henhaobio
Sign in

Nature MethodsMolecular BiologyPeer-reviewed journal article

Spatial single-cell profiling of intracellular metabolomes in situ

Luca RappeziD, Mira StadleriD, Sergio TrianaiD, Rose Muthoni GathunguiD, Katja Ovchinnikova, Prasad B. PhapaleiD, Mathias F HeikenwälderiD, Theodore AlexandroviD

PDFDOIPreprintCodeData
In-depth full-text digest available · Jump to digest ↓

Abstract

The recently unveiled extent of cellular heterogeneity demands for single-cell investigations of intracellular metabolomes to reveal their roles in intracellular processes, molecular microenvironment and cell-cell interactions. To address this, we developed SpaceM, a method for in situ spatial single-cell metabolomics of cell monolayers which detects >100 metabolites in >10000 individual cells together with fluorescence and morpho-spatial cellular features. We discovered that the intracellular metabolomes of co-cultured human HeLa cells and mouse NIH3T3 fibroblasts predict the cell type with 90.4% accuracy and revealed a short-distance metabolic intermixing between HeLa and NIH3T3. We characterized lipid classes composing lipid droplets in steatotic differentiated human hepatocytes, and discovered a preferential accumulation of long-chain phospholipids, a co-regulation of oleic and linoleic acids, and an association of phosphatidylinositol monophosphate with high cell-cell contact. SpaceM provides single-cell metabolic, phenotypic, and spatial information and enables spatio-molecular investigations of intracellular metabolomes in a variety of cellular models.

In-depth summary

Claude Opus 5.5 · xhigh effort · 8 Nov 2026 · from the full text of version 1 (2 Jan 2019) · CC BY 4.0

SpaceM integrates MALDI-imaging mass spectrometry with pre- and post-MALDI microscopy to measure >100 metabolites in >10000 individual cultured cells alongside fluorescence, morphology and spatial context.

Prior mass spectrometry approaches detected metabolites in single cells or small groups of cells, but they could not connect those metabolites to cell phenotypes or to how neighboring cells are arranged. SpaceM maps MALDI laser ablation marks onto cells segmented from microscopy, normalizes by sampled area and keeps only intracellular signals, so cultured monolayers can be analyzed in situ for both molecules and space.

Key results

  1. ShownIntracellular metabolomes alone separate co-cultured human HeLa cells from mouse NIH3T3 fibroblasts. A single fibroblast surrounded by HeLa cells was also classified correctly, which supports single-cell resolution.

    88 metabolites (FDR<10%) were profiled in 1624 co-cultured cells (958 HeLa, 666 NIH3T3). A Gaussian-kernel SVM with 10-fold cross-validation reached 90.4% accuracy, and 87.6% using a random half of the metabolites (1000 repetitions). Top t-test markers: PI(34:1) and PI(34:2) for HeLa, PE(40:6) for NIH3T3. · SpaceM predicts cell types with single-cell…; Figure legends

  2. SuggestedIn co-culture, each cell type's metabolome becomes more like the other type's. The effect depends on nearby cells of the other type and is stronger in NIH3T3 fibroblasts than in HeLa cells.

    On an LDA discriminant where the monocultures do not overlap, 39.2% of co-cultured NIH3T3 and 48.2% of HeLa cells fall in the mixing region or inside the other type's interval. Accuracy is 90.4% in co-culture vs 96.6% in monoculture. The effect peaks at 58 µm for NIH3T3 (rs=0.43, -log10 p=17.5) vs 107 µm for HeLa (rs=-0.18, -log10 p=3.8). · Intermixing of metabolomes of co-cultured HeLa…; Figure legends

  3. SuggestedControl analyses argue against technical artifacts as the cause of the intermixing. Most metabolites stay within cell boundaries, and classification accuracy changes little when the analysis is restricted to well-localized metabolites and to cells without co-sampling ablation marks.

    65 of 88 metabolites were well-localized; delocalized ones such as LPE(18:0) had a median distance outside cells of <5 µm. Accuracy was 88.9% with well-localized metabolites only, and 90.7% after also excluding 878 co-sampled cells (444 HeLa, 332 NIH3T3 kept), vs 90.4% overall. 72.25% of cells were sampled by one ablation mark. · The SpaceM method; Intermixing of metabolomes of co-cultured HeLa…; Discussion; Figure legends

  4. ShownIn human dHepaRG hepatocytes stimulated with TNFa plus fatty acids, single-cell lipid profiles follow lipid-droplet (LD540) staining. Steatotic cells preferentially accumulate longer-chain phospholipids.

    167 metabolites (FDR<10%) in 2370 cells; principal components correlated with LD540 fluorescence. TGs, DGs and PCs were the most correlated lipid classes, with TG(50:1) the top lipid (p<0.001). Carbon number correlated positively with LD540 for PCs, sphingomyelins and PEs, and negatively but not significantly for DGs and TGs. · SpaceM discovers accumulation of long-chain…; Figure legends

  5. SuggestedAcross control, fatty-acid, LPS and TNFa conditions, oleic and linoleic acid levels move in opposite directions yet stay tightly coupled within individual cells. PIP(38:4) and PIP(38:5) are higher in hepatocytes with high cell-cell contact.

    136 metabolites in 22258 cells (CTRL n=5654, FA n=4972, LPS n=5062, TNFa n=6570; three batch-corrected replicates per condition). Oleic acid rose and linoleic acid fell with stimulation, and the two correlated across all cells. The contact association with metabolite levels is shown for one TNFa replicate (1830 cells); AMP showed no association and oleic acid a negative one. · SpaceM discovers intracellular co-regulation of…; SpaceM discovers association of PIP(38:5) with…; Cell filtering and batch correction; Figure legends

Methods, data and reproducibility

Nuclear-stained cell monolayers are imaged by bright-field and fluorescence microscopy, then by MALDI-imaging mass spectrometry with false discovery rate (FDR)-controlled METASPACE annotation. Post-MALDI microscopy locates ablation marks, which are co-registered to segmented cells for area-weighted normalization, selection of intracellular metabolites and cell filtering.

  • Co-culture of HeLa Kyoto H2B-mCherry and NIH3T3-GFP cells on glass chamber slides, cell type labeled by constitutive fluorescence · human (HeLa) and mouse (NIH3T3) · n = 1624 cells (958 HeLa, 666 NIH3T3)
  • Monoculture controls of HeLa and NIH3T3 cells for comparison with the co-cultures · human (HeLa) and mouse (NIH3T3) · n = 2197 cells (1603 HeLa, 594 NIH3T3)
  • Differentiated HepaRG hepatocytes stimulated with TNFa plus oleic and palmitic acids, LD540-stained for lipid droplets · human · n = 2370 cells
  • dHepaRG hepatocytes under four conditions (CTRL, FA, LPS plus FA, TNFa plus FA), three technical replicate wells each · human · n = 22258 cells (after removing 1240 of 23498)
  • Lipid (Folch) and metabolite (80% methanol) extracts analyzed by LC-MS/MS to validate METASPACE annotations
Methods & tools
METASPACE FDR-controlled annotation against HMDB v2.5, AP-SMALDI source with Q Exactive Plus mass spectrometer, TM-Sprayer robotic matrix sprayer (DHB and DAN matrices), Nikon Ti-E microscope with NIS Elements; FIJI stitching plugin, CellProfiler (cell segmentation), Matlab and numpy (Fourier filtering, Otsu thresholding of ablation marks), scipy 0.18.1 basin-hopping (image co-registration, thresholding), scikit-learn 0.19.1 (LDA, PCA), Support Vector Machine with Gaussian kernel, 10-fold cross-validation, neuroCombat (ComBat batch correction), ImageQuest v1.1.0 (RAW to imzML conversion), Vanquish UHPLC-Q Exactive Plus LC-MS/MS with LipidBlast, LIPID MAPS, mzCloud
Software & environment
Python 3.6.2 with scikit-learn 0.19.1, scipy 0.18.1, matplotlib 2.1, seaborn 0.8.1; Matlab and CellProfiler (versions not stated); Thermo Tune 2.5 Build 2042; ImageQuest v1.1.0; METASPACE annotation against HMDB v2.5 at 3 ppm
Reproducibility gaps
No repository is given for SpaceM's own integration code; the only GitHub link is the third-party neuroCombat package. The per-dataset METASPACE links are unfilled 'URL for' placeholders, and the annotation validation details are in Supplementary Data S1.
Limitations (6)
  • Highly confluent monolayers are nearly impossible to sample without the laser hitting several cells at once, and applying the MALDI matrix can reduce spatial resolution. · stated by the authors
  • How far a metabolite spreads outside cells depends on the metabolite (LPE(18:0) spreads, the similar LPE(18:1) does not), and the cause is unresolved; matrix spraying is only a hypothesis. · stated by the authors
  • The method needs adherent cells grown as monolayers, because cells growing on top of each other increase co-sampling. Applying it to tissue sections is left for future work. · stated by the authors
  • PIP(38:4), one of the two lipids most associated with cell-cell contact, was not confirmed by LC-MS/MS. · stated by the authors
  • The text contradicts itself in places: the HeLa intermixing distance is 107 µm in the main text but 120 µm in the Figure 4 legend, and monoculture replicates are described as two in the results but one in the methods. · noted in this summary
  • The intermixing is read as a biological effect of neighboring cells, but no mechanistic test was run (for example, conditioned medium or physically separated co-culture). It rests on two co-culture replicates and correlations (rs=0.43 for NIH3T3). · noted in this summary

Read this if you want to link single-cell metabolite or lipid profiles to fluorescence phenotypes and neighbor arrangement in cultured monolayers using MALDI-imaging and microscopy.

Citations
393
FWCI
19.21
Field percentile
top 1%

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISDelafiori '25Blanc '24Nunes '23Potthoff '25Tian '22Buglakova '23
70% semantic matchMolecular Biology
Jeany Delafiori, Mohammed Shahraz, Andreas Eisenbarth+7 more·Cell·Sep 5, 2025· 35 citations