First, I have a very naive / basic question about the way that transcriptomics data can be used. Would it be possible to use it to plot the density of cell types across brain areas? (Similar to what you can do with histological staining, as in Kim et al. 2017 Neuron, https://doi.org/10.1016/j.cell.2017.09.020.) Or are there intrinsic limitations due to the way the cells are sampled?
If so, could you help me to locate the relevant data file in mice, with the number of cells for each gene per brain area?
This should be possible using either mouse MERFISH data set available as part of the ABC Atlas. What you get from these data is the X, Y, Z coordinates, cell type assignments, and gene expression levels (for a subset of genes) for every cell identified, and so in principal you know the distribution of all cell types in the brain. I would recommend waiting to do this, as eventually these sections will be aligned to the CCF and therefore each cell will also have an assigned brain area.
Such an analysis is typically not possible across the brain using single cell/nucleus RNA-seq data since the tissue is sectioned by brain region prior to sequencing. You can, however, get a good estimate of cell type density within each anatomical dissection, which is (arguably) at least accurate as what you get using MERFISH. This response is my opinion, and so I’d encourage other folks to weigh in!
Thank you for the quick reply. So if I understand correctly, the nucleus RNA-seq data should be well suited if I would just be interested in an average density per brain region (not differences within brain region)? Is it possible to get access to such a datafile, already giving the average results per brain region?
I think I may have confused what I meant by density. With single nucleus RNA-seq you can estimate the relative proportion of cell types within a dissection (with respect to other cell types) but NOT the absolute numbers or the relative numbers with respect to a given area. With single cell RNA-seq the estimates are a bit less reliable due to technical biases but is probably still reasonable. I believe you’d have all the information you need if you follow “Cell Metadata” part of this Jupyter Notebook.
Correct. We’ve found that single nucleus RNA-seq collects cells of all types with minimal bias, where single cell RNA-seq can collect some cells more successfully than others. In human tissue the differences are dramatic: the vast majority of neurons don’t survive the process (which is why we nearly exclusively perform single nucleus RNA-seq in human tissue!). I’m not sure of the details in mouse brain (maybe others can comment?) but I think it varies by brain region from minimal bias to high bias.
Happy to help! In theory, stereo-seq should also work for this and could be used in a manner similar to MERFISH. I suspect there are biases, but am not familiar enough with this technique to comment.