Integration of mass cytometry and mass spectrometry imaging for spatially resolved single-cell metabolic profiling

Abstract

The integration of spatial omics technologies can provide important insights into the biology of tissues. Here we combined mass spectrometry imaging-based metabolomics and imaging mass cytometry-based immunophenotyping on a single tissue section to reveal metabolic heterogeneity at single-cell resolution within tissues and its association with specific cell populations such as cancer cells or immune cells. This approach has the potential to greatly increase our understanding of tissue-level interplay between metabolic processes and their cellular components.

Fig. 1. Identification of single-cell metabolic profiles in CRC.

a, Integrated workflow of consecutive MALDI-MSI and IMC analyses. Fresh-frozen tissue sections were cut (1) and treated with MALDI matrix (2). MALDI-TOF-MSI was performed to obtain spatial metabolomics data (3), followed by matrix removal (4). Sections were then labeled with metal-conjugated antibodies, and IMC was used to analyze the areas previously imaged by MALDI-TOF-MSI (5). Both datasets were preprocessed (6) and coregistered using visual landmarks (7). Cell segmentation and phenotype identification were performed (8), and MALDI-MSI-derived metabolic abundances were assigned to each cell (9), enabling downstream analysis (10). Created with BioRender.com. b, Scaled metabolite abundance profiles across cell populations identified in CRC1. Hierarchical clustering was guided by glycerophospholipid abundances (heatmaps of CRC2 and CRC3 are shown in Extended Data Fig. 1e). c,d, UMAP embedding of all cells (c) or immune cells (d) of CRC1, clustered by glycerophospholipids. Phenotypes of labeled cells are visualized on the UMAP embedding. e, Top differentiating glycerophospholipid features between cancer cells and the stromal–immune cell compartment, calculated for all images in the dataset. f, Left: representative IMC image of CRC1 highlighting cancer cells and stromal cells (keratin in red, vimentin in green and DNA in blue). Middle and right: MALDI-MSI image of the same region of interest showing differentially abundant metabolites in the stromal–immune cell compartment (PC(37:5)) and in cancer cells (PI(34:1)) displayed using viridis colors with saturated pixels above the 99th percentile (n = 2). Source data

Fig. 2. Glycerophospholipid landscape of myeloid cells.

a, Differential glycerophospholipid features distinguishing CD204⁺ macrophages from other cells in the stroma, calculated for all images in the dataset. Differentially abundant metabolites were identified by Wilcoxon test, and features with a FDR-adjusted P value below 0.05 are shown in black. b, Violin plots showing the relative abundance per cell of the two most differentially abundant metabolites in CD204⁺ macrophages. Each dot represents an image. c, UMAP embedding of macrophages/monocytes in CRC1, utilizing glycerophospholipids as features. Cells are labeled by IMC phenotypes. d, k-Means clusters of macrophages/monocytes based on glycerophospholipids, visualized on the UMAP embedding of c. e, The relative abundance of glycerophospholipids in the k-means clusters, visualized in d (enlarged heatmap with metabolite annotation is available in Extended Data Fig. 3). Source data

Extended Data Fig. 1. Integration of MALDI-MSI metabolite data with cellular phenotypes.

a. Approach for aligning MALDI-MSI and IMC data using visual landmarks present in both datasets. b. Schematic illustrating the approach for calculating metabolite abundance in each cell as pixel sizes differ between MALDI-MSI and IMC. Five 5 × 5 µm MALDI-MSI pixels are shown, each containing 25 1 × 1 µm IMC pixels. Cells within the pixels are colored with a red border, and a cell can span multiple MALDI-MSI pixels. To calculate metabolite abundance per cell, the peak intensity of the overlapping 5 × 5 µm MALDI-MSI pixel was assigned to each 1 × 1 µm IMC-pixel. The IMC cell segmentation mask (red borders) and the assigned peak intensities of each 1 × 1 µm were combined, and the relative metabolite abundance per cell was calculated using the mean of all pixels within a cell. c. Relative marker expression and abundance per image of cellular phenotypes determined by IMC. d. distribution of the number of cells contained in a single MSI pixel for each image. e. Glycerophospholipid profiles across distinct cellular phenotypes, in CRC2 and CRC3. Hierarchical clustering was guided by glycerophospholipid abundances. Source data

Extended Data Fig. 2. Glycerophospholipid heterogeneity within cell populations.

a. Density of cancer cell phenotypes and immune/stromal phenotypes corresponding to the UMAP of Fig. 1c. b. Pairwise distances between cancer cells, cancer cells and the stromal/immune compartment or the stromal/immune compartment only, visualized in a boxplot with median distance, interquartile range and min-max whiskers. n = 9066 cancer cells and 4889 stromal/immune cells. c. k-means clustering of all immune cells in CRC1 based on glycerophospholipid features, visualized on the UMAP embedding shown in Fig. 1d. d. Confusion matrix comparing IMC phenotypes with k-means clusters to determine clustering by cell type or glycerophospholipid features. The heatmap indicates the number of cells that overlap between the two clustering methods. e. Density of CD204⁺ macrophage phenotypes corresponding to the UMAP of Fig. 1d. Source data

Extended Data Fig. 3. Glycerophospholipid abundance in myeloid cell k-means clusters.

Enlarged heatmap of Fig. 2e: Relative abundance of glycerophospholipids in the k-means clusters, visualized in Fig. 2d. Source data