The field of spatial metabolomics exists because there is no alternative route to it. No antibody binds a small organic acid, and no hybridisation probe recognises a phospholipid — so mapping metabolites and lipids in tissue is a mass spectrometry problem or it is nothing. That gives analytical scientists an unusually clear claim on the territory, and an unusually specific difficulty: the data is comparatively easy to generate and genuinely hard to interpret.
Key Takeaways
|
Why Map Metabolites and Lipids in Space?
Because metabolism is spatially organised and a homogenate destroys that information. Concentration differences between tissues are informative; gradients within a tissue are mechanistic. A tumour with elevated lactate on average tells you less than a tumour with a lactate gradient tracking the hypoxic core, and only one of those measurements can distinguish cause from consequence.
The absence of alternatives is the structural point, and it is developed in Spatial Analysis in Analytical Science: Mass Spectrometry Imaging and Spatial Omics. In brief: antibodies require epitopes, which small molecules do not present, and hybridisation requires sequence, which lipids do not have. Spatial transcriptomics and antibody-based imaging are not slower routes to metabolite distribution; they are not routes at all.
Working in analytical science?
Register for a FREE Separation Science account to subscribe to the Separation Science Newsletter.
Two features of this class make it particularly rewarding to image, and one makes it difficult.
- Lipids are abundant, ionise well, and are strongly compartmentalised. Membrane composition varies between cell types and subcellular structures, so lipid images frequently recapitulate histology without any staining. This is why most method development in this field uses lipids.
- Metabolites report activity rather than potential. Transcript abundance indicates what a cell might do; metabolite distribution reflects what it did. For questions about function in place, that is a more direct measurement.
- Chemical diversity is the difficulty. The metabolome spans an enormous range of polarity, stability, and ionisation efficiency, so no single set of conditions detects all of it and every experiment is a partial view.
MSI Workflows for Small Molecules
Small-molecule imaging differs from peptide or protein imaging in ways that change the workflow rather than merely the settings. Four decisions dominate, and they interact.
Decision | Why It Matters for Small Molecules | Practical Guidance |
Ionisation approach | Matrix-derived ions crowd the low mass range, which is precisely where metabolites sit | MALDI with high mass resolving power, or DESI to avoid matrix interference entirely |
Ion mode | Acidic and basic metabolites favour opposite polarities, and one acquisition commits you to one | Run both modes on adjacent sections rather than compromising on a single one |
Matrix or solvent | Determines which chemical classes are extracted and ionised at all | Select for your target classes; expect to run more than one condition for broad coverage |
Pixel size | Coverage narrows as pixels shrink, and metabolites are less abundant than lipids | Coarser than instinct suggests for untargeted work, as set out in the trade-off article |
Table 1. The four workflow decisions that most affect small-molecule imaging. Each is treated in depth in the technique articles linked from this section.
The first row deserves emphasis because it is a genuine structural problem rather than a nuisance. In MALDI, the matrix is a small organic compound present in large excess, and it produces clusters and fragments across the low mass region where metabolites are found, as discussed in MALDI Imaging Mass Spectrometry: How It Works. That is a strong argument for high mass resolving power in metabolite work specifically, and part of why DESI and Ambient Ionization Imaging is attractive here despite its coarser resolution: no matrix means no matrix interference.
The fourth row runs against instinct and is worth stating plainly. For untargeted metabolite work, finer pixels reduce the number of species you detect, because the species that survive smaller sampling volumes are the abundant and readily ionised ones. Coverage is usually the objective in a discovery experiment, and coverage argues for coarser pixels. The reasoning and the arithmetic are in Spatial Resolution vs. Sensitivity in MS Imaging: The Fundamental Trade-off.
Why Is Annotation the Central Problem?
Because the measurement returns mass — and mass does not determine identity. This is the honest headline for the field, and it is worth stating in the field’s own terms rather than softening it.
The Field Calls It Dark Matter The team behind METASPACE, the standard community annotation engine for imaging MS, put the position directly in their 2024 Nature Communications paper on machine learning annotation: imaging mass spectrometry is a powerful technology enabling spatial metabolomics, yet metabolites can be assigned to only a fraction of the data generated, and the vast majority of imaging MS data, which they term dark matter, cannot be molecularly annotated with existing tools. That is an unusually candid statement from the people building the tools, and it should shape expectations for a first experiment. You will generate more detected features than you can name, and the gap is not a reflection of your method development. It is the current state of the field. |
The cause traces directly to a structural feature of imaging established elsewhere in this cluster. As a 2025 paper combining imaging with capillary electrophoresis states, the presence of isomers and isobars constitutes a bottleneck in molecular annotation, and imaging modalities operate without any chromatographic separation step, so they are not immune to this problem. The same absence of separation that forces reliance on mass resolving power, discussed in Mass Spectrometry Imaging: Principles, Techniques, and Applications, also forces annotation to rest on MS1 data alone.
What that means concretely is documented in the annotation literature. METASPACE performs annotation from MS1 data against a chosen database of molecular formulas, reporting ions at a controlled false discovery rate by ranking them against implausibly generated decoy ions. Because the evidence is MS1, it reports multiple possible isomers and isobars for each annotated ion, with the number determined largely by the database chosen, so a single annotated ion may be associated with a large number of molecular structures. An annotation, properly understood, is a formula with a confidence value and a list of candidates, not a molecule.
This makes the useful question not whether you can annotate, but what evidence supports each annotation. These are not equivalent claims.
Evidence Level | What It Establishes | What It Cannot Do | Typical Source |
Accurate mass alone | A set of candidate molecular formulas | Distinguish anything sharing that formula | The measurement itself |
Plus isotope pattern | Narrows formula assignment considerably | Separate isomers, which share a formula exactly | Spectral and spatial isotope scoring |
Plus spatial coherence | Confidence that the signal is real rather than noise | Anything about structure | Spatial chaos scoring within FDR-controlled annotation |
Plus collision cross-section | Discriminates many isomers and isobars | Resolve all structural ambiguity | Ion mobility, with reported accuracy under 0.4 percent against database values |
Plus fragmentation | Structural evidence approaching identification | Always be acquired, since on-tissue MS2 costs time | On-tissue tandem MS |
Table 2. An annotation confidence ladder for MS imaging. The ladder is our framing; each rung reflects evidence types documented in the cited annotation literature. Report which rung an annotation sits on.
Two routes up that ladder are worth knowing about. Ion mobility supplies collision cross-section values, and a study using cyclic ion mobility imaging of renal carcinoma tissue found that multipass experiments yielded cross-section accuracy better than 0.4 percent relative to database values, improving the filtering threshold used in earlier cross-section-based annotation workflows and enabling correct assignment where mass alone was ambiguous. Machine learning approaches have improved annotation coverage: the METASPACE-ML model was trained and evaluated on 1,710 datasets from 159 researchers across 47 laboratories, spanning animal and plant contexts, and reports higher precision and better identification of low-intensity metabolites than its rule-based predecessor. Annotation depth, database choice, and confidence reporting are treated fully in the annotation spoke linked at the end of this article.
Quantification Challenges
Signal intensity in an ion image is not proportional to concentration in any straightforward way, and the reason is chemical rather than instrumental. Ionisation efficiency depends on the local chemical environment, and that environment varies across a tissue section by definition, since heterogeneity is what you are imaging.
The mechanism is worth understanding because it explains why the problem is hard to correct. Ion suppression occurs when co-eluting or co-desorbing species compete for charge, so a region rich in abundant lipids will suppress signal from a trace metabolite relative to a region that is not. Work comparing imaging with capillary electrophoresis makes the inverse point instructively: part of the reason electrophoretic separation improved metabolite detectability was that concentrating an analyte into a band presents it for ionisation in a simpler chemical matrix, so there is less ion suppression. Imaging has no such simplification available, because the matrix is the tissue.
Four consequences for experimental design.
- Treat images as qualitative by default. Relative comparison of the same species between regions of similar composition is defensible. Comparison across regions of very different composition, or between species, is not.
- Normalisation is necessary and insufficient. Total ion current and related normalisations correct for some variation and can introduce artefacts of their own where composition varies strongly.
- Internal standards must be introduced deliberately. An isotopically labelled analogue applied uniformly to the section, or incorporated into the matrix, provides a reference that varies with local suppression as the analyte does.
- Validate against a bulk measurement. Extracting and quantifying a region by a conventional workflow provides the anchor that an image alone cannot.
This is an area where the field has been explicit about its own limitations, and where reporting practice matters more than technique. Describing an image as showing higher abundance in a region carries an implicit claim about ionisation being comparable between regions, and that claim should be justified rather than assumed. The subject is treated in full in the quantification spoke linked at the end of this article.
Applications in Disease and Pharma
Four application areas account for most published work, and they differ in how much of the annotation and quantification difficulty above actually bites.
Application | What Spatial Adds | Why It Works Despite the Limitations |
Tumour metabolism and margins | Metabolic gradients across a tumour and its margin, rather than a bulk average | Lipid signatures discriminating tissue types are abundant and reproducible, so annotation depth matters less |
Drug distribution | Where a compound and its metabolites actually reach in tissue | The analyte is known, so annotation is not the problem; the compound is administered rather than discovered |
Lipid biology and membrane structure | Compartment-specific lipid composition, often recapitulating histology | Lipids ionise well and are abundant, making them the most tractable class |
Microbial and plant metabolism | Exchange of specialised metabolites at interaction zones | Community databases for specialised metabolites now support annotation in these contexts |
Table 3. Established application areas. Note the pattern in the third column: the applications that have matured fastest are those where the annotation problem is smallest.
That pattern in the third column is the practically useful observation. Drug distribution studies are the most industrially established application of MS imaging precisely because the analyte identity is known in advance, which removes the field’s hardest problem from the experiment entirely. Lipid work has matured for a different reason: abundance and ionisation efficiency make the measurement robust even where structural assignment remains ambiguous.
The microbial case shows what targeted database work can achieve. A study applying METASPACE to microbial specialised metabolites from agar-based imaging annotated 53 ions corresponding to 32 specialised metabolites that were verified against taxonomic classification, while noting that these annotations remain putative on MS1 evidence including exact mass, isotopic pattern, spatial chaos, and co-localisation of isotopic ions. That is the ladder in Table 2 being climbed explicitly, and reported as such.
What Should You Expect From a First Experiment?
Calibrated expectations prevent a successful experiment being read as a failure. Five things to anticipate.
- More features than identities. You will detect substantially more distinct ions than you can confidently name. This is the field’s normal condition, not a shortcoming of your workflow.
- Lipids first. Your clearest and most reproducible images will almost certainly be lipids, even if metabolites were the objective. Treat that as a positive control that the workflow functions.
- Mode-dependent results. Positive and negative ion mode will return substantially different species lists. A single-mode experiment has seen part of the picture.
- Ambiguous annotations. Expect candidate lists rather than molecules, and record which evidence level supports each assignment from the outset rather than reconstructing it later.
- Qualitative conclusions. Plan the first experiment to answer where something is, not how much of it there is, unless quantification was designed in.
The productive framing for a first study is discovery followed by confirmation. Use imaging to find which species and which regions are worth attention, accepting ambiguity at that stage, then confirm the specific findings that matter by an orthogonal route: on-tissue fragmentation, ion mobility, or extraction of the region of interest followed by a conventional separation-based workflow. That sequence plays to the technique’s strength, which is telling you where to look, and compensates for its weakness, which is telling you exactly what you are looking at.
This section develops each stage. Untargeted method development end to end is in Untargeted Spatial Metabolomics Workflows; the most tractable analyte class and its specific isomer problems in Spatial Lipidomics in Tissue; annotation, databases, and confidence reporting in Metabolite Annotation and Databases for MS Imaging; and what quantitative can defensibly mean in Quantitative Mass Spectrometry Imaging: Challenges and Approaches.
This article was produced under Separation Science's AI Editorial Guidelines.
