BMDB: a comprehensive database and web server for integrated single-cell bone marrow microenvironment transcriptomic data

doi:10.21203/rs.3.rs-3931582/v1

Download PDF

Method Article

BMDB: a comprehensive database and web server for integrated single-cell bone marrow microenvironment transcriptomic data

https://doi.org/10.21203/rs.3.rs-3931582/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Background

The bone marrow (BM) niche is a specialized microenvironment housing the hematopoietic stem and progenitor cells (HSPC) and orchestrating hierarchical hematopoiesis pathways. However, understanding its molecular and cellular intricacies remains incomplete. Single-cell RNA sequencing (scRNA-seq) technology has developed into a powerful tool for investigation of BM niche heterogeneity and functional diversity.

Methods

Here, we present BMDB, the first dedicated web-based data resource of BM niche transcriptome and tools for interactive data analysis; available at http://bmdb.jflab.ac.cn:18083/app/bmdb.

Results

BMDB features two sections: (ⅰ) a well-managed database compiling 123,915 single cells derived from 45 healthy and 20 diseased samples and (ⅱ) versatile data analysis tools that allows users to perform customized analysis on built-in datasets and users’ data. BMDB also provides access to the aforementioned high-quality mouse and human single-cell reference atlases of the BM niche. The core objectives of the analysis module are to identify distinct cell types, delineate their developmental trajectories, and elucidate their potential roles in hematopoiesis, including molecular mechanisms.

Conclusion

BMDB will serve as a significant resource for the study of the BM niche by virtue of its extensive functionality and analytical capabilities.

bone marrow niche

single-cell RNA sequencing

web server

reference atlas

Hematopoiesis is the process by which all mature blood cell types are produced from hematopoietic stem and progenitor cells (HSPCs). The bone marrow (BM) niche refers to a specialized microenvironment within bones where the HSPC pool resides, facilitating the coordinated functioning of various hierarchical hematopoiesis pathways(1, 2). The concept of BM niche was first proposed by Schofield in 1978, and it has gained widespread acceptance(3, 4). The investigation into the BM niche commenced at an earlier time. These studies have identified two primary niches within the BM, namely the endosteal and vascular niches(5, 6). In more recent times, the vascular niche has been subdivided into the arteriolar and sinusoidal niches, based on distinct cellular characterization, localization, and function(7–9). In the diverse niches of BM, HSPCs engage in interactions with a range of cell types, including osteoblasts, endothelial cells, macrophages, and mesenchymal stromal cells (MSCs)(10). These interactions serve to either maintain HSPCs in a quiescent state or support their processes of proliferation, differentiation, and trafficking, depending on the physiological requirements of the body(11, 12). However, due to the inherent complexity and heterogeneity of the BM niche, numerous aspects of its functioning remain incompletely understood(13).

In recent years, the advancement of single-cell RNA sequencing (scRNA-seq) technology has presented an unparalleled chance to investigate the intricate cellular diversity and complex microenvironment linked to various developmental, physiological, and pathophysiological processes(14–17). These processes include hematopoiesis, inflammation, neurodegenerative diseases, and cancer(18, 19). Notably, in the field of cancer research, there has been a significant surge in both the quantity and quality of scRNA-seq data over the past few years(20, 21). The extensive amount of public scRNA-Seq data necessitates the implementation of efficient quality control measures, platform and analysis evaluation, data mining and integration, and information management. Consequently, databases like CancerSCEM and TISCH have been established to curate the accumulating data and offer single-cell gene expression profiles in the field of cancer biology(22, 23). In comparison to solid tumor tissue, hematopoietic tissue, such as bone marrow, presents a more convenient option for scRNA-seq investigations due to its accessibility as a single-cell suspension. Furthermore, the presence of established molecular markers and functionally validated cell types adds to the appeal of studying the hematopoietic system(24–26). Recent significant studies have shed light on the molecular intricacies of the BM niche in mice, leading to a revised understanding(27, 28). For example, Baryawno et al. have identified 17 distinct cell types with varied functions in the murine BM stroma, emphasizing its complexity(29). However, unlike cancer research field, there is currently no available database to effectively utilize the abundant scRNA-seq data resources pertaining to the dynamic BM niche, especially for comprehensively analyzing the heterogeneity of BM microenvironment and dissecting the hematopoiesis mechanisms.

In this study, we have developed a specialized web-based portal called BMDB, which has been designed to gather, analyze, and visually represent scRNA-seq data pertaining to the bone marrow microenvironment. As of March 2023, BMDB successfully collected public BM niche datasets encompassing 123,915 single cells derived from 45 normal and 20 pathological samples, encompassing both human and mouse subjects. BMDB provides two interactive modules, "Exploration" and "Reference Atlas", for the exploration and analysis of the built-in data, including cell clustering and annotation, cell differentiation trajectory, pathway enrichment, cell-cell communication. Furthermore, BMDB offers a comprehensive online tool called "De novo Analysis" to facilitate users' data analysis. In conclusion, it is anticipated that BMDB will serve as a significant asset in the study of the biochemical metabolomics niche.

2.1. Contents of BMDB

The current version of BMDB (http://bmdb.jflab.ac.cn:18083/app/bmdb) was established from 14 bone marrow niche-related scRNA-seq datasets in homo sapiens and mus musculus, involving 123, 915 single cells (Fig. 1A, Supplementary Table 1). Among them, 45 samples were generated from normal tissues and 20 samples were generated from pathological tissues (Fig. 1A, Supplementary Table 1).

2.2. Features and utilities

Framework of BMDB The framework of BMDB web server encompasses four primary functional groups: "Project Catalog" for data management, "Exploration" for display of integrated datasets, "Reference Atlas" for investigating reference atlases, and "De novo Analysis" for users' data analysis (Fig. 1B). Within the "Exploration" group, the "Project Information" sample manager is comprised of "Publication Info" and "Sample Info", along with four interactive visualization modules: "Clusters Visualization", "Developmental Trajectory", "Pathway Enrichment", and "Cell-cell Communication" (Supplementary Fig. 1A). Detailed information can be acquired by accessing the DOI link leading to the publication's homepage. The utilization of "Sample Info" facilitates the management of data and associated reference information. The establishment of the "Clusters Visualization" module enables fundamental data visualization, encompassing cell type, percentage of a particular cell type, and gene expression within cell clusters. The "Developmental Trajectory" module serves the purpose of visualizing the cell-fate trajectory within the chosen lineage. Furthermore, the "Pathway Enrichment" module is employed to visualize the enrichment of pathways enriched with highly variable genes among distinct cell clusters. The term "cell-cell" is employed to visually represent the activation of intercellular signaling pathways between various cell types.

Construction of reference atlas The construction of a high-quality atlas for mouse bone marrow niche cells was accomplished by utilizing a pipeline that integrates canonical analysis methods (Fig. 1C, Supplementary Fig. 1B and Supplementary Fig. 1C), based on three representative datasets(29–31). The UMAP plots were used to present the outcome of the initial cell clustering, which encompassed all cells. Subsequently, a manual filtering process was conducted to attain unbiased cell type annotation, utilizing classical gene markers from a previously published article(29). This annotation categorized the cells into six types, namely MSCs, Chondrocyte-like cells(CLCs), Osteolineage cells(OLCs), fibroblasts, Endothelial cells(ECs), and pericytes (Supplementary Fig. 2). Further refinement of the annotation was carried out for each cell type, employing marker genes identified by Findmarkers (Supplementary Fig. 3, Supplementary Table 2). The subsequent procedure for identifying cell types involves a sequential set of actions, encompassing the elimination of cells with excessive potential, verification through developmental trajectory analysis, and integration of clusters through manual intervention. Subsequently, a final manual correction of cell types was executed to establish a comprehensive transcriptome reference, which will be utilized in the BM landscape project. A comparable approach was employed to construct the human fetal single-cell reference atlas of BM niche, drawing upon the datasets that have been previously published(32). Based on established cell markers, the human fetal BM niche was found to contain three primary cell types: hematopoietic cells, ECs, and stromal cells (Supplementary Fig. 4A). Following the removal of hematopoietic cells, further annotation was conducted on both ECs and stromal cells using previously published marker genes(32). This analysis revealed four subsets of ECs (aECs, ECs_prolif, lyECs, and sECs) and three types of stromal cells (Bone marrow stromal cells(BMSCs), Chondrocytes, and OLCs) (Supplementary Fig. 4B). Subsequently, meticulous annotation was conducted in BMSCs, Chondrocytes, and OLCs with the objective of delineating the subclusters based on documented cell markers. Specifically, two distinct subsets of BMSCs, namely Cxcl12-abundant-reticular cells(CARs) and fibroblast, were identified, along with two subsets of Chondrocytes, namely Chondrocytes and Chondrocytes-prolif, and two subsets of OLCs, namely Osteoblasts(OBs) and Osteoprogenitor cell(OPs), were respectively assigned annotations (Supplementary Fig. 4C, Supplementary Fig. 5). The "Reference Atlas" page lists all the processed datasets and associated information. For further analysis and visualization, users can view the UMAP of integrated atlas and detailed demonstration of cell types composition and genes expression (Fig. 2).

Data curation and basic display The dataset management process was conducted by BMDB utilizing an internally developed pipeline. BMDB exhibits a user-friendly interface and offers a versatile access pathway, enabling users to efficiently retrieve information on individual datasets and their associated details. Upon selecting the "Dataset" option on the homepage, the Dataset catalog will appear, presenting a comprehensive collection of currently amassed datasets. Pertinent information such as publication details (author, journal, and year), species, tissue, and cell count will be readily displayed within the “Project Catalog”. By clicking the "Info" button (Fig. 2A), users are directed to the last column, which provides access to detailed publication information and sample information. The "Publication information" page displays the DOI, title, and abstract of the publication. On the "Sample Information" page, users can access comprehensive and structured details about the sample, including Data ID, species, tissue, cell count, and library construction method. Furthermore, clicking the "View" button in the last column of the "Sample Table" reveals the cell clustering and annotation.

Users can have the option to select from three clustering types, namely main annotation, fine-tune annotation, and Louvain cluster (Fig. 3B). Fine-tune annotation, in comparison to main annotation, notably enhances the accuracy of identifying cell types. The outcomes of cell clustering will be visually depicted using UMAP plots, while the distribution of each cell type will be represented using a pie chart. In order to alleviate the computational burden on the system, BMDB offers the capability to sample cells by a predetermined percentage. This allows for proportional sampling of cell clusters, facilitating visualization. Furthermore, by choosing a gene from the provided drop-down menu, the expression pattern of the selected gene across various cell types will be displayed (Fig. 3C).

Extended display of built-in data The "Exploration" functional module allows users to conduct various complex analyses, such as cell differentiation trajectory, pathway enrichment, and cell-cell communication (Fig. 4). On the trajectory analysis page, users have the option to customize several parameters in order to obtain the temporal-spatial development patterns of the chosen cell lineage through a drop-down box (Fig. 4A). By selecting the "State of Trajectory" using the "State Type" drop-down box, the monocle state of the selected cell lineage will be displayed on DDRTree. Furthermore, the subcluster of the cell lineage can also be projected onto the DDRTree when “Clusters” are selected via the drop-down box. Pseudotime analysis is conducted by utilizing monocle2 and slingshot. As previously stated, users have the option to sample cells based on a predetermined percentage for this analysis. To investigate the involvement of the gene of interest in the development of a specific cell type, users can choose the gene from the activated gene list. DDRTree will display the gene expression in each individual cell, along with the demonstration of gene expression dynamics along the pseudotime trajectory.

The "Pathway Enrichment" module facilitates the execution of pathway analysis on cell clusters. Through the selection of specific pathways, such as GO:0030953 and GO:0002062, from the provided drop-down box, users are able to visually represent their expression patterns using UMAP plots (Fig. 4B). For instance, the expression of pathway GO:0030953 is observed across all cell clusters, whereas the expression of pathway GO:0002062 is predominantly found in EC cells.

The module "Cell-cell Communication" provides users with the ability to analyze the process of cell-cell communication (Fig. 4C). To illustrate this concept, we have utilized the MK signaling pathway as an example, which can be selected from the drop-down box. On the bubble chart page, the y-axis displays the various pairs of Mdk and their respective receptors, while the x-axis represents the pairs of two different cell types. The color of each bubble corresponds to the level of contribution that the Mdk-receptor pair on the y-axis has towards the communication between the cell pair on the x-axis. Mdk-sdc4 exhibits a significant role in facilitating communication between MSC.5–19 and various cell types, such as CLCs.3, CLCs.4-14-27, fibroblasts.2, and fibroblasts.13–20. Furthermore, the comprehensive impact of the integrated signaling pathway can be visually represented through the utilization of a heatmap, network graph, and circle plot. The findings suggest that the Mdk signaling network primarily serves as a mediator for cell communication within the bone marrow niche, connecting MSCs or OLCs with other cellular entities. Violin plots can effectively depict the expression levels of the individual gene within each cell type along the pathway. In particular, Mdk exhibits predominant expression in MSCs and OLCs. The receptors associated with Mdk display diverse expression patterns among bone marrow cells. Notably, sdc1 is primarily expressed in MSCs, fibroblasts, and OLCs, while Ncl and ltgb1 exhibit widespread expression across all cell types within the bone marrow niche.

De novo analysis for users’ data BMDB platform incorporates an online de novo analysis function for users' data, enhancing its capabilities as a robust tool for the exploration of bone marrow scRNA-Seq data (Fig. 5). As previously stated, the current version of BMDB offers a comprehensive collection of high-quality mouse and human fetal single-cell reference atlases of the bone marrow niche. The availability of high-quality reference atlases enables the realization of online data analysis for user data in BMDB. BMDB supports both H5AD and RDS file formats. Upon uploading the files to the "De novo Analysis" section, cell mapping can be conducted by selecting the appropriate reference atlas and mapping method from the provided drop-down box (Fig. 5A). Two optional mapping methods, namely WNN and scANVI, are available for cell mapping (Fig. 5B). Additional analysis can be conducted using the function modules incorporated within the "De novo Analysis" group, encompassing cell clustering and annotation, pathway enrichment, and cell-cell communication.

In the physiological state, the regulation of hematopoiesis occurs with precision and speed in order to accommodate the fluctuating needs of the body(33). The bone marrow HSPC niche assumes a crucial function in this hematopoietic process(34). This specialized microenvironment primarily consists of mesenchymal stem/stromal cells and vascular endothelial cells(1, 35). In addition, osteoblasts, chondrocytes, macrophages, and even non-myelinating Schwann cells also contribute to BM HSPC niche(4, 36–38). Due to the advancements in scRNA-seq technology, researchers have acquired extensive knowledge regarding the molecular, cellular, and spatial arrangement of the bone marrow (BM) niche(39, 40). This understanding has been facilitated by the substantial accumulation of scRNA-seq data over the past decade(30, 31, 41). Nevertheless, the absence of a comprehensive platform that consolidates the scRNA-seq data pertaining to the BM niche from various experiments, samples, and species remains a significant gap. Consequently, the development of BMDB, a specialized database with web tools for exploring the BM niche at the single-cell level, was undertaken. To the best of our understanding, BMDB represents the initial specialized web-based portal aimed at characterizing cellular identities and functions of distinct elements within the BM niche at a resolution of individual cells. It encompasses the most comprehensive collection of single-cell transcriptomics data pertaining to the BM niche up until the present time. BMDB is comprised of three primary functional categories, namely "Exploration," "Reference Atlas," and "De novo analysis." The former two categories are designed for integrated data visualization and exploration of the reference atlas, respectively. The construction and presentation of a comprehensive single-cell reference atlas of the mouse and human fetal bone marrow niche has been accomplished and made available in BMDB. This resource can be downloaded by users and facilitates local data analysis. Notably, BMDB distinguishes itself from other databases through its exceptional "De novo Analysis" functional group, which offers online analysis capabilities for user-generated data. Furthermore, BMDB has successfully implemented visual analytics for all data within its platform.

Currently, BMDB has successfully accumulated 14 scRNA-seq datasets pertaining to the BM niche, encompassing 10 normal and 4 pathological datasets. We hold the belief that BMDB will prove to be an invaluable resource in facilitating our comprehension of the heterogeneity within the BM niche, as well as the underlying mechanisms of hematopoiesis, both in physiological and pathophysiological states, at a single-cell resolution. The majority of extant scholarly literature concerning the BM niche predominantly originates from inquiries conducted on animal models. Consequently, out of the 65 samples amassed in the present BMDB, 29 are derived from mouse BM samples, with the remaining dataset sourced from human fetal BM samples (Supplementary Table 1). As a result, the reference atlases within the BMDB solely pertain to the mouse and human fetal BM niche. Although the mouse and human datasets encompass data from numerous shared cell types, certain cell types exhibit discernible differences(42–44). Currently, there is a lack of comprehensive understanding regarding certain crucial components within the human BM, primarily because the investigation of stromal stem/progenitor cells has proven to be challenging(45, 46). However, recent studies have started shedding light on the intricate nature of the human BM niche and the potential varied functional capacities of BM stromal cells. Consequently, we intend to persistently monitor BM niche scRNA-seq studies and enhance the BMDB database by incorporating additional datasets, particularly those derived from human BM samples. Furthermore, future efforts will be directed towards enhancing the web tools by exploring alternative approaches. Specifically, a comprehensive reference atlas of the BM niche will be constructed, with a particular focus on human adults and various pathological conditions. As the inaugural scRNA-seq database and web tools for the BM niche, BMDB will continue to interpret additional datasets, develop user-friendly analytical modules, and ultimately facilitate further investigations into the HSPC niche within the bone marrow.

4.1. Data collection

We searched PubMed and Google scholar for the bone marrow niche-related single-cell transcriptome datasets using the following keywords: (‘bone marrow niche’ AND ‘single cell’) OR (‘bone marrow’ AND ‘single cell’) OR ‘bone marrow niche’. A total of 14 bone marrow niche-related datasets were obtained from this systemic search, which comprise 45 datasets of normal samples and 20 of pathological samples. All datasets were collected before March 2023.

4.2. Data processing pipelines

Data processing encompasses the tasks of reference construction and sample handling, which entail the unification of data formats and the execution of extended analysis. In addition, reference construction primarily entails the amalgamation of superior data obtained from diverse sources.

Data filtering For further processing and analysis, high-quality data from each dataset were meticulously chosen. Seurat (v4.3.0) was employed to generate a unified format, known as SeuratObj, for these data. Raw transcript counts of gene-cell matrices were subjected to filtration to eliminate cells with transcript counts below 500 and cells with a proportion of mitochondrial genes exceeding 10%. Genes expressed in less than three cells were also removed from the analysis. Subsequently, outlier cells containing more than 5,000 genes were also removed due to the possibility of being cell doublets. Additionally, a gene list of specific hematopoietic cell markers was chosen, and cells expressing any of these markers were eliminated from the pre-filtered data.

Data normalization The filtered data was then normalized using LogNormalize, and the standard variance was calculated using the VST algorithm. The top 3000 genes (selection.method = 'vst', nfeatures = 3000) was considered as high variable genes (HVGs) for further downstream principal component analysis (PCA).

Cell clustering We initially selected the top 10 principal components (dims = 1:10) to construct a k-nearest neighbors (kNN) graph. Subsequently, we utilized this kNN graph to construct a shared nearest neighbors (sNN) graph by computing the Jaccard index, which measures the neighborhood overlap, between each cell and its 20 nearest neighbors (k = 20). The Louvain algorithm with multilevel refinement (algorithm = 2) was then employed to identify communities (clustering), and more refined clusters were obtained by setting the resolution at 2.0. Similarly, the top 10 principal components were employed to calculate the cosine distance between two cells (metric = 'cosine'), followed by UMAP dimensionality reduction using uwot (v0.1.14).

Cell type annotation For the datasets without the information of cell type annotation, we carried out the cell type prediction for the SeuratObj in unified format by using SingleR (v2.0.0). HPCA data (HumanPrimaryCellAtlasData) and mouse bulk expression data(MouseRNAseqData)was used as reference data for human and mouse samples, respectively. Based on Louvain clusters, cell type annotation was performed by using two groups of labels (main and fine). The cell type derived from main annotation was defined as lineage, which was used for subsequent data processing. For the cell cluster, which could not be annotated, we used Unknown + number for labeling. As for the dataset with the information of cell type annotation, we used its original one. The preprocessed and annotated data were stored in the format of rds.

4.3. Advanced data processing

Advanced data processing consists of trajectory inference, cell-cell communication inference, and pathway enrichment analysis. Data undergoing the processing will be stored in the format of h5ad.

Trajectory analysis To conduct the analysis of developmental trajectories, we employed monocle2(47) and slingshot(48) algorithms to establish two distinct analytical frameworks. Subsequently, lineages comprising more than 50 cells were selected for subsequent investigation. The examination of developmental trajectories encompassed all identified lineages. Within each lineage, the genes expressed in a minimum of three cells, as well as the HVGs, were identified and considered as ordering genes. To reduce dimensionality, the DDRTree algorithm was employed. The identical datasets will be employed for conducting slingshot analysis in order to derive pseudotime. Furthermore, the integration of SPRING into BMDB facilitated the construction of a force-directed graph, serving as a complementary tool to pseudotime. In terms of integration analysis, BMDB enables SPRING Viewer to map monocle2-derived pseudotime. In the construction of a force-directed graph, the functional parameters were set to default values, which included the removal of cells with total UMI counts lower than 1000, the removal of genes with mean expression less than 0.1, the removal of genes with Fano Factor less than 3, the utilization of the top 20 principal components to construct k-nearest neighbors, where k was set to 5, and a fold-change of 1 for the coarse-grain.

Cell-cell communication The CellChat(49) ligand-receptor repository was employed to forecast communication patterns among cell types within each tissue, relying on interactions between receptors and ligands. In order to obtain a mean value of co-expression for each receptor-ligand pair, a comparison of receptor-ligand expression was conducted between every two cell types within each tissue. Subsequently, the cell type labels were randomly permuted multiple times to disrupt the biological significance between cells, thereby generating a background null distribution of expression values that is specific to each interaction pair of the two cell types. Interaction pairs with co-expression values that were found to be significantly higher than the background (P < 0.05) were selected and subsequently included in the BMDB, along with receptor-ligand profiles for each cell type in all tissues.

4.4. Construction of single-cell BM niche reference atlas

The construction of a comprehensive atlas of the normal mouse bone marrow niche at the single-cell level was based on three representative datasets(29–31). Following data collection, the construction of a reference atlas was carried out using a "five-step" strategy, which included: 1) conducting quality control of the scRNA data from each individual sample, 2) normalizing the scRNA data from each individual sample, 3) integrating the data and correcting for batch effects, 4) reducing dimensionality and clustering the integrated data, and 5) annotating cell types in the integrated data. The detailed procedure is described as follows. We performed a two-step quality control analysis of the scRNA data. Initially, we eliminated hematopoietic cells by utilizing specific marker genes, namely CD3A, CD3B, CD3C, CD3D, CD3E, CD4, CD8A, CD8B, CD8E, CD8G, CD19, TFRC, LY6G, LY76, VWF, PTPRC, and ITGAM. Subsequently, we excluded cells with an expression of genes below 500 or above 5000, as well as those with mitochondrial genes exceeding 5%. Additionally, we disregarded genes expressed in less than three cells. In the second phase of our methodology, we performed data normalization using the LogNormalize technique, followed by data scaling within the range of 0 to 10. To mitigate batch effects and enable dataset integration, we utilized the SelectIntegrationFeature function in Seurat to identify 3,000 features. Subsequently, data scaling and Principal Component Analysis were applied to each individual sample dataset using the selected features. Anchors were determined through the use of robust principal component analysis (RPCA) algorithms, facilitating dataset integration and batch effect correction. After completing data integration and normalization, the integrated data underwent dimensionality reduction and cell clustering. The selection of the top 3,000 variable genes as HVGs was performed using the vst method, which were subsequently utilized for PCA. The standard deviation (SD) of expression levels for the relevant genes in the principal components (PCs) was calculated, and the top 10 PCs based on SD were chosen for UMAP dimensionality reduction. To ensure a high-quality display effect, the minimum distance (min.dist) was set at 0.05. For cell clustering, the initial step involved selecting the top 10 PCs (dims = 1:10) to construct a kNN graph. Based on the aforementioned information, the Louvain algorithm with multilevel refinement (algorithm = 2) was employed to identify communities (clustering) with a resolution of 0.8. Subsequently, cell type annotation was conducted following the clustering of cells. Utilizing established cell markers, we assigned the normal mouse BM niche cells to six primary lineages, namely MSCs, OLCs, Fibroblasts, ECs, CLCs, and Pericytes. In order to annotate the subclusters within the main cell types, we initially identified marker genes using Findmarkers due to the absence of well-established markers. we conducted trajectory analysis on individual cell lineages using Louvain clusters, excluding any outliers. Subsequently, we merged the Louvain clusters that were positioned similarly in DDRTree, resulting in the generation of subclusters within the cell lineages. By calculating the differential expression of genes between these subclusters, we identified marker genes specific to each subcluster. Lastly, we annotated the subclusters within the primary cell lineages found in the normal mouse bone marrow niche.

A comparable approach was employed to create the atlas of the human fetal BM niche, albeit with minor adjustments. The scRNA data of the human fetal BM, encompassing 36 samples, was obtained from a previously published paper. Samples with a cell count below 200 were excluded in the construction of the human fetal BM niche atlas. The marker genes utilized for annotating the primary cell types or subclusters were sourced from published literature.

Bone marrow

HSPCs

Hematopoietic stem and progenitor cells

MSCs

Mesenchymal stromal cells

scRNA-seq

single-cell RNA sequencing

OLCs

Osteolineage cells

ECs

Endothelial cells

CLCs

Chondrocyte-like cells

BMSCs

Bone marrow stromal cells

CARs

Cxcl12-abundant-reticular cells

SMCs

Smooth muscle cells

OBs

Osteoblasts

OPs

Osteoprogenitor cells

HVGs

High variable genes

kNN

k-nearest neighbors

sNN

shared nearest neighbors

PCA

Principal component analysis

RPCA

Robust principal component analysis

Standard deviation

PCs

Principal components

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare that they have no competing interests.

Data availability

All data including data sources, as well as online website, are freely available at http://bmdb.jflab.ac.cn:18083/app/bmdb and there is no login requirement. The detail information of all datasets are listed in Supplementary Table 1.

Fundings

National Key R&D Program of China [2022YFA1103300]; National Natural Science Foundation of China [82020108004]; Translational Research Grant of NCRCH [2020ZKZC02]; Youth Talent Development Program from Second Affiliated Hospital, Army Medical University [2022YQB014].

Authors’ contributions

ZW and XZ designed the project. JC, HY, CB, YH and KS developed BMDB. JC and CB and KS wrote the manuscript and organized all figures. JC and HY designed and developed the project website, and edited the online tutorial. All authors discussed results and commented on the manuscript.

Acknowledgements

Not applicable.

Morrison SJ, Scadden DT. The bone marrow niche for haematopoietic stem cells. Nature. 2014;505(7483):327–34.
Reagan MR, Rosen CJ. Navigating the bone marrow niche: translational insights and cancer-driven dysfunction. Nat Rev Rheumatol. 2016;12(3):154–68.
Schofield R. The relationship between the spleen colony-forming cell and the haemopoietic stem cell. Blood cells. 1978;4(1–2):7–25.
Nakamura-Ishizu A, MacArthur BD, Suda T. Beginning of a new era: mapping the bone marrow niche. Cell. 2019;177(7):1679–81.
Méndez-Ferrer S, Bonnet D, Steensma DP, Hasserjian RP, Ghobrial IM, Gribben JG, et al. Bone marrow niches in haematological malignancies. Nat Rev Cancer. 2020;20(5):285–98.
Oh I-H, Kwon K-R. Concise review: multiple niches for hematopoietic stem cell regulations. Stem Cells. 2010;28(7):1243–9.
Fujino T, Asada S, Goyama S, Kitamura T. Mechanisms involved in hematopoietic stem cell aging. Cell Mol Life Sci. 2022;79(9):473.
Kunisaki Y, Bruns I, Scheiermann C, Ahmed J, Pinho S, Zhang D, et al. Arteriolar niches maintain haematopoietic stem cell quiescence. Nature. 2013;502(7473):637–43.
He N, Zhang L, Cui J, Li Z. Bone marrow vascular niche: home for hematopoietic stem cells. Bone marrow research. 2014;2014.
Stucky A, Gao L, Sun L, Li SC, Chen X, Park TH, et al. Evidence for AJUBA-catenin-CDH4-linked differentiation resistance of mesenchymal stem cells implies tumorigenesis and progression of head and neck squamous cell carcinoma: a single-cell transcriptome approach. Blood genomics. 2021;5(1):29.
Pinho S, Frenette PS. Haematopoietic stem cell activity and interactions with the niche. Nat Rev Mol Cell Biol. 2019;20(5):303–20.
Zhao X, Zhang C, Cui X, Liang Y. Interactions of hematopoietic stem cells with bone marrow niche. Stem Cell Renew Cell-Cell Communication: Methods Protocols. 2021:21–34.
Zhang X. Beyond the horizon: the newly found sinner disturbing mesenchymal stromal niche. Blood Sci. 2022;4(03):179–80.
Hedlund E, Deng Q. Single-cell RNA sequencing: technical advancements and biological applications. Mol Aspects Med. 2018;59:36–46.
Liu S-Q, Gao Z-J, Wu J, Zheng H-M, Li B, Sun S, et al. Single-cell and spatially resolved analysis uncovers cell heterogeneity of breast cancer. J Hematol Oncol. 2022;15(1):19.
Zhang X, Hou Z, Huang D, Wang F, Gao B, Zhang C, et al. Single-cell heterogeneity and dynamic evolution of Ph-like acute lymphoblastic leukemia patient with novel TPR-PDGFRB fusion gene. Experimental Hematol Oncol. 2023;12(1):19.
Wang Y, Zhang X, Wang Z. Cellular barcoding: From developmental tracing to anti-tumor drug discovery. Cancer Lett. 2023:216281.
Potter SS. Single-cell RNA sequencing for the study of development, physiology and disease. Nat Rev Nephrol. 2018;14(8):479–92.
Wen L, Li G, Huang T, Geng W, Pei H, Yang J et al. Single-cell technologies: From research to application. Innov. 2022;3(6).
Davis-Marcisak EF, Deshpande A, Stein-O’Brien GL, Ho WJ, Laheru D, Jaffee EM, et al. From bench to bedside: Single-cell analysis for cancer immunotherapy. Cancer Cell. 2021;39(8):1062–80.
Huang D, Ma N, Li X, Gou Y, Duan Y, Liu B, et al. Advances in single-cell RNA sequencing and its applications in cancer research. J Hematol Oncol. 2023;16(1):98.
Zeng J, Zhang Y, Shang Y, Mai J, Shi S, Lu M, et al. CancerSCEM: a database of single-cell expression map across various human cancers. Nucleic Acids Res. 2022;50(D1):D1147–D55.
Sun D, Wang J, Han Y, Dong X, Ge J, Zheng R, et al. TISCH: a comprehensive web resource enabling interactive single-cell transcriptome visualization of tumor microenvironment. Nucleic Acids Res. 2021;49(D1):D1420–D30.
Acosta J, Ssozi D, van Galen P. Single-cell RNA sequencing to disentangle the blood system. Arteriosclerosis, thrombosis, and vascular biology. 2021;41(3):1012–8.
Huang X-t, Li X, Qin P-z, Zhu Y, Xu S-n, Chen J. Technical advances in single-cell RNA sequencing and applications in normal and malignant hematopoiesis. Front Oncol. 2018;8:582.
Liu J, Jiang P, Lu Z, Yu Z, Qian P. Decoding leukemia at the single-cell level: clonal architecture, classification, microenvironment, and drug resistance. Experimental Hematol Oncol. 2024;13(12).
Campillo-Marcos I, Alvarez-Errico D, Alandes RA, Mereu E, Esteller M. Single-cell technologies and analyses in hematopoiesis and hematological malignancies. Exp Hematol. 2021;98:1–13.
Jacobsen SEW, Nerlov C. Haematopoiesis in the era of advanced single-cell technologies. Nat Cell Biol. 2019;21(1):2–8.
Baryawno N, Przybylski D, Kowalczyk MS, Kfoury Y, Severe N, Gustafsson K, et al. A cellular taxonomy of the bone marrow stroma in homeostasis and leukemia. Cell. 2019;177(7):1915–32. e16.
Sivaraj KK, Jeong H-W, Dharmalingam B, Zeuschner D, Adams S, Potente M et al. Regional specialization and fate specification of bone stromal cells in skeletal development. Cell Rep. 2021;36(2).
Tikhonova AN, Dolgalev I, Hu H, Sivaraj KK, Hoxha E, Cuesta-Domínguez Á, et al. The bone marrow microenvironment at single-cell resolution. Nature. 2019;569(7755):222–8.
Zheng Z, He H, Tang XT, Zhang H, Gou F, Yang H, et al. Uncovering the emergence of HSCs in the human fetal bone marrow by single-cell RNA-seq analysis. Cell Stem Cell. 2022;29(11):1562–79. e7.
Rasheed A, Arteriosclerosis. Thromb Vascular Biology. 2022;42(6):691–9.
Nemeth MJ, Bodine DM. Regulation of hematopoiesis and the hematopoietic stem cell niche by Wnt signaling pathways. Cell Res. 2007;17(9):746–58.
Crane GM, Jeffery E, Morrison SJ. Adult haematopoietic stem cell niches. Nat Rev Immunol. 2017;17(9):573–90.
Yamazaki S, Ema H, Karlsson G, Yamaguchi T, Miyoshi H, Shioda S, et al. Nonmyelinating Schwann cells maintain hematopoietic stem cell hibernation in the bone marrow niche. Cell. 2011;147(5):1146–58.
Winkler IG, Sims NA, Pettit AR, Barbier V, Nowlan B, Helwani F, et al. Bone marrow macrophages maintain hematopoietic stem cell (HSC) niches and their depletion mobilizes HSCs. Blood J Am Soc Hematol. 2010;116(23):4815–28.
Lévesque J, Helwani F, Winkler I. The endosteal ‘osteoblastic’niche and its role in hematopoietic stem cell homing and mobilization. Leukemia. 2010;24(12):1979–92.
Oetjen KA, Lindblad KE, Goswami M, Gui G, Dagur PK, Lai C et al. Human bone marrow assessment by single-cell RNA sequencing, mass cytometry, and flow cytometry. JCI insight. 2018;3(23).
Baccin C, Al-Sabah J, Velten L, Helbling PM, Grünschläger F, Hernández-Malmierca P, et al. Combined single-cell and spatial transcriptomics reveal the molecular, cellular and spatial bone marrow niche organization. Nat Cell Biol. 2020;22(1):38–48.
Dolgalev I, Tikhonova AN. Connecting the dots: resolving the bone marrow niche heterogeneity. Front cell Dev biology. 2021;9:622519.
Stumpf PS, Du X, Imanishi H, Kunisaki Y, Semba Y, Noble T, et al. Transfer learning efficiently maps bone marrow cell types from mouse to human using single-cell RNA sequencing. Commun biology. 2020;3(1):736.
Ayhan S, Nemutlu E, Uçkan Çetinkaya D, Kır S, Özgül RK. Characterization of human bone marrow niches with metabolome and transcriptome profiling. J Cell Sci. 2021;134(6):jcs250720.
Abarrategi A, Mian SA, Passaro D, Rouault-Pierre K, Grey W, Bonnet D. Modeling the human bone marrow niche in mice: From host bone marrow engraftment to bioengineering approaches. J Exp Med. 2018;215(3):729–43.
Nelson MR, Ghoshal D, Mejías JC, Rubio DF, Keith E, Roy K. A multi-niche microvascularized human bone marrow (hBM) on-a-chip elucidates key roles of the endosteal niche in hBM physiology. Biomaterials. 2021;270:120683.
Kristensen HB, Andersen TL, Patriarca A, Kallenbach K, MacDonald B, Sikjaer T, et al. Human hematopoietic microenvironments. PLoS ONE. 2021;16(4):e0250081.
Qiu X, Mao Q, Tang Y, Wang L, Chawla R, Pliner HA, et al. Reversed graph embedding resolves complex single-cell trajectories. Nat Methods. 2017;14(10):979–82.
Street K, Risso D, Fletcher RB, Das D, Ngai J, Yosef N, et al. Slingshot: cell lineage and pseudotime inference for single-cell transcriptomics. BMC Genomics. 2018;19:1–16.
Jin S, Guerrero-Juarez CF, Zhang L, Chang I, Ramos R, Kuan C-H, et al. Inference and analysis of cell-cell communication using CellChat. Nat Commun. 2021;12(1):1088.

No competing interests reported.

Download PDF

Version 1

posted

You are reading this latest preprint version

BMDB: a comprehensive database and web server for integrated single-cell bone marrow microenvironment transcriptomic data

Status:

Version 1

Abstract

Background

Methods

Results

Conclusion

Figures

1. Background

2. Results

2.1. Contents of BMDB

2.2. Features and utilities

3. Conclusions

4. Methods

4.1. Data collection

4.2. Data processing pipelines

4.3. Advanced data processing

4.4. Construction of single-cell BM niche reference atlas

Abbreviations

Declarations

References

Additional Declarations

Supplementary Files

Status:

Version 1