Genome-wide Spatial Expression Profiling in FFPE Tissues

Eva Gracia Villacampa, Ludvig Larsson, Linda Kvastad, Alma Andersson, Joseph W. Carlson, Joakim LundebergbioRxiv 2020 · DOI

TL;DR

This paper presents a novel protocol for performing genome-wide spatial mRNA expression profiling on Formalin-Fixed Paraffin-Embedded (FFPE) tissues, a widely available but challenging sample type. This method successfully maps gene expression to specific anatomical locations in mouse brain tissue and reveals the complex cellular landscape of ovarian carcinosarcoma, opening doors for deeper analysis of archival clinical samples.

What Problem Does This Paper Address?

Archival tissue samples, preserved using Formalin-Fixed Paraffin-Embedding (FFPE), represent a vast and invaluable resource for biomedical research. However, the FFPE process significantly degrades RNA integrity and introduces cross-linking, making it difficult to perform sensitive, genome-wide molecular analyses, especially those requiring spatial resolution. Existing spatial transcriptomics methods have largely been limited to fresh-frozen (FF) tissues, leaving a gap in our ability to leverage the wealth of FFPE samples for understanding tissue architecture and molecular mechanisms in a spatial context. This paper aims to bridge this gap by developing and validating a protocol for spatial transcriptomics on FFPE tissues.

Methods

The core of this work is a protocol adapted for FFPE tissue sections. Key steps include:

  • In situ recovery of mRNA: The procedure begins with deparaffinization and H&E staining of FFPE tissue sections placed on slides.
  • Cross-link reversal: A crucial step involves heat-induced retrieval in a Tris-EDTA buffer at pH 8.0 to reverse formaldehyde-induced cross-links, making the RNA accessible.
  • Spatial barcoding and library preparation: Similar to standard spatial transcriptomics, the protocol utilizes slides spotted with barcoded oligo(dT) probes to capture the 3' end of mRNA. This is followed by permeabilization, reverse transcription, and library preparation, with modifications to accommodate the shorter RNA fragments typical of FFPE samples.
  • Sequencing and data analysis: Libraries are sequenced, and the data is pre-processed to remove adapter sequences and poly(A) tails. The analysis involves standard spatial transcriptomics workflows, including normalization, dimensionality reduction (UMAP), clustering, and differential expression analysis.
  • Integration with single-cell data: For cell type mapping, the spatial transcriptomics data is integrated with single-cell RNA sequencing (scRNA-seq) data using the stereoscope tool to deconvolve spot transcriptomes into cell type proportions.
  • Non-negative Matrix Factorization (NNMF): For the carcinosarcoma data, NNMF was employed to explore gene expression patterns in an unbiased manner, identifying factors that correlate with histological features.

Datasets

  • Mouse Brain Tissue: FFPE coronal sections of mouse brain were used to validate the protocol. Fresh-frozen (FF) mouse brain tissue sections were also used for comparative analysis.
  • Ovarian Carcinosarcoma Tissue: FFPE samples of high-grade serous ovarian carcinosarcoma were used to assess the method's applicability in a clinical context.
  • Allen Brain Atlas: This resource was extensively used for anatomical referencing and for obtaining scRNA-seq data for cell type mapping.

Key Insights and Ideas

  • FFPE is amenable to spatial transcriptomics: The study demonstrates that despite RNA degradation and cross-linking, FFPE tissues can yield high-quality spatial transcriptomic data, unlocking the potential of archival samples.
  • Molecular delineation of anatomical structures: The mouse brain data showed that spatial gene expression patterns could accurately delineate known anatomical regions, validating the method's precision.
  • Unraveling tumor heterogeneity: Application to ovarian carcinosarcoma revealed distinct spatial patterns of gene expression, correlating with histological features like carcinoma, sarcoma, adipose tissue, and immune infiltration. This highlights the method's power in dissecting complex tumor microenvironments.
  • Integration with single-cell data is powerful: Combining spatial transcriptomics with scRNA-seq data allowed for sophisticated cell type mapping, providing a more nuanced understanding of tissue composition.
  • FFPE data correlates well with FF data: Despite differences in gene recovery, the FFPE and FF datasets showed a strong correlation, suggesting that FFPE data can provide complementary insights.

Key Results

  • The developed protocol successfully generated genome-wide spatial expression data from FFPE mouse brain tissue, allowing for the identification of 16 distinct clusters corresponding to anatomical structures.
  • Differential expression analysis in the mouse brain identified specific marker genes for various anatomical regions, which were validated using in situ hybridization (ISH) data.
  • Analysis of FFPE ovarian carcinosarcoma samples using NNMF identified 15 factors that spatially correlated with histological features, including carcinoma, sarcoma, and adipose tissue.
  • Pathway analysis of these factors revealed enrichment for processes like epithelial-to-mesenchymal transition (EMT), extracellular matrix organization, and immune responses, providing insights into the tumor's molecular mechanisms.
  • The FFPE and FF mouse brain datasets showed a Pearson correlation coefficient of 0.95 for bulk analysis, indicating strong agreement.

Limitations and Future Work

  • RNA quality and quantity: The paper acknowledges that FFPE samples generally yield lower RNA quality and quantity compared to fresh-frozen samples, which can impact the depth of analysis. The authors suggest that further investigation is needed to identify relevant QC metrics for FFPE spatial barcoding schemes.
  • Cell type deconvolution: While stereoscope was used for cell type mapping, the authors note that the spatial resolution of spots can encompass multiple cells, and the cell population within a spot may not be homogeneous. This means that a one-to-one relationship between spot and cell type is not guaranteed, and further refinement of deconvolution methods might be beneficial.
  • Exploration of clinical context: The authors encourage further exploration of the method's applicability in a clinical context, particularly for investigating the molecular heterogeneity of various cancer types and leveraging the rich resource of clinical biobanks.

Summary updated

Abstract

Abstract Formalin-fixed paraffin embedding (FFPE) is the most widespread long-term tissue preservation approach. Here we present a procedure to perform genome-wide spatial analysis of mRNA in FFPE tissue sections. The procedure takes advantage of well-established, commercially available methods for imaging and spatial barcoding using slides spotted with barcoded oligo(dT) probes to capture the 3’ end of mRNA molecules in tissue sections. First, we conducted expression profiling and cell type mapping in coronal sections from the mouse brain to demonstrate the method’s capability to delineate anatomical regions from a molecular perspective. Second, we explored the spatial composition of transcriptomic signatures in ovarian carcinosarcoma samples using data driven analysis methods, exemplifying the method’s potential to elucidate molecular mechanisms in heterogeneous clinical samples.
BibTeX
@misc{villacampa2020genome,
  author = {Eva Gracia Villacampa and Ludvig Larsson and Linda Kvastad and Alma Andersson and Joseph W. Carlson and Joakim Lundeberg},
  title = {{Genome-wide Spatial Expression Profiling in FFPE Tissues}},
  year = {2020},
  doi = {10.1101/2020.07.24.219758},
  howpublished = {bioRxiv},
}
Figure