Abstract
A common drawback of metabolic analyses of complex biological samples is the inability to consider cell-to-cell heterogeneity in the context of an organ or tissue. To overcome this limitation, we present an advanced high-spatial-resolution metabolomics approach using matrix-assisted laser desorption/ionization mass spectrometry imaging (MALDI-MSI) combined with isotope tracing. This method allows mapping of cell-type-specific dynamic changes in central carbon metabolism in the context of a complex heterogeneous tissue architecture, such as the kidney. Combined with multiplexed immunofluorescence staining, this method can detect metabolic changes and nutrient partitioning in targeted cell types, as demonstrated in a bilateral renal ischemia–reperfusion injury (bIRI) experimental model. Our approach enables us to identify region-specific metabolic perturbations associated with the lesion and throughout recovery, including unexpected metabolic anomalies in cells with an apparently normal phenotype in the recovery phase. These findings may be relevant to an understanding of the homeostatic capacity of the kidney microenvironment. In sum, this method allows us to achieve resolution at the single-cell level in situ and hence to interpret cell-type-specific metabolic dynamics in the context of structure and metabolism of neighboring cells.
Subject terms: Metabolomics, Metabolomics, Metabolism
Understanding dynamic metabolic changes in complex biological samples often overlooks heterogeneity in cell composition. Wang et al. combine mass spectrometry imaging, isotope tracing, and multiplexed immunofluorescence microscopy to study metabolic dynamics in the kidney upon ischemia–reperfusion.
Main
Spatial omics methods are becoming an increasingly important means to provide insights into in situ tissue cellular heterogeneity and how cells interact and behave within their microenvironments. The aim of analyzing the entirety of cellular metabolites at single-cell resolution is challenging owing to enormous structural diversity, rapid turnover, and low analyte abundances in a limited sample volume. Previously, MALDI-MSI-based metabolomics methods have been proposed as a way to measure spatial distribution of metabolites and lipids at single-cell resolution1,2. To date, MALDI-MSI-based single-cell metabolomics studies have, however, focused on analyzing single-timepoint metabolic ‘snapshots,’ which provide valuable information on in situ metabolic heterogeneity but crucially lack insight into the dynamic component of cellular metabolism3–5; hence, there is a need for approaches that provide a truly comprehensive understanding of the interplay between biochemical alterations and cell-type-specific functions, metabolic fluxes, and dynamic interpretations. To this end, isotope tracing has been used to model dynamic metabolic processes using bulk metabolomics6, in which complete tissues are homogenized and reduced to a single metabolic profile, disregarding tissue and metabolic heterogeneity. The combination of isotope tracing and subsequent analysis by MALDI-MSI could be considered as an approach to assess metabolic dynamics in situ7–9. Here, we describe a platform based on high-spatial-resolution MALDI-MSI (5 × 5 µm2 pixel size) that uses isotope tracing to allow for in situ cell-type-specific dynamic metabolism measurements to uncover cell metabolism within the architecture of the tissue that the cells reside in. The average diameter of most cells in kidney tissue is around 10 µm, and the proposed method provides a single measurement at subcellular-level resolution because each cell has a size of approximately 4 pixels in MALDI-MSI images. To gain the most information from the measured area, all pixels, not just those segmented into individual cells, will be used for the cell-type-specific metabolomics analysis that we describe below.
Our method uses 13C-labeled nutrients that allow tracing of the spatiotemporal incorporation of the 13C isotopes into the main intermediates of glycolysis and the tricarboxylic acid (TCA) cycle (Fig. 1a). Using different labeled nutrients allows not only visualization of dynamic changes over time, but also resolution of their contributions to specific metabolic pathways. To accomplish this, we used a tissue culture system in which 350-µm-thick vibratome slices of fresh mouse kidney10 (Fig. 1b) were incubated for up to 2 hours. During incubation, 13C-isotope-labeled nutrients were introduced in parallel to the tissue culture medium at selected timepoints, which led to efficient and biochemically meaningful labeling of metabolically active cells11. We could then measure metabolic changes in a non-steady state. Subsequently, tissue slices were quenched with liquid nitrogen and sectioned into 10-µm-thick sections, thaw-mounted on conductive glass slides, and coated with a chemical matrix for MALDI-MSI measurements. Negative-ion-mode MALDI-MSI at subcellular resolution (that is, pixel size of 5 × 5 µm2) was used to detect metabolites and (phospho)lipids12 (Extended Data Table 1) from the collected tissues. Following MALDI-MSI measurements, the sections were stained and subsequently imaged using multiplexed immunofluorescence (IF) microscopy for cell-type identification (Fig. 1c). By combining IF and spatial segmentation of the MALDI-MSI data (on the basis of uniform manifold approximation and projection (UMAP) of the lipid signals, in particular), cell-type-specific signatures were established using the following assumptions: (1) MALDI-MSI lipid profiles are cell-type specific and (2) major phospholipid species, important for cell typing, are mainly cell-membrane components13 and thus are stable during the 2 hours of isotopic labeling (Extended Data Fig. 1). Providing functional evidence for this approach, MALDI-MSI and molecular histology of tissues incubated with [U-13C]linoleate, [U-13C]glutamine, or [U-13C]glucose revealed similar distributions of lipid signatures for cell typing (Extended Data Fig. 1f). MALDI-MSI measurements were performed on sections at different timepoints and with different 13C-enriched nutrients to obtain the labeling datasets for each condition. Thus, in order to show all the dynamic metabolic changes within 1 pixel without batch effects, a two-step process was followed, as introduced by Stuart et al. in the Seurat package14: (1) single-pixel lipid profiles were used to identify anchors between two datasets; (2) 13C-labeled-metabolite production was imputed into the control dataset by transferring the abundance of all measured 13C-enriched metabolites from the labeling datasets using k-nearest neighbors (KNN) analysis (Fig. 1d). An imputed dataset containing the complete 13C-labeling information, representing the predicted values of mass isotopomers from the pixels with a similar lipid profile for each tracing experiment, from different timepoints and nutrients could thus be established. Dynamic metabolic calculations, including metabolic changes and pathway convergence, and assessment of 13C enrichment of isotopologues were performed on single-pixel data. Finally, the in situ heterogeneity of metabolic dynamics was visualized in pseudoimages generated from the calculated values (Fig. 1e).
Extended Data Table 1.
This approach allowed us to detect both endogenous and 13C-labeled metabolites from glycolysis and the TCA cycle, as well as branching metabolites, including hexose, glycerol-3-phosphate (G3P), 3-phosphoglycerate (3PG), ribose-5-phosphate/xylulose-5-phosphate (R5P/X5P), lactate, glutamate, glutamine, malate, aspartate, and linoleate (Extended Data Fig. 2). To validate metabolite imputation performance, a ‘leave-one-factor-out’ cross-validation15 was applied to a bisected MALDI-MSI dataset from a single kidney section after 2 hours of incubation with [U-13C6]glucose (Extended Data Fig. 3). The ‘new’ datasets (samples 1 and 2) created from the single MALDI-MSI measurement underwent identical sample handling and thus had comparable technical variation and metabolic activity (Extended Data Fig. 3a). The correlation between the imputed and measured metabolites at the single-pixel level was moderate (mean r = 0.49, P < 0.001, Extended Data Fig. 3b). To test imputation performance at the cell-type level, we divided all pixels into 29 clusters on the basis of their lipid profiles (Extended Data Fig. 3c). Following clustering, cell-type validation was performed using Spearman’s correlation analysis on the average value of each cluster. A very strong correlation (mean r = 0.94, P < 0.05) was observed between the imputed and detected values (Extended Data Fig. 3d). In addition, when comparing imputed and detected metabolite abundance, the average 13C enrichment of isotopologues in relation to endogenous metabolites in each cluster was highly accurate (ratio = 0.99 ± 0.04) (Extended Data Fig. 3e–g). We continued to test the MALDI-MSI data from two kidneys (samples 3 and 4) and found a strong correlation between imputed and detected metabolite intensity at the cluster level (mean r = 0.82, P < 0.001, Extended Data Fig. 3h–j). Following validation of the dynamic metabolism analysis, which revealed that the imputed cell-type-specific 13C-enrichment calculations were very reliable, pseudoimages were generated to show the spatial changes in dynamic metabolism within one tissue (Extended Data Fig. 4).
To demonstrate how the method can be used, we induced moderate ischemic injury to mouse kidneys, similar to that observed in a kidney-transplantation setting. We then studied the tissue metabolism 2 weeks after the injury. At this timepoint, part of the initial proximal tubular epithelial injury has regenerated (Extended Data Fig. 5a) and has taken up a normal phenotype again (characterized by Lotus tetragonolobus lectin (LTL) positivity and kidney injury molecule-1 (KIM1) negativity), while other epithelial cells may still have a maladaptive inflammatory phenotype, as exemplified by vascular cell adhesion molecular 1 (VCAM1) expression in the presence of LTL, or may have lost the tubular epithelial phenotype all together (LTL negative)16. To visualize overall metabolic heterogeneity in whole kidney tissue sections, we used MALDI-MSI at a 20 × 20 µm2 pixel resolution, which spatially and molecularly resolved all major renal structures on the basis of their specific lipid signatures (Fig. 2a and Extended Data Fig. 1). Two weeks after kidney bIRI, we found additional clusters (Injured_1, Injured_2, Injured_3, and Injured_4) with unique lipid signatures that corresponded with loss of LTL and E-cadherin staining (Fig. 2a–c and Extended Data Fig. 5b–d). A loss of the healthy proximal tubule (PT) lipid signature and an increased presence of oxidized lipid species (m/z 865.5; Extended Data Table 1) were observed in the areas with persisting epithelial injury (Fig. 2c). We then generated spatial segmentation images using an integrated 3D UMAP analysis, which displayed the kidney molecular histology. These images showed the presence of ongoing tubular injury in both the cortex and outer medulla of bIRI kidneys (Fig. 2d). There was substantial loss of PT structures from the outer stripe of outer medulla (PT-S3) compared with the renal cortex (PT-S1/PT-S2) (Fig. 2e). These observations were also confirmed using a histopathological staining (Fig. 2f).
Considering that a tissue’s microenvironment is an important cue for cell-specific metabolism within that tissue, we next used high-spatial-resolution MALDI-MSI measurements with a pixel size at the subcellular level (5 × 5 µm2) to investigate the metabolic heterogeneity in the PT of the kidney cortex and outer stripe of outer medulla and their response to injury. Following MALDI-MSI, IF staining was conducted to identify cell types (Fig. 3a). We could distinguish two PT-S1 and PT-S2 cortex segments and one medullary PT-S3 segment on the basis of their lipid profiles (Fig. 3a–c). PT-S1 and PT-S2 were combined for subsequent analyses because we could not annotate them on the basis of LTL staining or their location. Analysis of the isotopologues showed that PT-S3 had lower hexose M+6 ([U-13C6]hexose) and higher lactate M+3 ([13C3]lactate) enrichment than did PT-S1/PT-S2 (Fig. 3d). In addition, we found lower enrichment of R5P/X5P M+3 (glycolysis branch) as well as lower enrichment of glutamate ([13C2]glutamate; Glu M+2) from glucose carbons entering the TCA cycle in PT-S3 (Fig. 3d). To further evaluate the use of 13C-labeled nutrients in the TCA cycle, we measured the percentage enrichment of [U-13C5]glutamine-derived isotopologues, [U-13C5]glutamate (Glu M+5), [13C4]malate (malate M+4), and [13C3]glutamate (Glu M+3), all of which are intermediates of one oxidative TCA cycle. There was lower [U-13C5]glutamine and higher [U-13C5]glutamate enrichment in PT-S3 cells than in PT-S1/PT-S2 cells (Fig. 3e). Notably, the enrichment of [13C4]malate and [13C3]glutamate were relatively lower in PT-S3 than in PT-S1/PT-S2 from the same sample, pointing to a lower contribution of both [U-13C5]glutamine and [13C5]glutamate to the oxidative TCA cycle in PT-S3 (Fig. 3e). The contribution of [U-13C18]linoleate to the [13C2]glutamate isotopologue through fatty acid oxidation was lower in PT-S3 as well (Fig. 3f). By combining these data, we found that [U-13C5]glutamine, rather than [U-13C6]glucose or [U-13C18]linoleate, is glutamate’s main carbon source in all PT areas (Fig. 3g). However, usage of glutamate for the oxidative TCA cycle differed between PT segments, as shown by glutamine carbon tracing to characteristic mass isotopomers (Fig. 3e). Together, these data show that, in the normal kidney, there is a marked heterogeneity among PT segments with respect to TCA metabolite consumption and glycolysis: the PT-S3 segment appears to be adapted to the low-oxygen-tension environment of the outer medulla. These observations correspond to previous studies using isolated or micro-dissected PTs17,18.
We then sought to identify metabolic (mal)adaptation after bIRI by comparing healthy PT (LTL+VCAM1–) with failed maladaptive repaired PT (VCAM1+; FR_PT) (Fig. 4a). Following MALDI-MSI, we selected all pixels with a PT, determined on the basis LTL, VCAM1, and KIM1 IF staining (Fig. 4b and Extended Data Fig. 6). Four types of PT could subsequently be identified on the basis of lipid profiling: LTL+VCAM1–KIM1– (PT-S1/S2) and LTL+VCAM1 + (FR_PT1) in the cortex, and LTL–VCAM1 + (FR_PT2 and FR_PT3) and LTL+ in the outer stripe of the outer medulla (PT-S3) (Fig. 4c,d). The clusters present in the cortex areas showed a trajectory towards maladaptive repair on the basis of their lipid profiles (Fig. 4e). The spatial-trajectory map reflected progressive maladaptive repair in different areas (Fig. 4e), consistent with loss of LTL staining and increased VCAM1 expression. Notably, FR_PT and normal PT were localized in the same region (Fig. 4e), indicating that maladaptive failed repair could not be explained by a differential impact of only the ischemic insult. We then calculated the correlation of pseudotime scores with 13C enrichment of several isotopologues (Fig. 4f). Following the trajectory from healthy PT areas toward progressively impaired PT areas, the pseudotime score was negatively correlated with [U-13C6]hexose enrichment (mean r = –0.56, P < 0.001) and its downstream product [13C2]glutamate (mean r = –0.38, P < 0.001) but was weakly positively correlated with [13C3]lactate enrichment (mean r = 0.22, P < 0.001), in combination with more total lactate present in FR_PT (Fig. 4j). One possible explanation for this is a higher glycolytic flux to the production of lactate and a lower contribution of [U-13C6]glucose to the TCA cycle in FR_PT (Fig. 4f,g,i), although it cannot be ruled out that other sources contribute to total lactate. A reduction of both [U-13C5]glutamine enrichment and labeled-glutamine-derived isotopologues through the oxidative TCA cycle was also observed (Fig. 4f,h,j), but glutamine, compared with glucose and linoleate, remained the main carbon source for glutamate in FR_PT (Fig. 4k). From these data, we conclude that the maladaptive repair in the PT is characterized by differences in production of lactate, which could be possibly the result of higher glycolytic activity and, as injury progresses, a concomitant reduction of TCA cycle metabolite consumption.
We then focused on normal PTs in the bIRI kidneys. Strikingly, when compared with the sham kidneys, these epithelial cells still displayed abnormal nutrient partitioning to TCA metabolism, both in the cortex as well as in the outer medulla (Fig. 4l,m). Normal PTs, 2 weeks after bIRI, had lower [U-13C6]hexose and [13C2]glutamate enrichment than sham kidneys (Fig. 4l). In addition, although normal PTs exhibited higher [U-13C5]glutamine enrichment, we found a lower 13C enrichment in specific isotopologues that are produced through the oxidative TCA cycle, such as [U-13C5]glutamate, [13C4]malate, [13C4]aspartate, and [13C3]glutamate (Fig. 4n). This is remarkable, as the TCA is a central hub in cell function and a cell’s ability to adapt to its environment. Its metabolites not only serve in the biosynthesis of nucleotides and macromolecules such as lipids and proteins, but also have recently been recognized to control gene regulation through post-translational histone modifications and DNA methylation19; hence, it will be of relevance to study how the disbalance of TCA metabolite homeostasis, even in normal PT cells after bIRI, relates to chronic progressive kidney injury or loss of repair after a new injury.
The dynamic nature of metabolism, and the metabolic heterogeneity of different cell types and phenotypes within the architecture of a single tissue, call for single-cell-resolution measurements of metabolic dynamics. Currently available methods can provide singular spatial metabolic snapshots, bulk dynamic metabolism of homogenized tissue material, or isotope tracing on tissue with low spatial resolution1,2,6–9. Other strategies, based on single-cell transcriptomics20, cytometry by time of flight (CyTOF)21, or metabolic probes22, have yielded information on single-cell metabolism. However, these methods are based on the detection of surrogates for metabolic activity, such as metabolic enzymes and their transcripts. Here, we provide an in situ cell-type-specific measurement of metabolic dynamics through the direct detection of metabolic intermediates at single-cell resolution. Furthermore, to detect the dynamic metabolism of less abundant cell types in tissue, such as endothelial or immune cells, we introduced IF staining values to correlate with lipid profiles as a threshold for cell selection. For example, the combination of IF staining and lipid profiles allowed us to identify two populations of renal cortex endothelial cells: glomerular renal endothelial cell (gRECs) and cortical renal endothelial cells (cRECs) (Extended Data Fig. 7a–d). In line with known continuous VEGF signaling from podocytes23, gRECs showed higher [13C3]lactate enrichment than cRECs (Extended Data Fig. 7e). The combination of multiplexed IF staining on post-MSI-analyzed tissues with MALDI-MSI data shows great potential for multi-omics analysis by combining dynamic metabolomics with antibody-based spatial proteomics24,25. This will link dynamic metabolism to the cell genotype and phenotype, giving more insight into metabolic regulated cell behavior. The accuracy of image co-registration of the spatial distribution of clustered pixels with IF staining, as in the bIRI study in this paper, is less critical than co-registration at the single-pixel level. To combine IF staining values with an MSI lipid dataset, a more advanced method for co-registration could be used26. A single-cell segmentation step based on IF can also be used to study dispersed single-cell populations in tissues, such as immune cells27. Although in this study we used pixels for UMAP and trajectory analysis, rather than individual cells after single-cell segmentation, both cell typing and trajectory were in line with IF staining, which indicates that single-cell-resolution pixels can be used for cell-type-specific analysis, as has been previously shown using spatial transcriptomics technologies28,29.
To make this method potentially usable in human tissue biopsies, we used vibratome sectioning for stable-isotope tracing on tissue ex vivo, which provides the possibility of tracing different isotopes using a single tissue sample. However, the current 2-hour incubation period limits the dynamic measurement to central carbon metabolic pathways (glycolysis and TCA cycle) in metabolic non-steady state, as well as the detection of 13C-labeled TCA intermediates from one oxidative cycle. Nonetheless, this MALDI-MSI-based method for cell-type-specific measurement of metabolic dynamics can be applied to tissue samples or cultured cells for which longer incubation of stable isotopes is feasible, thus enabling the detection of 13C-labeled metabolites in a pseudo-steady-state. For instance, in endothelial cells cultured in vitro for 24 hours in the presence of [U-13C6]glucose, we could detect not only the full range of isotopologues of TCA intermediates, such as citrate (M+1, M+2, M+3, M+4, and M+5) and succinate (M+1, M+2, M+3, and M+4), but also13C-labeled metabolites from other metabolic pathways, such as ribose-5-phosphate/xylulose-5-phosphate (R5P/X5P), adenosine monophosphate (AMP), adenosine diphosphate (ADP), uridine diphosphate glucose (UDP-Glc), uridine diphosphate N-acetylglucosamine (UDP-GlcNAc), and glutathione (Extended Data Fig. 8). Owing to the limitations of MALDI-MS, we could not distinguish isomeric molecular species, such as glucose, galactose, mannose, and inositol9, as well as the confounders interfering with the quantification of labeling peaks, such as [13C2]aspartate and [13C2]malate. To solve this, more advanced instrumentation using ion-mobility separation is needed30. Delocalization of metabolites is a limitation of spatial metabolomics with MALDI-MSI in general, which could cause false negatives by decreasing the signal differences between cell types, but is less likely to cause false positives31.
In summary, we have developed a platform to detect cell-type-specific dynamic metabolic changes at single-cell resolution. We demonstrated that this platform, applied to tissue samples, can discern metabolic heterogeneity within individual cell types in relation to the microenvironment. We used the kidney as an example of a complex and metabolically heterogenous organ, but this method can be used to investigate tissue homeostasis in general. Our method could be applied to detect changes in cell-specific metabolism, which occur in processes such as inflammation, fibrosis, and cancer biology.
Methods
Reagents
All reagents were obtained from Sigma-Aldrich, unless stated otherwise.
Mouse studies
For studies of overall metabolic changes, post mortem material of 12-week-old male C57BL/6J mice (n = 3) culled as breeding surplus were used. Mice were kept and cared for in accordance with the Experiments on Animals Act (Wod, revision 2014, the Netherlands) and EU directive no. 2010/63/EU. Mice were housed at 20–22 °C in individually ventilated cages, humidity controlled (55%) with free access to food and water and a light/dark cycle of daytime (06:30–18:00) and nighttime (18.00–06:30).
For bIRI) experiments, we used 12-week-old male constitutional renin reporter (B6.Ren1cCre/TdTomato/J) mice32. Six mice were divided into two groups randomly (n = 3/group). In short, mice were placed under isoflurane anesthesia and controlled body temperature was kept at 36.7 °C, and renal arteries and veins were exposed through laparotomy with median incision on both sides. Subsequently, both arteries and veins were ligated using clamps for approximately 18 minutes, after which the clamps were removed to resume blood flow (reperfusion) and the abdomen was closed. For the sham controls, mice were opened up and closed similarly, but no ligations were performed. At day 14 after surgery, mice were perfused with cold PBS–heparin (5 IU/mL) via the left ventricle at a controlled pressure of 150 mmHg for 6 minutes to exsanguinate the kidneys before removal. Animal experiments were approved by the Ethical Committee on Animal Care and Experimentation of the Leiden University Medical Center (permit no. AVD1160020171145).
Vibratome sectioning and tissue slice incubation
Directly after euthanization, kidneys were collected and kept in ice-cold sterile Hanks’ Balanced Salt Solution (HBSS) with 5 mM glucose and penicillin–streptomycin before vibratome sectioning. Each kidney was embedded in 4% low-melting-temperature agarose gel, and 350-µm-thick tissue slices were obtained from fresh tissue under ice-cold HBSS with 5 mM glucose and penicillin–streptomycin using a Vibratome VT1200 (Leica Microsystems). Slicing speed was 0.1 mm/second, and vibration amplitude was 2 mm.
Tissue slices were placed into culture plates and incubated in a well-defined medium (nutrients-free Seahorse XF DMEM assay medium (Agilent, 103681), supplemented with 2% FCS, 3 mM linoleic acid (dissolved with addition of 1% BSA), 5 mM glucose, 500 µM glutamine, 100 µM sodium acetate, 50 µM sodium citrate, and penicillin–streptomycin (pH adjusted to 7.4)) for up to 2 hours at 37 °C and 5% CO2. During incubation, medium was changed to media containing various 13C-labeled nutrients at different time points. For the 13C-labeling incubation, the same amounts of either [U-13C6]glucose (99%, Sigma, 389374), [U-13C6]glutamine (99%, Cambridge Isotope Laboratories, Inc. CLM-1822-H), or [U-13C18]linoleic acid (98%, Cambridge Isotope Laboratories, CLM-6855-0) were used to replace similar unlabeled nutrients in each medium. In the end, tissue slices were quenched with liquid N2 and stored at −80 °C.
Tissue preparation and matrix deposition
Tissue slices were embedded in 10% gelatin and cryosectioned into 10-µm-thick sections using a Cryostar NX70 cryostat (Thermo Fisher Scientific) at –20 °C. The sections were thaw-mounted onto indium-tin-oxide (ITO)-coated glass slides (VisionTek Systems). Mounted sections were placed in a vacuum freeze-dryer for 15 minutes prior to matrix application. After drying, N-(1-naphthyl) ethylenediamine dihydrochloride (NEDC) (Sigma-Aldrich, UK) MALDI-matrix solution of 7 mg/mL in methanol/acetonitrile/deionized water (70, 25, 5% vol/vol/vol) was applied using a SunCollect sprayer (SunChrom). A total of 21 matrix layers were applied with the following flow rates: layer 1–3 at 5 µL/min, layer 4–6 at 10 µL/min, layer 7–9 at 15 µL/min and 10–21 at 20 µL/min (speed x, medium 1; speed y, medium 1; z position, 35).
MALDI-MSI measurement
MALDI-TOF/TOF-MSI was performed using a RapifleX MALDI-TOF/TOF system (Bruker Daltonics). Negative-ion-mode mass spectra were acquired at a pixel size of 5 × 5 µm2 or 20 × 20 µm2 over a mass range of m/z 80–1000. Prior to analysis, the instrument was externally calibrated using red phosphorus. Spectra were acquired with 15 laser shots per pixel (for 5 µm measurement) or 200 laser shots per pixel (for 20 µm measurement) at a laser repetition rate of 10 kHz. Data acquisition was performed using flexControl (Version 4.0, Bruker Daltonics) and visualizations were obtained from flexImaging 5.0 (Bruker Daltonics). All the samples from same slides were measured randomly. MALDI-FTICR-MSI was performed on a 12 T solariX FTICR mass spectrometer (Bruker Daltonics) in negative-ion mode, using 30 laser shots and 50-µm pixel size. Prior to analysis, the instrument was calibrated using red phosphorus. The spectra were recorded over a m/z range of 100–1,000 with a 2M data point transient and transient length of 0.5592 seconds. Data acquisition was performed using ftmsControl (Version 2.1.0, Bruker Daltonics), and visualizations were obtained from flexImaging 5.0 (Bruker Daltonics). The images of lipid and metabolite distributions on tissue were exported from flexImaging 5.0. Following the MALDI-MSI data acquisition, excess matrix was removed by washing in 100% ethanol (2 × 5 min), 75% ethanol (1 × 5 min), and 50% ethanol (1 × 5 min), after which this MSI-analyzed tissue section was used for IF staining, as described below.
Immunofluorescence staining
After MALDI-MSI tissues on the slide were fixed using 4% paraformaldehyde for 30 minutes, antigen retrieval was performed using antigen retrieval buffer (Dako, Agilent Technologies) in an autoclave. Slides were blocked with 3% normal donkey serum, 2% BSA, and 0.01% Triton X-100 in PBS for 1 hour at room temperature. Primary anti-mouse-KIM1 antibody (5 µg/mL, R&D Systems, MAB1817), anti-VCAM1 antibody (1:250, Abcam, ab134047), anti-CDH1 antibody (1:300, BD Biosciences, 610181), anti-NPHS1 (2 µg/mL, R&D Systems, AF4269), anti-mouse-pan-endothelial-cell-antigen (MECA32, 2 µg/mL, BD Biosciences, 553849), Lotus tetragonolobus Lectin (LTL, 1:300, Vector Laboratories, B-1325), or lectin from Bandeiraea simplicifolia isolectin B4 (BS1-TRITC, 1:200, Sigma, L5264) were incubated overnight at 4 °C, followed by corresponding fluorescent-labeled secondary antibodies (donkey anti-rat-IgG AF488 (1:300, Invitrogen, A21208), donkey anti-rabbit-IgG AF647 (1:300, Invitrogen, A31573), donkey anti-mouse-IgG AF488 (1:300, Invitrogen, A21202), donkey anti-sheep-IgG AF568 (1:300, Invitrogen, A21099), streptavidin–Alexa Flour 568 (1:300, Invitrogen, S11226), streptavidin–Alexa Flour 647 (1:300, Invitrogen, S32357)) for 1 hour at 4 °C, when necessary. Slides were embedded in Prolong gold antifade mountant with DAPI (Thermo Fisher Scientific, P36931). Fluorescent images of the slides were recorded using a 3D Histech Pannoramic MIDI Scanner (Sysmex). Digital scanned images were co-registered to the MALDI-MSI data.
MSI data processing and analysis
MSI data were exported and processed in SCiLS Lab 2016b (SCiLS, Bruker Daltonics) with baseline correction using convolution algorithm. All MALDI-TOF-MSI data were normalized to the total ion count (TIC). Peak picking was performed (signal-to-noise ratio > 3) on the average spectrum, and matrix peaks were excluded from the m/z feature list. The m/z features which were present in both MALDI-FTICR-MSI and MALDI-TOF-MSI datasets, and which had similar tissue distributions, were further used for identity assignment of lipid species (Extended Data Fig. 9). The m/z values from MALDI-FTICR-MSI were imported into the Human Metabolome Database33 (https://hmdb.ca/) after re-calibration in mMass and annotated for lipids species with an error < ±5 ppm. For the small molecules detected only in MALDI-TOF, the m/z values from MALDI-TOF were imported into the Human Metabolome Database (https://hmdb.ca/) after re-calibration in mMass and annotated for metabolites with an error < ±20 ppm. The 13C-labeled peaks were selected by comparing the spectrum of control and 13C-labeling experiments, and annotated on the basis of the presence of unlabeled metabolites and their theoretical m/z values. For measurement at 20 × 20 µm2 pixel size, a total of 230 high-molecular-weight features (m/z > 400, predominantly phospholipids12, Extended Data Table 1) that not co-localized with matrix peaks were selected (signal-to-noise ratio > 3). For measurement at 5 × 5 µm2 pixel size, a total of 16 metabolites (9 unlabeled and 11 13C-labeled), 227 high-molecular-weight features that were not co-localized with matrix peaks were selected (signal-to-noise ratio > 3). Peak intensities of the selected features were exported for all the measured pixels from SCiLS Lab 2016b, which were used for the following analysis. Natural isotope abundance correction was performed for metabolites using R package IsoCorrectoR34.
For targeted endothelial cell analysis at 5 × 5 µm2 pixel size, the high-resolution staining image of MECA32 and NPHS1 on the MALDI-MSI measured area was exported from CaseViewer (3DHISTECH). Staining in the areas not measured by MALDI-MSI was removed. The resolution of the exported image was changed to 5 × 5 µm2 pixel size using Matlab R2019a to comply with the MALDI-MSI resolution. By merging it with the previous high-resolution image, the pixels fully covering MECA32 staining were selected as endothelial cells, and non-fully-covered pixels were discarded from the image. Next, the staining values of all the pixels were exported using Matlab R2019a with a cutoff of 5% (values < 13 were replaced with 0). In the end, the staining values corresponding to the same pixels as MALDI-MSI measurements were integrated into MALDI-MSI data and used for data analysis as described below.
For UMAP analysis, the datasets were transformed into a count matrix by multiplying the TIC-normalized intensities by 100 and taking the integer. This count data matrix was normalized and scaled using SCTransform to generate a 2D UMAP map using Seurat 3.0 in R (version 4.0). The distribution of the pixels from different clusters on tissues were co-registered to the IF staining, and cell types were identified on the basis of both staining and their morphology on the kidney. Using measurements of endothelial cells, two endothelial cell clusters were identified on the basis of MECA32 staining and their position, and a podocyte cluster was identified on the basis of NPHS1 staining. For data imputation, the metabolite m/z features from the ‘control,’ or unlabeled, dataset were removed, so only lipid m/z features were left, which were used as the query. Then, the MALDI-MSI data from 13C-labeling experiments were used as a reference to transfer metabolite production into the query using FindTransferAnchors and TransferData function from the Seurat 3.0 package in R. Both the query and reference were normalized and scaled using SCTransform. Ultimately, all the imputed metabolite productions were combined into one dataset, which contained the 13C-labeling information over time.
After combining all the imputed data into one dataset, the fraction enrichment of isotopologues was calculated on the basis of the ratio of each 13C-labeled metabolite (isotopologue) to the sum of this metabolite abundance in each pixel. The calculated fraction enrichment of isotopologues was used to generate pseudoimages together with pixel coordinate information exported from SCiLS Lab 2016b. The average fraction enrichment values (AUC normalized to total time) of identified clusters were used for generating graphs and statistical analysis. Hotspot removal (high quantile 99%) were applied to all the pseudoimages generated from calculated values.
The integrated data matrix from 3 IRI kidneys from Seurat 3.0 was further used to for the trajectory analysis using Monocle3 (ref. 35) in R. A spatial trajectory map was generated using Matlab R2019a based on embedding information of UMAP and pseudotime values calculated by Monocle3 and pixel coordinate information exported from SCiLS Lab 2016b as descripted below.
For spatial UMAPs, the datasets were used to generate a 3D UMAP map using packages Seurat 3.0 and plotly. The embedding information of the 3D UMAP was translated to RGB color coding by varying red, green and blue intensities on the three independent axes. Together with pixel coordinate information exported from SCiLS Lab 2016b, a MxNx3 data matrix was generated and used to generate UMAP images in Matlab R2019a.
Data collection and analysis were not performed blind to the conditions of the experiments. A step-by-step protocol associated with this study is available in Protocol Exchange (10.21203/rs.3.pex-1912/v1)36.
Validation test of imputation performance
Validation of the imputation performance using ‘leave-one factor-out’ cross-validation is shown in Extended Data Fig. 3. In short, the metabolite production from MALDI-MSI data was taken out from one kidney slice (sample 1, Extended Data Fig. 3a) or two different kidney slices (sample 3, Extended Data Fig. 3h) and use samples 2 and 4 to impute the metabolite production in sample 1 and in sample 3, respectively, as described above. Next, Spearman’s correlation was calculated between the imputed and detected values in sample 1 and 3. Values higher than 0.8 were graded as a very strong positive correlation. Values between 0.6 to 0.8 were a strong positive correlation, and between 0.4 to 0.6 a moderate positive correlation.
Statistics and reproducibility
No statistical methods were used to predetermine sample sizes, but our sample sizes are similar to those reported in previous publications9,16. All the experiments and data analysis were performed in triplicate (3 animals per group). No animals or data points were excluded from the analysis. All the data are presented as mean ± s.d., unless indicated otherwise. The average fractional contribution values of identified clusters were used for statistical analysis. AUC was used for comparison of the changes following timecourse of isotope tracer incubation up to 2 hours between different groups. Data normality and equal variances were tested using the Shapiro–Wilk test. Differences between groups were assessed by paired two-tailed Student’s t-test, two-tailed Student’s t-test, when not normally distributed, or by two-tailed F-test. P < 0.05 was considered statistically significant.
Reporting summary
Further information on research design is available in the Nature Research Reporting Summary linked to this article.
Supplementary information
Source data
Acknowledgements
The Novo Nordisk Foundation Center for Stem Cell Medicine (reNEW) is supported by Novo Nordisk Foundation grants (NNF21CC0073729). Financial support from the Jaap de Graeff-Lingling Wiyadharma subsidy of the Leiden University Fund (LUF) and the China Scholarship Council grant to G. W. are gratefully acknowledged. Financial support from the Marie Skłodowska-Curie Individual Fellowship to S.J.D. is gratefully acknowledged. We thank M. van Weeghel (AMC, Amsterdam, the Netherlands) and J.B. van Klinken (LUMC, Leiden, the Netherlands) for discussing the possibility of flux analysis methods.
Extended data
Author contributions
G. W. designed research study, conducted experiments, processed data, and wrote the manuscript. B. H. provided the technical guidance of MALDI-MSI experiments and data processing and wrote the manuscript. S.K. and M.G. provided expert knowledge on metabolomics and lipidomics analysis, read the manuscript, and provided helpful comments. R.B., A.K., and L.A.K.v.d.P. performed the animal experiments, read the manuscript, and provided helpful comments. A.M. provided professional instruction of single-cell data analysis, read the manuscript, and provided helpful comments. R.G.J.R. and C.W.v.d.B. helped with the experiments, read the manuscript, and provided helpful comments. P.C. and S.J.D. read the manuscript and provided helpful comments. B.M.v.d.B. and T.J.R. organized funding, designed research study, and wrote the manuscript.
Peer review
Peer review information
Nature Metabolism thanks Roland Nilsson and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Alfredo Giménez-Cassina, in collaboration with the Nature Metabolism team.
Data availability
The exported and processed MSI data for this study were deposited in FigShare at 10.6084/m9.figshare.20227419.v1. Owing to the large size of all raw data, parts of the raw MSI data are deposited to provide the necessary information of 13C-labeled metabolites and spectrum quality. For full raw MALDI-MSI data related to this study, please contact G. W. (g.wang@lumc.nl) or B. H. (b.p.a.m.heijs@lumc.nl). Upon reasonable request, data will be made available. Source data are provided with this paper.
Code availability
The codes used in this article is available in Github (https://github.com/Gangqiwang/scDYMO).
Competing interests
The authors declare no competing interests.
Footnotes
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Extended data
is available for this paper at 10.1038/s42255-022-00615-8.
Supplementary information
The online version contains supplementary material available at 10.1038/s42255-022-00615-8.
References
- 1.Rappez L, et al. SpaceM reveals metabolic states of single cells. Nat. Methods. 2021;18:799–805. doi: 10.1038/s41592-021-01198-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Taylor MJ, Lukowski JK, Anderton CR. Spatially resolved mass spectrometry at the single cell: recent innovations in proteomics and metabolomics. J. Am. Soc. Mass Spectr. 2021;32:872–894. doi: 10.1021/jasms.0c00439. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Sun N, et al. Mass spectrometry imaging establishes 2 distinct metabolic phenotypes of aldosterone-producing cell clusters in primary aldosteronism. Hypertension. 2020;75:634–644. doi: 10.1161/HYPERTENSIONAHA.119.14041. [DOI] [PubMed] [Google Scholar]
- 4.Basu SS, et al. Rapid MALDI mass spectrometry imaging for surgical pathology. NPJ Precis. Oncol. 2019;3:17. doi: 10.1038/s41698-019-0089-y. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Abbas I, et al. Kidney lipidomics by mass spectrometry imaging: a focus on the glomerulus. Int. J. Mol. Sci. 2019;20:1623. doi: 10.3390/ijms20071623. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Jang C, Chen L, Rabinowitz JD. Metabolomics and isotope tracing. Cell. 2018;173:822–837. doi: 10.1016/j.cell.2018.03.055. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Sugiura, Y. et al. Visualization of in vivo metabolic flows reveals accelerated utilization of glucose and lactate in penumbra of ischemic heart. Sci Rep. 6, 32361 (2016). [DOI] [PMC free article] [PubMed]
- 8.Cao, J. H. et al. Mass spectrometry imaging of L-[ring-13C6]-labeled phenylalanine and tyrosine kinetics in non-small cell lung carcinoma. Cancer Metab. 9, 26 (2021). [DOI] [PMC free article] [PubMed]
- 9.Wang L, et al. Spatially resolved isotope tracing reveals tissue metabolic activity. Nat. Methods. 2022;19:223–230. doi: 10.1038/s41592-021-01378-y. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Roelants C, et al. Ex-vivo treatment of tumor tissue slices as a predictive preclinical method to evaluate targeted therapies for patients with renal carcinoma. Cancers. 2020;12:232. doi: 10.3390/cancers12010232. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Fan TW, Lane AN, Higashi RM. Stable isotope resolved metabolomics studies in ex vivo tissue slices. Bio Protoc. 2016;6:e1730. doi: 10.21769/bioprotoc.1730. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Wang J, et al. MALDI-TOF MS imaging of metabolites with a N-(1-naphthyl) ethylenediamine dihydrochloride matrix and its application to colorectal cancer liver metastasis. Anal. Chem. 2015;87:422–430. doi: 10.1021/ac504294s. [DOI] [PubMed] [Google Scholar]
- 13.Eiersbrock FB, Orthen JM, Soltwisch J. Validation of MALDI-MS imaging data of selected membrane lipids in murine brain with and without laser postionization by quantitative nano-HPLC-MS using laser microdissection. Anal. Bioanal. Chem. 2020;412:6875–6886. doi: 10.1007/s00216-020-02818-y. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Stuart T, et al. Comprehensive integration of single-cell data. Cell. 2019;177:1888–188. doi: 10.1016/j.cell.2019.05.031. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Abdelaal T, Mourragui S, Mahfouz A, Reinders MJT. SpaGE: spatial gene enhancement using scRNA-seq. Nucleic Acids Res. 2020;48:e107. doi: 10.1093/nar/gkaa740. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Kirita Y, Wu H, Uchimura K, Wilson PC, Humphreys BD. Cell profiling of mouse acute kidney injury reveals conserved cellular responses to injury. Proc. Natl Acad. Sci. USA. 2020;117:15874–15883. doi: 10.1073/pnas.2005477117. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Ruegg CE, Mandel LJ. Bulk isolation of renal PCT and PST. I. Glucose-dependent metabolic differences. Am. J. Physiol. 1990;259:F164–F175. doi: 10.1152/ajprenal.1990.259.1.F164. [DOI] [PubMed] [Google Scholar]
- 18.Uchida S, Endou H. Substrate specificity to maintain cellular ATP along the mouse nephron. Am. J. Physiol. 1988;255:F977–F983. doi: 10.1152/ajprenal.1988.255.5.F977. [DOI] [PubMed] [Google Scholar]
- 19.Martinez-Reyes I, Chandel NS. Mitochondrial TCA cycle metabolites control physiology and disease. Nat. Commun. 2020;11:102. doi: 10.1038/s41467-019-13668-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Damiani C, et al. Integration of single-cell RNA-seq data into population models to characterize cancer metabolism. PLoS Comput. Biol. 2019;15:e1006733. doi: 10.1371/journal.pcbi.1006733. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Hartmann FJ, et al. Single-cell metabolic profiling of human cytotoxic T cells. Nat. Biotechnol. 2021;39:186–197. doi: 10.1038/s41587-020-0651-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Madonna MC, et al. Optical imaging of glucose uptake and mitochondrial membrane potential to characterize Her2 breast tumor metabolic phenotypes. Mol. Cancer Res. 2019;17:1545–1555. doi: 10.1158/1541-7786.MCR-18-0618. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Dumas SJ, et al. Single-cell RNA sequencing reveals renal endothelium heterogeneity and metabolic adaptation to water deprivation. J. Am. Soc. Nephrol. 2020;31:118–138. doi: 10.1681/ASN.2019080832. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Lewis SM, et al. Spatial omics and multiplexed imaging to explore cancer biology. Nat. Methods. 2021;18:997–1012. doi: 10.1038/s41592-021-01203-6. [DOI] [PubMed] [Google Scholar]
- 25.Yagnik G, Liu ZY, Rothschild KJ, Lim MJ. Highly multiplexed immunohistochemical MALDI-MS imaging of biomarkers in tissues. J. Am. Soc. Mass Spectr. 2021;32:977–988. doi: 10.1021/jasms.0c00473. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Patterson NH, Tuck M, Van de Plas R, Caprioli RM. Advanced registration and analysis of MALDI imaging mass spectrometry measurements through autofluorescence microscopy. Anal. Chem. 2018;90:12395–12403. doi: 10.1021/acs.analchem.8b02884. [DOI] [PubMed] [Google Scholar]
- 27.Soliman K. CellProfiler: novel automated image segmentation procedure for super-resolution microscopy. Biol. Proced. Online. 2015;17:11. doi: 10.1186/s12575-015-0023-9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Chen A, et al. Spatiotemporal transcriptomic atlas of mouse organogenesis using DNA nanoball-patterned arrays. Cell. 2022;185:1777–1792. doi: 10.1016/j.cell.2022.04.003. [DOI] [PubMed] [Google Scholar]
- 29.Stickels RR, et al. Highly sensitive spatial transcriptomics at near-cellular resolution with Slide-seqV2. Nat. Biotechnol. 2021;39:313–319. doi: 10.1038/s41587-020-0739-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Soltwisch J, et al. MALDI-2 on a trapped ion mobility quadrupole time-of-flight instrument for rapid mass spectrometry imaging and ion mobility separation of complex lipid profiles. Anal. Chem. 2020;92:8697–8703. doi: 10.1021/acs.analchem.0c01747. [DOI] [PubMed] [Google Scholar]
- 31.Andersen MK, et al. Spatial differentiation of metabolism in prostate cancer tissue by MALDI-TOF MSI. Cancer Metab. 2021;9:9. doi: 10.1186/s40170-021-00242-z. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Pippin JW, et al. Cells of renin lineage are progenitors of podocytes and parietal epithelial cells in experimental glomerular disease. Am. J. Pathol. 2013;183:542–557. doi: 10.1016/j.ajpath.2013.04.024. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.Wishart DS, et al. HMDB 4.0: the human metabolome database for 2018. Nucleic Acids Res. 2018;46:D608–D617. doi: 10.1093/nar/gkx1089. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34.Heinrich P, et al. Correcting for natural isotope abundance and tracer impurity in MS-, MS/MS- and high-resolution-multiple-tracer-data from stable isotope labeling experiments with IsoCorrectoR. Sci. Rep. 2018;8:17910. doi: 10.1038/s41598-018-36293-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Cao J, et al. The single-cell transcriptional landscape of mammalian organogenesis. Nature. 2019;566:496–502. doi: 10.1038/s41586-019-0969-x. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36.Wang, G. et al. Analyzing cell type-specific dynamics of metabolism on kidney. Protoc. Exch.10.21203/rs.3.pex-1912/v1 (2022). [DOI] [PMC free article] [PubMed]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
The exported and processed MSI data for this study were deposited in FigShare at 10.6084/m9.figshare.20227419.v1. Owing to the large size of all raw data, parts of the raw MSI data are deposited to provide the necessary information of 13C-labeled metabolites and spectrum quality. For full raw MALDI-MSI data related to this study, please contact G. W. (g.wang@lumc.nl) or B. H. (b.p.a.m.heijs@lumc.nl). Upon reasonable request, data will be made available. Source data are provided with this paper.
The codes used in this article is available in Github (https://github.com/Gangqiwang/scDYMO).