Single-cell spatial metabolomics with cell-type specific protein profiling for tissue systems biology

Abstract

Metabolic reprogramming in cancer and immune cells occurs to support their increasing energy needs in biological tissues. Here we propose Single Cell Spatially resolved Metabolic (scSpaMet) framework for joint protein-metabolite profiling of single immune and cancer cells in male human tissues by incorporating untargeted spatial metabolomics and targeted multiplexed protein imaging in a single pipeline. We utilized the scSpaMet to profile cell types and spatial metabolomic maps of 19507, 31156, and 8215 single cells in human lung cancer, tonsil, and endometrium tissues, respectively. The scSpaMet analysis revealed cell type-dependent metabolite profiles and local metabolite competition of neighboring single cells in human tissues. Deep learning-based joint embedding revealed unique metabolite states within cell types. Trajectory inference showed metabolic patterns along cell differentiation paths. Here we show scSpaMet's ability to quantify and visualize the cell-type specific and spatially resolved metabolic-protein mapping as an emerging tool for systems-level understanding of tissue biology.

Fig. 1. The scSpaMet pipeline for integrated metabolite and protein profiling at the single-cell resolution.

a Overview of scSpaMet. Tissue samples on glass slides are labeled with metal-isotope conjugated antibodies followed by metabolic profiling with 3D-SMF and finally proteomic profiling using IMC. Created with Biorender.com. b Examples of scSpaMet generated data in lung cancer tissues. Left to right: PO3- channel ion, multiplexed metabolic data processed by pixel clustering, IMC imaging Histone H3 marker, multiplexed IMC data overlaid with pseudo-coloring, virtual reconstructed H&E staining from IMC multiplexed proteomic data. n = 7 biologically independent samples on 21 FOVs. Scale bar 100 μm. c Examples of scSpaMet generated data in tonsil tissues. Left to right: PO3- channel ion, multiplexed metabolic data analyzed by pixel clustering, IMC imaging of Intercalator marker for DNA, multiplexed IMC proteomic data overlaid with pseudo-coloring, virtual reconstructed H&E staining from IMC multiplexed data. n = 2 biologically independent samples on 11 FOVs. Scale bar 100 μm.

Fig. 2. Single-cell cross-modality registration pipeline.

a Single-cell protein-metabolite bi-modal registration pipeline. Input images consist of 3D-SMF and IMC-generated images. First, a template matching algorithm is used to find the corresponding matching region between the 3D-SMF region (smaller) inside the IMC region (larger). Next, the rotation offset between the two aligned and cropped images is calculated. Finally, the affine transformation of the two images is calculated to obtain bi-modal matched images. b Registration result from the comparison between random, rotation registration, and affine registration. Left: Examples of the three registration methods between IMC (gray) and 3D-SMF (red) images and their corresponding inset. Scale bar 100 μm. c Comparison of Structural/structure Similarity (SSIM) and Normalized Root Mean Square Error (NRMSE) between the three registration methods (n = 24 FOVs). Mann–Whitney-Wilcoxon test two-sided with Bonferroni correction (ns: 0.05 p, *: 0.01 p < = 0.05, **: 0.001 p < = 0.01, ***: 0.0001 < p < = 0.001, ****: p < = 0.0001). All box plots with center lines showing the medians, boxes indicating the interquartile range, and whiskers indicating a maximum of 1.5 times the interquartile range beyond the box.

Fig. 3. Metabolomic and proteomic modalities need to be integrated with data analysis.

a Overview of metabolomic and proteomic data generated by scSpaMet. Tissue samples on glass slides are labeled with metal-isotope conjugated antibodies followed by metabolic profiling with TOF-SIMS and sequential proteomic profiling using IMC. The resulting outputs are $n$ cells with distinct $p_m$ metabolomic feature and $p_p$ proteomic features. The modalities have distinct feature spaces due to differences in feature number and variability across datasets. Created with Biorender.com. b Examples of existing spatial-omics data integration pipelines show (i) same-modality integration, (ii) cross-modality integration with shared features, and (iii) cross-modality multi-modal on the same cell. (i) is applied to remove the batch effect from both metabolite and protein datasets. Our scSpaMet data lacked any shared features (markers) or cell types. Created with Biorender.com.

Fig. 4. The scSpaMet pipeline identifies metabolite differences between tumor and stromal regions in human lung cancer tissues.

a Unsupervised single cell clusters from protein profiles in human lung cancer tissues. Created with Biorender.com. b Spatial projection of corresponding cell clusters from a. c Left: definition of stroma and tumor region in tissue sample based on single cell phenotypes from (a). Right: The most expressed metabolite channels in stroma and tumor region from the differential analysis. d Comparison of single cell metabolite expression level for identified mass channels. Left: The metabolite channels related to Glucose pathway and Cholesterol fragments. Middle: The metabolite channels related to amino acid fragments. Right: Bar graph of selected metabolite channels (n = 19507 cells). Mann-Whitney-Wilcoxon test was two-sided with Bonferroni correction (ns: 0.05 < p, ****: p < =0.0001). All box plots with center lines showing the medians, boxes indicating the interquartile range, and whiskers indicating a maximum of 1.5 times the interquartile range beyond the box. e Metabolite channels related to identified lipid fragments. f Spatial projection of single cell metabolite expression level for selected metabolite channels.

Fig. 5. The scSpaMet pipeline quantifies local metabolite competition in lung cancer as a function of distance to the endothelial cells.

a Representative schematic showing the definition of distance to CD31+ cells in lung cancer tissues. Created with Biorender.com. b Pearson correlation of metabolic signals compared to single cell distance to CD31+ cells. Left: Heatmap showing Pearson correlation of selected metabolite channels compared to the distance to CD31+ cells. Right: Single cell spatial distance map to CD31+ cells in lung cancer tissues. c Local metabolite competition as a distance of CD31+ cells between T-cells and tumor cells. Left: Selected metabolite channels showing positive and negative correlation of tumor cells (top) and T-cells (bottom). Right: spatial projection of T-cells and tumor cells’ local metabolite competition for 48 m/z. d Local metabolite competition as a distance of CD31+ cells between CD68 positive cells and tumor cells. Left: Selected metabolite channels showing positive and negative correlation of tumor cells (top) and CD68 cells (bottom). Right: spatial projection of CD68 cells and tumor cells’ local metabolite competition for 74 m/z.

Fig. 6. The scSpaMet pipeline identifies spatial signatures of joint protein-metabolite signatures in lung cancer tissues across patients.

a Overview of spatial joint protein-metabolite signatures identification by the scSpaMet pipeline. Using VAE joint embedding, cells are clustered based on their joint protein-metabolite profiles. A neighborhood graph is constructed based on cell spatial location. The corresponding cell neighborhood cell type frequencies are used to determine spatial joint protein-metabolite signatures across patients. Created with Biorender.com. b Count of corresponding spatial joint protein-metabolite signatures across imaging regions in all the patients and their lung cancer tissues. c Frequency of corresponding spatial joint protein-metabolite signatures across patients in lung cancer tissues. d Spatial projection of cell corresponding spatial joint protein-metabolite signatures from (a).

Fig. 7. The scSpaMet pipeline identifies metabolite differences of B cell follicles regions in human tonsil tissues.

a Unsupervised single cell clusters from protein profiles in human tonsil tissues. Created with Biorender.com. b Spatial projection of corresponding cell clusters from a). c The top expressed metabolite channels in regions inside and outside germinal centers across tonsil tissues from the differential analysis. d Comparison of single cell metabolite expression levels for identified mass channels. Left: The metabolite channels related to Glucose pathway and Cholesterol fragments. Middle: The metabolite channels related to amino acid fragments. Right: Bar graph of selected metabolite channels (n = 31156 cells). Mann-Whitney-Wilcoxon test was two-sided with Bonferroni correction (****: p < =0.0001). All box plots with center lines showing the medians, boxes indicating the interquartile range, and whiskers indicating a maximum of 1.5 times the interquartile range beyond the box. e Metabolite channels related to identified lipid fragments. f Spatial projection of single-cell metabolite expression levels for selected metabolite channels.

Fig. 8. ScSpaMet pipeline quantifies cell type-specific local metabolite competition in germinal centers.

a Representative schematic showing the definition of local cell metabolite competition in human tonsil germinal center regions. Created with Biorender.com. b Local competition of metabolites between B cells and FDCs (top, n = 4371 cells), B cells and TFHs (middle, n = 2807 cells), and B cells in LZ with DZ (bottom, n = 1870 cells). Mann-Whitney-Wilcoxon test was two-sided with Bonferroni correction (ns: 0.05 <p, ****: p < =0.0001). All box plots with center lines showing the medians, boxes indicating the interquartile range, and whiskers indicating a maximum of 1.5 times the interquartile range beyond the box. c Comparison of selected metabolite channels between GC LZ and GZ DZ. Mann-Whitney-Wilcoxon test was two-sided with Bonferroni correction (ns: 0.05 < p, ****: p < =0.0001). All box plots with center lines showing the medians, boxes indicating the interquartile range, and whiskers indicating a maximum of 1.5 times the interquartile range beyond the box.

Fig. 9. The scSpaMet pipeline infers metabolites trajectory for B cell differentiation inside of germinal centers.

a Representative schematic showing the definition of germinal center B cell trajectories. Created with Biorender.com. b B cell trajectories in the germinal center from protein markers in scSpaMet. Left: Unsupervised clustering of B cell protein markers and corresponding TSNE plot. Middle: Pseudotime analysis of B cell protein phenotype. Right: Identified B cell trajectories from single-cell phenotype. (1) the GC DZ B-cells to GC LZ B-cells and (2) the GC DZ B-cells to the activated B-cells. c Identified germinal center DZ to LZ and DZ to Activated B-cell trajectory. Left: Variation of protein marker intensity along identified trajectories, right: Variation of selected metabolite channel along identified trajectories. d Spatial plot of germinal center DZ to LZ trajectory (left) and germinal center dark zone to Activated B cell trajectory (right).

Fig. 10. The scSpaMet pipeline characterizes proteins and metabolites in human endometrium samples.

a Unsupervised single cell clusters from protein profiles in human endometrium tissues. Created with Biorender.com. b Spatial projection of corresponding cell clusters from a. c The top expressed metabolite channels in identified cell types across endometrium tissues from the differential analysis. d Comparison of single cell metabolite expression levels for identified mass channels. Left: The metabolite channels related to Glucose pathway and Cholesterol fragments. Middle: The metabolite channels related to amino acid fragments. Right: Bar graph of selected metabolite channels (n = 8215 cells). Mann-Whitney-Wilcoxon test was two-sided with Bonferroni correction (****: p < =0.0001). All box plots with center lines showing the medians, boxes indicating the interquartile range, and whiskers indicating a maximum of 1.5 times the interquartile range beyond the box. e Metabolite channels related to identified lipid channels fragmentation. f Spatial projection of single-cell metabolite expression levels for selected metabolite channels in obese benign (top) and lean benign (bottom) tissues.