A spatially resolved multi-omics atlas of human intestinal metabolism

Introduction: : Metabolism orchestrates organ function and is shaped by unique, organ-specific contexts. In the intestine, metabolism maintains local homeostasis and regulates whole-body metabolic state, yet how metabolism varies across intestinal regions and individual cell types remains poorly understood. This gap results from a lack of integrative data linking metabolism across molecular, cellular, and tissue scales. To address this, we combined three complementary technologies—multiplexed spatial protein imaging (CODEX), spatial metabolomics (MALDI-IMS), and single-cell RNA sequencing (scRNA-seq)—to generate a comprehensive, spatially resolved metabolic atlas of the healthy human intestine (Fig. 1a). This integration leverages the distinct strengths of each method, simultaneously capturing spatial organization of cellular composition, detailed molecular profiles at single-cell resolution, and metabolic landscape information, thereby providing unprecedented resolution of intestinal metabolic processes. Our integrative atlas revealed distinct metabolic signatures among cell types, cellular neighborhoods, and across intestinal regions from the small intestine to the colon, uncovering specialized lipid metabolic pathways at interfaces and regional variations in nutrient processing. This analytical framework offers a robust template for dissecting multiscale metabolic programs within organs, highlights specific intestinal targets for metabolic modulation, and provides a healthy reference for understanding metabolic dysregulation in disease.

Materials and

Methods: : Fresh-frozen human intestinal tissue from nine donors was embedded in optimal cutting temperature compound, cryosectioned at 7 µm thickness, and stained for CODEX multiplexed imaging. Images were stitched, corrected, and aligned using Akoya software. Single-cell segmentation was performed using the Mesmer deep learning model in SPACEc, with DAPI, CD45, and CD90 signals. Cell clustering and annotation were performed using Leiden clustering, K-means, and manual curation based on marker enrichment and localization.
Cells were grouped into neighborhoods, communities, and tissue units based on proximity and cell composition. 2D boundary masks for each level were generated using parametric grouping of boundary cells and refined using Elliptic Fourier Analysis (EFA) of patch shapes. These masks were projected onto adjacent MALDI imaging mass spectrometry (MALDI-IMS) datasets, which were acquired from sequential tissue sections after matrix application. Spectra were collected at 5 µm resolution, allowing alignment of spatial ion distributions to defined tissue structures.
Surface lipidomics data were collected using LC-MC workflows from parallel tissue sections. Lipid features were extracted, identified, and analyzed across neighborhoods, communities, and tissue units using MaxQuant and R-based pipelines.
Single-cell RNA sequencing (scRNA-seq) data were analyzed by selecting metabolism-related genes from curated public databases. Seurat was used for unsupervised clustering, differential gene analysis (Wilcoxon, DESeq2), and pathway enrichment. Machine learning models (XGBoost, random forest, elastic net) were trained to classify aggregated cell groups and intestinal regions, and identify characteristic metabolic genes. Integration of scRNA-seq with CODEX spatial maps is underway using MaxFuse to align transcriptional states with imaging data.

Results, Conclusions, and Discussions:: With CODEX multiplexed imaging, we identified 36 cell types (Fig. 1b), 23 neighborhoods (Fig. 1c), 11 communities (Fig. 1d), and 3 tissue units (Fig. 1e) across biological scales of the intestine. We observed biologically consistent region-specific differences in intestinal cell type composition (Fig. 1f-h), such as CD8⁺ T cells more abundant in the small intestine (Fig. 1i–l). Because this is the first time that CODEX multiplexed data is co-registered with spatial-metabolic data, we segmented tissues into discrete patches to extract aligned metabolites (Fig. 1n-p) consistent across scales (Fig. 1q–t).
Aligned MALDI IMS (Fig. 2a-d) revealed metabolically distinct clusters of ion intensity profiles (Fig. 2e, f). Cluster 7, for example, contained m/z 446.3771 and 746.3272, enriched in both the small intestine and colon (Fig. 2h), while other clusters such as Cluster 3 showed region-specific feature abundance. Differential metabolite expression across anatomical regions (Fig. 2j), plasma cell-enriched neighborhoods (Fig. 2k), and innate immune-enriched communities (Fig. 2l), revealed distinct spatially restricted lipid zones.
Human intestine scRNA-seq data (Fig. 3a) using curated metabolic gene sets (Fig. 3b) revealed clear transcriptomic separation between small and large intestine (Fig. 3c-f), as well as among eight finer anatomical regions and major cell types (Fig. 3g). We identified signature metabolic markers, including CYP3A5 and GDA in enterocytes, and ABCA1, PIP4K2A, and TBXAS1 in myeloid cells (Fig. 3i, k, l), suggesting targets for cell- and disease-state modulation. Subclustering of epithelial cells revealed metabolically distinct states (Fig. 3j).
For the first time, we show differences in metabolism between intestinal regions were observed across multiple spatial scales—cell type, neighborhood, community, and tissue-level organization—within CODEX, MALDI-IMS, and scRNA-seq that identified conserved spatial patterns and patch clusters across distinct anatomical areas. These findings suggest that intestinal tissue metabolism is governed by both localized specialization (region and cell type) and underlying organizational templates with implications for intestinal health and disease. Our novel, multimodality omics framework can be applied across tissues and diseases serving to understand cell and multicellular tissue-based metabolism. In conclusion, this framework offers a scalable approach to map molecular function back into tissue context, enabling spatially-informed therapeutic strategies.

Acknowledgements and/or References (Optional):: John Hickey and Angela Kruse acknowledge funding from the HuBMAP Jumpstart grant.