Tracking early mammalian organogenesis – prediction and validation of differentiation trajectories at whole organism scale

ABSTRACT Early organogenesis represents a key step in animal development, during which pluripotent cells diversify to initiate organ formation. Here, we sampled 300,000 single-cell transcriptomes from mouse embryos between E8.5 and E9.5 in 6-h intervals and combined this new dataset with our previous atlas (E6.5-E8.5) to produce a densely sampled timecourse of >400,000 cells from early gastrulation to organogenesis. Computational lineage reconstruction identified complex waves of blood and endothelial development, including a new programme for somite-derived endothelium. We also dissected the E7.5 primitive streak into four adjacent regions, performed scRNA-seq and predicted cell fates computationally. Finally, we defined developmental state/fate relationships by combining orthotopic grafting, microscopic analysis and scRNA-seq to transcriptionally determine cell fates of grafted primitive streak regions after 24 h of in vitro embryo culture. Experimentally determined fate outcomes were in good agreement with computationally predicted fates, demonstrating how classical grafting experiments can be revisited to establish high-resolution cell state/fate relationships. Such interdisciplinary approaches will benefit future studies in developmental biology and guide the in vitro production of cells for organ regeneration and repair.


Advance summary and potential significance to field
In the manuscript entitled "Tracking Early Mammalian Organogenesis -Prediction and Validation of Differentiation Trajectories at Whole Organism Scale", Imaz-Rosshandler and colleagues built on their previous data and now provided a thorough transcriptomic characterization of murine embryonic development covering the critical window of time of early organogenesis.In fact, to the already publish atlas of single cell transcriptomes from mouse embryos between E6.5 and E8.5, they have now added more than 300,000 transcriptomes from single cells isolated between E8.5 and E9.5, always sampled in 6-hour intervals.This unique dataset was then exploited to infer cell trajectories and make cell fate predictions using the probabilistic model of Waddington-OT (W-OT).Via this bioinformatic analysis the authors found evidence of the existence of endotome in the murine embryo.Finally, the authors validated in vivo some of the predictions via the orthotopic grafting of specific primitive streak regions.Overall, the manuscript is of clear interest to the developmental biology community and, while the authors focus their analysis on the development of the hemato-endothelial system, the data presented will be very useful also to scientists working on different organs.While this reviewer is very supportive, there are some points that need to be addressed.

Comments for the author
Generally speaking, it would greatly benefit the manuscript if some important details were provided already in the text and not in the method section.A glaring example is figure 3d which is unreferenced in the main text, but instead is explained briefly in the methods (page 16).In addition, this manuscript includes a minimal number of references.Despite the limitations of space, etc, I encourage the authors to include many more references, including for genes that appear as markers of specific cell types (one example, but I can cite many others, when referring to Msx2).As literature review was pivotal for cluster annotations (Fig S1), this reviewer thinks that the literature used should be referenced, maybe as supplementary.Given that this dataset will clearly appeal also developmental biologists working in fields other than hematopoiesis, if there is a case of articles that benefit from an extensive reference list, this is one.
-Fig 1H,Fig S1E,and Materials and Methods "Expansion and refinement of cell population annotations": the literature or markers used for the cell type annotations are not provided or clearly explained.There is a mention of 2 papers in the legend of Fig S1 but it would be recommended to have a supplementary table(s) with the markers defining each cluster and/or FeaturePlots-Violin Plots highlighting the main markers defining each cell type for validation and confirmation.In this new refined and very thorough atlas, it is quintessential to provide more information for how the 88 clusters were annotated.Again, this will clearly help the non-blood-expert reader.
-The same needs to be applied throughout the manuscript.As an example, how the haematoendothelial populations that the authors have isolated for further analysis (Fig 2 ) were defined?How MEP, EMP, BP and HEP were separated?Markers and/or FeaturePlots should be provided.
-Fig 2 .Why were embryo proper endothelial cells, and allantois endothelial cells included in the YS analysis?As they appear to be located at the very origin of the blue trajectory, if these cells are removed from the analysis, does this affect the trajectory results?-Fig S5 ."Lymphoid and microglial-like progenitors can also be detected".This sentence must be revised since all is known from this analysis is that the cells the authors are referring to are cells expressing lymphoid or microglial markers.In the absence of fate experiments, tracing or even analysis of lineage potential, it is not possible to define them as lymphoid or microglial-like progenitors.
-The caption for figure 3a is the same as figure 2a.This should be remedied.
-In Fig S7 : it is mentioned that endothelial landscape revealed 3 different trajectories but this is not clearly presented/easy to see.It is very difficult to guess which are the 3 trajectories the authors are referring to.In addition can the authors add some interpretation of these data?Are those trajectories a result of time and or locations?This seems to be, at least partially, the case however without a clearer indication of the trajectories, the reader is left to guess.For the location point, can the author check if trajectories of EP ECs have a specific HOX code?Lyve1 seems to be a specific marker of YS cells in this dataset, however in E9.5 embryos Lyve1 expression, measured by Lyve1-Cre, is observed in placental vessels, likely allantois derivatives, in a previous report (Lee et al. Cell Reports 2016).Can the author comment on this?Why Meg 3 is an important gene to mention?Mef2c is assigned as a venous gene, however many reports show very high level of expression in arteries, including the dorsal aorta during embryonic development.
The observation regarding the existence of an endotome during mouse development is a very interesting one and would add a very important element of novelty, so this reviewer thinks that a deeper molecular characterization of these cells should be added.Beside the expression of Cxcl12, Pax3, Meox1, Foxc2, Pdgfra and Alcam, can the authors provide a better transcriptomic characterization?What are the DEG between the endotome vs neighbouring clusters (like anterior CPP) as well as ECs? Do the endotome express any endothelial genes or genes related to vasculogenesis/angiogenesis? Bias towards a specific subset of EC (venous vs arterial vs lymphatic)?Can this analysis differentiate between ECs of intersomitic vs other vessels?If so, can the authors predict whether the endotome can contribute to the formers?
Page 7 what does "fetal liver HSC induction" mean, for Hlf? -In the Materials and Methods: "Generating an integrated atlas": the identification of highly variable gene is not well explained.It is not mentioned the function or the parameters used (it is unclear what "as described above" in page 13 means).
-In the Materials and Methods: "Differential gene expression and Canonical Correlation Analysis of Endotome derived VACs": it is not clarified the test used for the findMarkers function or if there is a p-value threshold applied to the selection of genes.
-In the Materials and Methods: "Generating the landscapes of haemato-endothelial trajectories": two thresholds of log odd ratios have been chosen (log odds > 0 and log odds > -1).While the choice of the different threshold is explained, it is usually not a recommended practice changing the threshold in the same setup.Can the authors comment on this?Also the choice and identification of the complexity of a landscape is not well described.
-In Materials and Methods, it lacks a section explaining how the force-directed graphs, UMAPs, etc. were constructed.A vizualisation section here is recommended, similarly to what already provided in their previous paper (Pijuan-Sala et al. 2019) -In Materials and Methods, in many instances the version of the tool used (scran, fastMNN, igraph, Metacell... ) is not clarified.This can affect the reproducibility of the analysis given also that the code is not provided yet.
-The ending paragraph, in particular the last sentence, feels a bit strange and out of place.
-Merely aesthetic comment: Fig 1 .Adjust cartoon to match the image of how the dissection was done.

Advance summary and potential significance to field
In this manuscript by Imaz-Rosshandler et al., authors expanded a previous mouse early developmental atlas with 300,000 sc transcriptomes collected from 8 timepoints, every ~6h, from e8.5 to e9.5, covering organogenesis.Integrating this information with the previous atlas, they contribute the following highlights: 1) an extended atlas of morphogenesis in the mouse (8.5-9.5) 2) an extensive trajectory analysis of early blood and endothelial development 3) new evidence and trajectories for the mouse endotome (contributing to vascular-associated cells, which potentially support definitive hemogenesis).4) a spatially defined primitive stream single-cell atlas 5) a state-fate analysis through orthotopic grafting of primitive streak, with outcomes analyzed through single-cell sequencing This is a collection of technical achievements (which provide new datasets and benchmark methods of interest to the community) as well as biological insights (such as the discovery of a potential endotome-VAC trajectory in mammals).
The manuscript nicely evokes classic experimental approaches and uses modern phenotyping technologies, such as single-cell sequencing, to reach important novel conclusions.Put together, I think both the dataset and the classic fate-mapping analyses make for a very interesting contribution to the field that will reach a broad readership.The cherry-in-the-pie would have been some heterotopic transplants to study fate plasticity, but I understand this might be out-of-scope for the present study.
Comments for the author ****Extended atlas and endotome While some discussion is provided, I think the manuscript would still benefit a lot from better highlighting how it differs from existing published atlases, as there are several other ones available, e.g. the integrative dataset of Qiu et al.Nature Genetics 2022 and the new preprint of Qiu et al. bioRxiv 2023.Where the present manuscript has advantages, this should be highlighted.There is a constant trickle of developmental atlas efforts, and I find it increasingly unclear how they compare to each other.Eventually, this would help establish guidelines/tables that may be very useful for newcomers in the field and any other researchers interested in looking at mouse development to choose the right dataset for their question.
I do not understand the rationale of choosing all blood, blood+endothelium or all endothelium (E9.25/9.5)as the end-point to calculate the W-OT fate matrices.Why not using the individual clusters of interest at the endpoint?Wouldn"t this increase the power of fate predictions and clarify the trajectories, especially for those clusters that only contribute to a rare small population, such as the endotome?
The findings presented in Figures 2-3 are very interesting, but perhaps it would be important to better summarize them.Even a simple graphic representation of the author's interpretation of the manifold and OT analysis would serve to more easily follow this through the text.Overlaying significantly-enriched TFs, predicted GRNs, ligand-receptor-interactions, GO-terms and any other computational findings for that particular trajectory would be a definite plus.
Arguably, the biggest finding in the current data is the connection between the vascular associated cells (VACs), a putative HSC niche during early EHT, and a part of the somitic mesoderm (the "Endotome").This parallels results in zebrafish obtained from multiple independent groups (Nguyen et al., Sahai-Hernandez et al.).It is important to note (and highlight) that the present manuscript only provides an inference of this trajectory, and that there is currently no validation or additional lines of evidence of mammalian somitic-derived endothelial cells (SDECs) contributing to the aorta.Indeed, it is surprising that after almost a decade since the Nguyen et al. manuscript, there is still no lineage-tracing evidence for contribution of somitic mesoderm into the SDECs in mammals.Canonical Cre-based lineage tracing (e.g.Tbx6 or Pax3 Cre lines) should have been able to validate this connection.Recent papers by Catherine Robin (Hubrecht) and Oliver Stone (Uni Oxford) have looked carefully at the endothelial somitic derivatives using these lines (Stone and Stainier Dev Cell 2019, Lupu et al. biorxiv 2023Yvernogeau et al. biorxiv 2020).While the most thorough paper performing Pax3-Cre single-cell RNAseq analysis is yet to be peer-reviewed (Lupu et al.), the data in the preprint does not obviously suggest any contribution of Pax3-Cre cells to the aorta, although I find their singlecell data plots hard to interpret.Interestingly, the microscopy images of sections at e9.5 and e10.5 suggest some Pax3-lineage cells may be located very close to the DA (though not integrated in the monolayer).Similarly, Yvernogeau et al. detected a substantial contribution of Tbx6+ PSM to limb ECs, but very little contribution to visceral and aortic ECs (although it was not 0).Both manuscripts find cells that are peripheral to the aorta, which perhaps resembles the stromal cell populations from Murayama et al. 2023, which share multiple markers with the endotome VACs presented in the current manuscript.Of course, the best thing here would be to integrate the cells from Lupu et al into this atlas, but the dataset is not yet accessible.At the very least, I feel that some additional discussion and citation to these manuscripts is missing.
In chick and zebrafish this "endotome" population is supposed to be bipotent contributing to both muscle and endothelium.Does W-OT make the same prediction in the current dataset?
Is there any evidence here that the endotome-derived endothelium described here expresses hematopoietic-specifying/modulatory molecules?What about ligand-receptor predictions, with the other endothelial cells?****Orthotopic PS transplant & fate analysis The State-fate analysis through orthotopic grafting is really cool and interesting and, although not at single-cell resolution, it does highlight the power of these fate-predictive tools when applied at the population level.I just missed a few more statistical analyses to objectively quantify the accuracy of the predictions.I also do not quite understand why the labels are different in the predicted fates vs. the observed fates (Figure 5 c-d-e-f).I"m sure this makes some sense somehow but it makes it hard on the reader to interpret to what extent the predictions were accurate based on pre-grafted states and OT.
There seem to be some large differences between some of the predictions by morphology and by transcriptome.For instance: paired aortae and neural tube are overrepresented in the morphological analyses, compared to somites, which are overrepresented in the transcriptome analysis.However, it seems to me that the morphological analysis is based on # of embryos with ANY mTom contribution to that compartment, whereas the numbers on Fig 5d are based on total numbers of grafted cells.Is there any way to better compare predictions and fates, using the same metric?**** Minor points: Figures (especially 1-4) are really huge, with very small prints for legends, and tons of labels in some cases.I would highly recommend reformatting these for improving readability.

First revision
Author response to reviewers' comments

Response to Reviewers
We were delighted to read the positive and constructive reviewers" comments, all of which we have addressed in our revised manuscript.Please find below our point-by-point response, with the reviewers" comments in black, and our responses in blue.In the revised manuscript we have highlighted the changes we have made in red text.

Reviewer 1 Advance Summary and Potential Significance to Field:
In the manuscript entitled "Tracking Early Mammalian Organogenesis -Prediction and Validation of Differentiation Trajectories at Whole Organism Scale", Imaz-Rosshandler and colleagues built on their previous data and now provided a thorough transcriptomic characterization of murine embryonic development covering the critical window of time of early organogenesis.In fact, to the already publish atlas of single cell transcriptomes from mouse embryos between E6.5 and E8.5, they have now added more than 300,000 transcriptomes from single cells isolated between E8.5 and E9.5, always sampled in 6-hour intervals.This unique dataset was then exploited to infer cell trajectories and make cell fate predictions using the probabilistic model of Waddington-OT (W-OT).Via this bioinformatic analysis the authors found evidence of the existence of endotome in the murine embryo.Finally, the authors validated in vivo some of the predictions via the orthotopic grafting of specific primitive streak regions.Overall, the manuscript is of clear interest to the developmental biology community and, while the authors focus their analysis on the development of the hemato-endothelial system, the data presented will be very useful also to scientists working on different organs.While this reviewer is very supportive, there are some points that need to be addressed.
Reviewer 1 Comments for the Author: Generally speaking, it would greatly benefit the manuscript if some important details were provided already in the text and not in the method section.A glaring example is figure 3d which is unreferenced in the main text, but instead is explained briefly in the methods (page 16).
We have carefully checked the paper and made several small changes to provide more detail about the methods within the main text.As suggested, we have specifically updated the text to provide more description of the methods behind figure 3d.In relation to this, the text has been reworded to: "Intriguingly, in the Force Atlas representation of the hematoendothelial landscape a subset of endotome cells present in later stage embryos (E9.25-E9.5)are placed next to EMP blood progenitors (Fig. 3d, red box).Additionally, these cells cluster with EMPs (Fig. 3di, cluster 20) when louvian clustering is performed using the top 50 principal components from the hematoendothelial landscape.However, this association is only visually highlighted when a force directed layout is generated on a subset of the atlas.By contrast, clustering, cell type annotation and UMAP embedding over the entire atlas demonstrates these cells have a transcriptional identity aligned with earlier stage endotome cells that are distinctive of EMPs (Fig. 1h and Fig. 3dii)." In addition, this manuscript includes a minimal number of references.Despite the limitations of space, etc, I encourage the authors to include many more references, including for genes that appear as markers of specific cell types (one example, but I can cite many others, when referring to Msx2).As literature review was pivotal for cluster annotations (Fig S1), this reviewer thinks that the literature used should be referenced, maybe as supplementary.Given that this dataset will clearly appeal also developmental biologists working in fields other than hematopoiesis, if there is a case of articles that benefit from an extensive reference list, this is one.
We are grateful for this comment and agree it will help with readability and framing the paper in a broader context.We now provide a new, exhaustive list of references in (Supplementary Table S2).Table S2 includes marker genes that were critical for annotating many of the additional cell types (annotations that go beyond our original Pijuan-Sala et al., 2019 paper) and the references listed highlight the expression patterns of many of these marker genes in developing mouse embryos from the literature.We also include an additional supplementary table (Table S1) that lists all the marker genes that were identified using FindMarkers in Seurat (p<0.05,min.pct= 0.25, only.pos= TRUE, logfc.threshold= 0.25).
-Fig 1H,Fig S1E,and Materials and Methods "Expansion and refinement of cell population annotations": the literature or markers used for the cell type annotations are not provided or clearly explained.There is a mention of 2 papers in the legend of Fig S1 but it would be recommended to have a supplementary table(s) with the markers defining each cluster and/or FeaturePlots-Violin Plots highlighting the main markers defining each cell type for validation and confirmation.In this new refined and very thorough atlas, it is quintessential to provide more information for how the 88 clusters were annotated.Again, this will clearly help the non-blood-expert reader.
Again, we completely agree that the above additions will help with broadening the impact of our paper.As mentioned above, we have generated two new supplementary tables (Table S1 and Table S2) that list the marker genes that were utilized to annotate the 88 cell types.Additionally, as requested by the reviewer, we now include a new Supplementary Figure (Fig. S2) that shows the expression patterns of the top marker genes (scaled gene expression, ~2 markers/cell type) for the 88 annotated cell types in a heatmap, which has been split in 4 so that it is easier to read.Furthermore, we provide (Fig. S3a) that summarizes the distribution of the 88 cell types in space (anatomical location) as this spatial information was critical for the annotation of the 88 cell types.
-The same needs to be applied throughout the manuscript.As an example, how the haematoendothelial populations that the authors have isolated for further analysis (Fig 2 ) were defined?How MEP, EMP, BP and HEP were separated?Markers and/or FeaturePlots should be provided.
As mentioned above, we now provide a more exhaustive list of the marker genes and associated references used to annotate the different cell types, including the hematoendothelial cell types, in Table S1 and Table S2.We have also generated a new heatmap that highlights the expression patterns of marker genes used to distinguish hematoendothelial cell types in Fig. S3b.The expression patterns of various marker genes used to distinguish different hematoendothelial cell types (including MEP, EMP, BP and HEP) are also included in Fig. S4, S5, S6 and S8 visualized using FeaturePlots (expression patterns overlaid on force directed layout).Additionally, we now highlight many of the expression patterns of these genes, and some additional genes, in some of the additional trajectory analyses that we have performed along the various waves of primitive and definitive blood production in the developing YS in Fig. S7a,b.
-Fig 2 .Why were embryo proper endothelial cells, and allantois endothelial cells included in the YS analysis?As they appear to be located at the very origin of the blue trajectory, if these cells are removed from the analysis, does this affect the trajectory results?
Apologies, we should have explained this better.There are two main reasons (i, technical and ii, biological) to include these cells for some of the analyses.Firstly, due to relatively high transcriptional similarity between embryo proper endothelial and YS haematoendothelial cells, some embryo proper cells are assigned above the log odds threshold.The log odds is a nondeterministic criterion.The cost function of the optimal transport framework is the Euclidian distance between the transcriptomes (i.e., the cost function of the equation optimised for calculating cell coupling probabilities between time points that define trajectories).The optimisation procedure therefore favours cells with highly similar transcriptional profiles and as such some cells from non-yolk-sac locations, with highly similar transcriptomes to yolk-sac cells are included in the trajectory.However, these few false positives do not substantially change the trajectory/landscape even if you filter them out in a supervised fashion when generating the force directed layout.
Secondly, we included these cells as it opens the underexplored, but in the context of development, highly important issue of molecular convergence, whereby inferred differentiation trajectories finish at very similar molecular states (in this case the endothelial state in the various anatomical locations), but to get there, the cells traverse diverse molecular states.We believe that molecular divergence/convergence and lineage histories/relationships provide important baseline information for a better understanding of (i) normal development, (ii) production of more authentic cell types from pluripotent stem cells in vitro, and (iii) is likely to influence disease phenotype variation.We have amended the text to clarify these points.
-Fig S5 ."Lymphoid and microglial-like progenitors can also be detected".This sentence must be revised since all is known from this analysis is that the cells the authors are referring to are cells expressing lymphoid or microglial markers.In the absence of fate experiments, tracing or even analysis of lineage potential, it is not possible to define them as lymphoid or microglial-like progenitors.
We agree and have amended this sentence.The sentence has been changed to: "Cells expressing lymphoid and microglial-like progenitor markers are also detected, albeit at a low frequency (Fig. S5)." -The caption for figure 3a is the same as figure 2a.This should be remedied.
We are sorry for this mistake; it has been fixed.The new figure legend for 3a is: "UMAP layout of the mouse extended atlas displaying the log odds of fate probabilities associated with the complete haemato-endothelial landscape (top left).Cells with log odds > -1 were retained to generate a force directed layout.Cells are coloured by Log odds of fate probabilities of all hematoendothelial cells, embryo stage and anatomical regions."-In Fig S7 : it is mentioned that endothelial landscape revealed 3 different trajectories, but this is not clearly presented/easy to see.It is very difficult to guess which are the 3 trajectories the authors are referring to.In addition, can the authors add some interpretation of these data?Are those trajectories a result of time and or locations?This seems to be, at least partially, the case, however without a clearer indication of the trajectories, the reader is left to guess.For the location point, can the author check if trajectories of EP ECs have a specific HOX code?
We much appreciated this comment, because indeed time and location distinguish the trajectories, which we are now utilising to provide better visualization.We have added three arrows to the Fig. S10a (originally Fig. S7a) to better highlight for the reader the three distinctive waves of endothelial formation we are referring to.Additionally, we provide a new figure (Fig. S11a-c) that clearly highlights the predicted subset of cells that comprise each of these three inferred trajectories.Briefly, the first wave occurs early and forms endothelium in the developing yolk-sac, the two later waves initiate from distinctive anatomical locations in the embryo proper (posterior vs. anterior/medial sections).As requested by the reviewer, we have also explored the potential for a Hox code as differentiator in embryo proper endothelium, and now include this in Fig. S10d.This analysis demonstrates that the expression of different sets of hox genes are enriched in the embryo proper endothelium from different anatomical locations.The manuscript has been amended to introduce this new analysis, which we agree helps substantially to illuminate the complexity of endothelial development and better highlight the three inferred trajectories we are referring to.
Lyve1 seems to be a specific marker of YS cells in this dataset, however in E9.5 embryos Lyve1 expression, measured by Lyve1-Cre, is observed in placental vessels, likely allantois derivatives, in a previous report (Lee et al. Cell Reports 2016).Can the author comment on this?
It is important to note that the placenta and placental vasculature were not included in our extended mouse gastrulation atlas.In Figure 2 E and F in Lee at al. a minor proportion (~10%) of the placental vasculature is GFP+ and therefore derived from cells that at some stage during their formation expressed Lyve1-Cre.The reviewer comments that these placental vessels are likely derived from allantois precursors.On page 2290 of Lee et al. the authors state: "Analysis of the early concepti revealed LYVE1 (protein) expression in Tie2+CD31+ angioblasts already by E7.5 (Figure S1C).At E8.5, LYVE1 was present in YS Tie2+CD31+ angioblasts, but excluded from Ter119+ primitive erythroid cells (Figure S1C), the embryo, and allantois (data not shown)."Based on these data of protein expression and our RNA atlas, our interpretation is that Lyve1 mRNA is likely not expressed early-on in the mesodermal allantois precursors that will later on form Lyve1-Cre-GFP+ placental vessels.Instead, Lyve1 protein expression is likely induced in the developing vasculature of the placenta.As such, we do not find high levels of Lyve1 mRNA expression in the allantois cells in our atlas.As our dataset does not include placental vessels, Lyve1 mRNA expression is confined to the YS endothelium.In line with Lyve1 induction in the placental vasculature, the proportion of endothelial cells expressing Lyve1 protein in the placental vasculature dramatically increases from E13.5 to E16.5 (Lee et al.S1A).

Why Meg 3 is an important gene to mention?
Meg3 is a long noncoding RNA that has been reported to be involved in the DNA damage response and play a role in suppressing endothelial proliferation.We don"t think it is an important gene to mention in the text, therefore we have removed highlighting its expression in relation to different subtypes of endothelium.
Mef2c is assigned as a venous gene, however many reports show very high level of expression in arteries, including the dorsal aorta during embryonic development.
The reviewer points out that the expression pattern of Mef2c is broader than venous endothelium.In line with this, we find Mef2c is expressed in various endothelial subsets (please see figure above) as well as the cardiac lineage in our atlas, in line with WISH experiments by other research groups.We observe detectable levels of Mef2c mRNA expression in the posterior embryo proper endothelium, which likely includes endothelial cells of the developing dorsal aorta.Considering the reviewer"s comment, and to improve the clarity of our manuscript, we have decided not to mention the high Mef2c expression levels in venous endothelium specifically, as it is indeed also expressed in other endothelial cell types in our atlas as the reviewer rightly points out.
The observation regarding the existence of an endotome during mouse development is a very interesting one and would add a very important element of novelty, so this reviewer thinks that a deeper molecular characterization of these cells should be added.Beside the expression of Cxcl12, Pax3, Meox1, Foxc2, Pdgfra and Alcam, can the authors provide a better transcriptomic characterization?What are the DEG between the endotome vs neighbouring clusters (like anterior CPP) as well as ECs? Do the endotome express any endothelial genes or genes related to vasculogenesis/angiogenesis? Bias towards a specific subset of EC (venous vs arterial vs lymphatic)?Can this analysis differentiate between ECs of intersomitic vs other vessels?If so, can the authors predict whether the endotome can contribute to the formers?
We agree with the reviewer and have therefore provided a better characterization of the endotome by (i) identifying genes that are differentially expressed (DEG) in the endotome compared to nearby cell types (including the anterior CPP) in the endothelial landscape, as asked by the reviewer (Table S7) (ii) we additionally provide more information about the marker genes for the endotome versus all other cell types (these gene sets are also listed in Table S7).We have also performed GO Term enrichment analyses as well which are provided in Table S7.These analyses reveal that the endotome expresses genes related to vasculature development and angiogenesis including, Apoe , Nr2f2, Col3a1, Col4a1, Ednra, Eya1, Foxc2, Id1, Lrp1, Foxc1, Pdgfra, Pdgfrb, Prrx1, Prrx2, Cxcl12, Six1, Tbx1, Tead2, Vegfb, Vegfc, Smarca2, Plpp3, Cald1 and Cxcl12.To introduce these new analyses into the paper and provide visualization, we now show the expression patterns of these endotome marker genes in a new heatmap in supplementary (Fig. S12).This analysis highlights several endotome maker genes are also expressed in endothelial cells (white box, Igfbp4, Gng11, Cd302, Eva1b, Bgn, Dlc1, Plekho1, Id1, Mdfi).Additionally, the endotome cells express genes that are expressed in the dermomyotome (white box, Tcf15, Meox1, Foxd1, Foxc2, Foxc1).Furthermore, we have explored changes in gene expression along the putative endotome to endothelial trajectory that occurs in the anterior/medial sections (Fig. S11 a,b,d and Supplmentary Table S4).We do not think that based on gene expression alone, we can draw conclusions about a potential bias towards a specific subset of endothelial cell, nor does this analysis allow us to determine whether the endotome derived endothelial cells likely contribute to intersomitic versus other vessels.
Page 7 what does "fetal liver HSC induction" mean, for Hlf?
We mentioned Hlf induction as Hlf expression can induce HSC formation in committed blood progenitors in a paper by Derek Rossi"s group and we thought this finding might motivate future experiments.Furthermore, Hlfhi cells in the fetal liver are enriched for HSCs that repopulate the bone marrow.However, there is no evidence that under physiological conditions Hlf expression in the fetal liver induces HSC formation, therefore, we have removed "fetal liver HSC induction" from the text.
-In the Materials and Methods: "Generating an integrated atlas": the identification of highly variable gene is not well explained.It is not mentioned the function or the parameters used (it is unclear what "as described above" in page 13 means).
We have now included greater description in the material and methods section, stating: "Highly variable genes (HVGs) were calculated using "trendVar" and "decomposeVar" from the scran R package, with loess span of 0.05.Genes that had significantly higher variance than the fitted trend (Benjamini-Hochberg-corrected P < 0.05) were retained.Genes with mean log2 normalized count <10−3; genes on the Y chromosome; the gene Xist; and the reads mapping to the tdTomato construct (where applicable) were excluded."-In the Materials and Methods: "Differential gene expression and Canonical Correlation Analysis of Endotome derived VACs": it is not clarified the test used for the findMarkers function or if there is a p-value threshold applied to the selection of genes.
We have now provided more details about our procedure for gene selection: The statistical test performed by findMarkers uses a general linear model and moderated tstatistics to perform differential expression, as implemented in the R package limma[ref1].The threshold established was FDR < 0.05 and LogFoldChange > 0.5.Furthermore, only genes among the top 20 rank genes are displayed in heatmaps (considering that some genes can have tied ranks, gene lists usually contain more than 20 genes).
-In the Materials and Methods: "Generating the landscapes of haemato-endothelial trajectories": two thresholds of log odd ratios have been chosen (log odds > 0 and log odds > -1).While the choice of the different threshold is explained, it is usually not a recommended practice changing the threshold in the same setup.Can the authors comment on this?Also the choice and identification of the complexity of a landscape is not well described.
The landscape in Fig. 2 was generated to provide a robust reconstruction of YS primitive and definitive waves to illustrate a widely studied process in embryonic haematopoiesis.It is a proof of concept for our approach to dissect complex trajectories using Waddington-OT.The landscape in Fig. 3 (and S10) aims to expand our understanding of the complex process of blood and endothelial formation more generally during embryonic development.When constructing the force layout from connectivity in the WOT graph with more complex cellular differentiation landscapes relevant cellcell connections spanning different cell types might be lost if hard thresholds are imposed.Threshold selection is ultimately heuristic, and as such we aimed to make careful statements about the thresholds and inferences we made in the text.Furthermore, we provide a layout where cells are coloured by log odds ratios for the reader to consider for all the blood and endothelial landscapes that we generate.
-In Materials and Methods, it lacks a section explaining how the force-directed graphs, UMAPs, etc. were constructed.A vizualisation section here is recommended, similarly to what already provided in their previous paper (Pijuan-Sala et al. 2019)

Visualisation
To generate the UMAP layout of the whole embryo, we used the top 50 batch corrected principal components to generate a BBKNN graph[ref1] and then scanpy"s implementation of UMAP, with parameter min_distance = 0.99.To generate the force directed layuouts of the haematoendothelial landscapes, we recomputed highlgy variable genes for each subset as well as batch correction of PCA manifolds (We again retained the top 50 principal components).We then built a KNN graph (K=50) using and used ForceAtlas2[ref2] implementation of forced directed layouts included in scanpy.
-In Materials and Methods, in many instances the version of the tool used (scran, fastMNN, igraph, Metacell... ) is not clarified.This can affect the reproducibility of the analysis given also that the code is not provided yet.
We now provided a more detailed methods section regarding these methods.It was stated that unless specified differently, tool versions are the same as in Pijuan-Sala et al., 2019.We are now providing full details in the current manuscript.
-The ending paragraph, in particular the last sentence, feels a bit strange and out of place.
We have amended the ending paragraph and final sentence, which now states: Our extended mouse gastrulation atlas provides the developmental biology community with a significant new resource to probe novel hypotheses concerning cell fate acquisition and lineage commitment during embryogenesis.
-Merely aesthetic comment: Fig 1 .Adjust cartoon to match the image of how the dissection was done.
This has been done.We have flipped the image of the embryo and the yolk-sac from left to right so that it matches the cartoon.

Reviewer 2 Advance Summary and Potential Significance to Field:
In this manuscript by Imaz-Rosshandler et al., authors expanded a previous mouse early developmental atlas with 300,000 sc transcriptomes collected from 8 timepoints, every ~6h, from e8.5 to e9.5, covering organogenesis.Integrating this information with the previous atlas, they contribute the following highlights: 1) an extended atlas of morphogenesis in the mouse (8.5-9.5) 2) an extensive trajectory analysis of early blood and endothelial development 3) new evidence and trajectories for the mouse endotome (contributing to vascular-associated cells, which potentially support definitive hemogenesis).4) a spatially defined primitive stream single-cell atlas 5) a state-fate analysis through orthotopic grafting of primitive streak, with outcomes analyzed through single-cell sequencing This is a collection of technical achievements (which provide new datasets and benchmark methods of interest to the community) as well as biological insights (such as the discovery of a potential endotome-VAC trajectory in mammals).The manuscript nicely evokes classic experimental approaches and uses modern phenotyping technologies, such as single-cell sequencing, to reach important novel conclusions.Put together, I think both the dataset and the classic fate-mapping analyses make for a very interesting contribution to the field that will reach a broad readership.The cherry-in-the-pie would have been some heterotopic transplants to study fate plasticity, but I understand this might be out-of-scope for the present study.

Reviewer 2 Comments for the Author: ****Extended atlas and endotome
While some discussion is provided, I think the manuscript would still benefit a lot from better highlighting how it differs from existing published atlases, as there are several other ones available, e.g. the integrative dataset of Qiu et al.Nature Genetics 2022 and the new preprint of Qiu et al. bioRxiv 2023.Where the present manuscript has advantages, this should be highlighted.There is a constant trickle of developmental atlas efforts, and I find it increasingly unclear how they compare to each other.Eventually, this would help establish guidelines/tables that may be very useful for newcomers in the field and any other researchers interested in looking at mouse development to choose the right dataset for their question.
We greatly appreciate this valuable suggestion, and as a response, we have expanded the discussion section to elucidate the distinctions more comprehensively between our work and existing published atlases.We believe this enhancement will effectively addresses your concern by highlighting the unique advantages of our dataset.

New text reads:
Previous single cell atlas efforts from us and others placed observed molecular states into the context of existing knowledge of mouse development (reviewed by [Tam and Ho, 2020] and further integrated by [Chengxiang Qiu et al., 2022 andQiu et al. bioRxiv 2023]), altogether contributing to a deeper understanding of the molecular heterogeneity accompanying lineage differentiation during mouse embryogenesis.Our present study advances beyond these efforts in several impactful ways.By incorporating WOT, spatial information and encompassing the developing yolk-sac region, our analysis achieves a comprehensive characterization of blood and endothelial formation spanning the critical developmental window of E8.5-E9.5.This spatial context has allowed us to uncover hitherto unobserved cell states and developmental nuances.Our dense sampling, deep sequencing and regional sub dissection of embryos allowed us to identify cell states not previously observed in early mouse development, such as the endotome and VACs.Moreover, our trajectory analyses suggest that these progenitor populations are connected with downstream endothelial cells, suggesting a greater intricacy in endothelial differentiation than previously appreciated.I do not understand the rationale of choosing all blood, blood+endothelium or all endothelium (E9.25/9.5)as the end-point to calculate the W-OT fate matrices.Why not using the individual clusters of interest at the endpoint?Wouldn"t this increase the power of fate predictions and clarify the trajectories, especially for those clusters that only contribute to a rare small population, such as the endotome?
We are grateful to the reviewer for raising this point, as it highlights that we clearly didn"t articulate the analytical complications arising from trajectories that display molecular convergence.As already discussed in one of the responses to reviewer #1, there is extensive molecular convergence in both the blood and endothelial endpoints.It is therefore challenging to cleanly separate the various destinations from single cell suspension scRNA-Seq, even though our spatial dissections and time-series samplings do help.Nevertheless, it is impossible to make high confidence calls for many of the "endpoint" cells, so it is safer to include all of them (or pooled by anatomical location), and then dissect the trajectories in the way we did.We have amended the manuscript to clarify this critical point.
The findings presented in Figures 2-3 are very interesting, but perhaps it would be important to better summarize them.Even a simple graphic representation of the author's interpretation of the manifold and OT analysis would serve to more easily follow this through the text.Overlaying significantly-enriched TFs, predicted GRNs, ligand-receptor-interactions, GO-terms and any other computational findings for that particular trajectory would be a definite plus.
We appreciate this comment, and now provide more intuitive summary figures that highlight the various inferred trajectories that we were referring to in new supplementary figures (Fig. S7a-c and Fig. S11 a-c).Additionally, we provide a deeper characterization of genes (including many transcription factors) that significantly change in expression along the various inferred trajectories (primitive blood trajectory (Fig. S7a,b), YS definitive blood and endothelial trajectory (Fig. S7c,d), YS endothelial trajectory (Fig. S7e,f) anterior/medial wave of endothelial production (Fig. S11b) and a posterior wave of endothelial formation (Fig. S11c).These analyses were performed using Tradeseq and the comprehensive gene lists generated by these analyses are now included in new Supplementary Tables (Table S3, Table S4, Table S5).For the latter two endothelial trajectories we also provide enriched GO-terms for various gene clusters that change along pseudotime (Fig. S11d,e, and Table S4 and Table S5).Additionally, we provide new supplementary tables that highlight significantly enriched ligand-receptor interactions (Table S6) for the different cell types that arise in the canonical landscape (Fig. 2) as well as the anterior/medial sections of the hematoendothelial landscape (Fig. 3), which were generated using CellComm to infer cellular communication.We appreciate this suggestion from the reviewer and hope these additional analyses will be helpful in clarifying the inferred trajectories we identified and for hypothesis generation by other research groups investigating the complexities of hematoendothelial formation during embryogenesis.
Arguably, the biggest finding in the current data is the connection between the vascular associated cells (VACs), a putative HSC niche during early EHT, and a part of the somitic mesoderm (the "Endotome").This parallels results in zebrafish obtained from multiple independent groups (Nguyen et al., Sahai-Hernandez et al.).It is important to note (and highlight) that the present manuscript only provides an inference of this trajectory, and that there is currently no validation or additional lines of evidence of mammalian somitic-derived endothelial cells (SDECs) contributing to the aorta.Indeed, it is surprising that after almost a decade since the Nguyen et al. manuscript, there is still no lineage-tracing evidence for contribution of somitic mesoderm into the SDECs in mammals.Canonical Cre-based lineage tracing (e.g.Tbx6 or Pax3 Cre lines) should have been able to validate this connection.Recent papers by Catherine Robin (Hubrecht) and Oliver Stone (Uni Oxford) have looked carefully at the endothelial somitic derivatives using these lines (Stone and Stainier Dev Cell 2019, Lupu et al. biorxiv 2023, Yvernogeau et al. biorxiv 2020).
For the revision, we have been sure to highlight/emphasize that the landscapes/trajectories we present in our manuscript are inferred.In line with this goal, the new supplementary figures we have generated (FigS7 and S11) that specifically highlight these various trajectories, have been labelled with the term inferred.We are very pleased that this reviewer values our identification of putative mouse endotome cells and VACs and the opportunities for future research that our results open.
While the most thorough paper performing Pax3-Cre single-cell RNAseq analysis is yet to be peerreviewed (Lupu et al.), the data in the preprint does not obviously suggest any contribution of Pax3-Cre cells to the aorta, although I find their single-cell data plots hard to interpret.Interestingly, the microscopy images of sections at e9.5 and e10.5 suggest some Pax3-lineage cells may be located very close to the DA (though not integrated in the monolayer).Similarly, Yvernogeau et al. detected a substantial contribution of Tbx6+ PSM to limb ECs, but very little contribution to visceral and aortic ECs (although it was not 0).Both manuscripts find cells that are peripheral to the aorta, which perhaps resembles the stromal cell populations from Murayama et al. 2023, which share multiple markers with the endotome VACs presented in the current manuscript.Of course, the best thing here would be to integrate the cells from Lupu et al into this atlas, but the dataset is not yet accessible.At the very least, I feel that some additional discussion and citation to these manuscripts is missing.
We are happy to read how this reviewer emphasizes that our datasets and discoveries raise many interesting new research questions.As stated by the reviewer, we cannot integrate our data with others currently referred to in preprints.We have nevertheless referred to these emerging papers in the discussion, where we have drafted several new sentences on the endotome and vascular associated cells.Of note, all our data are freely available, which will greatly benefit future integrated analysis.
New sentences added to the text: Intriguingly, more recent unpublished Tbx6+ [Yvernogeau et al.] and Pax3+ lineage tracing studies [Lupu et al.] have been performed at embryonic timepoints (E7.5-E11.5)that overlap with the emergence of endotome, VACs and endotome derived endothelial cells in our atlas.These studies suggest somitic mesodermal precursors give rise to endothelial cells of the limb and trunk region in developing mouse embryos, as well as stromal cells juxtaposed to the dorsal aorta, which may be similar to the VAC population we identify in this present study.In these studies, somite derived endothelial cells make only a minimal contribution to dorsal aorta endothelium by E9.5-10.5.
In chick and zebrafish this "endotome" population is supposed to be bipotent, contributing to both muscle and endothelium.Does W-OT make the same prediction in the current dataset?Our dataset does not contain an obvious downstream muscle cell population to use as a future cell state in the Waddington-OT analysis the review asks for.However, in response to a question from reviewer 1 we now present a more comprehensive analysis of the genes and associated GO terms that are enriched in the endotome cells in Supplementary Table S7.This analysis reveals endotome cells express several genes related to GO terms such as muscle development (logP -5.8) and muscle tissue morphogenesis (logP -4.9) including Col3a1, Ednra, Foxc2, Foxc1, Tbx1, Fzd2, Nr2f2, Six1, Tcf15, Twist1, Fzd2 and Foxp1.
Is there any evidence here that the endotome-derived endothelium described here expresses hematopoietic-specifying/modulatory molecules?What about ligand-receptor predictions, with the other endothelial cells?
As mentioned in a previous response, we now provide a new supplementary table (Table S6) that highlights predicted cellular communication (ligand-receptor interactions) between the various cell types in our hematoendothelial landscape, including the endotome and developing embryo proper endothelial cells.This analysis suggests that endotome cells likely secrete VEGF and CXCL signals.

****Orthotopic PS transplant & fate analysis
The State-fate analysis through orthotopic grafting is really cool and interesting and, although not at single-cell resolution, it does highlight the power of these fate-predictive tools when applied at the population level.I just missed a few more statistical analyses to objectively quantify the accuracy of the predictions.I also do not quite understand why the labels are different in the predicted fates vs. the observed fates (Figure 5 c-d-e-f).I"m sure this makes some sense somehow, but it makes it hard on the reader to interpret to what extent the predictions were accurate based on pre-grafted states and OT.
We are delighted to read that this reviewer liked our state-fate analysis.The confusion with the labels is due to the fact that in the predictive fates, the grafted and transcriptomically analysed cells were mapped onto the extended gastrulation atlas, whereby the closest neighbour cell was calculated and mapped to the atlas, therefore it has so many more labels to map to.On the other hand the morphologically observed fates were established based on anatomical hallmarks by looking at wholemount and sectioned embryos post grafting and culture; there it was not possible to distinguish more detailed cell populations without additional markers (only anti-CD31 antibody was used to highlight the vasculature of the grafted conceptus and thus provide an anatomical orientation).
There seem to be some large differences between some of the predictions by morphology and by transcriptome.For instance: paired aortae and neural tube are overrepresented in the morphological analyses, compared to somites, which are overrepresented in the transcriptome analysis.However, it seems to me that the morphological analysis is based on # of embryos with ANY mTom contribution to that compartment, whereas the numbers on Fig 5d are based on total numbers of grafted cells.Is there any way to better compare predictions and fates, using the same metric?
The large bias for neural tube and somites in morphological and transcriptomic fates, respectively, is likely due to the fact that in the morphological observations often a cluster of grafted cells would be lodged at the site of grafting (primitive streak) which 24h later would contribute to the neural tube.Since the morphological observations are only based on anatomical hallmarks there is no way to distinguish whether the mTom grafted cells that have lodged in the neural tube are actually of neural transcript or whether they have not been able to migrate and would have formed a small microenvironment, possibly mainly of somitic origin.This is difficult to assess by morphology, which is why we sorted the grafted cells and analysed transcriptomically.It must be noted that the cluster of cells that would be lodged was mostly seen in the neural tube (Fig 6 A,C,Fig S16b), while mTom contribution to other cell tissues was more of the true migratory quality due to the fact that mTom cells would appear individually and not always connected to other mTom+ cells (Fig S16b).Regarding the metric, all transcriptomically observed fates deal with single cells, while the morphologically observed fates deal with number of embryos where this mTom expression was observed out of the total number of embryos grafted for that particular primitive streak portion.Unfortunately, it is currently impossible to either accurately count the number of mTom+ cells during the morphological assessment or accurately assess the distribution of individual embryos within the mapped gastrulation atlas, making the comparison using the same metric, at this point, not possible.Advances in novel imaging and sequencing technology may resolve this question in the future.

**** Minor points:
Figures (especially 1-4) are really huge, with very small prints for legends, and tons of labels in some cases.I would highly recommend reformatting these for improving readability.
As we had only used 5 figures in the initial submission, we have now split the most crowded figure (figure 1) into 2 separate figures.We have also gone through all the main figures to improve readability.Comments for the author I am happy with the revisions provided and I am sure this study will benefit many in the field.

Adding the total # of cells in
Reviewer 2

Advance summary and potential significance to field
Early organogenesis is a critical window of animal development, where most of the cell types that will form the embryo, and eventually the adult individual become specified and begin to organize into tissues.In this manuscript, the authors have profiled over 400,000 high-quality deepsequenced single-cell transcriptomes from multiple anatomic sites in mouse embryos and integrated them with existing atlases.This resolution allowed them to clarify the diversity of endothelial cell trajectories in the embryo, helping them to identify the somitic precursors that putatively contribute to the aortic endothelial niche where definitive hematopoietic stem cells arise.This trajectory data will help the community design better experiments to challenge this hypothesis in the mouse, where this lineage has so far been elusive.Finally, proving that this atlas can provide useful predictions of cell fates (at least at the population level), the authors transplanted E7.5 cells from the anterior part of the primitive streak into a second embryo and then profiled them 24h after differentiation in embryo cultures.In the end, I believe this integrated & deep-sequenced single-cell transcriptome dataset will be one of the most used mouse embryo atlases for quite a few years to come.

Comments for the author
This revised version of the manuscript by Imaz-Rosshandler et al. has significantly improved upon the original version by incorporating comments from all reviewers.Seeing the response to reviewers, it seems clear that substantial attention has been given to both improving data presentation and discussion.Critically, the authors now discuss alternative interpretations and highlight the inferential nature of some of the results, further suggesting experimental possibilities to confirm them in the future.They also provide reasonable explanations regarding morphological vs transcriptomic fate classification differences in transplant experiments.All my additional comments were also addressed successfully.I think there's little else that could be asked of the study at this point.I would still suggest some minor figure editing issues that could be worked out during editorial processing: for instance, some legends and scale bars are too small for the figures (example,Fig 3c,Fig 6e), and some of the axes legends are too small to read (example, violin plots in Fig 5e).
-Alejo E. Rodriguez Fraticelli Fig 6f is condensed to the most appropriate cell types in order to compare it to the morphological fate map, the full transcriptomic fate map of post-grafted PS-derived cells is shown in Fig S16d.