Scientist Stories: Jens Hjerling Leffler, Single cell RNA sequencing in Psychiatric Disorders

So thanks to the organizer for letting me speak this session. We have a huge auditorium for not that many people. I wish we had a smaller auditorium so we could have more interactions. If you really want to ask a question, just stand up and shout, and I will try to repeat the questions to the audience. Okay, so I don't have a slide, just what do you call it, the conflict of interest or something like that.

I should say I have a small funding of $30,000 or so from Roche in order to do something that is not related to what I'm talking about.

It's— okay.

And I should also say that I'm a developmental neuroscientist.

I'm a bit nervous about talking to the human genetics crowd. So basically I do neuroscience and then I use your data to try to make sense of what we do and vice versa. So it might be that my question— my ability to answer some of the more genetic questions might be limited, but we'll see when we get there. OK, so cellular diversity in the brain. I use these two images. This is something really central to what I'm interested in my lab. And if you're a neuroscientist and you manage to publish a reasonably cool paper and the public press takes any notice, you either get one of these two pictures to illustrate what you've done. I use them as kind of like an example for two different ways of seeing it. This is kind of how modelers that model the brain, this is what they think it looks like. It's gray spheres floating in space, and they're all exactly similar to each other.

Sometimes if it's online, you'll see a small action potential in the form of a flash going along one of these axons.

And this is, of course, one extreme to see that all neurons are pretty much the same. The other one here is this is an image of a rainbow mouse, and I use this to say to the opposite view, which is that all neurons are different. And we call this the snowflaking approach.

And every neuron is different.

And what we do in my lab really goes from the assumption that the truth must lie somewhere in between. In order for us to have any chance of making sense of the brain, we need to have some kind of organizing principle behind the cells in the brain.

So we think about a lot of things about neuronal subtypes. And I want to point out that this is a practical concept.

It does not hold as a philosophical concept because where do you draw the limit? That makes you crazy.

And it can't really be done. But we see it as a practical concept. We see it as a hierarchy of identities.

We can probably all agree, or most neuroscientists at least agree that we have neurons and non-neurons. And then you take the neurons, you can divide them into excitatory cells and inhibitory cells. We all agree. And then you can probably divide them into further, but as soon as you start going down a bit more on the finer subdivisions, people start disagreeing.

So we think about this as a hierarchy. And the reason we do this is that we want to—

not maybe for you guys, but for the people that do neuroscience— we want to introduce what we call an informed reductionism. You have to choose which level you study the brain at. It's OK to choose a level, but when you do that, you should be informed. You should know what kind of information you're ignoring and what kind of principles you're looking at.

So my background is really in neural development of GABAergic interneurons, the inhibitory cells of the cortex. They're born in the bottom.

And I'm using this as an example of the hierarchies.

They're born in the bottom in one of the two regions, the medial ganglionic eminence or the caudal ganglionic eminence. And then they migrate throughout the brain up into the cortex where they reside nested inside among the excitatory neurons that are born locally. And we can use mouse genetics to label them when they're born and follow them as they migrate up.

And if you look at the inhibitory interneurons of the cortex, it turns out that they can be divided into three different flavors.

They either express parvalbumin, somatostatin, or serotonin receptor 5-HT3A.

And that, if you have those three markers, you can account for 100% of the neurons. Okay, that's great. So there's three types of GABAergic neurons. Actually, this is my postdoc work, and it seems that most of the field agree with me because they're happy that it was a simple image. But if we just look at the green ones and look a little bit more in detail, what you'll see is These neurons, they're all expressing 5H3 receptor.

They are residing in different layers of the cortex. They have very different electrophysiological profiles.

And they also have different morphologies. And there is residual variability when it comes to expression of other genes. So it was clear to me when we started this—

this is some 8 or 9 years ago— and that this Collecting this kind of data is very slow and painstaking.

You get maybe 5 cells a day, if you're lucky.

And it was clear that the molecular code that we had was not enough to capture the variability.

So I collaborated with a guy, Stan Linasson, who had just invented one of the chemistries to do single-cell sequencing. So since I studied the heterogeneity of these neurons, I teamed up with Stan.

And what we did was, you all know it, it's single-cell sequencing. You take a piece of the brain, you take out the cells, you generate single-cell expression profiles, and then you use a computer to identify the cell clusterings. So, we did this in two regions of the brain, the somatosensory cortex and the hippocampus. And this was published 2015.

This is the first large-scale study in the brain. And we found many different types of cells. So, I'm not gonna go more into this. There are now several different methods to do this.

We heard about SmartSeq in the talk we talked before. 10x Genomics, of course.

And there's maybe 20 different chemistries you can use. And depending on what kind of information you're looking for, it's going to be different methods that are best for you. But the important thing is, of course, that when we do this, we get information on the expression of every gene in every different kind of cell that you look at. So that's the key thing. So my lab now, one aspect, one side of the lab is really trying to find the limitations of the methodologies in terms of neuroscience. From a neuroscience perspective. And the other one is see, can we use this kind of transcriptional information to say something about psychiatric disorders? So this is a slide that is the field of single-cell biology. It's exploded. There's like thousands of papers over the last 4 years. And now we ask the question, how can we use this to gain some new insights on complex disorders? And since this is a teaching session, I'm not just going to talk about my data. I'm also going to talk about 3 new papers that just came out. So of course, the most immediate thing you can imagine of how to look at this is, of course, to take humans that suffer from a disorder and compare them to normal people and do single-cell expression analysis. This is similar to the bulk sequencing efforts that we've been doing for many years. But now you add single-cell resolution, which is great because you're not averaging across different types of cell types, et cetera.

And just over the last year, 3 papers came out, one on Alzheimer's, multiple sclerosis, and autism.

Fantastic work. This is some of the stuff that we're doing too. That's why Roche is paying me. But I just want to have a few comments on if you are to embark on this, or what are the pitfalls of these studies. So this is one. You can kind of get 2 great types of information. One is you can understand the changes in the cellular composition of the brain. So, in terms of schizophrenia, which is kind of where I work, we still don't know whether some cell types are missing or not in the schizophrenic brain. So, this is the kind of information you can get out. And the other one is, of course, cell type-specific changes in the expression then of the gene regulatory networks or individual genes. So, that's super interesting. And I think it's, you know, we're looking forward to a great next 5 or 10 years.

So, super powerful.

What the data you cannot really do is then separating compensatory or lifestyle changes from those causing the disorder. The second thing is this is super expensive to do. You need a lot of money. And if you're interested in disorder and you want to start this, you have to find some serious funders before you can get going. And then I want to issue a big warning, especially now because it's early days. The first papers are starting to come out.

Pay—

I'm happy to talk at a human genetics conference because you guys have been suffering with this for many years and you've been much better than us neuroscientists are. And that is you have to be very wary of the computational methods they're using in the paper. There has not really been set a standard in the field yet. A lot of people doing this are single-cell biologists or neuroscientists. And it's very tempting when you do single-cell neuroscience because single-cell expression because you get a lot of Ns if you treat each single cell as an N, you get thousands of Ns without just having a single patient. So at least some of these papers that I mentioned, or one of them, have treated each cell as a biological replicate rather than the patient. So this hyperinflates your p-values, and you are analyzing noise. So when these studies are starting to come out now, please pay close attention and treat them with a little bit of caution until us in the single-cell neuroscience field have kind of got our hands around what's proper use of data. Okay, good. So this is like the most, where a lot of effort is going right now, but I want to talk to you about a different kind of analysis which we have used in my lab. And this is work of, really a work of two postdocs. Nathan Skeen in my lab and Julien Briouard, who is working with Patrick Sullivan, is also at Karolinska and UNC. Nathan just started his own group at Imperial College, so he's no longer with me, unfortunately. One of these was published last year, and the other study is in bioRxiv if you want to go and find it. So with this approach, what we're trying to use is trying to use single-cell expression data from normal healthy individuals or mice, and then try to raster the findings from GWAS studies on top of that. So if we find which genes are carrying risk for disorder, we can ask a simple question, where are they expressed? So in terms of schizophrenia, which is kind of the interest of my lab, it's a very common disorder. It has a high heritability. There's not a single convincing gene, single gene cause. But instead, it's very spread out over the genome. So estimated 500-plus risk genes, depending on who you talk to. Some say 800. Some say 1,000. Some say 300.

But very many genes.

So each gene contributing a little bit of—

or each SNP contributing a little bit. So this is unique for schizophrenia, where we only have pretty this complex origin.

While for autism and intellectual disability, we have much more robust common finding. So for schizophrenia, this means when you have a known cause of a disorder, for us neuroscientists, we get happy because then experimentation and drug target identification becomes much easier. But in terms of schizophrenia, we have so much that is polygenic in origin that we don't really know where to start. So, if you look at the— you can, of course, do the gene set analysis, which people have done for schizophrenia.

And the risk genes for schizophrenia, they're basically, they're neuronal. They're genes intolerant for loss of function variation, genes that bind FMRP or CEL4, Rbfoxes, higher enrichment of synaptic genes. This has been debated right now.

But then also genes binding microRNA. OK, so that's kind of where we stand. So Nathan, when he came to my lab, he asked a really simple question. It's like, can we use the biology of neuroscience to structure the genetic data? Because it turns out that this genetic data is largely unlinked. It's like a cloud.

But he asked a very simple question then.

If you look at the literature of schizophrenia, you'll find papers that support that any one of the brain's cell types carry pathology. So you have a myelination deficit. You had changes in astrocytes. You have changes, definite changes in neurons, et cetera. But what if we take the information from the GWAS, the risk genes that we know increases the risk for schizophrenia, and we just ask which cell types use these genes? Because if you're not using the gene, chances are that you're not the cell type affected or behind the disorder. So to do this, We created— this is a good example of how quickly slides can be outdated. This slide is called the superset because it was the largest single-cell dataset at the time, but now it's a dwarf in comparison.

So we took— we were lucky in the lab because we had published several separate studies, all performed pretty much by the same people on the same pipeline.

So we had single-cell datasets from different regions that we know have been implicated in schizophrenia.

So we have the hippocampus CA1 region, somatosensory cortex, the striatum, hypothalamus, and the ventral midbrain.

And we said, can we just marry these datasets together?

And yes, if you look at different types of cells that you don't think should vary much between regions, you can see that there's largely no structures, so they're very much overlapping. The reason that the red cells are a bit on the outside is that these were much younger animals.

So, we said we can bunch this data together. So, that means that we have a lot of different cell types from different regions of the brain. So, that's the single-cell data. So, to do this analysis, we had to come up with a way to merge GWAS data with single-cell data. So, for the single-cell data, we created these kind of specificity of expression graphs for each cell type. This here is a medium spiny neuron of the striatum.

And what you can see here is the density of genes and then the specificity of expression of those genes in the cell types.

So, in green here and maybe in pink, you have the genes that you need to be a cell. This is nuclei, membrane, whatever, you know, the stuff that every cell has.

But as you move higher in specificity, you start getting into the genes that you need to be a neuron.

That's the ion channels, et cetera.

And as you move all the way to the top tier here, you start seeing the genes that are specific for certain subtype of neurons. And this is the dopamine 2 receptor that we use in our example, which is expressed by medium spiny neurons specifically. OK, so for each cell type, we can create this kind of like where different genes end up in which decile.

And then to marry this to the— to go from This is, of course, expression of a gene, but to go from GWAS, which is from SNP to gene, we use two different methods that uses different assumptions.

We use MAGMA and LD score regression, and we put a window here saying like if the SNP is close to this gene, we call this SNP to that gene. And then we just said—

so we didn't do a cutoff. We didn't take just the risk genes, just the genome-wide. All the genes and the risk that is associated with each gene. So, even if you have a low risk with your gene, that tiny amount of risk is in this analysis.

So, then we looked at— this is now the cell-specificity graph, this one here in a non-log scale. And then we said, okay, for each of these deciles here and 30 tiles here, we said how much of the inherited risk lies in each section. And then we plotted it like this. So this is now for LD score regression. You'll see as you move forward, this is for the medium spiny neuron.

And as you move up in specificity, you'll find an increased enrichment of schizophrenia heritability. So this tells us that there's more risk in the specific portion of medium spiny neurons than it is in the nonspecific portion.

And if you instead use the GWAS for height, for which we have no reason to believe that the striatum is involved, You'll see it's a flat line. And as you actually go more specific, it drops off.

You see less enrichment for height. So we did this for every cell type in the brain.

Here you see different examples of this.

This is for the LD score regression. And then we did the same thing for Magma. And then for LD score regression, I'm going to show you data.

And here we analyzed the enrichment in the last decile in red here.

For each cell type. But for the Magma, we fit a line and then we look at the slope of this line. So it's just two different ways and it's a way to try to work under different assumptions. And we want to find something that's robust, so we only consider something where both methods agree to be something that could be potentially of interest. Okay, so here's the actual data. So this is now mouse data.

Mouse single-cell data, and we look where the schizophrenia heritability is enriched. And it turns out it's actually mainly in 4 of these larger neuronal classes. It's in pyramidal cells, the excitatory cells of the cortex and the hippocampus, medium spiny neurons of the striatum, and also in cortical interneurons, in the GABAergic cells of the cortex.

But many of the usual suspects is not enriched.

So we don't see a good signal in oligodendrocytes, nothing in astrocytes. In dopaminergic neurons, there's some, but it's not significant.

So it means we know that dopaminergic signaling is a key feature of schizophrenia symptomatology or pathology, but it seems like the underlying risk might not hit the dopaminergic neurons. That might actually then be a secondary thing. And more importantly also— whoops, sorry, one step too far—

it's not in all neurons.

We do find neurons that are not enriched. So I would say that schizophrenia is largely a neuronal disease, but there is a hierarchy among along the neurons. Some neurons are heavily targeted by schizophrenia risk genes and some are not. OK, these are the larger classes of the brain. Of course, we can do the same thing with the very detailed, if you want to go down and look at many different cell types.

Here's all the level 2 cell types, what we call it. And you see now that we can say that the parameter neuron 1 of the CA1, median spinal neuron D2 receptor expression is a bit higher than the median spinal neuron D1. Et cetera.

And interesting here is the GABAergic cells. So a central dogma in schizophrenia neuroscience has been that the parvalminergic interneurons, the basket, fast-spiking basket cells, they are very severely affected when you look in pathology. So if you look at brain sections of people suffering from schizophrenia, these cells are affected. But all the changes that you can see in those cells are actually in line with changes that would be considered normal if you decrease the excitatory input to these cells. So, we haven't been able to say whether the pathology that we see of these cells, I think it's been pretty convincingly linked to the symptomatology of the patient, but we can't say that it's causative, because it might be downstream from other insults. And our data, actually, the cortical PV neurons does not really, in this case, doesn't make the list.

The striatal ones do, but they're very different. Instead, the cortical interneuron that is the most enriched is these reelin and DNF cells.

They're called neurogliaform cells, and they're a really interesting cell type, mainly because it's the one during development that integrates information about the surrounding brain areas. So the other neurons are pretty much hardwired in terms of their connectivity. But these guys here, these reelin neurons, if they don't receive the correct inputs, they actually change their developmental profile a lot. So anyways, be careful when we do these kind of things. But OK, so we did this. We see this cell type enrichment for schizophrenia. Now we ask the question, is this schizophrenia-specific or not?

Then you can compare it to other GWASes. And we made a cutoff that had to have more than 20,000 cases, more than 10 genome-wide significant hits. And then we just hide as a control. And you can see the first two ones here are two different schizophrenia sets. They hit the same cell types. I mean, they are subsets of each other. But then we look at educational years. We kind of see the same. And then we don't see a very significant signal in the other ones, but they don't recapitulate the same pattern.

In the end of my talk, I'm going to go through a new study where we're comparing this a bit more in detail. OK. Major caveat here is, of course, that we've been using mouse neurons to annotate the human genome data, human genetic data. So we want to try to do this now with single-cell data from humans. And to date, it's very hard to do human single-cell sequencing in the brain. So what you resort to is human nuclei. And the reason is that human neurons are so large that they tend to break when you try to dissociate them. So many groups have— I mean, there's a couple of papers, but most groups, including the Allen Institute and my group for many years, have given up. So we do nuclei, isolate the nuclei, and look at the RNA signal there instead.

So we've been using a mouse, which we have— at the time, we had larger data sets. We have better experimental control in terms of how we kill the mouse. And we can use single whole cell.

So now we wanted to ask the question, are we selectively missing something, or do we have some kind of bias in the data when we move from whole cell to whole nuclei?

And to do this, we used a dataset from Kahigas et al., published in Neuron, in which they took the pyramidal cell layer in the hippocampus.

This contains all the cell bodies. And then they cut out the neuropil, which contains very little cell bodies, and they just compared the gene expression between the two. And this is in an attempt to get those kind of transcripts that are preferentially exported into the dendrite and axons and compared to what you find in the cell bodies.

So, if we compare the blue and the red one here, the red being the cell body, what we could show is that comparing— if you compare two cell versus cell datasets, you actually see no difference in here.

But if you compare all the datasets that we could find for whole cell versus those that had whole nuclei, we see a completely very selective depletion of the transcripts that go out into the dendrites and axons. And for our kind of analysis, it's very unfortunate because the transcripts that go out to the dendrites and axons are, of course, heavily enriched for schizophrenia signal. These are the synaptic genes, et cetera.

So this is consistent across all the studies that we compared.

And again, if you compare the nuclei to nuclei, you don't see this difference. So I'm going to show you the human data now. This is work under construction. I mean, no, this is published, but with a caveat. So we have to be a bit careful that what we suspect is that we get much lower signal when we look at human nuclear data.

Even so, what we see is that we see enrichment in the same cells in the pyramidal neurons and cortical interneurons. These are two different sets, one from Allen Institute and one from the Broad Institute. And we see the same kind of enrichment there and see very little in other cell types. And the same thing in the second data set.

The big difference, I think, is that we see a much higher signal in the GABAergic interneurons than we did in the mouse data.

And we don't know whether this means that there's a loss of signal specific to pyramidal cells due to our selective depletion of mRNA species.

Or this represents a human specialization of GABAergic interneurons that is highly relevant for disorders. This is, we don't know yet, but data from Joe Ecker kind of suggests that it might be due to the loss of signal specifics, but we don't know yet. We repeated this in a new dataset now from the Allen Institute, and here we actually do see the GABAergic neurons being even more enriched than the pyramidal cells. So judge is out. We're trying to find ways to come over this. But the main thing holds true to our finding in mouse. OK, so in the beginning, I alluded to these gene sets that had been shown to be enriched for schizophrenia risk. It was neurons, synapses, things that bind mRNA, et cetera. And it turns out, if you go and look at— those gene sets, in which cell types are they enriched?

It turns out that these gene sets are now enriched in the very same cell types that was found to be enriched for schizophrenia heritability.

So it's kind of like a parsimonious explanation.

Why do we see these gene sets?

Well, both the gene set data and our data point towards the same cell types, although it's a bit circular. OK, so we have this schizophrenia heritability risk that is enriched in certain cell types. We now ask the question, is it the same genes that are conferring the risk to two different cell types, or do they have separate programs?

And to do this, we used a method where we normalized for the variability seen for the enrichment on one cell type and removed that portion from each cell type., and we actually found no, that each cell type remained significant when removing the other ones.

So, with one exception, and this is the red dot here, and that's, there's, if you remove the signal from the hippocampal pyramidal cells, there's nothing left in the pyramidal cells from the cortex. These two are very similar cell types, so this was maybe not so surprising.

But with the cortical interneurons, they have their own signal, and the median spinal neurons carry their own signal.

And this is something that is a very active research in my group right now, trying to understand—

whoops— trying to understand what are the biological insights we can get from this.

So one first is we took the different gene sets that we had, and we said, OK.

This is now just taking the top genes and see how much are they shared or are they individual for the four main cell types.

And you can see the largest numbers are these that are only expressed in a single cell type.

But there are some overlap in between. So we did this analysis and we can now say that the different gene sets, the Rbfox binding genes, they're enriched in the pyramidal cells. Anything that has to do with the dopamine pathway is completely specific for medium spiny neurons. Et cetera.

And this is, I think it's super interesting because what it suggests is that different parts of the brain, that the schizophrenia genetics is hitting different biological pathways. So, for instance, the classical synaptic pathway seems to be the cortex and hippocampus. The dopamine pathway, but also things that has to do with calcium integration seems to be more important in the striatum. And I think this is nice because it really gives us some kind of beginning to see how we could partition the signal from GWAS data into biology. Okay, this is the end part of the paper that we did publish. So I want to talk a little bit about the new study that we have on bioRxiv. We now moved from a dataset where we only had a few brain regions. In the meanwhile, we published together with my collaborator Sten Linnason published a new paper where we instead of 15,000 cells, we had 800,000 cells, single-cell sequencing from the entire nervous system of the mouse.

So, we took the entire nervous system. This is what it looks like when you do a t-SNE plot.

We classify them into cells, and then we say, where's the heritability? So, of course, unfortunately, the signal—

our findings hold, but it's a bit more complex. So when we have all these, this is now for schizophrenia, and this is the, if it's yellow, both methods are agreeing, if it's blue, only one method.

And as you can see now, the signal starts looking more panneuronal than it did before.

So it's something, the same neurons that we had before are still the ones that are the most enriched, and also when you start normalizing between subtypes, they are the ones that remain enriched. But we do see a larger panneuronal signal.

And we also did many other brain traits here. And I just want to give you a teaser so you can go and look at this.

This is all the different brain traits that we actually looked at. And this is how they hit—

how the cell type that they hit correlates with each other.

And you can see that all the psychiatric disorders are very highly correlated in terms of which cell types are being targeted— schizophrenia, bipolar, educational attainment, intelligence, ADHD, et cetera. While the brain traits that are like migraine, stroke, Alzheimer's, they hit very different cell types.

They are pretty much non-neuronal disorders when it comes to this. They are mainly— when it comes to stroke and migraine, they are mainly affecting the blood vessel cells. And Alzheimer's seems to be a microglia disorder when it comes to the genetic risk for them. So plenty of these kind of comparisons in our bioRxiv paper if you're interested. So, one prediction that this did was allowed us to do was we had a GWAS from the largest GWAS so far. We collaborated with 23andMe to get all their self-reported Parkinson's patients or data.

And if you took all that, that GWAS or some stats and you divided it per cell types, we found the things that we were expecting which is serotonergic neurons, dopaminergic neurons, etc. But we found a very strong signal for oligodendrocytes.

And this was a surprise to us. It turns out if you go back to really old data, there is some data on oligodendrocytes being affected in Parkinson. But what we did is that we went then to bulk sequencing samples that have been published. So when you have Parkinson, you can divide Parkinson into different Brock stages. And if you have a very early Braak stages, this is before you see a dopamine loss, dopamine neuron loss.

And it turns out if you look at the bulk sequencing samples that we— this is— what we can see is like the genes of oligodendrocyte lineages are actually upregulated prior to any effect on the dopaminergic neurons.

And this is, you know, this is in—

seen in all the— we looked at different studies and they all show, except that one, showed this upregulation of oligodendrocyte genes.

And it seems to precede the effects on the dopaminergic neurons. So this remains to be further elucidated, but I think it serves as a good example that going from genetics over cell types seems to actually be able to inform us and point us in new directions when it comes to complex disorders.

Okay, come to the conclusion slide.

So, you know, single-cell sequencing is hailed as us being able to find how many different types there are of neurons, but I would say that having quantitative measures of the diversity opens up for new measurements, new frontiers in science, and a possibility to merge human genetics with neuroscience. And the second one, this is very important, don't let your guard down when you read these papers. Don't trust them. Make your own analysis because it's early days for single-cell sequencing yet. We're still trying to figure out which is the best way to mathematically or statistically handle this data, and people are making too large claims constantly in our literature.

So be wary when you read the studies. And then, okay, to be then say, To try to be transparent, there are some really important caveats to the method that I just presented to you. One of them is, of course, the cell type coverage.

We have shown you from mice that are adult or late adolescence, maybe. But, of course, many of these disorders could be neurodevelopmental disorders. So we're lacking data on developmental stages. It could be that the genes are actually— the most important genes are expressed by different cell types during development. And then I think the other major caveat is that we have a very bad annotation of the non-coding regions. The SNPs for schizophrenia, for instance, are all in gene regulatory regions. Very few of them are actually in the exons. And understanding— having a better method to call which SNP actually affects the gene would make this method much more reliable.

Okay, but with those caveats, We see 4 larger classes mainly associated with schizophrenia, and this replicates in multiple datasets.

I think it's a new starting point for schizophrenia research. I'm actually banking on it because that's what my lab does now.

And the schizophrenia-associated pathways that have previously been described are specifically enriched in the same cell types.

And interestingly, this cell type enrichment for schizophrenia is largely due to non-overlapping genes. And then I showed you in the last couple of slides that oligodendrocytes seems to be implicated in risk for Parkinson's. We could confirm this in clinical samples. And one thing that we're doing, working with now is the possibility that different patients have different cell types affected, and we're seeing whether integrating this kind of information could affect the genetic risk score work, but that's early work, early days too. And then these are my acknowledgment slides. If you are a human geneticist and you're interested in merging it with neuroscience and you're about to go on your postdoc, please call me because I'm interested in getting your input onto our projects. With that, I'm happy to take questions.