Application of spatial transcriptomics across organoids: a high-resolution spatial whole-transcriptome benchmarking dataset

This study presents the first systematic benchmarking of Stereo-seq spatial transcriptomics across diverse stem cell-derived organoids, detailing assay optimization, multi-sample chip integration, and a novel regional analysis method to bridge the gap between organoid models and in vivo tissue characterization.

The Gist

The Big Picture: Building Better "Mini-Organs"

Imagine scientists are trying to build tiny, 3D models of human organs (like brains, hearts, and kidneys) using stem cells. They call these "organoids." Think of them as miniature, edible versions of a real organ, but instead of being made of cake, they are made of living cells.

Scientists want to use these mini-organs to test new drugs or study diseases. But there's a catch: Are these mini-organs actually built correctly? Do they have the right cells in the right places, just like a real human organ?

To check this, the researchers used a super-powered camera called Spatial Transcriptomics (specifically a technology called Stereo-seq).

The Problem: The "Pixelated" Photo

Usually, when you take a photo of a tiny object, if the object is too small, the photo comes out blurry or pixelated. You can't see the details.

• The Challenge: These organoids are very small. When the scientists tried to use the Stereo-seq camera on them, the "pixels" (which represent individual cells) were too big or the signal was too weak. It was like trying to read the fine print on a postage stamp with a magnifying glass that was slightly out of focus.
• The Result: They couldn't clearly tell which specific cell was doing what, making it hard to say if the mini-organ was built correctly.

The Solution: Packing a Suitcase and Drawing a Map

The researchers came up with two clever tricks to solve this.

1. The "Suitcase" Trick (Multi-Sample on One Chip)

Stereo-seq chips are expensive, like high-end camera sensors. Usually, you put one big tissue sample on one chip. But organoids are tiny. Putting just one on a huge chip is like putting a single grape in a giant suitcase—it's a waste of space and money.

• What they did: They figured out how to pack multiple organoids onto a single chip, like filling a suitcase with many small apples instead of one giant watermelon.
• The Glue: Some of these mini-organs (like heart muscle) were slippery and kept falling off the chip. The team coated the chip with a special "sticky glue" (poly-L-lysine) to make sure the tiny organs stayed put during the scan.
• The Win: They successfully scanned up to 12 different mini-organs at once, saving money and allowing them to compare different conditions side-by-side.

2. The "Neighborhood" Map (Region Analysis)

Since they still couldn't see individual cells clearly (the "pixels" were still a bit fuzzy), they changed their strategy. Instead of trying to identify every single person in a crowd, they decided to look at neighborhoods.

• The Analogy: Imagine you are looking at a city from a high drone. You can't see individual faces, but you can clearly see that the "downtown" area is busy with skyscrapers, while the "suburbs" are quiet with houses.
• What they did: They divided each organoid into two zones: the Core (the center) and the Border (the edge).
• The Discovery:
  • Brain Organoids: They found that the "Core" was busy with cells that eat sugar for energy (glycolysis), while the "Border" had cells that were more active and needed more power (ATP synthesis). This is exactly how a real developing brain works!
  • Heart Organoids: They compared hearts grown in normal food vs. "super-food" (directed maturation). The "super-food" hearts showed signs of being more mature and having stronger energy engines (mitochondria).

Why This Matters

This paper is like a user manual for taking high-resolution photos of tiny, living things.

1. It proves it's possible: They showed that you can scan many tiny organoids at once without ruining the data.
2. It fixes the "blur": Even if the camera can't see every single cell perfectly yet, their new "Neighborhood Map" method lets scientists see the big picture: Is the center of the organ different from the edge? Yes? Good! That means it's building correctly.
3. It saves money: By packing more samples onto one chip, this technology becomes affordable enough for more labs to use.

The Bottom Line

The scientists took a high-tech camera, figured out how to stick tiny, slippery mini-organs to it, and invented a new way to read the data. Instead of getting frustrated that they couldn't see every single cell, they looked at the "neighborhoods" within the organ. This helped them confirm that their mini-brains and mini-hearts are actually developing in the right way, bringing us one step closer to using these models to cure real diseases.

Technical Summary

1. Problem Statement

Stem cell-derived organoids are critical for modeling human development and disease, but their utility depends on how accurately they recapitulate in vivo tissue architecture and cellular composition. While Spatial Transcriptomics (ST) technologies like Stereo-seq offer high-resolution, whole-transcriptome profiling, their application to organoids faces specific challenges:

• Size and Shape: Organoids are often small and irregularly shaped, covering only a fraction of the ST chip, potentially leading to artificially low mRNA capture or "transcript diffusion" (RNA leaking outside tissue boundaries).
• Cost Efficiency: Standard ST workflows often require one tissue section per chip. Profiling multiple organoid conditions (e.g., different culture media or time points) is cost-prohibitive if each requires a separate chip.
• Resolution Limitations: Achieving true single-cell resolution in organoids is difficult due to lower RNA capture efficiency compared to larger in vivo tissues, hindering standard cell-type clustering.
• Lack of Benchmarks: There is a scarcity of studies systematically comparing organoid spatial profiles against in vivo reference tissues to validate their fidelity.

2. Methodology

The study employed a systematic workflow to benchmark Stereo-seq across eight human organoid models (brain, heart muscle, heart valve, kidney [transwell and suspension], lung, cartilage, and haematopoietic) and two in vivo mouse references (E13.5 head and 21 PCW heart).

Key Technical Steps:

• Sample Preparation & Optimization:
  • Multi-Organoid Loading: Multiple organoids of the same tissue type (and different conditions) were co-embedded in OCT and placed on a single Stereo-seq chip to maximize coverage and reduce inter-chip variability.
  • Adhesion Enhancement: To prevent tissue detachment (observed in cartilage, heart muscle, and kidney), chips were coated with poly-L-lysine.
  • Permeabilization Optimization: Tissue-specific permeabilization times were tuned (ranging from 12 to 24 minutes) to balance RNA release with the prevention of transcript diffusion.
• Sequencing: Libraries were prepared using the STOmics Transcriptomics kit and sequenced on a MGI DNBSEQ T7 sequencer.
• Bioinformatics Workflow:
  • Preprocessing: Data was processed using the SAW pipeline (v6.12) and aligned to GRCh38 (organoids) or GRCm39 (mouse).
  • Quality Control (QC): Metrics included UMI counts, gene counts per CellBin, DNA nanoballs, and sequencing saturation.
  • Bin Size Analysis: The study tested increasing "square bin" sizes (50, 75, 100 µm²) to determine the optimal resolution for organoids versus in vivo tissues.
  • Novel Regional Analysis: Since single-cell clustering failed due to low transcript capture, a bespoke region-specific analysis was developed. This involved:
    1. Outlier Removal: Using k-nearest neighbors (k-NN) to identify and remove bins with false signals (e.g., detached cells or background noise).
    2. Spatial Segmentation: Classifying remaining bins into Core, Border, and Edge regions based on Euclidean distance from the organoid centroid or convex hull.
    3. Pseudo-bulk Aggregation: Aggregating gene expression within these regions to boost signal strength for differential expression analysis (using pydeseq2).

3. Key Contributions

• First Systematic Benchmarking: Provides the first comprehensive comparison of Stereo-seq performance across diverse organoid types against in vivo references.
• Cost-Effective Multi-Sample Workflow: Demonstrates the feasibility of placing up to 12 organoid sections (including different culture conditions) on a single chip without compromising data integrity.
• Protocol Optimization: Establishes critical modifications for organoid ST, specifically poly-L-lysine coating for adhesion and tissue-specific permeabilization times.
• New Analytical Framework: Introduces a region-based analysis method (Core vs. Border) as a viable alternative to single-cell clustering when RNA capture is insufficient for cell-type annotation.

4. Key Results

• Data Capture & Adhesion:
  • Poly-L-lysine coating successfully improved adhesion for heart muscle, heart valve, and cartilage organoids.
  • Haematopoietic and suspension kidney organoids exhibited significant transcript diffusion (RNA captured outside tissue boundaries), likely due to their "liquid-phase" culture and lack of structural integrity. These were excluded from deep regional analysis.
  • Brain, transwell kidney, and cartilage organoids showed RNA capture quality comparable to in vivo references in terms of gene and UMI counts per bin.
• Bin Size Effects:
  • Increasing bin size (50 → 100 µm²) reduced the total number of bins disproportionately (more than 50% reduction for a 50% size increase) and led to a loss of peripheral tissue coverage.
  • Smaller bin sizes (50 µm²) were preferred to maximize tissue area inclusion, though they still fell short of true single-cell resolution for organoids.
• Regional Analysis Findings:
  • Brain Organoids: Successfully identified spatially distinct metabolic signatures. The border regions were enriched for ATP synthesis and electron transport (mature neurons), while the core was enriched for glycolysis (neural progenitors), mirroring in vivo cortical development.
  • Cartilage Organoids: Showed homogeneous gene expression across regions (no differentially expressed genes), consistent with a uniform chondrocyte population.
  • Heart Muscle Organoids: Comparison between Serum-Free (Control) and Directed Maturation (DM) conditions revealed:
    • DM Organoids: Upregulation of mitochondrial activity, ATP synthesis, and metabolic genes (e.g., MRPS12), indicating enhanced maturation.
    • Control Organoids: Enrichment of ion channel activities.
    • The regional analysis successfully filtered out low-density "pole" regions, focusing analysis on the contractile "center."

5. Significance and Limitations

Significance:

• This study validates Stereo-seq as a powerful tool for characterizing organoid heterogeneity and maturation states, provided specific protocols are followed.
• The regional analysis workflow offers a robust solution for extracting biological insights from organoids where single-cell resolution is unattainable due to technical limitations in RNA capture.
• It establishes a framework for future organoid quality control, allowing researchers to assess if their models recapitulate the spatial organization of native tissues.

Limitations:

• Resolution: Despite optimization, the platform did not achieve reliable single-cell resolution for organoids, necessitating the shift to pseudo-bulk regional analysis.
• Diffusion: Organoids cultured in suspension or liquid phases (haematopoietic, suspension kidney) remain challenging due to transcript diffusion.
• Generalizability: The study focused on Stereo-seq; other ST technologies (e.g., 10x Visium HD or imaging-based methods) might yield different results.
• Reference Tissues: The in vivo references were mouse tissues, which, while useful for benchmarking, differ developmentally and genetically from human organoids.

In conclusion, this paper provides a critical roadmap for applying high-resolution spatial transcriptomics to organoid research, balancing technical optimization with novel computational strategies to overcome current data capture limitations.