Combining xenium in situ spatial transcriptomics and imaging mass cytometry on a single tissue section

This study demonstrates the feasibility and biological value of a novel workflow that combines Xenium in situ spatial transcriptomics and Imaging Mass Cytometry on a single FFPE tissue section, revealing that while global RNA-protein correlations exist, integrating both modalities at the single-cell level yields the most comprehensive and biologically meaningful insights into cellular landscapes.

The Gist

Imagine you are trying to understand a bustling city. You have two powerful tools to study it, but each only gives you half the picture.

Tool A (The Transcriptome) is like a phone call log. It tells you who tried to make a call, what they said they were going to do, and who they were planning to meet. It's a list of intentions and blueprints.
Tool B (The Proteome) is like a security camera. It shows you who is actually standing on the street corner, what they are wearing, and what they are actually doing right now.

For a long time, scientists had to look at the phone logs on one day and the security footage from a different day (or a different slice of the city) to guess what was happening. This is like trying to match a phone call from Monday with a security video from Tuesday; the city might have changed, or the people might have moved.

The Big Breakthrough: The "Double-Exposure" Photo

This paper describes a new, clever way to take both the phone log and the security footage of the exact same spot at the exact same time.

The researchers combined two high-tech machines:

1. Xenium: A machine that reads thousands of genetic "phone calls" (RNA) inside cells.
2. IMC (Imaging Mass Cytometry): A machine that takes a "security photo" of proteins (the physical workers) using metal-tagged antibodies.

Usually, you can't use both machines on the same piece of tissue because the first machine is so intense (using heat and chemicals) that it destroys the tissue, making the second machine useless.

The Analogy: Imagine trying to read a letter written on a piece of paper, and then immediately trying to take a fingerprint of that same paper. Usually, the ink would smear, or the paper would tear.

What they did: They figured out a way to read the letter (Xenium) so gently that the paper didn't tear, and then they could still take the fingerprint (IMC) right after. They proved that the "letter" didn't ruin the "fingerprint."

What They Found: The "Intention vs. Action" Gap

When they looked at the data, they found something fascinating: What cells say they are doing (RNA) isn't always what they are actually doing (Protein).

• The "Ghost" Cells: Sometimes a cell has the "phone call" (RNA) saying "I am a T-cell (a soldier)," but the security camera (Protein) doesn't see the soldier's uniform yet.
• The "Silent" Soldiers: Sometimes the camera sees a soldier in full gear (Protein), but the phone log is silent. Maybe the soldier stopped talking but is still ready to fight.

If you only looked at the phone logs, you might think a cell is active when it's actually sleeping. If you only looked at the camera, you might miss a cell that is just about to wake up.

The Best Solution: The "Double-Check"

The researchers realized that the most accurate way to understand the city is to only count the people who appear on both the phone log and the security camera.

When they focused on the T-cells (immune soldiers) that were identified by both machines, they got the clearest picture of the battle. They could see exactly which soldiers were inside the "enemy territory" (the tumor) and which were outside. They found that the soldiers inside the tumor were truly "activated" and ready to fight, showing signs of both the "plan" (RNA) and the "gear" (Protein).

Why This Matters

This new method is like having a super-powerful microscope that lets you see the full story of a disease.

• Before: We were guessing if a cell was a "soldier" based on just one clue.
• Now: We can confirm it's a soldier because it has the uniform and the orders.

This helps doctors and scientists understand complex diseases like cancer much better. It stops them from being fooled by cells that look like one thing but act like another. By combining the "what they say" with the "what they do," we get a much deeper, more honest view of how our bodies fight disease.

Technical Summary

1. Problem Statement

Current spatial omics technologies generally operate in isolation, profiling either the transcriptome (RNA) or the proteome (protein) within a tissue microenvironment.

• Limitations of Single-Modal Approaches: While Imaging Mass Cytometry (IMC) is robust for defining cell phenotypes via protein markers (up to ~40 targets), it lacks the high-plex capability of transcriptomics. Conversely, spatial transcriptomics (like Xenium) offers high-plex mRNA profiling (up to 5,000 targets) but may suffer from issues like off-target binding or RNA diffusion, and protein expression does not always correlate perfectly with transcript levels due to post-transcriptional regulation.
• Spatial Context Loss: Existing multi-omic methods like CITE-seq lose spatial context. While some commercial platforms (e.g., CosMx, Xenium Protein) offer simultaneous RNA/protein profiling, they are limited to pre-validated, sub-optimal protein panels, preventing custom antibody design.
• Adjacent Section Variability: Previous attempts to combine modalities used adjacent tissue sections, which introduces spatial discrepancies in cellular composition and microenvironment structure.
• The Gap: There was no established, feasible workflow to perform high-plex Xenium spatial transcriptomics followed by custom 40-plex Imaging Mass Cytometry (IMC) on the same formalin-fixed paraffin-embedded (FFPE) tissue section to achieve true single-cell multi-omic integration.

2. Methodology

The authors developed a novel dual-platform workflow applied to FFPE melanoma lymph node tissue microarrays (TMAs).

• Sample Preparation:

  • Slide A: Underwent the full Xenium workflow followed by IMC.
  • Slide B: Underwent the full Xenium workflow followed by Hematoxylin and Eosin (H&E) staining for quality control.
  • Control: Adjacent sections underwent IMC alone (performed previously) for comparison.

• Xenium Workflow (Transcriptomics):

  • Performed using the 10x Genomics Xenium Human Multi-Tissue and Cancer Expression Panel (377 genes).
  • Included deparaffinization, decrosslinking, probe hybridization, ligation, rolling circle amplification, and cyclic imaging.
  • Cell segmentation masks were generated onboard using nuclear expansion (15 µm).

• IMC Workflow (Proteomics) - Post-Xenium:

  • Crucial Step: Performed after Xenium imaging on Slide A.
  • Antigen Retrieval: Required a second heat-induced antigen retrieval (microwave) after the Xenium protocol.
  • Staining: Applied a custom 40-plex metal-tagged antibody panel (Table 1) targeting immune cells (T, B, myeloid), tumor cells (SOX10), and functional markers (Ki67, PD-1, Granzyme B, etc.).
  • Acquisition: Performed on the Hyperion Imaging System using laser ablation and time-of-flight mass cytometry.

• Data Integration & Analysis:

  • Registration: Xenium DAPI images and IMC DNA-Ir images were manually aligned using Visiopharm software (320 landmarks) to correct for spatial shifts.
  • Segmentation: The Xenium cell segmentation mask was applied to the aligned IMC data to ensure consistent cell boundaries across modalities.
  • Cell Annotation: Two approaches were used:
    1. Threshold-based: Single marker gating (e.g., CD3E+ for T cells).
    2. Semi-supervised Clustering: FuseSOM and clustSIGNAL algorithms used multiple markers to define cell types.
  • Concordance Analysis: Cells were categorized as identified by "Both," "Xenium only," or "IMC only."

3. Key Contributions

• First Feasibility Demonstration: This is the first study to comprehensively demonstrate the technical feasibility of performing Xenium followed by IMC on a single FFPE section.
• Validation of Tissue Integrity: Proved that the rigorous Xenium protocol (including heat cycles and chemical washes) does not degrade tissue quality or prevent subsequent high-quality IMC protein staining.
• Customizable Multi-omics: Overcomes the limitation of fixed protein panels in commercial spatial proteomics by allowing researchers to use custom IMC panels after high-plex transcriptomics.
• Analytical Framework: Established a pipeline for registering, segmenting, and analyzing co-registered RNA and protein data from distinct platforms.

4. Key Results

• Tissue Integrity: H&E staining on Slide B and visual comparison of DAPI (Xenium) vs. DNA-Ir (IMC) on Slide A confirmed minimal tissue loss and high morphological concordance. The tissue withstood two rounds of heat-induced antigen retrieval.
• Protein Signal Quality: Comparison of "Post-Xenium IMC" vs. "IMC Only" controls showed comparable signal intensity and cellular quantification for key markers (e.g., SOX10, CD45), confirming the workflow does not compromise protein detection.
• Transcript-Protein Concordance:
  • Global Scale: High visual correlation between transcript and protein expression for markers like CD3E, CD20, and Ki67.
  • Single-Cell Level: Significant discordance was observed. For example, CD20/MS4A1 showed moderate correlation ($r=0.48$), while CD3E showed low correlation ($r=0.15$). Many cells were positive for protein but negative for transcript, and vice versa.
• Biological Insight via Integration:
  • Cell Identification: Using semi-supervised clustering improved the concordance of cell type identification between platforms compared to single-marker thresholds.
  • Functional Analysis: T cells identified by both platforms (RNA+ and Protein+) within tumor regions showed the most biologically meaningful activation signatures (high expression of MKI67, LAG3, GZMB, FOXP3 transcripts and corresponding proteins).
  • Single-Modality Bias: T cells identified by Xenium alone showed transcript upregulation without protein confirmation, while IMC-only cells showed protein upregulation without transcript confirmation. This highlights that relying on a single modality yields incomplete biological pictures.

5. Significance

• Enhanced Phenotyping: This workflow enables a more accurate and comprehensive characterization of cell phenotypes by requiring concordance between RNA and protein, reducing false positives caused by technical artifacts (e.g., RNA diffusion or antibody cross-reactivity).
• Functional Depth: By integrating both data types, researchers can better understand the functional state of cells (e.g., T cell exhaustion or activation) within specific spatial niches (e.g., tumor core vs. stroma).
• Workflow Validation: The study provides a validated protocol for labs to combine high-plex transcriptomics with custom proteomics, bypassing the limitations of current all-in-one commercial solutions.
• Future Directions: The authors note challenges such as registration difficulties due to different nuclear stains (DAPI vs. Ir) and potential RNA/protein degradation in older FFPE samples. They suggest future work should focus on automated registration and improved cell segmentation algorithms (e.g., using membrane stains) to further refine this multi-omic approach.

In conclusion, this paper establishes a robust, novel methodology for spatial multi-omics, demonstrating that combining Xenium and IMC on a single section yields superior biological insights into cellular landscapes and interactions compared to single-modality approaches.