Explain spatial transcriptomics

Overview

Spatial transcriptomics is a revolutionary technique that measures gene expression (which genes are being transcribed into RNA) while preserving the spatial location of cells within a tissue. Traditional RNA sequencing (RNA-seq) grinds up tissue into a "soup" of cells, destroying all spatial information. Spatial transcriptomics answers: "What genes are active WHERE in the tissue?"

Why this matters:

• Cells behave differently based on their neighbors and microenvironment
• Diseases like cancer create spatial patterns (tumor core vs. invasive edge)
• Development follows spatial gradients (head vs. tail in embryos)
• The same cell type can have different gene expression in different tissue locations

Without spatial context, we miss critical biology: how cells organize, communicate, and create functional tissue architecture.

Core Concepts

The Central Problem

Traditional RNA-seq workflow:

1. Dissociate tissue → single cell suspension
2. Extract RNA from each cell
3. Sequence and count transcripts
4. ❌ Spatial information is lost forever

The innovation: Capture RNA molecules on a spatially-barcoded surface, where each location has a unique DNA barcode. When you sequence the RNA+barcode together, you can trace each transcript back to its tissue location.

How It Works: First-Principles Derivation

Method 1: Spatial Barcoding (Visium/10x Genomics)

Step 1: Create a spatially-barcoded slide

• A glass slide is printed with millions of capture probes
• Each probe has structure: [Spatial Barcode] - [UMI] - [polyT tail]
• Probes are organized in spots (55 μm diameter, ~10 cells per spot)
• Each spot has a unique spatial barcode

Why polyT? mRNA molecules have polyA tails. The polyT binds them specifically.

Step 2: Place fresh-frozen tissue on the slide

• Tissue section (10 μm thick) is placed directly on barcoded spots
• Tissue is permeabilized → mRNA molecules diffuse out and bind to nearest capture probes

Step 3: Reverse transcription captures spatial information

mRNA-AAAA (from tissue)
     ||||
Probe-TT-[UMI]-[Spatial Barcode]-[on slide]
          ↓ Reverse transcription
        cDNA-[UMI]-[Spatial Barcode]

Why this step is critical: The spatial barcode is now physically attached to the cDNA copy of the mRNA. Even when we remove the cDNA from the slide for sequencing, it carries its location information.

Step 4: Sequence and decode

• Remove all cDNA from slide, amplify, sequence
• Each read contains: Gene identity + UMI + Spatial barcode
• Map reads back to spot coordinates → gene expression matrix with XY positions

For spot $i$ at position $(x_i, y_i)$, the expression level of gene $g$ is:

$E_{g,i} = \sum_{j \in \text{UMIs at spot } i} \mathbb{1}(\text{UMI}_j \text{ maps to gene } g)$

Why count UMIs not reads? During PCR amplification, one mRNA molecule creates many identical copies (PCR duplicates). The Unique Molecular Identifier (UMI) is a random barcode added before amplification. Duplicates share the same UMI, so we count unique UMIs to get the true molecule count.

Resolution: Each spot = ~5-10 cells, so this is near-cellular resolution, not single-cell.

Method 2: In Situ Sequencing

Instead of capturing RNA on a slide, sequence it directly in the tissue.

Step 1: Fix tissue to preserve structure

• Crosslink proteins/RNA with formaldehyde
• RNA molecules are locked in their original positions

Step 2: Reverse transcribe mRNA → cDNA inside tissue

• Use padlock probes that circularize around target RNA sequences
• Create rolling circle amplification (RCA) → bright fluorescent spots

Step 3: Sequence by synthesis in tissue

• Add fluorescent nucleotides one base at a time
• Image the tissue after each addition
• Record color at each XY position → decode sequence

Step 4: Align sequences to genome

• Each fluorescent spot = one mRNA molecule at a specific XY coordinate
• Achieve subcellular resolution (~100 nm) but for a limited gene panel

Trade-off: In situ methods have higher spatial resolution but can only measure 100-1000 pre-selected genes (not whole transcriptome).

Worked Examples

Scenario: Map spatial gene expression in mouse hippocampus (memory formation region).

Approach: Use 10x Visium (55 μm spots, whole transcriptome).

Results interpretation:

• Spot A (in CA1 pyramidal layer): High expression of Camk2a, Grin1 (excitatory neuron markers)
• Spot B (in dentate gyrus): High expression of Prox1, Calb2 (dentate granule cells)
• Spot C (in white matter): High expression of Mbp, Plp1 (oligodendrocyte myelin genes)

Why this step matters: These expression patterns confirm known anatomy. A researcher studying Alzheimer's could now ask: "Which spatial regions show abnormal gene expression?"

Calculation: If spot A has 500 UMIs for Camk2a and the spot contains ~8 cells, average expression per cell ≈ 500/8 = 62.5 UMI/cell. (This is approximate since not all cells in the spot are neurons.)

Scenario: Breast cancer biopsy shows a tumor mass surrounded by normal tissue.

Approach: Spatial transcriptomics to measure gene gradients from tumor center → edge → normal tissue.

Finding:

• Tumor core: High MKI67 (proliferation), VEGFA (angiogenesis)
• Invasive edge: High MP9 (matrix degradation), SNAI1 (epithelial-mesenchymal transition)
• Immune infiltrate zone: High CD8A, PDCD1 (T cells, exhaustion markers)

Why spatial matters: Bulk RNA-seq would average these three zones together, hiding the critical biology of tumor invasion. Spatial data reveals that invasive cells at the edge have a different expression program than proliferating cells in the core.

Therapeutic insight: If you target proliferation (tumor core) but ignore invasion genes (tumor edge), cancer may recur from invasive cells that weren't killed.

Common Mistakes & Misconceptions

Why it feels right: The word "spatial" sounds like it means "precise."

Reality: Most commercial platforms (Visium) have spots containing5-10 cells. You measure the average expression across those cells.

The fix:

• Visium ≈ 55 μm spots = near-cellular resolution
• In situ methods (MERFISH, seqFISH) = subcellular resolution (~0.1-1 μm) but fewer genes
• True single-cell spatial = newer methods like Xenium, CosMx, or computationally deconvolving Visium spots

Steel-man: The confusion arises because single-cell RNA-seq exists, and "spatial" sounds like an upgrade. But spatial barcoding trades off: you gain location, but lose single-cell resolution (unless you use specialized in situ methods).

Why it feels right: If spot A has 1000 UMIs and spot B has 500 UMIs for gene X, spot A has 2× more expression, right?

Reality: Total UMI counts vary by spot due to:

• Tissue thickness variation
• Cell density differences (10 cells vs. 5 cells per spot)
• RNA quality (degraded regions have fewer counts)

The fix: Normalize before comparing: $\tilde{E}_{g,i} = \frac{E_{g,i}}{\sum_{g'} E_{g',i}} \times 10^4$

This converts to counts per 10,000 (like TPM in RNA-seq), making spots comparable.

Why this step is essential: Without normalization, you might think spot A expresses a gene more, when really it just captured more total RNA.

Why it feels right: We think of gene expression as a dynamic process in living cells.

Reality: Tissue must be fresh-frozen (Visium) or formaldehyde-fixed (in situ methods). Both kill cells. We're measuring a snapshot of RNA at the moment of preservation.

The fix: Spatial transcriptomics captures the steady-state RNA levels that existed when the tissue was harvested. For dynamic processes (e.g., circadian rhythms), you'd need to harvest tissue at multiple time points.

Implication: You see correlation between spatial patterns and function, but not causation. If gene X is high in region Y, it doesn't prove X causes Y's phenotype—could be a marker of cell type, a response to environment, etc.

Active Recall Questions

#flashcards/biology

What is spatial transcriptomics? :: A technique that measures gene expression (RNA levels) while preserving the XY coordinates of cells within a tissue, creating a spatial map of gene activity.

Memory Aids

Think: GPS navigation tells you "what stores are WHERE on the map." Spatial transcriptomics tells you "what genes are active WHERE in tissue."

Imagine you have a big Lego city (that's your tissue, like a piece of brain or skin). Every Lego brick is a cell, and each cell has a recipe book (genes) telling it what to build.

Now, if you put all the Lego bricks in a blender and ask "what recipes are being used?", you'd lose important information: Are the red bricks (fire trucks) near the fire station? Are the blue bricks (police cars) near the police station?

Spatial transcriptomics is like taking a photo of the Lego city from above and labeling each brick with which recipe it's currently reading. Now you can see patterns: "Oh! All the fire truck recipes are active near the fire station!" or "Uh oh, the villain's lair has weird recipes that shouldn't be there."

In real biology, this helps us understand diseases. Cancer cells (villain's lair) might have different recipes at the center vs. the edge where they're invading. If we only looked at the blended soup, we'd miss that the edge cells are using "break through walls" recipes while center cells use "multiply fast" recipes. To cure cancer, we need to stop BOTH, and spatial transcriptomics shows us where each problem is happening.

Connections & Further Learning

Prerequisites:

• 6.1-DNAstructure-and-replication
• 6.2.3-Transcription-and-gene-expression
• 6.3.5-RNA-sequencing-basics

Related Concepts:

• 6.5.8-Single-cell-RNA-seq
• 6.5.9-Immunofluorescence-microscopy
• 6.5.10-FISH-fluorescence-in-situ-hybridization

Advanced Topics:

• 6.5.12-Spatial-proteomics
• 6.5.13-Computational-deconvolution-of-spatial-data
• 6.5.14-Spatial-multi-omics

Clinical Applications:

• 7.2.5-Tumor-microenvironment-mapping
• 7.3.2-Brain-region-specific-gene-expression

Summary

Spatial transcriptomics revolutionizes gene expression profiling by answering "what genes are active WHERE?" It uses spatially-barcoded capture (Visium: whole transcriptome, near-cellular resolution) or in situ sequencing (MERFISH/seqFISH: selected genes, subcellular resolution). The key insight: cells with the same identity can have different gene programs depending on their tissue location and neighbors.

Critical points:

1. Spatial barcodes physically tag RNA with location coordinates
2. UMIs count true molecules, not PCR duplicates
3. Normalization is essential before comparing spots
4. Resolution vs. gene coverage is a trade-off
5. Enables discovery of spatial gradients, zones, and microenvironments invisible to bulk RNA-seq

This technique is transforming cancer biology, neuroscience, developmental biology, and immunology by revealing how gene expression creates tissue architecture.

Concept Map

Hinglish (regional understanding)