Advances in mass spectrometry imaging for spatial cancer metabolomics

Abstract

Mass spectrometry (MS) has become a central technique in cancer research. The ability to analyze various types of biomolecules in complex biological matrices makes it well suited for understanding biochemical alterations associated with disease progression. Different biological samples, including serum, urine, saliva, and tissues have been successfully analyzed using mass spectrometry. In particular, spatial metabolomics using MS imaging (MSI) allows the direct visualization of metabolite distributions in tissues, thus enabling in-depth understanding of cancer-associated biochemical changes within specific structures. In recent years, MSI studies have been increasingly used to uncover metabolic reprogramming associated with cancer development, enabling the discovery of key biomarkers with potential for cancer diagnostics. In this review, we aim to cover the basic principles of MSI experiments for the nonspecialists, including fundamentals, the sample preparation process, the evolution of the mass spectrometry techniques used, and data analysis strategies. We also review MSI advances associated with cancer research in the last 5 years, including spatial lipidomics and glycomics, the adoption of three-dimensional and multimodal imaging MSI approaches, and the implementation of artificial intelligence/machine learning in MSI-based cancer studies. The adoption of MSI in clinical research and for single-cell metabolomics is also discussed. Spatially resolved studies on other small molecule metabolites such as amino acids, polyamines, and nucleotides/nucleosides will not be discussed in the context.

1 |. CURRENT MASS SPECTROMETRY (MS) METHODS IN CANCER RESEARCH

MS plays a crucial role in current studies of cancer biology, biomarker discovery efforts, cancer screening, and drug discovery. Advancement in various “omics” studies using MS has enabled a more in-depth and comprehensive understanding of cancer progression, providing solid molecular foundations for the development of better drugs and therapies. Additionally, diagnosis of cancers, especially early-stage diagnosis, remains one of the major challenges in cancer prevention. Discovery and identification of cancer biomarkers using MS-based techniques has yielded numerous biomarker candidates. Emerging advances in MS imaging techniques now allow visualization of the metabolic alterations associated with cancer directly in tissues, even in single cells, thus enabling pathway enrichment analysis in situ during cancer progression.

1.1 |. Nonimaging MS methods

Several MS techniques are nowadays commonplace in cancer research. One of earliest, gas chromatography-MS (GC-MS) has been extensively applied to biomarker analysis for many types of cancers, including prostate cancer (Lima et al., 2018a, 2018b; W. Wang, Lee, et al., 2021), breast cancer (Phillips et al., 2018; Tan et al., 2020), and several other cancer types (Barberini et al., 2019; Dai et al., 2020; Fang et al., 2022; Hirata et al., 2017). However, although being a simple, inexpensive, and mature analytical technique with high sensitivity and low limits of detection (LOD) (Gao & Lee, 2019; Lubes & Goodarzi, 2018), the major limitation of conventional GC-MS lies in its chemical coverage, as only volatile, thermally stable compounds with low boiling points can be separated and analyzed (Lubes & Goodarzi, 2018). Typically, only compounds with molecular weights (MW) below 1000 Da are amenable to GC-MS (Lubes & Goodarzi, 2018). Nevertheless, analysis of larger, nonvolatile biomolecules is many times not possible, even with derivatization approaches.

On the other hand, liquid chromatography-MS (LC-MS) and capillary electrophoresis-MS (CE-MS) are free from limitations concerning analyte volatility, MW, and thermal stability (Dai et al., 2020). Consequently, LC-MS and CE-MS methods have been more prevalent in cancer studies (Bian et al., 2018; A. Chen et al., 2021; Frantzi et al., 2019; Igarashi et al., 2021; Y. Liu et al., 2018; Łuczykowski et al., 2021; van Winden et al., 2021; M. Zhou et al., 2021). LC-MS and LC-MS/MS are the key techniques behind protein biomarker studies using proteomic approaches and the majority of metabolomic studies (An et al., 2019; Chang et al., 2017; Kuzyk et al., 2021). The ease of sample preparation for LC-MS makes this technique preferable over GC-MS (Lubes & Goodarzi, 2018).

One common trait of the techniques mentioned above is that they do not provide direct visualization of the distribution of the molecules in a variety of samples with any sort of spatial resolution, unless combined with spatially resolved sampling such as laser capture micro-dissection (LCM) (Hondius et al., 2018). Therefore, MS imaging (MSI) techniques were developed to fill this niche, becoming valuable tools for biological sample analysis in a spatially resolved fashion over the last few decades.

1.2 |. MSI methods: Overview

With the application of matrix-assisted laser desorption/ionization (MALDI) to imaging peptides and proteins in biological samples in the late 1990s (Caprioli et al., 1997), and the introduction of desorption electrospray ionization (DESI) in the early 2000s (Wiseman et al., 2006), MSI emerged as a powerful analytical approach. As a broadband label-free imaging approach, MSI allows spatial visualization of a variety of biological molecules, including proteins (Garza et al., 2018; Keener et al., 2021; H. Liu et al., 2021; Piehowski et al., 2020), peptides (Kakuda et al., 2017; Kaya et al., 2017), lipids (Anderson et al., 2020; Claes et al., 2021; Kaya et al., 2017; Unsihuay et al., 2020), polysaccharides (glycans) (Arnaud et al., 2020; Heijs, Holst-Bernal, et al., 2020; Heijs, Potthoff, et al., 2020; McDowell, Klamer, et al., 2021; McDowell, Lu, et al., 2021), amino acids (Esteve et al., 2016; J. He et al., 2019), and oligonucleotides (Nakashima & Setou, 2018; Yokoi et al., 2018) with softer ionization than preceding techniques such as secondary ion MS (SIMS).

MSI provides direct spatial distribution information for molecules of interests on a thin tissue section, displaying its molecular heterogeneity. Currently, most MSI techniques using MALDI or DESI for ion generation are able to produce molecular images at a spatial resolution (defined by the effective pixel size) ranging between 10 and 200 μm, depending on the microprobe (i.e., laser, liquid jet) used (Buchberger et al., 2018). There are also more recent reports of MSI methods achieving <10 μm spatial resolution through atmospheric-pressure (AP) MALDI and nanoDESI techniques (Kompauer et al., 2017; Yin et al., 2019). The development of MALDI postionization (MALDI-2) techniques has pushed the spatial resolution of MSI experiments down to 0.6 μm (600 nm) (Niehaus et al., 2019). Coupling of various MSI ion generation techniques to high-resolution Orbitraps and Fourier-transform ion cyclotron resonance (FTICR) mass spectrometers allows generating data sets with the highest mass accuracy, leading to more accurate compound annotations. Artificial intelligence approaches such as machine learning and deep learning have also been recently implemented in MSI workflows for better data processing and interpretation (Alexandrov, 2020).

Despite challenges associated with generating reproducible data for long-term studies, ion suppression and visualization and analysis of sophisticated multi-dimensional data sets (Buchberger et al., 2018; Chughtai & Heeren, 2010), MSI has been extensively applied to numerous biological and clinical studies due to its high spatial resolution, comprehensive coverage of different classes of biological molecules and its ability to directly map molecular alterations in tissues without labels (J. Zhang et al., 2021). In particular, the development of MSI techniques has promoted spatial metabolomics and proteomics studies for different types of cancers in the past 10 years (Arentz et al., 2017). As shown in Figure 1, publications in the field of cancer research using MSI have grown almost exponentially since 2010. Therefore, the goal of this review is to discuss the basic principles and technical aspects of MSI experiments, outline its applications in cancer spatial metabolomics and describe key advancement of MSI techniques since 2017, with an emphasis on the more popular MALDI MSI methods.

FIGURE 1.

Number of publications reported by Google Scholar since 2010 with the keywords “mass spectrometry imaging” and “cancer.”

2 |. BASIC PRINCIPLES OF MSI

MSI is one of the major techniques used to conduct spatially resolved metabolomics studies. MALDI, DESI and secondary ion MS (SIMS) (Chughtai & Heeren, 2010) are the three major ionization methods used in MSI experiments, each of which has its own unique strengths (Figure 2 [Aoki et al., 2016; Sturtevant et al., 2016; Takáts et al., 2004] and Table 1, see Sections 2.3 and 2.4 for details). In addition to the more popular methods, other types of promising ionization techniques, including MALDI postionization (MALDI-2) (Niehaus et al., 2019; Soltwisch et al., 2015), matrix-assisted laser desorption electrospray ionization (MALDESI) (Robichaud et al., 2013), and laser ablation electrospray ionization (LAESI) (Nemes et al., 2010) have been developed, finding unique applications in the field of MSI (see Section 2.4.3). Introduction of ion mobility spectrometry (IMS) has further enabled the ability to distinguish isomeric compounds with different collisional cross sections (CCS) (see Section 2.4.4) (McLean et al., 2007; Sans et al., 2018). In this section, basic MSI concepts as related to spatial metabolomics, will be discussed, together with a basic description of the above-mentioned techniques. Mass analyzers commonly used in MSI experiments will also be introduced. Additionally, we also discuss MSI data processing and other tasks commonly encountered during MSI experiments.

FIGURE 2.

Schematic illustrations of the ionization steps in (A) MALDI, (B) DESI, and (C) SIMS MSI experiments. Adjusted and reprinted with permissions from Elsevier (A: D. Sturtevant et al., 2016 and C: D. Aoki et al., 2016) and the American Association for the Advancement of Science (B: Z. Takáts et al., 2004).

TABLE 1.

Figures of Merit for MALDI, DESI, and SIMS MSI

	MALDI	DESI	SIMS
Required sample preparation (after tissue embedding and sectioning)	Matrix deposition	No	Freeze fracture and drying for subcellular imaging (Chandra, 2008)
Ionization conditions	Atmospheric/medium/high vacuum	Open air/ambient	Ultrahigh vacuum
Tissue conservation methods	Fresh frozen/formalin-fixed paraffin embedding	Fresh frozen	Fresh frozen
Typical spatial resolution	5–200 μm (Buchberger et al., 2018)	50–200 μm (Qi et al., 2021) (<10 μm for nanoDESI) (Yin et al., 2019)	0.05–100 μm (Anderton & Gamble, 2016) (subcellular level imaging possible)
Compound coverage	Small molecule metabolites, lipids, peptides, and proteins (<50 kDa) (Iakab et al., 2021)	Small molecule metabolites, lipids, peptides, and proteins (<50 kDa) (Iakab et al., 2021)	Chemical elements, small molecule metabolites, and lipids, fragment ions (<1000 Da) (Iakab et al., 2021)
Destructive/conservative nature	Minimally destructive (Francese et al., 2013)	Minimally destructive (Francese et al., 2013)	Destructive (Francese et al., 2013)

2.1 |. Introduction to spatial metabolomics

Spatial metabolomics involves the investigation of the spatial distributions and alterations of metabolites (mainly molecules with MW < 2000 Da, such as lipids, amino acids, and sugars) in tissues, organs and cells (Alexandrov, 2020; Lundberg & Borner, 2019; Petras et al., 2017). Such studies help visualize and pinpoint the locations of these biomolecules, providing an extra dimension to understand biological processes of interest such as disease progression mechanisms. Generally, information obtained from spatial metabolomics experiments is tightly associated with the functions of biomolecules present in certain tissue regions or sub-structures, organs, and cells. The changes in the tissue or appearance of abnormal substructures (e.g., heterogeneity of the tissue) observed by other spatial techniques can be connected to the molecular-level alterations of these compounds. In cancer research, the spatial information and alterations of biomolecule abundances reflect cancer progression (M. J. Taylor et al., 2021). Spatially resolved information can also be correlated to other studies such as time-resolved protein, peptide, metabolite and lipid abundances in biofluids measured via GC-, LC- CE-MS, and NMR (Lane et al., 2020; Lu et al., 2020; Sah et al., 2022). Therefore, MSI has gained popularity in cancer research (Figure 1), becoming a valuable addition to other experimental approaches.

2.2 |. Sample preparation for MSI

Normally, MSI experiments are carried out by probing the surface of a thin section of the animal or plant tissue under investigation, usually with a thickness of tens of microns (Figure 3). Therefore, the first step in an MSI experimental workflow is to obtain a high-quality tissue section, as the condition and the uniformity of these sections highly determine the quality and reproducibility of MSI data (Goodwin, 2012).

FIGURE 3.

Schematic drawing of sample preparation process for MSI experiments, created with license obtained from BioRender.com (agreement number ZW23CMI9GA). MSI, mass spectrometry imaging.

Many factors influence the quality of MSI data, including biometric characteristics of the animal or plant, tissue collection time, and anesthetic techniques used for animal sacrifice (Goodwin, 2012). Sample preparation is also a major source of variance in MSI studies (Balluff et al., 2021). Following tissue collection, tissue samples are usually stabilized and preserved at a low temperature (–80°C) via two major approaches: formalin-fixed paraffin embedding or fresh frozen in dry ice. The fresh frozen approach is a modified snap-freeze method. Snap freezing suffers from tissue cracking and fragmentation caused by differences in freezing rates in liquid nitrogen (–196°C), which may lead to poor-quality MSI measurements (Goodwin, Nilsson, et al., 2012). As an alternative approach to avoid tissue cracking, a modified fresh frozen method utilizes a solution containing ethanol and isopentane in dry ice as the bath for freezing, leading to more homogeneous freezing environment. Before freezing, the tissue is embedded in an aqueous carboxymethylcellulose and gelatin solution. The embedded tissue is then frozen, and can then be stored at −80°C for at least a year (Patel, 2017).

Formalin-fixed paraffin embedding (FFPE) is an alternative technique for tissue stabilization, where formalin and paraffin are used as the embedding media (formalin is used for tissue fixation and conservation, and paraffin is used for tissue embedding). Formalin induces crosslinking of molecules present in the sample. Crosslinking is beneficial for imaging some classes of analytes such as N-glycans (Angel, Mehta, et al., 2018), lipids (Neef et al., 2020), and some small-molecule metabolites (Buck et al., 2015), but the covalent crosslinking of proteins during FFPE leads to poor sensitivity for protein analysis (Goodwin, 2012). Molecular diffusion during the fixation process may also cause problems with small molecule metabolite analysis (Goodwin, 2012). Protocols developed for the extraction of metabolites, lipids, and glycans from FFPE fixed tissues before MSI have enabled its broader application to large sets of biobanked samples, also allowing the coregistration of histology and MS images (Gorzolka & Walch, 2014).

Regardless of the method chosen, tissue embedding is typically followed by tissue sectioning with a cryostat or a microtome. Folding, cracking, and displacement of the tissues should be carefully avoided during tissue sectioning, as it can partially prevent analysis. Several strategies have been reported to robustly produce high-quality tissue sections for MSI experiments (Nelson et al., 2013; Shimma & Sugiura, 2014).

Tissue washing and derivatization of the molecules in the tissue are sometimes necessary steps in MSI workflows that allow removing interferences or increasing sensitivity for specific chemical species. For example, washing with organic solvents (e.g., 70%–100% ethanol; Chaurand et al., 2006; Schwartz et al., 2003) is used when imaging larger molecules to prevent ionization suppression caused by salts and lipids (Goodwin, 2012). MSI of most small molecules such as lipids normally does not require any washing steps before matrix deposition. However, washing is sometimes still applied to remove high-abundance species, so that molecules with lower abundances in the tissue can be ionized and detected more efficiently (Vu et al., 2021). Removal of large molecules such as proteins, also helps avoid unwanted interactions with small molecule metabolites.

On-tissue enzymatic digestion is commonly used for MSI of larger molecules (usually proteins with MW > 30 kDa) (Cole et al., 2011; Goodwin, 2012; Groseclose et al., 2007). With enzymatic digestion, large proteins can be converted to peptides that can be more easily desorbed and ionized, improving sensitivity. A number of other on-tissue chemical derivatization methods have been described for MALDI MSI metabolomics studies (Harkin et al., 2021). For specific classes of small molecules, such as glycopeptides, fatty acids (FA), and N-glycans, targeted derivatization protocols have been developed (Holst et al., 2016; X. Liu et al., 2010; Q. Wu et al., 2016; H. Zhang et al., 2020). More recently, on-tissue chemical derivatization has been applied to locate C=C bonds in lipid isomers (Feng et al., 2019; G. Li et al., 2022; Lu et al., 2021; H. Zhang et al., 2021).

2.3 |. MALDI MSI

2.3.1 |. Basic principles of MALDI

A solution containing the matrix molecule is air-brushed, sprayed, or sublimated onto a tissue section and cocrystallized with the analyte molecules on its surface (Gemperline et al., 2014). Different matrix deposition techniques affect image quality to various degrees, as discussed in the next section.

During MALDI, the excess of matrix cocrystals absorbs the majority of the energy from the laser, becoming excited and vaporizing into the gas phase, while carrying with them the analytes (Karas & Krüger, 2003). Although the full extent of the MALDI mechanism remains unclear, one of the main MALDI ionization processes involves gas-phase proton transfer between the analyte and the matrix molecules (Karas & Krüger, 2003). Therefore, protonated molecules are the dominant ions observed in positive mode experiments, although sometimes sodiated and potassiated molecules ([M + Na]⁺ and [M + K]⁺) are also observed (Karas & Krüger, 2003). In negative ion mode experiments, deprotonated molecules ([M – H]^–) are usually generated via gas-phase proton transfer. Usually, the amount of matrix deposited is in large excess with respect to the analyte to ensure high desorption and ionization efficiency.

2.3.2 |. MALDI matrix selection

It is well established that the MALDI matrix plays a crucial role in the ionization process. Matrix selection depends on the goal of the experiments, as different classes of biomolecules are best ionized by different matrices. The ionization efficiency for a given compound class strongly varies with the matrix of choice. Sometimes, a mixture of two or several matrix compounds is used for enhanced ionization of specific compound classes, for example, lipids and oligosaccharides (Feenstra et al., 2016; Guran et al., 2017; Noh et al., 2019; Schröter et al., 2018; Shanta et al., 2011). MALDI MSI experiments using matrix mixtures are commonly performed in dual polarity settings, so molecules that can be ionized in either positive or negative ion modes can be studied simultaneously. Exploration of new types of matrices such as nanoparticle matrices for MALDI MSI experiments remains an active research topic (Guan et al., 2018; H. Zhao, Li, et al., 2018). For instance, a binary matrix system consisting of sinapinic acid or 2-nitrophloroglucinol together with silica nanoparticles enabled the generation of higher charge-state proteins in MALDI MSI experiments, leading to higher compound coverage (Banstola & Murray, 2021). Reactive matrices can act as both the ionizing reagent and the derivatization reagent, and are used to modify the structures of biomolecules for structural elucidation purposes, such as Paternò–Büchi matrices for identification of C=C bond positional isomers of lipids (Ma & Xia, 2014; Wäldchen et al., 2019; Wäldchen et al., 2020). Table 2 showcases a number of matrices and binary matrix mixtures of interest used in MALDI MSI experiments.

TABLE 2.

Matrix molecules and binary matrix systems used in different MALDI MSI experiments

Matrix	Ion polarity	MSI applications
2,5-Dihydroxybenzoic acid (DHB)	Positive/negative	Lipids (Perry et al., 2020) (Enhanced signal with salt doping) (Dufresne et al., 2019) Polysaccharides (Arnaud et al., 2020) Peptides (Au-Hanrieder et al., 2012) Amino acids (Esteve et al., 2016)
α-Cyano-4-hydroxycinnamic acid (CHCA)	Positive	N-glycans (Heijs, Holst-Bernal, et al., 2020; McDowell, Klamer, et al., 2021) Proteins (Angel et al., 2019; Angel, Comte-Walters, et al., 2018) Peptides (Angel, Mehta, et al., 2018; Touve et al., 2019) (Enhanced signal with salt doping) (Dufresne et al., 2021) Lipids (Fu et al., 2020)
Sinapinic acid (SA)	Positive	Peptides (Kakuda et al., 2017) Proteins (Angel et al., 2017)
1,5-Diaminonaphthalene (DAN)	Positive/negative	Lipids (Anderson et al., 2020) Peptides (Kaya et al., 2017)
9-Aminoacridine (9-AA)	Positive/negative	Lipids (Cerruti et al., 2012; Perry et al., 2020)
Norharmane	Positive/negative	Lipids (McMillen et al., 2020) N-glycans (Heijs, Potthoff, et al., 2020)
2-Mercaptobenzothiazole (MBT)	Positive/negative	Lipids (Garate et al., 2015; Perry et al., 2020; Thomas et al., 2012)
2,5-Dihydroxyacetophenone (2,5-DHA)	Positive/negative	Lipids (Bien et al., 2020; Claes et al., 2021) Proteins (Kaya et al., 2020; Zavalin et al., 2015)
2,6-Dihydroxyacetophenone (2,6-DHA)	Negative	Lipids (Jackson et al., 2018)
3-Hydroxypicolinic acid (3-HPA)	Negative	Oligonucleotides (Yokoi et al., 2018) Lipids (Thomas et al., 2012)
Caffeic acid (CA)	Positive	Proteins (H. Liu et al., 2021)
2′,4′,6′-Trihydroxyacetophenone (THAP)	Positive/negative	Lipids (Basu et al., 2019) Proteins (alkylated THAP used) (Fukuyama et al., 2016)
1,8-Bis (dimethylamino) naphthalene (DMAN)	Negative	Metabolites (Ye et al., 2013)
1,8-Di (piperidinyl)-naphthalene (DPN)	Negative	Lipids (Weißflog & Svatoš, 2016)
1,8,9-Anthracentriol (DIT)	Positive/negative	Lipids (Thomas et al., 2012)
Quercetin	Positive/negative	Lipids (X. Wang et al., 2014)
N-(1-naphthyl)-ethylenediamine dihydrochloride (NEDC)	Negative	Lipids (J. Wang et al., 2015; C. Zhao, Li, et al., 2018)
4-Phenyl-α-cyanocinnamic acid amide	Negative	Lipids (Fülöp et al., 2013)
2-Nitrophloroglucinol (2-NPG)	Positive	Proteins (Banstola & Murray, 2021)
Silver nanoparticles (AgNPs)	Positive	Lipids (Guan et al., 2018)
N- and P-doped graphene	Positive/negative	Metabolites (H. Zhao, Li, et al., 2018)
Graphene oxide (GO)	Negative	Lipids (D. Zhou et al., 2017)
DHB + SA	Positive	Proteins (Guran et al., 2017)
DHB + CHCA	Positive/negative	Lipids (Noh et al., 2019; Shanta et al., 2011)
DHB + DHA	Positive/negative	Lipids (Schröter et al., 2018)
DHB + Fe3O4	Positive/negative	Lipids (Feenstra et al., 2016)
Caffeic acid + Graphene oxide	Positive	Metabolites (T. Wang, Lee, et al., 2021)

Note that modifications of the matrix mixture by addition of doping reagents such as salts can promote ionization efficiency of the compounds of interest, thus leading to enhanced sensitivity. For instance, pre-coating of the tissue sections with salt solutions has led to an increased number of the sodium and/or potassium adducts of phospholipids and neutral lipids, further increasing lipid coverage (Dufresne et al., 2021, 2019). The choice of matrix deposition technique has a critical impact on MS image quality and MSI reproducibility (Goodwin, 2012). It is well known that the physicochemical properties of the MALDI matrix compound influence the quality of the deposited matrix layer, as different compounds crystallize at different rates yielding different crystal morphologies, even using the same deposition parameters. Most used matrix deposition techniques involve automated spraying or sublimation. To achieve homogeneous matrix deposition, parameters including spray solvent flow rate, drying gas flow rate, nozzle velocity, temperature (spraying nozzle temperature or sublimation chamber temperature), and humidity must be carefully optimized and controlled. Another experimental aspect to be considered is the amount of water in the matrix solution during automatic spraying, as it leads to diffusion of the analytes on the surface. Sublimation, a dry matrix deposition technique, has the advantage of avoiding these diffusion effects, but it can show low ionization efficiency for some type of analytes. Usually a special device (commercial or custom-built) is required for automated spraying to ensure the high quality of the matrix deposition. The operation of the device is relatively simple if the optimal spraying conditions are determined. For the sublimation approach, a custom-built chamber is typically used. However, compared to the automated spraying device, more accurate control of sublimation conditions, such as the temperature of the chamber, is required. Alternative matrix deposition approaches, such as inkjet printers (Baluya et al., 2007), electrospray devices (S. Li, Zhang, et al., 2016), pulsed spraying of the matrix solution (Hoffmann & Dorrestein, 2015), factorial design of experiments optimization of spraying condition (Tressler et al., 2021) and a combination of sublimation with recrystallization (Morikawa-Ichinose et al., 2019) have all been used to improve matrix deposition quality.

The site-to-site reproducibility of MALDI MSI has been studied and coupled to standard MALDI MSI workflows in clinical studies. FFPE samples obtained from different sites were studied to evaluate reproducibility. The relative standard deviation (RSD%) was found to range from 17.2% to 35.9% (Ly et al., 2019). Efforts have been made to improve the site-to-site reproducibility of MALDI MSI experiments (Boskamp et al., 2021).

Matrix-free MALDI MSI techniques, sometimes referred to as surface-assisted laser desorption/ionization (SALDI) methods, eliminate the interference of matrix ions with small molecule metabolites in the low mass range (normally ions with mass-to-charge, or m/z < 350). Molecular diffusion on the tissue surface during the matrix deposition process (especially for “wet” deposition conditions; Gemperline et al. 2014) can also be avoided, resulting in more accurate compound localization, increased ion signal, and higher compound coverage (Fincher et al., 2019; Kuwata et al., 2020). In addition to SALDI, nanoassisted laser desorption/ionization (NALDI) has also been developed as a matrix-free technique used in MSI studies of different tumor tissues (Tata et al., 2012). However, matrix-free MALDI requires specialized substrates that may be difficult to produce reproducibly. A recent review describes different matrix-free LDI methods for MSI (Müller et al., 2022).

2.3.3 |. Instrumentation

The most used instrument for MALDI MSI experiments is the time-of-flight (ToF) mass spectrometer, due to its high spectral acquisition speed and excellent mass range. The relatively lower mass resolution of ToF compared to other current mass spectrometers, however, is its major limitation. Recently, a growing number of MSI studies have been conducted using FTICR mass spectrometers, as their ultrahigh mass resolution capabilities provide highly accurate mass measurements with sub-ppm mass errors.

In MALDI-ToF MSI experiments, either linear or reflector mode can be used. Reflector mode yields mass spectra with higher mass accuracy and mass resolution (e.g., ~20,000 resolving power for reflector mode vs. ~5000 for linear mode) (Aichler & Walch, 2015). Linear mode, however, is normally used for analysis of larger molecules (MW larger than 2000 Da, up to 25,000 Da; Chaurand et al., 2008; Cornett et al., 2007), mainly peptides and proteins. For other smaller molecules (MW < 2000 Da), such as metabolites or lipids, reflector mode is more commonly used (Mezger et al., 2019). The signal of higher mass ions is enhanced in linear mode compared to reflector mode. The spatial resolution of ToF MSI experiments is typically 20–200 μm and can be as low as several μm (Zavalin et al., 2013, 2014). One major advantage of MALDI-ToF is the high spectral acquisition rate. Although it is largely dependent on the highest m/z acquisition limit, acquisition rates of ~1–2 pixels s^–1 at 100 μm spatial resolution can still be maintained. These rates result in acquisition times of several hours for tissue sections with an area of ~1 cm² (Prentice et al., 2015; Spraggins & Caprioli, 2011; Spraggins et al., 2016). By optimizing the laser repetition rate, sample stage velocity, laser spot diameter, and raster sampling methods, higher acquisition rates (50 pixels s^–1) are now achievable (Prentice et al., 2015; Spraggins & Caprioli, 2011; Spraggins et al., 2016). For this reason, MALDI-ToF mass spectrometers has been widely used in three-dimensional (3D) MSI experiments where multiple tissue sections are imaged consecutively to reconstruct the 3D anatomy of the entire tissue (Liang et al., 2021; Paine et al., 2019; Quanico et al., 2018). In cancer studies, MALDI-ToF MSI has become a useful tool for biomarker discovery, diagnosis and pathway analysis in various types of cancers, as discussed in detail in the following section (Berghmans et al., 2019; Briggs et al., 2019; Denti et al., 2021; Randall et al., 2019).

Recently, MALDI-FTICR MSI experiments have become more popular due to its inherent ultrahigh mass resolving power. Usually, FTICR mass spectrometers can achieve mass resolution that ranges from several hundred thousand to several millions, with sub ppm or ppb-level mass accuracy (Bowman, Blakney, et al., 2020). Therefore, FTICR has seen extensive use in MSI experiments, especially for metabolite imaging (Cornett et al., 2008). Research is now being conducted to enhance its mass range (Piga et al., 2019), but FTICR MS is not typically used for imaging large molecules, such as proteins. The spatial resolution of images obtained with FTICR mass spectrometers is not as high as that of images obtained by ToF. To achieve ultrahigh mass resolution, the acquisition time in FTICR is usually much longer than ToF MSI experiments, so the spatial resolution (and therefore the number of pixels) must be kept at a lower setting (less pixels). Despite these limitations, FTICR MSI is still a powerful tool for spatial metabolomics as it provides unparalleled mass resolving power, which leads to much more accurate spectral feature annotations in the studies of cancers such as lung carcinoma (Cao et al., 2021), renal cell carcinoma (Drake et al., 2020), breast cancer (Tucker et al., 2019), and so forth. With more reliable annotations, metabolic pathway analysis becomes more reliable. The ability of conducting tandem MS (MSⁿ) experiments in FTICR also allows structural elucidation of isomeric metabolites (Cornett et al., 2008).

2.4 |. Other MSI techniques

2.4.1 |. DESI MSI

DESI desorbs and ionizes molecules by spraying charged, fast-moving droplets (~150 m s^–1) (Venter et al., 2006) onto the tissue surface via an ESI nozzle. The secondary droplets generated from the tissue surface are charged and directed to the MS inlet (Ifa et al., 2010; Takáts et al., 2004). The likely mechanism of DESI has been described as “droplet pickup,” (Cooks et al., 2006) where the surface is wetted, splashed, followed by the extraction of the molecules. Ionization efficiency can be tuned by switching to different spray solvents (Green et al., 2010). Unlike MALDI, desorption and ionization take place under ambient conditions, that is, in an open-air environment. No matrix solution is needed; therefore, sample preparation for DESI experiments is minimal, enabling high throughput analysis, and direct applications of MSI in clinical studies (Ifa & Eberlin, 2016; Ifa et al., 2010). The removal of the matrix prevents the loss of detectable features due to poor matrix deposition. Ion suppression caused by matrix ions can also be avoided.

DESI MSI studies are mainly aimed at the analysis of small molecules, and MALDI has been used together with DESI to increase compound coverage (Banerjee et al., 2017; Eberlin et al., 2011; Garza et al., 2018; Zemaitis et al., 2021). The major limitation of DESI MSI experiments is the relatively low spatial resolution. Normally, the spatial resolution of a DESI MSI experiment is 50–200 μm (Table 1) (Qi et al., 2021). Some efforts have been made to improve pixel size, for instance, the use of nanoDESI has improved the spatial resolution to <10 μm level (Yin et al., 2019). Various types of mass analyzers have been coupled to DESI for imaging experiments, including Q-ToF, Orbitrap, and FTICR (Kooijman et al., 2019; Manicke et al., 2010; Towers et al., 2018). A limitation of DESI-FTICR experiments is their lower duty cycle resulting from the low scanning rate of the FTICR mass spectrometer (Zemaitis et al., 2021). To improve this while maintaining the high mass accuracy of the FTICR measurement, an external high-performance data acquisition unit was used in parallel with the standard data acquisition system, making the transient match the total ion detection time (Kooijman et al., 2019). Studies on the repeatability and reproducibility of DESI MSI experiments for measuring lipids in human cancer tissues have been conducted. The mean standard deviation for repeatability and reproducibility were 22 ± 7% and 18 ± 8% under optimized DESI settings and solvent parameters, which is acceptable for lipid analysis in human cancer tissues and further clinical studies (Abbassi-Ghadi et al., 2015).

DESI MSI has been applied to cancer metabolic profiling and diagnosis studies, such as lipidomic profiling of breast cancer (Santoro et al., 2020), oral squamous cell carcinoma (Yang et al., 2021), colorectal cancer (Mirnezami et al., 2016), diagnosis of prostate cancer (Banerjee et al., 2017), and clear cell renal cell carcinoma (Vijayalakshmi et al., 2020).

2.4.2 |. SIMS MSI

Instead of an ablation laser or a charged spray, SIMS sputters the tissue surface with a focused ion beam (the “primary ion beam,” usually consisting of metal, fullerene or gas clusters; Breuer et al., 2019) for desorption and ionization of molecules. Molecules from the tissue surface become ionized generating a secondary ion beam, that is, transferred to a mass analyzer. Compared to MALDI and DESI MSI, SIMS has the highest spatial resolution (50–100 nm, Table 1) (Anderton & Gamble, 2016), which makes it suitable for imaging of single cells (Brison et al., 2013; C. He et al., 2017; H.-W. Li et al., 2019; Nygren et al., 2007). The introduction of nanoSIMS to MSI further increases the spatial resolution of the experiment (C. He et al., 2017). SIMS is also a versatile ionization technique that can be coupled to various mass analyzers, for example, ToF and FTICR. SIMS is specific to only the very first molecular layer of the tissue, thus it is commonly used in 3D MSI experiments (Brison et al., 2013; Castellanos et al., 2019; Nygren et al., 2007). Low SIMS ionization efficiency (Fletcher et al., 2011) and extensive fragmentation limits the analysis of low-abundance, intact biomolecules (Yoon & Lee, 2018). Introduction of gas cluster ion beams for SIMS MSI has largely improved the ability of analyzing biological samples (Yoon & Lee, 2018). Ionization can also be enhanced by treating the tissues with agents such as trifluoroacetic acid vapor (Angerer et al., 2015, 2016).

Although being a matrix-free technique, sample preparation has considerable impact on SIMS imaging quality. The temperature of the tissue is the main source of molecule relocation in SIMS MSI experiments. Migration of molecules and lowered ionization efficiency caused by temperature change during SIMS MSI analysis result in distorted ion images and loss of features (Yoon & Lee, 2018). Frozen-hydrated analysis and tape-supported mounting and freeze drying are normally used to maintain the intact state of the tissues (Angerer et al., 2016; Kim et al., 2017).

Together with MALDI and DESI, SIMS MSI has also been extensively used in cancer studies, including multi-omics studies of breast cancer (Tian et al., 2021), pancreatic β cell islet tumors (Bluestein et al., 2018), and basal cell carcinoma (Dimovska Nilsson et al., 2020).

2.4.3 |. Other ionization methods used in MSI experiments

Several newer ionization methods for MSI have been described. Among them, LAESI (Nemes & Vertes, 2007), matrix-assisted laser desorption electrospray ionization (MALDESI) (Sampson et al., 2006), and MALDI-2 (Figure 4) have emerged as powerful imaging tools in recent years. All these techniques have been widely used for the imaging of cancer biomarkers such as metabolites, lipids, glycans, and peptides in tissues and single cells (Bien et al., 2020; Heijs, Potthoff, et al., 2020; Hieta et al., 2020; Nazari et al., 2018; Pace et al., 2022; Stolee & Vertes, 2013).

FIGURE 4.

Schematic illustrations of (A) LAESI (C, capillary; CCD, CCD camera with short-distance microscope; CE, counter electrode; CV, cuvette; FL, focusing lenses; HV, high-voltage power supply; L-N₂, nitrogen laser; L-Er:YAG, Er:YAG laser; M, mirrors; OSC, digital oscilloscope; PC-1 to PC-3, personal computers; SH, sample holder; SP, syringe pump), (B) MALDESI and (C) MALDI-2 setups for MSI experiments. Adjusted and reprinted with permissions from American Chemical Society (A: P. Nemes & Vertes, 2007, B: J. S. Sampson et al., 2006, and C: A. Potthoff et al., 2020).

As shown in Figure 4A,B, both LAESI and MALDESI involve an ESI source coupled with a laser pulse at a 90° incidence angle respect to the ESI emitter. The tissue sections are typically held near the emitter tip on a xyz-movable stage.

In LAESI, the ablation laser (usually a mid-IR laser) generates plumes ejecting from the tissue section surface under ambient conditions. This plume is then intercepted and ionized by the charged ESI solvent (so called “postablation ionization”) (Vertes et al., 2008). The generation of the ablation plume is induced by the absorption of laser energy by the naturally existing water in the tissue section. No matrix is required in the process, and sample pretreatment is minimal. LAESI is a more sensitive technique compared to typical AP-MALDI MSI experiments, where a 10²- to 10⁴-fold enhancement in ions’ signal has been reported (Nemes & Vertes, 2007). However, the application of LAESI is limited to water-rich samples because of its ionization mechanism. As a matrix-free technique, the spatial resolution of LAESI MSI experiments is not limited by the size of cocrystals formed in the normal MALDI MSI experiments, but dependent on the size of the laser spot. Its spatial resolution has been improved from 300 to 400 μm (Nemes & Vertes, 2007) to 70 μm (Hieta et al., 2020). Besides, LAESI tends to favor ionization of more polar lipids (Hieta et al., 2020). A matrix compound is used in MALDESI experiments to facilitate desorption of the molecules from the tissue surface (Sampson et al., 2006). The desorbed molecules are then ionized by the ESI solvent. The organic matrix compound can be replaced by endogenous water or deposited thin ice layers on the tissue surface (Robichaud et al., 2014). The ability to generate multiply charged ions is another feature of MALDESI that makes it very appealing for large molecule imaging (Sampson et al., 2006). Different mass analyzers such as ToF, Orbitrap, and FTICR have been successfully coupled to all the aforementioned ambient ionization techniques (Barré et al., 2019; Barry et al., 2015; Hieta et al., 2020).

Briefly, a two-step postionization process is involved in MALDI-2, including ionization of the matrix molecules by the primary MALDI laser beam and chemical ionization-like charge transfer to the analytes initiated by the secondary laser beam (Potthoff et al., 2020). The ionization efficiency and sensitivity of various molecules can be enhanced by 2 to 3 orders of magnitude in both ionization modes (Barré et al., 2019; Potthoff et al., 2020), with optimized instrumental parameters. Mass analyzers including ToF and Orbitrap have been coupled with MALDI-2 sources (Niehaus et al., 2019; Soltwisch et al., 2015). Ion mobility instruments have also been incorporated with MALDI-2 devices (Soltwisch et al., 2020).

In addition to all the ionization techniques mentioned above, laser ablation inductively coupled plasma (LA-ICP) has also been used in MSI studies for the imaging of trace amount of metals and nonmetals as biomarkers in cancer diagnosis, for example, human malignant mesothelioma tissues (Voloaca et al., 2022), but this elemental imaging technique is not covered in detail in this review. A more detailed description of the development of the LA-ICP-MSI technique and its applications in biology can be found elsewhere (Doble et al., 2021).

2.4.4 |. IMS-MSI

IMS separates ions based on their size, or CCS. Therefore, it provides an additional dimension of separation complementary to MS analysis. It is especially powerful for differentiation of isomeric ions, bringing more confidence for accurate compound identifications (Mesa Sanchez et al., 2020). Many commercial and custom-designed instruments that couple IMS with MS have been reported and applied to MSI studies (Sans et al., 2018).

Both vacuum-based (MALDI) and ambient ionization (DESI, LAESI, etc.) techniques have been successfully coupled to various types of IMS techniques (Figure 5) (Sans et al., 2018), and small molecules such as metabolites and lipids have been extensively imaged. For instance, for the imaging of lipids, separation, identification and spatial localization of phospholipids (e.g., phosphatidylethanolamines and phosphatidylcholines) from isobaric and isomeric ions has been achieved (Jackson et al., 2014; McLean et al., 2007). Field asymmetric IMS filtering in DESI MSI has enabled enhanced signal-to-noise ratios, leading to increased lipid coverage (Feider et al., 2016). Trends in the measured IMS collision cross sections (CCS) of the different lipid classes have also been used for lipid identification (Snel, 2019; Unsihuay et al., 2021). Trapped IMS (TIMS), a relatively newer technique, has shown great potential for separating isobaric and/or isomeric analyte ions and has been increasingly used in the MSI field (Rivera et al., 2020). Several platforms, including timsToF-MALDI and timsToF-MALDI-2 have been successfully applied to MSI studies (Soltwisch et al., 2020; Spraggins et al., 2019). Moreover, parallel accumulation-serial fragmentation (PASEF) performed during TIMS analysis has enabled higher throughput MS/MS capabilities for MSI, yielding structural information for more analyte ions in a single MSI experiment (Rivera et al., 2020). Recently, a cyclic ion mobility instrument has been reported with improved ion mobility resolving power. In this instrument, ions perform multiple passes through the cyclic ion mobility cell, improving resolution (Giles et al., 2019). This technique is now being coupled with DESI MSI experiments in newly-designed commercial platforms.

FIGURE 5.

Schematic illustration of IMS and MSI techniques that have been integrated for imaging of metabolites, lipids, and proteins. Adjusted and reprinted with permission from Elsevier (M. Sans et al., 2018). IMS, ion mass spectrometry; MSI, mass spectrometry imaging.

2.5 |. Practical considerations

2.5.1 |. Mass calibration in MSI experiments

Mass calibration is a crucial step in MSI workflows for reliable compound identification. Even though high-resolution mass spectrometers such as FTICR can provide the most accurate mass measurements (on average, <1 ppm mass errors can be achieved) (D. F. Smith et al., 2012), proper calibration techniques are still required to ensure there is minimal mass shift during data collection. Mass calibration is even more important for measurements carried out in mass spectrometers that show more pronounced drift, such as ToF.

Mass calibration can be conducted either externally or “internally” (or sometimes referred to as “online” calibration). A calibration mixture is normally used for external mass calibration. For instance, CalMix solution from Thermo, sodium trifluoroacetate, sodium acetate, and sodium formate solutions. For “internal” or “online” calibration, a compound deposited on top, under, or next to the tissue section, or endogenous molecules from the tissue section itself, are used as the calibrant (s). For MALDI MSI, red phosphorus is commonly spotted on the glass slide for online calibration. Matrix ions or specific analytes may also be used, as long as they distribute evenly throughout the entire tissue region with reasonable ionization efficiency (Treu & Römpp, 2021). Chemical background ions can also be used for calibration in MALDI MSI (Boskamp et al., 2020). For DESI MSI, the calibrants are usually added to the spray solvent and sprayed directly from the ESI nozzle (Inglese et al., 2021). Normally, external and internal calibration are combined for better results.

Some modified calibration methods have been introduced for MSI, minimizing mass errors. For example, by using the average frequency shift of ambient ions in a MALDESI FTICR MSI experiment, ppb mass accuracy was achieved (Barry et al., 2013). A “Mosaic cube” approach that generates multiple spatially distributed data cubes in a single data analysis session can also improve the mass accuracy to sub-ppm levels (D. F. Smith et al., 2012).

2.5.2 |. Ion suppression in MSI

Most of the tissues studied by MSI are highly heterogeneous, leading to different ionization efficiencies for different compound classes. Therefore, ion suppression is a common issue that needs to be expected when interpreting results, especially for imaging of small molecules such as lipids (A. J. Taylor et al., 2018). Competition among the endogenous molecules for charge, different extraction efficiencies of the molecules on the heterogeneous tissue surface and presence of high salt contents in the tissue are the major sources of ion suppression. Specifically, ion suppression in MALDI MSI experiments due to heterogeneous or poor matrix deposition is a key disadvantage of this technique (Heeren et al., 2008; Tomlinson et al., 2014), causing poor spectral reproducibility that limits quantitation ability and adds extra steps to the spectral normalization process.

Research has been conducted to address ion suppression in MSI. Rinsing steps can be added to the sample preparation workflow to alleviate or remove ion suppression caused by salts (Heeren et al., 2008) and high salt tolerance matrices have also been used (S. Wang et al., 2016). In most cases ion suppression is highly localized; tissue extinction coefficients have been used to assess ion suppression on-tissue relative to off-tissue regions (A. J. Taylor et al., 2018). These coefficients were used as regional normalization factors for specific tissue features. Tools like LCM can be combined with MSI experiments to validate ion abundances (Heeren et al., 2008). Until now, ion suppression remains a complex, unresolved issue in MSI.

2.5.3 |. Batch effects

Batch effects are a series of systematic variations that influence a certain group of samples in the same way (Balluff et al., 2021). In MSI experiments, batch effects can result from tissue collection, storage, sectioning, matrix application (for MALDI MSI experiments), and data collection. Variations in the laboratory environment, instrument conditions, and operations are also sources of batch effects. Batch effects exist from pixel to pixel, section to section, and slide to slide. Normally, batch effects bring bias to some measured variables, such as peak abundances, masking true biological variance from sample to sample (Balluff et al., 2021). Several efforts have been made to reduce the influence of batch effects in MSI experiments. Randomization, performing technical replicate experiments, and blocking play important roles in removing such technical artifacts (Oberg & Vitek, 2009). Therefore, a careful study design must be conducted before MSI experiments. In addition to study design, detection and removal of outliers can further reduce variances. Several software packages for detecting outliers have been developed (Balluff et al., 2021), and multivariate statistical models have been used (Wehrli et al., 2019). In 3D MSI experiments, outlier detection can be achieved by regression analysis (Vos et al., 2019). Efforts are still underway to attempt to address MSI batch effects, including designing more robust workflows, selecting quality control samples, and developing better normalization methods (Balluff et al., 2021).

2.6 |. Data processing

2.6.1 |. Image visualization

Following MSI experiments, features can be visualized as heatmaps that show the distribution and abundances in the tissue pixel by pixel. It is crucial that images reflect the actual molecular distributions in the entire tissue. Moreover, the size of MSI data sets is usually quite large, especially for FTICR MSI. Therefore, conversion of mass spectral features to images requires handling of large data sets accurately, and many commercial (e.g., SCiLS Lab) and open source (e.g., BioMap, MSiReader [Bokhart et al., 2018] and Cardinal [Bemis et al., 2015]) software platforms have been developed to this purpose. These platforms include additional functions for further data analysis. For 3D imaging experiments, reconstruction of the entire tissue volume can be challenging. It not only requires alignment of the various regions of different tissue sections, but also the m/z values themselves. To this end, some methods such as 3D reconstruction by multivariate segmentation have been reported (Patterson et al., 2016). Image realignment can also be accomplished with the help of a series of optical images, but it requires linkage between optical and MS images (Vos et al., 2021). Cardinal, for example, is capable of reconstructing 3D images efficiently (Bemis et al., 2015). SCiLS Lab is also capable of processing 3D images but requires purchase of additional licenses.

2.6.2 |. Feature annotation and identification

MSI mass spectral features must be annotated correctly for the purpose of metabolic pathway analysis. It is crucial in cancer research that metabolic pathway analysis correctly reflects the mechanism of cancer pathogenesis without false positive stemming from incorrect annotations. From the full mass spectra generated in MSI experiments, limited structural information is obtained. Therefore, only putative feature annotations are possible by comparing measured m/z values to databases. At this level of analysis, accurate mass measurements are critical. FTICR MSI provides the most accurate mass measurements with the highest possible mass resolving power, enabling more reliable putative feature annotations compared to ToF MSI.

Further experiments, such as tandem MS (MS/MS, MS², or MSⁿ) are typically required for better spectral feature annotations. For instance, in lipidomic profiling studies, isomeric phosphatidylcholines (PC) and phosphatidylethanolamines (PE) are indistinguishable simply based on accurate mass measurements. MS/MS experiments can fragment these lipid ions upon isolation and provide signature fragment ions related to their head groups. Trapping instruments such as Orbitrap and FTICR, and hybrid instruments are all capable of conducting MSⁿ experiments. However, more detailed identification of the lipid features, such as pinpointing the position of C=C bonds in lipid alkyl chains, requires additional efforts. As mentioned previously, on-tissue derivatization, where a reagent is used to react with C=C bonds, and diagnostic fragment ions are produced in the mass spectrum, is a useful tool for this purpose. For example, ozonolysis (Claes et al., 2021), Paternò–Büchi reactions (Wäldchen et al., 2019), and epoxidation reactions (H. Zhang et al., 2021) have been extensively used for C=C bond localization.

2.6.3 |. Statistical analysis

Following data processing steps such as normalization, data compression (e.g., spatial segmentation using hierarchical clustering), and scaling, statistical analyses are conducted on MSI data. Data analysis leads to the identification of changes in abundances for specific metabolites or lipids that can then be correlated with cancer progression, for example, alterations of some lipids are indicators of cell proliferation, differentiation, and apoptosis (Jiménez-Rojo et al., 2020). Proper statistical models based on altered metabolite features can also be a powerful tool in cancer diagnostics.

Typically, univariate analysis is performed first to evaluate if significant differences exist between each data set. A t test (Krzywinski & Altman, 2013) is used for comparison between two data sets and analysis of variance (ANOVA) (Kaufmann & Schering, 2014) is used with three or more data sets. Dimensionality reduction and classification methods, including principal component analysis (PCA) and partial least squares discriminant analysis (PLS-DA) can then be performed to identify and differentiate different classes in the data sets, such as differentiation of cancerous and healthy tissue regions. Other classification methods, for example, t-distributed stochastic neighbor embedding (t-SNE) (Schwarz et al., 2022) and uniform manifold approximation and projection (UMAP) (Smets et al., 2020) have been increasingly used in MSI studies. Furthermore, to properly discover biomarkers within hundreds of features, receiver operator characteristic (ROC) curves (Fawcett, 2006) are usually computed. The ROC curve is a plot of the percentage of the true positives (sensitivity) against the percentage of the false positives (specificity). The areas under the ROC curves (AUC, ranging from 0 to 1) are used to examine if a specific feature could serve as a univariate marker for cancer detection. Normally ROC curves of features with an AUC ≥ 0.80 are considered significant (English et al., 2015). In MSI, multiple significant biomarkers are required to confirm a cancer diagnosis, and ROC curves can be plotted based on several correlated features. Regression methods, such as least absolute shrinkage and selection operator (LASSO) (Tibshirani, 1996) have been widely used for feature selection and predictive model development (Bensussan et al., 2020; Eberlin et al., 2016; Tu et al., 2021; Vijayalakshmi et al., 2020). Many software packages are now able to perform statistical analysis, on MSI data, including SCiLS Lab (Bruker), and R-based packages such as Cardinal (Bemis et al., 2015), MALDIQuant (Gibb & Strimmer, 2012), and SPUTNIK (Inglese et al., 2019).

2.6.4 |. Quantitation

Quantitation in MSI is challenging because it is influenced by many factors including ion suppression, matrix effects, and the lack of proper standards (Unsihuay et al., 2021). Relative quantitation by directly comparing two or more tissue sections is the most common approach. To compare the abundances of interesting features in different tissue sections, normalization is usually performed to address the effects of ion suppression and drifting ionization efficiency. Normalization to the total ion current is usually used in MSI data preprocessing. Addition of an internal standard can help the normalization process, where ions’ signals can also be normalized to this compound (Cobice et al., 2016). The performances of several internal standard deposition techniques has been evaluated and compared for DESI MSI (Perez & Ifa, 2021). A detailed protocol for nanoDESI MSI quantitation has been described in the literature (Yin et al., 2019). Addition of internal standards is even more crucial for absolute quantitative analysis. In many cases, a calibration curve is generated to obtain the absolute quantity of a given analyte. For MALDI MSI experiments, internal standards are usually spotted on the tissue to correct for the influence of the matrix and tissue heterogeneity (Chumbley et al., 2016). Isotope-labeled internal standards can also be sprayed with the matrix for absolute quantitation purposes. Quantitative MSI has been used in a 3D MSI study to help understand tumor heterogeneity (Giordano et al., 2016). A study comparing glutathione levels between hen ovarian cancer and healthy tissues has also been reported (Nazari et al., 2018).

3 |. APPLICATIONS OF MSI IN CANCER RESEARCH

MSI is a powerful, label-free, in situ imaging technique that has seen increased application to cancer studies in the last 5 years. In this section, applications of MSI to spatial cancer metabolomics since 2017 will be discussed. Implementation of MSI techniques in clinical settings has also been rapidly increasing, and the popularization of machine learning has made MSI a mighty cancer diagnostic tool. Reconstruction of 3D tissues using 3D MSI enables more comprehensive understanding of the spread of cancers on highly heterogeneous tissues, and MSI at single cell levels can provide details on cancer progression without averaging a whole population.

3.1 |. MSI spatial cancer lipidomics

Lipids are a class of compounds closely involved in cancer pathogenesis (Butler et al., 2020). It is estimated that more than one million molecular species are lipids. Lipids are classified based on their hydrophilic head groups (8 main lipid categories and 86 classes) (L. Li et al., 2014). Variations in their hydrophobic alkyl tails (or side chains) such as chain length adds a variety of biochemical properties to lipids. For instance, studies have shown that the side chains of phosphatidylinositol phosphate (PIP) lipids are altered in cancer cells bearing p53 mutations (Naguib et al., 2015). The amphipathic nature of lipid molecules makes them play essential roles as important biological building blocks in promoting cell growth, proliferation, and division (Storck et al., 2018). For example, cancer cells consume, produce and transform various types of lipids to support cancer cell growth and provide energy for cancer metastasis (Butler et al., 2020). Meanwhile, immune responses of the body to cancer also alter specific lipid pathways (Yu et al., 2021). Therefore, alterations of lipid abundances during cancer development are viewed as the result of the direct impact of the disease on the organs and their environments, and also a reflection of the corresponding responses of the body to mediate cancer cell death. Lipid profiling studies can provide a more in-depth understanding of the cancer development mechanism. Further, lipids have extensively been proposed as early-stage cancer biomarkers and are also used in surgical decision making (Butler et al., 2020; L. Li et al., 2014).

Nevertheless, unlike proteins and nucleic acids, studies on cancer lipid profiles are still at relatively incipient stages. The complex combinatorial nature and diversity of lipid molecules leads to difficulties in their detection and quantitation, especially in complex biological systems. The development of MS-based techniques is filling this gap rapidly. For example, LC-MS studies on lipid profiles of different types of cancers using serum (Cheng et al., 2020), plasma (Z. Chen et al., 2022), urine (Lima et al., 2021), and tissue (Neef et al., 2020) samples has shown many fundamental and significant mechanisms related to cancer development. In a complementary fashion, MSI has enabled label-free, direct visualization of the distributions and abundance alterations of lipids directly in the cancer tissues, gaining traction as a valuable tool. Typically, MSI workflows for cancer lipidomic studies involve sample collection and pretreatment, acquisition and visualization of mass spectral features, feature annotations, and data processing including (but not limited to) lipid pathway analysis, development of statistical models for classification and biomarker analysis (Figure 6).

FIGURE 6.

A typical FTICR MSI lipidomics workflow for cancer studies. Part of the figure was created with license obtained from BioRender.com (agreement numbers XX23PNM7XW, RA23PNM7ZG, TI23PNM80Q, and IF23PNN56N). FTICR, Fourier-transform ion cyclotron resonance; MSI, mass spectrometry imaging.

MSI has been broadly used to understand mechanisms of cancer pathogenesis and progression in breast cancer (Santoro et al., 2020), prostate cancer (Randall et al., 2019), bladder cancer (Tu et al., 2021), and gastric cancer (A. Smith et al., 2017). Specifically, lipid MSI is an effective way to identify and analyze different subtypes of a given cancer and study the tumor microenvironment. For example, lipid profiling of molecular subtypes of breast cancer, including invasive breast cancer (IBC) and ductal carcinoma in situ (DCIS) were studied by using DESI MSI operated in negative ion mode with a 200 μm spatial resolution (Santoro et al., 2020). In this study, several important FA and glycerophospholipids were identified as key molecular species that mediate cell signaling and apoptosis in these different subtypes, providing insights into the progression of breast cancer. Recently, IR-MALDESI MSI (operated in both positive and negative ion modes) was used for in-depth lipidomic profiling of muscle-invasive bladder cancer (Tu et al., 2021). The alterations and biofunctions of several classes of phospholipids, glycerides, and FA were investigated and explored. In a recent study on prostate cancer, lipid isomers were identified using MALDI MSI coupled with ozone-induced dissociation, and the lipid isomer ratio was used for classification of the cancer tissues (Young et al., 2021). More importantly, comprehensive understanding of the molecular level lipidomic alterations in cancerous tissue defined the basis for differentiation of cancer from healthy tissues, and could thus be used as the starting point for diagnostic tools, especially for early-stage cancer diagnosis. MSI, together with the development of proper statistical models, has been broadly used in diagnosis of many other cancer types (Sans et al., 2017; Young et al., 2021; J. Zhang et al., 2017). PCA and PLS-DA are typically used for classification of tumor and healthy tissues, and ROC curves for biomarker selection. Several multivariate statistical models have been applied in MSI-based cancer studies. For example, in studies on muscle-invasive bladder cancer, different cancerous and healthy regions were classified using t-SNE and discriminating lipid features were selected by LASSO (Tu et al., 2021). DESI MSI and LASSO have been also used to diagnose and subtype non-small cell lung cancer (Bensussan et al., 2020).

To enhance the detection of specific lipid classes, on-tissue chemical modifications and derivatizations have been used in targeted lipid MSI studies. For example, to map low-abundance free FA (FFA) in thyroid cancer tissue sections, N,N-dimethylpiperazine iodide was used as a modifying agent. Nine low-abundance FFAs were successfully detected and imaged using MALDI MSI, and identified as the main products of de novo fatty acid synthesis in cancer tissues (S.-S. Wang et al., 2019). Furthermore, a recent study has shown that by adding F^– (NH₄F) into the negative ion mode nanoDESI extraction solvent, a 10- to 110-fold lipid signal enhancement could be reached for the detected [M – H]^– ions (Weigand et al., 2022). In theory, more lipid pathways and new biomarkers could be identified with the enhancement of the ions’ signals.

MSI experiments may also be coupled with established clinical tools and procedures for a more detailed evaluation of cancer progression and diagnosis. For instance, fine needle aspiration was combined with MALDI MSI for detection and characterization of lipid biomarkers of breast cancer (Cho et al., 2017). Another study combined MALDI FTICR MSI with histological grading through the Gleason system for the purpose of prostate cancer diagnosis (Randall et al., 2019). DESI MSI has been coupled with immunohistochemistry and LASSO modeling for lipidomic profiling of oral squamous cell carcinoma (Yang et al., 2021). Histology staining, laser microdissection, and MSI are also commonly used together to isolate regions of interests from the entire tissue section, helping to define the tumor region more accurately in both MALDI and DESI MSI experiments. However, spatial coregistration between these two methods must be carried out with great attention to detail (Dewez et al., 2019). For more information on the applications of MSI techniques in clinical studies, refer to Section 3.5.

3.2 |. Spatial glycomics studies in cancer research using MSI

Besides lipids, glycans are another group of essential molecules in cells that participate in many physiopathological processes involved in cancer progression, including cell adhesion, cell–cell interactions, signaling, inflammation, tumor cell invasion, and metastasis (Nardy et al., 2016). Specifically, epithelial–mesenchymal transition (EMT) is the main process where epithelial cells start to lose polarity and gain migratory and invasive properties, eventually resulting in tumor metastasis. Alterations in glycan levels have been proven to be closely related to such processes (X. Li, Zhang, et al., 2016). For instance, N-glycans with high mannose chains were proved to be key intracellular intermediates in tumor tissues (Loke et al., 2016), and branching of N-glycans was found in many tumorigenic processes (Drake et al., 2017). Some glycans have also been found to be involved in immune recognition processes (Silva et al., 2020). Normally, glycans link to proteins and lipids through glycosylation processes, forming N- or O-linked glycans (formation of glycosidic bonds with either asparagine or serine/threonine residues in proteins) and glycolipids. Abnormal levels of N- and O-glycans have been detected in different types of cancers, for example, prostate cancer (Scott & Munkley, 2019), gastric cancer (Freitas et al., 2019), colorectal cancer (Doherty et al., 2018), bladder cancer (Jian et al., 2020), and colon cancer (Pothuraju et al., 2020). Therefore, understanding the metabolism of glycans in cancer is crucial to study the mechanisms of cancer cell migration, communication, and metastasis. Like lipidomic studies, glycans have also been investigated extensively by MSI. The workflow shown in Figure 6 is also applicable to glycans, with one additional sample preparation step: the tissues are normally stabilized using FFPE, and an enzymatic deglycosylation step is needed to detach glycans from the proteins and peptide backbones.

Spatial glycomics via MSI has emerged as a popular research area drawing significant attention. In a recent study on human clear cell renal cell carcinoma (ccRCC) tissues, N-glycans with bisecting N-acetylglucosamine (GlcNAc) and multiple fucosylated residues were found to associate with ccRCC development, and mainly localized in the proximal tubule region of the kidney. Further, a comprehensive mapping of N-glycans in the entire kidney tissue, including cortex, medullar, glomeruli, and proximal tubule, indicated several ccRCC-specific features that could be used to delineate the tumor (Drake et al., 2020). Several N-glycans have been detected by MALDI MSI in serous ovarian cancer (SOC) tissues, closely following the various cancer stages. Late stage SOC was distinguished from early stage SOC by the higher abundances of oligomannose, complex neutral, bisecting, and sialylated N-glycans in the tissue (Briggs et al., 2019). Besides high-grade SOC carcinoma, spatial glycomics on other subtypes of ovarian cancer has also been studied using MSI (Grzeski et al., 2022). Currently, MSI studies on N-glycans are more common than O-glycans. The reason is the lack of proper enzymes for deglycosylation. Besides, the high heterogeneity of the glycan core in O-glycans makes their analysis even more challenging (Wilkinson & Saldova, 2020).

On-tissue chemical derivatization has been used to enhance glycan signals or increase sensitivity of MSI experiments. For instance, Girard’s Reagent P (a positively charged hydrazine) was used to form conjugates with N-glycans. The abundances of the analytes were enhanced by 230-fold for glucose and 28-fold for maltooctose. This method has also been successfully applied in MSI experiments on human ovarian cancer tissues (H. Zhang et al., 2020). In another study, the sensitivity for the detection of deprotonated N-glycans was improved by three orders of magnitude when MALDI-2 was used (Heijs, Potthoff, et al., 2020). Recently, it has been shown that native N-glycans can be studied using IR-MALDESI MSI without additional chemical derivatizations after enzymatic digestions (Pace et al., 2022).

3.3 |. 3D MSI in cancer research

Reconstruction of the 3D tissue structure is important for better understanding the spatial distribution of metabolites, biochemical interactions of metabolites in the entire tissue, signaling pathways, and cancer progression (Seeley & Caprioli, 2012). Most commonly used 3D MSI methods are MALDI and SIMS, with DESI MSI gaining more attention recently (Vos et al., 2021). Generally, in 3D MSI experiments, 2D images are obtained from a series of adjacent tissue sections, and the 2D images are then aligned and combined into a 3D image data set. Therefore, the improvement of strategies for 3D model reconstruction has been crucial in MSI studies (Vos et al., 2019). The most straightforward path to 3D image assembly is to embed fiducial markers onto the tissue sections as reference points (Chughtai et al., 2012). The anatomical features of the optical or MS images have also been used to align 2D MS images. However, the alignment is mostly conducted manually, and only tissues with well-defined and visible structures can be accurately aligned (Dueñas et al., 2017; Paine et al., 2019). Multivariate techniques such as t-SNE have also been used to simplify the 2D image dimensional complexity by segmenting the tissue sections into spectrally similar regions (Abdelmoula et al., 2019). Currently, alignment of the 2D images for 3D reconstruction remains challenging. The assignment of the tissue subregions in the 3D model is another major challenge in 3D MSI experiments, and the workload for 3D MSI data analysis is still relatively heavy (Vos et al., 2021).

Despite these challenges, 3D MSI has been increasingly used in cancer research. A 3D brain model was constructed by aligning 49 mouse brain tissue sections to visualize lipid markers of metastasizing medulloblastoma (Figure 7A). The distribution of several lipids and their roles in cancer metastasis were discussed, providing insights for diagnosis and treatment (Paine et al., 2019). In a separate study, a 3D model was reconstructed from 162 consecutive human oral squamous cell carcinoma tissues, coregistration of the 2D MS images with H&E images and immunohistochemistry images was used to construct the 3D model (Figure 7B) (Lotz et al., 2017). The distribution of the tumor drug imatinib was also mapped through a 3D MALDI MSI model (Morosi et al., 2017). MSI-based clinical studies in cancer research will be reviewed in Section 3.5. 3D DESI MSI and a neural network-based deep learning were used to study the metabolic heterogeneity of human colorectal adenocarcinoma (Inglese et al., 2017). Applications of machine learning/artificial intelligence in MSI studies will be discussed in detail in Section 3.7.

FIGURE 7.

(A) 3D MS images constructed by alignment of 49 metastasizing medulloblastoma tissue sections; different colors indicate distributions of three lipids in the tissue. (B) Reconstructed 3D H&E image (left) and MS image (right) of 162 human oral squamous cell carcinoma tissue sections. Adjusted and reprinted with permissions from Springer Nature (A: Paine et al., 2019) and Elsevier (B: Lotz et al., 2017). 3D, three-dimensional; H&E, hematoxylin and eosin; MS, mass spectrometry.

3.4 |. Multimodal imaging techniques

Combination of other imaging modalities with MSI is now considered a more powerful approach that maximizes the chemical and biological information obtained (Neumann et al., 2020). Many imaging modalities have been successfully combined with MSI techniques, and those multimodal imaging techniques have been widely used in cancer research, including microscopy (Prade et al., 2020), transcriptomics (Sung et al., 2019), spectroscopy (Neumann et al., 2018), and electrochemistry (Baker & Jagdale, 2019). Figure 8 shows a few examples of multimodal imaging workflows.

FIGURE 8.

(A) Coregistration of a MS image with a H&E-stained optical image, (B) MS (top) and immunofluorescence (bottom) images of drug metabolites and DNA damage markers in pancreatic cancer tumors, (C) cluster image of the different anatomical regions in a rat hippocampus tissue (top left), average IR absorption spectra per cluster (top right) and average MS spectra per cluster (bottom), and (d) coregistration of a DESI-MS image with Raman image of a mouse brain tissue. Adjusted and reprinted with permissions from Springer Nature (A: Frédéric Dewez et al., 2019) and American Chemical Society (B: Strittmatter et al., 2022, C: Neumann et al., 2018, and D: Bergholt et al., 2018). H&E, hematoxylin and eosin; IR, infrared; MS, mass spectrometry.

Among all mentioned techniques, microscopic tools have mostly been used together with MSI in cancer research. For example, hematoxylin and eosin (H&E) staining has been used extensively in cancer pathology and coupled with various MSI techniques. H&E-stained microscopic images can help visualize the morphology of the tissue, and specific histologic features seen after staining can help the identification and differentiation of tumor regions in the tissue (Schwamborn, 2017). Usually, tissues can be reused for H&E staining following DESI and MALDI MSI, therefore, optical images of the same tissue section can be easily acquired, the overlay of chemical and morphological information obtained, and the distribution of certain metabolic features co-registered (Cordeiro et al., 2020; Heijs et al., 2016; Y. Wang et al., 2020). Immunofluorescence (IF) microscopy has also been coupled with MSI. In the study on the treatment of pancreatic ductal adenocarcinoma, gemcitabine and its metabolites were first visualized by DESI MSI. IF analysis was performed on the same tissues and coregistered with the MS images to evaluate drug-induced DNA damage in the tumor (Strittmatter et al., 2022). Other microscopic techniques such as imaging mass cytometry (Porta Siegel et al., 2018; Strittmatter et al., 2022) and electron microscopy have also found applications in multimodal MSI studies (Klingner et al., 2019). In these cases, MSI can help microscopy identify the cancer borders by imaging of specific signature lipids (Woolman et al., 2021).

Several spectroscopic techniques have been used together with MSI. Nondestructive, label-free, and high spatial resolution (<50 μm) (Iakab et al., 2021) spectroscopic methods provide additional molecular information for the features detected by MSI, such as bonding information and functional groups of specific features in the mixture (Neumann et al., 2020). Vibrational spectroscopy methods, including Fourier transform infrared spectroscopy (FT-IR) and Raman spectroscopy are the most commonly used spectroscopic techniques in MSI studies. IR and Raman spectroscopy are not only used independently for cancer diagnosis (Lazaro-Pacheco et al., 2020; Wrobel & Bhargava, 2018), but have also been used to improve the quality of MS images. Improvement of MS image quality (sharpening) has been achieved by performing IR spectroscopy on the same tissue, and combining IR absorption bands with MS images (Neumann et al., 2018). Images of mouse brain tissues obtained by Raman have also been co-registered with DESI MS and immunofluorescence imaging. Such “heterospectral” lipidomic profiling approach enabled an improved molecular understanding of the remyelination process (Bergholt et al., 2018). Other in vivo imaging techniques, such as magnetic resonance imaging (MRI), positron emission tomography (PET), and computed tomography (CT) have also been coupled to MSI to further facilitate the studies of on prognosis, diagnosis, and treatment of cancers (C. Zhao et al., 2021).

3.5 |. Applications of MSI in clinical studies

MSI has been increasingly used in clinical studies in the recent 5 years, it provides unique cell-type specific molecular profiles, direct visualization of the heterogeneity of the tumor, and expansive compound coverage, making it a valuable add-on to current clinical diagnostic tools. Using the diagnostic molecular features discovered by MSI to identify tumor regions and differentiate them from the adjacent healthy regions has shown high accuracy. Together with histology staining techniques and multivariate statistical tools, MSI has been frequently used for accurate surgical margin assessment. A DESI MSI/LASSO workflow has been reported and evaluated as a tool for assessing and defining pancreatic cancer surgical resection margins (Eberlin et al., 2016). Another intraoperative DESI MSI method has shown its ability to help estimate and determine the resection margins for glioma in 3 min with 93% sensitivity and 83% specificity (Pirro et al., 2017).

The application of MSI to pharmacokinetics studies has enabled the evaluation of the efficacy and efficiency of anticancer drugs by visualizing their spatial distribution in the target tissue (Nishidate et al., 2019). Therefore, MSI has been used to guide anticancer drug design and to monitor and optimize cancer therapeutics. Examinations of the drug’s pharmacokinetic drug resistance, that is, whether the drug can be delivered to the tumor with sufficiently high concentrations and the spatial distribution of such drug has been investigated using MSI. For example, the distribution of an anti-angiogenic drug, sunitinib, was mapped in rabbit liver tumor tissue using MALDI MSI, and the heterogeneous drug distributions in several distinct tumor regions were probed (Fuchs et al., 2018). MALDI MSI has also been used to monitor the drug delivery process at different time points in a pancreatic xenograft model. The maximum accumulation of the active compound in the cancer cells were determined, allowing the optimization of the drug design (Fujiwara et al., 2016). In another study, the distribution, metabolism, and uptake of irinotecan in colorectal tumor organoids were systematically studied, providing insights for optimal and personalized medicine recommendations (X. Liu et al., 2018). Furthermore, the toxicology of anticancer drugs has also been evaluated by MSI. An example is a MALDI MSI study on the nephrotoxicity of dabrafenib and its metabolites in juvenile rat kidney tissues (Groseclose et al., 2015). A more detailed description of DESI and other ambient MSI studies in clinical cancer research can be found elsewhere (N. Li et al., 2020).

3.6 |. Single cell MSI in cancer research

With recent advancements in high spatial resolution and high sensitivity MSI techniques, imaging experiments at the single cell level have become more approachable, showing tremendous potential for cancer studies. Spatial metabolomics at the single cell level enables the visualization of intercellular metabolic differences within a tissue section and provides detailed metabolic profiling and biochemical changes of individual cancer cells, which is especially interesting at the early stages (M. J. Taylor et al., 2021). To cover the various types of cancer tumor cells and their broad size distributions, different ionization approaches have been applied. Among them, SIMS MSI and MALDI MSI are the most commonly used techniques (Gilmore et al., 2019). With the application of gas cluster ion beams (e.g., C₆₀⁺ and Ar_n⁺), the sensitivity of SIMS MSI experiments has improved drastically (Gilmore et al., 2019), and the level of fragmentation has been significantly reduced (Seah et al., 2014). The development of nanoSIMS has pushed the spatial resolution to 50 nm and better (Schoffelen et al., 2018), and several commercial instruments are now able to achieve sub-μm spatial resolution. ToF is the most common mass analyzer used in single-cell SIMS MSI, and a recent instrumental modification coupling an Orbitrap to a ToF-SIMS instrument has enabled high spatial and mass resolution MSI experiments. Combining trapping instruments with ToF also adds MS² capability to the system, so structural information can be obtained (Kotowska et al., 2020).

In addition to SIMS, MALDI has also been adapted to single-cell MSI experiments. A recent study has shown that 10 μm spatial resolution is readily achievable in a MALDI-ToF mass spectrometer (Ogrinc Potočnik et al., 2015). In some studies, the geometry of the MALDI setup has been optimized for better spatial resolution. A transmission geometry, where the laser beam is coaxial with the MS inlet and penetrates the analyte from the back of the sample stage, allowed the pixel size to be decreased to 5 μm (Gilmore et al., 2019). Similarly, an AP-scanning MALDI (SMALDI) ion source has been developed and the spatial resolution has been further improved to 1.4 μm (Kompauer et al., 2017). Application of MALDI-2 in MSI experiments has resulted in an increase in spatial resolution to better than 10 μm, necessary for most single-cell analysis (Bowman, Bogie, et al., 2020). Other instrumentation advancements, such as the development of nanoDESI and LAESI, has also facilitated single-cell MSI under ambient conditions (M. J. Taylor et al., 2021). The coupling of various ion mobility instruments to various mass analyzers has been proven to further increase sensitivity and throughput (Hebert et al., 2018), which can be potentially useful for single-cell MSI studies.

Due to the molecular complexity of the cellular metabolome, molecular coverage remains a challenge for single-cell MSI experiments. Decreased pixel sizes lead to smaller sampling areas, thus lowering the number of metabolites extracted and ionized during the MSI event. Therefore, higher sensitivity must be achieved to enable single-cell MSI studies. For MALDI MSI, MALDI-2 has shown an increase in sensitivity of 2–3 orders of magnitude compared to conventional experiments (Potthoff et al., 2020). Molecular coverage is increased in MALDI-2 by the addition of a second laser (Soltwisch et al., 2015). For SIMS MSI, the highest sensitivity at present is achieved by using water clusters as the primary ion beam (Sheraznée Rabbani et al., 2015).

A single cell MALDI ToF MSI study has been recently conducted on a human gastric cancer specimen (Ščupáková, Dewez, et al., 2020). Using coregistered H&E imaging, the molecular contents related to different cell morphologies were probed, and the influence of the cancer cells on their adjacent healthy cells was investigated. In another study, 14 different breast cancer cell lines that belong to different cancer subtypes were imaged using a trapped ion mobility-ToF mass spectrometer with both MALDI and MALDI-2. Numerous lipids were identified at the single cell level using data-dependent acquisition in an Orbitrap mass spectrometer. A total of 79 lipids were detected with different ratios in different cancer subtypes. This study provided important information for recognizing cancer cells in the tissue and understanding intracellular metabolic pathways in breast cancer. This MSI approach demonstrated its potential as a robust digital pathology tool compared to IHC, which is more time-consuming and suffers more technical variance (Cuypers et al., 2022; Ščupáková, Balluff, et al., 2020). Several anticancer drugs have been imaged at the single cell level using laser ablation (nano) SIMS, and ICP MS, providing information on the interaction between anticancer drugs and the cancer cells themselves, visualization of the drug uptake and distribution, and further facilitating studies on drug action mechanism for improved drug design (Meng et al., 2020; Meng et al., 2021; K. Wu et al., 2017).

3.7 |. Machine learning in cancer MSI studies

The rapid development of MSI in recent years requires more efficient data handling and processing tools, as very large multi-dimensional data sets are now routinely generated with each experiment, especially under high (mass and spatial) resolution conditions. Extraction of important biochemical information from the data sets is a real challenge. It is beneficial to build robust machine learning models that can predict cancer subtypes, model cancer progression, discover cancer biomarkers, and diagnose cancer at early stages. Therefore, machine learning and deep learning have been widely applied in MSI studies to interpret MSI data and build proper predictive models for the analysis of new clinical data sets.

Several machine learning and deep learning algorithms have been developed to provide more efficient approaches to MSI data analysis and answers to questions in cancer biology. For example, a highly sparse MSI data set collected from a human prostate cancer tissue section was compressed and clustered using a deep learning algorithm based on a variational autoencoder neural network. This network was trained to learn and capture spectral features that were used for prediction of the original data set, inducing minimal spectral feature loss after the reduction in data dimensionality. The reduced data set was then used for prostate tumor region identification by clustering of tumor-related features. The algorithm was further tested on a 3D MSI data set of a mouse glioblastoma tissue section, also exhibiting efficient data compression and high-quality spectral prediction (Abdelmoula et al., 2021). Five different 3D MSI data sets from different cancer types were used in the study, each of which consisted of more than 800,000 preprocessed spectra with more than 7600 data points per spectrum. As mentioned in Section 3.3, a deep learning 3D DESI MSI workflow was established to probe the metabolic heterogeneity of a human colorectal adenocarcinoma biopsy. In this study, an unsupervised parametric t-SNE algorithm was applied to a data set containing >200,000 spectra with 391 features to map 3D MSI data sets onto a 2D manifold, enabling the clustering of data that is not visible in PCA (Inglese et al., 2017). In a different study, a convolutional neural network algorithm was used to classify different lung tumor subtypes in a MALDI MSI experiment (Behrmann et al., 2018). Overall, machine learning continues to be extensively used in MSI-based cancer studies for data compression and classification, with almost unlimited potential for developing even more efficient algorithms for cancer status prediction and biomarker discovery. We believe that machine learning approaches coupled with MSI will largely improve our ability to extract meaningful biological information from complex data sets in the field of cancer research.

4 |. CHALLENGES AND PERSPECTIVES

MSI has become one of the most promising techniques for spatially resolved metabolomics studies and is now starting to be widely used in cancer research. However, challenges remain that would need to be addressed for improving the analytical performance of the methods involved. One of the major challenges in MSI-based cancer studies is low reproducibility, especially in MALDI MSI experiments. These issues arise mainly from variance introduced during the sample pretreatment steps and technical variability occurring from laboratory to laboratory. Such limitations hinder long-term MSI studies, such as longitudinal spatial metabolic profiling of cancer tissues. Therefore, efforts have been made to either minimize the possible sources of variance in the sample pretreatment step, such as improvements in matrix deposition techniques (see Section 2.3.2) (Baluya et al., 2007; Hoffmann & Dorrestein, 2015; S. Li, Zhang, et al., 2016; Morikawa-Ichinose et al., 2019; Tressler et al., 2021), or performing cross-normalization on collected MSI data to improve site-to-site reproducibility (Boskamp et al., 2021). Multicenter studies are normally performed to evaluate the reproducibility of MSI experiments (Ly et al., 2019). Despite these advances, more standardization of experimental protocols is still needed. Development of more robust MSI techniques that require minimal sample preparation is an alternative approach to solve these issues. The growth of subcellular MSI experiments conducted at sub-μm spatial resolution and 3D MSI studies has increased the urgency for methods to efficiently compress, process, and analyze the massive data sets generated. Therefore, single-cell level MSI studies may continue to grow with the improvements of the mass spectrometers and data processing platforms in the foreseeable future. Additionally, further improvements in MS platforms used for MSI could yield remarkable benefits in terms of increasing data acquisition rates without sacrificing spatial or mass resolution. For clinical studies, higher throughput MSI analysis is critical so that larger sample cohorts can be screened rapidly, enabling MSI to become a more practical clinical tool. More research is still needed to further improve lipid annotation in MSI, a key aspect of cancer tissue typing, we believe there will be a rapid expansion in the development of different approaches for accurate feature annotations. We are also confident that the application of new machine and deep learning techniques will continue to grow in MSI-based cancer studies, leading to more efficient data compression, filtering, classification, and pattern recognition and may become one of the mainstreams in MSI studies in the next 5 years.

Biographies

Xin Ma is a postdoctoral fellow working in Professor Facundo Fernández’s lab at the Georgia Institute of Technology since February 2021, after obtaining his PhD from Purdue University under the guidance of Professor Hilkka Kenttämaa in December 2020. His current research involves spatially resolved lipidomic and glycomic profiling of ovarian cancers using ultrahigh resolution FTICR mass spectrometry and developing new mass spectrometry imaging platforms for lipid identification. In Professor Kenttämaa’s lab, he studied fundamental aspects and applications of gas-phase ion-molecule reactions using ion trap mass spectrometry, including gas-phase reactivity studies of organic polyradicals, C–H bond activations using metal-free reagents, and development of functional-group-selective ion-molecule reactions for metabolite structural elucidation and drug impurity identification.

Prof. Facundo M. Fernández is the Vasser-Woolley Professor in Bioanalytical Chemistry and Associate Chair for Research and Graduate Training in the School of Chemistry and Biochemistry at the Georgia Institute of Technology. He is the author of 195+ peer-reviewed publications and numerous invited presentations at national and international conferences in the field of mass spectrometry, metabolomics, and analytical chemistry. He has received several awards, including the NSF CAREER award, the CETL/BP Teaching award, the Ron A. Hites best paper award from the American Society for Mass Spectrometry, and the Beynon award from Rapid Communications in Mass Spectrometry, among others. He serves on the editorial board of The Analyst and as an Associate Editor for the Journal of the American Society for Mass Spectrometry and Frontiers in Chemistry. His current research interests span the field of metabolomics and chemical evolution, and the development of new ionization, imaging, machine learning, and ion mobility spectrometry tools for probing composition and structure in complex molecular mixtures.