Spatial Transcriptomics

Unlike single-cell RNA sequencing (scRNA-seq) which requires tissue dissociation, spatial transcriptomics retains the native spatial coordinates of each transcript. This allows researchers to overlay gene expression data directly onto hematoxylin and eosin (H&E) stained images, correlating molecular signatures with visible histological features such as tumor margins, necrotic cores, or immune infiltrates.

Spatial methods fall into two categories: unbiased whole-transcriptome capture and targeted in situ hybridization. Unbiased methods like Visium capture polyadenylated mRNA across the entire transcriptome, enabling discovery. Targeted methods like MERFISH or Xenium use pre-designed probe panels to quantify hundreds to thousands of specific genes with higher sensitivity and subcellular resolution.

The resulting data requires specialized computational frameworks. Spatial autocorrelation metrics like Moran's I quantify gene clustering. Spatial deconvolution algorithms (e.g., RCTD, cell2location) estimate cell-type proportions within each capture spot by integrating scRNA-seq reference data. Spatial graph neural networks learn context-aware representations by treating tissue as a graph where nodes are cells or spots and edges represent physical proximity.

A core application is identifying cell-cell communication in situ. By analyzing spatially proximal cell-type pairs, algorithms detect instances where a ligand gene in one cell type and its cognate receptor gene in a neighboring cell type are co-expressed. This moves beyond simple co-expression to infer active paracrine signaling within tissue niches like the tumor microenvironment.

Modern spatial biology integrates transcriptomics with spatial proteomics (e.g., CODEX, MIBI) and spatial epigenomics from adjacent tissue sections. Spatial registration algorithms computationally align these disparate modalities into a common coordinate framework, enabling the simultaneous analysis of mRNA, protein, and chromatin accessibility within the same tissue architecture.

Biological processes like development or tumor invasion occur along spatial gradients. Spatial trajectory inference algorithms order cells based on both their gene expression profiles and their physical coordinates, reconstructing dynamic processes in situ. This reveals the molecular progression of cells as they migrate from one tissue compartment to another, such as the epithelial-to-mesenchymal transition at a tumor's invasive front.

A foundational multiplexed capture technique where spatially indexed oligonucleotide arrays bind mRNA from tissue sections. Each spot on the array contains probes with unique positional barcodes, enabling transcriptomic data to be mapped back to its original tissue coordinates after sequencing.

• Key platforms: 10x Visium, Slide-seq, Stereo-seq
• Resolution range: 2-100 µm spot diameters
• Workflow: Tissue staining → imaging → permeabilization → reverse transcription → library preparation → sequencing
• Enables unbiased, genome-wide spatial profiling without prior target selection

A targeted spatial method that directly reads RNA sequences within preserved tissue using padlock probes and rolling circle amplification. Sequencing-by-ligation chemistry is performed on-site, generating fluorescent signals decoded by microscopy.

• Subcellular resolution: Detects transcripts at single-molecule level
• Multiplexing: Hundreds of genes simultaneously via sequential barcode rounds
• Key advantage: No tissue dissociation required; preserves native spatial architecture
• Variant: HybISS uses hybridization-based decoding instead of enzymatic ligation

A computational inference method that disentangles mixed gene expression signals from multi-cellular capture spots into constituent cell-type proportions. Essential when spatial resolution exceeds single-cell dimensions.

• Reference-based: Uses scRNA-seq signatures (e.g., RCTD, SPOTlight)
• Reference-free: Employs matrix factorization (e.g., STdeconvolve)
• Probabilistic frameworks: Bayesian models like cell2location estimate per-spot cell abundance
• Outputs a cell-type proportion matrix mapped to spatial coordinates

Genes whose expression exhibits statistically significant spatial patterning rather than random distribution across tissue. Identifying SVGs is a primary analysis goal, revealing genes that define anatomical structures or functional gradients.

• Detection methods: SpatialDE (Gaussian process), SPARK-X (non-parametric), Moran's I
• Expression patterns: Gradients, patches, or periodic distributions
• Biological relevance: Often mark developmental organizers, tumor boundaries, or neural layer markers
• Distinguished from spatially differentially expressed genes which compare predefined regions

Deep learning architectures that operate on graph representations where nodes are cells or spots and edges encode spatial proximity. These models learn context-aware embeddings by message-passing between neighbors.

• Architectures: GraphSAGE, GAT, GCN applied to spatial neighborhood graphs
• Applications: Spatial domain detection, gene imputation, tissue segmentation
• Key advantage: Captures higher-order spatial dependencies beyond pairwise correlations
• Tools: SpaGCN, STAGATE, GraphST integrate histology images with expression data

A cell-cell communication analysis identifying spatially proximal cell-type pairs where a ligand gene in one cell type and its cognate receptor in another are co-expressed. Maps intercellular signaling networks onto tissue architecture.

• Tools: CellPhoneDB, NicheNet, stLearn, Giotto
• Requires: Spatial coordinates + cell-type annotations + ligand-receptor databases
• Outputs: Interaction probability scores and spatial communication maps
• Key insight: Reveals signaling hubs in tumor microenvironments and developmental niches