Scientist Stories: Raheleh Rahbari, Single-cell heterogeneity in the human brain

So good evening, everyone. Before starting my talk, first I would like to thank the organizers for inviting me to give a talk and contribute to this very nice conference. So here are my slides and declaration of conflict of interest. I have no commercial disclosure. And since this is an educational session, I think I will start with a few slides on introduction about somatic mutations and neurodegeneration. So as we all know, during the normal natural process of aging, a lot of mutations occur and accumulate in different tissues of our body. And essentially, the natural process of aging is a very complex mechanism which contains altered cellular processes, oxidative damage, and defect in DNA repair, and many more. However, it is not really clear what are the genetics impact on essentially neurological mechanism that underlie normal and pathological aging. So I should confess that I'm not a neurobiologist, and my interest is essentially in mutational burden during natural process of aging and what are the mechanism that underlying this type of mutational burden. So to my idea, actually, brain is one of the most interesting tissues in order to study mutational burden during the lifetime and the normal process of aging, mainly due to the fact that majority of the cells in the brain are post-mitotic, as you may know. So these cells have a very limited ability of regeneration. Hence, any landmark of aging essentially recorded in this tissue, and one can study brain in order to understand more about the mutational burden during aging. So as the schematic diagram shows here, for instance, postzygotic mutations during development do inherit to daughter cells, and they essentially always stay there. And— oh, sorry, goes back. And also during the normal essentially stage of aging and adulthood, mutations can acquire where they essentially stay in the brain architecture. And this actually causes a brain mosaic sort of tissue where it has different type of mutations accumulated in an adult and makes adult brain different from one and another. And there are many studies which have looked into, for instance, mutational burden in brain, but using a bulk sample is really, really hard and it doesn't give us enough resolution to look into the level of mosaicism in the brain. And therefore, single-cell analysis come handy. However, there are quite a bit of still methodological problem with false positive rate with whole genome and transcriptome amplification, it makes the process quite harder. However, recent years, there are some new bioinformatics and statistical analysis that really enable to try to overcome some of these high error rate due to whole genome and transcriptome amplification. And now some groups have started actually looking to different cellular processes by looking at the mutational burden, i.e., SNVs accumulation in single brain or single neurons. So, a study published by Christopher Walsh's group last year, they have actually tried to make an attempt in terms of what are the mutational signature or processes that can actually identify or show a landmark of different aging stage. So, they have defined different mutational signature. One of the two major one is Signature B, which is really not age-related. And this is essentially— they have claimed or they suggest that this mutation is mainly postzygotic mutations that happen during development. And another mutation that they have identified, which is quite interesting, is Mutation A. And this mutation essentially accumulates during aging, and this is essentially a triplet mutational signature. Composition analysis. And when they look at the different brains at different ages, for instance, an adult brain at 80 years old compared to the younger brain, they observe this nice correlation of accumulation of this type of signature with age. And actually, this signature is not a new signature. It has been previously published and well-characterized in cancer study, and it is quite abundant across different tissues, which is mainly related to deamination type of signature. So, as I said, this is really nice that we can overcome some of the false positive error rates and start exploring what are the cellular processes that may cause different genome mutational burden at different stage of aging. Another very interesting study that I really like is a recent paper published by Matthias et al. this year, earlier this year, and actually they have emphasized on kind kind of relation of the cell type and sex-specific differences that they observe in Alzheimer's disease brain. So, in the first plot here, they highlighted how different genes essentially differentially express in Alzheimer's pathology depending on how or which progress of the Alzheimer's they have sampled the patient from. And they clearly see differentially expressed genes that are grouped in different cell types quite differently. So what they observe is there are many genes that express differently, especially in neuronal cell types comparing to astrocytes, oligocytes, and all the other cell types that they have analyzed. And more interestingly, they have observed differential expression pattern, especially in a female who had Alzheimer pathology. And here they have shown the pathological traits. So all the red ones are the pathological traits and the blue one is the cognition trait. And what they show is different cell types such as excitatory and inhibitory neurons demonstrate different type of expression compared to the median expression across the similar cell types in female comparing to male. So this is actually really interesting what they have highlighted so far in terms of the level of mosaicism, in terms of somatic mutation accumulation when they look at the genome or when you look at the transcriptome and you observe that actually mutation accumulation are very dependent on different cell types and it might also depend on sexes also differently. So all of this evidence highlights the importance of doing a single-cell analysis in order to really identify what are the underlying mechanisms of neurodegenerative disorders. But really, before getting into neurodegenerative, I think it is really important to do a more deep study to identify actually what are the levels of heterogeneity in normal brain. With that in mind, I'm going to present some of the ongoing work in our group in order to identify or better explain the level of heterogeneity in natural aging brain. So, I will present two approaches we have used so far. One of them is using 10x Chromium DNA pipeline, and the other one is single nucleotide genome and transcriptome approach. So, I will first start with the 10x Chromium experiment. Maybe some of you are familiar with this new pipeline that has become available by 10x Genomics quite recently. Essentially, the single-cell suspensions are entered into these cell beads, and these cell beads then bind with all the enzymes and produce this cell bead gel bead structure, which they then uniquely barcoded, and we can do a parallel sequencing of quite a lot of cells at the same time. So, one question we had is, what is the level of single-cell heterogeneity across different brain regions in an adult brain? And for that reason, we applied this 10x Chromium approach on just two random tissues that first was available from this normal donor. So, he was 28 years old at the at the age when we got the sample. And the reason of death was natural cause, so there was no neurodegenerative sort of brain or disease involved in this brain. And we have sequenced about 3,500 single cells across these two regions, cerebellum and cerebrum, and we kept the difference between the left and the right lobe differently in case we are observing any type of the single-cell heterogeneity across different lobes of these regions. So first we look into— we did look into CNV patterns, copy number variation patterns across these four brain regions. So on the x-axis we have the genome and on the y-axis every row contains all the single cells that we have analyzed for each of these regions. So the gray whitish area usually have a copy number neutral and when it is blue it's usually deletion of the genome and the orange color depending on the gradient, it means there is an amplification of the genome in that region. So I'm not going to go through too much of the details, but just to highlight the fact some of these nodes up here shows that there are more than one single cell that have a similar copy number profile for these sort of— in these regions. So if we put all this together, What the data show actually is a bilateral clonal mosaicism for CNV profile across these two brain regions of this normal person that we have analyzed. So this is interesting, but obviously when we have only genome data, it makes our kind of inferences a bit limited. With the fact that I show you, or presented earlier, that groups have shown there are cell types specific, for instance, association with Alzheimer's disease, for instance. And also it is important to know what are the functional impact of this CNV in terms of maybe neurodegeneration or if they have any protective effect or et cetera. However, with only genome, it is really hard to interpret this data further. And for that reason, we used single nucleotide genome and transcriptome method Some of you might be familiar with GNT-seq approach which we have published a few years ago now. This method, the single nuclei, is a new modified method where we can actually isolate single nuclei from a cell and essentially do a parallel DNA and RNA sequencing. This method is very useful for very cell-dense essentially tissues such as muscles and brain where isolation of a single cell actually is quite hard and you cannot isolate a single cell intact. So it is safer actually to isolate a single nuclei and study the genome and transcriptome as it has shown that it recapitulates the kind of genome and transcriptome profile of the whole cell. So just to give you a brief upshot of how the method works, we identify or select the region of the brain that we want to study. We homogenize the tissue and essentially extract the nucleus. Also, it is possible to use different nuclei marker to only select the cell types that we are interested to study. So, for our purpose of the study, we use neuron-positive cells in order to only study neurons. So, for the remainder of this talk, I'm going to show you all the study that we've done on neurons only. So, once we have this single nucleus isolated, we can actually dissociate mature RNA and DNA from each other. And subsequently, we can perform whole genome or whole transcriptome amplification followed by sequencing to study what is the genome and transcriptome of this single nucleus. From neuron, from normal brain at the same time. So, for whole genome amplification, there are two big major kind of amplification methods that some of you might be familiar with. One of them is PCR-based genome amplification such as MALBAC or Picoplex. This method is really good and useful for copy number and big structural variation analysis mainly because the genome amplification is quite smooth across the genome, breadth of the genome, so it is more reliable for copy number analysis. Another method is isothermal amplification such as MDA approach, and this is proven to be quite good and reliable for SNV calling because it has lesser false positive rate. So we know that already de novo CNVs represent rare variants; however, their effect is quite actually strong, and it is already reported correlation relationship of de novo CNVs with genetic risk of neurological disorders. However, it is still not really clear what are the impact of these CNVs and how they contribute to brain development and also function and if they actually lead to any neurodegeneration. So with that in mind, we studied prefrontal cortex across 5 human brains. Their ages vary from 41 all the way to 72, so there is a bit of about 30 years of age gap between the youngest and the oldest donor. And we sequenced a whole genome and whole transcriptome of nearly 600 single cells across these brains together. So, for the whole genome amplification, first to look at the CNV profile, we use Picoplex method, as I said, because it is more reliable for copy number analysis. Using read depth, focal read depth analysis, we can show the CNV profile, and here I have demonstrated 4 of these neurotypical brains across these normal prefrontal cortex neurons. So, as one would expect, the copy number is neutral, and because these are post-mitotic cells with lesser possibility of regeneration, hence you don't expect to observe a lot of copy number aberration. However, story does not stop here, and we observe some neurons with actually megabases of copy number gain and losses. And this is not the first time actually we are reporting these changes. Last year, McConnell group have also identified unreported copy number aberration in neurotypical brains. And what are the underlying mechanisms and if they have any functional impact still really is unclear. So when we put this copy number profile information together per brain, we observe on average about 4% of the neurons have copy number aberration. So here I'm just showing one of the copy number profile across 100 single cells from one of the brain that the majority have copy number neutral and they look yellow, but there are some of them, about 4% of them, which have copy number aberration. So that is fine, but to make the story a little bit more complicated, we started observing some also other peculiar things about these copy number changes. So here I'm showing a copy number profile from one of these CNV neurons, and this time I have plotted the copy number profile per chromosome. So here we have chromosome 1 all the way to chromosome Y down here. So some of you might have already spotted something different from this copy number profile, but if you haven't done, I'm gonna distract you for a minute to introduce you to a phenomenon called chromotripsis event. Maybe some of you are familiar if you are working on cancer genomics. So chromotripsis is essentially a chromosome shattering event, which then the chromosome shattered into pieces and then stitched back to each other in a kind of a random manner. And it is first identified in chronic lymphocytic leukemia,, and subsequently has been reported afterwards in different types of cancer. Actually, it is one of the hallmarks of cancer evolution. So surprisingly, the pattern is very similar to what we observed in here. So as I said, this is the CMV profile of one single neuron where we actually show the copy number profile per chromosome. And as you can see, only in the highlighted chromosome we see quite a lot of chromosome deletion and amplification at the same time, which is very similar to the phenomenon that people have observed in chromotripsis-like event. Again, what are the mechanisms and what are the functional impacts of these chromotripsis-like event in a normal brain? It's really hard to explain. But we have some information, as I mentioned, about the transcriptome. So, we can add a little bit more information, and at least in terms of the cell types and if we can identify any functional impact on the transcriptome of these CNV neurons and chromotrypsis-like neurons. So, for all these 600 single cells, we have also whole transcriptome amplified their or mature RNA using SmartSeq2 library. So, this is just to give you a brief idea about the level of gene expression. Per each neuron, we identify about 3,000 genes that express. And when we do a cluster analysis, we identify two major clusters for these neurons. When we look at the cell markers, actually we can divide these two clusters into two main clusters, excitatory excitatory and inhibitory neurons. And actually, some of these gene markers are very standard gene markers that have been identified and used to mark these type of neuronal sub-cell types previously. So, it is great. Now we have some cell type information and we know the CNV profile of these neurons and we can put this together. And actually, we did that. When we put the CNV profile in the context of the cell type specificity, we cannot see any cell type-specific CNV neurons. So the CNV neurons is actually quite abundant across both cell types equally, and it is not exclusive to one or the other. Also, we looked into copy number variation on the transcriptome, and this is work done by a student in the group who has been looking into the gene expression of these CMV neurons and chromotrypsis-like neurons by comparing the level of their expression and subtracting it from a median gene expression across all the cells within the same cell type. And what he has highlighted here is all the chromosomal or CMV aberration on the transcriptome that is different from the median gene expression from the given cell type. To look at it a bit more closely, this is one of the examples of the neurons. When we look at the structural variation and different translocation duplication in this neuron which had a CNV in the neuron, we observe quite a lot of actually structural variation and translocation, et cetera. So it's really from the genome, it is actually quite a messed up genome. And when we look at the transcriptome, we observe quite a lot copy number gain and loss comparing to the similar cell type gene expression profile. So, this is just to give you an idea. For instance, in chromosome 8, which is kind of easier for this cell to detect, we observe quite a lot of loss of this chromosome, at least one copy of the chromosome from the genome data. And actually, when we look at the chromosome 8 transcriptome data, also we identify quite a bit of a deletion or less expression when we compare it to the similar inhibitory cell types. Moving forward, this is again another example. This time we have a chromotrypsis-like event where possibly you can see there is quite a lot of structural rearrangement that we can observe in these single cells. And Also, the number of the copy number changes in the transcriptome is quite substantially higher than what we have observed in the similar cell types. So, it looks like that actually there are some correlation between the DNA CNV profile and transcriptome. And obviously this is just the tip of an iceberg, and we still need to explore a little bit more deeper in order to identify cellular enrichment or pathway enrichment that might explain whether these copy number aberrations underline any specific cellular pathway, et cetera. So moving forward, as I mentioned earlier, we have also opportunity to use isothermal type of amplification such as MDA to study the single nucleotide variation. And obviously, the single nucleotide variation represents a more substantial source of somatic mutations. And for that, it is quite actually interesting to look at what are the mutational burden in this brain. So, for that reason, we selected 10 neurons per each of the same brain from the same region, prefrontal cortex, and we performed the G&T method this time For the gene method, we use MDA approach. So here we show that deep bulk sequence data for these 5 brains, and when we look at— we call the SMVs on the DNA layer only across these 50 neurons from 5 brains, and also we separately call the SMVs on the RNA separately. As you can see, obviously, or you would expect, there is a different level of the SMVs that we can code on DNA versus RNA. Something which makes this study more interesting is when we look at the mutational spectra in these normal brains using SMVs only from the DNA data, we generated mutational spectra. So this is the 6 mutational spectra. We do not have the resolution or enough sample size to do 96, signature analysis, but we can look at the mutational spectra. And when we look at the DNA SMVs, this is the spectra of the DNA. We can also do the similar analysis— oh, sorry— on the RNA data, and the mutational spectra actually is quite different. However, when we overlap DNA and RNA data together, so we select a subset of SMVs that are called in the DNA and also confirm in RNA, we have quite a bit of a different mutational spectra. And I think, to my opinion, might be biased, this is one of the more closer to the truth in terms of the mutational spectra in single-cell analysis in brain. This is mainly due to the fact that there is a high level of false positive rate due to amplification methods that we cannot actually so far extract from the data. So again, I would like to emphasize how much multilayer information can help us here in terms of the essentially false positives that we encounter with the amplification method. Lastly, I want to just quickly share a study that we did in terms of comparing human brain with mouse brain to identify actually if developmentally these brains are— how much they are different in terms of the somatic mutational burden and level of heterogeneity. For that, we sequence again genome and transcriptome of 100 neurons from mouse hippocampus. This time we actually selected deliberately hippocampus mainly due to the fact that there are there reside some dentate gyrus neurons where they have regeneration abilities. So with that in mind, we thought possibly we can capture a lot more heterogeneity in this region of mouse brain. So we perform cluster analysis, and we can identify actually 4 neuronal cell types in the hippocampus, including the dentate gyrus CA1, 2, and 3. When we look at the copy number profile, it is actually quite surprising. We almost find 0 to 1% copy number variation in mouse brain, in normal mouse brain. And it looks like that mouse brains seem to be more stable in compared to human brain in terms of at least level of heterogeneity in their brain. There are some speculation to explain this observation. Maybe this is something to do to the to the volume of the brain, imagining that the mouse brain volume is a lot smaller than human. So during development, actually, the neuron progenitor goes into a massive expansion to essentially generate the whole volume of a human brain comparing to mouse brain. So maybe this rapid expansion can cause more copy number aberration and cellular heterogeneity in human brain when compared to the mouse. But again, this is something that we need to look into it deeper. So to kind of wrap up, I would like to summarize what I have talked about. Essentially, single-cell resolution is very crucial for studying essentially brain heterogeneity and how it can impact or cause neurodegeneration in general. Also, multi-omics approach is really important to identify functional impact and also overcome some of the systematic limitations that we— that limit us in order to correctly estimate mutational burden in this tissue. Also, we observe some fundamental neurodevelopmental differences in in terms of genome heterogeneity between human and mouse. And obviously functional consequences of CNV neurons and its relation to aging, it is really still remain at its infancy and we need to do more work in order to identify what are the reason of observing these CNV neurons, if they have any protective sort of phenotype or actually they may be more— make more vulnerable to neurodegeneration during aging and adulthood. So with that, I just wanted to say I didn't show you a lot of data on actually neurodegeneration, so this is ongoing work. Now we are expanding the work to compare the normal observation in normal brain versus different types of neurodegeneration in human, and also we have different mouse, and also fly model in order to recapitulate some of the observation that we made so far. So with that, I should thank quite a lot of my colleagues and groups across the world, really, and our funders. And above all, I should thank all the brain donors who, without them, we could not make this study happen. And with that, I thank you for listening, and I'm happy to take any questions.