Spatial transcriptomics measures gene expression while preserving the spatial architecture of the tissue. Unlike scRNA-seq — which loses location information when you dissociate the tissue — spatial methods tell you not just which genes are expressed, but where.
Why spatial information matters
Tissue function depends on spatial organization: which cells touch which, how far signaling molecules travel, how immune cells infiltrate tumors, how stem cell niches are organized. A tumor with the same scRNA-seq profile can have wildly different prognoses depending on whether immune cells are excluded, surrounding the tumor, or infiltrated. Spatial context is biology you can’t recover from dissociated single-cell data.
The two technological approaches
Sequencing-based
Tissue is placed on a slide with spatially barcoded capture spots. mRNA from each spot is captured, the spatial barcode tags it, and the result is sequenced. Each spot’s transcriptome is positioned in 2D coordinates.
Examples: 10x Genomics Visium (55 µm spots, ~1–10 cells per spot), Visium HD (2 µm bins, near single-cell), Stereo-seq (sub-cellular resolution).
Imaging-based (in situ)
Targeted gene panels are detected directly in tissue using fluorescent probes, with multiple rounds of imaging encoding gene identity. Subcellular resolution and multiplex imaging in the same section.
Examples: 10x Xenium, NanoString CosMx, Vizgen MERSCOPE/MERFISH.
Comparison
| Feature | Visium | Xenium / CosMx / MERFISH |
|---|---|---|
| Resolution | ~55 µm spots (HD: 2 µm bins) | Subcellular |
| Genes profiled | Whole transcriptome | Targeted panels (100–5,000 genes) |
| Throughput | Lower (per slide) | Higher (single section) |
| Cost per sample | Higher | Variable |
| Sample type | Fresh frozen + FFPE | FFPE-friendly |
Standard workflow
- Sample preparation — fresh frozen tissue sections or FFPE
- Tissue placement on slide
- Permeabilization and capture (sequencing-based) or hybridization (imaging-based)
- Library prep and sequencing, or imaging cycles
- Computational alignment of expression to histology image
- Analysis: clustering, deconvolution, neighborhood analysis
Analysis approaches
- Spot-level clustering: Identify spatial domains corresponding to different tissue regions
- Cell-type deconvolution: For sequencing-based methods, infer cell types per spot using paired scRNA-seq references (Cell2location, RCTD, SPOTlight)
- Spatially variable gene detection: Genes whose expression depends on location (SpatialDE, SPARK)
- Cell-cell interaction inference: Use neighbor relationships to identify likely paracrine signaling
Common applications
- Tumor microenvironment mapping
- Brain atlases (regional gene expression patterns)
- Developmental biology (spatial gradients during organogenesis)
- Inflammation studies (immune infiltrate organization)
- Drug response in tissue context
Common pitfalls
- Tissue quality matters enormously. RNA integrity drops fast in fresh-frozen tissue handling and in old FFPE blocks
- Sample alignment errors — image registration to expression data must be checked carefully
- Resolution mismatch — Visium spots contain multiple cells; deconvolution is required for cell-type inference
- Targeted panels miss biology — imaging-based methods only detect what’s on the panel
The future
Subcellular-resolution sequencing (Stereo-seq, Visium HD) and large-panel imaging (1000+ genes) are converging. Pair this with spatial proteomics (CODEX, IMC), and you can describe tissue architecture in extraordinary detail.
Spatial transcriptomics complements scRNA-seq, doesn’t replace it. Use scRNA-seq for cell-type granularity and dynamics; use spatial for architecture and tissue-level context. Many of the strongest current studies pair the two.
