Scientist Stories: Luca Pinello and Karen Mohlke, Playing genomic battleship with CRISPR

Welcome, everyone. We have registrants with us from 25 countries today, and we're so glad to have you here for the first in our new seminar series called Genomic Innovators. So I'm Dr. Chris Gunter, and I serve as NHGRI's Senior Advisor to the Director for Genomics Engagement. And I'm very pleased to host today with my longtime friend, Dr. Lisa Chadwick.

Hi, I'm Dr. Chadwick. I'm a Program Director in NHGRI's Division of Genome Sciences, and I'm the lead of several of our programs that try target early-stage investigators, which includes the Genomic Innovator Award.

And we also want to strongly thank the team at NHGRI who helped us organize this series, particularly Susan Vasquez and Gerald Samani, who are both wonderful and have helped us do a number of these. So thank you to them. So on this first slide here, you can see, as I mentioned, today is the first one in this new series. We will be hosting them quarterly. That's our plan. So the next one will be January-ish. But you can find out more details for the series that are listed on the web page here on the bottom if you're interested in the future, and we'll make sure to let you know about them as well. They'll be posted on our YouTube channel after, so if you enjoy today's talk, feel free to share it later with your friends who could not make it because we all know there are competing meetings. We named this series both after the extramural funding program we had in the past, which Lisa was responsible for, called the Genomic Innovator Awards, and after the fact that we view both of these speakers as true innovators in the field of genomics. So on the next slide, you can see more about our speakers, and I'm going to introduce both of them to you. So Dr. Luca Pinello is an associate professor at Massachusetts General Hospital and Harvard Medical School. He earned his PhD in computer science and mathematics from the University of Palermo in Italy. His research program uses computational approaches, which we'll tell you about today, to systematically analyze the sources of variation that affect gene regulation, including epigenetic variation, genetic variation, and single-cell gene expression variability. And our second speaker, Dr. Karen Mulkey, is Professor of Genetics and Associate Chair for Research at the University of North Carolina School of Medicine Chapel Hill. She earned her doctoral degree in human genetics at the University of Michigan. Her research program is identifying genetic variants that influence common human traits with complex inheritance patterns and also seeks to understand the biological function of identified variants and interactions between genes and environmental risk factors in disease pathogenesis. So while we're talking, we encourage you to put your questions for the speakers into the Q&A section, which you should be able to see at the bottom of your screen. And then Dr. Chadwick will ask those questions of the speakers during the question period when the talks are finished. And with that, I'll turn it over to Dr. Pinello.

Hi everyone, it's a great pleasure to be here. I'm really excited to kick off this seminar series. So today we're going to play genomic battleship with CRISPR technologies to uncover non-coding functional elements and their phenotypic effects. So can you please enable the control of the slide? No, it's not, it's not working. Should I share my screen instead? Let me try. Okay, now it's working. So yes, I want to, you know, first start with the disclosure. My slide of disclosure. And I want to now briefly outline what we're going to talk about today. So I want to briefly mention my transition from computer science to computational biology and how NHGRI really enabled this. And then I want to talk about the DNA puzzle and where we can focus in the genome to understand function and then propose a framework to essentially have a multi-scale exploration of the non-coding genome using CRISPR tiny screens. And then the second segment that is related to what I will present, Karen instead will highlight disease variants in non-coding functional elements, CRISPR technologies for variant element function, and also an outlook on future opportunities. So my lab is trying to combine computational experimental strategies to understand gene regulation. And we're working with different technologies and assays including epigenomics, CRISPR genome editing, and single-cell omics. And we are trying to answer two specific questions. The first one is, how can we uncover and dissect functional and coding elements in the genome? And the second one, how can we model cell fate choices? And today we'll focus on the first one. And this is just a brief slide about my career trajectory. I'm from this little town in Italy called Altavilla. And I did my PhD in computer science in Palermo. And as you can see from this slide, my first transition to computational biology was in 2011 during my postdoc. And throughout the years, as you can see, NHGRI has been really important providing funding to make me to the next stage. So as a postdoc, I received a K99, and then I was promoted to assistant professor. And then in 2019, I received this Genomic Innovator Award that gave me so much freedom to explore my research direction. 2020, I was promoted to associate professor. In 2021, I joined the NHGRI iGVF Consortium leading a characterization center. So just wanted to highlight this because there are a lot of funding sources that may enable other people to make the same transition. And this really helped me to create, you know, have the freedom to create over the years several computational tools and to do team science and to enable other researchers to do cool analysis related to genomics, CRISPR genome editing, and also nice and crazy visualization in virtual reality for single cell genomics. So going back to the question, I was really passionate about the genome because there are so many puzzles yet to solve. And as you know, the first draft The first good draft was completed in 2001, but it was not until this year that we have the full complete sequence of the genome. There are several parts in the phylite, a random region of the genome, and they ask you, what is the function of this small segment? If you have a coding sequence, maybe easier to spell out the function, but if you are in an encoding sequence, maybe it's not so easy. The non-coding part is really important because 98% of our genome is non-coding. Still, 80%, depending on which assay you use, has some biochemical activity. And most importantly, many of the disease-associated variants are in the non-coding space. And Karen will expand on this point more. So what really made me passionate as a computer scientist about studying genome function was essentially this puzzle. How come almost all the cells all the cells of our body share almost the same code, but they have different output and phenotype and gene expression programs. To me, it was really fascinating. Over the years, I learned that one of the explanations to this puzzle is the chromatin structure. In fact, the way the DNA is folded is really different in different cell types, and different mechanisms essentially help to specialize this code. And these are, for example, nucleosome positioning, DNA methylation, and histone modification that you can think as a way of enabling or disabling part of this genetic code and to create these gene expression programs that are cell type specific. So this can also help us to think on where we want to focus in exploring this genome. Because the genome is quite large, we have 3 billion nucleotide pairs, So we need to think about where we want to focus in this exploratory game. And what do we need to be smart? So one first thing is we need some mechanism to highlight regions that are important. Second, if we really want to understand the function, we need to perturb this region. And for this, we have gene editing technologies like CRISPR.. And this can lead us to understand progressively, and we can iterate this framework. So in order to, the first step, highlight, there are many ways to highlight based on genetic variants, conservation. Here, what I'm proposing, or many people are proposing, is to use essentially chromatin accessibility of particular histone modifications, because these can highlight regions that are active and cell type specific. In fact, in over the years, we have learned that, for example, different histone modifications are associated with different biological functions. For example, if you have a gene, you can immediately find promoters that may active or poised or repressed, and genes that are transcribed. But one important mark is multiple marks that essentially allow you to highlight this distal region that are called enhancers. These are really important region in gene regulation because they are cell type specific and that to fine-tune the gene expression in different cell types. This is the way we can potentially highlight. I want to highlight that we have so much data thanks to roadmap epigenomics and ENCODE. That profile histone marks, chromatin accessibility, and also binding of different proteins for many, many cell types, also primary cells. So in terms of perturbation, there are different genetic technologies. One that really changed the way we can do large-scale screens is CRISPR genome editing that is composed by a Cas9 protein that there's some cleavage activity, and then, you know, guide RNA that essentially can be programmed. So you can synthesize this region of around 20 base pairs, and by simple complementarity rules, you can essentially target this system almost anywhere in the genome. And this system will introduce double-strand break that the cell will try to repair. And, you know, in this process, you will introduce deletion or short deletion or insertion, or you can also trigger other mechanisms, provide a small piece of DNA to actually have precise replacement. So this is a really powerful technique. You can design easily thousands of perturbations if you wish. And then, you know, over the years people have modified the Cas9. They disabled, you know, the cleavage activity, and there's this dead Cas9, and now you can fuse other factor or epigenetic modifier. And there are like these different versions of this, but there are like names for the different versions. A global name is like CRISPRi that essentially allows you to repress epigenetically a region in the genome, or CRISPRa that allows you to activate. And for example, you can repress genes targeting other promoters or creating new enhancers in non-coding regions. So really exciting technology. In terms of understanding, I want to tell you a brief story. This really shaped the way I think about this, also my career. During my postdoc in Guoqing Yuan lab, I had the fortune to collaborate with Stuart Turk and Daniel Bauer. They were really interested in studying blood diseases and sickle cell. In 2013, they uncovered one enhancer that was enriched for genetic variants associated with fetal hemoglobin. At that time, we essentially decided to actually delete the enhancer using gene editing. We saw that this enhancer was for a gene called BC11A, indeed was controlling BC11A. BC11A being a repressor for fetal hemoglobin, the perturbation of this enhancer element was able to reactivate in a cell type-specific fashion fetal hemoglobin. And this is really important because, you know, as you know, we have different hemoglobin. We have the fetal one that's only at the fetal stage, and then we have the adult one. So if you find, you know, a genetic switch or, you know, a perturbation that re-enabled the fetal hemoglobin, you can ameliorate, you know, for example, sickle cell disease, providing a backup, you know, option. In 2015, we used again CRISPR technologies. This time, you know, instead of deleting the full enhancer, we did the saturation mutagenesis screen. We uncovered that actually you don't need the entire enhancer to activate BCL11A, but we uncover a small critical element that was a binding site for a combo transcription factor GATA-1 and TAL-1. You need a single perturbation around 10 base pairs to essentially reactivate this fetal hemoglobin. In 2019, we showed that this potentially can be a potential therapeutic avenue. You can envision that you can modify human hepatobatic stem cell ex vivo with this CRISPR technology with a single perturbation, and then reinfuse these cells, and this can potentially ameliorate this sickle cell. This is actually a reality. In fact, in 2000— I cannot advance. Oops. Sorry, the advancement is not working. Yeah, this is, I was saying, is really, it's truly translational. In fact, in 2020, the first patient was treated with this idea. And there are like several clinical trials, one led by Daniel Bauer, the Boston Children's Hospital. And today, like this year, there are already 20 patients that have like a really positive response using this. And I was really lucky to be part of this project developing all the computational tools that enable some of this exploratory analysis. So now expanding on that, these are really powerful frameworks that we can apply to other regions of the genome. And there is this concept of CRISPR tiling screen that, as my title mentioned, is kind of playing battleship against the genome. So imagine you have a board and there are a few things that you want to discover. And now you have this high-throughput way to throw perturbation on this board. And you have a way actually to see which perturbation is associated to a phenotype of interest. And you can potentially use target enhancer variants and so on. So in my Genomic Innovator Award, I was essentially proposing, well, you can do this in multiple ways, but how can we optimally do this. So the idea is we can combine different CRISPR genome editing technologies to efficiently first uncover and then dissect these regulatory elements. And what I mean, imagine you have like in a locus a gene of interest associated to a phenotype. You can design different perturbations and you can start with low resolution. So you can use CRISPRi and CRISPR aided with 1 kb to 100 base pair resolution to uncover the first large elements. Then you can use genetic perturbation like Cas9 to tile this region precisely and uncover the sub-element that are important. You can envision even to use base editing or prime editing to go up to 1 base pair resolution. This is something we're developing. We're really excited. But there are already CRISPR tiling screens that our group have performed. For example, this is one for this piece that I was mentioning. Also other groups are developing these screens. There is a clear need to develop the technology to enable other people to do this at high scale and for different phenotypes. In this slide, you see 3 papers using 3 different technologies, Cas9, CRISPR, and CRISPRi. For different goals, you know, enhance the detection as we did, or enhance discovery as, you know, these two other papers propose. In terms of the analysis, there are different ways to do this, and I will propose one way that essentially unifies all the different strategies that we have so far. So just to provide a little bit more detail on the CRISPR TALEN screen, the idea is we start from a locus of interest, We design, that means that we take all the small sequences that we can use to deploy our perturbation to target our perturbation. Then we create our library of perturbation and we can lentivirally transduce our cells. We can introduce the perturbation in the cells, these circle are the cells. We want to have one perturbation per cell. Then you can sort based on a phenotype of interest. For example, can be fetal hemoglobin level, can be the expression of a gene of interest, can be viability, a cell surface marker. This depends on the phenotype that you want to study. Now that you have these two populations, you can try to recover the guide enrichment in this high and low population. Just to provide a little bit more detail, what you do essentially, you sequence. In each cell, you have a lentiviral integrant that is integrated in the genome. The part around the guide RNA, this perturbation is the same. You can just design primer, amplify, and you can create this nice table and calculate based on the log fold change between these two numbers. You can get a 1D track that is the enrichment score. Then based on this, you can essentially understand which part is important associated to your phenotype. And this seems all simple. So high level, we start from reads of ASCII file that what we want is something like this, this denoised signal that tells us this association and maybe some statistical test that tells you like, okay, this region is the one that you should follow up. And there are a lot of steps that you need to do right in between. So in order to do that, we created a tool called CRISPR-SERF that is based on the idea of the convolution. This was led by Jonathan Xu, who was a PhD student in my lab and Kit Zhang lab. The idea of the convolution, we want to uncover the true underlying regulatory signal in the genome, but we have an imperfect measurement system that is our CRISPR perturbation. Then this essentially introduce some distortion to our scene. With the convolution, essentially we want to go back to this signal. This depends, of course, on the perturbation that you use. If you use genetic perturbation, you have resolution around 10 base pairs, and these are the small indels. If you use epigenetic perturbation, you have much larger perturbations, so you have a lower resolution of around 200 base pairs. In the measurement problem, What we do is you can envision that this CRISPR perturbation is like you essentially you can measure this signal here. If you have the perfect tool, you will observe essentially the perfect reconstruction. But with this Cas9 and CRISPRa, you get a smoothed version and this operation is called convolution. Now you can frame this as a convolutional model. You have what you want to recover, You have your perturbation with CRISPR. Of course, we have noise. And this is what we observe, the noisy signal. We want to reconstruct the original signal. And for this, the good news is there are several frameworks. For the convolution here, we propose to use generalized LASSO that essentially is a way to go from something like this to something that parsimoniously tries to explain what we observe in the signal. And you can learn more in our paper. So, we tested this in multiple CRISPR screens. And, you know, we showed that this is a generalisable approach, no matter which perturbation type you use, you can recover the same validated elements and also new candidates. We also did the dual screen targeting, you know, the Bcl-11A locus and several chromatin accessible sites with CRISPRi and Cas9. And one thing that is nice when you have this dual data on the, on exactly the same region, you can clearly see the value of this multi-scale resolution. For CRISPRi, you recover these 3 enhancers, but with Cas9, you recover the critical element. You have a much finer resolution. And also, if you use genetic perturbation, you can assess the effect of coding region, whereas with epigenetic perturbation, you cannot do that. And then finally, I just want to say that our lab is really passionate about building tools that are usable and user-friendly. So we built this website that can assist you in the design of the experiment and as well in the deconvolution and then the follow-up annotation and analysis. So this is online, the code is open source, and if you want to try, please let me know and we're happy to provide support. The last slide, there are exciting things I did. I just show you some tools based on CRISPRi, CRISPRa, and Cas9. But now we are using this new editor, base editor and prime editing that allows you to have really high resolution to potentially change up to a single nucleotide. Also, we are exploring a new framework for CRISPR tolerance screen. Where we envision that instead of reading the guide RNA enrichment, we can read directly the alleles and we can potentially develop tools to leverage this allele frequency in the different population. With this, I would like to acknowledge current collaborators, past collaborators, and people in my lab, and especially also my funding source in the NIGRI that really helped me to make this transition, supported my career in multiple stages. And for people that are excited about CRISPR technologies, there is a free conference next month, and we have stellar speakers. So if you want to learn more, please join us. We'll be online, and you can also send an abstract if you want. And I will stop here. I will hand off to Karen. Thank you.

Okay, uh, I will now add some context to the use of CRISPR technologies to identify mechanisms responsible for common disease that include variants in non-coding regulatory elements, and then some perspectives on the future. So many common variants associated with common diseases and their related traits have been identified by genome-wide association studies or GWAS analysis. An example shown here is a recent results of a genome-wide association study screen for variants associated with LDL cholesterol levels. However, most of the variants that may be responsible for these association signals are located in non-coding regions. Thus, a key goal to understand the heritable contributions to disease risk is to identify the genes responsible for the associations. And identifying the specific variants and their molecular mechanisms rigorously links disease risk to gene function. So for non-coding variants, a frequently applied strategy is to examine the non-coding variants located in regulatory elements that correspond to disease-relevant cells or tissue types. So I'm gonna show two examples where experimental manipulations with CRISPR-based technologies provided needed validation of the hypothesized links between variants and genes. And these hypotheses are based on genetic strategies to evaluate how variants exhibit regulatory effects. So first we hypothesized that variants that are associated with gene expression levels and also with a trait or disease may function through that effect on that gene to influence the disease. And then to further understand how the variants may be affecting gene expression levels, examine the chromatin context of those variants. So variants associated with chromatin characteristics such as accessibility that are also associated with gene expression and a trait or disease may be acting through that effect on the regulatory element to influence the gene and the trait. So here's a first example. Starting with a genome-wide association study for glucose levels zoomed into this one region. This was performed in about 58,000 individuals. The same set of variants are also associated with islet RNA-seq-based expression levels of the gene in this region, ADCY5. The variants associated with lower gene expression level were associated with higher glucose levels and higher risk of diabetes. Now, the function of this gene and its expression in pancreatic islets suggested that the variant may act to alter insulin secretion. So when examining the set of variants that could be responsible for this association signal, We compared the locations of those variants to regulatory elements in especially pancreatic islets. And you can see this variant located here is located within a region of accessible chromatin based on two technologies, FAIRE and DNase hypersensitivity, and in a region that is an enhancer-based chromatin state in islets and really limited to not many other cell types. So we first set out to, as a first step to understand mechanism, tested that particular variant, the trait-associated variant in its regulatory element for effects on transcriptional activity. So shown here, we analyzed that variant in a reporter assay in the regulatory element And both the regulatory element with either allele shows enhancer activity. However, the A allele at the variant shows reduced enhancer activity. And that's consistent with the lower levels of ADCY5 expression in the eQTL analysis. Now while this experiment shows that the variant is capable of altering transcriptional activity, it does not validate the link from the variant to the gene or to gene function. And so for that manipulation, we turned to CRISPR. The regulatory element spanning this variant is conserved across species, allowing us to use a rat cell model that secretes insulin well. So in this experiment, we generated two double-stranded breaks, deleting the intronic regulatory element. From the genome. And we obtained clonal lines with homozygous deletions. And we then measured the effect of gene expression, the effect of these deletions on gene expression and on cell function. So here, I showed that when the regulatory element is deleted in the homozygous enhancer deletion clonal lines, the gene expression level of the ADCY5 gene is reduced compared to lines that were created that were not targeting— that did not generate deletions. A nearby gene, SEC22A, was not affected. Its gene expression was not affected, suggesting that that deletion was limited in its impact. What about function of those cells?

Does reducing—

does removing this regulatory element alter the cell capabilities? And here I show that those deletion enhancer— the deletions of the enhancer show reduced secretion of insulin compared to the mock-edited cells. So we can build a mechanism Looking at the variant, its effects on enhancer activity, that association with expression level in pancreatic islets from individuals, the association with glucose levels and risk of diabetes. And this evidence now with the CRISPR-based deletion of the regulatory element makes that connection between the variant and the regulatory element and the gene, as well as its function in disease. Now, this strategy is— this type of application of CRISPR technology is pretty low throughput. Targeting efficiency is low, requiring extensive screening of clonal lines and really setting priorities among variant-containing elements. So in this case, we were motivated by that evidence that the variant affects transcriptional activity, that the element was limited in the cell types in which it was an enhancer. It has characteristics of enhancers. That there were really few other candidate variants in elements that could be responsible for this GWAS signal based on those annotations. That this element was conserved across species and some other experimental and annotation support, some allelic imbalance in ChIP-seq data and evidence of specific binding of proteins. I'll now turn to an example of using CRISPR technologies to understand genetic effects on regulatory elements, gene expression, and disease. So here, we started by generating chromatin accessibility data in a pilot study of human liver samples from 20 individuals. We performed ATAC-seq to identify the accessible chromatin regions, identified peaks spanning the genome, and then tested variants that were located within 1 kilobase of the site of one of these peaks or regulatory elements to ask whether the alleles at that variant were associated with the level of accessibility. So in this study, we identified more than 3,000 regions across the genome where this was the case. We then compared these chromatin QTL signals to genome-wide association signals to identify potential functions. So an example of that chromatin association is shown here. These are profiles of ATAC-seq data from 9 different individuals shown across the width of the screen. And the peaks then that are called as regulatory elements are shown below. Now, you can see that most of these elements, most of these chromatin accessibility is similar across individuals. But in one region, the chromatin Accessibility differs between individuals. And the genotype of a variant located within that regulatory element is shown over here on the left. And you can see that individuals with— that are homozygous for the T allele have stronger evidence of accessibility than individuals homozygous for the C allele. So this is a chromatin QTL variant. These elements are more often found— the ones that are genetically regulated are a little bit more often found in enhancers and in promoters when compared to chromatin states from Liver Tissue in the Roadmap Epigenomics Consortium. The presence of some additional elements in otherwise termed quiescent regions suggests that perhaps the greater genetic diversity of the individuals profiled here helped identify additional regulatory elements in the liver. So this region that I showed that's a chromatin QTL is also a signal that is associated with LDL cholesterol levels and with expression level of the LITAF gene in liver samples. So the LDL data is from around 1 million individuals, that GWAS study. The eQTL data is from around 1,000 individuals. And here the same variant that was the lead most strongly associated with LDL cholesterol and with gene expression was also the most strongly associated with the accessibility of this peak here at the 3' end of the LITAF gene. This gene potentially plays a role in more than one cell type within the liver tissue that we examined. And the variant within the regulatory element is really— there are 3 variants that are found all close together located in the same regulatory element that are all proxies from one another. So again, as a first step, we tested these variants within the regulatory element to see if they could alter transcriptional activity. So here tested as a haplotype. And we examined 4 different cell types that could be playing a role in the liver tissue. In all 4 cell types, we observed the same results that both haplotypes, both that the regulatory element serves as an enhancer compared to empty vector. And that the alleles that were associated with greater chromatin accessibility showed greater transcriptional activity. These are the same set of alleles that are associated with higher liver expression of LITAF in the eQTL analysis and higher cholesterol levels in the GWAS data. So again, this, does not demonstrate— the data do not yet validate the role of the regulatory element on the gene. So here to do that test, we use CRISPR interference in a hepatocarcinoma cell line. Luca introduced this strategy earlier. And we used guide RNAs spanning the enhancer and compared them to non-targeting guide RNAs. And in this preliminary experiment, the knockdown leads to a significant reduction in expression level of the LITAF gene. So these data provide important validation of the hypothesized link to the gene. The CRISPRi system is much more amenable to higher throughput studies, as the guide RNAs can be designed to many regulatory elements, including promoters, and to measure the effect of altering gene regulation as well as gene expression. So what does the future hold for CRISPR technologies to identify non-coding functional elements and their phenotypic effects? Well, first, there will be improved CRISPR technologies, many technological advances, continued development of the biochemistry of editing, and the design of effective guide RNAs. Guide design that better considers the diversity of human variants will improve both the efficiency of targeting and reduce off-target effects, especially as more diverse genomes are edited. Newer technologies, as Luke introduced, of base editing and prime editing, hold promise for more precise nucleotide changes with higher efficiency. Improved design of the biochemical aspects of CRISPRi and its counterpart CRISPR activation can enable study of the consequences of a wider range of changes to gene expression levels. Screens can become more comprehensive and test a larger number of variants and elements in a wider set of cell types, cell states, and organisms. And improved signal cell Technologies will enable DNA and RNA to be profiled at scale, facilitating understanding of genetic effects in these specific cell types. And finally, we can anticipate more in vivo applications, including to treat and reduce the impact of human disease. The future will also hold a more comprehensive understanding of non-coding functional elements and their phenotypic effects. So non-coding regulatory elements differ by cell environment, by specific cell states, and their— the cell types and their various cell states that differ during development and in response to perturbations of the cell environment. And non-coding elements also differ by genotype, and we'll learn about these effects through the CRISPR saturation mutagenesis screens, as Luca mentioned, and the study of natural human genetic variation in more and more individuals. One community that is performing this type of research is the new IGVF Consortium. This consortium has multiple goals to understand the impact of genomic variation on function, and I've listed two of them here. And one is to apply technologies like the ones that we have described to systematically perturb the genome and figure out the effects of variation. Of variants and elements on genome function and phenotype, and also to perform high-resolution identification of where and when genes and regulatory elements function. Luca is a program director for one of the projects, as described here, and he and his team are using base editing CRISPR screens to identify and examine the function of variants responsible for blood and heart diseases. And in one aspect of this project, Luca's team is collaborating with a team at UNC to examine variant function results across technologies and strategies. Finally, I'd like to acknowledge the people who contributed to the projects I showed, including the people in my group, past and present, our collaborators at UNC and elsewhere.

Great. Thank you both so much for those great talks. We have some time now to answer questions from the audience. So remember, if you have a question for our speakers, go ahead and post it in the Q&A box and I will ask it. I'm going to start with a question and then I'll ask some of the other ones we've gotten in the chat while you both were talking. So Luca, you know, the work that you talked about uses tools and technologies that are still relatively new. How have those kind of changed the way that you both have been able to address the questions that your lab is interested in? And how accessible are these kinds of technologies to other researchers?

Yes, very good question. So I think, you know, like we have seen a revolution, you know, like for example, with CRISPR, you know, single cell technologies and Initially, they were not so accessible, but now so many labs can easily deploy this technology. One barrier sometimes is having the computational tools, but now there are several programs including this Genomic Innovator that help researchers like me that are excited about developing tools to help others. I think, you know, like, it's a nice, you know, framework, right? You know, there are new technology, and I think now, like, we're much faster in adopting new technologies. And, you know, with, you know, sharing, you know, it's like more open, you know, we share, you know, reagents. And so I, I think, you know, like, I see like a really positive future, you know. There will be new technologies, but we have, you know, already a system to share, you know, like this technology as well, you know, computational tools. So that's my short, you know, view on this. Great.

So one of the questions that we had in the chat was about the difference between discovering enhancers and dissecting enhancers. I think probably like, what kind of tools would you use? What kind of information are you looking for? Can you guys speak to that a little bit?

I can start with that one and say that identifying where the enhancers are located and which cell types they are active in requires looking at those different and having access to the different cell types, tissues, cell states of interest. I think that the dissection of the enhancer is identifying how does that— how does changes to that enhancer have an effect? How do— we talked a bunch about DNA variants that have an influence on the enhancer. They may be acting by changing accessibility to the chromatin, by changing which transcription— the availability of the sequence to transcription factors that can bind and have their effects.

Yeah. Maybe in terms of technology, one point I was trying to highlight in my talk you can delete completely one sequence and see, do you see that the gene that you think is regulating change expression? You can have this direct assessment. You can use CRISPRi and you can try to shut down that enhancer again, you see if the gene is changing. Then dissect means that you already know that maybe this is an enhancer, and now you want to know, what is the critical part, what are the maybe 10 base pairs or 20 base pairs that are important for this activity. And one example that we show in the Bcl-11A story was you just need a single binding site to perturb a single binding site to shut down that enhancer. So that's the difference between the discovery and then the dissection.

Cool.

So I think the next 3 questions that I saw in the chat kind of all relate to the types of things that are biology that will still make it more difficult for you to do these kinds of experiments. So the first one is about the repetitive nature of the genome. Of course, there is everything from small duplications in the genome to even large large-scale duplications like segmental duplications. How do you deal with that repetitive nature of the genome when you're designing these CRISPR perturbations? Does that make it more complicated for you to analyse the data? How do you sort of deal with that in these kinds of experiments?

I avoid those regions.

So I think Luca can answer this question.

Yeah, I mean, this is really, you know, a good question, you know, depending on the perturbation size, you can say, well, I can try to stay on the side. Hopefully, if you have, for example, CRISPRi, this thing will spread and you feel like it will still shut down or activate the region. But if you have more precise perturbation and if you have the same region in thousands of places of the genome, then it's really hard to understand the effect of the perturbation because you don't know anymore what you're perturbing. Unfortunately, I don't have a solution for this, but it's a really good question.

Sorry, go ahead, Karen.

Well, the related question about cell type and developmental stage offers opportunities to evaluate the cells at different stages of differentiation, either cells in a dish and so from multiple individuals, like iPSC-derived cells that are differentiated or in the process of differentiation into different cell types, you can examine both the effect of variants and any perturbations to those cells and the change in developmental stage to understand their impacts.

Mm-hmm. And I think sort of the next part of that question about cell type context and how important that is, is like, how do you even know necessarily whether you have the right cell type in your assay? And how do you deal with that uncertainty when you're designing these kinds of studies. So I'm looking— oh, go ahead.

Yeah, go ahead, Karen, please.

I'm looking forward to and making use of the single-cell data that is helping identify what the underlying cell types are. And because the data that we generate in bulk tissue leaves that still a bit of a mystery. And so identifying, say, chromatin QTL in bispecific cell types can help elucidate some of that, where the action is happening.

Yeah. Oh, sorry, go ahead, Luca.

Well, you know, there are like, you know, papers, you know, that show like already, you know, chromatin QTL or, you know, QTL, the value actually decomposing this, you know, in different cell types or, you know, subpopulation, even, you know, continuous like, you know, like trajectories, right? So you have way more power, you know, to detect these signals.

So Karen's been doing a great job of leading into the next part of the question, which is, as we move into things like single-cell assays, that generates a ton more data and probably a ton more computational challenges. Can you talk about sort of the biggest challenges of analyzing this data and sort of what maybe is the next important advancement in the field that's going to facilitate analysis of these sort of large-scale CRISPR screens in a single-cell sort of assay environment? That's probably a Luca question.

Yeah, I can start from that. You know, if we think about not computation, but, you know, what is my dream, you know, I think what what we miss right now, especially if we want to assess element or variant perturbation to phenotypes, you can think maybe as the easiest one, changing gene expression. The current assays are still based on the guide count. You actually don't see the alleles that you're creating in this single cell. You have this RNA readout. It's really hard to have a definitive or high-resolution answer if you cannot read what you're doing in individual cells. So in terms of technology, I know there are some efforts in this direction, but we are still not yet there. They are not scalable. But that's the one thing that can change a little bit again. In terms of computational analysis, it's becoming harder and harder because the scale is increasing. We have several companies and labs that are innovating. And, you know, like the kind of skills that you need to develop, you know, for example, if you're a postdoc or a trainee or, you know, a research scientist, you need now to learn, you know, like things related to the big data or, you know, start to learn how to use GPUs or, you know, other techniques that, you know, before were not necessary. But at the same time, it's a an exciting challenge, right? I'm not complaining. Actually, I see this as an opportunity to actually leverage these new computational technologies and data structure that were adopted in different fields.

So there was a question in the chat about combinatorial regulatory interactions. So of course, one of these regulatory elements it's probably not necessarily working by itself. But that adds a whole other layer of complexity to the whole thing. So how do you use these kinds of tools and methods that you talked about to look at the combinatorial effects of different regulatory elements? We're really stumping you guys here.

Yeah, yeah.

No, these are good questions. I mean, I think this is a super exciting area And there are some technical limitations. If you think about single cell as a readout, how many cells can you really profile and how many combinations you can explore? So if you have already an hypothesis about a small number of elements that can have some synergistic effect, you can design these screens. But if you want to think broad, I don't think we have the technology and the funding level to explore, you know, the time to explore, you know, this large, you know, the large combinatorial space. So we need to be really careful in, you know, rational design, you know, the pairs, you know, or, you know, the combination that we want to explore. So yeah, that's my take on this.

I agree. Prioritizing, starting with elements that show an effect and then designing specifically which combinations to evaluate is the practical short-term way forward to be able to investigate those combinations.

So I see a question in the chat that reminded me of one that I had planned to ask you about. So Luca, you received the Genomic Innovator Award for the— to help you fund this work. That is, as you know, a grant mechanism that's a little bit different from sort of the standard grant mechanisms. This is kind of the boring talk about NIH funding, but it uses the R35 mechanism, which is much broader in scope and less tied to like very specific research goals. So how has that helped you pursue those— this kind of work? Has that flexibility been important in sort of helping advance the work in your lab?

Yeah, I mean, absolutely. I think this grant has been really transformative for me. I'm a really collaborative person. I'm excited about too many things. So having the flexibility to do things that will advance maybe one area, that can be gene regulation, but having the flexibility to collaborate broadly with many labs, these are really enable me to create many tools because the way we envision— I'm really excited about new technologies related to gene regulation broadly. And then you can contact different labs that generate data or design together experiments. And I think this is really different if you have a rigid structure where you have a particular disease or a particular phenotype that you want to study. That is really important research, and I like that model as well, But for team science, I think the Genomic Innovator Award really enables to do that without worrying too much about completing this sub-bain. So yeah, I'm really grateful for this, and it's really helping my career in the past few years.

Yeah, you mentioned the team science aspect. That was another sort of unique feature of the Genomic Innovator was that it was open to people who had been involved in some team science kind of projects. And I think, you know, some of your previous work in ENCODE and what you guys will be doing in IGBF are good examples of how those team science approaches can be so important in helping us address big scientific questions that we couldn't really do quite as well on our own in individual labs.

Yeah, totally agree.

So the other question here then is also about funding, and it's about funding and maintaining software tools. That's also a challenge, and you've developed a number of software tools. You know, how do you in your lab maintain them? How do you support them? How do you fund them in your lab?

Yeah, no, that's a really good question. It's really also hard, you know, question. So It's really hard because, you know, like, you know, I don't think we— there are like specific mechanisms. Maybe, you know, this is my ignorance, but I don't know many mechanisms that are specifically designed to maintain softwares. And, you know, unless, you know, your software is like so famous and, you know, foundational, you know, if you have like multiple softwares, it's really hard to maintain them. And, you know, my, my strategy, you know, my practical strategy is trying to use technologies like containers. People probably have heard about Docker, that essentially that thing that you did when you develop, it will always work. So this is a partial solution. To maintain, then you have to stay in this, if we can say, in this cage. You have this small— so it's really hard then to maintain long-term, but at least would be always reproducible and can always run the way you were intended. So I think it's a really hard question. And maybe something that we can say, we need more funds for boring things like maintaining software without no innovation, just like helping. I'm not the funding agency, I think it's a really important problem. So yeah, whoever raised this, you know, like, yeah, really good point. And you know, I think we should all be aware that, you know, some software will be abandoned, you know, if there are no funding that, you know, will support the, you know, the, the development and the maintenance, right?

So, um I see it. There are so many good questions in the chat now. So one of them is about something that NHGRI feels is really important and it has been talking about a lot lately, which is the importance of including participants with diverse genetic ancestries in our research and how we really need to build that in from the start so that we can make the work that we do applicable to all populations. How do you both think about that when you're developing studies in this sort of area?

I can start there. We learn a lot from every genetic variant that can be studied, and there are more variants present across the world in individuals of all populations than are found in just single populations. And so we follow up and focus on, try to identify the basis for the association signals identified across those different populations. All of the biology that gets discovered is relevant to all of us humans. It's just the opportunities that we have to characterize them and identify them through the presence of those variants. And it means that when we test and evaluate the roles of variants, that we can learn more about how elements, variants and elements impact genes.

No, I agree with that, Tosin. You know, like for example, Talking about iGVF, this was one specific point that we had to think together, thinking about being inclusive. And for my side, I'm working with genetics that can help us to think, to say, what are the data? So just think broadly. And there are some gaps that we need to fill, but it might be to see that these teams are more and more important and maybe also required for some funding mechanisms.

So I'm going to ask another one of my questions. So Karen and Luca both know that I'm involved in another NHGRI program called GREGER. What GREGER is doing is trying to address a problem that we have in human disease genetic research, which is that the standard way that we look for the mutations that cause different genetic disorders, which is called exome sequencing, is very often like more than half the time not able to find the mutation that causes the disease. And exome sequencing is one of those look under the lamppost type assays where it's looking in the most likely place, the exons. But that's of course only a very small portion of the genome. These other mutations that are harder to find could be anywhere, and I'm sure a lot of them will end up being in things like these regulatory regions like enhancers. You both talked about some great examples of how those kinds of mutations could lead to disease. But one problem is that when you're looking at such a big part of the genome, it's really hard to look through all of the variants and figure out which ones might be most important and help decide which ones to follow up on. So I wonder if either of you could comment about how these kinds of tools or the resources that might be developed by programs like IGVF might help researchers who are trying to look for those harder-to-find disease mutations?

Yeah, that's a really good question. I think, you know, we have different ways to prioritize variants. You know, IGVF, we have a lot of, you know, active discussions about, you know, including, you know, like existing annotation, you know, fine mapping or other tools. And then thinking really hard about you have too many things that you can test, for example, if you want to perturb. So what is the optimal way to rank them if you want to efficiently explore this space? And then there are, for example, within the GVF predictive modeling groups that are creating machine learning methods to help us to potentially predict the impact. And we acknowledge that the tools are not perfect, but if we have this effort of creating perturbation and readout, we can improve. Something that within AGV we call the active learning strategy. So we're envisioning that it's not something we are going to solve in one year, but over the years we become better and better with these efforts. So I don't know, Karen, you know, if you want to comment on that.

I think the continued discovery of regulatory elements, recognizing that what we've discovered so far is mostly still at the basal sort of cell state or tissue state, and that identifying the elements across more cell types and in these perturbed environments, maybe those relevant to the rare diseases, could help identify, characterize those elements, make the elements available. So then when characterizing, when looking through those large number of potential genetic variants responsible for a rare disease, to have some guidance, some priorities perhaps based on the most relevant regulatory elements.

And I see another question in the chat that's sort of related to this. I think you've probably sort of answered it, but like these kinds of methods could also be used to help understand some of the variability that we see in these Mendelian disorders, even when we we can find the mutation, we still know that there's a lot of difference in the phenotypes that you see in different individuals, even with the same mutation. And so these kinds of tools and understanding variants in these kinds of elements could help explain that. So I see one question in here. Oh, so the Genomic Innovator Program, of course, is a program targeted at early career researchers. Researchers, so people who had not had a major NIH grant before. Luca, what advice do you have for an early career person who's maybe still working on their PhD who's starting to get into this kind of an area?

Yeah, I think, you know, like my suggestion— I mean, some are obvious, like, you know, reading a lot of papers, but, you know, one thing that, you know, You know, it's important. It was really important for me try to follow the technology, right? You know, trying to see what are the new technologies and trying to envision like what, what may happen, you know, with this new technology. Because there are like these waves, you know, that, you know, we, we all experience. They were all, you know, like related to new technologies, you know, sequencing technology, single cell, you know, perturbation. And then, you know, they were like, they were enabling so many new things. So, you know, I think this really helped me to see, you know, a little bit where the field, you know, was going. And there were like surprises, of course, but I think, you know, like this is one, you know, one suggestion they have. And then also being really open, you know, like trying to, you know, you know, like going to specific conferences and, you know, where, you know, you have a mix of different, you know, like, you know, technology, biological talks, and, you know, just connecting and talking to people, I think is like really important. I mean, two that, you know, I really like, and, you know, I'm not associated with them, you know, one is Biology of Genomes at Cold Spring Harbor, and, you know, ASAG, that, you know, is also fantastic, you know, like, so these are, you know, two meetings I try to attend, and they're really broad, and, you know, you can connect with many people, and, you know, so these are, you know, some suggestions.

So I cannot let this kind of a question pass without telling everyone who's on this webinar too that another important thing in being successful in this kind of research is talking to the program officers at the institute that is most relevant to your work. There are a lot of colleagues of mine, including me and other people in my division, who are here really to help you think about your research, how to get it funded, what are the best opportunities for that. So make sure that you contact a program officer. And I also am putting a link in the chat, and I'm hoping that this shows for everyone, to a web page that we have put together that has some information about funding opportunities for new and early stage investigators. There's some, just some general information, links to different events that we have, like a grant workshop at ASH and funding opportunities that are open specifically to early career researchers.

And, you know, talk at the conferences. You know, that's another thing that is so important. You know, often you go to these conferences, so it's sometimes it's so nice just to chat in person.

So, let me look through these questions here. So, I saw one about— okay, we talked earlier about the importance of cellular context. Complex on looking at functional elements like enhancers. Of course, there's like also the level of how different cells within an organ interact with each other and how that might influence how different functional elements work. Can you apply technologies like this in more complex systems like organoids? And if not, sort of what are the challenges in doing that? Or maybe none of you have tried that yet.

You know, I'm not, you know, I've not tried that. I mean, it's a nice question, you know, like, again, you know, I don't do personally experiment in my lab, but I think it is a, for sure, it's an important direction.

So CRISPR editing has been implemented in model organisms, And so model organisms are maybe that step beyond organoids. And so it is feasible and I think important when it's not just one cell type that— and you're not looking for some cell autonomous result to be able to look at the relationship between cells and the impact of a change in one cell to how it influences the others.

So I see a couple of questions in here that are related to another sort of important area for NHGRI, and that is about sharing data. So the one question is sort of how do you share this kind of data that comes out of your lab from these kinds of experiments? And then how will programs like IGBF share the data that comes out of that program so that it can be of use for people, for example, who are trying to interpret whether a specific variant might be important in a disease?

Yeah, no, that's a really good question. I think, you know, I'm always saying that we always need to share the raw data, you know, the closest to the raw data. And then we need to share as well the methods that we use to process this data and, you know, to, you know, reach our conclusions. And, you know, for this, there are like, you know, for raw data, you know, we have Geo, for example, for sequencing data. For code, we have GitHub or other repos. For methods, like assemble methods, there are different avenues where you can put the, the, the, the short version of your software that you use. And then journals have been more and more demanding on this in a good way, right? You need to actually share, you know, like, you know, and for some other, you know, option, they are like, you know, Figshare, you know, Zenodo. And so, you know, we have like a system in place, you know, to share. So, you know, if we don't do that, you know, it's our fault. And, you know, like, I think, you know, it's ridiculous now, like, to see some people, you know, we will share upon request, you know, like, this is something that I think, you know, we should go away from that, you know, should be the default that you share. And you should not have this statement. I know it's a lot of work to reach the point that you are confident in showing your code and all the data, but I think the standard should be higher. It should not be possible to say, well, if you contact me, I will share the data. In terms of AGVF, it's really nice to see that there are a lot of, again, discussion, we're at the planning year, thinking really hard about How can we maximize the data we are going to generate and share? And how to make sure that we share to a level that would be used, right? And so there would be the raw data or preprocessed data, and then there would be something that we call the catalog that will essentially be something really accessible to the community that will also evolve over the years. I don't know, Karen, do you want to add your view?

Right. So IGVF, like many NHGRI consortia, is very interested in getting that data available and available as soon as possible. And so being able to obtain— it will be feasible to obtain data that's generated even prior to being published. And then this catalog idea is to try to to combine the data from multiple different types of experiments and types of data characterization and predictions to make that data more accessible. That's quite the challenge. So I think that it is a long-term process. One of the goals of IGVF is to also make it possible for others to perform the kinds of experiments that are being done in that consortium. And so making available protocols and trying to, if there are standards identified, between members of that community to, you know, get feedback on them, combine them, make those available to be able to— right, there are a lot of cell types, cell states, and variants to cover.

So it's more than just one consortium's work. Yeah, yeah, no, I agree. And I think you also highlighted sort of one of the major challenges. We always at NHGRI want our data to have the broadest impact possible. But it is hard to understand or even think about all of the potential uses of the data and to sort of put it out there in a way that makes it most accessible for people. So I think if you see these kinds of datasets and you have thoughts about what would make it easier for you to use, I'm sure that there will be opportunities to provide that sort of feedback. So I think we're almost out of time. I didn't get to all the questions, but I just want to thank everyone for asking them in the chat. I also want to remind you that this is a quarterly seminar series. We haven't scheduled the next one yet, but we imagine it's going to be in January, I think is what Chris said. So we hope you'll be able to join us for future talks. Keep an eye on our webpage and our social media to find updates about that. And I just want to again thank our speakers, Luca Pinello and Karen Mulkey, for these great talks and this this really exciting discussion, and we hope to see you again at the next Genomic Innovator Seminar.