Spatial single-cell profiling of intracellular metabolomes in situ

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.

To bridge this gap, we designed SpaceM, a method for spatial single-cell metabolomics of cell 55 monolayers that integrates MALDI-imaging mass spectrometry with bright-field and 56 fluorescence microscopy. Integration with microscopy enables associating metabolites with 57 fluorescence and morphological cell properties (fluorescent reporter intensity, area, compactness, 58 shape) as well as with spatial features quantifying multi-cellular organization. The integration 59 was enabled by a method for precise detection of parts of cells sampled by MALDI laser with the 60 help of sequential microscopy, novel image analysis, and a novel cell-ablation marks 61 normalisation strategy. Using the False Discovery Rate-controlled metabolite annotation, and 62 novel methods for unbiased selection of intracellular metabolites and for filtering out poor 63 quality cells allowed us to perform high-throughput analyses with >100 metabolites detected in 64 >10000 individual cells, with a high reproducibility between replicates. We validated SpaceM by 65 investigating metabolomes of co-cultured HeLa and mouse fibroblasts cells as well as of 66 differentiated human steatotic hepatocytes stimulated with pro-inflammatory factors that 67 provided rich metabolic, phenotypic, and spatial information.  (Baker et al., 2017). MALDI-imaging is increasingly used for spatial metabolomics 85 (Palmer et al., 2016) and was demonstrated to achieve the femtomolar-levels sensitivity 86 (Soltwisch et al., 2015). This, together with soft ionisation preventing excessive in-source 87 molecular fragmentation makes it a perfect choice for single-cell metabolomics as demonstrated 88 by others (Do et al., 2017;Ibáñez et al., 2013). The experimental part of SpaceM combines 89 MALDI-imaging with microscopy as well as with collecting supporting information to integrate 90 these two sources of data ( Figure 1; for a detailed workflow see Figure S1). The cells for 91 SpaceM are cultured on a labtek chamber glass slide in a monolayer, with the cell confluence 92 sufficient to allow cells to interact between each other but at the same time preventing the growth 93 of cells on top of each other. After washing, cells are fixed to halt enzymatic activity, stained 94 with a fluorescent dye with the staining protocol compatible with metabolomics, and dried in a 95 desiccator following regular cell preparation protocols. SpaceM requires the Hoechst (or any 96 similar) staining for nuclei detection. For investigation of the steatotic hepatocytes, we also used 97 the lipophilic LD540 staining to detect lipid droplets (Spandl et al., 2009). Then, bright-field and 98 fluorescence microscopy images of cells are collected with the following two aims in mind. First, the cell segmentation of the microscopy images provides cell outlines and enables cell 100 localization. Second, microscopy provides rich phenotypic information about single-cell 101 fluorescence, immunochemistry and spatio-morphological properties of the cells. In the next 102 experimental step, MALDI-imaging is applied to the dried cells to collect mass spectra across 103 cells and extracellular areas. MALDI-imaging procedure starts with application of an ionisation-104 enhancing matrix. Similar to MALDI-imaging of tissues, we used a robotic sprayer for enhanced 105 extraction, high spatial resolution, and high reproducibility. MALDI-imaging generates big 106 datasets with millions of mass-to-charge channels. For finding metabolic signals in this data, we 107 exploited the False Discovery Rate-controlled metabolite annotation implemented as the 108 METASPACE cloud software (http://metaspace2020.eu) (Palmer et al., 2017). METASPACE is 109 an essential step as it reduces millions of mass-to-charge (m/z)-values to ~100 metabolite 110 annotations, filters out signals representing matrix and contaminants, ensures quality control and 111 represents metabolite images a user-friendly way. In the last experimental step, we performed 112 post-MALDI microscopy to determine which cells were sampled by the MALDI-imaging laser 113 and to associate MALDI-imaging spectra with the cells. Next, we performed data integration 114 with the first step associating ablation marks with individual cells. We detected MALDI laser 115 ablation marks in post-MALDI microscopy images using a customized 2D Fourier 116 Transformation image analysis method that exploits similarities between ablation marks and the 117 regularity in spacings between them. Then, we obtained positions of the MALDI-imaging 118 ablation marks within the cell areas by co-registering pre-MALDI microscopy images 119 (containing cell outlines) with post-MALDI microscopy images (containing ablation marks 120 outlines) ( Figure S2). For a majority of cells, a cell was sampled with just one ablation mark. In 121 our benchmarking experiment with HeLa cells and NIH3T3 fibroblast (described later), 72.25%, 122 8 23.8%, 2.8%, 0.9% cells were sampled with 1, 2, 3, 4 ablation marks, respectively. To integrate 123 metabolic profiles from several ablation marks co-sampling the same cell as well as to reduce the 124 confusion of co-sampled cells, we developed a cell-ablation marks normalization strategy 125 ( Figure 2). The normalization provides the metabolite intensity normalized by the area that is a 126 natural readout for metabolite concentration. Next, we developed a strategy to distinguish 127 intracellular from extracellular metabolites by requiring an intracellular metabolite intensities to 128 be highly correlated with the cell spatial distribution. Finally, similar to cell filtration strategies   In order to assess whether the metabolic profiles can predict the cell type with a single-cell 170 resolution, we evaluated a case of a fibroblast surrounded by HeLa cells ( Figure 3D). As shown 171 in Figure 3E, its cell type was predicted correctly. The phosphoethanolamine PE(40:6), the  we could predict the cell type of co-cultured cells with 90.4% accuracy, for the monocultured 212 cells we could predict their cell type with a higher accuracy of 96.6%.

213
Furthermore, we discovered that the metabolic intermixing between cells of two types happens 214 locally and can be considered a short-distance effect. The extent of the metabolic intermixing for 215 13 NIH3T3 cells depends on the presence of HeLa in their close vicinity ( Figure 4D). We estimated 216 that the metabolic intermixing is the strongest at the distance of 58 µm for NIH3T3 and 107 µm 217 for HeLa, with NIH3T3 affected most (rs=0.43, -log10(p-value)=17.5 compared to rs=-0.18, -218 log10(p-value)=3.8). We evaluated whether the observed metabolic intermixing can be observed 219 either due to metabolite delocalization during sample preparation or due to co-ablation of cells.

220
Sample preparation is known to be a key to achieve high spatial resolution MALDI-imaging. accuracy. The fact that the cell type prediction accuracy did not increase after considering 230 localized metabolites only suggests that delocalization does not explain the observed intermixing 231 between the cell types. Next, we evaluated whether co-ablation of cells can be a reason for the 232 observed metabolic intermixing. We considered only the cells which had uniquely-associated 233 ablation marks and excluded 878 cells which were co-sampled (having co-sampling ablation 234 marks, Figure 4H). Still, considering both well-localized metabolites and cells without co-  Figure 5D shows a single-cell scatterplot for the most correlated lipid TG(50:1)  Interestingly, for PCs, sphingomyelins, and phosphoethanolamines, the LD540 fluorescence was 304 found to be positively correlated with the number of carbons suggesting that LDs in steatotic 305 hepatocytes preferentially accumulate long-chain species of these phospholipids ( Figure 5E).

306
The opposite effect (negative correlation) was observed for di-and triglycerides albeit not 307 significant ( Figure 5F).      PIPs with high cell-cell contact. 463 We expect SpaceM to be broadly applicable to any adherent cells cultured in a monolayer, 464 avoiding growing on top of each other that can lead to increased co-sampling. In our experience,     Bremen, Germany) supplemented with 10% FBS, 100 U/ml penicillin, 100 µg/ml streptomycin 619 (Gibco) and 1 mM sodium pyruvate (Gibco). Cells were trypsinized with 0.25% trypsin-EDTA 620 (Gibco) and split 1:10 twice a week. Two technical replicates for the co-cultures and one 621 replicate for monoculture were used. Trypsinized cells were counted and cells were seeded on 4-622 well-glass labtek chamber slides (Lab-Tek II, CC2) (ThermoFisher Scientific). In the co-culture 623 experiment, equal number of cells of each cell type was added into each well (4x10 5 cells/well).

624
After 48h of incubation cells were washed with PBS. After washing, the cells were fixed for 15 625 min with 4% paraformaldehyde (Sigma Aldrich, Darmstadt, Germany) at room temperature.

626
Then the cells were stained with DAPI (1µg/ml) (ThermoFisher Scientific) in PBS for 20 min at 627 room temperature.