4.2.1 Metabolomics contribute to target identification via bottom-up approach

The introduction of exogenous compounds into cells can lead to competitive binding with specific protein receptors, some of which can catalyze the metabolite transformations. As a result, the downstream changes in the metabolome provide insights into the enzyme targets that interacted with these compounds. This concept underpins the metabolomics-driven “bottom-up” approach for identifying drug targets, enabling the discovery of enzyme targets and clarifying the MoAs of drugs through the analysis of metabolic pathway alterations.¹⁶⁶ For example, Zhao et al.¹⁶⁷ analyzed the db/db mice samples (e.g., urine, serum, and kidney) following acteoside administration via UHPLC–LTQ–orbitrap MS-based nontargeted metabolomics. They found that the dysregulation of amino acid metabolism due to diabetes can be reversed by drug (ACT), such as restoring phenylalanine levels. Combined with network pharmacology, drug targets such as AKT2 and AKT1 were discovered.¹⁶⁷ Zhou et al.¹⁶⁸ employed targeted metabolomics to analyze the metabolic profile of elemene treatment on PC3 prostate cancer cells. Changes were observed in circulating metabolic intermediates within glycolysis and the TCA cycle, which further validated that 6-phosphofructo-2-kinase and fructose-2,6-biphosphatase 3 are critical targets in mediating the elemene-induced inhibition of glycolysis in prostate cancer cells.¹⁶⁸ Similarly, Liu et al.¹⁶⁹ conducted metabolomic analysis on Tao-Hong-Si-Wu decoction extract and treated mouse tissue samples, identifying 11 key metabolites. Integrated with network pharmacology analysis followed by relevant experimental validation, the study predicted Vegfa, IL-6, and TNF genes as the primary targets of Tao-Hong-Si-Wu decoction in treating sepsis.¹⁶⁹

In our group, we used a bottom-up approach to study triphenyl phosphate (TPhP), employing metabolomics and lipidomics to screen dysregulated metabolites and lipids (Figure 3). Using UHPLC–Q-TOF–MS, we identified 72 dysregulated metabolites, highlighting key metabolic pathways through the construction of an interaction network by Metscape.¹⁶⁶ Subsequently, molecular docking and surface plasma resonance analysis validated TPhP's interaction with enzymes in the β-fatty acid oxidation pathway. A series of validation experiments confirmed the interaction of TPhP with hydroxyl-coA dehydrogenase and hydroxysteroid 17-beta dehydrogenase 10 (HSD17B10).¹⁶⁶ This approach demonstrates the efficiency and cost effectiveness of metabolomics in large-scale target discovery.

4.2.2 Dose–response metabolomics contribute to target identification

The therapeutic efficacy and safety of a drug are inherently dose dependent, as both beneficial and adverse responses vary with drug concentrations.^{170, 171} Relying on a single-dose metabolomics study may overlook critical metabolic responses that occur at other dose levels. For instance, while acetaminophen is effective at therapeutic doses, higher exposures can deplete GSH, induce oxidative stress and mitochondrial dysfunction, and lead to hepatotoxicity.¹⁷² Multidose metabolomics addresses these limitations by providing a comprehensive assessment of metabolic changes across various drug doses, allowing for a deeper understanding of MoAs and potential off-target effects. Conventional metabolomics may be insufficient for the evaluation of chemicals that could potentially affect multiple proteins with varying potencies.^{173, 174} Therefore, dose–response metabolomics has emerged as a powerful method to quantify drug concentration effects, elucidate metabolic pathways, and identify therapeutic targets.

Dose–response metabolomics quantifies the relationship between drug concentration and its effects on metabolites within a sample,¹⁷⁴ allowing researchers to construct dose–response curves that clarify potential dose-dependent metabolic responses and provide vital insights into MoAs of drugs and possible off-target effects. For example, Patti et al.¹⁷⁴ conducted dose-effect curve modeling to find 908 features exhibiting statistically significant monotonic rises or declines in the medication etomoxir. The conclusion that at least two separate targets for etomoxir was inferred by analysis of these dose-dependent metabolite alterations using three-dimensional (3D) PCA, which offers potential explanations for its hepatotoxicity.¹⁷⁴ Similarly, Pink et al.¹⁷⁵ characterized the dose–response relationship of 3-nitrobenzanthrone (3-NBA) in a human urothelial carcinoma cell line. At low doses, the cells activated adaptive mechanisms and elevated antioxidant levels in response to the toxic 3-aminobenzanthrone metabolite derived from 3-NBA. At higher doses, cellular adaptation decreased, leading to a greater reliance on metabolic reprogramming to sustain homeostasis, which resulted in increased oxidative stress, reduced proliferation, and mitochondrial hyperpolarization, with significant involvement from the pentose phosphate pathway and folate metabolism.¹⁷⁵ In summary, dose–response metabolomics effectively elucidates the intricate connections between drug exposure, metabolic alterations, and potential therapeutic or off-target effects, thereby offering critical insights for drug development and optimization.

The identification of drug targets through dose–response metabolomics involves a systematic four-stage process (Figure 4): (1) Metabolomic analysis at various doses. The initial step encompasses the determination of an appropriate dose gradient for drug administration through pre-experimentation. This step established a foundation for the subsequent construction of dose–response curves. Metabolomic analyses of the distinct dose groups are typically conducted using LC–MS techniques. (2) Construction of dose–response curves. The large nontargeted metabolomics datasets pose computational challenges. To extract and categorize the dose–response curves for each measured metabolic feature, specialized software tools, such as TOXcms were used.¹⁷⁶ The dose–response curves can be classified into four main response trends: “no difference,” “increase,” “decrease,” and “inflection” (further divided into V-shaped and inverted V-shaped).¹⁷⁴ Based on TOXcms, Fang et al.¹⁷⁶ identified and obtained 29 significantly dysregulated metabolites of bisphenol A (BPA) exposure to cancer cell lines and concluded that the purine metabolism is the most important target pathway of BPA. (3) Identification of key target proteins. These dysregulated metabolic characteristics exhibit a clear dose–response relationship. These features can be used to identify key metabolites and enzymes related to the mechanism of drug action. Then schematic indications of the potential number of targeted proteins can be provided based on the response trends and effective dose (ED₅₀) values.¹⁷⁷ (4) Experimental validation. The correctness of the identified targets is subsequently verified through a combination of molecular docking studies, gene knockout experiments, and animal models to confirm the predicted drug–target interactions and their functional consequences. This systematic, multistep approach enables researchers to utilize dose–response metabolomics to uncover the molecular targets underlying the observed phenotypic changes induced by drug treatment.

The application of dose–response metabolomics extends beyond the investigation of individual compounds, demonstrating its versatility in the identification of targets within complex mixtures, such as TCMs and environmental exposures.¹⁷⁸ Our group conducted a dose–response metabolomics of the TCM Venenum Bufonis (VBF), known for its significant anticancer properties, demonstrated a dose-dependent decrease in DNA synthesis-related metabolites such as UTP, ATP, GTP, CTP, and UDP, suggesting that VBF primarily targets the mTOR pathway to inhibit both salvage and de novo pyrimidine and purine synthesis, ultimately leading to cell cycle arrest.¹⁷⁹ Colas et al.¹⁸⁰ investigated the alterations in river biofilms following cobalt (Co) exposure, identifying two primary concentration ranges of metabolic responses. The CRIDeR stage exhibits a biphasic trend indicative of a defensive response, while the CRIDaR stage exhibits a monotonic trend associated with cellular damage and metabolic dysregulation. The authors further elucidated the dose-dependent changes in the activities of metabolic enzymes, such as catalase and superoxide dismutase, during the CRIDeR stage, revealing the underlying defense mechanisms against the elevation in intracellular Co content.¹⁸⁰ Similarly, Lu et al.¹⁸¹ investigated the microstructural and physiological characteristics of biofilms exposed to different concentrations of Mg²⁺. The research shows that Mg²⁺ significantly affects biofilm formation in a concentration-dependent manner by regulating the microstructure of Escherichia coli and inducing metabolic modifications in nucleotide and amino acid metabolism.¹⁸¹ These studies collectively illustrate the application potential of dose–response metabolomics in the multitarget screening of TCMs, environmental exposures, and other complex mixtures, providing valuable insights into their MoA and therapeutic potential.

4.2.3 Stable isotope-resolved metabolomics contribute to the target identification

The inherent complexity of metabolic networks can limit the accurate representation of metabolic activity, even though standard metabolomics provides valuable insights into static metabolite abundance.¹⁸² Metabolite accumulation may result from increased production or decreased consumption. The actual metabolic fluxes and the contribution of substrates may differ under the intertwining of multiple pathways, despite the same level of intermediate metabolites.²⁶ In contrast, metabolic flux analysis (MFA), enabled by stable isotope-resolved metabolomics (SIRM), offers a more comprehensive understanding of metabolic regulation by directly measuring metabolic reaction rates (fluxes).¹⁸³ Pertinent biomarkers and therapeutic targets can be determined by accurately identifying and comprehending the primary metabolic pathways that are involved in drug treatment. This will provide a scientific basis for clinical diagnosis and drug–target therapy. MFA enables the observation of metabolic activity through dynamic means, thus offering a more comprehensive and in-depth understanding of the mechanisms and roles of metabolic regulation.¹⁸⁴

The utilization of isotope tracing technology in metabolic studies has evolved over the decades.¹⁸⁵ With researchers transitioning from the use of radioactive tracers to the safer application of stable isotopes, such as ¹³C, ²H, and ¹⁸O,^187-189 these isotopic labels are used to track metabolic pathways at varying labeling ratios. The quantification of isotope enrichment in cellular metabolites (M+0/M+1/M+2/…) to assess metabolic activity.^190-192 By integrating experimental data from isotope labeling¹⁹³ with metabolic network models,¹⁹⁴ researchers can mathematically determine metabolic fluxes and analyze the changes as well as distribution of compounds in metabolic pathways. Thereby, the understanding of metabolic reprogramming can be advanced. Metabolic network models are typically associated with searchable and queryable databases, such as KEGG, Virtual Metabolic Human, and MetaCyc.¹⁸³ Two common experimental designs, stationary and nonstationary MFA (i.e., isotopically nonstationary MFA), are utilized in MFA.^{183, 195-197} Among these techniques, ¹³C-MFA is considered the primary technique for determining intracellular fluxes. It offers a significant advantage in enhancing flux accuracy due to the substantial redundancy constraints provided for flux estimation.^{194, 198} Additionally, various software platforms, including eiFlux, INCA, OpenMebius, NSMFRA, ScalaFlux, and KFP, are available for flux analysis.^199-202 For instance, Zou et al.²⁰³ utilized ¹³C-labeled Gln for metabolic tracking, analyzed metabolic fluxes in vitro and in vivo, and combined with transcriptomics to uncover key reprogrammed metabolic pathways in the fibrotic microenvironment. McConville et al.²⁰⁴ utilized multiple ¹³C-labeling to detect atypical metabolic pathways in Plasmodium falciparum, highlighting the activity of its metabolite repair pathway and providing new research directions for drug discovery.²⁰⁴ The continued advancement and application of isotope tracing technology have enabled researchers to unravel the complex mechanisms underlying metabolic reprogramming, paving the way for more comprehensive target identification in biological systems.

MFA provides a more precise assessment of metabolic pathway activity, making it an essential tool for identifying key molecules as potential drug targets for further validation (Figure 5). Locasale et al.²⁰⁵ elucidated the variable mitochondrial metabolism subsequent to drug administration by scrutinizing the metabolic profiles of the drug treatment group. Through the collective use of isotopically labeled substrates, including U-¹³C glucose, U-¹³C palmitate, and U-¹³C Gln, the authors traced the alterations in mitochondrial substrate utilization patterns upon metformin treatment. Their results demonstrated that metformin induced changes in substrate utilization within mitochondria and anabolic processes. These alternation impact heightened levels of tricarboxylic acid (TCA) cycle intermediates, NADH accumulation, and the generation of ROS.²⁰⁵ This MFA-based approach suggested the mitochondrial metabolic targets underlying the anticancer activity of metformin and the factors influencing metformin resistance. Similarly, Tian et al.²⁰⁶ used SIRM to reveal multiple targets of Xiaoyaosan for improving TCA cycle disorders. The decrease in isotopic abundance of key metabolites, such as citric acid and glutamate, at M+0 indicated the role of XYS in repairing the TCA cycle. Furthermore, the isotopic abundance of metabolites in the TCA cycle at M+2 and M+3 pointed to the critical involvement of enzymes like pyruvate carboxylase and pyruvate dehydrogenase in regulating the entry of pyruvate into the TCA cycle.^205-207 In another example, Lyssiotis et al.²⁰⁷ found that glucose and uridine promote metabolism to a similar extent, confirming the catabolism of pancreatic ductal adenocarcinoma to uridine in vivo due to the high uridine, UMP, and UTP labeling (M+5) by SIRM analyses. This suggested the potential of uridine phosphorylase 1 (UPP1) as a new target for cancer therapy. Additionally, DeBerardinis et al.²⁰⁸ employed ¹⁵NH₄⁺ labeling to demonstrate that the knockout of carbamoyl phosphate synthetase-1 (CPS1) reduced the isotope abundance of thymidine at M+1, revealing the critical involvement of CPS1 in thymine synthesis and the dependence of non-small-cell lung cancer genotypes on pyrimidine metabolism. These examples illustrate the power of MFA and SIRM in elucidating the dynamic alterations in metabolic pathways upon various treatments. So that they could identify key metabolic targets for drug development and therapeutic interventions.

4.2.4 High-throughput metabolomics contributes to the target identification

Metabolomics presents a cost effective and easily implemented approach to detect thousands of metabolite components in samples, offering significant advantages in the analysis of large numbers of samples.²⁰⁹ Metabolomics allows for the direct detection of drug-induced metabolic changes associated with phenotypic effects,²¹⁰ driving the demand for high-throughput methods in multidose studies, active molecule screening, and the exploration of individual differences in drug metabolism.²⁰⁹ Traditional MS-based metabolomics face limitations due to chromatographic separation speed, which determines the time required for sample detection. However, the development of higher-throughput LC–MS and CE–MS technologies has significantly enhanced analysis efficiency. Advancements in LC–MS, such as ultra-performance LC (UHPLC), optimized chromatography columns, have significantly improved analysis efficiency by reducing solvent consumption²¹¹ and analytical cycle times, enabling large-scale metabolic phenotyping.²¹² For example, Rainville et al.²¹³ achieved human urine analysis in just 3 min using UHPLC–MS, detecting 10,400 features. While Link's team quantified primary metabolites in under 2 min with a short column and a sensitive QqQ instrument.²¹⁴ Higher-throughput LC–MS methods may sacrifice peak capacity and total feature detection, but incorporating ion mobility (IM) can mitigate these issues.²¹² Alternative strategies like DI and flow injection analysis (FIA) further enhance throughput, which abandon chromatographic separation.^{210, 215} DIMS analysis can be up to 20 times faster than LC–MS, though it may require additional cleaning steps to address system contamination and sample carryover.^216-218 FIA allows for the injection of 1000–2000 samples per day, with small injection volumes and short reaction times, which also elevate the analysis range and sensitivity, enabling high-throughput detection.^{210, 219, 220} Innovations like flow injection time-of-flight MS (FIA TOF MS) have further reduced analysis time to under 1 min, enabling the analysis of over 1400 samples per day.²²¹ Additionally, high-throughput NMR-based metabolomics offers simple sample preparation and broad metabolite detection capabilities.²²² The introduction of higher-density microplates and droplet microarrays,^223-225 along with advances in miniaturization and automation,^{226, 227} has solidified high-throughput metabolomics as a crucial tool in drug discovery and development.

High-throughput metabolomics provides an efficient method for rapid identification of metabolic targets and reactions. It enables us to move beyond the traditional bottom-up approach that focus on complex metabolite changes, driving researchers to develop innovative experimental designs like high-throughput screening (HTS) for discovering new drugs and uncovering unknown enzyme functions, including applications in in vitro toxicology. Clare et al.²²⁸ conducted HTS using a 384-well format to evaluate five types of snake venoms, identifying four snake venom metalloproteinase inhibitors and two lead compounds. Sauer et al.²²⁹ characterized the functions of 1725 overexpressed or purified E. coli proteins by monitoring changes in metabolite abundance in supplemented metabolome extracts, which revealed potential enzyme catalysis; they established evaluation criteria based on empirical Z-score thresholds derived from prior analyses of 189 known enzymes, ultimately identifying 241 potential new enzymes.²²⁹ Subsequent experiments successfully confirmed 12 of these newly identified enzymes,²²⁹ suggesting this approach could aid in discovering drug target information by identifying affected metabolic enzymes.

A key application of high-throughput metabolomics is HTS utilizing drug libraries (Figure 6A), which distinguishes itself from conventional methods by using the metabolome as a direct indicator of cellular biochemistry, thereby minimizing metabolism-related issues for screened drugs.²³⁰ By comparing spectral similarities, high-throughput metabolomics can predict drug–target relationships across large batches. For instance, Zampieri et al.²³¹ developed a method that integrates experimental calculations with metabolic analyses of drug responses and the metabolic and chemical genome profiles of single-gene deletion strains. This approach allows for the classification of drugs according to their metabolomes and MoAs based on the screening results. Additionally, a weighted scoring function is employed to correlate enzymes with metabolic changes, subsequently linking genes to construct drug–target relationships. Similarly, Sauer et al.²³² utilized high-throughput metabolomics to compare the metabolomic profiles of compound treatments with those from induced membrane protein overexpression in yeast cells, identifying several new targets, including ibuprofen's effect on GPR1 signaling (Figure 6B).

High-throughput metabolomics enhances sample analysis efficiency by optimizing instrumentation and reducing sample volume requirements, allowing researchers to analyze thousands of samples, a growing standard in the pharmaceutical industry that necessitates more efficient throughput methods. Zamboni et al.²¹⁰ propose that efficiency can be significantly enhanced without compromising sensitivity by optimizing the dead time between injections. Advances in injection techniques, such as droplet-based microfluidics and surface acoustic sampling, have made it possible to analyze more than 10,000 samples per day.²¹⁰ For example, Kim et al.²³³ describe an innovative analytical method that integrates time-of-flight secondary ion MS (ToF-SIMS) with a microfluidic cell system, referred to as the system for analysis at the liquid vacuum interface). They introduce a rapid, high-throughput platform for the quantitative analysis of amino acids in matrix deposits, which are formed from freeze-dried solution drops through ice sublimation on a parylene film microarray substrate.²³⁴ However, the drive for higher throughput necessitates the sacrifice of the separation of compounds with mass-to-charge ratios (e.g., isomers).²³⁴ This challenge demands not only higher resolution but also more advanced analytical methods. It is crucial to exclude the influence of the matrix effect and accurately identify key metabolite characteristics from complex signal peaks. According to the study by Zamboni et al.,²¹⁰ the former issue can be addressed through effective metabolite extraction protocols that ensure a constant concentration of analytes and sample clean-up. While the latter urgently needs the development of a new computing method for data integration.

4.2.5 Multiomics contributes to targets identification

Metabolomics currently faces several limitations, including limited coverage of metabolites, lack of annotations for specific compounds, and potential degradation of active metabolites, which can obscure critical metabolites and pathways, hindering a comprehensive understanding of biological mechanisms. Single-omics approaches also typically capture only a small subset of biological changes,²³⁵ making the integration of multiomics data a powerful strategy to address these challenges.²³⁶ By elucidating the origin and changes of metabolites, transcriptomics and genomics enhance the understanding of metabolic regulation at the genetic level to complement metabolomics. While proteomics reveals the relationships between metabolites and proteins through the analysis of protein expression and activity. Additionally, epigenomics helps identify epigenetic factors that influence metabolic levels.

Integrating metabolomics with other omics provides researchers with a comprehensive and systematic understanding of disease mechanisms and MoAs of drugs, thereby improving the predictive accuracy of therapeutic targets' roles and clinical effects.²³⁷ For example, Jiang et al. categorized triple-negative breast cancer (TNBC) into three distinct subtypes based on metabolic characteristics identified in the polar metabolome and lipid group. This metabolomics-driven classification offered more insights compared with transcriptome-based subtype categorization. Concurrently, genomics was employed to explore potential genomic drivers underlying metabolomic characteristics. By associating the metabolome with the genome, key metabolic genes were identified, and the accuracy of classification was validated. This multiomics approach not only revealed metabolic heterogeneity within TNBC but also enabled more precise target identification and understanding of the underlying causes of phenotypic variability in complex diseases such as chronic kidney disease and idiopathic pulmonary fibrosis.²³⁸ In another example, Cai et al.²³⁹ combined metabolomics with chemical proteomics and revealed that acetyl-CoA carboxylase 1 and acetyl-CoA carboxylase 2 are the direct molecular targets for perfluorooctanoate. While Jiang et al.²⁴⁰ demonstrated the neuroprotective effects of scutellarin via targeted metabolomics and 13C-MFA, highlighting the role of pyruvate dehydrogenase kinase 2 (PDK2) in mitochondrial glucose oxidation. Additionally, our group's research identified SOD3 and GPX4 as novel targets for hepatotoxicity induced by Benzo[α]pyrene through a combination of metabolomics and chemical proteomics.²⁴¹

Integrating metabolomics with other omics strategies has significantly improved the understanding of drug repositioning opportunities through new target identification. For example, Zou et al.²⁴² conducted a metabolomics analysis of metformin, which has shown remarkable efficacy in inhibiting tumor growth, revealing significant changes in metabolic pathways such as glutathione metabolism and polyamine synthesis in thyroid cancer. By integrating transcriptomics data, the researchers confirmed that metformin inhibited glycolysis by downregulating key glycolytic enzymes while upregulating the expression of isocitrate dehydrogenase, thereby exerting anticancer effects. Additionally, metabolomics revealed reduced levels of short-chain acylcarnitine in tumor tissues of metformin-treated patients, indicating that metformin inhibits fatty acid oxidation, likely through direct effect on mitochondrial complex I.²⁴³ Multiomics strategies also pave the way for combination therapy. For instance, Rosner et al.²⁴⁴ found that BACH1 deficiency in TNBC cells promotes mitochondrial respiration, and subsequent analyses suggested that combining metformin, a mitochondrial respiratory inhibitor, could enhance this effect. Overall, compared with single metabolomics, multiomics analysis construct a dynamic multidimensional networks of biochemical interactions rather than static maps.²⁴⁵ Overall, multiomics analysis creates dynamic, multidimensional networks of biochemical interactions rather than static maps, allowing researchers to gain a comprehensive understanding of drug effects, identify novel therapeutic targets, and explore rational combination therapies, thus advancing drug repositioning and development efforts.

Multiomics strategies indeed present unique challenges, particularly in sample collection and processing, as the response times and lifetimes of biomolecules can vary greatly within and across different omics layers. Determining the optimal sampling time to account for these variations is critical.²³⁵ Additionally, advancements in MS imaging and single-cell technologies highlight the impact of tissue and cellular heterogeneity on experimental outcomes. Developing a well-considered sample extraction and distribution scheme is essential for ensuring even representation and accurate analysis across multiple omics layers. For example, tissues may be homogenized or sectioned to provide samples for various analyses. When working with limited samples, separate extractions may yield inadequate sample sizes.

Integrate multiple omics data can enhance comprehensive analysis through two main methods: conceptual and statistical integration. Conceptual integration analyzes each omics dataset separately and synthesizes the results, similar to a 1+1 approach.²⁴⁶ Statistical integration (synchronization strategy), however, involves simultaneous analysis of data from multiple omics layers, revealing associations between features and providing a more comprehensive view.²⁴⁷ This method amplifies relationships and changes that may be less evident in single-omics analyses. The synchronization strategy for multiomics integration faces three major computational challenges: (1) correlating and tracking molecular components across different layers and databases; (2) ensuring reliable data normalization and conducting multilevel data analysis; and (3) effectively visualizing the results.²⁴⁸ In multiomics strategies centered on metabolomics, the goal is to integrate data from other omics layers into a metabolic network model, allowing exploration of related genes, proteins, and differential metabolites. Achieving this necessitates ongoing advancements in fields like computer vision and natural language processing to support unbiased integration of high-dimensional data.²⁴⁹ Relevant data integration methods and application tools have been detailed in previous reviews (e.g., Refs. ^{235, 248-250}) and will not be further discussed here. However, there remains significant potential for improvement, particularly in addressing objective differences caused by external factors (e.g., batch effects).²⁵⁰

4.3.1 MSI contributes to small molecule target identification

MSI is a crucial technique in spatial omics, providing a detailed view of metabolite distribution and helping elucidate their regulatory mechanisms. As a powerful analytical tool, MSI ionizes individual pixels from biological samples, generating mass spectral maps that are reconstructed into high-resolution images.²⁵¹ This method directly captures the spatial distribution of molecules in thin tissue sections, with spatial resolutions ranging from 10 to 200 µm.²⁵² MSI effectively reveals molecular heterogeneity across biomolecules, including proteins, peptides, lipids, polysaccharides, amino acids, and oligonucleotides.^{253, 254}

Matrix-assisted laser desorption/ionization (MALDI) has been the predominant MSI method since its introduction in 1994. MALDI uses a laser-absorbing matrix to interact with analytes, forming cocrystals that enhance imaging and detection.^{255, 256} The ionization process, triggered by laser irradiation of tissue sections, ejects compounds for subsequent mass spectrometric analysis.²⁵⁷ More recently, desorption electrospray ionization (DESI), introduced in 2007, has enabled direct analysis of living surfaces or tissue sections without prior sample preparation.^{251, 252, 258} Advances in spatial resolution and instrument sensitivity have focused on improving ionization, transmission, and detection efficiencies.²⁵⁹ MALDI–MSI typically achieves moderate spatial resolution (5–50 µm) in a nontargeted manner.²⁶⁰ While the SpaceM approach, integrating MALDI–MSI with optical microscopy and digital image processing, facilitates the correlation of metabolite data with the cells of origin.²⁶¹ The development of MALDI-2 has improved resolution to 600 nm by enhancing the ionization efficiency,^{262, 263} allowing for single-cell analysis.^{258, 259} Other techniques like air-flow assisted desorption electrospray ionization (AFADESI–MSI)^{264, 265} and surface-assisted laser desorption/ionization (SALDI–MSI)²⁶⁶ have further expanded MSI applications, enhancing ionization efficiency and reducing background noise. Recent advancements, such as 3D MSI, have extended spatial resolution to include both lateral and depth dimensions.²⁵⁹ Several comprehensive reviews detail these technological advances, applications, and operational principles.^{252, 256, 258, 267-270}

MSI is especially valuable in revealing metabolite discrepancies across various spatial contexts, making it a promising tool for target discovery in clinical metabolomics. It provides high coverage, sensitivity, throughput, reproducibility, and nonspecificity, allowing spatial analysis of highly complex biological samples.^{266, 267} In contrast, conventional metabolomics often overlooks spatial distribution within tissues, limiting insights into metabolic heterogeneity, such as in tumor environments.²⁵³ MSI's spatial metabolomics capabilities allow for the investigation of chemical distribution across different scales, illuminating complex biological systems and aiding in the understanding of drug mechanisms and therapeutic predictions. For instance, the different metabolic activity between the external regions and the center in esophageal cancer multicellular tumor spheroids²⁷¹ and the heterogeneous distribution of active metabolites of gemcitabine in pancreatic cancer tumor tissues²⁷² illustrates the value of spatial metabolomics in providing insights into complex biological systems.

MSI has proven useful in studying metabolic reprogramming linked to specific structures, offering pathways for target identification.^{102, 268} For instance, Abliz et al.²⁷³ used MSI to uncover metabolic heterogeneity in a diabetic cardiomyopathy (DCM) model, identifying hundreds of metabolites distributed within the heart via AFADESI–MSI and MALDI–MSI analyses (Figure 7). They focused on fatty acid metabolism, oxidative stress pathways, and ATP metabolism due to the significant accumulation of differential metabolites in the lesion area.²⁷³ These heterogeneous distributions of metabolites suggest their role in maintaining the unique structural and functional characteristics of the cardiac region and may also serve as specific structural markers to guide investigations into localized metabolic activity. The findings suggested metabolic markers linked to lesion areas and potential therapeutic interventions, such as the reversal of DCM symptoms with ferulic acid through inhibiting AMPK activity. Supporting literature indicates that ferulic acid-mediated AMPK activation improves cardiac function, reduces oxidative stress, and prevents myocardial damage in diabetic hearts through its anti-inflammatory and antioxidant activities.

Similarly, AFADESI–MSI has identified abnormally expressed enzymes in esophageal squamous cell carcinoma based on the spatial discrepancies in metabolites. Abliz group revealed metabolic vulnerabilities that could be targeted for therapy, such as targeting enzymes pyrrolidine-5-carboxylate reductase 2 and uridine phosphorylase 1.²⁷⁴ The advantage of this approach lies not only in its high-throughput characterization but also in the fact that it does not require the definition of a specific target of interest beforehand. Abliz et al.²⁷⁵ also mapped carnitine metabolism reprogramming in breast cancer, uncovering novel findings regarding the aberrant expression of carnitine palmitoyltransferase 2 and carnitine/acylcarnitine translocase. They also employed MSI to map the microregional distribution patterns of the sedative-hypnotic drug candidate YZG-331 in brain tissue sections, revealing its prominent distribution in the pineal gland and suggesting glutamate decarboxylase as a potential drug target.²⁷⁶ Sun et al.²⁷⁷ employed MALDI–MSI combined with IM to establish a high dimensionality reduction and spatial clustering method, effectively identifying the pathological features of advanced pulmonary fibrosis, including the specific accumulation of glycogen in myofibroblasts. They ultimately revealed glycogen as a promising target for pulmonary fibrosis. Spengler et al. introduced a high-resolution (HR2) MSI method, which demonstrated an enrichment of the anticancer drug imatinib and isocyclophosphamide in the kidney, hinting at the location of potential drug targets.²⁷⁸

A novel target discovery approach combines spatiotemporal metabolomics with isotope tracing, addressing the limitations of traditional metabolic flux technologies.²⁷⁹ This method identifies drug-related metabolic changes and interference points in metabolic pathways. For instance, Davidson et al.²⁸⁰ employed spatially resolved isotope tracing to elucidate metabolic activities in different tissues, which revealed regional differences in nitrogen source utilization for brain synthesis of glutamate. Patti et al.²⁸¹ employed this integrated approach to investigate metabolite abundance and nutrient contribution in the brains of GL261 glioma mice. They also developed a spatial isotopologue spectral analysis method to evaluate the spatial distribution of metabolic turnover fluxes.²⁸¹ Furthermore, future research will likely focus on integrating spatial metabolomics with multiomics strategies and computational advancements.²⁸² He et al.²⁸³ demonstrated the potential of this integration by visualizing cellular and metabolic features in gastric cancer, identifying metabolic remodeling and interactions that could serve as therapeutic targets, including arginine deiminase or arginase.

In conclusion, MSI's ability to elucidate metabolite discrepancies across spatial contexts offers significant potential for advancing target discovery and clinical applications in metabolomics.

4.3.2 Single-cell metabolomics in identifying small molecule targets

Cells serve as the fundamental units of life, performing diverse functions essential for maintaining vital processes and transmitting genetic information in living organisms. While seemingly identical, cells can exhibit significant heterogeneity in their gene expression patterns and phenotypes.^{284, 285} This cellular heterogeneity unveils intricate biological phenomena, such as tumor heterogeneity.^{286, 287} Cancer stem cells (CSCs) play a key role in tumor heterogeneity, driving tumor growth, proliferation, metastasis, and resistance to therapy,^{288, 289} such as the varied expression of the AC133 epitope in colonic CSCs.²⁹⁰ Similarly, metabolic differences have been observed between prostate cancer cell lines, PC3 and DU145, despite both originating from the same cancer type.²⁹¹ To fully understand disease mechanisms and MoAs for drugs, it is critical to investigate cellular activity at the single-cell level.

Single-cell analysis is commonly used to describe population diversity, track cell lines, classify cell types, and perform genomic analysis of rare cells.²⁹² Single-cell sequencing can effectively elucidate genomic, epigenomic, and transcriptomic heterogeneity within cell populations.^{285, 293, 294} For instance, single-cell RNA sequencing (RNA-seq) has demonstrated intra-tumor heterogeneity in primary glioblastoma.^{295, 296} Additionally, single-cell technologies can be integrated with proteomics, using methods like cytometry by time-of-flight to analyze the expression of surface and intracellular proteins.^{297, 298} These techniques are vital for identifying key cell subtypes relevant to drug targeting and disease progression. For example, single-cell TargEt analysis, which combines transcriptomics and chemical proteomics, can identify cell-specific targets and pathways. This approach has been used to uncover cell-specific targets of aristolochic acid I in proximal tubular cells and aspirin in brain tissues.²⁹⁹ Single-cell analysis is flourishing in various omics disciplines. However, the correlation between genotype and phenotype remains a pressing issue that requires further investigation.

The development of single-cell technologies has led to the emergence of single-cell metabolomics (SCM), which examines the distinct metabolite compositions of individual cells in varying environments. SCM provides insights into the metabolic patterns of cancer cells and helps identify potential therapeutic targets (Figure 8). Studies have shown that by understanding the metabolic patterns observed in cancer cells, immune cells can be rendered more functional by adapting to the tumor microenvironment.²⁸⁴ For example, Yang et al.³⁰⁰ used single-cell MS (SCMS) to study irinotecan (IRI)-resistant cancer cells and discovered that combining IRI with metformin significantly inhibited fatty acid synthase, overcoming IRI resistance. They also identified distinct lipids and metabolites exclusively present in live single cells. while the upregulation of stearoyl-coenzyme A desaturase 1 (SCD1) in IRI-resistant cells demonstrated that inhibiting SCD1 activity increased the sensitivity of drug-resistant cells.³⁰¹ Furthermore, Yang et al.³⁰² combined SCM and utilized a single-probe SCMS technique to gather metabolic profiles of living cancer cells exhibiting varying levels of chemotherapy-induced resistance. While SCM platforms have also been used to discover cell-type-selective microbial metabolites³⁰³ and reveal distinct lysosomal isoforms with unique metabolic profiles,³⁰⁴ offering new avenues for drug discovery.

Despite its potential, SCM still faces challenges, particularly in sampling strategies.⁹⁷ Common single-cell isolation methods include pipette-based micromanipulation, fluorescence-activated cell sorting (FACS), and microfluidic arrays.^305-308 FACS and microfluidics are preferred for preserving cell morphology and scalability,^{305, 307} while micromanipulation minimizes disruption to the cell environment.³⁰⁷ SCM requires higher sensitivity than conventional metabolomics due to the limited sample size of a single cell. The ability to resolve this sensitivity issue will determine the feasibility of using SCM in discovery-based studies to gain deep biological insights.^{284, 309} Advances in MS, MSI, microfluidic platforms,³¹⁰ and fluorescent biosensors have improved detection sensitivity, allowing for the simultaneous determination of hundreds of metabolites in single cells.³¹¹ For example, the nano-electrospray ionization MS technique developed by Shi et al.²⁹¹ enables simultaneous detection of nonpolar and polar metabolites at low concentrations (10 pg/mL), significantly enhancing detection coverage. Qian et al.³¹² have successfully employed a combination of flow cytometry separation and nanoparticle-enhanced MS for high-throughput profiling of hematopoietic stem cells, known as Hi-scMet.

In recent years, the integration of SCM and other omics techniques has enabled the analysis of multiomics data at a finer resolution, facilitating the visualization of intricate interactions among various cellular populations.³¹³ Fang et al.³¹⁴ developed a “one-shot” method for simultaneous proteome and metabolome analysis, uncovering the role of the NASP gene in tumor cell response to adriamycin. This technique minimizes sample loss and enhances detection sensitivity compared with traditional methods, holding significant promise for future research.³¹⁴

However, SCM still faces limitations, such as challenges in isolating single cells without metabolite leakage and the rapid transformation of metabolites, which can lead to undetectable or fluctuating concentrations.^{305, 307} Additionally, the lack of standardized data-sharing platforms hinders collaboration and further advancement.³⁰⁷ Despite these obstacles, SCM's ability to analyze cellular heterogeneity holds great promise for guiding therapeutic approaches, discovering new drug targets, and understanding disease mechanisms.

4.3.3 AI-assisted metabolomics to identify small molecule targets

The vast amount of data generated in metabolomics presents a significant challenge in extracting meaningful insights. Recent advances in AI have increasingly impacted the life sciences, particularly in identifying metabolomics-related targets.³¹⁵ AI, through machine learning (ML) and deep learning (DL) algorithms,³¹⁶ excels at analyzing large datasets and complex biological networks, facilitating the integration of metabolomics data and uncovering intricate metabolic relationships (Figure 9A). ML algorithms like random forest, Naive Bayes, and support vector machines (SVM) have been applied in drug development, target prediction, and disease modeling.^{317, 318} For instance, the Bayesian ANalysis to determine Drug Interaction Targets platform spans the drug discovery process, from target identification to drug development.³¹⁹ Similarly, Dong et al.³²⁰ introduced iMSEA, an enrichment analysis strategy that enhances understanding of drug-metabolite interactions and predicts drug side effects. ML has also proven effective in disease classification and prognosis,³²¹ as demonstrated by Hu et al.³²² in gastric cancer, offering a precision medicine tool.

ML has shown immense potential for predicting unknown metabolic pathways. Shah et al.³²³ proposed a DL-based method for metabolic pathway prediction, utilizing annotated genomic data to train models for accurate prediction of unknown metabolic pathways, offering a novel way to unveil the intricate metabolic networks within organisms. Valentino et al.³²⁴ focused on the application of neural networks in metabolome pathway analysis, efficiently processing complex metabolome data and mining potential metabolic pathway information. Additionally, Hero et al.³²⁵ and Liu et al.³²⁶ have worked on improving the prediction accuracy and efficiency of metabolic pathways through the optimization of model structures and parameters. Marti et al.³²⁶ proposed a dynamic prediction method to analyze time-series omics data, providing insights into the dynamic nature of metabolic processes.³²⁷

DL, using neural networks to extract features from large datasets, has also been employed in metabolomics research.³²⁸ PandaOmics, an AI-driven drug discovery platform, combines metabolomics and multimodal DL techniques to accelerate target identification and optimize drug discovery³²⁹ (Figure 9B). Tools like Microsoft's BioGPT leverage large language models to mine biomedical text, linking diseases, genes, and biological processes to rapidly identify potential drug targets.³³⁰ To address the challenges in clinical metabolomic data analysis, such as high dimensions, small sample sizes, and class imbalances, Sha et al.³³¹ proposed MetDIT, a deep convolutional neural network. MetDIT can efficiently process clinical metabolomics data, converting 1D data into 2D images for better analysis, though it highlights the need for target enrichment.³³¹ Furthermore, OTTER, a novel tool, enriches and ranks targets by integrating omics and textual data, providing more accurate results for pharmacological research. OTTER provides a “keyword” that represents the pharmacological features observed in cells treated with active compounds, enabling OTTER to provide more accurate and relevant target enrichment results.³³²

AI has dramatically accelerated metabolomics research, improving data analysis accuracy while reducing costs. The application of AI in metabolomics target discovery not only significantly accelerates the research process, improves the accuracy of data analysis, but also greatly reduces the research cost. With advanced technologies such as ML and deep neural networks, the researchers are able to deeply analyze the complex relationships in metabolic processes, providing strong support for the diagnosis and treatment of diseases. However, the widespread adoption of AI in metabolomics research also faces several challenges, including ethical considerations, data privacy protection, data availability, and ensuring AI fairness and transparency.^{333, 334} As technologies such as interpretable AI evolve and more metabolomics data becomes available, AI is expected to revolutionize drug discovery and target identification.