Pan-Organ Spatial-omics Insights Built on First Unified Single-cell Spatial-omics Atlas

Introduction: : Understanding how cells assemble into functional tissues and organs is a central challenge in biology. Spatial-omics technologies, such as immunohistochemistry, immunofluorescence, and in-situ hybridization, now enable the mapping of cellular organization across multiple spatial scales. These methods are particularly powerful for studying immune responses and disease progression, where the microenvironment critically influences cellular behavior. Recent studies have shown spatial patterns of cell interaction are conserved across tissue types and even species, underscoring the need for large-scale spatial analysis.

Efforts like the NIH's Human Biomolecular Atlas Program and the NCI's Human Tumor Atlas Network aim to construct tissue atlases using spatial-omics, yet limitations in data uniformity and accessibility hinder cross-study comparisons. Most datasets lack standardized formats, complete segmentation, or consistent annotations, posing challenges for broader integrative analyses.

Having a fully standardized pan-organ spatial atlas will enable comparative studies of cellular organization, revealing conserved and context-specific interaction patterns in both healthy and diseased tissues. Therefore, we propose a framework for aggregating spatial-omics datasets across experimental platforms. Our approach standardizes data processing, consistent cell type annotation, and integration across metadata information. This resource will serve as a critical foundation for advancing spatial biology and precision medicine by offering a comprehensive pipeline and database for multi-cellular spatial investigations.

Materials and

Methods: : To construct a comprehensive spatial-omics atlas, we systematically curated 58 multiplexed imaging datasets from human tissues. An automated literature search using PubMed and Google Scholar with human examinations identified 60 studies published in the past five years with single-cell resolution data. Inclusion criteria required x/y spatial coordinates, cell type annotations, and imaging region identifiers. Associated metadata such as age, sex, tissue type, and disease status were collected to support integrative analyses (Fig.1a).

Cell type labels were harmonized using a hierarchical merging process to map 487 unique labels into 13 broader categories, enabling multiscale resolution and cross-study comparability (Fig.1b-d). Studies were categorized into three types: (1) single-cell with annotated cell types, (2) single-cell without annotations, and (3) imaging-only. Only datasets in category (1) were included. Public datasets were directly downloaded; others were acquired by contacting authors (Fig.1c).

To characterize tissue microenvironments, we computed 20-cell nearest-neighbor windows for each cell and performed unsupervised clustering using enrichment-guided optimization to define spatial neighborhoods (Fig. 2a). 18 neighborhood clusters were annotated based on cell type compositions and spatial topologies (Fig. 2c). In addition, follicular microstructures were segmented using HDBSCAN-based patch analysis on tissue regions exceeding 500 cells (Fig. 3b). These standardized pipelines facilitate high-resolution characterization of multicellular architecture across diverse human tissues.

The PanOrgan atlas provides a scalable platform for spatial-omics data integration and analysis. A cloud infrastructure is under development to support ongoing dataset curation and public contributions.

Results, Conclusions, and Discussions:: The PanOrgan atlas integrates 55 single-cell spatial-omics datasets encompassing over 147 million cells, 5,598 imaging regions, and 1,226 donors (Fig.1e-h). Across studies, 487 unique cell types were standardized into 13 categories via hierarchical graph mapping to unify granularity (Fig.1i). This enabled robust inter-tissue comparisons and downstream spatial analyses.

Neighborhood clustering identified 18 biologically meaningful multicellular neighborhoods(Fig. 2b-d). We observed differential neighborhood abundances between both tissues and disease states, as shown with just one example comparing all intestine samples (Fig. 2e). Another example are immune follicle neighborhoods (Fig. 2f) that were differentially abundant across organs (Fig. 2g) and increased in males compared to females (Fig. 2h). These differences highlight organ-, disease-, metadata-specific multicellular organizations described for the first time.

Further analysis using patch-level segmentation of follicular regions uncovered diverse compositional patterns across disease states and tissues (Fig. 3a, b). As expected, lymphoid tissues had high levels of follicles (Fig. 3c, d) and B cell density (Fig. 3e). In contrast, cancerous breast and lung tissues, as well as inflamed gastrointestinal regions, showed increased epithelial cell infiltration within follicles and a concurrent reduction in B and CD4+ T cells (Fig. 3f). These trends are quantitatively supported by comparisons of B cell percentages across normal, inflamed, and cancerous tissues, where cancer samples exhibited significantly lower B cell densities than inflammation and normal groups (Fig. 3e, f). These findings suggest possible follicular disruption or invasion in disease contexts, aligning with known mechanisms of immune evasion and tumor progression.

By combining spatial neighborhoods, cell type distributions, and metadata, PanOrgan atlas enables rich characterization of tissue microenvironments across health and disease. Future work will continue to explore atlas comparisons, expand the atlas with new datasets, and support public engagement through a cloud-based platform. This standardized resource establishes a foundation for comparative spatial-omics studies, offering novel insights into tissue organization, immune architecture, and disease pathology.

Acknowledgements and/or References (Optional)::