Spatial transcriptomics
A capture-array method: the section is mounted on a slide printed with spots that each carry a spatial barcode, imaged by brightfield through a fiducial frame, permeabilised so mRNA binds the probes beneath it, reverse transcribed on the slide and sequenced. Whole transcriptome, spot resolution, one or several cells per spot.
Our ruleSimulations prepare you for the bench but don't replace hands-on practice.
Visium
A capture-array method: the section is mounted on a slide printed with spots that each carry a spatial barcode, imaged by brightfield through a fiducial frame, permeabilised so mRNA binds the probes beneath it, reverse transcribed on the slide and sequenced. Whole transcriptome, spot resolution, one or several cells per spot.
How it differs from the standard method
- Works from fresh-frozen or FFPE sections, with a probe-based chemistry for the degraded case.
- Resolution is set by the printed spot size and pitch, so each spot is a mixture that usually needs deconvolution.
- Permeabilisation time is tissue-specific and is normally optimised on a dedicated slide before the real run.
- The brightfield image and the fiducial alignment are what give the barcodes their coordinates.
1Measuring expression without throwing the coordinates away
Every expression method before this one started by destroying the tissue. You lyse a biopsy, or you dissociate it into single cells, and in both cases the first step discards the thing a pathologist actually looks at: which cell sat next to which.
2The one trade the whole field runs on
There are two ways to keep the position, and they fail in opposite directions. Capture-based methods bring the transcripts to a barcoded surface: the section sits on an array where every spot carries its own spatial barcode, the tissue is permeabilised so released mRNA binds the probes directly underneath, and the library is sequenced like any other.
Whole transcriptome, coarse
Every gene, no panel to choose, so it is the right tool when you do not yet know what you are looking for. A spot holds several cells, so each profile is a mixture.
Single molecule, narrow
Molecule coordinates inside segmented cells, so cell types and sub-cellular localisation are real. Nothing outside the panel exists in the data.
Unmixing a spot
Statistical methods estimate the cell-type proportions in a spot against a matched single-cell reference. The answer is only as good as the reference.
Whose transcript is it?
In imaging methods, drawing cell boundaries is the step that decides whose expression a molecule counts towards, and getting it wrong invents cell states.
3Permeabilisation is the parameter that decides everything in capture methods
In a capture experiment the tissue has to be opened enough that mRNA leaves the cells and reaches the probes a few micrometres below. Too short and most transcripts never arrive, so UMIs per spot collapse and the map is empty.
4What limits the data before you start
RNA integrity in the section gates everything and cannot be recovered afterwards, so it is measured on an adjacent section before the capture area is committed: a cryostat used for routine histology, a block that thawed in transit or an old FFPE block can all give a beautiful-looking section with half its RNA gone. Section thickness trades transcripts per spot against how many cells a spot holds and how evenly the permeabilisation reaches through the depth.
5How it misleads
- A spot's apparent co-expression of two markers is usually two neighbouring cells, not one cell doing both things.
- A blurred boundary from over-permeabilisation looks like a real expression gradient, and nothing in the standard QC flags it.
- In imaging methods, a fixed nuclear expansion makes small cells inherit their large neighbours' transcripts, which produces clusters that look like hybrid cell types and scale with local cell density.
- A high negative-control-probe rate means a fixed fraction of every gene's counts are invented, and it rises with plex.
- Comparing counts per gene between a low-plex and a high-plex run on the same tissue is not valid: detection efficiency differs.
Key terms
Practise these as flashcards →A sequence unique to one position on a capture array, added to every transcript caught there, so a read can be traced back to a place.
Printed markers around the capture area used to align the brightfield image to the barcode coordinates. Without the alignment there are no coordinates.
Percentage of RNA fragments longer than 200 nt; the integrity metric used for degraded and FFPE material, where RIN is uninformative.
How long the section is digested before capture. Too short loses transcripts, too long lets them diffuse sideways and blurs the map.
Estimating the cell-type composition of a multi-cell spot by fitting it against a single-cell reference of the same tissue.
The fraction of reads that are duplicates of molecules already seen; once it is high, deeper sequencing adds no new molecules.
The barcode scheme used in imaging methods. Spare codewords at a distance of four let a single-round error be corrected and a double error be rejected.
A probe matching no sequence in the sample, or a codeword matching no probe; its count is the false-decode rate of the run.
Drawing the boundary that decides which cell a transcript is counted towards, from nuclei, a fixed expansion, or an imaged cell boundary.
When to use it
- The question is about where in the tissue, or about which cells are neighbours, and dissociation would destroy the answer.
- You have a cell-type atlas from single-cell sequencing and need to place those types back into the intact organ.
- You want an unbiased, whole-transcriptome view of a tissue region and can live with several cells per measurement.
- You want single-cell and sub-cellular resolution for a defined set of genes you can specify in advance.
- Only sectioned material exists, including archival blocks, and a dissociation protocol is not an option.
Compared with alternatives
Prefer it when you need whole transcriptomes per cell and can accept losing the tissue coordinates; the two are usually run together, one as the reference for the other.
Cheaper and sharper for one or a few transcripts you have already chosen; spatial transcriptomics is what you use when you do not yet know which genes to probe.
Bulk sequencing gives far deeper, cheaper coverage of a whole sample, but one averaged profile with no position at all.
Still the standard for reading protein in a section with familiar morphology, and the obvious orthogonal check on a spatial result.
The quality of the section (thickness, flatness, RNase-free handling) sets the ceiling on everything a spatial run can give.
Sub-techniques
Choosing one here switches the whole page, including the protocol and the lab, to it.
A capture-array method: the section is mounted on a slide printed with spots that each carry a spatial barcode, imaged by brightfield through a fiducial frame, permeabilised so mRNA binds the probes beneath it, reverse transcribed on the slide and sequenced. Whole transcriptome, spot resolution, one or several cells per spot.
- Works from fresh-frozen or FFPE sections, with a probe-based chemistry for the degraded case.
- Resolution is set by the printed spot size and pitch, so each spot is a mixture that usually needs deconvolution.
- Permeabilisation time is tissue-specific and is normally optimised on a dedicated slide before the real run.
- The brightfield image and the fiducial alignment are what give the barcodes their coordinates.