Pursuing multimodal mass spectrometry imaging alongside spatial transcriptomics means combining what MS sees — metabolites and lipids — with what sequencing-based platforms see, transcripts, at the same location in the same tissue. Each modality alone gives a partial view. Together, and only together, they let a researcher ask whether a transcriptional program and its downstream metabolic consequence occur in the same cells.
Key Takeaways
|
Why Combine MSI With Transcriptomics?
Because the molecular layers do not simply track one another, and that disagreement is informative rather than merely a nuisance to be averaged away. A same-section spatial transcriptomics and proteomics study of human lung cancer resolved transcript and protein levels at cellular resolution for the same physical cells and found systematic low correlation between the two layers for the same gene, consistent with decades of prior findings from bulk tissue. If a gene’s transcript abundance reliably predicted its protein abundance, and by extension its downstream metabolic effect, there would be limited value in measuring more than one layer. It does not, reliably, which is precisely the justification for spatial multi-omics rather than a single modality read in isolation.
Working in analytical science?
Register for a FREE Separation Science account to subscribe to the Separation Science Newsletter.
Applied to MSI plus transcriptomics specifically, the same logic holds: a transcriptional signature for a metabolic pathway does not guarantee that the corresponding metabolite is actually present at elevated levels in the same location, because transcription, translation, and metabolic flux are all independently regulated steps. Only a measurement that captures both layers, at the same physical position, can test whether they align in a specific tissue and a specific disease state.
Same-Section or Serial-Section Integration?
This is the first and most consequential design decision — it trades destructiveness for precision in one direction, or non-destructiveness for registration uncertainty in the other.
| Same-Section | Serial-Section |
Spatial correspondence | Exact. Both modalities measure the identical physical cells | Approximate. Adjacent sections differ by at least the section thickness, and small offsets sample different cells |
Destructiveness | MS ablation affects the tissue before the second modality runs, reducing its yield | Non-destructive to either modality, since each has its own section |
Registration approach | Ablation marks or other in-tissue features serve as shared landmarks | Requires a dedicated non-rigid registration algorithm between genuinely different sections |
Best suited to | Studies where single-cell correspondence between layers is the point | Studies where tissue is abundant and modality-specific preparation requirements conflict |
Table 1. Same-section against serial-section multimodal integration. Most published same-section workflows accept a documented reduction in transcript yield as the price of exact correspondence.
One described same-section workflow makes the mechanism concrete. MALDI-MSI integrated with Xenium spatial transcriptomics on a single tissue section exploits something already present on the slide: laser ablation spots from the MS run are readily visible in the fluorescence channel used for the sequencing platform’s own cell segmentation, providing precise markers for pixel-scale alignment without any additional labelling step. Single cells detected by the transcriptomics platform’s segmentation were then linked directly to the MALDI spectrum overlapping each cell boundary, achieving per-cell extraction of both a metabolic profile and a transcriptomic one from the same physical cell. This is the same self-generated-landmark principle covered in Coregistration With Histology and Multimodal Imaging, applied here to link two molecular layers rather than an ion image to histology.
That same study quantified the cost of the destructive step directly, validating on mouse brain and demonstrating the approach on human glioblastoma tissue: the number of transcripts recovered per cell decreased by approximately 30 percent following MSI, while cell recovery and cell-type assignment accuracy were preserved. The practical reading is that same-section integration is not free, but the cost is a moderate reduction in sequencing depth rather than a loss of the biological signal the second modality is meant to provide.
Where same-section analysis is not viable, because tissue is limited or the preparation requirements genuinely conflict, serial-section integration is the alternative, and it requires its own dedicated registration step rather than a simple overlay. The Multi-Omics Imaging Integration Toolset, MIIT, a Python framework for integrating spatially resolved multi-omics data from serial sections, addresses this with a purpose-built non-rigid registration algorithm named GreedyFHist, validated against 244 images from fresh-frozen serial sections and reported to achieve state-of-the-art registration performance on that benchmark. As a proof of concept, the same framework was used to integrate spatial transcriptomics and MSI data from prostate tissue, correlating a gene signature for a specific metabolic pathway derived from the transcriptomic data against the corresponding metabolic measurements from MSI on the adjacent section.
Resolution Mismatch Between Platforms
MSI and sequencing-based spatial transcriptomics are fundamentally different kinds of spatial measurement, and reconciling them is a distinct problem from registration itself. MSI produces a regular grid of pixels, each an independent measurement at a fixed spacing set by the raster step. Imaging-based transcriptomics platforms, by contrast, segment the image into irregular cell boundaries first and report one measurement per cell, while sequencing-based spot platforms report one measurement per spot at their own fixed spacing, which rarely matches an MSI pixel exactly.
Three practical situations follow, and each demands a different resolution strategy.
- MSI pixel size is finer than the transcriptomics unit. Aggregate multiple MSI pixels to match each cell or spot, typically by averaging or summing the spectra falling within its boundary. This is the situation in the pixel-to-cell-boundary approach described above.
- MSI pixel size is coarser than the transcriptomics unit. Multiple cells or spots will fall within a single MSI pixel, so the metabolic measurement necessarily represents an average over that population, and single-cell metabolic resolution is not recoverable regardless of how finely the transcriptomic side was measured.
- The two grids do not align at all. Requires an explicit spatial interpolation or resampling step onto a shared coordinate system before any per-unit comparison is meaningful, introducing its own approximation.
The first situation is the most favourable and increasingly achievable as MSI resolution improves, which is why the resolution-versus-sensitivity trade-off covered in Spatial Resolution vs. Sensitivity in MS Imaging: The Fundamental Trade-off has direct consequences here: an MSI acquisition run at coarse resolution to preserve sensitivity may simply be incompatible with single-cell multimodal integration, independent of any registration accuracy, because the pixel already averages over multiple cells before registration is even considered.
Joint Analysis Approaches
Once registered and resolution-matched, a joint dataset supports analysis that neither modality alone can, and the appropriate method depends on what question is being asked of the combined data.
- Per-unit correlation. For each matched cell or region, correlate a specific transcript or gene signature against a specific metabolite or lipid, testing directly whether the expected biological relationship holds in that tissue rather than assuming it from prior literature.
- Joint clustering or classification. Combine both feature sets before clustering, which can reveal cell states or regions that neither modality distinguishes on its own, since a subtle transcriptional difference and a subtle metabolic difference together may separate populations that either alone leaves merged.
- Pathway-level integration. Rather than comparing individual transcripts to individual metabolites, aggregate to a pathway or signature level, as in the prostate tissue proof of concept above, which is often more robust to noise in either individual modality.
Machine learning approaches increasingly sit inside this stage as well, particularly for joint clustering across high-dimensional combined feature sets, and are treated in depth in Machine Learning for Mass Spectrometry Imaging Data. The broader statistical and software foundation this all sits on is covered in Analyzing Mass Spectrometry Imaging Data: Processing, Statistics, and Multimodal Integration.
Emerging Multimodal Platforms and Toolsets
The tooling for this specific combination is young relative to either modality individually, and it is developing along two lines that mirror the same-section-versus-serial-section split above.
For same-section work, the enabling development is not new registration software so much as the recognition that the ablation process itself provides usable landmarks, removing the need for separately introduced fiducial markers. For serial-section work, dedicated non-rigid registration algorithms benchmarked specifically on this problem, rather than adapted from general medical image registration, are what make reliable cross-section correspondence achievable at all.
For a reader building this capability, the practical starting point is the same as for any spatial data analysis question: match the tool to what the question actually requires, described generally in MS Imaging Data Processing and Software, then verify format compatibility between the MSI side and whichever transcriptomics analysis ecosystem is in use before committing to a specific pipeline. Broader computational context for the transcriptomics side of that pipeline, including cell segmentation and typing conventions, is covered in Spatial Multi-Omics: Combining Transcriptome, Proteome, and Epigenome in Tissue.
For where multimodal integration sits within 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.
