May 30, 2024

Medical Trend

Medical News and Medical Resources

Science: Atlas of Human Fetal Gene Expression Cells

Science: Atlas of Human Fetal Gene Expression Cells


Science: Atlas of Human Fetal Gene Expression Cells.   These data provide rich resources for exploring gene expression in different tissues and cell types of the human body.



The gene expression program behind the human cell type has fundamental significance. This paper constructs a human cell map of gene expression and chromatin accessibility in fetal tissues. For gene expression, the author applied three-level combinatorial indexing to >110 samples representing 15 organs, and finally analyzed about 4 million single cells. The author uses literature and other atlases to identify and annotate hundreds of cell types and subtypes within and between tissues. The author’s analysis focuses on the widely distributed cell types (such as blood, endothelial cells, and epithelial cells), the organ specificity of fetal red blood cell production sites (especially including the adrenal gland), and the integration with mouse development maps (such as the conservative characteristics of blood cells). These data provide rich resources for exploring gene expression in different tissues and cell types of the human body.


Paper ID

Original name: A human cell atlas of fetal gene expression

Translation: Atlas of Human Fetal Gene Expression Cells

Journal: Science

IF: 41.845

Publication time: November 2020

Corresponding author: Jay Shendure

Author Unit: Department of Genomics, University of Washington School of Medicine

DOI number: 10.1126/science.aba7721


Experimental design


Science: Atlas of Human Fetal Gene Expression Cells


Research Results: 


1 Identification and annotation of 77 main cell types

The author described the median of 72241 cells or nuclei per organ [Figure 1A; largest, 2005512 (brain); smallest, 12611 (thymus)]. Although the sequencing is relatively shallow compared to other large-scale single-cell RNA sequencing (scRNA-seq) maps (about 14,000 raw reads per cell), the authors recovered a considerable amount of UMI (median 863 UMI and 524 genes, excluding cultured cells; Figure S1E). As expected, the ratio of UMIs in nuclei to introns was higher than that in cells (56% in nuclei; 45% in cells; P<2.2E-16, two-sided Wilcoxon rank sum test). Hereafter, unless otherwise stated, the author will use the term “cell” to refer to cells and nuclei.

Through sex-specific gene expression (Figure S1F), it is easy to determine whether the tissue is of male (n=14) or female (n=14) origin. Each of the 15 organs is represented by multiple samples (median 8), and each sample includes at least two genders (Figure S1G) and a series of estimated post-gestational weeks (Figure 1B). The Pseudo-bulk transcriptome is clustered by organs rather than individuals or experiments [Figure S1H; GEO documents S4 and S5 (GSE156793)]. About half of the expressed protein coding transcripts are differentially expressed in the pseudo-bulk transcriptome [11766 of 20033; false discovery rate (FDR) is 5%; Table S2].

The authors used Scrublet to detect a 6.4% probability of double cells, which corresponds to a double estimate of 12.6%, including double cells within and between clusters (Figure S1I). The scalable strategy previously developed by the authors was then applied to remove low-quality cells, double-enriched clusters, and peaks in HEK293T and NIH/3T3 cells. All the analyses below focus on the 4062980 human single-cell gene expression profiles extracted from 112 fetal tissue samples, which still exist after filtering.


Science: Atlas of Human Fetal Gene Expression Cells
Figure 1 Data generation and identification of the cell types of 15 human organs. (A) The project workflow (left) and the bar graph (right) show the number of cells distributed on a log10 scale in each organ. (B) The bar graph shows the distribution of estimated weeks after pregnancy for tissue samples corresponding to each organ. (C) After filtering low-quality cells and clusters rich in double peaks, 4 million single-cell gene expression profiles were visualized by UMAP on a per-organ basis, and Louvain clustering was performed with Monocle 3.



Using monocle3, the author performed UMAP visualization and Louvain cluster analysis on single-cell gene expression profiles. In summary, the author initially identified and annotated 172 cell types based on the expression of cell-type-specific marker genes [Figure 1C and Table S3; GEO documents S6 and S7 (GSE156793)]. After folding the common annotations across tissues, these cells were reduced to 77 main cell types, 54 of which were only observed in a single organ (for example, Purkinje neurons in the cerebellum), and 23 in multiple organs Observed (for example, vascular endothelial cells in each organ). There are 15 cell types that the author failed to annotate (a subset named by a pair of markers in Figure 1C); these will be discussed further below. Each of these 77 main cell types is represented by a median of 4829 cells, ranging from 1258818 cells (excitatory neurons in the brain) to 68 cells (SLC26A4- and PAEP positive cells in the adrenal glands) ( Figure S2A). Each major cell type was observed in multiple individuals (median 9; Figure S2B). Despite differences in species, developmental stages, and technology, the authors have restored almost all the major cell types identified in previous atlasing work for the same organ. The authors identified a median of 12 major cell types for each organ, ranging from 5 (thymus) to 16 (eyes, heart, and stomach). The authors did not observe a correlation between the number of atypical cells and the number of determined cell types (Spearman=-0.10, P=0.74).

On average, the authors identified 11 marker genes for each major cell type (minimum value is 0; maximum value is 294; defined as differentially expressed genes, the expression difference between the first and second cell types is at least 5 Times; FDR is 5%; Figure S2C and Table S4). Due to the highly related cell types in other organs (such as gut glial cells and Schwann cells), several cell types lack marker genes at this threshold. For this reason, the author used the same procedure to determine a set of marker genes in tissues, but organ-based (average 147 markers per cell type; minimum 12; maximum 778; Figure S2D and Table S5 ). An interactive website helps to explore this data by tissue, cell type or gene (

Although typical markers are usually observed and play a vital role in the author’s annotation process, as far as the author knows, most of the observed markers have not been identified in previous studies. For example, OLR1, SIGLEC10 and non-coding RNA RP11-480C22.1 are the strongest microglia markers, as well as more mature microglia markers such as CLEC7A, TLR7 and CCL3. As expected, considering that these tissues are developing, many of the 77 major cell types include the process from precursors to one or more terminally differentiated cell types. For example, excitatory neurons in the brain show a continuous trajectory, from PAX6+ neuron progenitor cells to NEUROD6+ differentiated neurons, and then to SLC17A7+ mature neurons. In the liver, hepatic progenitor cells (DLK1+, KRT8+, and KRT18+) showed continuous trajectories leading to functional hepatocytes (SLC22A25+, ASS2+, and ASS1+) (Figure S2, G, and H). Compared with mouse organogenesis, in these data, the maturation of the transcription program is tightly coupled with developmental time, and the cell state trajectory is not consistent with the estimated number of weeks after pregnancy (Figure S2, I, and J). One possible explanation is that gene expression is significantly more dynamic during the embryonic period than during the fetal development period. However, the inaccuracy of the estimated number of gestational weeks may also confuse the author’s solution.

In addition to the manual annotation of these cell types, the author also used Garnett to generate a semi-automatic classifier for each organ. The generation of the Garnett classifier has nothing to do with the previous clustering, and the marker genes are compiled separately from the literature. Garnett’s classification is consistent with manual classification (Figure S3A). Using the Garnett model trained on this data, the author was able to accurately classify cell types from other single-cell data sets, including data generated by different methods and data from adult organs. When the authors applied the pancreas classifier to the inDrop-scRNA-seq data, Garnett correctly annotated 82% of the cells (cluster expansion; 11% incorrect, 8% unclassified) (Figure S3B). These models can be widely used for automatic cell type classification of single cell data from different organs (Figure S3C;

Next, the author evaluates the specificity of the author’s main cell types by using cross-validation within the data set of the support vector machine (SVM) classifier. In this framework, high cross-validation accuracy and recall value indicate that cells from a given cluster can be stably redistributed to that cluster; therefore, the author uses a high F1 score as a marker to identify a cell cluster as a valid type, at least in the recognition This is the case in the settings of your organization. The author first applied this method to the kidney. As expected, the annotated kidney cell types had a much higher specificity score (median 0.99) than the control cell types, where the cell markers were ranked before cross-validation (median 0.17) [Figure 2A (leftmost panel only) , Figure 2B (left picture only), Figure S4A and Table S3].

Then, the author applied this method to the cells of each organ. Again, the annotated main cell types showed much higher specificity scores than the replacement cell types (Figure 2C and Figure S4B; median value 0.99 vs. 0.10; P<2.2×10-16, two-sided Wilcoxon rank sum test). Despite the small number of cells, most of the 15 initially unannotated cell types also showed high specificity scores (median 0.98). Exceptions may be better described as subtypes of other cell types. The author also applied this method to a collection of 77 major cell types (ie, not organ by organ), and the results were similar (Figure S4C).



2 Automatic preliminary annotation of cell subtypes

To identify cell subtypes, the authors performed unsupervised clustering of major cell types with >1000 cells in any given tissue. For each major cell type in each tissue, the authors first applied batch correction, followed by dimensionality reduction and Louvain clustering (Figure 2A). After passing the above-mentioned internal cross-validation procedure to distinguish clusters that are not easy to distinguish, a total of 657 cell subtypes were identified in 15 tissues, with a median value of 824 cells in each tissue. All subtypes consist of cells contributed by at least two individuals (median 7). Unsurprisingly, considering the procedures used to form clusters, these subtypes have higher specificity scores than replacement controls (median 0.77 vs. 0.13; P<2.2×10E-16, two-sided Wilcoxon rank sum test; Figure 2C ).

The author next tried to use the existing mouse cell map to automatically annotate the cell type cross-matching method previously developed by the author of these human subtypes. The author can compare 605 of the 606 (99%) human cell subtypes with those from the small group. Mouse cell atlas (MCA) corresponds to at least one cell type in the corresponding fetal and/or adult tissues (specificity score β>0.01, the same threshold as the author used to align with MCA before; 51 adrenal glands) The subtype was excluded because the corresponding MCA organization was not available) (Table S6 and Figures S5 to S8). In addition, 77 (52%) of 148 (52%) brain or cerebellar subtypes matched at least one adult cell type in the Mouse Brain Cell Atlas (MBCA) (Figure S9).

Despite species differences, many human cell subtypes match 1:1 with mouse cell types. For example, the different epithelial subtypes in the human kidney matched the annotated MCA cell type 1:1 (Figure 2A), and the different neuron subtypes in the human brain matched the annotated MBCA cell type 1:1 (Figure S9). It is worth noting that although there are many human subtypes that match a single MCA or MBCA cell type (for example, hepatoblasts in Figure S5 and oligodendrocytes in Figure S9), these may reflect true heterogeneity , As evidenced by its specificity score (Figure 2C). Additional work is required to annotate subtypes with greater granularity.



Science: Atlas of Human Fetal Gene Expression Cells
Figure 2 Identification of cell subtypes. (A) Cell subtype identification process. (B) The cell type cross-validation confusion matrix in the data set of the SVM classifier for the main cell types in the kidney (left) and metanephric subtypes (right). (C) The box plot shows the distribution of cell-specific scores (F1 scores) for the main cell types and subtypes from the cross-validated replacement controls in the data set.



3.  Research on integration between tissues and initial unannotated cell types

Next, the author tried to integrate data from all 15 organs and compare cell types. In order to reduce the impact of sampling errors, the author randomly selected 5000 cells from each organ (or <5000 cells representing one cell type in an organ, take all cells), and performed UMAP visualization (Figure 3A and Figure S10A) ). As expected, cell types represented in multiple organs and cell types related to development tend to co-locate. Many surface proteins (4565 out of 5480), secreted proteins (2491 out of 2933), transcription factors (1715 out of 1984), and non-coding RNA (3130 out of 10695) in 77 major cell types Differential expression in types (FDR is 0.05, Figure 3B and Table S4, The expression pattern of non-coding RNA is clearly sufficient to divide cell types into developmentally related groups (Figure S10, B and C).

As mentioned above, there are 15 cell types that the author failed to annotate (a subset named by a pair of markers in Figure 1C). To clarify these, the authors checked their distribution in the global UMAP (Figure 3A), whether they match the cell types annotated in MCA or MBCA (Figures S5-S9), and their distribution in different individual tissues (Figure 3A). S11A) and its maternal origin similarity (Figure S11B).

These further analyses allowed the authors to annotate 8 of the 15 cell types. For example, the rare CSH1 and CSH2 positive cells in the lung and adrenal glands (the two deepest contoured organs) are highly similar to placental trophoblast cells, for example, expressing high levels of placental prolactin, chorionic gonadotropin and aromatase (Figure 3A) ). The AFP- and ALB-positive cells in the placenta and spleen are similar to hepatoblasts, for example, express high levels of serum albumin, alpha-fetoprotein, and apolipoprotein (Figure 3A) [at least in the placenta, observed in mice Similar hepatoblast-like AFP- and ALB-positive cells (Figure S5)]. Subsequent immunostaining studies supported the presence of these trophoblast-like cells and hepatoblast-like cells in the adrenal gland and spleen, respectively (Figure 3, C and D, and Figure S12). Given that these cell types are rare but only recurring in a few organs, they may have corresponding circulating trophoblast cells and circulating hepatoblasts.

In men, IGFBP1- and DKK1 positive cells in the placenta and PAEP- and MECOM positive cells all expressed significant levels of XIST or TSIX (Figure S12B); further review of the markers, they correspond to decidual stromal cells and maternal endometrium, respectively Epithelial Cells. This conclusion is supported by the maternal genotype of the corresponding cell type in the chromatin accessibility data.
Several other cell types were annotated through strong matches with MCA or MBCA (Figure S13) or through their position in the global UMAP plus additional literature review (Figure 3A); these cells include STC2- and TLX1- Positive cells, which are abundant in the spleen and express genes related to mesenchymal precursor cells or stem cells. Of the remaining 7 cell types that were not initially noted, 4 may be better classified as subtypes (these subtypes tend to have lower specificity scores), and 3 have higher specificity scores , But still ambiguous.


Science: Atlas of Human Fetal Gene Expression Cells
Figure 3 Integrated visualization of cell types in all contoured tissues. (A) Sampling 5000 cells from each organ. (B) The heat map shows the relative expression of surface and secreted protein-coding genes, non-coding RNA (ncRNA) and TF (column) in 77 main cell types (rows). (C) Representative fluorescence microscope image of human fetal adrenal gland or (D) spleen tissue.



4 Developmental characteristics of blood lineage between organs

The nature of this data set creates an opportunity to systematically study differences in organ-specific gene expression in a wide range of cell types, such as blood cells. The author reclassified 103766 cells from all 15 organs corresponding to hematopoietic cell types (Figure 4A). Wethen performed Louvain clustering and further annotated fine-grained blood cell types, identifying very rare cell types in some cases (Figure 4B). For example, myeloid cells are divided into microglia, macrophages, and various dendritic cell subtypes [CD1C+, S100A9+, CLEC9A+ and plasmacytoid dendritic cells (pDCs)]. Microglia clusters are mainly derived from brain tissue and are well separated from macrophages, which is consistent with their different developmental background. Lymphocytes aggregate into several groups, including B cells, natural killer (NK) cells, ILC3 cells, and T cells, the latter including the thymus. The authors also recovered very rare cell types, such as plasma cells (139 cells, mainly in the placenta, accounting for 0.1% of all blood cells or 0.003% of the complete data set) and TRAF1+ antigen presenting cells (APCs) (189 cells, Mainly in the thymus and heart, accounting for 0.2% of all blood cells or 0.005% of the complete data set).

To verify these annotations, the authors integrated the scRNAseq profiles of fetal blood cells from all organs and blood cells from the fetal liver (Figure 4C, left panel and Figure S14A). Although the methods are different, the corresponding cell types of the two data sets are highly overlapped; this is also true when performing integration analysis with another scRNA-seq data set of 1231 human embryonic blood cells (Figure S14B). It is worth noting that some extremely rare cell types (such as VCAM1+EI macrophages, monocyte precursor cells and neutrophil-myeloid progenitor cells) identified by CD45+ fluorescence activated cell sorting (FACS) enrichment ) There is no comment in the author’s data. On the other hand, the authors captured fetal blood cells from tissues other than the liver, such as microglia in the brain, and T cells and B cells in the thymus and spleen. In addition, because they span multiple organs, the author is able to better capture the cell state transition path from hematopoietic stem and progenitor cells (HSPC) to lymphocytes, rather than a single organ study (Figure 4C, right panel).

Although gene expression markers for different immune cell types have been extensively studied, these markers may be restricted by their definition through a restricted set of organs or cell types. Here, the authors found that many traditional immune cell markers are expressed in a variety of cell types. For example, conventional markers of T cells are also expressed in macrophages and dendritic cells (CD4) or NK cells (CD8A), consistent with other studies (Figure S14C). The authors calculated pan-organ cell type-specific markers for 14 blood cell types (Figure 4D and Table S7). From this the authors observed that T cells specifically express CD8B and CD5 as expected, but also express TENM1 (Figure 4D and Figure S14C). ILC3 cells, whose annotation is determined based on the expression of RORC and KIT, are more specifically labeled by SORCS1 and JMY (Figure 4D and Figure S14C). The identification of these and other markers through panorgan analysis may help label and purify specific blood cell types.

As expected, different organs showed different proportions of blood cells (Figure 4E). For example, the liver has the highest proportion of red blood cells, which is consistent with its role as the main site of fetal red blood cell production, while T cells in the thymus and B cells in the spleen are more abundant. Almost all blood cells recovered from the cerebellum and brain are microglia. At the same time, the tissue distribution of ILC3 cells and dendritic cell subtypes was obtained (Figure 4E and Figure S14D). Pan-organ analysis can also identify rare cell populations in specific organs. The authors found rare heat shock cells in the liver, but also found rare cells that are transcriptionally similar to heat shock cells in the lung, spleen, thymus, heart, intestine, adrenal gland, and other organs (Figure S15). Subclustering analysis showed that HSPCs outside the liver and a subset of liver HSPCs express differentiation markers, such as LYZ, ACTG1 and ANK1, while most liver HSPCs express MECOM and NRIP1, both of which are maintained by normal static HSPC And function necessary (Figure S15).

Focusing on erythropoiesis, the author observed a continuous trajectory from HSPC to intermediate cell types, erythroid basophilic megakaryocytes toward progenitor cells (EBMP), and then divided into erythroid, basophil megakaryocytes, and megakaryocyte trajectories (Figure 5A and Table S8), consistent with recent studies on mouse fetal liver. Although there are differences in species (humans and mice), technology (sci-RNA-seq3 and 10x genomics), and tissues (only pan-organs and liver), this consistency still exists. Through unsupervised clustering and using the terminology in the study, the author further divided the continuum of erythroid status into three stages: early erythroid progenitor cells (EEPs) (marked by SLC16A9 and FAM178B), erythroid progenitor cells ( CEPs) (marked by KIF18B and KIF15) and erythroid terminally differentiated cells (ETDs) (marked by TMCC2 and HBB) (Figure 5B). The early and late stages of megakaryocytes are also easy to identify (Figure 5A and B).

As expected, given their established role in fetal red blood cell production, a portion of the blood cells in the liver and spleen correspond to EEPs, CEPs, and megakaryotic progenitor cells. It is worth noting that the authors also observed EEP, CEP, and megakaryotic progenitor cells in the adrenal glands of each study sample (Figure 5C and Figure S16A). Because the authors did not observe the more common cell types in the liver and spleen, slight contamination of the adrenal glands during recovery is an unlikely explanation. Although extramedullary hematopoietic islands are occasionally observed in the adrenal glands of human embryos, the consistency among individuals led the authors to further investigate whether the adrenal glands can be used as a normal site for mammalian red blood cell production. The immunohistochemical analysis of human fetal adrenal gland tissue showed that CD34+ extravascular nucleated GYPA+ cells (Figure 5D and Figure S16B). The author further used image flow cytometry to observe and count the mature red blood cell precursors and enucleated red blood cells of the mice during the perinatal period. Approximately 8% of live free cells in the adrenal gland are composed of mature red blood cells, while the number of live free cells in the kidney is 0.2% (Figure 5E). The adrenal gland is also a site of ongoing red blood cell production, and the distribution of immature to mature red blood cells is very similar to that of adult mouse bone marrow (Figure 5E and F).

Macrophages are more widely distributed. The authors compared all macrophages with microglia in the brain, and performed UMAP visualization and Louvain clustering on them, regardless of other cell types (Figure 5G and H; Figure S16C; and Table S9). It is worth noting that microglia are divided into three subgroups, one of which is marked by IL1B and TNFRSF10D, which may represent activated microglia expressing pro-inflammatory cytokines involved in the normal development of the nervous system. Other signs of microglia clusters are the expression of TMEM119 and CX3CR1 (more commonly in the brain) or PTPRG and CDC14B (more commonly in the cerebellum).

Macrophages outside the brain are divided into three categories (Figure 5G and H; Figure S16C; and Table S9): (i) Antigen-presenting macrophages, mainly present in the gastrointestinal organs (intestine and stomach), are presented as antigens (Such as HLA-DPB1 and HLA-DQA1) and high expression of inflammation-activating genes [such as AHR]; (ii) perivascular macrophages found in most organs, with specific expression markers, such as F13A1 And COLEC12, and markers, such as RNASE1 and LYVE1; (iii) phagocytic macrophages, which are enriched in liver, spleen and adrenal glands (Figure 5I), with specific expression markers, such as CD5L, TIMD4 and VCAM1. Phagocytic macrophages are essential for the elimination of nucleoprotein progenitor cells (so-called extruded nuclei) after late red blood cell removal to form reticulocytes; their observation in the adrenal gland and their potential as additional sites for normal fetal erythropoiesis Same role. Next, the author uses the integration with the mouse organogenesis atlas to study the conservative procedures of the blood cell characteristics and developmental origin of microglia and macrophages.

Science: Atlas of Human Fetal Gene Expression Cells
Figure 4 Identification and characterization of blood cell subtypes and developmental trajectories. (A and B) UMAP visualization and marker-based annotation of blood cell types colored by organ type (A) and cell type (B). (C) UMAP blood cell visualization, integrated into the scRNA-seq map of all organs and human fetal liver blood cells in this study. (D) Dot plot showing the expression of two selectable marker genes for each cell type. (E) The bar graph shows the estimated cell fraction of each organ from each of the 17 annotated blood cell types.




5.   Characteristics of endothelial and epithelial cells across organs

As the second analysis of a single cell category across multiple organs, the authors reclassified 89,291 endothelial cells (ECs), which correspond to vascular endothelial cells (VECs), lymphatic endothelial cells (LECs), or endocardium. These three groups are easily separated, and vascular endothelial cells further aggregate through the organ at least to some extent (Figure S17A-C). Compared with the differences between arteries, capillaries, and veins, organ-specific differences are easier to detect, which is consistent with the previous cell atlas of adult mice. The author performed a comprehensive analysis of endothelial cells from human fetal tissue (this study) and mouse adult tissue (Figure S17, Dand E). Human and mouse endothelial cells are separated first through blood vessels, lymphatic vessels, and endocardium, and then through organs. Vascular endothelial cells from the same tissue usually cluster together, despite differences in species, developmental stages, and technology. Conservative markers of organ-specific endothelial cells are easy to identify (Figure S17F).

Differential gene expression analysis identified 700 markers specifically expressed in ECs subgroups (FDR of 0.05, the expression difference between the first and second clusters was more than two-fold) (Figure S17G and Table S10). Approximately one-third of these encode membrane proteins, many of which seem to correspond to potentially specialized functions. Consistent with the observations in mice, brain endothelial cells specifically express amino acid transport (q=5.6E-10) and carboxylic acid transport (q=4.2E-8) gene sets; lung endothelial cells specifically express 3,5 single The gene set of adenosine phosphate (cAMP) and cyclic nucleotide (q=1.4E-3) catabolism, and the gene set related to stem cell differentiation specifically expressed by vascular endothelial cells from the gastrointestinal tract, heart and muscle ( q=3.7E-2). On the basis of these differences, human fetal endothelial cells express different transcription factors (TFs) (Figure S17H). For example, LECs specifically express TBX1, brain VECs specifically express FOXQ1 and FOXF2, and liver VECs specifically express DAB2, all of which are consistent with observations in mice.

As the third analysis of the widely distributed cell types, the authors reclassified 28262 epithelial cells, which are derived from all organs, and performed UMAP display on them (Figure S18A and B). Although certain epithelial cell types are highly organ-specific, such as acinar (pancreas) and alveolar cells (lung), epithelial cells with similar functions are usually clustered together (Figure S18C).

Figure 5 Identification and characterization of adrenal erythropoiesis and macrophage differentiation. (A) An enlarged view of the portion of the erythropoiesis trajectory in Figure 4B, colored by the red blood cell or megakaryocyte subtype. (B) The graph similar to (A) is colored by the normalized expression of cell type-specific genes (FDR is 0.05, the expression difference between the first and second ranked cell types is more than two times), and the cell type used is specific The number of sex genes and the first few genes displayed. (C) The dot plot and box plot show the proportion of EEP blood cells in each sample of different organs. (D) Representative fluorescence microscopic examination of human fetal adrenal gland tissue, stained with endothelial cells (CD34+) and erythroblasts (nucleated and GYPA+); cell nuclei were stained with blue DAPI. (E) (Left) Percentage of dissociated kidneys and adrenal glands of newborn (P0) mice composed of enucleated red blood cells and mature red blood cells. (F) Representative images of mature red blood cells in the P0 adrenal gland and adult bone marrow. (G and H) UMAP visualization and marker-based annotation of macrophage subtypes colored by organ type (G) and subtype name (H). (I) The dot plot and box plot show the proportion of blood cells that have phagocytosed macrophages for each sample of different organs.



Within the epithelial cells, two neuroendocrine cell clusters were identified (Figure S18C). The simpler one corresponds to adrenal chromaffin cells and is marked by the specific expression of HMX1 (NKX-5-3), which is a TF involved in the differentiation of sympathetic neurons. The other cluster is composed of neuroendocrine cells from multiple organs (stomach, intestine, pancreas, and lungs), and is characterized by the specific expression of NKX2-2, which plays a key role in pancreatic islet and enteroendocrine differentiation TF. The author further analyzed the latter group and identified five subgroups (Figure S18, D to F): (i) islet β cells, marked by insulin expression; (ii) islet α and γ cells, expressed by pancreatic polypeptide And glucagon expression as a marker; (iii) pancreatic islet delta cells, marked by somatostatin expression; (iv) pulmonary neuroendocrine cells (PNECs), marked by the expression of ASCL1 and NKX2-1, these two TFs Play a key role in specifying this lineage in the lung; and (v) enteroendocrine cells. Enteroendocrine cells further include several subpopulations, including pancreatic islet epsilon progenitor cells expressing NEUROG, enterochromaffin cells expressing TPH1 in the stomach and intestine, and G, L, K, and I cells expressing gastrin or cholecystokinin. Finally, the authors observed ghrelin-expressing enteroendocrine progenitor cells in the stomach and intestines, and also observed ghrelin-expressing endocrine cells in the developing lung (Figure S18F). The multiple functions of neuroendocrine cells are closely related to their secreted proteins; the authors identified 1086 secreted protein-coding genes that are differentially expressed in neuroendocrine cells (FDR is 0.05) (Figure S18G and Table S11). For example, PNEC exhibits specific expression of trefoil peptide factor, which is involved in mucosal protection and lung ciliary cell differentiation; gastrin release peptide, which stimulates gastric G cells to release gastrin; SCGB3A2, a type related to lung development The surface active agent.

The authors used these data to explore cell trajectories and further studied the ways in which the diversification of epithelial cells leads to renal tubular cells. Combining and reclassifying ureteral buds and metanephric cells, the authors determined the types of precursor and end-renal epithelial cells, and their differentiation pathway is highly consistent with recent studies on human fetal kidneys (Figure S19A). Through differential gene expression analysis, the authors further determined that TFs may regulate their specifications (Figure S19B and Table S12). For example, nephron progenitor cells in the metanephric trajectory specifically express high levels of mesenchymal and meis homeobox genes (MEOX1, MEIS1, and MEIS2), while renal sac visceral cells specifically express MAFB and TCF21/POD1. Another example is that HNF4A is specifically expressed in proximal tubule cells. Mutations in this gene cause Fanconi’s renal tubular syndrome, a disease that specifically affects the proximal tubules. HNF4A has recently been shown to be required for the formation of proximal tubules in mice.



6 Integration of human and mouse developmental maps

The transition from embryonic development to fetal development is quite important, but obtaining human embryonic tissue is more limited than obtaining fetal tissue. In order to reuse the mouse, the author tried to integrate these human fetal data with the Mouse Organogenesis Cell Atlas (MOCA). The author had previously analyzed 2 million cells from unisolated embryos between E9.5 and E13.5. In this context, this window corresponds to the 22nd to 44th day of human development, and the tissues studied here are estimated to be from the 72nd to the 129th day.

First, the author compared the 77 main cell types defined here with the organogenesis and development trajectory defined by MOCA through the method of cell type cross-matching. Most human cell types strongly match a major mouse trajectory and subtrajectories (Figure S20 and Tables S13 and S14). These are usually in line with expectations, although some differences help to correct MOCA (see Figures S20 and S21). Many human cell types and mouse cell types that lack a strong 1:1 match [summed non-negative least squares (NNLS) regression coefficient <0.6] correspond to tissues excluded from other data sets (for example, mouse placenta and human skin) And gonads). Other ambiguities may come from gaps in the developmental window studied (such as adrenal cell types), rarity (such as bipolar cells), and/or complex developmental relationships (such as fetal cell types derived from multiple embryonic trajectories).

Second, the author tried to co-implant human and mouse cells directly together. In short, the author extracted 100,000 mouse embryonic cells and about 65,000 human embryonic cells (up to 1,000 cells for each of 77 cell types) from MOCA (random), and conducted a comprehensive analysis of them. The resulting distribution of mouse cells in the UMAP visualization is similar to the author’s overall analysis of MOCA (Figure 6A-C and Figure 6S21 to S23). In addition, despite species differences, the distribution of human embryonic cells is absolutely dominant in a way that respects the developmental relationship between cell types. For example, human fetal endothelial cells, hematopoietic cells, hepatocytes, epithelial cells, and mesenchymal cells all map to the corresponding mouse embryo trajectories (Figure 6B and Figure S21). In each main trajectory, mouse cells are arranged in consecutive time points, while human embryonic cells appear to project from the last (E13.5) mouse embryo time point (Figure 6C). At the lower foregut level, the seniscal map includes human fetal intestinal epithelial cells from the lower hindgut of the mouse midgut; human fetal parietal cells and principal cells (stomach), as well as acinar cells and ductal cells from the lower foregut epithelium of mice ( Pancreas); human fetal bronchioles and alveolar epithelial cells from mouse lung epithelial trajectories; human fetal ureteric buds and metanephric cells from mouse embryonic kidney epithelial trajectories; and many other cells (Figure S21=S23).

However, there are some surprises. For example, although central nervous system (CNS) neurons map to neural tube trajectories, enteric nervous system (ENS) glial cells and Schwann cells map to peripheral nervous system (PNS) glial cell trajectories, some neural crest derivatives include ENS neurons, visceral neurons, sympathetic neuroblasts, and chromaffin cells aggregate separately from the corresponding mouse embryo trajectories (Figures S21 to S23), which may be due to excessive differences in developmental stages or species. Human embryonic astrocytes accumulate on the trajectory of mouse embryonic neuroepithelium [mouse astrocytes do not develop until E18.5]. Human fetal oligodendrocytes overlap with a rare mouse embryonic lower segment (Pdgfra+ glial cells). In retrospect, this cell is more likely to correspond to oligodendrocyte precursors (Olig1+, Olig2+ and Brinp3+) , Which made the author’s previous comments on different Olig1+ subsections as precursors of oligodendrocytes questioned. These and other unexpected relationships deserve further investigation.

In order to evaluate the relationship between mouse embryonic cells and human embryonic cells in more detail, the authors applied the same strategy to extract cells from hematopoietic (Figure 6D and Figure S24), endothelium (Figure S25) and epithelium (Figure S26). In these visualizations, the authors observed that organ-parsed human data deconvolved the entire embryonic mouse data into a more fine-grained subset. For example, subpopulations of mouse leukocyte embryonic subregions map to specific human blood cell types, such as HSPC, microglia, macrophages (liver and spleen), macrophages (other organs), and dendritic cells (DC ) (Figure 6D). These subpopulations were further verified by the expression of relevant blood cell markers (Figure S24C), and annotated in co-implantation based on their human k-nearest neighbors (k=3) (Figure S24D).

Among 1087 human fetal blood cell type-specific gene markers, there are 337 genes that are differentially expressed in the same cell type (FDR is 0.05) (Figure 6E and Table S15; in comparison, only 12 genes cross after label arrangement) . Among the 337 conserved markers, 28 are TFs, of which 24 have been previously reported to be involved in early blood cell differentiation or maintenance of target cell types. For example, HLF is a key regulator of HSPC quiescence, MITF is a factor that drives mast cell differentiation. PAX5 is the main regulator of B cell development, and SOX6 promotes the differentiation of erythroid progenitor cells. However, 4 of the 28 conservative markers TFs have not been described in the relevant context: NR1D2 in IL3 cells, TCF7L2 in macrophages, FHL2 in megakaryocytes, and NUAK1 in microglia.

In the same analysis, human fetal macrophages and microglia formed different clusters, but they were connected by a subset of mouse cells from the leukocyte trajectory (Figure 6D), which is in contrast to previous studies showing that these two These cell types are consistent with the differentiation of yolk sac progenitor cells. To explore this further, the authors used unsupervised trajectory analysis to extract and reanalyze 4327 mouse embryonic microglia and macrophages. The authors observed three smooth cell differentiation trajectories, from a common progenitor cell in the brain to microglia, phagocytic macrophages (TIMD4+ and CD5L+; mainly in the liver, spleen and adrenal glands) and perivascular macrophages (F13A1+ And LYVE1+; widely distributed) (Figure S27A and Figure 5). The direction of progression along the trajectory of each macrophage through the false time is consistent with the actual development time (Figure S27B). There are a total of 1412 genes, including 111 TFs, which are differentially expressed in the three macrophage branches (Table S16). For example, the microglia trajectory showed increased expression of BACH2 and RUNX3 and the known microglia regulatory factors SALL1 and MEF2A, perivascular macrophages DAB2 and TCF7L2, and phagocytic macrophages MAFB and NR1H3 (Figure S27C). Collectively, these analyses illustrate how fetal annotations can be used to identify and describe the characteristics of progenitor cells of a particular lineage at developmental time points where they may be difficult to resolve individually or even across species.

Figure 6 Integrating human and mouse embryonic cell maps. (A) The cells are colored by the source species. (B) Mouse cells are colored by the identity of the main mouse embryo trajectory. Human cells are gray. (C) Cells are colored according to source and developmental stage. (D) The author used Seurat to analyze 103766 human and 40606 mouse hematopoietic cells. (E) A graph similar to (D), colored by the normalized expression of human-mouse conserved cell type-specific genes.







All living things are composed of cells, and cells are the most basic unit of life. Two centuries after this cell theory was put forward, researchers are classifying and characterizing all cell types that make up the human body, including health and disease. For this reason, the field of single-cell biology is developing at an alarming rate, which is driven by the synergy between new technologies and new computing methods. In the past few years alone, this synergy has made the single-cell maps of many human organs and entire model organisms attractive and informative.

Human development is an extraordinary process, starting from a fertilized egg, going through a germinal stage, and then embryogenesis. By the end of the 10th week, the embryo has acquired its basic shape and is called a fetus. During the next 30 weeks, all organs continued to grow and mature, and different terminally differentiated cell types were produced from their progenitor cells. Although the single-cell method has been used to conduct in-depth research on the germinal and embryogenesis stages of humans and mice, the research on the fetal stage is more challenging. Although there have been some single-cell studies on human fetal development recently, these studies are limited to a single organ or cell lineage and cannot obtain a comprehensive view.

In this study, the authors set out to use different tissues obtained during human embryo development to generate single-cell maps of gene expression and chromatin accessibility. From 15 different organs, the author successfully analyzed the gene expression of about 4 million single cells and the chromatin accessibility of about 800,000 single cells. The limitations of these data sets include uneven sampling (that is, there are more cells in some organs than in others), tissue loss (most notably bone marrow, skin, bone, and gonads), relatively low sequencing depth, and The sparseness of the single-cell molecular map. Nevertheless, the authors identified hundreds of cell types and subtypes, which are supported by a framework for quantifying specificity, and by comparing almost all cell types and subtypes with published mouse The cell types or subtypes in the map match.

In contrast to organ-specific studies, the diversity of tissues described here makes cross-tissue comparisons of widely distributed cell types possible. The author’s process of annotating cell types has benefited greatly from the large number of single-cell maps of specific human organs or other mammals that have been produced so far. Of course, the decision in the annotation process may be subjective (for example, over-clustering and under-clustering), and the cell type and sub-type annotations made here should be considered preliminary and need to be modified.

The obvious hematopoiesis observed by the author in the fetal adrenal glands is consistent with the fact that the adrenal glands and many other organs (such as the spleen, liver, and lymph nodes) can be used as sites for hematopoiesis outside the bone marrow of adults. These pathological factors lead to an increase in the demand for blood cell production. Especially hemoglobinopathies. Although extramedullary hematopoietic islands can occasionally be seen in the adrenal glands of human embryos, the author’s findings in humans and mice provide quantitative evidence that the adrenal glands are normal, albeit small. The hematopoietic transition of the bone marrow overlaps the erythropoiesis site.

The authors integrated the single-cell characteristics of mouse organogenesis and human embryonic development, especially considering that these represent different stages of mammalian development, not to mention the evolution of more than 100 million years of separation between humans and mice. The relatively straightforward arrangement of the data set highlights the degree of evolutionary limitations on the molecular programs of individual cell types, and it further supports the long-term use of the mouse as a powerful model system for studying human development.

Looking to the future, the authors envision that the narrow window of human development in the second trimester studied here will be supplemented by additional maps at early and late time points (such as embryos and adults), as well as similar comprehensive analysis and integration of data from model organisms. Continuous development and application of methods for determining gene expression and chromatin accessibility consistent with spatial, epigenetics, proteomics, pedigree history and other information are essential for the time to obtain the diversity of human cell types starting from single-cell zygotes A comprehensive view of the expansion is necessary.

So far, most of the research on human development is indirect. The key molecular factors are nominated by human genetics and then studied in model organisms and/or in vitro systems. In vivo studies on gene expression and regulation are still limited. In the process of filling this gap, the author hopes that this map can better understand the molecular and cellular basis of rare and common developmental disorders in humans, and at the same time provide information for successful treatment.






(source:internet, reference only)

Disclaimer of