Executive Summary
Spatial transcriptomics (ST) bridges the gap between histology and genomics, allowing researchers to map gene expression back to tissue architecture. However, the high cost of reagents (particularly for platforms like 10x Genomics Visium) means there is zero room for error.
Key Takeaways:
- Sample Prep is Deterministic: The success of a Visium experiment is often decided before the slide enters the sequencer. Tissue optimization and RNA quality are critical.
- The Resolution Choice (Standard vs. HD): Standard Visium spots (55µm) require deconvolution to resolve cell mixtures. The newer Visium HD (2µm) offers single-cell resolution but generates significantly larger datasets.
- Integration is Key: The most powerful insights come from integrating spatial data with matched single-cell RNA-seq (scRNA-seq) references.
Phase 1: Experimental Design & Sample Prep
The "garbage in, garbage out" rule applies tenfold to spatial transcriptomics. Unlike bulk RNA-seq, where you can sometimes salvage poor libraries, spatial assays rely on the physical integrity of RNA within a tissue slice.
1. The CytAssist Revolution (FFPE & Fresh Frozen)
Historically, Visium required placing fresh frozen tissue directly on the slide, which was technically unforgiving. The Visium CytAssist instrument has changed the game. It allows you to prepare tissue on standard glass slides (both FFPE and Fresh Frozen) and then transfer the analytes to the Visium slide.
Decision Matrix:
- Visium Direct Placement (Fresh Frozen only): Choose this only if you need unbiased poly-A capture (e.g., for non-standard model organisms) and have pristine tissue. Requires difficult tissue optimization.
- Visium CytAssist (FFPE or Fresh Frozen): The preferred route for most human/mouse studies. It uses probe-based chemistry (transcriptome-wide), which is more robust to RNA degradation and eliminates the need for tissue optimization slides.
3. The Tissue Optimization Step (Direct Placement Only)
If you are using the Direct Placement method for fresh frozen tissue (not CytAssist), you cannot skip the Tissue Optimization (TO) slide. This step determines the optimal permeabilization time.
- Visium Standard (v2): Features 55µm capture spots. Great for mapping general tissue architecture and domains. Requires deconvolution to infer cell types.
- Visium HD: Features 2µm continuous bins. Provides near single-cell resolution without gaps. Note: HD datasets are massive. Ensure your computational infrastructure (RAM/Storage) is prepared for the scale increase.
2. The Tissue Optimization Step
For fresh frozen samples, you cannot skip the Tissue Optimization (TO) slide. This step determines the optimal permeabilization time to release RNA from the tissue onto the capture slide.
- Under-permeabilization: RNA stays trapped in the cell; you get low sensitivity.
- Over-permeabilization: RNA diffuses laterally before capture; you lose spatial resolution and get "blurry" data.
4. Imaging: The Unsung Hero
Your gene expression data is only as good as the image you map it to. High-resolution H&E (Hematoxylin and Eosin) or Immunofluorescence (IF) imaging is required for accurate spot alignment. Ensure your microscope settings (exposure, focus) are perfected before the actual experiment.
Phase 2: Sequencing Strategy
Sequencing costs are a significant portion of the budget. Planning your depth correctly is essential to avoid under-powered analysis.
Sequencing Depth Strategy
For a standard Visium capture area (~5,000 spots), 10x Genomics recommends a minimum of 25,000 read pairs per spot. However, for robust analysis, we recommend aiming higher.
- Minimum (< 25k reads/spot): Sufficient for defining major anatomical regions and highly expressed genes.
- Recommended (> 50k reads/spot): Ideal for deconvolution, receptor-ligand analysis, and detecting low-abundance markers (transcription factors).
Note for Visium HD: Due to the binning approach, sequencing requirements are calculated per area or per bin. Consult the latest user guide, as HD libraries often require significantly deeper sequencing (often 500M+ reads/slide).
Phase 3: The Computational Pipeline
Once the FASTQ files arrive, the real work begins. The bioinformatics workflow for spatial data is distinct from standard scRNA-seq.
Step 1: Alignment & QC (SpaceRanger)
The 10x Genomics SpaceRanger pipeline aligns reads to the genome and maps them to the image via fiducial markers.
Expert Tip: Always manually inspect the `web_summary.html` output. Automated alignment frequently fails if tissue covers the fiducial frame. Manual alignment (using the Loupe Browser) is often necessary. See our guide on visualization best practices.
Step 2: Spot Deconvolution (Standard Visium)
A standard Visium spot is 55µm in diameter. Depending on tissue density, one spot can contain 1 to 10+ cells. To resolve this, we use Deconvolution algorithms that leverage a matched single-cell reference (scRNA-seq) to estimate the proportion of cell types in each spot.
ey interacting, or just neighbors?To resolve this, we use Deconvolution algorithms that leverage a matched single-cell reference (scRNA-seq) to estimate the proportion of cell types in each spot.
- RCTD (Robust Cell Type Decomposition): Excellent for handling technical noise and platform effects.
- Cell2Location: A Bayesian model that is computationally intensive but highly accurate for mapping fine-grained cell subtypes.
- Spotlight: Uses NMF (Non-negative Matrix Factorization) to seed cell type signatures.
Step 3: Spatial Clustering & Domains
Standard clustering (like K-means or Leiden) treats every spot as independent. But in tissue, neighbors matter. Spatial clustering tools like BayesSpace or PRECAST use the physical location of spots to refine clusters, smoothing out noise and identifying robust "spatial domains" (e.g., tumor core vs. invasive margin).
Step 4: Spatially Variable Genes (SVGs)
We don't just want to know what is there, but where it is active. Algorithms like SpatialDE or SPARK identify genes with non-random spatial patterns, revealing gradients of expression (e.g., hypoxia markers increasing as you move away from blood vessels).
Common Pitfalls to Avoid
❌ The "Blind" Reference
Attempting deconvolution without a good single-cell reference from the same tissue type. Using a public PBMC dataset to deconvolute a solid tumor will lead to erroneous results.
❌ Ignoring Batch Effects
Placing all controls on Slide A and all treatments on Slide B. If possible, randomize samples across capture areas to distinguish biology from slide-to-slide technical variation.
❌ Over-interpreting "Co-localization"
Just because Ligand A and Receptor B are in the same 55µm spot doesn't prove they are interacting. It only proves proximity. Validation (e.g., high-res IF) is crucial.
❌ The "One-Sample" Study
Spatial heterogeneity is massive. n=1 is a pilot, not a study. Plan for biological replicates to make statistically sound claims.
Why Start Now?
Spatial transcriptomics is rapidly moving from "nice to have" to "need to have" for high-impact publications in oncology, neuroscience, and immunology. By establishing these workflows now, you position your lab at the cutting edge of tissue biology, ready to see biology in high definition.