Scientist Stories: Nadav Ahituv, Characterization and Targeting of Gene Regulatory Elements

My lab is interested in characterizing regulatory elements and how mutations in them can lead to various phenotypes. In particular, we're interested in enhancers that are sort of the promoters of the promoter. In the example that you see here, you see a mouse embryo and that's expressing the gene wherever it's dark blue, so in the limb, in the neural tube, in the brain. And in addition to the promoter, this gene would likely have enhancers that tell it to turn on in the limb, in the neural tube, and in the brain, and combined with the promoter give this complex expression pattern. So the enhancers could be dubbed as the promoters of the promoter. And these enhancers could lead to differences in gene expression, expression could lead to phenotypes like human variation, which I'll talk about in length here, differences between species, all of which our lab has interest in. For this meeting today, I'll talk about how mutations in these elements could lead to human disease. In the midst of revolutionary times where the number of genomes that are being sequenced is going up like crazy and the cost of sequencing is drastically going down, allowing us to sequence millions of genomes. Sequence a human genome, only 2% of it will be coding for protein exons, while the 98% is actually non-coding sequence. So if we look at the sequences here depicting the genome, you see that only the two nucleotides here are actually exons, and everything else is non-coding, suggesting that we have this vast genomic space to characterize these regulatory elements. Whenever we sequence a human genome, we get on average 3.5 million variants per genome, And that small little red line here is the amount of variation we get in exons, 0.65%, suggesting that most variants that we find are actually in non-coding space and could affect gene regulation. We have a code, and so we could easily figure out the consequences of nucleotide changes to that code. For example, here, if you change CCC to CTC, proline will change to a leucine, and that could lead to some sort of phenotype. Regulatory elements don't have this code, and that's a big problem for us in the field because if I find a mutation, for example, even in a known regulatory element, I currently don't have the sense of what it does in terms of the phenotype. So we really don't have a good genotype to phenotype correlation in regulatory elements. How problematic this is, I'll show you an example from probably the most characterized enhancer that leads to human disease, and that's the sonic hedgehog limb enhancer. So mutations in sonic hedgehog, the gene here on the left, lead to holoprosencephaly, cyclopia, kids are born with one eye, limb defects, a lot of bad things. And then 1 million base pairs away from sonic hedgehog in an intron of another gene is the sonic hedgehog limb enhancer. That is a non-coding sequence. Sequence. Enhancer in mice, you get mice with truncated limbs, as you can see here on the right, versus their wild-type littermates on the left. And again, this is a non-coding sequence, an enhancer, not a gene. Patients who have an extra finger, a condition called polydactyly, you can see here in both these families that whoever has a mutation, point mutation, in this enhancer has polydactyly. And now we have over 20 different mutations in this enhancer that lead to various limb phenotypes. So we can try to ask the question whether we can tell phenotype based on genotype, and the short answer is no, we really can't do a great job of that. And just to illustrate how problematic this is, I'll show you an example from a family that went into our lab that we characterized. For this enhancer. Family that had an extra finger, polydactyly, just like the previous mutations that I've shown you. And so the first thing that we did for it is sequence the sonic hedgehog limb enhancer to see if it has a mutation. Found that all affected individuals had a 13 base pair insertion in the enhancer compared to the unaffected individuals, which for me was a bit surprising as I was expecting a 13 base pair insertion to lead to a much more severe phenotype than a point mutation. Yet what we got was a point— a similar phenotype to the point mutation. Characterize how this 13 base pair insertion leads to this phenotype. And so we could do, for example, a bioinformatics screen to look what changed in terms of the transcription factors that are binding to the sequence. And we see that a lot of transcription factors now can bind to the sequence that before couldn't. But this really doesn't tell me why I'm getting this phenotype. A mouse enhancer assay, so we could take the wild-type sequence of this enhancer, stick it in front of a minimal promoter and reporter gene, and if this sequence is an enhancer, it will turn on that minimal promoter and turn on the reporter gene, and we can see where it's expressed. Same sequence, but now with the 13 base pair insertion, and compare the results to one another to to see if there's a difference in enhancer activity. You get here on the top right for the wild-type sequence exactly where Sonic hedgehog is expressed. Once we see that the enhancer is expressed where Sonic is expressed, but also on the top portion of the limb, suggesting that that might be the cause that leads to that extra finger. That's great. However, this assay is still limited in the fact that it's not a quantitative assay. Each embryo gets a different number of integrations of the transgene, and so you can't really measure a quantity of the enhancer, how strong it is. You can only measure differences in timing or the location. And in addition, as I'll show you in the next slide, it's also still limiting in various ways. Mutations in this enhancer, for example here in position 402 or 404, that lead to a much more severe phenotype in this enhancer, not just an extra finger but the radius and ulnar affected, The legs could be affected. And so we took one of these mutations, 402, and tested it for enhancer activity. Mutation in the minimal promoter and reporter gene and stick it into mice, this is what we get here on the right, where you can see ectopic expression of sonic hedgehog is what we expect that we're seeing. But this really doesn't tell me why this mutation is much more severe than the previous mutation. Compare it to this mutation and look at the expression pattern, I really can't get a sense why the other one is more severe than the other, again suggesting that this assay is quite limited. With the fact that we have tons of genomic data, a limited understanding of the regulatory code, and a limited functional assay that is not quantitative in nature, and also cannot test the hundreds of thousands of sequences that we are finding to be candidate regulatory elements or mutations within them. What I'll show you today is a collaboration of over 8 years with the lab of Jason Duer in the University of Washington and is also in collaboration with Greg Cooper in HudsonAlpha, Daniela Witten in UW, and spearheaded by Fumitaka Inawa, an extremely talented postdoc in my lab, and Martin Kircher, who used to be in Jay's lab and now has his own lab in Berlin. The classic enhancer assay vector, you have an enhancer, a minimal promoter, and a reporter gene. And so we set out to make this more high-throughput. The trick is to stick in a barcode of about 15 base pairs right after the reporter gene. And so if the sequence is an enhancer, it will turn on the minimal promoter. Turn on the reporter gene, but also provide a 15-base pair barcode RNA readout that could be measurable. What that allows you is now you can clone hundreds of thousands of sequences on the left and hundreds of thousands of different barcodes to match them on the right. And with paired-end reads, where you're sequencing the same sequence from the right and from the left, you can assign which barcode goes with which candidate sequence. Take a library of hundreds of thousands of sequences and stick them into cells or into mice. Sequence the barcodes so the DNA will let you know how much of your barcodes, which barcodes went into the cells or the mice. And the RNA of the barcodes will let you know if that sequence is an enhancer. Did it turn on the mammal promoter reporter gene and lead to RNA of that barcode. Then plot your enhancer activity as RNA over DNA. For example, for this red sequence, we saw DNA, meaning it went into the cell, but we also see a lot of RNA, suggesting it's a strong enhancer, versus the yellow sequence where we just saw DNA, meaning it went into the cell, but we don't see RNA, meaning it likely didn't turn on the minimal promoter and reporter gene, suggesting it's not an enhanced We use this assay in various ways, for example, synthesizing hundreds of thousands of sequences to see if there are enhancers to learn about the regulatory code, or just testing hundreds of thousands of ChIP-seq datasets, sequences, or others to find enhancers. What I thought I'd show you today is how we can use this to try to pinpoint how mutations lead to phenotypes. In these enhancers, in particular human disease. What we did was take known sequences, sequences that we know are promoters or enhancers, and see what effects mutations can have on them to learn how those could affect their activity. So for example, we could take a known enhancer and do all possible mutations in a certain position. So if it has an A in a certain position, we could change it to a C, to a T, or a G. And look at the effects of those mutations on its activity using this assay. 10 promoters that are known to cause human disease and 10 enhancers that are known to cause human disease, and you can see the various diseases here on the right. With the idea and the advantage here is that there's known mutations that could serve as positive controls where we expect them to affect the promoter or enhancer activity. For example, I'll show you the telomerase promoter where mutations in this promoter are known to lead to various types of cancer. So mutations in the telomerase promoter, the TERT promoter, are extremely common in some types of cancers. In particular, mutation -146 or -124, either of these mutations could be found, for example, in 83% of cases in glioblastoma 71% of melanoma, and so forth. So these mutations are extremely common, and they're also activating mutations. They make the promoter more active and lead to telomerase being expressed and cancer progression. Telomerase promoter and did saturation mutagenesis on it. And here below you can see the results of that assay using the massively parallel reporter assay. And so 0 is plotted as the wild-type activity here, and you could see that some mutations reduce telomerase activity, likely, and some increase telomerase activity. Where mutations -146 and -124 are that are activating, you clearly see that these are the top activating mutations for this promoter. And in addition, you see a lot of other activating mutations, and some patients actually don't have these 2 but may have these other ones. And so by having all possible mutations for this promoter, the next time you find a novel mutation in this patient, you can go to the SIRT cheat sheet and see where these mutations are and what they cause in terms of the phenotype. Of interest also are the mutations that reduce promoter activity. These, for example, could have some transcription factors that bind to that region, and if we could turn if we target those specific transcription factors, we may be able to use these as a treatment for the patients who have the increased expression of this promoter. I've shown you this massively parallel reporter assay that allows us to test thousands of sequences and mutations in them for enhancer or promoter activity. I'll just end this part by mentioning that we have also developed this about 4 years ago to work with lentivirus.. And the advantage of lentivirus is that it integrates into the genome, and so it provides a readout that was within the genome versus an epizomal readout. And in addition, it could infect hard-to-transfect cells like neurons or others, providing us a wider range of cell types where we can assay using this technology. So now I'll switch to the second part of the talk of how we can use enhancers as therapeutic targets. This work was spearheaded by an extremely talented postdoc in the lab, Navneet Matharu, who has now transitioned to be an assistant professor at UCSF. Navneet set out to try to tackle haploinsufficient diseases. So we usually have two copies of a gene, one from our mom, one from our dad, and that leads to normal protein levels. Every once in a while, we could have a loss-of-function mutation in one copy, which provides only one functional copy, and that provides 50% of the gene and subsequent protein being expressed. This could be perfectly fine for most genes, but some genes when you have half the amount, that actually can lead to human disease. It is actually estimated that over 660 genes lead to human disease due to haploinsufficiency. And when people look at whole genome datasets like NOMAD or ExAC, whole exomes, they estimate that the number could be much higher where they, for over 3,000 genes, do not see a heterozygous loss of function mutation, suggesting that they are haploinsufficient. Of course, some of these could be haploinsufficient lethal, and that's why we don't see them. But some could also be leading to disease that is observable. Current treatments for haploinsufficient disease are standard drugs, The problem there is that it's timely, costly, and these are primarily to treat the symptoms but not actually cure the disease. We could use gene therapy where we insert extra copies of the gene to try to cure the disease. The problems there could be dosage, tissue specificity, or gene length, which I'll talk about a bit later on. CRISPR could be another approach to fix these mutations. However, it's currently limited by low HDR frequency, and in addition, for every mutation, we'd have to custom tailor the approach to fix it. For example, the donor plasmid or the guide if it's base editing, and so forth, and that could be problematic. So about 7 years ago, beautiful work from Jonathan Weissman's lab, Stanley Qi, Wendell Lin and Luke Gilbert at UCSF and others, took Cas9, basically CRISPR, and mutated the Cas9 so it wouldn't be able to cut. So the scissors are not working, but you can use Cas9 to bind to specific regions in the DNA and connect other things to it. So you could basically use it like a delivery truck to bring things to specific locations in the DNA. For example, you can connect a transcriptional repressor to it, and that would reduce the expression of the gene, and that's called CRISPRi, for CRISPR interference. Or you can connect a transcriptional activator to it to increase the expression of the gene, and that's been called CRISPRa, for CRISPR activation. And the minute we heard about this, we thought, hey, this could be a great therapeutic for haploinsufficient diseases. So going back to this model, we still have one existing copy that's perfectly fine. It's just giving 50% of the amount of mRNA and then subsequently protein. So what if we could boost that expression up? So the idea is to target the existing copy with CRISPR-A and force it to generate more RNA and then subsequently lead to more protein from that existing normal copy. This could lead to increased expression of the gene, of the RNA, and subsequent protein, and might be able to cure the disease. We thought hard about what disease to go after. We wanted something quantitative, something that we can measure, that even if we just increase a little bit the expression of the existing copy, we might be able to see some sort of phenotype. And so we came up with obesity in this case. We chose a gene called SIM1 for our initial studies. It's a transcription factor that's involved in the leptin pathway. So leptin gets secreted from adipose tissues, binds to the leptin receptor in the hypothalamus, and through a signaling cascade basically signals us that we're not hungry. And mutations in many parts of the leptin pathway—leptin, leptin receptor, POMC, MC4R, SIM1—are known to lead to severe obesity in humans. SIM1 is actually the second most common mutation in humans that leads to obesity. So if you seek sequence the top 1 percentile of the obese population, about 2 to 3% of them will have SIM1 mutations, and the majority of these mutations are loss of function that lead to haploinsufficiency. So to summarize this, if you have 2 existing copies of SIM1 at the top, you're fine. If you have 1 existing copy, you're haploinsufficient and you are obese. And if you have two mutated copies that are nonfunctional, it's lethal, and there's actually nobody walking around with SIM1 mutations. Same holds true for mice. If you have mice with two copies of SIM1, they're fine. Mice with one copy of SIM1 are extremely obese, and no SIM1 is lethal also in mice. So going back here, the idea would be that we target the SIM1 promoter And in this case, we used a dead Cas9 that doesn't cut the DNA, bound to a weak activator called VP64. The reason for using that is: one, it isn't as strong, so we don't want to get too high of expression levels. And two, it's small enough that we could later fit it into an AAV, as I'll show you. And if that works, that will lead to increased SYM1 expression from the existing copy. This would lead for more protein and hopefully rescue the obesity phenotype in these mice. In addition, we also targeted a hypothalamus enhancer of Cym1 that resides 270 kb away from this gene, with the same idea that if we increase the expression of Cym1 by targeting its enhancer now, not promoter, we might also be able to increase protein expression and rescue the obesity phenotype. So we got hold of knockout mice for SIM1, and the heterozygous mice of course become extremely obese after 16 weeks. We then generated mice that have the dead Cas9 and the VP64 transcriptional activator, and mate them to the Sim1 heterozygous mice, and those of course become extremely obese because there's no guide to tell the Cas9 where to bind in the genome. We then made mice on another locus that either have a promoter guide for Sim1 or an enhancer guide for Sim1 and mate all three lines together, in the hope that having all three now will reduce the body weight of these mice. These are the results for the promoter CRISPRa-targeted mice. You can see that in orange, the heterozygous mice become extremely obese after 16 weeks compared to the wild type, which are in light pink. And you can see that in yellow are the CRISPRa-rescued mice, showing a body weight that is very similar to the wild-type mice, suggesting that this approach could rescue their body weight phenotype. Below now is the results for the enhancer CRISPRa mice, and you can see again in orange the heterozygous mice that become extremely obese, in pink the wild-type mice, and in green the enhancer-rescued mice showing a body weight very similar to the wild-type mice, suggesting that by targeting the enhancer of this gene in the hypothalamus, we could also rescue the obesity phenotype in these mice. It's always nice to show a picture, so here you can see these mice. On the left is a heterozygous mouse for SYM1 that is, as you can see, is extremely obese. And in the middle is a mouse with the same genotype that should be as obese as the mouse on the left, but because of the CRISPRa approach, you can see that it looks, in terms of its body weight, very similar to the wild-type mice on the right, suggesting again that we could rescue their obesity phenotype. Another interesting finding here was tissue specificity. So Cym1 is expressed in the hypothalamus, in the kidney, but not in the lung and the liver. Our dCas9 VP64 mice have a CAG promoter, which is a ubiquitous promoter, meaning that Cas9 is expressed in most tissues. And you can see here in the hypothalamus, kidney, lung, and liver on the right that Cas9 is expressed in all these tissues, and the same for the guide. So technically, we could be upregulating Cym1 in all tissues. We looked at SYM1 expression in these tissues, and here I'm showing you below the expression in wild type of SYM1 that is plotted as 1 for the RNA of SYM1. And you can see that SYM1 is expressed in the hypothalamus and the kidney, but below detectable levels, BDL, in the lung and the liver. And if we look at the heterozygous mice, we can see that we get a about half of that, as we would expect. Now what do we get for the promoter or the enhancer rescued mice? So for the promoter rescued mice, you can see that we're upregulating in the hypothalamus, in the kidney, but not in the lung, in the liver, even though Cas9 and the guides are expressed in all these tissues, suggesting that the upregulation is specific to the tissue where this regulatory element is active. For the hypothalamus CRISPRa results, you can see that we're only upregulating Cym1 in the hypothalamus but not in the kidney, and if you recall, this is a hypothalamus enhancer. Combined, these results suggest that the regulatory element that we target could provide tissue specificity, meaning if we target the hypothalamus enhancer in this case, we only get upregulation of SYM1 in the hypothalamus but not in the kidney. Or if we target the promoter that's active both in the hypothalamus and in the kidney, we only get upregulation there. And this, we think, is maybe due to open nucleosome regions through whole exome datasets where people use CRISPR-A or CRISPR-I that see that open nucleosome positioning is really important for guides that work versus ones that don't. And it could be that because, for example, the hypothalamus enhancer is only open in the hypothalamus, Cas9 can only bind there and lead to upregulation in this tissue. However, of note, this is really an n of 2, and more work needs to be done to really show why this is happening. So everything I've shown you is transgenic, and here we really wanted to go the distance and develop this as a therapeutic. We thus turned here to AAV, adeno-associated virus, that is widely used now in clinical trials for gene therapy, and the reason is that it can integrate into the nucleus but does not integrate into the genome and could lead to long-term expression of the transgene. However, AAV has a limitation that it can only package optimally 4.7 kb, 4,700 base pairs of DNA. And if you add in the ITRs and the regulatory elements that you need for AAV, it's about 3.5 the size of the transgene that it can package. And that could be limiting in some cases. For example, if we look at all the 660 genes that are known to lead to haploinsufficiency, or the 3,000 or so predicted ones, and look at their cDNA length, and we draw the line, the red dotted line that you can see here, as what could fit into an AAV, you can see that a lot of genes, their cDNA will not be able to fit into an AAV, and these are genes that could lead to haploinsufficient disease. Some of them quite severe, and so a classic gene therapy approach where we just package them into an AAV will not be helpful here. So Navneet worked hard to try to get both the dCas9 and the VP64 to fit into one AAV. In addition, she made another AAV that has the promoter guide with mCherry for SYM1, and another AAV that has the enhancer guide for SYM1, with mCherry, and the idea is we inject the dCas9 and VP64 AAV along with the promoter guide, or instead along with the enhancer guide, and try to see whether that might rescue the obesity phenotype in these mice. So we did stereotactic injections of these viruses into the hypothalamus, and here in red you can see the mCherry expression suggesting that we are injecting into the right place where Cym1 is expressed in the PVN of the hypothalamus. We injected into Cym1 heterozygous mice at 4 weeks of age, and then we waited 7 weeks to see how much weight they gained after that with the different injections. Here on the bottom left, you can see the expression of Cym1 And again, we plot wild type as 1, and all the other mice are heterozygous mice. So in orange are uninjected heterozygous mice, and you can see they're about half. In blue are heterozygous mice that were injected just with the activator without the guide, and you— as a negative control, and you can see that they're also about half. And in yellow and green are the promoter or the enhancer rescued mice, and you could see that we're getting Sim1 levels that are similar to wild type in the hypothalamus. What about body weight? So you can see that the wild type in this 7-week period gained about 5 grams of weight, heterozygous about 12, and in blue the negative controls with just the activators, similar amounts, versus the ones that were injected with the promoter guide or the enhancer guide that gained very similar weight, to the wild-type mice, suggesting that this approach can rescue their body weight phenotype. We also took these mice all the way to 9 months post-injection, as you can see here on the right, and measured their body weight. And you can see that after 9 months from that single injection at 4 weeks of age, we have wild-type mice weighing about 30 grams, heterozygous and the negative control in blue about 50 grams versus the CRISPRa rescued promoter or enhancer mice that weigh about 30 grams similar to their wild-type mice, suggesting that this approach of using AAV could be a long-term therapeutic approach to rescue the obesity phenotype in these mice. So in summary, using our transgenic approach, we showed that we can rescue by targeting the promoter or enhancer of SIM1 the obesity phenotype phenotype in these heterozygous knockout mice. What was good about the transgenic approach is that it also showed us that the tissue specificity could be determined by the regulatory element you target. For example, if we target a hypothalamus enhancer, we only saw upregulation in the hypothalamus but not in other tissues. A third point is that we have these mice and these vectors and happy to share for various uses. And finally, and most importantly for me, was that that AAV approach actually managed to rescue the obesity phenotype in these mice, suggesting that it could be used as a therapeutic for many other haploinsufficient diseases. To show that this approach could be used for other genes and not just SYM1, we went after another leptin-associated gene, and this is MC4R, which is actually the topmost mutated gene in humans that leads to obesity. If we go back to the 1 percentile population of BMI and sequence it, about 5% of them will have mutations in MC4R, and a lot of these mutations are loss of function that lead to haploinsufficiency. So we got hold of MC4R heterozygous mice that also become extremely obese. We made guides that target the MC4R promoter and inject them at 4 weeks of age and similar weight, 8 weeks this case, and see how much weight these mice gain. If we look at MC4R expression, and again we plot wild type as 1, we can see that heterozygous mice in the dark purple, have about half the amount of Mc4R expressed in the hypothalamus. Heterozygous mice that were injected just with an activator as a negative control have a similar amount of Mc4R versus ones that were injected with the activator and the promoter guide, the light purple on the right, which now have much higher expression of Mc4R in the hypothalamus. In terms of body weight, we can see that wild-type mice gain in those at that 8-week period, about 6 grams of weight. And heterozygous mice gain much more, same as the negative control injected with just the activator, versus the light purple mice that you can see both for females and males that gain similar weight to the wild type, suggesting that we managed to rescue their obesity phenotype. And again, this is a different gene, different locus, showing that this approach could be used for other genes and other diseases. And what we're doing now in the lab is using this approach to go after other haploinsufficient diseases, in particular very severe diseases that lead to various neurological abnormalities. So to sum up this part, I've shown you that CRISPRa could be used to treat haploinsufficient diseases by upregulating the existing normal copy of the gene. But in addition, this could be used for other diseases. For example, there are a lot of microdeletions in humans that lead to various diseases where it's not just one gene that is missing, but a few genes. And so by trying to upregulate a few genes in parallel using CRISPRa, we might be able to rescue subsets of that phenotype. In addition, we might be able to use this approach to upregulate an alternative gene. For example, if a disease is caused by mutations in both copies of a gene—gene A is messed up in both copies—there may be another gene, for example gene B, that has a similar function that is not expressed in a certain tissue or is it's expressed at lower levels in that tissue. And if we could upregulate its expression in that disease-associated tissue, we might be able to rescue the disease. And some diseases that come to mind are spinal muscular atrophy, where SMN1, both copies are loss of function. But there's SMN2, for example, that is a similar gene that has a splicing acceptor mutation that to 10% of expression. So if we could upregulate that expression for SMN2, we might be able to rescue that disease. The same goes for utrophin, for dystrophin or fetal globin for beta globin and many other potential diseases where upregulating an alternative gene might be beneficial for treating those diseases. So I'll end with the most important slide acknowledging members of my, Amazing Lab for all the work that they've done on this project. In addition, our collaborators in UCSF, in particular Christian Weiss for the obesity work, and in UW, Jay Shendure for the massively parallel reporter assays, along with Martin Pircher in the Berlin Institute of Health. And also want to acknowledge all my funding, and thank you very much for listening.