GBSBC Seminar Series: Jennifer Doudna, PhD - March 3rd, 2022

Wonderful. Thank you so much, Terry, for that generous introduction and also to Joe as the sponsors of this seminar series. I'm really thrilled to be here. Delighted to have a chance to talk with all of you about our science and get your feedback and talk to you afterwards with questions. So, let's dive in. And what I wanted to talk to you about today is basically what we're learning about the chemistry and the mechanism of CRISPR-Cas systems and also ways that we can employ these systems to do new biology. And the first part of the talk is going to be a little bit more on the biophysical side of things. And the second part a little bit more on thinking about how we use bioinformatics to address questions in microbial biology using CRISPR. And so just by way of introduction, just to make sure we're all on the same page, I just want to point out that CRISPR is a system that has become known as a technology for programmed genome editing because of its ability to use an RNA guide, which is the yellow molecule shown here, with a protein, CRISPR-Cas9, to trigger targeted DNA chemistry, in this case DNA cleavage. And so I love this image because it's actually based on structural data for this enzyme that illustrates fundamentally how it interacts with double-stranded DNA, pries apart the two strands, and generates a double-stranded break. And this system originated in bacteria as a bacterial immune system. So we know that in bacteria that have a CRISPR locus encoded in the genome, when they are infected, when these cells are infected by a virus, they can acquire a small piece of DNA, insert it into the CRISPR sequence in a very particular way that maintains the array structure of this CRISPR sequence with repeat elements that flank each of the inserted sequences. And then those sequences are transcribed. The RNAs combine with CRISPR-Cas proteins like Cas9 and trigger targeted DNA cutting. And then over here on the right you can see that in a technology sense this system is extremely useful because it can be employed to target DNA sequences in any cell type, like for example in eukaryotic cells, for cleavage. And in eukaryotes the DNA cleavage event is typically repaired. So this leads to targeted genome editing opportunities with CRISPR-Cas9 and we'll talk more about that as we go through the talk today. And then just to put this into a video format, we imagine that CRISPR-Cas9 operates by searching. We know it has to search through the genome to identify a sequence that has a 20 nucleotide match to the RNA guide. And when that match occurs, the DNA unwinds and this is a property of this enzyme. It opens up the duplex, forms an RNA-DNA helix inside the protein. That's the recognition mechanism and then triggers targeted cleavage of the two strands of DNA that in eukaryotic cells can then be handed off to repair enzymes. They can process the breaks by inserting a small change to the DNA sequence during non-homologous end joining like in that example or by integrating a new sequence of DNA for homology directed repair. And over the last decade since Emmanuelle Charpentier and I first published our work on this system, it's been widely adopted for many different kinds of applications including in medicine and agriculture, as Terry mentioned in the introduction. So, it's really an exciting time and I still feel that in many ways we're at the very beginning of the field because there are so many opportunities ahead, whether it's to cure rare genetic diseases, whether it's to address challenges of climate change by making alterations to plants and perhaps to environmental microbes that can help with carbon capture and food production. But also thinking in the future about ways that it might affect many people in terms of proactively helping us deal with challenges in human health such as cardiovascular disease and neurodegeneration. So there's lots to be done for sure and I love this particular collection of attendees at this talk because I think we need all of your expertise to realize these opportunities. We need people that are clinicians and are thinking about how to treat patients effectively. We need people that think about molecules and how they work. And then we need bioinformatics and computational scientists who can deal with large data sets and the kinds of information that are coming out of CRISPR screens that can increasingly tell us what is actually encoded in our genomes and how to manipulate that information productively. So, you know, what's been really interesting over the last decade is that increasingly this system that started off as a bacterially evolved mechanism for defending against viral infection has been harnessed as a very wide-ranging technology not only for disrupting genes or replacing genes but also for things like controlling transcriptional output of specific genes, using it to diagnose disease including SARS-CoV-2, and using it for imaging of particular parts of the genome. And all of these types of applications fundamentally rely on the RNA-guided mechanism of the system. And so in our own lab, you know, we've continued to explore this mechanism and to investigate how these types of enzymes actually function. And I want to talk to you today a little bit about some of that research and then I'll turn to a particular application that we're excited about as well. So the two things I'm going to talk about today, first of all, thinking about this question of how genome surveillance works. In other words, how is it that an enzyme loaded with an RNA guide can sift through the three billion base pairs in the human genome, find a target site with pretty good accuracy and make a chemical change there or do something that controls transcriptional output. So, we'd like to understand that surveillance process. How does the system read DNA? And then I'll turn to this question of how we can think about new microbial biology at the level of whole microbiomes that can be accessed potentially using a CRISPR approach. So, to the first question about reading the genome, we know that Cas9 is able to read DNA sequences that match the 20 nucleotides of the RNA guide in a fashion that relies on initial binding to this little motif that we call the PAM sequence. And for the enzyme that Emmanuelle Charpentier and I first began investigating, this is a pair of GC base pairs as shown here. So we've got this little motif next door to a potential target sequence. And the target sequence, if it is an actual target for this enzyme, has to match the nucleotides found in the RNA guide molecule. And from some research that we did a few years ago now in collaboration with Eric Greene's lab at Columbia, the work of Sam Sternberg and Sai Reddy showed that you could use a single molecule approach to actually measure the kinetics of Cas9's interaction with DNA. And this was a really kind of a cool experiment because it took advantage of the DNA curtains system that Eric Greene's lab had developed to study things like DNA polymerases and how they move along DNA. And we were using phage lambda DNA molecules. So, you know, a true substrate for the Cas9 system, if you will, and looking at how the enzyme is able to traverse the DNA to look for potential target sequences. And those experiments showed that the protein when it doesn't have a guide RNA loaded onto it, it has virtually no detectable interaction with DNA. The off rate presumably is very high. As soon as we have a guide RNA loaded onto the protein, we know that there's a structural change that occurs in the enzyme that puts it into a structural state where it's now competent to spend a little bit more time associated with DNA. So the off rate goes down. And furthermore, we found that the interaction with the DNA involves initial recognition of this PAM motif. So that would be a motif that would occur quite frequently in a typical genome, a pair of GC base pairs. And that transient association somehow is able to open the DNA duplex next door to the PAM for long enough and with sufficient length to allow this kind of base pairing interrogation to occur. And then from data from a number of labs not only our own we know that the DNA then unwinds directionally from the direction of the PAM to the distal end of the duplex. And so the formation ultimately of this structure that we call the R-loop, which is a fully engaged Cas9 in which the guide RNA is fully base paired with its target sequence in the DNA, this structure forms in this directional fashion and there's a lot of data showing that mismatches between the guide RNA and the DNA trigger changes in the kinetics of binding such that those sites tend to be excluded and also that there are conformational changes in the enzyme that accompany that process of recognition. And there's some really nice literature on this whole process as well as a number of publications in which people have, and we've done some of this work ourselves but a number of other labs too, have been able to show that you can manipulate the accuracy of binding by playing around and making mutations in the protein that alter that balance. And this is something that still really blows my mind, this whole process which involves DNA melting in the context of a genome is an ATP-independent process. So there's no ATP hydrolysis going on that drives that unwinding. So you may know that there are lots of enzymes that are typically helicases and things that are able to melt DNA and a lot of DNA repair proteins work this way where they require ATP hydrolysis as an external energy source to melt the DNA. Somehow these CRISPR proteins do not do that. They have an intrinsic capability to melt the DNA. So that's been something that we've been intrigued about and trying to understand. And then finally I'll just point out that there's a lot of nice data over the last few years that shows that the rate-limiting step in the whole genome editing process using CRISPR-Cas9 is at the level of the target search. So the target search typically takes minutes to hours versus the very rapid chemistry that can happen as soon as Cas9 is actually fully engaged on a target sequence. So we think that it's very important to understand what's happening in that target search process. Why? Well, because that is fundamentally determining the accuracy of the system. It's determining the rate at which editing can occur. And it really does underscore and underlie everything that all these different types of applications that these systems are now employed for in different cell types. So whether we're using these proteins for clinical purposes or whether we're using them in plants or microbes, in every case this target search mechanism underlies how well these tools will actually function in different cell types. So, we've been keenly interested in understanding exactly what's going on during that genome surveillance process. And there are a couple of clues that I'll just bring to your attention and these are from work that's been published over the last few years. One clue is that we know that PAM binding, this recognition of this little dinucleotide, you know, GC base pair motif in the DNA is an essential part of the target recognition process and biologically it makes sense because in bacteria this is actually utilized as a way to avoid cleaving the host genome and I won't go into the details there but Eric Sorek and others have done, and Luciano Marraffini have done nice work originally showing this and now of course many others. But I'll show you the effect biochemically and so that's in an experiment I'm showing here that was performed by Sam Sternberg when he was a student in the lab and what Sam did was an experiment, this is a typical biochemical experiment to test Cas9 RNA-guided cutting of DNA. So we have a DNA molecule, it's a double-stranded piece of DNA, short DNA that's radiolabeled. And if we add Cas9 and a matching guide RNA, guide RNA that matches the DNA and we have a PAM sequence present, you can see that over time we get DNA cutting and we get the cleavage pattern that we predict based on Cas9 recognition and cleavage of each DNA strand. And then if we do this experiment in the middle where we take an unlabeled version of that DNA substrate, still has the PAM and the 20 base pair match to the guide RNA. This is effectively a perfect competitor for the enzyme. It binds just as well as the radiolabeled form of the substrate that's there in limiting amount. And this excess competitor can then bind up all the Cas9 and we get very little cutting going on. So that's a control reaction. And then on the right-hand side is the experiment where now we take a DNA molecule that has the 20 nucleotides that match the guide RNA sequence but we've mutated the PAM motif. So there's no PAM. And you can see that in this experiment this DNA is invisible. It's basically ineffective as a competitor because the enzyme can't see it. And you see on the right-hand side this principle kind of illustrated a bit further where we can test these different molecules shown on the x-axis as competitors. And what you find is that remarkably even if you have no sequence in the DNA matching the RNA guide at all but you have increasing numbers of PAM sequences, those molecules can work increasingly well as competitors. Really suggesting that the way the enzyme first engages with DNA really involves PAM binding. And then the other clue that I want to point out here is that several studies early on looking at the structures of Cas9 as it forms a complex with its guide RNA and then interacts with double-stranded DNA showed that the protein undergoes a large conformational rearrangement. And this was first shown by work that Martin Jinek did when he was a postdoc in our lab and has been now shown in a number of other structural studies. We didn't know what that structural change signified initially, but as I'll show you today, we think we now understand why the protein does this and that it's really a very fundamental part of the way that the enzyme reads DNA. And so, the question that we wanted to ask, and this was really the project of Josh Kofsky, a very recently graduated student from the lab, was whether he could trap this complex that we call the PAM encounter complex, which by definition would be a very short-lived complex that would happen probably extremely fast and extremely frequently during the search process of this enzyme with its RNA guide in a typical genome. So we imagine that very frequently it's encountering this situation where there's a PAM motif, a couple of GC base pairs next door to a potential target binding site, but that sequence doesn't have complementarity to the RNA guide. So it actually doesn't progress to form this type of an R-loop structure. And so we imagine lots and lots of interrogation like that going on. And so the question was could we actually trap this encounter complex and not perturb the structure in a way that would make it not interesting to study but also not let it progress to this state over here. And so Josh's strategy for doing this was to introduce a cross-link. And I want to point out that, you know, I'm showing you this R-loop structure in the cartoon here, but please know that the way this work was done was to use a piece of DNA that had a PAM sequence and then no complementarity to the RNA guide on this end. So there's the PAM but no ability to form this type of an R-loop structure. And so to trap that Josh put a single modified nucleotide into the DNA immediately adjacent to the PAM sequence and a cysteine in the protein in a position that we thought would be very close to that modified nucleotide so that we could get a disulfide cross-link. And I'll just quickly show you here in this little morphing video where that's located. So the PAM sequence is in yellow. You can see the two G's of the PAM being contacted by arginines in the protein and then right next door is this disulfide cross-link between the DNA and the protein. And so the idea here was not to perturb this interaction but simply to trap it using this type of a cross-linking strategy. And I won't show you the data for the controls here, but we got very high levels of cross-linking between 50 and 70% typically for these samples and they retained full DNA cutting activity. So, we were confident that this cross-link was trapping a structure that wasn't somehow preventing Cas9 from going about its business when it had a DNA molecule bound like this. And if you want to read the details there, there's a bioRxiv paper that we've posted on this subject. So, without material in hand, well this is just showing the kinetics of cutting of these different samples that the cross-linked material has the same kinetics of DNA cutting. And so Josh went on to work on structures using cryo-electron microscopy and I don't have her name here but I want to mention Kasia Sake and also Professor Evan Egelman who have been close collaborators of ours on the project. And so what they found was that these samples existed primarily in two conformational states. One in which the DNA was bound in a linear form to Cas9 and the other where the DNA was bent and sat a bit differently in the protein as I'll show you. And so what was really interesting to see was that the DNA binding occurred again sort of with this interaction with the PAM sequence which is shown in yellow color here and resulted in a bending and twisting of the DNA when we compared the linear and the bent forms of DNA in these structures. So here I'm showing you a little morphing video that illustrates the conformational change that happens when the DNA is bent in Cas9. And we were fascinated to see that the bending is occurring immediately adjacent to the PAM and involves the nucleotides in the DNA that have to first contact the RNA guide when that R-loop forms. And importantly, these nucleotides are being flipped out of the helix and made available for potential base pairing to the RNA guide. And so what causes these dynamics? And here we went back to the structural data that were published previously for different forms of Cas9 in complex with its RNA guide or bound to a DNA molecule. And we found something very interesting which I'm going to show you by comparing these two little morphing videos here. So on the left-hand side I'm showing you the structural states of Cas9 that we observed with the linear versus the kinked form of DNA. And you'll see that there's a large rearrangement in the Cas9 protein structure when we go from the linear to the kinked DNA conformation. So the protein is effectively changing conformation we think to induce this transient bend adjacent to the PAM. Over here on the right I'm showing you conformational states of Cas9 that had been observed previously and I think we first described in 2014 in fact and very similar conformational change. We did not understand at the time what the significance of that structural rearrangement might be. We wondered whether it might have something to do with forming the R-loop in DNA or perhaps unwinding the DNA duplex. And we now think that this structural change is actually inducing the kind of bending that is critical for the enzyme to sample the adjacent sequence in the DNA to look for potential base pairing with the RNA guide. So our thinking right now is that this is really a fundamental aspect of the enzyme and that these potentially isoenergetic forms of the protein are why we don't need ATP to achieve that. That it's really this probably very rapid switching between these conformational states of Cas9 that triggers that kind of DNA dynamics. And this just shows again a little bit more close-up a different view. You can see how that bending occurs. When this portion of Cas9, this region of the protein swings over, it results in just a physical bending of the helix and the exposure of these nucleotides over here. And in structures that I won't show you here in the talk but I want to mention Josh was able to actually trap structures of the enzyme with just two or three complementary base pairs between this DNA strand and the RNA guide that nicely kind of bridge the structure you're seeing here, the full R-loop formation with this intermediate form in which just two or three base pairs have formed between the DNA and the RNA. We really think that is the initial encounter complex that this enzyme is sampling. Very quickly I want to just show you two experiments that Josh did to explore this a bit further. So we had these nice structures and the question was is the enzyme actually doing this in solution? And so the two experiments that Josh designed to test this, first was to test whether Cas9 alters the chemical reactivity of nucleotides next to a PAM. So that was one prediction of this model was that we would imagine that those PAM-adjacent nucleotides would be transiently exposed in solution and you could observe this by reacting them chemically with something. And secondly, whether Cas9 in fact induces a PAM-dependent bend in the DNA which again would be our prediction based on these structural models. So I'll just show you these experiments and data very quickly here. But the first experiment kind of goes back to some old nucleic acid chemistry where we can use permanganate probing to detect unpaired DNA and this is just illustrating the reaction that you can use. So that with single-stranded DNA for T residues that are exposed in a single strand or melted out of a double-stranded duplex we can react with permanganate and followed by piperidine and denaturant to lead to strand cleavage. And so this is a reaction that you can easily detect in a biochemical type experiment. And for our purposes, the goal was to ask whether T's adjacent to a PAM sequence in the presence of Cas9 become more reactive in this type of an assay. And so this is summarizing some of the data that Josh was able to measure using this assay. So we have a system set up where we have a duplex DNA molecule with a PAM sequence and a potential target recognition site here shown in red. When we have a mutation in one of these base pairs, so we destroy the PAM motif for the molecule, we see no change in chemical reactivity for any of these T's as a function of increasing Cas9-RNA complex titration. No change in chemical reactivity to speak of. Down here when we now have the PAM motif intact in the same otherwise identical DNA molecule, what we found was that both of the T's that were immediately adjacent to the PAM, these labeled plus one and plus two, are much more reactive as a function of increasing Cas9 concentration. And so that really does suggest that this interaction with Cas9 triggers transient duplex melting that can be detected by this chemical reaction. And the other experiment, oh, I'm just noticing this is quite pixelated, so sorry for that. But this is based on some beautiful work done years ago by Don Crothers when he was at Yale and studying A-tract bending in DNA molecules. And they came up with a very nice assay that takes advantage of DNA ligase that simply is an end-proximity assay where if bending occurs like this where the two ends of the DNA are brought close together then you get rapid DNA ligation kinetics whereas if bending occurs like this where the ends are far away from each other then ligation kinetics are very slow in general. And so the strategy in the case of Cas9 was simply to use this kind of molecule that's shown here where we have an A-tract that intrinsically bends the DNA in a particular direction adjacent to two PAM sequences that were a variable distance apart. And so the idea is that when you add this type of molecule to Cas9 with its RNA guide, if these PAMs are bound by Cas9 and bent in the same direction as the A-tract and that will be a function of the distance here, then we're going to see rapid ligation. And if bending occurs in the opposite direction like over here, then we're going to see slow kinetics. So you could imagine sort of an almost cyclical type of plot that you could get from these types of data. And that's what is published in this original work by Crothers' lab for a different system. And so in the case of Cas9, we found very beautifully that Cas9 induces a phase-dependent DNA cyclization. And so essentially what this is showing you is that as a function of the distance between the PAM position, so the site of Cas9 interrogation, and the A-tract in the DNA we could see this very nice relationship of ligation kinetics for the DNA. And furthermore, and I won't show you this explicitly here, but if you're curious, you can look in the bioRxiv posting, we also can map the direction in which the DNA is bent. And when we match that up with our structural studies, we get a very nice match. So we really have, I think, very nice evidence both biochemically and structurally that Cas9 interrogation involves this transient melting that is PAM-dependent in the DNA. So we think that Cas9 is bending and twisting DNA to read the sequence next to PAMs, that this target search speed is probably a function of intrinsic protein dynamics, and importantly that this search speed in DNA is really distinct from either the binding or cutting speed. So it really is something that is probably a fundamental property of this conformational change that we're observing in Cas9. And so of course in the future we'd like to understand whether target search speed is a limiting factor for genome editing efficiency and safety. We'd also like to know whether the natural diversity of target search speeds, you know, what it is and whether that can be harnessed in a way that makes it easier to identify really good genome editing proteins. There's now lots of different Cas9 and Cas9-related types of enzymes out there but only some of them are good genome editors. Why is that? And we don't really know but I think this probably holds the key. And furthermore, it may be possible to access even faster searchers if we can understand how to engineer these enzymes to carry out this kind of interrogation quickly. So I want to now turn in the last few minutes to this question of how we might be able to use CRISPR in essentially its natural setting. So thinking back to, you know, a lot of the focus on genome editing has been with eukaryotic cells but I think there's a really interesting opportunity to turn CRISPR back to its origins in a way and use it to understand new biology in the microbial world, in particular in understanding microbiomes. So you probably know that a lot of the research that's been done on microbes, and by the way, a lot of what we understand about very fundamental biology comes from this type of research as well as some of the most useful tools that we have come from microbes. And yet the vast majority of these organisms are what we call the uncultured majority. They're not cultivatable or at least haven't been cultivated in laboratories. And in many cases, as shown in this phylogenetic tree shown over here, which I got from the lab of Jill Banfield, by the way, we have no isolated representative. So for the vast majority of these types of organisms, we don't have an isolated example that we can investigate in the lab. And certainly in many cases this is probably because these organisms don't grow that way. They actually grow in the context of communities of different microbes that share a biological niche. They probably have interesting interactions with each other that would be important to understand and that probably affects their behavior in the biosphere. And yet we haven't really had good ways to start understanding that biology at this level. And so because we don't have these types of tools right now for the most part, I think there's some really interesting opportunities if those tools can be developed, namely to think about how we can understand the microbiomes that are involved in food production, in things like wastewater treatment, as well as really accessing the biology of the carbon cycle in ways that haven't been really possible up until now. And so these are some of the areas that we're interested in and are motivating for thinking about how we can use CRISPR to access these new aspects, new areas of biology. So we've wanted to build what we call sort of a generalizable tool set for

There we go. Now it shouldn't be muted. Sorry about that if I clicked the wrong button. Right. So, you know, we're sort of thinking about how to develop this. And it's really been a two-part process. And I want to point out this is the work, it's been a wonderful and close collaboration with primarily Jill Banfield but others that I'll also mention at the end of the talk and three scientists in our labs who have been focused on this effort. One is Ben Rubin, another is Brady Crest. Both of them are postdocs in our lab and Spencer Diamond is a project scientist in the Banfield lab. And so they've really teamed up to work together on this challenge. And so the idea was to first for a given microbial community ask which organisms are able to acquire foreign DNA because if you want to manipulate a genome you obviously got to be able to do that. And so to ask that question, we used a non-targeted transposon system coupled to what we call environmental transformation sequencing or ET-seq that would essentially just allow us to measure the transposon insertion efficiencies in these different organisms as a function of different ways of transforming the cells. And then in the second step we wanted

To use CRISPR RNA-guided transposons, we could target individual genomes and loci in those genomes to do targeted editing in the context of this whole population of organisms.

So first, just to briefly describe this ET-seq strategy. The idea here is to get information about the genetic accessibility of these different organisms. So the idea was to treat a sample, and of course we have to have some kind of a sample that we can manipulate. We've been able to do this initially with some contrived microbial communities of nine-member or twenty-member systems, and then more recently with actual human fetal gut microbiomes that are much closer to what you would want to be able to do ultimately with this type of a system.

But I'm going to show you the data for this nine-member community initially. And again, the idea here is to use a non-targeted transposon with different methods of introducing the DNA into these cells and then measure insertions and insertion efficiency using a sequencing-based strategy. And we can divide the frequency of insertions again by the abundance of these different genomes to get a plot like is shown on the right.

And so just to show you what some of the data for this look like. So in a control reaction that's shown on the left, this is an experiment where we take this nine-member community and we spike in a pre-edited organism that is one of the members of this community and then use ET-seq to measure its abundance. So we know what it should be and then we use ET-seq to measure what we actually get. And we get a plot like this. So it looks like ET-seq is actually a pretty good measure of the insertion frequencies that we get as a function of the abundance of the organism in the sample.

Then over here on the right is the actual data that we get when we apply this approach to the whole nine-member community. And I want to just point out a couple of things here that I think are interesting. So first of all, this is comparing three different approaches to introducing the transposon DNA into the cells and you can see that in some cases it doesn't matter that much but for some of these organisms it matters a lot where you get some pretty decent insertion efficiencies for some transformation methods and none for others.

We also notice that this is an approach that's very much dependent on the abundance of the genome in the sample. And so as we move across to the right and these are less and less abundant in the sample, it's harder and harder to detect insertions. And this is one of the things that we're grappling with currently is how do we either increase the sensitivity of our detection or increase the actual frequency of insertions by increasing the uptake of foreign DNA in these organisms so that we can actually get observable editing in these organisms that are very inabundant in the sample. But that's an ongoing challenge.

And then I just want to mention briefly about using the CRISPR-based transposition system to do targeted integrations into these organisms. So the great thing here is that these RNA-guided transposons effectively use the RNA guide, at least the one that I'll show you is very good at this. This is a system that was first published by Sam Sternberg's lab at Columbia where they've been able to use the RNA-guided system here to get very nice precise insertions into the targeted genome.

And I'll just illustrate this very briefly here. So this is just showing a cartoon of the way the system works. And by the way, the transposase that we're using is actually not hooked up to a single Cas protein like in this cartoon, but it's actually working in context of a multi-protein assembly with an RNA guide. This is called a type I CRISPR-Cas system that uses a much larger protein structure to interact with the RNA guide. That's something just to bear in mind. It's not a single CRISPR-Cas like Cas9 in this case and it's hooked up to a transposase.

And so the way it works is that the RNA guide allows this Cas complex in this case to interact with the DNA forming a similar kind of R-loop structure with the DNA and then a short distance away the transposase is able to sit down on the DNA and catalyze DNA integration. So by changing the RNA guide, you can target this to a particular genomic locus and to a particular genome. So in principle, if you had a microbial population that you wanted to edit, you could use this as a way to target just a particular locus in a particular genome and nobody else would get touched by this transposon if things are working as one hopes.

And so, Sternberg's lab published this. Originally, they had all of these different component proteins that were encoded on three separate plasmids. And what Brady Crest in the lab did was to combine these into a single vector that's shown here that he called VchDart, Vch for Vibrio cholerae, which is the origin of this system originally. And this in principle would allow introduction of this single plasmid into organisms where transposition could then occur in an RNA-guided fashion.

And just to show you one example of this and there's a lot of other data that we've published on this recently but we know that this VchDart system is really precise. It's highly specific. It's truly dependent on its RNA guide. This is one example here just in E. coli showing that when you program the system to recognize the lacZ gene we get very nice transposon insertions at that position with a very narrow window of insertions that's shown in the enlargement here on the right and essentially no insertions anywhere else.

So it really is a very specific kind of system and there was another Cas transposon published around the same time as the Sternberg lab work and we tested that one too and it's really imprecise. So for whatever reason this one is really dependent on its RNA guide.

So then I just finally, and this is my last slide here, I just really wanted to show you what this looks like when we apply it to editing in the human fetal gut microbiome. And this is a slide that just summarizes a really huge amount of work on the part of many people but in particular Ben Rubin, Brady Crest, Spencer Diamond in our lab and in collaboration with a couple of groups at Stanford to identify isolates from roughly 90-day-old fetuses.

And what they did initially was just to show that these samples are quite stable. So we can take different isolates from this inoculum and we get roughly the same abundance of organisms each time that we look at it. So that's good. And then the experiment was then to ask whether we could target two closely related but distinct E. coli strains that are found in this sample at a couple of different genomic loci using the CRISPR transposase.

And this is just some data here shown on the right illustrating the results that we get. And so if we look at three different samples that are grown up without any editing, you can see the range of variation that we typically observe in the abundances of these different component organisms. And then in the middle and right sets of lanes in triplicate here you're seeing the experimental data that we get for targeted editing of one or the other of these closely related E. coli strains.

And what you can see is that when we use the transposon to insert a selectable marker, we can very nicely get outgrowth of one or the other of these E. coli strains over time. And so we're excited about this. We're hopeful that this type of strategy can be extended to additional types of organisms, something that we're working hard on doing right now.

And what I guess I'm most excited about is being able to do new biology with this type of a tool because we really would love to understand how these types of organisms are interacting and not only in this fetal gut system but also in some of the work that Jill Banfield that her laboratory is doing on environmental microbiomes and looking at how these organisms interact symbiotically and in other ways where we imagine that the biology is going to be different when you study them in the context of their natural communities versus when you study isolates, something that we're actively exploring currently.

And so I'll just close by pointing out that going back to what I said in the first part of the talk, really continuing to understand how these types of proteins interact with DNA we think is going to continue to drive the development of the technology as well as help us to understand just the fundamental biology of these CRISPR-Cas pathways in bacteria.

We also think that using the combination of environmental transformation sequencing and the CRISPR transposase system will enable microbiome editing that we hope will actually help us to start unlocking new biology of not only the human gut microbiome but ultimately also of some environmental samples that we're studying.

And I'll just close by thanking various people that have been involved in the work. So huge thanks to the lab of Jill Banfield and many of her lab members. I mentioned Spencer. These three guys really have been a great team working together on the work that I talked about in the second part of the talk. And we've had also great partnership with Adam Deutsch, Rudolph Barango, and Trent Northern who've been part of a large DOE-funded effort that is focused in part on doing this kind of microbial community editing.

And then I also want to give a big shout out to Josh Kasha, Gavin not who I didn't mention but is a recently matriculated postdoc I guess you could say. He's just started his own lab at Monash University in Australia. So these three folks really teamed up and worked together on the structural biology of Cas9 that I talked about and we've had of course wonderful collaborative work that we've been able to do with Jill, with Eva who I mentioned as well as giving a big shout out to Emmanuelle Charpentier with whom we started the Cas9 research over a decade ago now. So, I'll stop there and would be delighted to answer questions if you have them.

Thank you for that question. It's not naive at all. I think that you cut right to the, you know, that's one of the important questions that's being explored currently with using CRISPR for clinical purposes. I think one thing that's very interesting is that, and I sort of implied this I guess in the beginning of the talk, is that CRISPR is great for making knockouts. It's harder to make knock-ins right now. So for actually correcting a disease-causing mutation, that's one of the forefronts of the field currently is really figuring out how to do that kind of correction precisely and in enough cells where you have a clinical benefit.

And the other point I'll mention is just that it's been fascinating to see the data for a number of diseases or clinical trials that have been published so far showing that, and some of this comes from mouse animal models like mouse models as well, that in many cases you don't need to correct every cell. It's enough to correct 10 or 20 or 50% of the cells to see a benefit for patients and I think that's good to know because in principle it probably be pretty hard to correct every cell, but I think getting to a reasonable percentage of cells is actually something that's quite achievable and is already being done in a number of cases.

Yeah thanks for that question and I'm thrilled to know that you're focusing some of your effort on bioethics. It's a very important area as these kinds of new technologies advance we have to be on top of the ethical challenges that are coming up. So with CRISPR you probably are aware that it can be used in the germline including in the human germline and it has been done that way, used that way. And I think that that really raises a whole fascinating set of ethical and societal challenges. Should we be supporting and encouraging that type of use in the future? I think currently the answer is no. But, you know, I'm also open to, I think we want to continue to keep an open mind about whether that type of application of CRISPR could, you know, if it could reduce human suffering in the future. That's something we have to be open to.

So, there's that. There's also applications in the environment and you could imagine challenges there, especially in the context of something I didn't talk about today called gene drive where you can use CRISPR in a fashion that in some of this work is being done at San Diego as you probably know to spread a trait quickly through a population of insects let's say and could be beneficial for controlling human disease spread but also could have environmental impacts that might be difficult to control. So I think those are areas and there definitely are others where we definitely need people like yourself that are focused on the ethical challenges who also understand the science to think about it.

Yeah, great question, Erin. So, there's some nice work that's been published on that addressing that question in more of a kind of a bulk setting, you know, looking at kind of overall editing efficiencies and different types of chromatin, for example. ChIP-seq type studies that have looked at how nucleosome positioning can affect the efficiency of editing. I think it's fair to say that there's no absolute effect there. In other words, it's not as though nucleosomes absolutely prevent editing or that there really isn't evidence that chemical modification of DNA like methylation does not seem to really affect CRISPR-Cas9 induced editing.

However we know that certainly the chromatin dynamics absolutely have effects that people have detected. So going forward in our own work for example we're actually very keenly interested in that question and I have somebody in my lab right now that's using live cell imaging for example to start really trying to look at that in some detail.

Exactly.

Well, I guess I would just say that we haven't looked in cells at this mechanism yet. We're doing all of this work in vitro, you know, with biochemical samples. However, I think it's likely that this mechanism underlies all of the activities of Cas9. We have no reason to think otherwise right now and that's something that will be tested going forward. But, you know, I think it's important to think about this because it does mean that Cas9 as it moves through a genome, it's transiently melting the DNA. Why does that matter? Well, for example, when you think about using Cas9 in the context of base editing, which for those of you that are aware of that type of application, it's hooking up an editing domain to Cas9 so that you get a chemical modification to a particular DNA nucleotide when it is exposed transiently by Cas9.

It just means that you have to think about that in the context of the way the enzyme is moving through the genome if you want to avoid non-specific or off-target base editing. So that's one of the things that we've worked on in the past and that we're thinking about going forward that argues that you need to know these things about these enzymes if you want to be able to use them productively.

Oh, that's because in the insert we're putting in a selectable marker, right? So, we're putting in a gene that those bugs now require under selectable conditions. And so, that is just for us, it's just a way to detect quickly bugs that have acquired that insertion.

Right, well so there's a lot of nice work some of it done by people like Kira Makarova and Eugene Koonin if you're interested in the evolutionary origins of CRISPR-Cas I would refer you to some of their work but fundamentally it looks like these systems evolved from primordial transposons that became RNA-guided over time. And then you alluded to the other essential evolutionary step in creating the CRISPR pathway which is becoming capable of acquiring new sequences that can serve as guides, right? I mean that's, I didn't talk about that today, but I think that's also a fascinating aspect of these systems is that they're adaptive, right? So they really allow bugs over time in real time to acquire immunity to new phage and to do that very very quickly.

And how do they do that? Well, they encode a special integrase that is responsible for capturing those little phage sequences and inserting them into the genomic CRISPR locus. So it's a really interesting system where you've got this acquisition going on on the one end and then you're quickly transcribing those sequences and using them to guide the targeting part of the system on the other.

Well because we understand that self versus non-self mechanism well enough that we can override it essentially right so the PAM sequence is essential for determining self versus non-self and so once you know that you can then manipulate the system so that you can target the genome if you want to.

Yeah, it's a really good question. My suspicion is that it's not going to be ascribed to a particular gene in a host for example. It's probably more of a fundamental kind of intrinsic property of the editor itself. So that's kind of what I was implying in the first part of the talk is that we think that, I didn't say this explicitly but one of the things that I'm still scratching my head over is that the very first protein CRISPR-Cas9 protein that Emmanuelle Charpentier and I started working on more than a decade ago is still in many ways one of the best genome editors right and that by Murphy's law that probably shouldn't have been true but somehow it is. Why is that? Well so we're fascinated to know what it is. What are the properties of this enzyme compared to some of the other CRISPR-Cas proteins that make it a superior genome editor? And one of them is we think the kinetics of this DNA search process, but there probably other things too.

Yeah well that's you're taking a big leap forward which is great to think about. I think that one thing to appreciate is that right now these transposon systems are big. You know the one I showed you has I don't know it's something like seven protein components to it. So it's big. So delivering that in a clinical setting we'll have to figure out how to do that ultimately if that's desirable. But I agree that, you know, that's in the longer term, that's where we'd like to see this go is that it becomes possible to actually do that kind of targeted genome editing in a native microbiome. I mean, that would be awesome. I mean, imagine if you could use it in a way that would impact the human gut microbiome without requiring fecal transplants or something. I mean, this would be really interesting.

Yeah, I think both. I mean, I'd love to see both of those types of things move forward. And, you know, we've had some really great discussions over the past few weeks since we published some of this work with folks on both sides there, right? Some people that are doing really fundamental research on microbial communities where they have ideas about how to use this approach in their work in their research as well as with clinicians who are actively exploring the microbiome in connection with various kinds of bowel disease and even neurodegenerative disease and thinking about how we might be able to use this approach to manipulate those clinical samples in ways that could allow future applications in medicine. So, lots of opportunity here.

Yeah, that's my suspicion. I think it's literally that you have this enzyme that's kind of doing this, you know, if you think back to the first part of my talk and that those structural states are isoenergetic that there's very low energy barrier for moving back and forth. And we have some evidence from single molecule studies that that's true. And so you could imagine that essentially because that's fundamentally the way the enzyme works that it's able to perturb DNA. It's got a PAM binding pocket in the enzyme that locks it onto PAMs as it's doing the search process. So you get bending adjacent to PAMs. That's what I think is going on.

Yeah it's a great question. I think in that paper and perhaps what you're referring to is the fact that when Cas9 engages with not only a PAM but an actual target sequence where you've got binding that sort of RNA-DNA hybrid that forms in the enzyme that has a pretty long lifetime. It actually lasts quite a long time even after the DNA is cut and so on the order of many minutes and maybe longer but certainly in that experiment we could detect out to many minutes and so it does suggest that in cells you could imagine that there probably are other proteins, replication machinery etc that might strip Cas9 off the DNA but if you did in the absence of that those engagements with a target sequence have quite a long lifetime.

Yes, that is a very perceptive question and the answer is yes. And we haven't done that, but there's a lot of nice work on this. One of the nicest papers I think is actually from Andy May and Rachel Horwitz and others at a company called Caribou Biosciences that full disclosure I'm affiliated with, but I think they just did some really nice biochemistry. It's a Molecular Cell paper from a few years ago and they looked at this question in a lot of detail and did a lot of sequencing and they showed that depending on the target site and the particular Cas protein that you're using and to some extent as a function of the guide context that you could predict in many cases the editing outcomes that were most prevalent.

And I think for all of you that are bioinformaticians in the audience this is a really interesting opportunity in the future, I think, to ultimately be able to predict editing outcomes before or without even having to do an experiment once we have enough of those types of data.

You know, I go back and forth on that. Do you want a longer PAM or a shorter one or no PAM? I don't know. I think you could argue it either way. So that's something that needs to be tested.

Well, I think one of the ways is as I just mentioned, I think over time it will become possible to predict editing outcomes with accuracy in different cell types and that's very important because it will mean that you can predict not only the on-target editing but potentially off-target editing and that's already being done to some extent. So I think that's one way where bioinformatics will be really important. Another use though I think of computational approaches is going to be helping to understand the vast amount of data that are coming out of CRISPR screening where people are using CRISPR to identify gene functions and increasingly whole sets of genes that interact together but those data sets are very complex and so having good tools for doing that will be essential.

Well I just want to thank you Terry for hosting a really wonderful opportunity for me certainly and thanks to all the attendees. It was great to talk with you those of you that asked questions. I really appreciate it.

Fantastic. And keep up with the great work. We'll be in touch regarding the program.