Question: What copy number variant (CNV) calling methods are compatible with Visium for FFPE data? With Visium for fresh-frozen?
Answer: CNV calling methods compatible with Visium for FFPE will be a subset of those compatible with Visium for fresh-frozen.
Visium for FFPE uses RNA-templated ligation (RTL) probes targeting the whole transcriptome. The assay does not capture transcript directly, but captures probes via a capture sequence, e.g. poly-A for Visium for FFPE probes. The sequencing readout reflects the designed probe sequences. Polymorphisms and mutations are absent from the readout. Thus, RTL gene expression data is not amenable to secondary analyses that assume expression data represent germline SNP and indel variants.
One coverage-based method to call copy numbers is CopyKat (https://github.com/navinlabcode/copykat). Another is InferCNV (https://github.com/broadinstitute/infercnv).
In contrast, for Visium for fresh-frozen, which directly captures 3’ ends of transcripts via poly-A tail, we suggest trying copy number calling methods that incorporate allele information, e.g. Numbat (https://www.biorxiv.org/content/10.1101/2022.02.07.479314v1; https://github.com/kharchenkolab/numbat). Another tool for Visium for Fresh-Frozen is souporcell (https://github.com/wheaton5/souporcell). Souporcell calls genotypes de novo and in some cases it can call a tumor genotype from single cell and Visium data.
To reiterate, RNA-templated ligation probe-based gene expression data excludes germline variant information and so only coverage-based CNV methods are applicable and CNV methods that depend on allele-fraction information are not.
Other general advice
For copy number variant calling, in general, consider single-cell methods for spatial spot data. Data interpretation should keep in mind Visium resolution is not single-cell but averages ~10 cells per spot. Each Visium spot has a 55 µm diameter and the spot-to-spot distance is 100 µm. If data is H&E stained or includes a DAPI stain image at high enough resolution, then it will be possible to estimate the number of cells per spot, e.g. using cell segmentation methods (unsupported). This information can also aid in data interpretation.
Depending on the tumor, researchers can additionally identify tumor versus normal regions based on differential expression patterns and tissue anatomy, e.g. pathologist annotations. Having anatomical annotations for the aberrant areas of the tissue will aid in analysis. For some tumors, tumor versus normal tissue gene expression should vary enough that PCA and projection as well as clustering results will differentiate these areas. Furthermore, if the tumor type is not migratory nor admixed, then the tumor versus normal cells should be spatially distinct, i.e. somatic cells should be physically proximal. If so, then the spatial grouping should aid in data interpretation, especially towards tumor purity estimates.
Useful links
- Instructions to import pathologist annotations into Loupe are at https://kb.10xgenomics.com/hc/en-us/articles/360058943912.
- We describe RTL probe design at https://support.10xgenomics.com/spatial-gene-expression-ffpe/index/doc/technical-note-visium-spatial-gene-expression-for-ffpe-rna-templated-ligation-probe-design
- Introduction to integrating single-cell data and Visium spatial data https://www.10xgenomics.com/resources/analysis-guides/integrating-single-cell-and-visium-spatial-gene-expression-data. Integration of single-cell data is just one approach to deconvolving cell types. Methods that focus on the spatiality to annotate regions of cell types without needing single-cell truth data also exist.
- A resource that collects methods relevant to spatial data is the Museum of spatial transcriptomics from Lior Pachtor's lab at https://www.nature.com/articles/s41592-022-01409-2. The github IO equivalent is at https://pachterlab.github.io/LP_2021/ and drilling down eventually you will end up at the googlesheet at https://docs.google.com/spreadsheets/d/1sJDb9B7AtYmfKv4-m8XR7uc3XXw_k4kGSout8cqZ8bY/edit?usp=sharing. For example, the googlesheet's Analysis tab lists some deconvolution methods such as RCTD and SPOTlight.
Last modified June 1, 2022
Disclaimer: This article and its resources are provided for instructional purposes only. 10x Genomics does not support or guarantee the contents.
Products: Visium for FFPE, Visium for fresh-frozen