Multi-Modal Spatial Metabolomics with Mid-Infrared Microscope Guided Mass Spectrometry Imaging

Summary

We present a multi-modal spatial metabolomics workflow that integrates mass spectrometry imaging (MSI) and mid-infrared (MIR) imaging on the same tissue section for spatially resolved metabolic and biochemical analysis.

Abstract

Spatial metabolomics is a rapidly evolving field to map the distribution of metabolites within tissues, organs, and even single cells. This approach provides contextual metabolic information, which is critical for understanding the biochemical heterogeneity of biological systems. Mid-infrared (MIR) imaging and mass spectrometry imaging (MSI) have emerged as powerful approaches for spatial metabolomics, each offering unique and complementary advantages. In this study, we present a workflow for performing MIR imaging and matrix-assisted laser desorption ionization MSI (MALDI-MSI) on the same tissue section, encompassing experimental procedures, imaging co-registration, data integration, and bioinformatics analysis. MIR imaging is employed as the upstream modality, allowing for non-destructive biochemical analysis of tissues. Subsequently, a matrix is deposited in the tissue for MALDI-MSI, guided by the spatial information obtained from MIR imaging. Images are integrated following co-registration, enabling multi-modal spatial metabolomics analysis using advanced bioinformatics tools such as Seurat. The integration of MIR imaging and MSI represents a transformative advancement in spatial metabolomics, offering unprecedented opportunities to explore the spatial and chemical complexity of metabolic processes in both health and disease. This multi-modal approach holds significant potential for driving innovations in biomarker discovery, disease diagnostics, and therapeutic development.

Introduction

Spatial metabolomics involves mapping metabolites within their native tissue context by integrating mass spectrometry imaging (MSI), or other spatially resolved techniques, with metabolic analysis, offering insights into metabolism at sub-tissue, cellular, or subcellular resolution^{1,2,3,4,5,6,7,8}. Among these imaging modalities, mid-infrared (MIR) microscopy is gaining increasing attention for its ability to chemically map biomolecules based on molecular vibrational contrast^{4,7,9,10,11,12}. The non-destructive, label-free, high-speed, and cost-effective nature^{4,7,9,10,11,12} makes MIR imaging particularly well-suited for integration with MSI, forming a promising multi-modal approach for comprehensive spatial analysis.

Matrix-assisted laser desorption ionization MSI (MALDI-MSI) is one of the most widely used techniques in spatial metabolomics (Figure 1A)¹³. It enables the detection of a broad range of metabolites, including lipids, amino acids, sugars, nucleotides, and secondary metabolites, with high sensitivity¹. Advances in matrix selection (e.g., 2,5-dihydroxybenzoic acid [DHB], 9-aminoacridine [9AA], or N- (1-naphthyl) ethylenediamine dihydrochloride [NEDC]) and ionization techniques (e.g., post-ionization methods) have further expanded the detection coverage, particularly for low-abundance and poorly ionizable species¹⁴. Modern MALDI-MSI systems support imaging speeds typically ranging from 5 to 50 pixels per second, depending on spatial resolution and the type of mass analyzer used (e.g., time-of-flight [TOF], Orbitrap, or Fourier transform ion cyclotron resonance [FT-IRC]). Notably, high-speed MALDI-TOF instruments have achieved rates exceeding 100 pixels per second, enabling large-area imaging within a few hours¹⁵. The spatial resolution of MALDI-MSI generally falls within 5 to 50 µm, sufficient to resolve cellular and even subcellular structures^16,17. However, higher spatial resolution typically comes at the cost of longer acquisition times, reduced signal-to-noise ratio, and greater computational demands during data processing.

Unlike MSI, MIR imaging is a wide-field optical approach⁴, making it inherently non-destructive and high-throughput. Recent advancements in high-power quantum cascade lasers (QCLs) enable a new framework known as discrete frequency infrared (DFIR) imaging, which significantly increase the imaging speed up to video rate (Figure 1B)^12,18. Additionally, QCL sources have made uncooled microbolometer detectors viable for MIR imaging, enabling the use of larger focal plane array (FPA) detectors¹². As a result, a single-frequency MIR image over a 1 cm² area can be acquired with cellular resolution (~5 µm) in just a few seconds. MIR imaging can operate in both label-free and labeled modes. In previous studies^4,19,20, small IR-active vibrational probes have been introduced to monitor the kinetics of various metabolic activities in situ. These probes enable real-time tracking of metabolic pathways by reporting global chemical transformations through unique MIR spectral bands⁴. Thus, MIR imaging provides both static biochemical profiling and dynamic metabolic monitoring with high spatial resolution and rapid acquisition speed.

In this manuscript, we introduce a comprehensive workflow to perform MIR-guided MALDI-MSI spatial metabolomics, demonstrating the feasibility of acquiring both modalities from the same specimen (Figure 2, Figure 3). By integrating MIR imaging with MSI, we establish a multi-modal approach that leverages the complementary strengths of each technique. MIR imaging offers rapid, label-free biochemical mapping of the whole slide at cellular resolution, serving as a valuable guide for region-of-interest (ROI) selection in subsequent MSI acquisition. This strategy enables a more targeted and efficient MSI workflow, reducing acquisition time and cost while enhancing the interpretability and biological relevance of the spatial metabolomics data. In turn, MSI provides highly detailed molecular annotations of the MIR hyperspectral images, enabling a deeper understanding of tissue biochemistry. Overall, this integrated approach holds great potential for advancing biomarker discovery, disease diagnostics, and therapeutic development.

Protocol

The protocol was conducted in accordance with the animal experimental protocol (DSF-2024-004) approved by the Institutional Animal Care and Use Committee (IACUC) of the Laboratory Animal Resources Center at Fudan University.

NOTE: This study provides a workflow for performing MIR imaging and MALDI-MSI on the same tissue section. This integrated approach encompasses experimental procedures, image co-registration, data fusion, and downstream bioinformatics analysis (Figure 1C). MIR imaging is utilized as the upstream modality, enabling non-destructive, label-free biochemical mapping of the tissue. Guided by the spatial information from MIR imaging, a matrix is subsequently applied for MALDI-MSI acquisition. Following precise co-registration, the datasets are integrated for multivariate co-analysis (Figure 4), enabling comprehensive spatial metabolomics through advanced computational tools.

1. Sample preparation

1. Tissue sample preparation
   1. Mouse tissue collection
      1. House C57BL/6J mice (20 weeks old, 40 g) under specific pathogen-free (SPF) conditions, maintained on a 12-h light/dark cycle, and provided with ad libitum access to food and water to ensure optimal health before the experiment.
      2. Anesthetize the mouse by intraperitoneal injection of pentobarbital sodium at a dose of 50 mg/kg body weight to ensure deep anesthesia. Monitor the animal and confirm the absence of pedal and corneal reflexes before proceeding with perfusion.
      3. Perfuse approximately 20–30 mL of cold PBS containing heparin (5 IU/mL) through the left ventricle at a controlled pressure of 150 mmHg for 6 min to flush out the blood and exsanguinate the organ before removal.
      4. Immediately quench all collected tissues with liquid nitrogen and store them at -80 °C for further analysis.
         NOTE: Rapid freezing is essential to prevent ice crystal formation, which can distort tissue morphology.
   2. Tissue embedding and cryo-sectioning
      1. Set the microtome to -20 °C and decontaminate the chamber and blade with ethanol (EtOH) to eliminate residual embedding medium. Allow the CaF₂ slides, slide mailer, specimen holder, and tissue sample to equilibrate within the cryostat.
      2. Trim the tissue to approximately 5 mm × 5 mm × 3 mm. Embed it in 10% gelatin within a standard cryomold at -20 °C, ensuring proper orientation for sectioning. Snap-freeze the embedded sample on dry ice and store at -80 °C until cryosection.
      3. Extract the tissue from its mold, trim excess material to optimize sectioning, and affix it to the specimen holder using ultrapure water to ensure stable adhesion.
      4. Cut 10-µm-thick sections and immobilize the sections onto pre-coated CaF₂ substrates and store them at -80 °C.
      5. Place the mounted tissue sections in a vacuum freeze-dryer prior to mid-infrared imaging. Set the chamber pressure to 0.1 mPa and the temperature to -100 °C. Dry the samples for 30 min.
2. Cell sample preparation
   1. Seed 1.25 × 10⁵ HeLa cells onto 13 mm diameter circular CaF₂ substrates placed in standard culture plates. Use high-glucose Dulbecco's Modified Eagle Medium (DMEM) supplemented with 10% fetal bovine serum (FBS) and 1% penicillin-streptomycin. Incubate the cells under standard culture conditions (37 °C, 5% CO₂) for 24 h.
   2. After 24 h, carefully remove the media and fix the cells by incubating them in 4% paraformaldehyde (PFA) dissolved in PBS buffer for 20 min at room temperature (RT).
   3. Rinse the fixed cells three times with HBSS buffer, then wash them three additional times with ddH₂O.
   4. Air-dry the CaF₂ substrates completely at RT before proceeding to MIR imaging.

2. Multi-modal image acquisition

1. Quantum cascade laser (QCL) MIR image acquisition
   1. Perform QCL MIR imaging microscopy using a QCL-MIR microscope equipped with a non-N₂-liquid cooled microbolometer 520 × 480 focal plane array (FPA) detector and spatial coherence reduction technology.
   2. Select spectral range from 950 cm^-1 to 1800 cm^-1.
   3. Use a 4× objective with a 0.6 numerical aperture (NA), resulting in a nominal pixel size of 4.25 µm.
   4. Collect a background spectrum on a clean CaF₂ substrate at a 4 cm^-1 spectral resolution.
   5. Select the acquisition area based on tissue morphology and collect spectral data at a spectral resolution of 4 cm^-1. At this spectral resolution, the image acquisition speed is approximately 0.3 mm²/s.
      NOTE: All measurements are performed in transmission mode using CaF₂ slides, with manual focus adjustment.
2. Mass spectrometry image acquisition
   1. Matrix application
      1. Use the post-MIR tissue sections for MALDI-MSI, apply fiducial markers around the tissue using a Tippex pen, and capture a pre-MSI optical image with a flatbed scanner for image registration.
      2. Prepare NEDC matrix at 7 mg/mL in methanol/acetonitrile/deionized water (V/V/V=75:25:5), then filter through a 0.2 µm membrane to remove particulates.
      3. Load the marked tissue slides into an automated matrix sprayer. Dispense the NEDC matrix onto the sample using the predefined parameters.
   2. MALDI-MSI measurement
      1. Deposit 0.2 µL of red phosphorus suspension onto the CaF₂ slide to serve as an external mass calibration standard.
      2. Secure the matrix-coated slide in the slide adapter, ensuring optimal electrical contact with the holder. Load the plate into a MALDI-MSI system.
      3. Align the optical image of the slide with the XY-motor coordinates of the MALDI stage using fiducial markers. Define ROIs for tissue analysis and include a "matrix-only" area as a negative control.
      4. Configure the instrument in negative-ion mode with a pixel size of 20 × 20 µm², covering an m/z range of 50-1000. Perform mass calibration using the red phosphorus spot before initiating data acquisition. Perform data acquisition using a commercial data analysis platform.

3. Image data preprocessing and analysis

NOTE: The MATLAB custom codes are uploaded as Supplementary File 1.

1. QCL-MIR data preprocessing and analysis
   1. Data preprocessing
      1. Export hyperspectral data in an appropriate format for further analysis.
      2. Perform data preprocessing using CytoSpec and MATLAB, following these steps 3.1.1.3-3.1.1.8
      3. Import the MIR spectral data into CytoSpec v2.00.07 for preprocessing.
      4. Evaluate spectral quality using the built-in quality tests module in CytoSpec. Select criterion 2 (sample thickness) to assess sample consistency. Set the spectral integration range to 950-1800 cm^-1. Define the absorbance thresholds between 50 and 1000 units to exclude spectra from regions that are excessively thin or thick.
      5. Reduce noise using PCA-based filtering in CytoSpec. Retain the first 10 principal components (PCs) to eliminate high-frequency noise while preserving key biochemical variance.
      6. Export the preprocessed spectral data from CytoSpec in .mat format for further analysis in MATLAB.
      7. Perform baseline correction in MATLAB using a rubber band correction algorithm across the 950-1800 cm^-1 spectral range to eliminate baseline drift and improve spectral accuracy.
      8. Apply spectral normalization using min-max scaling over the same spectral range to standardize intensities across all spectra and ensure comparability throughout the dataset.
   2. Data analysis
      1. Identify protein-associated features at 1,550 cm^-1 and 1,650 cm^-1, corresponding to amide II and amide I bands, respectively. Identify lipid-associated features at 1,740 cm^-1, representing ester C=O stretching vibration.
      2. Calculate the intensity ratio of the Amide I band (1650 cm^-1) to the ester C=O stretch (1740 cm^-1) and generate ratio maps to visualize spatial lipid-to-protein distribution.
      3. Unsupervised cluster analysis: Perform hierarchical clustering analysis (HCA) on hyperspectral data using Euclidean distance and Ward's linkage to classify regions based on spectral similarity. Generate cluster maps to identify biochemically distinct regions and reveal structural or pathological differences.
2. MSI data preprocessing and analysis
   1. Import the raw MSI data into a commercial data analysis platform.
   2. Perform peak picking with relative intensity > 0.01%.
   3. Exclude matrix-derived peaks from the experimental peak list by comparing it to the spectrum obtained from a matrix-only control region. Remove any peaks that appear in the control spectrum and have a signal-to-noise ratio (SNR) less than 3.
   4. Normalize the dataset to Root Mean Square (RMS) to account for variability in ion signal intensity across pixels and samples.
   5. Refine metabolite identification by filtering features based on their tissue distribution, match with entries in the Human Metabolome Database (HMDB), and agreement with theoretical m/z values.
   6. Select relevant metabolites based on experimental goals and tissue context.
   7. Export the intensities of the selected features for all pixels as a CSV file.
   8. Apply dimensionality reduction techniques (e.g., PCA, t-SNE, or UMAP) to reduce data complexity and identify clusters of pixels with similar metabolic signatures using bioinformatics tools such as Seurat and Banksy, facilitating the visualization of spatial metabolic heterogeneity.

4. Multi-modal image co-registration and co-analysis

1. Select the MIR image acquired at 1650 cm^-1 as the spatial reference for co-registration. This specific wavenumber highlights the amide I protein band, offering strong tissue contrast.
2. Normalize the intensity values of both the MIR and MSI images to enhance visual comparability. Apply contrast enhancement techniques, such as histogram equalization using histeq function in MATLAB, to redistribute image intensities and improve the visibility of structural features critical for landmark selection.
3. Load both the MIR and MSI images into MATLAB using the cpselect function from the Image Processing Toolbox. This function opens an interactive interface for landmark selection.
4. Manually annotate five to ten anatomically relevant and well-distributed control points on both the MIR and MSI images using cpselect function. These landmarks should correspond to distinct morphological features such as tissue boundaries.
   NOTE: Distribute control points evenly across the image field to minimize localized distortion.
5. Export and record the coordinates of the annotated control points from both images. Store them as two matching arrays for transformation computation (e.g., movingPoints and fixedPoints).
6. Compute an affine transformation matrix using the MATLAB function fitgeotrans in Affine mode. This transformation accounts for scaling, rotation, translation, and shearing.
7. Apply the computed transformation to the MSI image using the imwarp function in MATLAB. This step generates a spatially registered MSI image that is geometrically aligned to the MIR reference frame.
8. Visually inspect the alignment using the imshowpair function in blend mode to overlay the MSI and MIR images and confirm correct registration of structural features.
9. Convert both the registered MSI image and the reference MIR image into binary masks using imbinarize function in MATLAB.
10. Compute the Dice Similarity Coefficient (DSC) to quantify the spatial overlap between the two binary masks. The DSC is defined as: DSC = 2 × (Area of Overlap)/(Total Area of Both Masks). A DSC value approaching 1.0 indicates high spatial overlap and registration accuracy.
    NOTE: If necessary, refine the registration by adjusting the number or placement of control points and repeating the transformation process. Consider non-rigid registration methods if local misalignments persist.
11. Save the registered image, transformation parameters, and landmark coordinates for downstream multi-modal image analysis, such as spatial correlation of biochemical and structural features.

Results

To validate the above described multi-modal spatial metabolomics strategy, we implemented it in a mouse model of cancer. Tumor cells were genetically modified to express green fluorescent protein (GFP), enabling precise localization of the tumor region through fluorescence imaging. Figure 2A shows label-free QCL-MIR imaging results with a spectral coverage of 950-1800 cm^-1, known as "fingerprint region". Representative MIR bands are marked in Figure 2A

Discussion

Here, we present a multi-modal spatial metabolomics strategy that integrates MSI and MIR imaging to enable high-throughput, molecularly specific metabolic profiling on the same specimens. MIR imaging enables label-free, non-destructive chemical mapping based on molecular vibrations, providing detailed structural and functional group information. In contrast, MSI delivers high sensitivity and specificity for metabolite identification and quantification, establishing it as a cornerstone of metabolomics research. By integra...

Materials

List of materials used in this article

Name	Company	Catalog Number	Comments
4% Paraformaldehyde, tissue fixation solution	Sangon Biotech	E672002
Bovine serum albumin	Sigma Aldrich	126615
C57BL/6JNifdc mice	Charles River	219
CaF2 slides	Crystran	CAFP25-1
Cryostat microtome	Dakewe Biotech	CT520
Cytospec	Peter Lasch	Version 2.10.01	Software for MIR data analysis, including baseline correction, normalization, image generation, and hyperspectral analysis.
DPBS buffer	Fisher Scientific	14-190-250
Fetal bovine serum	Gibco	10270106
Freeze dryers	LABCONCO	FreeZone
Gelatin	Sigma-Aldrich	G1890	Type A, from porcine skin
HBSS buffer	Sangon Biotech	A003210
High-glucose DMEM	ThermoFisher Scientific	10564011
Matlab	MathWorks	Matlab R2024a	Software for MIR data analysis, including baseline correction, normalization, image generation, and hyperspectral analysis.
Matrix-assisted laser desorption ionization mass spectrometry imaging (MALDI-MSI) system	Bruker Daltonics GmbH	timsTOF fleX MALDI-2
Methanol	Sinopharm Chemical Reagent	10014108
mMass	Institute of Entomology, Biology Centre of the Czech Academy of Sciences, Czech Republic	Version 5.5.0	Softwares for MSI data analysis. Import the raw MSI data into SCILS Lab for preprocessing.  Perform peak picking in mMass with relative intensity > 0.01%.
N-(1-naphthyl) ethylenediamine dihydrochloride (NEDC) matrix	Sigma Aldrich	222488
Penicillin-streptomycin	ThermoFisher Scientific	15140122
Quantum Cascade Laser  mid-Infrared (QCL-MIR) microscopy	Bruker Optics GmbH	LUMOS II ILIM
Red phosphorus	Sigma-Aldrich	8.0727
SCiLS Lab	Bruker Daltonics	SCiLS Lab 2024b	Softwares for MSI data analysis. Import the raw MSI data into SCILS Lab for preprocessing.  Perform peak picking in mMass with relative intensity > 0.01%.
Sprayer	HTX Technologies	TM
Trypsin, 0.25% EDTA	ThermoFisher Scientific	C25200072