The project started at the cross-section of two disciplines. Jonas Frisén from the Karolinska Institute, who was using microscopy to study neurogenesis, joined forces with Joakim Lundeberg, head of the Genomics Core Facility at the Science for Life Laboratory (SciLifeLab) in Sweden, to bring together the worlds of imaging and sequencing.

Gene expression heterogeneity in breast cancer: clustered spatial transcriptomic data overlap with morphological features (violet, normal mammary glands; red, fat tissue; green, cancerous regions). Credit: J. Lundeberg

Traditional RNA sequencing experiments provide quantitative information on expression levels but lose spatial information. The goal of Frisén and Lundeberg was to retain this information, not only for a few transcripts but for close to the entire transcriptome. The idea behind spatial transcriptomics was conceptually simple, but the approach proved challenging in practice. The protocol begins with easy steps: a tissue is sectioned and immobilized on a chip prior to staining and imaging. The tissue is then permeabilized to release RNA, which is captured by poly-dT oligos immobilized on the array. Once bound to the chip, the RNA is reverse-transcribed and imaged or sequenced. It is the capture step that was difficult to perfect. “If you permeabilize too long,” says Lundeberg, “you lose the morphology and then you lose the purpose of the technology.”

After a lot of optimization, Patrik Ståhl from the Frisén lab and Fredrik Salmén from Lundeberg's group, who spearheaded the project, found optimal conditions for several tissues. To prove the validity of their approach, the researchers used fluorescent nucleotides during cDNA synthesis to monitor the position of the RNA, and they observed the same tissue structure that is revealed in a histological stain with hematoxylin and eosin. “That proved that we could capture mRNA without any diffusion,” recalls Lundeberg. The researchers compared sections from different tumors and saw different areas of gene activity.

To generate a permanent spatial record of the mRNA molecules, the team incorporated positional barcodes and unique molecular tags into their capture oligos, reverse-transcribed the RNA and cleaved it off the array for amplification and sequencing. The correlation with standard RNA-seq was high, and the spatial barcodes allowed them to trace each transcript back to its spatial origin in the tissue section. Comparison of gene expression in spatially defined domains in brain sections revealed genes specific for certain regions and allowed the researchers to compare adjacent sections, as well as the same section in different animals.

The resolution of spatial transcriptomics is limited by the density of features (areas with the same spatial barcode), which currently stands at 100 micrometers. If they could get this density down to 10 micrometers, says Lundeberg, they would have single-cell resolution. Efforts are under way to increase the resolution.

Lundeberg is happy about the convergence of disciplines, which, he says, was the idea behind the founding of the SciLifeLab in 2010. He hopes to use the technology to compare protein maps, partly generated at SciLifeLab as part of the Human Protein Atlas, to a molecular atlas.

Another goal is to get a 3D view of cancer. The researchers are taking sequential sections through a tumor to determine its spatial transcriptome. The heterogeneity they see between sections underscores how difficult it is to get a representative example of a tumor and how much is missed by sequencing of pooled RNA from the entire tumor. “Down the line,” says Lundeberg, “I hope that this can be a complement to histology in clinical labs.”