Attempting quantitative mass spectrometry imaging means accepting that an ion image is a map of signal, not of concentration, and then doing the method development that closes the gap between them. The goal is worth stating precisely: distribution analysis and quantitation from a single mass spectrometry experiment. Achieving it depends less on the instrument than on how the calibration is constructed.
Key Takeaways
|
Why MSI Quantification Is Difficult
Because the sample is the interference — and that is not a figure of speech. In a conventional workflow, extraction and separation deliver the analyte to the ion source in a comparatively simple and reproducible chemical environment. In imaging, the analyte is ionised in situ, surrounded by whatever else is present at that position, and the composition of that surrounding material is exactly what varies across the tissue.
Four consequences follow, each of which has to be addressed rather than tolerated.
- Response varies spatially. The same concentration of analyte in two compositionally different regions will not produce the same signal, so intensity differences between regions conflate concentration with local ionisation efficiency.
- Extraction efficiency varies. How readily an analyte is released from tissue into the matrix layer or solvent depends on the tissue itself, not only on the analyte.
- Depth is not controlled. The signal comes from an ablated or extracted volume of uncertain depth, so converting to concentration requires an assumption about the volume sampled.
- Signal drifts. Over an acquisition that may last hours, instrument response and matrix condition change, so early and late pixels are not directly comparable without correction.
The general mechanism and its design consequences are set out in Spatial Metabolomics and Lipidomics by Mass Spectrometry Imaging. What follows here is how the method development community has responded to each.
Working in analytical science?
Register for a FREE Separation Science account to subscribe to the Separation Science Newsletter.
Ion Suppression and Matrix Effects
Ion suppression is competition for charge. When many species desorb together, those that ionise most readily take a disproportionate share of the available charge, and the rest are suppressed. Because tissue composition varies regionally, suppression varies regionally too.
Two implications are worth being explicit about, because they constrain what comparisons are legitimate even before calibration is considered. Comparing the same species between compositionally similar regions is reasonably defensible, since suppression should be broadly comparable. Comparing the same species between very different regions, for instance a lipid-rich white matter region against grey matter, conflates concentration with suppression. And comparing two different species to each other is not defensible at all, since their ionisation efficiencies differ by unknown factors.
One structural point worth noting for method selection: ambient approaches avoid the matrix-derived interference that MALDI introduces in the low mass range, which removes one contributor to suppression in metabolite work, as discussed in DESI and Ambient Ionization Imaging. That does not eliminate tissue-derived suppression, which is the dominant term, but it simplifies the problem slightly.
Which Calibration Approach Should You Use?
This is the decision that determines whether a quantitative claim holds, and the literature is refreshingly explicit about what each route assumes.
A tutorial on quantitative mass spectrometry imaging in the Journal of Mass Spectrometry notes that absolute quantitative imaging protocols all use calibration curves, but that these differ from the calibration curves generated in a conventional experiment. Work on quantitative imaging of bleomycin in skin states the position directly: the two most common calibration methods in imaging are deposition of standards on control tissue and preparation of spiked tissue homogenates.
| Droplet Deposition on Control Tissue | Mimetic Tissue Model |
Method | Standards applied as droplets, typically 1 to 2 microlitres, onto control tissue sections | Tissue homogenate spiked with standard at known concentration, set in a mould, then sectioned at the same thickness as the sample |
Effort | Relatively quick to perform | Considerably more work, including homogenate preparation and moulding |
Assumption 1 | That the deposited compound distributes homogeneously through the entire thickness of tissue below the droplet, since average concentration is calculated from the area covered | None equivalent. The standard is distributed through the homogenate by construction |
Assumption 2 | That a compound deposited on top of tissue undergoes the same interactions and ion suppression as if it were an integral part of the tissue | None equivalent. The standard is an integral part of the matrix |
Extraction behaviour | Surface-deposited analyte may be extracted more readily than analyte within tissue | Better represents analyte extraction, because the standard is fully integrated with the tissue |
Best suited to | Rapid work of limited quantitative scope, such as establishing limits of detection | Absolute quantification where accuracy matters |
Table 1. The two established calibration architectures and the assumptions each requires. The assumptions attributed to droplet deposition are those stated in the cited literature, not inferred.
Why the Assumptions Decide It Both droplet deposition assumptions are questionable in ways that matter. A droplet applied to a section surface has no reason to migrate uniformly through its full thickness, yet the concentration calculation assumes it does. And an analyte sitting on top of tissue is in a materially different chemical environment from one distributed within it, so its suppression will differ. Mimetic tissue models are described as one of the most accurate ways to perform quantitative imaging precisely because they sidestep both: the analyte is present at a well-defined concentration in a sample matrix highly similar to that of the unknown sample. As the tutorial literature puts it, because the standard is fully integrated with the tissue, this gives a better representation of analyte extraction than on-tissue spotting. The practical rule follows. Use droplet deposition when the quantitative scope is limited, for instance to establish a detection limit. Use a mimetic tissue model when you intend to report concentrations. |
One practical warning on mimetic models: tissue type materially affects feasibility. Skin, because of its high collagen content, produces homogenates that become extremely viscous, and the published protocol addresses this by adding water and handling the homogenate at elevated temperature where viscosity is lower. Expect to develop the homogenate preparation for your tissue rather than adopting a protocol unchanged, and note that variants of the droplet approach, including the use of a tissue extinction coefficient, remain in use where mimetic models are impractical.
Internal Standards and Normalisation
Calibration establishes the relationship between signal and concentration. An internal standard corrects for the fact that the relationship drifts — both across an acquisition and across a section.
An isotopically labelled analogue of the analyte is the ideal, because it experiences the same suppression and the same extraction behaviour while remaining distinguishable by mass. Published work applies exactly this: a study using targeted imaging in a multiple reaction monitoring mode employed a deuterated internal standard to normalise data from pixel to pixel, with a mimetic in-tissue model providing the calibration curve, reporting good linearity with an R squared of 0.9953. The bleomycin skin study likewise included an internal standard specifically to compensate for signal drift and for inhomogeneities in the skin.
Strategy | What It Corrects | Limitation |
Isotopically labelled analogue applied uniformly | Regional suppression and drift, with the closest possible match to analyte behaviour | Requires a labelled standard to exist and be affordable; uniform application is itself a method development problem |
Structural analogue as internal standard | Drift and gross suppression differences | Behaves similarly but not identically, so correction is approximate |
Internal standard incorporated into the matrix solution | Drift and matrix application variation | Applied above the tissue rather than within it, so extraction differences remain |
Total ion current normalisation | Broad acquisition-level variation | Can create artefacts where total composition varies strongly between regions, which is common in tissue |
Median or root mean square normalisation | Similar scope to TIC with different sensitivity to outliers | Same fundamental caveat: it assumes total signal is comparable between regions |
Reference ion normalisation | Variation tracked by an endogenous species assumed uniform | Requires a genuinely uniform endogenous species, which is difficult to establish |
Table 2. Internal standard and normalisation strategies. The top row is the strongest option where a labelled standard is available; the lower rows correct less and risk more.
On normalisation specifically, the honest guidance is to test more than one approach and check whether your conclusions survive the choice. Normalisation that assumes total signal is comparable between regions is doing something questionable in a sample whose regional composition differences are the object of study. If a finding appears under total ion current normalisation and disappears under another, that is information about the finding rather than about the normalisation.
How Do You Validate a Quantitative Result?
Against an orthogonal measurement on comparable tissue, and this is the step that separates a quantitative claim from an intensity map with units attached.
The bleomycin study provides a good template. Quantification was performed at several regions in a cross-section of skin at the injection site, and the results were compared against quantitative LC-MS on a neighbouring tissue biopsy from the same animal experiment. The overall tissue concentration determined by LC-MS fell within the range of the different regions quantified by imaging. The authors concluded that because the model gives results of the same order of magnitude as LC-MS, it can either replace LC-MS in skin studies where the two are currently performed in combination, or add quantitative information to studies otherwise carried out by imaging alone.
Three features of that design are worth copying.
- The comparison was regional against bulk. A bulk measurement cannot validate a single pixel, but it can be checked for consistency with the range of regional values, which is the appropriate test.
- The comparison tissue was neighbouring and from the same experiment. Using a different animal or a different experiment introduces biological variation that would obscure the methodological question.
- The claim was calibrated to the evidence. Same order of magnitude, stated as such. That is a defensible conclusion from a bulk-against-regional comparison and does not overclaim precision.
Where absolute figures are required for a protein or peptide, similar approaches have been applied with surface extraction methods: quantitative work using liquid extraction surface analysis used spiked liver homogenate as a control matrix, frozen and sectioned for calibration curves, having first optimised extraction solvent composition, tissue thickness, and solvent contact time, and compared results against LC-MS/MS. A broader survey of quantitative imaging of therapeutics and biomolecules documents limits of quantification determined by mimetic tissue model for a protein, at 68.6 and 163 nanomoles per gram, which illustrates that the approach extends beyond small molecules.
What Should You Report?
Enough that a reader can judge the claim rather than take it. Six items, and the first two are the ones most often missing.
Report | Why It Is Necessary |
Which calibration architecture was used | Droplet deposition and mimetic tissue models carry different assumptions, so the reader cannot assess the result without knowing which applies |
The internal standard and how it was applied | Correction quality depends on whether the standard was labelled, structurally analogous, or absent, and on how uniformly it was delivered |
The normalisation applied, and alternatives tested | Normalisation choice can create or remove apparent structure, so stating one without alternatives leaves the result unverifiable |
Section thickness for both sample and calibrant | Concentration conversion depends on sampled volume, and mimetic models require thickness matching to be valid |
Validation against an orthogonal method, if performed | Without it, a quantitative claim rests entirely on the internal consistency of the calibration |
The claim at the precision the evidence supports | Order of magnitude agreement is a legitimate finding. Stating concentrations to three significant figures from an uncalibrated intensity map is not |
Table 3. A reporting checklist for quantitative imaging results. The last row is a matter of scientific honesty rather than convention.
The wider point is that quantitative imaging is best understood as a method development undertaking rather than a setting. It requires a calibration architecture chosen deliberately, an internal standard strategy, a normalisation decision that has been tested, and ideally an orthogonal validation. Programmes that treat quantification as something to attempt after the images look good generally find they need to repeat the acquisition.
Sample preparation decisions upstream of all of this, including sectioning thickness and mounting, are covered in Sample Preparation for Spatial Analysis: From Tissue to Data. The related question of what an annotation claims, which is separate from how much of it is present, is treated in Metabolite Annotation and Databases for MS Imaging, and the untargeted discovery workflow that usually precedes quantitative work is in Untargeted Spatial Metabolomics Workflows. For where quantification sits in the wider spatial landscape, see Spatial Analysis in Analytical Science: Mass Spectrometry Imaging and Spatial Omics.
This article was produced under Separation Science's AI Editorial Guidelines.
