As a partial answer to your question (and I encourage others to add onto this), I will say that we do not see any obvious bias in cell type abundances of neurons in our snRNA-seq data sets at the Allen Institute. More specifically, in all cases where we have performed in situ hybridization (or related spatial transcriptomics methods) to assess cell abundances, the values across the two modalities are consistent. This is in contrast to single cell RNA-seq, were we find strong biases against collection of certain neuronal cell types.
That said, the abundances that are measured depend heavily on the accuracy of the dissection, which is often challenging in human tissue. For example, cortical layer 1 is known to contain a wide variety of interneuron types, but is devoid of excitatory cells and is relatively cell sparse; however, our layer 1 dissections in several cortical areas contain both excitatory and inhibitory cells, likely due to slight clipping of layer 2 in these dissections. Dissections become even more challenging in regions with lots of white matter or when you need to worry about 3D structures.
Evan Macosko told us that his lab hypothesized that the high abundance of oligodendrocytes that they measured is “because of the thick white matter tracts that surround the SNc, and the difficulty in dissecting them away from postmortem tissue blocks.”