Mapping Cell Atlases at the Single‐Cell Level

“

TL;DR

This review summarizes the rapid progress in constructing single-cell atlases across species, highlighting how high-throughput transcriptomic, genomic, proteomic, and spatial methods have enabled comprehensive mapping of cellular diversity. Key findings include the creation of whole-organism atlases for humans, mice, and other model organisms, which have revealed previously unknown cell types and states. The clinical significance lies in the potential of these atlases to improve diagnostics, prognostics, and personalized medicine by comparing healthy and diseased tissues at single-cell resolution.

Full summary

4,495 CHARS

**Background** Cells are the fundamental structural and functional units of animals, with diverse types and functions across tissues and organs. Recent advances in single-cell technologies have enabled the construction of comprehensive cell atlases that provide detailed information about every cell type in different organisms. Global projects such as the Mouse Cell Atlas and Human Cell Atlas have greatly expanded knowledge of cell biology by creating whole-organism cell atlases across various species. These atlases offer an integrative view of biological systems, enabling researchers to observe cell type hierarchy from a global view and facilitating the study of cell diversity, genetic networks, tissue organization, and disease processes.

**Methods** The review discusses various high-throughput methods for cell atlas mapping, including transcriptomic, genomic, proteomic, and spatial methods. Transcriptomic methods involve three major strategies: droplets (e.g., Drop-seq, inDrop, 10× Genomics), microwells (e.g., Microwell-seq, Seq-well), and split-pool barcoding (e.g., sci-RNA-seq, SPLiT-seq). Genomic methods include single-cell epigenomics sequencing such as scATAC-seq and snM-seq. Proteomic methods include sequencing-based approaches like CITE-seq and REAP-seq, as well as mass spectrometry-based methods like CyTOF. Spatial methods include sequencing-based approaches (e.g., spatial transcriptomics, HDST, Slide-seq, Seq-Scope, Stereo-seq) and imaging-based methods (e.g., MERFISH, seqFISH, STARmap). The review also covers the integration of machine learning techniques, including deep learning models for analyzing single-cell data.

**Key Results** The review presents an overview of cell atlases across multiple organisms. For humans, the Human Cell Atlas (HCA) has profiled more than 50 million cells from over 30 types of human organs. The human cell landscape (HCL) included more than 100 major cell clusters and over 800 defined cell subtypes. The Tabula Sapiens consortium constructed a multiple-organ atlas with 475 distinct cell types. At the chromatin accessibility level, Zhang et al. reported the first scATAC-seq atlas of over 1 million nuclei in 222 distinct cell types across 30 adult human tissues, providing ≈1.2 million candidate cis-regulatory elements (cCREs). For mice, the Mouse Cell Atlas (MCA) includes nearly half a million cells from more than 40 organs and tissues. The Tabula Muris Senis highlighted aging hallmarks such as mutational burden, genomic instability, and immune system gene expression phenotypes. The review also covers cell atlases of other organisms, including monkeys, zebrafish, amphibians, livestock, flies, worms, and marine organisms. In disease biology, single-cell atlases have been established for eight major systems, including urinary, respiratory, digestive, circulatory, and neurological disorders. For example, integrated healthy kidney atlases revealed a subpopulation of proximal tubules expressing VCAM1 (PT_VCAM1) as an injured cell state. The integrated human lung cell atlas (HLCA) combined 49 datasets and revealed altered cell states in lung disease. Deep learning models such as Geneformer, pretrained on 30 million single-cell transcriptomes, and Nvwa, which predicts single-cell gene expression across species, have been developed to analyze cell atlas data.

**Clinical Implications** Single-cell atlases have profound implications for clinical practice. They offer the potential for improved diagnostics, prognostics, and personalized medicine by enabling comparison of healthy and diseased tissues at the single-cell level. Researchers can identify specific cell types or states associated with diseases, aiding in early detection and targeted interventions. For example, the identification of PT_VCAM1 as an injured cell state in kidney disease and the mapping of disease-associated cell states in lung disease provide insights into disease mechanisms. The integration of GWAS data with cell atlases facilitates the identification of key genetic factors and regulatory regions contributing to disease development. Deep learning models can predict the regulatory effects of DNA variants and map genetic variants to cell-type levels, potentially identifying driver cell types and functional mechanisms of trait-causal and disease-causal genetic variations. The synergy among single-cell omics technology, cell atlases, and AI has great potential for advancing genomics research and predictive biology.