Spatial pharmacology using mass spectrometry imaging

Abstract

The emerging and powerful field of spatial pharmacology can map the spatial distribution of drugs and their metabolites, and effects on endogenous biomolecules, including metabolites, lipids, proteins, peptides, and glycans, without the need for labeling. This is enabled by mass spectrometry imaging (MSI) technology, providing previously inaccessible information in diverse phases of drug discovery and development. We provide here a perspective on how MSI technologies and computational tools can be implemented to reveal quantitative spatial drug pharmacokinetics and toxicology, tissue subtyping and associated biomarkers. We also highlight emerging potential for comprehensive spatial pharmacology by multimodal data integration of MSI with other spatial technologies. Finally, we describe how to overcome challenges including improving reproducibility and compound annotation to generate robust conclusions to improve drug discovery and development process.

Mass spectrometry imaging technologies enable spatial pharmacology

Drug discovery and development typically involve a pre-discovery phase of understanding dysregulated disease mechanism, biomarker discovery, and disease classification; a drug discovery phase of identifying targets and therapeutic agents that interfere with pathological processes; a preclinical phase of evaluating drug mechanism of action, efficacy, and toxicity; a clinical phase of investigating the effect of drugs on humans; and a regulatory phase for approval of the drug¹. Assessment of drug exposure at a target site using efficacious and safe drug concentrations is critical for advancement of drugs through the development pipeline. In a pharmaceutical setting, liquid-chromatography-coupled mass spectrometry (LC-MS) and tandem mass spectrometry (MS/MS), and whole-body autoradiography (WBA) using radiolabeled compounds are key analytical techniques for quantitative assessment of ADMET (absorption, distribution, metabolism, excretion and toxicity) properties of drugs and their metabolites². While the sensitivity and specificity of LC-MS assays are valuable for evaluation of pharmacokinetic and toxicological properties of drugs in tissue homogenates, the heterogeneity of tissue and analyte distribution is not preserved in these bulk tissue analysis. On the other hand, while WBA provides a sensitive way to assess ADMET properties of drugs within a spatial context, the parent drug and its metabolites cannot be discriminated, and drugs require cumbersome and costly radiolabeling. New techniques and technologies to improve understanding of therapeutic molecules early in the drug development process are sought after – to improve attrition rate and accelerate time to market.

Mass spectrometry imaging (MSI) unlocks new avenues for label-free spatial mapping of drugs and metabolites within tissues and cellular sub-compartments, while simultaneously shedding light on biomarkers of drug efficacy and toxicity by capturing effects on endogenous biomolecules (measured as mass-to-charge or m/z values)³. Distribution of small metabolites, lipids, glycans, proteins, and peptides can be visualized, without a priori knowledge of the molecules of interest⁴. Using an array of MSI technologies, spatial pharmacology of drugs can be determined at different scales – from whole body to single cell and subcellular compartments (Figure 1a).

Figure 1. DESI, MALDI and SIMS MSI across spatial scales enable spatial pharmacology.

Schematic illustration of (a) DESI, MALDI, and SIMS MSI exhibiting complementarity in spatial resolution. MSI allows imaging of whole body and organ level down to single cell and subcellular locations, using small rodent body size as a reference. (b) DESI, MALDI and SIMS ionization process. DESI is an ambient ionization technique using charged solvent sprayed on a sample surface for desorption and ionization of surface molecules. MALDI ionizes molecules with a pulsed laser beam rastered across a crystallized matrix-coated tissue surface. SIMS uses a high energy primary ion beam that sputters surface material. The secondary ions generated are accelerated, resolved, and measured by the mass spectrometer.

While the MSI field mainly focused on technology development in recent decades, the latest advances in instrumentation and artificial intelligence (AI) to analyze high-dimensional MSI data make this an emerging and valuable tool impacting the life cycle of drugs⁵. In this review, we provide a concise perspective on the most recent advances in the last few years in MSI technologies, and their implications for different phases of drug discovery and development process. We focus on the most commonly used ionization techniques including matrix-assisted laser desorption ionization (MALDI)⁶, desorption electrospray ionization (DESI)⁷, and secondary ionization mass spectrometry (SIMS)⁸ (Figure 1b). We discuss these technologies in terms of critical parameters in pharmaceutical research for obtaining reproducible spatial information with high sensitivity, spatial resolution, and assay scalability. We also discuss MSI-based quantitative spatial ADMET profiling, as well as emerging potential for discovery-based approaches for patient stratification and biomarker discovery. The high-dimensionality of MSI data, while rich in information content, also brings data analytic challenges. Machine learning (ML) and deep learning (DL) approaches which provide insights into tissue heterogeneity for therapeutic selection and treatment response, and multimodal imaging and data integration for comprehensive spatial pharmacology analysis are also covered.

Developments in MSI technologies and application in targeted drug and metabolite imaging

MSI technologies providing specific analytical capabilities are available to aid investigations with specific study objectives. The parameters that are of highest priority in drug discovery and development include the sensitivity with which compounds of interest can be detected, spatial resolution (See Glossary), and data acquisition speed (Table 1). These are determined by a combination of ionization source and mass analyzer on a mass spectrometer. While it is mostly desirable to achieve high spatial resolution, sensitivity, and throughput in a single experiment, one typically comes at the expense of the other. Therefore, critical parameters for drug evaluation need to be pre-determined. Recent innovations in instrumentation allow (a) improved sensitivity for different classes of analytes, allowing detection of a number of drugs, and also greater coverage of endogenous classes of molecules^4,9 (b) improved spatial resolution to detect compounds within smaller features, such as cellular and subcellular structures, for improved understanding of disease and drug mechanism of action¹⁰, (c) improved peak capacity and accuracy in identification of drugs, metabolites, and biomarkers, by coupling ion sources to ultra-high-resolution mass analyzers for better spectral resolution¹¹, and orthogonal ion mobility separation of complex mixtures¹². High spatial resolution and ultra-high mass resolution, however, come at the cost of throughput, which is an important factor for time-sensitive decision-making in drug discovery and development pipeline. Instrumentation developments as well as advanced computational strategies for post-acquisition data processing can aid in addressing this issue¹³.

Table 1.

Comparison of the DESI, MALDI and SIMS mass spectrometry imaging technologies.

Technique Feature	DESI	nano-DESI	MALDI-TOF	AP-MALDI	SIMS-TOF	nano-SIMS
Ionization source	Electrospray of highly charged droplets	Electrospray of highly charged droplets	Laser beam	Laser beam	High energy primary ion cluster beam such as Arn+, Bi3+, C60+, (H2O)n+	High energy primary ion beam such as Cs+ and O−
Molecular class detected	drugs lipids metabolites	drugs lipids metabolites glycans peptides	drugs lipids metabolites glycans peptides and proteins	drugs lipids metabolites glycans peptides and proteins	drugs lipids metabolites peptides	stable isotope labeled molecules
Spatial Resolution (μm)	30–200 (lowest ~ 20μm)	10–200 (lowest ~ 7μm)	5–100 (lowest ~ 1μm)	5 (lowest ~ 1.4μm)	1–100 (lowest ~ 0.5μm)	~0.05
Mass range (Da)	50–1200	50–1200	100–75,000	50–6000	100–10,1000	1–400
Throughput	High	High	Medium-High	Medium-High	Low	Low
Temperature condition	Ambient	Ambient	Medium-High vacuum	Ambient	High vacuum	High vacuum
Sample preparation	No pretreatment	No pretreatment	Matrix coating	Matrix coating	No pretreatment	No pretreatment
Mass analyzers	Q-ToF* TQ* MRT* FTICR Orbitrap LTQ	Q-ToF Orbitrap	ToF* ToF/ToF* FTICR* MRT* MRMS* Orbitrap	Orbitrap* Q-ToF*	ToF* Orbitrap	Magnetic sector*
Ion mobility	TWIMS* Cyclic-iMS*	TIMS	TIMS*	-	-	-
Advantages	- Minimal sample preparation - High throughput	- Minimal sample preparation - High throughput	- Broad class of molecules - Medium to high spatial and spectral resolution -Low to high throughput	- Broad class of molecules -High spatial and spectral resolution -Medium to high throughput	- Minimal sample preparation - Single cell resolution - 3D depth profiling	Subcellular resolution
Limitations	Spatial resolution	- Sensitivity - In-house setup	- Sample preparation critical - Matrix signal interference for low m/z region	- Sample preparation critical - Matrix signal interference for low m/z region	- Low mass resolution - Low throughput	- Maximum 5–7 labeled analytes - Complex sample preparation - Low throughput
Commercial availability	Waters	No	Bruker, Waters	Thermo Fisher, Shimadzu	Ionoptika, Iontof	Cameca
References	7,11,14,66	16	6,12,21,83	22,23	8,27,54	29,84

Abbreviations Q-Tof: quadrupole time-of-flight; TQ: triple-quadrupole; FTICR: Fourier-transform ion cyclotron resonance; MRT: Multi Reflecting Time-of-Flight; MRMS: Magnetic Resonance Mass Spectrometry; IMS: Ion Mobility Spectrometry; TWIMS: Traveling-wave Ion Mobility Spectrometry; TIMS: Trapped Ion Mobility Spectrometry;

^*commercially available with corresponding MSI source.

DESI MSI for rapid imaging of drugs, metabolites, and their effects

DESI MSI is an ambient ionization method enabling rapid and direct analysis of samples without pre-treatment, allowing high-throughput data acquisition. When a spatial resolution of 30–100 μm is sufficient to assess the distribution of analytes within tissue structures, DESI provides an effective means for rapid imaging of drugs, metabolites, and lipids across large areas with high sensitivity⁷(Table 1). While a previous generation of DESI methods suffered from poor reproducibility, the latest solvent sprayer configuration reported in 2022 implements a controlled flow of solvent for increased sensitivity, spatial resolution, and robustness, enabling quantitative analysis of drugs^14,15. In addition, nanospray DESI MSI allows high spatial resolution imaging and is amenable to detecting drugs and its metabolites¹⁶.

MALDI MSI for versatile mapping of drugs and their effects

MALDI imaging implements matrix-assisted laser desorption and ionization, where the choice of matrix and sample preparation is critical for the classes of analyte detected, spatial resolution, quantitation, and reproducibility¹⁷ (Table 1). The low ionization efficiency of small molecules caused by matrix ion interferences occurring at low mass range, or a low amount of material ionized in high spatial resolution imaging, when using conventional MALDI, has been addressed to some extent with post-ionization enhancement using a second laser, MALDI-2. In the last four years, MALDI-2 has shown improved ionization efficiency for drugs, small metabolites, glycans, and lipids by one to three orders of magnitude^18–20.

In addition, MALDI instrumentation developments have seen many prototype customizations for enhanced spatial resolution. The latest implementing a combination of transmission-mode geometry and MALDI-2 with an Orbitrap mass analyzer can achieve spatial resolution down to 600 nm, and increased sensitivity for small metabolites and lipids²¹. Similarly, an ambient AP-SMALDI source coupled to a high resolution Orbitrap MS can image drug and metabolite distribution at high spatial resolution of 1 μm²². When combined with an optical microscope, it also allows high resolution and high-speed imaging of drugs and their metabolites²³.

SIMS MSI for subcellular resolution imaging of drugs and their effects

Single-cell resolution imaging of biological samples can be performed using SIMS MSI, which uses gas cluster ion beams (GCIB) to facilitate 2D imaging, as well as depth profiling of drugs, lipids, and peptides, within cells and tissues (Table 1)²⁴. The molecular fragmentation caused by a high-energy ion beam and preferential sensitivity for compounds with high logP value had previously limited its use in drug imaging. The development of high-energy water GCIB with cryogenic handling reduces fragmentation and enhances sensitivity for lipids, metabolites and peptides²⁵, as well as drugs with low logP value²⁶. Researchers developed a hybrid mass analyzer combining the high acquisition rate of time of flight (ToF) and the high mass resolution of an Orbitrap detector²⁷, and a microscope-mode SIMS²⁸ to increase throughput. Isotope imaging using nanoscale SIMS (nanoSIMS) analyzes 5–7 isotope-labeled analytes of interest, including drugs and metabolites, with subcellular spatial resolution of 50 nm²⁹. Spatial resolution can be improved to 15 nm, using a magnetic sector mass spectrometer combined with a scanning electron microscope³⁰. Representative examples highlighting the use of DESI, MALDI and SIMS MSI for different applications within the drug development pipeline, including pharmacokinetics, toxicology, and pharmacodynamics are shown in Table 2.

Table 2:

Examples of quantitative and non-quantitative MSI methods used to understand drug distribution, pharmacokinetic, metabolism, toxicology and mechanism.

Small molecules	Tissue	Ion Source	Mass analyzer	Quantitative-Calibration Curve	Application/Key Findings	Ref
Ulixertinib	Mouse brain	DESI	qToF	Spotting on slide, tissue	Quantitative MSI for drug distribution analysis	14
STVNa and Pirfenidone	Rat Lung	DESI	q-ToF	Mimetics	Quantitative drug delivery of inhaled drugs in different regions of lungs	85
Citalopram	Mouse brain	MALDI	Tof/Tof and FTICR	Spotting on tissue	Quantitative MSI and cross validation of pharmacokinetics and its induction of serotonin and its metabolites	86
Colzapine and its metabolite	Rat liver	MALDI	FTICR	Spotting on tissue and mimetics	Quantitative pharmacokinetics and drug metabolism	34
Acetaminophen and its metabolite	Mouse Kidney	AP-MALDI	qToF		Pharmacokinetics, drug metabolism and nephrotoxicity of the drug performed without derivatization	87
Bleomycin	Skin	AP-MALDI	Orbitrap	Mimetics	Quantitative MSI for drug distribution analysis	88
13C L-DOPA and 13C Dopamine	PC12 cells	nanoSIMS	magnetic sector	Number of molecules counted by electrochemistry and imaging	Quantitative MSI of dopamine and its precursor in distinct subvesicular compartments, correlated with TEM images	89
14C-labelled Climbi-36 and its metabolites	Rat brain	DESI	LTQ (linear ion trap)	Spotting on tissue	Quantitative MSI of PET tracer and its correlation with autoradiography and PET	74
Amiodarone	Rat lung	DESI	q-ToF		Drug metabolism and organotoxicity through induction of markers of lipidosis	90
Efavirenz, tenofovir, and emtricitabine	Rat brain	MALDI	ToF	Spotting on tissue	Differential distribution of the anti-retroviral drugs in brain suggesting potential benefit from combination therapeutics	91
Amiodarone	Rat alveolar macrophage cell	SIMS	Orbitrap		Mechanism of toxicity of the drug through increased phospholipidosis	92
Retigabine	Rat eye	MALDI	FTICR		Mechanism of toxicity of the drug through dimerization with its metabolite in melanin containing layers of the eye	93
Doxorubicin	Varied	MALDI	TOF/TOF		Differential delivery of drug-loaded nanocarrier within tumor versus normal organs	94
Cisplatin	Hela cells	nanoSIMS	magnetic sector		Mechanism of drug resistance mediated through its nuclear localization within open and closed chromatin regions probed using metal tagged antibodies	95
Aristolochic acids		Air flow assisted-DESI			Nephrotoxicity related to arginine-creatinine, choline and lipid metabolism	96
Scopolamine	Rat brain	AFA-DESI			Spatially resolved metabolic alterations in drug induced alzheimers disease model	97
Notoginsenoside R1 (NG-R1)	Rat brain	MALDI	TOF/TOF		Mechanism of action of drug though regulation of metabolic pathways	98
Tacrine	Mouse brain	MALDI	FTICR		Heterogeneous age-related metabolic perturbations and its response upon tacrine treatment	99

Quantitative and targeted MSI of drugs and metabolites

Conventionally, tissue-based quantitative pharmacokinetic and toxicological studies are performed using LC-MS/MS assays. However, the analysis of bulk tissue samples obscures the heterogeneity of drug uptake and its metabolism within discrete histopathological and structural correlates. This can be overcome by implementing targeted and quantitative MSI during the early phases of drug discovery and development, providing critical insight into drug pharmacokinetics and metabolism. However, quantitative MSI of drugs and metabolites is challenging because of matrix effect and ion suppression, resulting in heterogeneous analyte ionization efficiency within different tissue sub-structures. Normalization of ion suppression effects can be performed by correcting the intensity of drugs and metabolites to an isotope-labeled internal standard of the same drug or metabolite, uniformly deposited on the sample surface³¹. However, this relies on the availability of isotope-labeled compounds, which might not be readily available or may be expensive to synthesize. Absolute quantification can be performed using a calibration curve with serial dilution of compounds (a) spotted on the slide (b) spotted on control tissue, and/or (c) in a mimetic model. Though more labor-intensive, the mimetic model, where the calibration curve is created by spiking drugs on the tissue homogenate with the same background matrix, thus mimicking the ion suppression effects, is currently the most accurate technique for quantitative MSI of drugs and metabolites³².

Quantitative MSI relies on instrumentation with high sensitivity and reproducibility, and robust sample preparation, data normalization, and calibration measures³³. Recently, a multicenter study designed to assess reproducibility and accuracy of quantitative MSI of drugs normalized to internal standards has shown comparable results between centers, as well as correlation with results from LC-MS/MS assays³⁴. This suggests robustness of quantitative MSI and its potential for being an effective method within the drug development pipeline. The sensitivity of LC-MS/MS assays, however, exceeds that of MSI for quantitative analysis. MSI systems offering high sensitivity for a variety of drugs are desired, especially for potent compounds administered and accumulating at low concentrations within tissue. Assessing limits of detection early in a study to design MSI experiments to dose animals to have sufficient sensitivity without inducing toxicity is recommended³⁵.

Recent studies showed higher sensitivity and speed for AP-MALDI MSI compared to DESI and MALDI for different classes of pharmaceutical compounds^23,36. More MSI studies are needed to benchmark classes of small molecules of interest in pharmaceutical research. Currently, a number of studies have shown quantitative and non-quantitative drug distribution for a variety of drugs and metabolites using DESI, MALDI, and SIMS, with varied mass analyzers elucidating pharmacokinetics, metabolism, and toxicology³⁷ (Table 2). While LC-MS/MS will still likely remain the primary means to assess quantitative pharmacokinetics, metabolism and toxicology, complementation with MSI could help bolster understanding of drugs in early drug discovery and development efforts.

MSI for discovery-based and biomarker-driven drug development

The high attrition rate in drug discovery and development necessitates exploring new strategies to aid the process. One such approach is to implement biomarker-driven drug discovery and development. Biomarkers facilitate an adaptive drug development paradigm, impacting different stages of drug development³⁸. Biomarkers with high predictive power are desired to identify subgroups of patients that are likely to benefit from therapeutics in precision trial designs. Companion diagnostics that can reliably measure biomarkers are required, and spatial biology is currently being explored to support such endeavors³⁹. In that regard, untargeted MSI data has potential to improve understanding of spatial relationships between biomolecules and cells for diagnostics and discovery of spatially-informed biomarkers.

An array of AI, ML and DL tools are available to gain meaningful insight from the high-dimensional (in both spectral (10³-10⁷ m/z spectral bins) and spatial (10⁴-10⁶ pixels) domains) MSI data. Details on crucial components of MSI data analytics including data preprocessing, normalization, dimensionality reduction, classification, statistical analysis, and post-acquisition data transformations, that are beyond the scope of this review are discussed in these references^5,40–42 (Figure 2). This section discusses the emerging developments within the cross-section of discovery-oriented MSI for clinical diagnostics and biomarker identification, challenges and strategies for implementing AI, ML and DL algorithms, and multimodal imaging for comprehensive spatial profiling.

Figure 2. Experimental and computational MSI workflows related to pharmacology.

(a) A typical workflow of MSI sample processing includes cryosectioning of the tissue and placing it on the modality specific substrate. In case of MALDI, uniform matrix coating is applied on tissue surface. Data are acquired in the mass spectrometer. (b) Sample preprocessing includes baseline subtraction, smoothing, peak picking, peak alignment, and normalization to retrieve highly informative peaks. (c) Unsupervised (left panel) or supervised (right panel) machine learning and deep learning approaches can be used to derive a segmentation map within the tissue or for classification of tissues (d) Multivariate statistical analysis allows identification of biomarker of spatial clusters or classes obtained from (c).

Computational approaches for tissue classification and biomarker discovery

The high-dimensionality of MSI data with non-linear relationships between features makes it hard to manually identify and interpret underlying patterns. AI, ML, and DL strategies enable identification of spatially-informed biomarkers capturing metabolic heterogeneity and cellular interactions. Supervised and unsupervised ML algorithms for dimensionality reduction, classification, and visualization implemented in analyzing omics data have been translated to spatially-aware MSI data analytics^5,43 (Figure 2c). In addition, architecture-based and multiple-instance-learning-based DL approaches has been successfully implemented in MSI data, enabling rapid distinction between tissue types and disease states⁴². Similarly, unsupervised ML methods, including matrix factorization method, manifold learning and clustering, are widely used for exploratory analysis to reveal underlying structure within MSI data⁴¹. More recently, DL methods such as variational autoencoder neural network for self-supervised peak learning without data pre-processing⁴⁴ and algebraic topological framework⁴⁵ have been implemented for efficient denoising, data structure inference, and tissue sub-regions classification.

ML algorithms have been instrumental in diagnosing challenging tumors, in which pathologies of different disease etiologies exhibit histological similarities and lack discriminating markers for diagnosis⁴⁶. For example, MALDI MSI of peptides discriminated PDAC versus cholangiocarcinoma⁴⁷; DESI MSI of lipids and metabolites discriminated a benign renal tumor from renal cell carcinoma⁴⁸. Classification of glioma samples based on SIMS MSI profiling protein and metabolites could identify cluster of cells indicative of the pathophysiology⁴⁹. Similar approaches could be implemented in preclinical study designs for tissue characterization and classification for therapeutic selection, as well as to understand the effect of therapeutics. A few examples are highlighted in Table 2.

Once intersample and intrasample heterogeneity is captured as distinct classes, statistical analysis to determine the biomarkers that are highly predictive of the classes of interest can be determined. Receiver-operating characteristic (ROC)-curve analysis approach has been the gold standard for evaluating the performance of a biomarker as a classifier⁴², with shapley additive explanations⁵⁰ as a recent addition. For example, classification of MALDI MSI data, followed by discriminatory biomarker analysis, showed that L-carnitine and short-chain acylcarnitines were significantly reprogrammed in breast cancer⁵¹. Multiple studies, including multicenter benchmarking and validation studies, have shown the potential of MSI in disease diagnosis as well as biomarker identification⁵².

Highly multiplexed antibody-based MSI to probe for spatial phenotypic biomarkers

In addition, recent developments with significant implications for drug discovery are the implementation of multiplexed antibody imaging to identify cellular phenotypes and interactions in health and disease. MSI is one of the new avenues being explored, where different antibody tags are amenable to ionization and detection with different MSI technologies. Highly multiplexed lanthanide-metal-tagged antibodies have been used with SIMS^53,54, halide-tagged antibodies with nanoSIMS⁵⁵, photocleavable peptide-tagged antibodies with MALDI imaging⁵⁶ and boronic acid mass (BMTs) tagged-antibodies with DESI⁵⁷. With these tools, cellular and biomarker distribution information at the single-cell and cellular neighborhood level can now be obtained. For example, a panel of metal-tagged antibodies against subsets of tumor and immune cells using SIMS-multiplexed ion beam imaging enabled classification of triple-negative breast cancer tissues into subtypes associated with distinct spatial organization of immune and tumor cells, which correlated with survival⁵³. This new class of spatial phenotypic biomarkers account for cell densities and their interaction, compared to classical biomarkers, and several are being tested in clinical trials as companion diagnostics³⁹.

Challenges in integrating biomarker-driven MSI studies in drug discovery and development

Most discovery-based MSI examining spatial metabolomics are driven towards classification of disease samples; there aren’t many studies evaluating pharmacodynamic effects of drugs. This is likely because of challenges in untargeted spatial metabolomics studies to distill reliable and biologically meaningful insights. First, robust study designs with enough statistical power and methods to minimize sources of variability and artifacts for highly reproducible results is required. Second, the high-dimensionality of untargeted MSI data, while rich in information content, also means that it could lead to spurious associations between features; only biomarkers with high accuracy and reproducibility that can be quantitatively validated and implemented in a time and cost-effective manner have clinical utility. Large-scale cross-cohort longitudinal studies are required to derive robust validation of methods and AI/ML/DL algorithm derived biomarkers. Open-source and proprietary computational tools for MSI data analysis⁴² require cross-validated pipelines and clear guidelines for use. One approach involves the integration of open-source MSI data analysis tools, such as Cardinal⁵⁸, into a cohesive framework that is coupled with quality control, visualization, preprocessing, and statistical analysis⁵⁹. Repository infrastructures such as those built for proteomics to store assay metadata, workflows details, experimental design, and data acquisition and analysis methods⁶⁰, should be implemented in the MSI field to improve rigor and reproducibility. Clinical validation of MSI as a clinical assay⁶¹, combined with iterative training of classification models based on computational and expert evaluation will help integration of MSI into clinical diagnostics and drug discovery workflow. Another critical component of annotating and identifying the hits from these studies are described below.

Strategies to improve molecular annotation and identification for untargeted MSI

To derive meaningful biological insights and accurate biomarker annotations from the untargeted MSI spatial metabolomics studies, the m/z values need to be accurately assigned to unique biomolecules such as lipids and metabolites. Several factors, including lack of separation capabilities during MSI data acquisition, mass resolution of instrument, in-source fragmentation and influence of sample preparation on native compound modification makes accurate annotation challenging⁶². Orthogonal validation of ions by LC-MS/MS on the same sample is feasible⁶³, though the lack of chromatographic separation in MSI experiments could lead to inaccurate assignments of ions between the methods. The gold standard is to perform in situ MS/MS analysis to confirm fragmentation patterns for verification of molecular identity. However, MS/MS is feasible only for targeted studies by performing single ion fragmentation on commonly used ToF instruments. Recent instrument configurations, such as targeted DESI MSI using multiple reaction monitoring (MRM) for multiple ions on triple quadrupoles in imaging mode enable quantitative imaging of drugs metabolites and lipid biomarkers of interest^31,64. Adding another dimension of analysis, such as ion mobility separation, enables resolving isobaric compounds, aiding confidence in annotations^65,66. Lipid class, chain length and degree of unsaturation prediction using computational methods, such as Kendrick mass defect (KMD) analysis, can also complement these analysis⁶⁷. Proprietary and open-source software are available to assist in annotation and identification⁶². However, clear guidelines for reporting MSI ion annotations are still lacking. A standardized way to report how annotations are performed along with degree of confidence would provide transparency within the MSI community and increase confidence in the findings from the studies.

Next-generation multimodal imaging for integrated biomarker discovery and biological insights

The implementation of discovery-based approaches in the drug development pipeline also depends on whether mechanistic insights can be gleaned from these studies. Multi-omics ML-driven network-based approaches has shown promise for providing holistic views of systems to infer drug-target interactions, patient stratification, biomarkers, and molecular drivers of phenotypes⁶⁸. Similar spatial multimodal studies implementing developments in MSI technologies, as well as other spatial technologies including spatial transcriptomics⁶⁹ and medical imaging⁷⁰, opens up an opportunity for multimodal imaging to capture the complex dynamics in tissues (Figure 3).

Figure 3. MSI multimodal imaging reveals new insights.

Multimodal imaging using different technologies, such as (a) metabolomics using individual or integrative multimodal MSI, (b) proteomics using highly multiplexed antibody imaging for cellular phenotyping, and (c) transcriptomics and epigenomics using corresponding spatial imaging. (d) Data processing includes multimodal image alignment and registration, followed by cellular or cellular neighborhood segmentation for multimodal feature extraction. Intra and inter-modality molecular network analysis can be performed to infer spatial features and biological pathways impacting precision medicine.

Multimodality can be implemented both within the MSI field and in connection with other imaging fields. The complementarity of different MSI technologies in terms of the classes of molecules detected and spatial resolution, makes them ideal for combining, to acquire a comprehensive map of tissues at different scales. For example, multimodal imaging on the same sample and instrument, such as MALDI and DESI⁷¹, or MALDI and SIMS⁷², reveals metabolically complementary classes of small metabolites and lipids at different scales. Similarly, different primary ion beams allow lipids and metabolites, as well as multiplexed antibody imaging to be acquired to derive single-cell metabolomics by SIMS on the same tissue section⁵⁴. Correlative MSI using different instruments on the same or consecutive tissue sections using SIMS and DESI⁶³ has allowed researchers to dissect tissue organization at different scales.

Integrating other methods, such as in situ hybridization probes against mRNA markers and multiplexed antibody imaging,^63,73 positron emission tomography (PET)⁷⁴, surface enhanced Raman Spectroscopy (SERS)⁷⁵, magnetic resonance imaging (MRI)⁷⁶ with different MSI modalities has allowed researchers to correlate metabolic maps to anatomical and molecularly defined structures. Multimodality can also work as an additional validation tool. For example, detection of some classes of metabolites that are laser-sensitive benefit from infrared-assisted SERS imaging for visualization of cancer biomarkers without auto-oxidation⁷⁵. Additionally, recent advances in other imaging modalities such as spatial transcriptomics and epigenomics⁶⁹ create new avenues to examine the mechanistic relationship between omic layers. Fresh frozen tissue sections are amenable to spatial molecular imaging techniques, including MSI, spatial transcriptomics, and multiplexed antibody imaging, allowing multimodal data acquisition and analysis to be done on the same or consecutive tissue sections. By implementing complementary technologies, the limitations of each technology can be overcome to provide an integrated and systems perspective on tissues under study. The biological insights gained from these studies aid in precision medicine (Figure 3).

Multimodal MSI is in the early phases of development. Challenges in data integration and analysis using images acquired with different spatial resolutions, data structure, and features are currently being explored⁷⁷. Lessons learned from non-spatial integrated omics research could be tested within the spatial domain⁷⁸. Network and deep-learning-based integrative approaches for multi-omics data are emerging to gain comprehensive views of systems^68,79. Multimodal imaging in drug discovery and development are being explored⁷⁰, while challenges in multimodal image registration and guidelines are becoming evident⁷⁷.

Conclusions and future perspectives

Spatial pharmacology has the potential to dramatically impact human health by decoding the molecular and cellular content and interactions in healthy and disease biological systems, providing new strategies to diagnose and prevent disease, explaining response to treatment, and insights into drug pharmacology. Researchers have improved MSI instrumentation for sensitivity, spatial resolution, throughput, and chemical coverage, such that there is now an array of tools that perform well for complementary classes of molecules with impressive spatial resolution. ML and DL algorithms in image analysis have been translated in MSI to dissect and comprehend spatially and spectrally rich imaging data. Multimodal imaging has the potential to change the face of biomedicine.

Challenges that need to be addressed in the emerging field of spatial pharmacology include rigorous validation of data analysis workflows, data integration for multimodal imaging, and annotation of ions (See Outstanding questions). Clear guidelines for MSI methods for research and clinical studies are needed. Increased adoption of these technologies by diverse research labs will improve the learning process. The knowledge gained from application of these technologies and computational workflows will guide the next generation of instrumentation, and experimental and computational pipelines to improve rigor and reproducibility. Community efforts to provide resources and standardization, as recently shown for building antibody panel maps for different tissues⁸⁰, should be implemented in the MSI field. Consortia efforts such as The Human BioMolecular Atlas Program (HuBMAP)⁸¹ and Human Tumor Atlas Network (HTAN)⁸², which aim to build spatial multi-omic maps of healthy and diseased tissues, play an important role in bringing together researchers using different technologies, computational biologists, pathologists, and clinicians for interdisciplinary collaborations to decode human health and disease. Continued advancements in technological developments, data analytics, cross-cohort validation studies and community-building efforts will propel the field of MSI into mainstream biomedical and pharmaceutical research, transforming biology and medicine.

Outstanding questions.

• Where in the drug discovery process does MSI provide the most value?

• What are the key modalities to include in multimodal MSI for integrative systems biology?

• Can MSI workflows be standardized to improve rigor and reproducibility?

• When should MSI be used as a companion diagnostic or clinical diagnostic?

• Can MSI become accessible to more companies and research labs?

Highlights.

• The spatial distribution of drugs, metabolites, and endogenous biomolecules within cells and tissues can be visualized by mass spectrometry imaging (MSI)

• Targeted MSI deciphers spatial pharmacokinetics, metabolism, and toxicology of drugs

• Untargeted MSI elucidates disease stratification, disease subtyping, mechanism of pathophysiology, drug-related efficacy and toxicology, and associated biomarkers.

• Machine learning and deep learning reveal hidden structures within high-dimensional MSI data

• Multimodal imaging integrating complementary spatial biology information has the potential to provide improved biological insight and integrated biomarkers aiding precision medicine.

Glossary

Spatial resolution

A measure of the dimension of pixel size. For example, a pixel of 5 μm × 5 μm is considered to have a spatial resolution of 5 μm. The smaller the pixel size, the higher the spatial resolution.

Spectral resolution

Also termed as mass resolution or resolving power refers to the ability of the mass analyzer to separate peaks of similar m/z values. The higher the spectral resolution, the higher the probability of accurately identifying ions based on accurate mass.

LogP value

This term measures hydrophobicity/lipophilicity of compound sand is defined as the partition coefficient of a molecule between the aqueous and organic phases. A lower or negative value denotes more hydrophilicity (water soluble) and higher value denotes hydrophobicity (water insoluble).

Multiple reaction monitoring

This term refers to a highly specific and sensitive tandem MS/MS method performed on triple-quadrupole and linear ion trap MS in a multiplexed manner. Here, each predefined precursor ion is selected and subjected to fragmentation, and the product ion of predefined m/z is selected and detected.

Matrix effect

This is the suppression or enhancement in ionization efficiency of the analyte based on other analytes that are present in the sample or biological matrix.

Ion suppression

This term refers to the decrease in ionization efficiency and detection of analytes of interest caused by competing ions in the sample matrix in mass spectrometry analyses.

Isobaric interference

This term refers to molecules with different composition and structure, sharing the exact mass (or m/z values) resulting in overlapping spectra, confounding annotation.

Ion mobility

This refers to a gas-phase technique in which the speed with which a molecule travels in the presence of inert gas depends on mass, charge, and shape of the molecule.

Receiver-operating characteristic (ROC)-curve analysis

This analysis is a graph-based analytical method for assessing the performance of the binary classifier as the discrimination threshold is changed over a range. The ROC curve is the plot based on true positive rate against false positive rate at each threshold setting.

Spatial metabolomics

This refers to omics profiling of spatial distribution of small metabolites, lipids and small molecules.