h (Fig. 1D). In our experience, observation of early polarised activation of the Flk1 locus at 96h (Fig. 1B) was predictive of CD41⁺ detection 2 days later, suggesting that TGF-β activation in early patterning is required for later hematopoietic development, putatively by modifying mesoderm specification. We extended the protocol by incorporation of a 24-hour pulse of sonic hedgehog (Shh) between 144-168h to mimic the aortic patterning that precedes cluster formation (Fig. 1A) (Rybtsov et al., 2014). Continued exposure to VEGF after the Shh pulse resulted in specification of CD45⁺ hematopoietic cells progenitors (Fig. 1D-E), which increased after 168h. The proportion of CD45⁺ cells was enhanced by the addition of SCF, FLT3L and TPO (Fig. S1B-C), a combination of cytokines normally used for HSC maintenance; conversely, it did not require continued addition of FGF2 after 168h (Fig. S1D), which was omitted. CD45⁺ cells appeared in a small number of ‘cluster’-like structures adjacent to CD31⁺ endothelium (Fig. 1E), which were reminiscent of HSC and progenitor cell budding from the aortic HE, and suggested recapitulation of hemogenic tissue architecture. Time-dependent specification of C-Kit⁺, CD41⁺ and CD45⁺ cells in hGxs could be obtained with other mouse ES cell lines, including commonly used E14Tg2a (E14) cells (Fig. S1E-G), with successful specification of comparable frequencies of CD45⁺ progenitors (Fig. S1H), attesting to the reproducibility of the protocol.

Figure 1

Overall, our extended hGx model could capture successive specification of HE, CD41⁺ and CD45⁺ cells, with temporal and spatial congruence relative to the embryo, thus showing initial promise as a tool for the exploration of normal and perturbed developmental hematopoiesis.

To better characterize the extent and progression of developmental hematopoiesis in hGx, we performed single-cell RNA-sequencing (scRNA-seq) time-course analysis of gastruloid cell specification. We sorted cells from 2 independent hGx cultures at 120, 144, 168, 192 and 216h, and profiled a total of 846 cells using the Smart-Seq2 protocol (Picelli et al., 2014) (Fig. S2A). We analysed unfractionated live single hGx cells at 120-192h and enriched the dataset with sorted CD41⁺ cells at their peak time of 144h, and sorted CD45⁺ cells at 192 and 216h. At timepoints of significant enrichment in CD41⁺ (144h) and CD45⁺ cells (192h) (Fig. 1D), we also sorted C-Kit⁺ cells. Library preparation and sequencing generated an average of 120000 reads/cell, which were mapped to an average of 4000 genes/cell, with minimal signs of cell stress / death as measured by mitochondrial DNA fraction (Fig. S2B). Read and gene counts were similar between biological replicates, cell types and at different time points (Fig. S2C). The exception was 120h unfractionated hGx cells, which despite similar sequencing depth (average read counts), were mapped to twice the number of genes (Fig. S2C), possibly reflecting multi-gene program priming at the onset of hemogenic specification. We selected highly varying genes (HVG) before principal component analysis (PCA) dimensionality reduction and retained the most relevant dimensions. In the PCA reduced space, we constructed a KNN graph and used uniform manifold approximation and projection (UMAP) to visualize the data on 2 dimensions, looking for cell communities using Leiden clustering (Fig. 2A). We identified 12 cell clusters of which 2 (clusters 0 and 5) almost exactly mapped to C-Kit⁺ cells (also Sca-1⁺), 2 (clusters 1 and 8) contained CD45⁺ sorted cells, and cluster 4 uniquely captured the CD41⁺ cells observed at 144h (Fig. 2A). Although some unfractionated cells overlapped with the hemogenic cell clusters (Fig. 2A - center), most occupied different transcriptional spaces, reflecting the frequency of hematopoietic cells and suggesting the presence of potential hemogenic niches.

Figure 2

To explore tissue and lineage affiliation of the different clusters, we performed differential gene expression against all other cells using Wilcoxon ranking test, and established cluster classifier gene lists (Supplemental File S1). Top classifier genes for clusters 0 and 5 had endothelial affiliation (Supplemental File S1), and the cellular composition of the clusters broadly reflected their time signature (Fig. 2A): cluster 5 comprised 144h cells and cluster 0 included most C-Kit⁺ cells at 192h. Cluster 5 occupied the vicinity of 144h-CD41⁺ cells (Fig. 2A), which carried an erythroid-biased signature (Supplemental File S1). On the other hand, 192/216h-CD45⁺ cells in cluster 8 expressed myeloid and lymphoid genes (Supplemental File S1). The remaining clusters largely corresponded to time-restricted unfractionated cells (Fig. 2A - right), with clusters of cells present at later time-points matching mesenchymal stromal cell and autonomic neuron signatures (Fig. S2D), potentially configuring incipient formation of niches relevant to production of HSC (Kapeni et al., 2022; Fitch et al., 2012).

To further characterise the identity of hGx cells, we interrogated lineage priming in individual hemato-endothelial cells (Fig. 2B-C), to probe the presence of putative multilineage YS (EMP) and AGM (MLP/MPP) progenitors, and HE cells. The heatmaps (Fig. 2B-C) show expression of lineage markers and regulators, with unsupervised clustering of cells (Fig. S2E-F) and genes. The gene list used was identical in both analyses, with only expressed genes shown. Concordance between sorting antigens and expression of the respective genes (Kit, Itga2b [CD41] and Ptprc [CD45]) was partial potentially reflecting low levels of transcript expression in the hGx system. Summary of endothelial, co-expressing erythro-myeloid (EM) and co-expressing myelo-lymphoid (ML) programs (Fig. 2D) clearly associates the first with C-Kit⁺ and the latter with CD45⁺ cells; CD41⁺ cells capture a heterogeneous population, which includes a proportion of cells co-expressing erythroid and myeloid genes that constitute candidate EMPs.

Clustering of cells sorted as C-Kit⁺ did not separate the 144h and 192h time-points (Figs. 2B and S2E), indicating that the relative separation on the global UMAP was not based on endothelial identity genes. We observed widespread expression of arterial genes (e.g. Flt1, Efnb2) with relative reciprocity with venous endothelial genes (e.g. Flt4) supporting specification of both types of endothelia (Fig. 2B). There was limited co-expression of hematopoietic genes, but genes associated with endothelial-to-hematopoietic transition (EHT), e.g. Gata2, Notch1, and Procr, were significantly more frequent within the C-Kit⁺ population (Fig. 2E). Specific associations of Cd44 and Procr with the 192h time-point (Fig. 2F) putatively represent aorta-like EHT.

Clustering of CD41⁺ and CD45⁺ cells (Fig. 2C and S2F) more clearly separates unique hematopoietic signatures. A subset of cells, predominantly CD41⁺, expresses high levels of globin genes. While embryonic β-globin chains are highly expressed, there is a gain in fetal Hbb-y and adult Hbb-bt and Hbb-bs (C57Bl/6 strain-specific β major and β minor, respectively) in CD45⁺ cells at 216h, indicative of AGM-like progenitor identity (Fig. 2C). 192 and 216h cells with high expression of Ptprc have an MLP profile (Fig. 2C-D), co-expressing myeloid and lymphoid genes including immunoglobulin chains. There is progression of the lymphoid signature between 192 and 216h (Fig. 2D and S2G). Other CD45⁺ cells have low-frequency expression of HSC-associated transcription factors, including Myb, Hlf and Hoxa9 (Fig. 2G and S2I-J), suggesting incipient specification of AGM-like MPP and pre-HSC, albeit with incomplete programming.

Altogether, single-cell RNA-seq data suggest that hGx specifies YS-like and AGM-like progenitors in a time-dependent manner, with concomitant emergence of hematopoietic-supporting niche cells. The data suggested putative incipient specification of rare candidate MPP or pre-HSC, which we investigated further in comparison with mouse AGM datasets and at a functional level.

We made use of a dataset recently published by one of us (Thambyrajah et al., 2024) that interrogates the mouse AGM at the point of HSC emergence from the dorsal aorta HE. In this study, a combination of surface markers and Gfi1 locus reporting of EHT, tested functionally through transplantation, identified HE and HSC-enriched clusters (Fig. S3A - clusters 1 and 2, respectively, with cluster 4 a putatively heterogenous cluster at the transition between the 2 states). We projected cell transcription profiles from hGx to AGM using a version of scmap algorithm (Kiselev, Yiu and Hemberg, 2018) with 1-cell and 3-cell neighbourhoods for lower and higher levels of confidence on similarity projections (respectively gray and black circles on Fig. S3A - left). The numbers of cluster-specific hGx cells projected onto AGM clusters is quantified in Fig. S3B. Top differentially expressed genes (DEG) between AGM HE and HSC-projected hGx cells are represented in Fig. S3C and Supplemental File S2.

With respect to HE specification, we observed that a subset of hGx C-Kit⁺ cells present in endothelial clusters 5 and 0 (Fig. 2A) projected onto the AGM HE-identity cluster 1 (Fig. S3A), with more frequent similarity amongst hGx 192h cluster 0 (41%) than 144h cluster 5 (21%). Overall, this supports the notion that specification of endothelial and HE cells in hGx follows with time-dependent developmental progression into putative AGM-like HE. Lineage identity genes supported the nature of HE-projected hGx cells (Fig. S3D), which expressed Gata2 but not Runx1, Myb, or Gfi1b, indicating that EHT occurs downstream of these cells. We then asked if any of the hGx cells in hemopoietic clusters bore resemblance to cells from the HSC-enriched AGM cluster 2. We observed that most (85%) hGx CD45⁺ cells in cluster 8 (Fig. 2A) projected onto the HSC-enriched cluster 2, which also showed similarity with 76% of CD41⁺-enriched hGx cluster 4 cells. The relative positioning of hGx cells cluster 4 and cluster 8 progressively further away from the AGM HE cluster was in line with the time-dependent specification in the hGx. However, the lineage marker programs (Fig. S3D) expressed by the hGx cells, although definitive in nature (e.g. there is expression of immunoglobulin chains and adult β-globins), lacked detection of HSC genes like Hlf, Gfi1, or Hoxa9. On the other hand, Mllt3 was present (Calvanese et al., 2019), even in the absence of an erythroid signature (Pina et al., 2008), suggesting a developing pre-HSC program compatible with the co-expression of multi-lineage markers (Fig. S3D).

Late-stage hGx cells can engraft hematopoietic tissues upon maturation

We inspected the cellularity of hGx at the 216h time-point more closely through a combination of multi-parameter flow cytometry (Fig. 3A-B), colony-forming progenitor assays (CFC) (Fig. 3C-D) and in vivo engraftment of haematopoietic tissues (Fig. 3E-G). In order to facilitate the analysis of engrafted animals, we engineered constitutive expression of BFP from the Rosa26 locus in the Flk1-GFP mES cell line (Fig. S4A), which we also used in some of the in vitro analyses. Phenotypic characterisation of 216h-hGx (4 replicates, 1 with Rosa26-BFP::Flk1-GFP line) (Fig. 3A, S4B) showed the presence of pre-HSC-like CD45⁺CD41^loc-Kit⁺VE-Cadherin⁺ cells (1.8% of all CD45⁺CD41^lo cells) (Fig. 3B), corresponding to an estimated 8 cells per hGx. Analysis of the hGx progenitor content using multipotential CFC assays (Fig. 3C-D) revealed the presence of >50 CFC/hGx, of which an average of 7 have mixed-lineage granulocytic-monocytic-erythroid (GEM) potential, numerically matching the estimated number of pre-HSC. Other progenitors were predominantly granulo-monocytic (GM) and erythroid (E) progenitors (Fig. 3C-D). Differentiated cells of G, M and E lineages (Fig. S4C-D) were also detectable in 216h hGx, putatively reflecting the progeny of EMP-like cells specified earlier at the 144h timepoint.

Figure 3

As an ultimate test of the presence of pre-HSC, we tested whether late-stage 216h hGx contained cells capable of hematopoietic engraftment of immunodeficient mice. In early experiments, we transplanted dissociated hGx cells into irradiated immunocompetent C57Bl/6 recipients, sub-lethally irradiated immunodeficient NSG mice, or non-conditioned NSGW41 hosts, but failed to detect signs of engraftment of peripheral blood, spleen (Spl) or bone marrow (BM) at 12 days (including absence of spleen colony formation), at 4-6 weeks (short-term engraftment), or beyond 8-10 weeks (long-term engraftment) by either flow cytometry detection of CD45.2 or by genomic DNA (gDNA) PCR analysis of the Flk1-GFP locus. However, given the absence of clear HSC-like transcriptional programs in 216h CD45⁺ cells, as analysed by scRNA-seq (Fig. 2 and S2), and in light of the late specification of mesenchymal or sympathetic neuronal cells (Fig. 2A and S2D) putatively supportive of HSC formation, we hypothesized that hGx pre-HSC required additional maturation in a relevant supportive niche. We thus tested implantation of undissociated hGx in the adrenal gland of Nude mice (Fig. 3E), a topography commonly used to support tumor development, and recently shown to support adult haematopoietic development (Schyrr et al., 2023). Given the coincidence of CD45 isoform between Flk1-GFP mouse ES cells and Nude mice, we made use of the Rosa26-BFP::Flk1-GFP mES cell line (Fig. S3A). We implanted 3 hGx each unilaterally in the adrenal gland of unconditioned recipient mice. Control mice were injected with an equivalent volume of PBS. Animals were sacrificed at 24 hours, 12 days, and 4 weeks after implantation, with 5 experimental animals and 1 control collected at each timepoint. We checked Spl and BM BFP engraftment at the different timepoints and analyzed the morphology of adrenal glands receiving implants (Fig. 3F). We did not observe BFP engraftment, spleen enlargement or colony development, or macroscopic changes in the adrenal gland at the earlier timepoints. However, 3 of 5 adrenal glands were enlarged, and their histology suggested in situ development of the hGx (Fig. 3F), particularly mesodermal tissues – somite derivatives (Fig. 3F - left, red box), and lateral plate and intermediate mesoderm derivatives, respectively blood cells and renal tubules (Fig. 3F - left, green box). The control, PBS-injected glands showed normal tissue architecture (Fig. 3F - right). It should be noted that none of the specimens contained glomeruli in addition to the renal tubules, pointing to an origin from the hGx, and not from the adjacent kidney tissue. BFP engraftment of the Spl and BM by flow cytometry was very low level albeit consistently above control (Fig. S4E), and we opted for analysing engraftment by PCR analysis of gDNA using primers directed against the BFP insert (Fig. S4A). We measured the sensitivity of the amplification through serial dilution of Flk1-GFP Rosa26-BFP mES cell gDNA into that of human K562 cells, and determined detection of 10pg/100ng gDNA, equivalent to 2-3 cells (Fig. S4F). PCR analysis of spleen and BM gDNA detected BFP amplicon in both tissues in 3 of the 5 samples (#11, 12 and 15) (Fig. 3G). We sampled the gDNA of each recipient 4-5 times and detected the presence of the amplicon in only a fraction of the replicate samples of the positive animals, reflecting the low-level detection by flow cytometry. Replicate sampling of the control was consistently negative (Fig. 3G). Sorting of erythroid Ter119⁺ and lympho-myeloid CD45⁺ cells (Fig. S4G) showed PCR engraftment in both lineage fractions (Fig. S4H).

Altogether, the data suggest that 216h hGx generate AGM-like pre-HSC capable of at least short-term multilineage engraftment upon maturation. The low-level production of engrafting cells recapitulates their rarity in vivo, in agreement with the embryo-like qualities of the gastruloid system, which accommodates modular qualitative bias, but not quantitative changes in tissue specification.

The ability of hGx to reconstitute YS and AGM-like HE and hemopoietic progenitors with temporal coherence lends itself as an attractive model to understand developmental leukemia. In particular, forms of AML unique to the first 2 years of life – infant (inf)AML – have been shown to transform fetal liver (FL) but not adult haematopoietic cells (Waraky et al., 2024; Mercher et al., 2009; Chen, W. et al., 2011; Ragusa et al., 2023), compatible with targeting of a transient embryonic cell. In the case of t(7;12) AML (herein MNX1-r) (Ragusa et al., 2023), which results in a transformation-driving ectopic expression of MNX1 (Waraky et al., 2024), analysis of patient transcriptional signatures (Ragusa et al., 2022) for over-representation of cell atlas-defined affiliations, indicates enrichment in cardiomyogenic, endothelial, HSC/progenitor and mast cells (Fig. 4A). These cell enrichments are compatible with an early hemogenic cell derivative, capturable in the hGx model. Other cell type-enrichments reflect MNX1 functions in pancreatic and neuronal development (Ragusa et al., 2022) (Fig. S5A).

Figure 4

We overexpressed MNX1 (MNX1-OE) in Flk1-GFP mouse ES cells by lentiviral transduction (Fig. 4B) and analyzed progression of hGx generation (Fig. 4C). We used human MNX1 cDNA to distinguish from the endogenous gene, but the degree of homology is nevertheless high (84%), supporting functional equivalence. We confirmed that MNX1 overexpression in hGx is maintained at endpoint (Fig S5B). MNX1-OE hGx activated polarized Flk1-GFP expression and elongated with similar kinetics to empty-vector (EV) control (Fig. 4C), but consistently produced larger hGx (Fig. 4D) denoting increased cellularity (Fig. S5C). From 192h onwards, MNX1-OE hGx had a higher frequency of spontaneously contractile structures (Fig. S5D), compatible with the cardiogenic cell association of patients’ transcriptomes (Fig. 4A). We interrogated the time-dependent hemogenic cell composition of hGx by flow cytometry, quantifying C-Kit, CD41 and CD45 (Fig. 4E-G and S5E-F). We observed a significant expansion of the C-Kit⁺ compartment specifically at 144h (Fig. 4E and S5E). All cell markers were relatively unchanged at later timepoints, although absolute cell numbers were higher in MNX1-OE hGx (Fig. S5C).

To better understand the consequences of MNX1-OE on hemogenic development, we performed RNA-sequencing (RNA-seq) of hGx at 144h and 216h, matching YS and AGM-like hemopoietic waves, respectively (Fig. 4H). We confirmed expression of the human MNX1 transgene in the sequenced reads (Fig. S6A); in contrast, we could not detect endogenous mouse Mnx1 in any of the samples (Fig. S6A), confirming that it does not normally play a role in hemogenic development. DEGs between MNX1-OE and EV (Supplemental File S3) configured distinct MNX1-driven programs at 144h and 216h, with >80% genes timepoint-specific (Fig. 4H). Both programs include up-regulation of hematopoietic-associated genes, capturing EHT and progenitor regulators (e.g. Runx1, and Epor and Zfpm1, respectively) at the earlier timepoint, and more differentiated effectors, e.g. Hba-a1, Hbb-bh1 and Il3ra at 216h, compatible with selective targeting and/or expansion of newly specified progenitors at the YS-like stage (Fig. 4H). An overview of transcriptional regulatory programs at both timepoints denotes a clearer hemogenic enrichment at 144h, with over-representation of GATA2, GATA1 and RUNX1 binding targets (Fig. 4I), putatively capturing an expansion of the HE-to-EMP transition, also represented in enriched GO categories related to VEGF signalling and epithelial-to-mesenchymal transition (EMT) (Fig. 4J). Compatible with the increase in contractile foci in MNX1-OE late hGx, there was an enrichment in cardiogenic gene ontologies (GO) at the 216h timepoint (Fig. 4J). No hemogenic transcription factors (TF)-ChIP or GO categories were enriched in MNX1-OE hGx at 216h (Fig. 4J-I).

We attempted to deconvolute the cellular heterogeneity underlying bulk RNA-seq DEGs by interrogating the cluster signatures obtained from the single-cell analysis of hGx development (Fig. S6B-C; refer to Fig. 2A). In support of enhanced YS-stage hemato-endothelial specification, MNX1-OE transcriptomes showed enrichment of hGx HE cluster 5 and 0 signatures, and of the EMP-like cluster 4 (Fig. S6B). MNX1-OE DEGs at both timepoints also enriched for the relatively less differentiated cluster 10, which predominantly captures unfractionated hGx cells at 120h (Fig. S6B-C; refer to Fig. 2A). Significantly, clusters 5 and 0 HE signatures were enriched in MNX1-r infAML patient signatures (Ragusa et al., 2022) (Fig. S5B-C), positioning MNX1-OE at early stages of hemogenic specification. MNX1-OE DEGs at 216h show enrichment in late-stage clusters 7 and 8 (Fig. S6B-C), denoting enhancement of hGx development, including MLP/MPP in cluster 8, but diverging from MNX1-r infAML signatures. Projection of infAML subtype-specific gene signatures onto hGx development separates the lineage affiliations of MNX1-r from, for example, the most frequent infAML driver – KMT2A (MLL) rearrangements (Fig. S6B-C), which carries myelo-monocytic and/or mixed-lineage affiliation, highlighting the utility of the hGx system for interrogating infAML biology.

We further interrogated the MNX1-r-like leukemia-initiating potential of MNX1-OE hGx through serial replating of CFC assays, a classical in vitro assay of leukemia transformation. We compared 144h and 216h hGx to match functional and transcriptome data, and plated unfractionated hGx, given the transient nature of C-Kit⁺ enrichment. Both 144h (Fig. 5A-B) and 216h (Fig. 5C-D) MNX1-OE hGx replated for at least 5 plates, with significant differences to control (EV) (Fig. 5B,D). EV-hGx replating could not be sustained beyond plate 3 in most experiments started from 216h (Fig. 5C); 144h EV-hGx exhibited some low-level replating until plate 5 (Fig. 5A), but the colonies were formed by few scattered cells clearly distinct from the well-demarcated colonies obtained from MNX1-OE hGx. We analyzed early-stage (plate 1) and late-stage (plate 5) colonies by flow cytometry to identify the nature of the replating cells. Cells selected through replating were C-Kit⁺ (Fig. 5E), matching the cell phenotype transiently enriched at 144h. Early replating of MNX1-OE hGx also enriched for erythroid-affiliated Ter119⁺ cells particularly from the 144h time-point (Fig. 5F), reflecting the differential erythroid potential of EMP and MLP/MPP at 144h and 216h, respectively. This enrichment is not sustained upon replating, compatible with the undifferentiated nature of MNX1-r infAML, which is also reflected by the low levels of CD11b, a marker affiliated with the myelo-monocytic lineages (Fig. 5E). Notably, the cells obtained are not CD45⁺ (Fig. 5F), suggesting perpetuation of an early hemogenic state.

Figure 5

We performed RNA-seq analysis of MNX1-OE hGx CFC cells obtained through re-plating and integrated the data with the 144h and 216h hGx time-points to identify genes enriched through the process of transformation. We observed relative enrichment in HsMNX1 reads in MNX1-OE CFC samples, supporting selective expansion of transduced cells (Fig. 6A). Hierarchical clustering of all DEGs in pairwise comparisons (Fig. 6B), identified a subgroup of genes up-regulated upon CFC replating which was also up-regulated at MNX1-OE hGx at 144h (K14), thus matching the putative time-dependent enrichment of the MNX1-OE propagating population. To understand the relevance of these genes in MNX1-r leukemia, we performed Gene Set Enrichment Analysis (GSEA) of MNX1-r patients in comparison to other AML in the same infant (0-2) age group. K14-genes were enriched in MNX1-r all comparisons (Fig. 6C), with considerable overlap between leading-edge (i.e. significantly-enriched) genes (Fig. S6D). Interrogation of cell-type-specific signatures within the K14 MNX1-r leading edge genes (Fig. 6D) against the single-cell PanglaoDB atlas captured similarities to hemogenic and non-hemogenic cell types, including interneuron, and other neural, signatures, which reflect MNX1 biological functions (Harrison et al., 1999; Thaler et al., 1999; Ragusa et al., 2022). The most highly significant hemogenic signatures capture early developmental components, such as ‘endothelial cells’, which were specifically up-regulated within MNX1-OE hGx (Fig. S6B). ‘Erythroid cell precursors’ also reflect MNX1-OE hGx effects (Fig. 4E-J). We were intrigued by the ‘mast cells’ affiliation of MNX1-r AML signatures within the hGx K14 cluster, which was more specifically aligned with MNX1-OE hGx isolated at 144h (Fig. S6E). Given the characteristic morphology of these cells, we analysed Giemsa-Wright-stained cytospins of CFC replating assays (Fig. 6E and S6F). Consecutive plates of 144h MNX1-OE-initiated in vitro transformation showed a shift from blast-like at plate 3 to precursor cell morphology with some differentiated mast cells at plate 5 (Fig. S6F). However, the extent of differentiation was greater, and apparent earlier, in 216h-MNX1-OE initiated colonies (Fig. 6E). This suggests that the MNX1-OE-responsive hemogenic C-Kit⁺ progenitor may experience maturation between 144h and 216h, and is more likely targeted for transformation at the YS EMP stage. The data support the utility of the hGx in dissecting ontogeny of embryonic origin in hematological cancers, allowing us to specifically narrow down the initiating cellular targets of MNX1-r AML.

Figure 6

Discussion

In this study, we developed and characterized a hGx model from mouse ES cells, which successfully recapitulates YS-like and AGM-like waves of definitive blood progenitor specification, and can engraft adult hematopoietic tissues upon in vivo maturation. We took advantage of the hGx’s nearly full recapitulation of embryonic hemogenic specification, including critical early stages of HE production and EHT to EMP and MLP/MPPs, to interrogate the biology of MNX1-r AML, a poorly characterized form of leukemia exclusive to the infant age group (Ragusa et al., 2023). We showed that MNX1-OE transiently expanded a C-Kit⁺ cell with hemato-endothelial characteristics at the YS-like stage, and could selectively sustain the phenotype through serial replating of CFC assays. Replating CFCs could be initiated beyond the YS-like stage, albeit at the relative expense of an early differentiation state and of the similarity to patient transcriptome, putatively defining a window of susceptibility to MNX1-driven transformation. The hGx MNX1-OE propagating cell bears transcriptional resemblance to patient MNX1-r AML, overall positioning the hGx as an attractive tractable model in which to study developmental leukemias.

MNX1-r has been difficult to characterize due to a combination of rarity, diagnostic difficulty (the underlying t(7;12) can be missed if not specifically looked for) (Tosi et al., 2015), and the lack of a fusion protein that can be ectopically expressed and tracked through development, as is the case for other translocations unique to infAML (Alexander and Mullighan, 2021). However, a recent paper using MNX1-OE has shown that MNX1 can initiate a transplantable leukemia in immunodeficient mice by targeting FL, but not BM, cells (Waraky et al., 2024). This suggests that ectopic expression of MNX1 transforms an embryonic cell, rather than depends on an embryonic niche. It also suggests that MNX1-OE can cause leukemia as a single genetic hit, although it cannot be excluded that other mutations are required for, and may indeed have contributed to, full transformation in vivo, as observed in a recent report confirming the in utero origin of the leukemia (Bousquets-Muñoz et al., 2024). However, the mouse model does not address the nature of the embryonic cell targeted by MNX1-OE. Like the MNX1-OE hGx cells selected through serial replating, MNX1-OE mouse leukemia are also C-Kit⁺ (Waraky et al., 2024). Interestingly, clusters of genes up-regulated in the mouse MNX1-OE leukemia (Fig. S7A – K1, 2, 5, 6, and 9) do not robustly separate MNX1-r from other infAML by GSEA analysis (Fig. S7B), and have diverged more from the patient data in their cell-type signature analysis than the original MNX1-OE cells in the FL (Fig. S7C), putatively selecting a more generic program of leukemia transformation. We propose that the divergence from MNX1-r infAML signatures increases from MNX1-OE hGx 144h (YS-like), to 216h (AGM-like) to FL, positioning the origin of the AML at the YS stage, likely at the HE-to-EMP transition. This cell may persist in the hGx culture system to the 216h AGM-like time-point, and migrate to the FL in the embryo, eventually extinguishing through differentiation, as suggested by the MNX1-OE hGx CFC assays. Persistence of the original MNX1-OE target cell will determine the sensitivity to MNX1-OE, and can be addressed in the hGx system through lineage tracing. Whether additional mutations, namely a reported association between MNX1-r and tri(19) (Espersen et al., 2018; Ragusa et al., 2023), contribute to prolonging the undifferentiated state of the candidate YS target cell with increased probability of transformation, or otherwise expand the targeted cell in a more classical second-hit fashion, can potentially also be screened in the hGx system, in combination with transplantation experiments. Finally, although our study has positioned the MNX1-OE target cell within the YS-EMP stage, the exact nature of the cell requires better definition. The mast cell affiliation, transcriptional and by CFC morphology, is more likely to reflect the period of susceptibility to MNX1 and the multilineage potential of the target cell, than the lineage commitment of the latter. Indeed, blast cells in the leukemia mouse model are morphologically undifferentiated (Waraky et al., 2024) and a mast cell morphology has not been described in patients (Ragusa et al., 2023; Espersen et al., 2018). Although adult HSC have mast cell potential in vitro, this is not employed in vivo under physiological conditions (Chia et al., 2023). Tissue resident mast cells are formed in the embryo at 2 periods: the YS EMP stage, which we propose to have targeted with MNX1-OE in the hGx model; and a transient AGM FL-HSC population which does not persist in post-natal life (Chia et al., 2023; Yoshimoto et al., 2022), and that may be incipiently specified at the end of the current hGx protocol. Prospective isolation of these cell types from embryos and over-expression of MNX1 followed by transformation in vitro and in vivo, will precisely define the embryonic time-window of susceptibility to MNX1-r and the nature the leukemia-initiating cell. Harnessing the high-throughput nature of the hGx culture system will help identifying therapeutic vulnerabilities that can improve the prognosis of this often-fatal disease.

The hGx model furthermore constitutes an advanced platform to probe physiological development of the hematopoietic system. Currently, in vitro platforms do not recapitulate in vivo blood formation sufficiently faithfully (Dijkhuis et al., 2023; Fidanza and Forrester, 2021). Establishment of a tractable in vitro model of hematopoietic development which faithfully recapitulates time-dependent emergence of hematopoietic progenitors in their physiological niche would propel translational applications by enabling high-throughput studies of HSC emergence. Most ESC-based models of blood specification do not achieve concomitant recapitulation of the surrounding microenvironment. Protocols typically remove ESC-derived hemogenic cells from an early embryonic-like niche to direct differentiation through genetic manipulation (Sugimura et al., 2017) and/or empirically-determined growth factor addition (Lis et al., 2017; Vo and Daley, 2015; Fowler et al., 2024; Nafria et al., 2020; Sroczynska et al., 2009; Piau et al., 2023), with minimal probing of intermediate steps. When hematopoietic waves have been specifically inspected, emergence of YS and AGM-like stages were conflated in time (Pearson et al., 2015; Murry and Keller, 2008), suggesting a bias towards intrinsic hematopoietic cell potential, at the expense of cell fate modulation by time-coherent extrinsic physiological cues, affecting the nature of cellular output. The present hGx system, in contrast, successfully deconvolutes YS-like and AGM-like stages of blood production, including capturing differences in the HE generation, suggesting a level of organisation amenable to further development into an integrated hemogenic niche. Although the system is shy of achieving HSC maturation in situ, it can mature in a supportive microenvironment, in this case provided by the adrenal gland (Schyrr et al., 2023), to produce engrafting multilineage progenitors, and potentially HSCs. This suggests that access to the relevant hematopoietic molecular programs is preserved. Incipient specification of non-hematopoietic tissues shown to support HSC emergence, including PDGFRA⁺ mesenchymal (Chandrakanthan et al., 2022) and neuro-mesenchymal (Miladinovic et al., 2024) cells, and neural-crest derivatives (Kapeni et al., 2022; Fitch et al., 2012), occurs in sync with development of AGM-like hematopoiesis, to indicate the presence of cues that with further maturation may facilitate HSPC specification at the appropriate developmental time.

We have observed that the nature of the symmetry-breaking pulse affects the specification of hemato-endothelial programs, with a more robust Flk1/Kdr activation in the presence of Activin A, potentially balancing mesodermal derivatives and the anterior-posterior axis. The hematopoietic output of cardiogenic gastruloids (Rossi et al., 2022) does not seem to require Activin A, but their blood formation is of lower efficiency, at least at the comparable timepoints of 144 and 168h, and apparently biased towards erythroid formation, reflecting its YS, and potentially primitive, nature. In the absence of additional cytokines after the CHI99021 pulse, standard gastruloid protocols (Beccari et al., 2018) capture a similar transcriptional erythrogenic wave. Although Rossi et al. (2022) reported the formation of spleen colonies, these are difficult to ascertain from the data published, and the engraftment capacity of cardiogenic gastruloid-derived blood remains uncertain. Our hGx system has more robust CD41⁺ cell production at 144-168h, which may reflect distinct tissue induction by the CHI99021+Activin A pulse. Interestingly, through extending the hGx culture, and in addition to AGM-like hematopoiesis, we also captured the potential to form renal tubular structures upon in vivo maturation, suggesting induction of lateral plate and intermediate mesoderm from which the AGM region derives, and putative reconstitution of the intra-embryonic hematopoietic niche. An additional challenge is tissue organisation: non-hematopoietic gastruloid models benefitted from mechanical, or chemo-mechanical, support of artificial ECM embedding (Veenvliet et al., 2020; van den Brink, Susanne C et al., 2020), to match temporal coherence and tissue polarisation, with detailed tissue architecture illustrated by somite formation. Prior to HSC emergence in the AGM, critical signals such as SCF and Shh, which we have delivered to the cultures, are secreted by different cells on opposite dorsal-ventral locations (Souilhol et al., 2016), indicating the potential relevance of tissue organisation to improve and potentially extend hematopoietic formation. Moreover, HSPC formation in the AGM is facilitated by circulatory shear stress (North et al., 2009; Diaz et al., 2015; Lundin et al., 2020), suggesting that introduction of flow may improve hemato-endothelial organization and boost blood production.

Over the last decade, the effort to achieve robust hematopoietic production from PSC, mouse and human, has been centred on the generation of HSC, a critical translational need (Chang et al., 2023; Dijkhuis et al., 2023; Fidanza and Forrester, 2021). There are good indications that this is achievable, both with (Vo and Daley, 2015; Dang et al., 2002) and without (Piau et al., 2023) genetic manipulation. However, success remains episodic, mainly because mechanistic understanding of processes underpinning blood generation during development is limiting, exacerbated by limiting access to human embryonic and fetal material. On-going optimization of gastruloid models and the success at integration of organoid cultures in microfluidic systems (Saorin, Caligiuri and Rizzolio, 2023; Quintard et al., 2024) will allow us to further optimize hGx culture. In conclusion, the hGx represents a faithful in vitro reference to replace and complement embryo studies, holding promise to understand the biology of healthy and malignant developmental haematopoiesis and advance blood production for clinical use.

Experimental procedures

Materials

Methods

Cell culture

Flk-1-GFP (Jakobsson et al., 2010), Flk-1-GFP::Rosa26-BFP, Sox17-GFP (Niakan et al., 2010), T/Bra::GFP (Fehling et al., 2003), and E14Tg2A (Hooper et al., 1987) mouse embryonic stem cells (mES) lines were cultured in ES+LIF medium in gelatinized (0.1% gelatin) with daily medium change, as previously described (Turner et al., 2017). The ES+LIF medium contained 500 ml Glasgow MEM BHK-21 (Gibco), 50 ml of fetal bovine serum (FBS, Embryonic Stem Cells tested, biosera, Nuaillé, France), 5 ml GlutaMAX supplement (Gibco), 5 ml MEM Non-Essential Amino Acids (Gibco), 5 ml Sodium pyruvate solution (100 mM, Gibco), and 1 ml 2-Mercaptoethanol (50mM, Gibco). Murine LIF (Peprotech) was added at 1000 U/ml. HEK293T cells were grown in Dulbecco’s modified Eagle medium (Gibco) supplemented with 10% FBS (Gibco) and 1% penicillin/streptomycin (Gibco). All cultures were kept at 37°C and 5% CO₂.

Hemogenic gastruloids assembly, culture, and dissociation

mES were maintained in ES+LIF medium and transferred to 2i+LIF (containing Chiron and MEK inhibitor PD03) for 24 hours prior to the assembly into gastruloids. 250 untransduced mES cells (or 400 for GX-MNX1 and GX-EV) were seeded in each well of a U-bottom, cell-repellent 96-well plate (Greiner Bio-One, Stonehouse, UK) in 40 μl of N2B27 medium [Takara Bio or home-made as Cold Spring Harbour Protocols ‘N2B27 Medium’ (2017)]. The plate was centrifuged at 750 rpm for 2 minutes to promote deposition and aggregation of the cells and was then incubated at 37°C, 5% CO2 for 48 hours. After 48 hours, 150 μl of N2B27 medium supplemented with 100 ng/ml Activin A (QKine, Cambridge, UK) and 3 μM chiron (Peprotech) was added to each well. At 72 hours, 150 μl of medium were removed, without disrupting the gastruloids in the wells. 100 μl of N2B27 with 5 ng/ml of Vegf and Fgf2 each (Peprotech) were added to each well. From 72 h to 144 h, each day 100 μl of medium were removed and replaced with N2B27 + Vegf + Fgf2. At 144 h, the medium was further supplemented with Shh at 20 ng/ml. From 168 h to 216 h, the medium was N2B27 + 5 ng/ml Vegf, plus 20 ng/ml mTpo, 100 ng/ml mFlt3l, and 100 ng/ml mScf (Peprotech). To dissociate cells from the gastruloid structures, medium was removed and individual gastruloids were collected using a pipette and precipitated at the bottom of a microcentrifuge tube. The remaining medium was aspirated and the bulk of gastruloids was washed in PBS. 50 μl of TrypLE express was added to pelleted gastruloids to be incubated at 37°C for 2 minutes to dissociate cells.

Animals and implantation

Female athymic nude-Foxn1^nu (nu/nu) mice (Envigo, Bresso, Italy) were housed under pathogen-free conditions. In accordance with the “3Rs policy”, experiments were reviewed and approved by the Animal Welfare Body (OPBA) of IRCCS Ospedale Policlinico San Martino and by the Italian Ministry of Health (n. 883/2020-PR).

Intact hGx (3 per mouse in 10 μL of PBS) were inoculated in the adrenal gland of five-week-old mice. Briefly, mice were anesthetized with a mixture of xylazine (ROMPUN, Bayer) and ketamine (LOBOTOR, Acme S.r.l.) and injected with hGx, after laparotomy, in the capsule of the left adrenal gland, as previously described (Brignole et al., 2023; Pastorino et al., 2003). All mice survived surgery. Control mice (CTR) received vehicle only. Mice body weight and general physical conditions were recorded daily. One month after hGx inoculation, mice were sacrificed and spleen and bone marrow (BM) from femurs collected. BM cells and spleens, mechanically reduced to cell suspension, were stored at -80°C until use. Adrenal glands were paraffin-embedded for subsequent IHC studies.

Methylcellulose colony forming assays (CFC)

Disassembled gastruloids cells were plated on Mouse Methylcellulose Complete Media (R&D Systems). Cells were first suspended in Iscove’s Modified Dulbecco’s Medium (IMDM, Gibco) with 20% FBS (Gibco) before addition to the methylcellulose medium. Cells were plated in duplicate 35 mm dishes with 2×10⁵ cells/plate. Plates were incubated at 37°C and 5% CO2 for 10 days, when colonies were scored. For serial replating experiments, cells in methylcellulose were collected and washed in phosphate buffer saline (PBS) to achieve single-cell suspensions and replated as described above. Dissociated gastruloid cells from CFC were collected as cell suspensions in PBS and centrifuged onto slides (Shandon Cytospin 2 Cytocentrifuge) at 350rpm for 5-7min, with a PBS-only pre-spin at 1000rpm, 2min. Slides were stained in Giemsa-Wright stain for 30 seconds followed by addition of phosphate buffer pH 6.5 for 5 min.

Immunofluorescence staining

Immunostaining of whole gastruloids was performed as described before (Baillie-Johnson et al., 2015). Briefly, gastruloids were fixed in 4% paraformaldehyde (PFA) dissolved in PBS for 4 hours at 4°C on orbital shaking and permeabilised in PBSFT (10% FBS and 0.2% Triton X-100), followed by one hour blocking in PBSFT at 4°C on orbital shaking. Antibody dilutions were made in PBSFT at 1:200 for primary and 1:500 for secondary antibodies. Antibody incubations were performed overnight at 4°C on orbital shaking, and subjected to optical clearing overnight in ScaleS4 clearing solution. Individual gastruloids were then mounted on glass coverslips by pipetting as droplets in ScaleS4 and DAPI nuclear stain.

Histology and H&E staining

Samples were fixed with 10% formalin at room temperature for 48 hours followed by an ethanol series of ethanol and paraffin; ethanol 70% (90 minutes), ethanol 80% (120 minutes), ethanol 95% (180 minutes), 3 times ethanol 100% of 120, 120 and 180 minutes, 50% ethanol/paraffin (3x120 minutes) and 3 times paraffin (60, 90 and 120 minutes) before embedding in histological cassettes. Blocks were sectioned with a HistoCore BIOCUT Microtome (Leica) at 10 mm and slides were stained with Haematoxylin/Eosin.

Imaging

Images of cultured gastruloids and CFC plates were captured using the Cytation 5 Cell Imaging Multi-Mode Reader (Biotek) plate reader using bright field and FITC channels. ImageJ (Schneider et al., 2012) was used for gastruloid size quantification. Confocal microscopy was performed on LSM700 on a Zeiss Axiovert 200 M with Zeiss EC Plan-Neofluar 10x/0.30 M27 and Zeiss LD Plan-Neofluar 20x/0.4 M27 objective lens. Tissue sections were imaged with a THUNDER Imager 3D microscope system (Leica).

Lentiviral vector packaging and transduction

The lentiviral overexpression vector pWPT-LSSmOrange-PQR was used to clone the MNX1 gene cDNA. The viral packaging vectors pCMV and pMD2.G, described in Pina et al. (2008), were used to assemble lentiviral particles using HEK293T cells via transfection using TurboFect Reagent (Invitrogen). Transduction of mES cells was performed overnight by addition of lentivirus to culture medium and washed the following day (Moris et al., 2018). Transduced cells were sorted for positivity to LSSmOrange.

Flow cytometry

Surface cell marker analysis was performed by staining using the antibodies listed in Key Resources Table. Disassembled gastruloids cells were resuspended in PBS, 2% FBS and 0.5 mM EDTA and stained at a dilution of 1:100 for primary antibodies for 20 minutes at 4°C. When indicated, streptavidin was added at a dilution of 1:200. Analysis was performed on ACEA Novocyte (Agilent) or AttuneNxT (Thermo) analyzers, using the respective software packages. Cell sorting was performed using a CS&T calibrated BDFACS Aria III system (488nm 40 mW, 633nm 20 mW, 405nm 30 mW, and 561nm 50 mW), set with the 100μm nozzle at 20PSI and a 4-way purity mask or, in the case of hGx adrenal implants, a Beckman Coulter CytoFlex SRT (488nm 50 mW, 633nm 100 mW, 405nm 90 mW, and 561nm 30 mW), set with 100μm nozzle at 15PSI, also using 4-way purity. Single-cell deposition in 96-well plates was performed using single-cell sorting mode. Intact cells were gated on FSC-A vs SSC-A plot, followed by doublet exclusion on FSC-A vs FSC-H and SSC-A vs SSC-H, prior to gating on fluorescent parameters for the markers described in the results.

Single cell RNA sequencing of time-resolved hemogenic gastruloids

Gastruloids were collected at different timepoints of the protocol, disassembled and FACS deposited into 96-well plates, either as unsorted (global) or sorted by CD45, CD41 and C-Kit/ScaI markers (sorted). RNA from the cells were extracted from single cells using Smart-seq2 technology at 500Kb and depth of 151 bases. Sequencing reads were quality-checked using FastQC (v0.11.9). We trimmed the samples using trimGalore! (0.6.6) with a cutoff of 30, clipping 15 base pairs and retaining reads of more than 100 bases. Alignment was performed on STAR (2.7.8a) and Mus musculus annotations from GENCODE vM26. Aligned BAM files were annotated using featureCounts (v2.0.1) and count matrices were computed in python by directly accessing the BAM files with gtfparse packages and collapsing lanes.

For quality control, based on the histogram of counts and multimodality distributions, we set a minimum count threshold of 200000 counts, a minimum threshold of 1000 expressed genes, and a maximum threshold of 20% mitochondrial fraction per cell. We performed the same procedure of the second dataset with a more restrictive threshold of 400000 counts and similar expressed genes and mitochondrial thresholds. Cells that did not pass the quality control metrics were omitted from analysis. We normalized the cells to the mean count number per dataset and applied a plus-one-log transformation of the data before proceeding to the downstream analysis.

Dimensionality reduction was performed on feature selection of the gene space using the function scanpy.highly_varying_genes with default parameters. Selection of principal components in principal component analysis (PCA) was performed by heuristic elbow method. Nearest neighbor analysis was constructed by KNN graphs using a correlation metric and 10 nearest neighbors. Data was projected for low dimensional visualization using the UMAP algorithm with default parameters as implemented in scanpy.tl.umap. We used the leiden algorithm as implemented in scanpy.tl.leiden to partition the data into clusters. In order to assess the election of the resolution parameter we used Newman-Girvan modularity as a metric of clustering quality. Differential expressed genes were computed comparing each cluster against the rest using the Wilcoxon test with Benjamini-Hochberg correction.

Annotation and projection to additional datasets

Raw counts matrices from Thambyrajah et al. (2024) were downloaded and processed following the same pipeline as described above for scRNA-seq of gastruloids to generate UMAP and clustering. We implemented the scmap algorithm to compare our scRNA sequencing of gastruloids with the available datasets. We reduced the dimensionality of the space by selecting highly varying genes from the annotated dataset. Then, we constructed a KNN classifier with correlation metric and computed the nearest neighbors of the target data. If neighbors with correlation metrics below 0.7 default standards, the projected cells were not projected onto any cell from the annotated dataset. To visualize the cells over the UMAP plots of the other datasets, we constructed a KNN regressor with three neighbors and a correlation metric. Confusion matrices were used to visualize the overlap between gastruloid clusters and the embryo dataset assigned clusters using scmap.

Bulk RNA sequencing

Total RNA was extracted from disassembled gastruloid cells at 216 hours. Sequencing was performed on NovaSeq PE150 platform, at 20M paired-end reads per sample. Tophat2 with Bowtie2 were used to map paired-end reads to the reference Mus musculus genome build GRCm39. GENCODE Release M30 (Frankish et al., 2019) was used as the reference mouse genome annotation. Aligned reads were filtered by quality using samtools (Li, H. et al., 2009) with a minimum threshold set at 30 (q30). Transcript assembly and quantification was achieved using htseq (Putri et al., 2022). Differential expression between sample and control was performed by collapsing technical replicates for each condition using Deseq2 (Love, Huber and Anders, 2014) in R environment (Deseq2 library v 1.32.0). The differential expression was expressed in the form of log2 fold change and filtered by false discovery rate (FDR) of 0.05.

Gene list enrichment analyses

Gene ontology (GO) analysis was performed in ExpressAnalyst (available at www.expressanalyst.ca) using the PANTHER or GO Biological Process (BP) repository. GO terms and pathways were filtered by false discovery rate (FDR) with a cut-off of ≤ 0.1 for meaningful association. EnrichR (Xie et al., 2021; Kuleshov et al., 2016; Chen, E. Y. et al., 2013) was used for cell type analysis using the Panglao DB (Franzén, Gan and Björkegren, 2019) databases using a FDR threshold of ≤ 0.1 or p value ≤ 0.01, where specified, as well as transcription factor (TF) binding site enrichment using the ENCODE and ChEA Consensus TFs from ChIP database.

Gene Set Enrichment Analysis (GSEA)

Custom gene signatures were used as gene sets for GSEA analysis (Subramanian et al., 2005) on the GSEA software v4.2.3 on RNA sequencing expression values in counts units. GSEA was ran in 10000 permutations on gene set using the weighted Signal2Noise metric. Enrichment metrics are shown as normalized enrichment score (NES) and filtered by FDR ≤ 0.05. Leading edge genes (LEGs) are genes with a “Yes” values for core enrichment. For AML patient analysis, clinical phenotype and expression data (in counts units) were extracted from the GDC TARGET-AML cohorts in the Therapeutically Applicable Research to Generate Effective Treatments project (TARGET, https://ocg.cancer.gov/programs/target), downloaded from the University of California Santa Cruz (UCSC) Xena public repository (last accessed 31st August 2022). Patient samples were selected according to the reported karyotype to include t(7;12), inv(16), MLL, normal karyotype, and t(8;21). GSEA was performed comparing RNA sequencing counts of t(7;12) samples against pooled AML subtypes (inv(16), MLL, normal karyotype and t(8;21)) as “other AML”.

Real-time polymerase chain reaction (qPCR)

Extracted RNA was reverse-transcribed into complementary DNA (cDNA) using High-Capacity RNA-to-cDNA Kit (Applied Biosystems). QPCR was performed using FastGene 2x IC Green Universal qPCR Mix (Nippon Genetics, Duren, Germany) using primers for human MNX1 (forward 5-GTTCAAGCTCAACAAGTACC-3; reverse 5-GGTTCTGGAACCAAATCTTC-3) (Gulino et al., 2021) and Ppia for endogenous control (forward 5-TTACCCATCAAACCATTCCTTCTG-3; reverse 5-AACCCAAAGAACTTCAGTGAGAGC-3) (Moris et al., 2018). Differential gene expression was calculated using the delta delta Ct (ΔΔCt) method.

Polymerase chain reaction (PCR)

Genomic DNA was extracted using Monarch Genomic DNA Purification Kit (New England Biolabs, Hitchin, UK) using manufacturer’s instructions. PCR on genomic DNA was performed using Phire PCR Master Mix (Thermo Scientific) to amplify the BFP locus using forward primer 5’-GCACCGTGGACAACCATCACTT-3’ and reverse primer 5’-CAGTTTGCTAGGGAGGTCGC-3’.

Statistical analysis

Experiments were performed at least in triplicates, unless specified otherwise. Data are plotted to include standard deviation (+/-SD) between replicates. Statistical significance was set at a threshold of p value < 0.05. Statistical analysis was performed in R environment (version 4.1.3) or using GraphPad Prism 8.0 software and is detailed in respective figure legends.

Data availability

Raw data as well as processed count matrices and post-processed files from single-cell RNA-seq for the time-resolved data is available at E-MTAB-12148. Bulk RNA-seq of MNX1 overexpressing gastruloids is available at Array Express with accession code E-MTAB-12173. The post-processing was performed in Python on DockerHub: dsblab/single_cell_analysis:0.5. Scripts are available in https://github.com/dsb-lab/blood_gastruloids and Zenodo (https://doi.org/10.5281/zenodo.7053423). The results published here are partly based upon data generated by the Therapeutically Applicable Research to Generate Effective Treatments (TARGET) (https://ocg.cancer.gov/programs/target) initiative, of the Acute Myeloid Leukemia (AML) cohort GDC TARGET-AML. The data used for this analysis are available at https://portal.gdc.cancer.gov/projects and https://xenabrowser.net/.

Acknowledgements

This project was funded by a start-up grant and a BRIEF award from Brunel University London to CP, by a Little Princess Trust (LPT) Project Grant (CCLGA 2023 22 Pina) to CP, and by an ERC Advanced Grant (MiniEmbryoBlueprint 834580) to AMA. LPT research is funded in partnership with CCLG through the CCLG Charity Research Network. DR received funding from the Royal Society of Biology (MRSB Travel Grant). GTC was funded by grant FPU18/05091 from the Spanish Ministry of Universities. AJ is funded by a Lady Tata Memorial Trust International Scholarship (2022). SvdB was funded by an EMBO Postdoctoral Fellowship (ALTF 195-2021) and is in receipt of a HFSP Postdoctoral Fellowship (LT0047/2022-L). Work in the CP lab was also funded by a KKLF Intermediate Fellowship (KL888), a Leuka John Goldman Fellowship for Future Science (2017-2019), and a Wellcome Trust / ISSF Bridge Funding award at the University of Cambridge (2019). JGO acknowledges financial support from the Spanish Ministry of Science and Innovation and FEDER (grant PGC2018-101251-B-I00), by the Maria de Maeztu Programme for Units of Excellence in R&D (grant CEX2018-000792-M), and by the Generalitat de Catalunya (ICREA Academia programme). MP acknowledges funding from the Italian Ministry of Health (Ricerca Finalizzata 5 per mille and Ricerca Corrente to MP). Library preparation and next-generation sequencing for single-cell RNA-seq analysis were performed by the Single Cell Genomics Group at the National Centre for Genomic Analysis – Centre for Genomic Regulation (CNAG-CRG), Barcelona. The Authors wish to acknowledge Tina Balayo, Ana Filipa Domingues, Shikha Gupta, Oliver Davies, Kristen Place and Remisha Gurung’s technical support at different stages of the project.

Additional information

Author contributions

Conceptualization: CP, AMA; Methodology: CWS, DR, GTC, FP, SvdB, MC,MP, KRK, VH-H, AB, JGO, AMA, CP; Software: GTC, JGO; Validation: CWS, DR, AJ, YC, SvdB, CP; Investigation: CWS, DR, GTC, FP, AJ, YC, LD, CB, G-AI, CP; Formal Analysis: GTC, DR, JGO, CP; Resources: GTC, JC, MP, JGO, AMA; Data curation: DR, GTC, JGO; Writing – Original Draft: CP, DR, GTC; Writing – Review and editing: CP, DR, GTC, AB, JGO, AMA; Visualisation: DR, GTC, CWS, CP; Supervision: CP; Project administration: CP, AMA, JGO; Funding acquisition: CP, AMA.

Additional files

Supplemental Figures

Supplemental File S1

Supplemental File S2

Supplemental File S3