Well, it’s inadvertently been sort of a Neuroscience Week here. This latest paper is a very interesting addition to the field indeed, just out from a very large team centered at the Allen Institute, where some rather large-scale work in the field has been done in the past. This one continues their tradition: it’s a look at single-cell-nucleus RNA sequencing in one brain region (the middle temporal gyrus), and seeing how much these standard classifications look when you get down to the gene expression level. And what’s more, they do the exact same in both human and mouse brain samples and compare across species. The MTG was chosen because it’s a region that’s often the subject of brain surgery for epilepsy, and there are thus frozen post-mortem samples available (as well as, in this study, a few that were provided immediately after such surgery itself). Isolating cell nuclei is also much easier then teasing apart individual neurons, and smaller single-cell studies have already shown that cell types can be distinguished from RNA in the nucleus.
There are a lot of results to think about. First off, it’s important to stipulate that the cellular architecture of the human and mouse brain tissues are “surprisingly well-conserved“, as the authors put it. This has been apparent from anatomical studies (at increasing levels of detail) for many decades. In other words, if you were handed a small brain tissue sample of this sort and some high-quality microscopes, you would have to bear down on them to say if you were looking at a human brain or a mouse one. They’re organized in very similar fashion with a similar apparent diversity of cell types. There are some distinct astrocyte types in human brain tissue as opposed to mouse and different distributions of other types (such as “rosehip neurons”), but these differences are not always easy to spot.
Now for the RNA sequencing. For the human side of the experiment, eight donor brains furnished 15,928 cell nuclei, which were taken layer by layer, and the general transcriptional profile sorted these into 10,708 excitatory neurons, 4,297 inhibitory neurons and 923 cells that weren’t neurons at all. Person-to-person variation didn’t seem to throw the results around too much, because the pooled samples still showed very robust clustering into transcriptional types. (They could, however, see some small but real differences between the post-mortem samples and the freshly derived ones, interestingly, although these still binned into the major clustering scheme just fine). Overall, there were about 75 transcriptionally distinct cell types in those >15,000 cells, and they fit (roughly) into the known developmental lineages. The upper-layer excitatory neurons had the most duplicates, but there was definitely a “long tail” distribution over the entire samples, with most of the types being rather rare.
Most of the neuron variations were quite spatially restricted. That said, they noticed particular excitatory types spanning several layers (while others showed up only in particular locations). The inhibitory neurons didn’t have as much layer-spanning, but definitely showed similar enhancements in different cluster types across the layers. The non-neuronal cells were more evenly distributed spatially (except for one particular astrocyte class), but the authors note that the clustering for these is more provisional since the sample size was so much smaller, and that a larger sample would surely reveal more distinct classes of (for example) astrocytes than we know about now. It’s worth noting that the broad features still match up very well:
“Despite differences across datasets, alignment based on expression covariation reveals a cellular architecture that is largely conserved between cortical areas and species, as anatomical studies have shown for the last century. . .Beyond similarities in overall diversity and hierarchical organization, most cell types mapped at the subclass level, seven cell types mapped one-to-one, and no major classes had missing homologous types despite the last common ancestor between humans and mice living at least 65 million years ago and despite the thousand-fold difference in brain size and number of cells. . .”
But even at this level, the differences are still daunting enough. It certainly fits the general neuroscience template of “the more you look, the more you see” – brain tissue seems to have a really terrifying level of differentiated detail no matter what technique you examine it by. But now the group compared the mouse and human cells more directly, focusing on the primary visual cortex and the anterior lateral motor cortex. The same genes tended to work as classifiers in both species, although less well in the non-neuronal types. There were three of the rare types in mouse that were not present in the human samples, although the authors note that these could well show up in larger samples. Overall, some particular classes were more diverse in the human samples, and some were more diverse in the mice.
Bearing down, the paper compares the expression of over 14,000 RNA transcripts across the two species, and they found that nearly 10,000 of them had divergent expression in at least one of 37 homologous cell types they identified. Moreover, many of these had such changes in only one of those types. The non-neuronal types had the greatest differences, suggesting that these have had the biggest evolutionary changes between mice and humans. Overall, though, the mice and human cells have very different expression patterns, even when you match down to particular cell types. While the overall organization holds quite well, levels of individual transcripts are quite different, and entire genes show up in one but not the other.
Most interestingly (and alarmingly?) the biggest divergences were seen in things like neurotransmitter receptors and ion channels, which of course have been extremely active areas in drug discovery. This is direct evidence for how poor mouse CNS models are, and why. That is not news in general to anyone who’s done that sort of work, since everyone knows that yeah, mice ain’t humans. But this shows that even down at deeply reductionist levels, mice really are not humans. Our serotonin receptors, our glutamate receptor subtypes – they’re expressed very differently and in different ratios to each other. If you’re making ligands for those sorts of things, they are going to do different things in mouse tissue and in live mice than they will in humans, and there appears to be no way to get around this. Our brains, mouse and human, are organized in very similar ways, but the tiniest gears and pulleys are hooked up in very different fashion. On a philosophical level, it’s quite interesting to see how such broadly similar architecture has been adapted to such different species, and also to see how that architecture can also contain and accommodate such varying mechanisms at the most fundamental levels. We now have a lot more for our own brains to work on!