Spatial transcriptomics and spatial multi-omics

Dissociating a tissue destroys the information that made it a tissue. Spatial methods keep the coordinates, and they do it in two fundamentally different ways whose limitations are almost mirror images.

Capture-based methods: unbiased, but the unit is not a cell

A sectioned tissue is placed on a surface carrying spatially barcoded capture probes. Permeabilisation releases messenger RNA, which binds the probe beneath it and carries that position’s barcode into the sequencing library. Because capture uses the poly-A tail, the readout is transcriptome-wide: no target list is chosen in advance, and unexpected genes can be found.

The compromise is spatial unit and efficiency. Early array formats had capture spots tens of micrometres across containing several cells, so each measurement is a mixture. Making the spots smaller raises resolution but reduces the number of transcripts captured per spot, because there is less area and less material — resolution and sensitivity trade against each other directly. High-definition formats now reach the scale of a cell or below, at correspondingly sparse counts per unit, which are usually binned back up for analysis.

Where a spot contains several cells, the observed profile is a weighted sum of cell types. Deconvolution estimates those weights using a single-cell reference from the same tissue, and its output is a composition estimate, not an observation. It fails predictably where the reference lacks a cell type present in the section, and its confidence depends on how distinct the reference profiles are.

Imaging-based methods: single molecules, but a chosen panel

The alternative labels transcripts in place with fluorescent probes and reads them by microscopy, using combinatorial rounds of hybridisation and imaging so that each gene has a distinct code across rounds, with error-correcting design so misreads can be detected. Resolution is that of the microscope: individual molecules are localised, and with a cell boundary stain the assignment to cells is direct rather than deconvolved. Detection efficiency per molecule is generally higher than capture.

The compromise is the panel. Every gene must be designed in as probes, so the experiment measures hundreds to a few thousand pre-selected transcripts and is blind to everything else. Panel choice therefore encodes the hypothesis. Optical crowding limits how many molecules can be resolved in a dense cell, and segmentation errors — assigning a molecule to the wrong neighbouring cell — are a real and quantifiable source of apparent co-expression.

Adding protein and other layers

Protein can be measured on the same or an adjacent section using antibodies with oligonucleotide tags read out in sequencing, or cyclic immunofluorescence, in which antibodies are imaged then stripped or bleached and the cycle repeated. This is where “spatial multi-omics” usually begins, and its practical difficulty is registration: aligning sections and modalities accurately enough that a joint claim about one cell is warranted.

Chromatin-state measurements in tissue are a different technology with different constraints, covered separately under spatial epigenomics; this page does not describe them.