What led you to be an early adopter of spatial transcriptomics? What was your first spatial project?
During my time at UPenn, we quickly realized the power of single cell RNA sequencing (scRNAseq), but more recent analyses looking at ways to predict cell-cell communication emphasized the need to put these cell populations within their in vivo spatial context. The first spatial project I worked on was looking at the role of sensory nerves in the developing calvarial sutures. We had shown that lack of nerve input was causing premature suture closure but understanding the changes in transcriptional landscape within this small, spatially organized structure was challenging. While overall signaling in the suture showed relatively modest changes, spatial showed us that activation of major morphogenetic pathways such as BMP and TGFβ was spatially altered, activating osteogenic differentiation gene programs within the stem cell containing midline. These results guided our downstream experiments to discover how nerves could potentially regulate these pathways to maintain the balance between stem cell self-renewal and osteoblast commitment within the tight spatial confines of the expanding suture.
Have you ever been surprised by the cell populations revealed in a spatial experiment?
The Visium spatial transcriptomics system that I am more familiar with has a resolution of 55um, so I would say it is not well suited for discovery of new cell types compared to other spatial technologies such as MERFISH. However, our more recent submitted work suggests that deconvolution packages, combined with scRNAseq can determine the position of unique cell types of interest. Even further, we have used the large spot sizes to our advantage, thinking of it less as an individual cell spot and more as a niche spot. Using this concept, you can use spatial to determine which cell types often exist within close proximity of each other and which genes are specifically expressed within this niche microenvironment.
How are you currently applying spatial transcriptomics in your research program?
Outside of plastic samples, sectioning of mineralized bone tissue can be a challenge. Initially, our studies made use of fresh frozen samples and relied on optimization of blades, sectioning temperature, and embedding practices, combined with tissues of relatively low mineral content, to overcome this limitation. Traditionally, the field gets around this inability to section bone by decalcifying with EDTA. However, this process is not designed to maximize RNA integrity. My lab currently focuses on ways to process bone samples, both from mouse and human sources, to decalcify the tissue to permit sectioning while simultaneously preserving RNA abundance and quality. Once optimized, I believe spatial transcriptomics will become as common place to musculoskeletal research as single cell RNA sequencing has become over the past 5 years.
What are advantages and challenges associated with using spatial transcriptomics in biological research?
Many people who have analyzed scRNAseq for ligand-receptor interactions acknowledge that while these tools are efficient at predicting mechanisms through which cell types could communicate to each other, these predictions don’t hold if the cell types of interest are no where near each other. The addition of spatial helps overcome this limitation, refining the list of possible signaling mechanisms down to those more biologically relevant within the in vivo tissue. But like all technologies, spatial has its limitations. As discussed above, the Visium system has a resolution of 55um, meaning this technique is not well suited for single cell analysis. Though deconvolution packages are being actively developed, I think there is still some ways to go to make this concept readily accessible. The other major limitation lies in the number of spots when using spatial. From my experience, each spatial spot typically contains about the same amount of information as a cell from a scRNAseq experiment. In scRNAseq, missing information is filled in by the 1000s of neighboring cells. In contrast, spatial analysis typically gives us 100s of spots of interest. As a result, fewer neighbors are available to fill in the gaps. Outside the box thinking and specialized tools are sometimes required to overcome this. In contrast, a technique like MERFISH is better suited to single cell spatial information gathering and provides much more sensitive read out of the transcriptome, but requires more initial troubleshooting, specialized equipment and an upfront knowledge of what genes you would like to probe for. Both techniques have their place and should be considered in terms of the biological question you hope to answer.
Is there a specific barrier that once addressed might bring spatial transcriptomics technology to the next level?
I believe the 3 biggest barriers currently existing are the limited resolution of a platform like Visium, the relatively low efficiency of transcriptome capture (similar to scRNAseq), and the cost of the specialized reagents. I think addressing any one of these 3 issues is likely to result in a large uptake in the utility of spatial approaches.
What advice would you give investigators who want to incorporate spatial transcriptomics into their research program? What learning resources would you recommend?
The large cost of the reagents and uncertainty of what you will get for this investment are certainly daunting. At some point, you just need to try it to find out for yourself. But don’t underestimate the sample preparation steps. In my experience, good quality RNA going in will mean much better results coming out. Conversely, no amount of slide efficiency and computational know how can overcome poor tissue input. So take the time to make sure you’re giving yourself the best chance for success. For resources, 10x customer support has been great over the years, from our initial scRNAseq experiments to our ongoing spatial experiments. So don’t be afraid to reach out to them. Finally, I initially learned the computational analysis side from online tutorials from the Seurat website vignettes. There is a barrier to entry for someone not familiar with R or coding, but it has been a skill set well worth the time to develop.