Re: Right Hemisphere Deep Paint 3d Build 2309 Full Version
Renzo Hayami
Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.
Single-cell transcriptomics by single-cell RNA sequencing (scRNA-seq) or single-nucleus RNA sequencing (snRNA-seq) provides unprecedented profiling depth and scalability, enabling comprehensive quantitative analysis and classification of cell types at scale2,3,7,8,9. Transcriptomically defined cell types have been shown to exhibit concordant morphological and physiological properties10,11. Single-cell transcriptomics has been used to categorize cell types from many different regions of the mouse nervous system and increasingly in human and non-human primate brains2,12. The BRAIN Initiative Cell Census Network (BICCN) and the Human Cell Atlas (HCA) are representative community efforts that use single-cell transcriptomics to create cell-type atlases for the brain and body of human and other mammals8,13,14,15,16.
An essential next step is to create a comprehensive and high-resolution transcriptomic cell-type atlas for the entire adult brain from a single mammalian species. The mouse (Mus musculus) is the most widely used mammalian model organism and is therefore a natural first choice for a comprehensive definition of mammalian brain composition and architecture. To define the anatomical context for cell types, another critical requirement is to characterize the precise spatial location of each cell type using single-cell-level spatial transcriptomics analysis17,18,19,20 covering the entire mouse brain. In addition to describing a complete, brain-wide cell-type atlas of a mammalian brain, this analysis will enable us to address questions on how the brain-wide transcriptomic landscape of cell types relates to the anatomical and circuit organization and its ontology rooted in development and evolution, and how coordinated gene expression specifies cell-type identity and functional properties.
As part of the BICCN, we set out to build a comprehensive, high-resolution transcriptomic cell-type atlas for the entire adult mouse brain. We systematically generated two types of large-scale, single-cell-resolution transcriptomic datasets for all mouse brain regions, using scRNA-seq and MERFISH21. We used the scRNA-seq data to generate a transcriptomic cell-type taxonomy, and the MERFISH data to visualize and annotate the spatial location of each cluster in this taxonomy, based on the Allen Mouse Brain Common Coordinate Framework version 3 (CCFv3)22 (Supplementary Table 1 provides the anatomical ontology with full names and acronyms of all brain regions).
By performing all pairwise cluster comparisons in this initial transcriptomic taxonomy, we derived 8,460 differentially expressed genes (DEGs) (Supplementary Table 5) differentiating all pairs of clusters. We then designed two gene panels for the generation of MERFISH data, with each gene panel containing a selected set of marker genes with the greatest combinatorial power to discriminate among all clusters. The first gene panel contained 1,147 genes and was used by the X.Z. laboratory to generate MERFISH datasets from several male and female mouse brains using a custom imaging platform24. The second gene panel contained 500 genes (Supplementary Table 6 and Methods) and was used to generate a MERFISH dataset from one male mouse brain at the Allen Institute for Brain Science (AIBS) using the Vizgen MERSCOPE platform (Extended Data Fig. 2). The AIBS MERFISH dataset contained 59 serial full coronal sections at 200-m intervals spanning the entire mouse brain, with a total of around 4.3 million segmented and QC-passed cells (Extended Data Fig. 2), subsequently registered to the Allen CCFv3 (Methods).
To hierarchically organize the transcriptomic cell-type taxonomy and delineate the relationship between clusters, we first computed Pearson correlations of gene expression between each pair of clusters using all or a subset of DEGs as a measure of similarity between clusters (Extended Data Fig. 3). We found that clusters have different degrees of similarities between them and can be grouped into smaller or larger categories. Furthermore, transcription factor marker genes provide the lowest correlation values across the brain compared with functional marker genes, adhesion molecules and all marker genes, and can best resolve the global relationships among clusters. Therefore, we used transcription factor marker genes to computationally build a cell-type hierarchy, grouping the clusters into putative classes, subclasses and supertypes (Methods).
We used the AIBS MERFISH dataset and one of the MERFISH datasets from the X.Z. laboratory to annotate the spatial location of each subclass, supertype and cluster. To do this, we developed a hierarchical mapping approach (Methods) to map each MERFISH cell to the transcriptomic taxonomy and assign the best matched cluster identity along with a correlation score to each MERFISH cell. The spatial location of each cluster was subsequently obtained by the collective locations of majority of the cells assigned to that cluster with high correlation scores. We annotated each subclass with its most representative anatomical regions and incorporated these annotations into subclass nomenclature for easier recognition of their identities. In this way, the high-level distribution of cell types across the entire mouse brain is described. As the anatomical annotations at subclass level are largely consistent between the X.Z. laboratory and the AIBS MERFISH datasets, the AIBS MERFISH dataset is used to illustrate our results and findings in the subsequent sections of this manuscript.
Thorough analysis revealed extraordinarily complex relationships among transcriptomic clusters and their associated regions. We further fine-tuned and adjusted class, subclass and supertype memberships of a small fraction of clusters to reach the final definition (Extended Data Fig. 4 and Methods). To organize the complex molecular relationships, we present a high-resolution transcriptomic and spatial cell-type atlas for the whole mouse brain with four nested levels of classification: 34 classes, 338 subclasses, 1,201 supertypes and 5,322 clusters or types (Fig. 1, Extended Data Fig. 5e and Extended Data Table 1). We also grouped the classes into seven neighbourhoods for more in-depth analyses of related subsets of cell types. The neighbourhoods recapitulate to a great extent the molecular and anatomical relatedness among cell types, but they are not part of the cell-type hierarchy because they do not strictly follow the distance relationship among cell types and they contain partially overlapping memberships.
To facilitate the wide dissemination of data and utilization of the comprehensive mouse whole-brain cell-type atlas, we have developed the Allen Brain Cell Atlas. This platform, accessible at -map.org/atlases-and-data/bkp/abc-atlas, is designed to visualize extensive scRNA-seq, snRNA-seq and MERFISH datasets, organized according to the whole-brain cell-type taxonomy, along with accompanying metadata. The Allen Brain Cell Atlas leverages a service-oriented architecture and is hosted on Amazon Web Services, ensuring efficient access and robust performance.
The Allen Brain Cell Atlas enables researchers to explore the landscape of cell types across various hierarchical levels and brain regions. Users can delve into specific cell types, examine their spatial distributions, study gene expression patterns, explore co-expression relationships, or investigate the composition of cell types within distinct brain regions. Additionally, the Allen Brain Cell Atlas provides valuable links to related resources, including an open source project repository for data download, complete with comprehensive documentation and a Jupyter Notebook that illustrates data retrieval and analysis techniques (available at _atlas_access/intro.html). To foster a supportive research community, we offer a dedicated community forum where users can find a user guide, seek assistance and exchange knowledge. This forum, which is monitored by members of the Allen Brain Cell Atlas team, can be accessed at -map.org/c/how-to/abc-atlas/19/l/top.
Furthermore, we have developed the MapMyCells tool ( -map.org/atlases-and-data/bkp/mapmycells), which enables researchers to upload and use our cell-type mapping solution based on the hierarchical mapping tools that we have developed ( ). This tool facilitates integrating and comparing their scRNA-seq and/or snRNA-seq data with the reference taxonomy of cell types in whole brain of mouse, including high-quality single-cell transcriptomes. By doing so, researchers can gain valuable insights into their data mapped against a reference and accelerate their investigations.
Neuronal cell types constitute a large proportion of the whole-brain cell-type atlas, including 6 neighbourhoods, 29 classes (85%), 315 subclasses (93%), 1,156 supertypes (96%) and 5,205 clusters (98%) (Extended Data Table 1 and Supplementary Table 7). Neuronal types have high regional specificity and exhibit highly variable degrees of similarities and differences. To further investigate the neuronal diversity within each major brain structure, we generated re-embedded UMAPs (in 2D and 3D) for the neighbourhoods of neuronal types described above, to reveal fine-grained relationships between neuronal types within and between brain regions in conjunction with the MERFISH data. The results shown in Fig. 2 reveal a marked correspondence between transcriptomic specificity and relatedness and spatial specificity and relatedness among the different neuronal subclasses.
We systematically assigned neurotransmitter identity to each cell cluster on the basis of the co-expression of canonical neurotransmitter transporter genes and synthesizing enzymes and considering alternative neurotransmitter release mechanisms (Figs. 1e and 3, Extended Data Figs. 5e and 9, Supplementary Table 7 and Methods).
7fc3f7cf58