Hi. My name is David Bikard and I’m the head of the Synthetic Biology Group at the Pasteur Institut in Paris. And today I’m going to talk to you about CRISPR technologies for bacteria and some of the work we’re doing in my lab. But first, I want to give you a brief history of the CRISPR system and how these fascinating systems were discovered. And the story begins in 1987, when a Japanese group made this interesting observation, where they sequenced a piece of the Escherichia coli chromosome. And next to this iap gene, they identified this interesting locus with repeated sequences that were interspaced by these variable sequences, here, but that had a very conserved size. And it was only much later that we really understood what the systems were actually doing. And in basically the next 25 years, with the advent of genome sequencing, we started to have a lot more sequences of archaeal and bacterial genomes, and people were identifying these interesting loci in a lot of places. And they were identified independently many times and given many different names, like DVR, TREP, LTRR, and the list goes on. And it’s only in 2002 that, basically, the name CRISPR was coined and unified in the field, and this stands for Clustered Regularly Interspaced Short Palindromic Repeats. And in this paper by Jansen in 2002, they also described the existence of a set of genes that were constantly associated with these CRISPR loci. And they named them cas genes, for CRISPR-associated, here. So, the way these loci look is you have the CRISPR locus itself with the repeats and spacers, and the cas genes that are very frequently found next to the CRISPR. And it was only in 2005 that three papers were published the same year making the same discovery, that the spacer sequences between the repeats, some of them are actually matching foreign DNA, and specifically bacteriophage DNA. So, you have these… these three papers that made this observation. And in particular, also, the paper from Francisco Mojica, where they found this to be consistently the fact in many bacterial and archaeal species, and they also observed that bacteria that carried CRISPR spacers that were matching a bacteriophage could not usually be infected by that bacteriophage. But this was still all theoretical and bioinformatics work. And the experimental evidence that the CRISPR system was actually an immune system to defend against foreign elements and bacteriophages only came with this Science paper in 2007, by Rodolphe Barrangou in the group of Philippe Horvarth. And this is an interesting paper in many respects. And in particular because it comes not from academia, but from an industrial company called Danisco, where they produced pheromones for cheese and yogurt production. In this industry, they always need to make sure that the strain, bacterial strains they are using cannot be infected by bacteriophages. They make milk cultures of very large volumes, thousands of liters, and it can cost a lot if you get bacteriophage contamination. And so, when selecting strains, and specifically in this paper it was Streptococcus thermophilus, that are resistant to bacteriophages, they found that the CRISPR locus was actually changing and adding novel repeat spacers units. And specifically, they did this experiment where they infected the strain with two different bacteriophages, and they noticed that the CRISPR system was capturing novel sequences that matched the genome of these bacteriophages, and then it would become resistant to the corresponding phage. So, this was the first experimental evidence that CRISPR systems are an adaptive immune system of prokaryotes. So, in the… in the next years, many groups started to look into these CRISPR systems and describe their mechanism and mode of action. And basically, when a bacteriophage injects… injects its DNA into the cell, then a group of cas proteins, and specifically cas1 and 2 are involved in capturing pieces of DNA from the bacteriophage and integrating it as a novel spacer in the CRISPR locus. The CRISPR locus is then transcribed. And the transcript is then processed into this small guide RNA, or CRISPR RNA, that is able to guide a complex of cas proteins to go and target and destroy target sequences. So, the first step is referred to as immunization, or CRISPR adaptation, CRISPR spacer acquisition. And then we have the immunity phase of CRISPR. You need to know that CRISPRs come in many shapes and sizes. They’ve been classified, now, in two classes: class 1 and class 2. Class 1 systems have an effector complex of cas protein that consists of many different proteins that you see here, whereas class 2 systems have just one protein that is able to carry out the whole immunity interference step. But then, class 1 and class 2 can then be further subdivided into many, many types and subtypes of CRISPR systems. And here I’m not going to go through all this in detail, but basically they are now around 20 subtypes that have been described of all these CRISPR systems. And very interestingly, they all differ in the way they carry out many aspects of their… of the CRISPR biology, including the RNA processing and the interference. And it’s also important to note that some of these CRISPR systems targets DNA, but others are able to target both DNA and RNA, and some seem to only target RNA. But here, we are mostly going to talk about the type II CRISPR system, and specifically type II-A CRISPR system, because those are the ones that are mostly used right now for all the biotechnological applications. And specifically, the system from Streptococcus pyogenes, which was described in the first place by the group of Emmanuelle Charpentier, is an interesting one. So, it carries this CRISPR array, here, and this set of four cas genes. And cas9 is the only one required for the interference step. Well, the way it works is that the CRISPR is first transcribed into this precursor RNA, but another RNA is also produced by the system. And this RNA, here, the tracrRNA, for transacting CRISPR RNA, carries a sequence that’s homologous to the repeats in the CRISPR array. So it’s able to form this duplex RNA that is recognized by Cas9 and by the host RNase III, and that will process it to form the final guide RNA, also called the CRISPR RNA. So, you form this nucleoprotein complex that basically acts as a surveillance complex in the cell. It’s constantly scanning the DNA present in the cell. And the way it does that is by first looking for a small… a small sequence motif known as a protospacer adjacent motif. In the case of this specific Cas9, the motif is just a GG dinucleotide. When it finds this GG dinucleotide, it then starts to unwind the DNA and try to pair it with a CRISPR RNA. If that can be done successfully, it will trigger a conformational shift in the Cas9 protein, bringing two catalytic domains in contact with the DNA and introducing a double-strand break And this leads to the target destruction. So, all… this system has been described, mostly in the papers listed here, and that you can have a look at. And this discovery led to a lot of interest, because being able to target a nuclease to a specific position by just changing the sequence of a small RNA, here, is of course something that’s very interesting for many biotechnological applications. And the year, basically, after the system was really described in vitro, so, in 2013, was what people called CRISPR craze. And many, many papers came out describing the use of this Cas9 nuclease to modify genomes. And so, I was happy to contribute to the work that was done to demonstrate this in bacterial genomes, Streptococcus pneumoniae and E. coli, but you see this whole list of papers where they did that in human cells, mice, frog, zebrafish, and the list goes on. So, now you basically have CRISPR tools for all your favorite organisms. The way these technologies work always basically starts by reprogramming the Cas9 nuclease, but with a guide RNA to target the position of interest. And this will introduce a break in the DNA. And then you can have different possible outcomes. If you provide a template DNA that can be used to repair the break through homologous recombination, you can use this to introduce many types of mutations — point mutations, insertions, deletions. You can very accurately modify the target sequence. If you don’t provide a template for homologous recombination, but if your cells are able to carry out another type of repair pathway that’s called non-homologous end joining, or NHEJ, then this NHEJ pathway can repair the break, but does that in a very error-prone way, very frequently making mistakes — insertions and deletions, indels. And that can be very useful, also, to introduce frameshifts in target genes and knockout genes. Of course, there is a third possible pathway, which is if the cell is not able to repair the break then you would expect the cells to die. And the picture that is starting to emerge in the literature is that these two pathways are really quite efficient in most eukaryotic systems, explaining the success of CRISPR tools to edit human cells, mice cells, and many more. While, in bacteria, Cas9 breaks seem to be very efficient at killing the cells. However, we showed that you can actually still use this as a… as a tool to introduce modifications to bacterial genomes. And just to give you an example of the work that was… that we published in 2013 in E. coli, what we did was to try to modify this rpsL gene. And the way we did it was, first, by putting a plasmid expressing Cas9 in the cell. And then transforming, together in the cell, another plasmid carrying a CRISPR programmed target, this rpsL gene, together with another oligonucleotide that carries a mutation we want to introduce. And we do this transformation not just in any E. coli cell, but in these HME63 cells that express some phage recombinases that will promote the integration of this oligonucleotide at the target position. And the way this works is that the phage recombinases introduce a mutation and the CRISPR system, by introducing the break, is actually going to kill all the cells that did not introduce this mutation. What you see here is that the mutation by itself is introduced at a frequency of around 10^-4 but then, by killing all the cells that do not introduce the mutation, we can select for a population where almost all of the cells carry the desired mutation. So, this tool has been improved upon by many groups and adapted to many bacterial species, but I just want to mention two interesting pieces of work. In this paper published a few months ago by Garst et al, they improve on the technique and design this editing construct, what they call the CREATE cassette, where they program a guide RNA and a template for repair, everything on a small piece of DNA that can fit in an oligonucleotide. And that’s very interesting, because if you can fit everything on an oligonucleotide, then you can make that as a library, generate that on the cheap and obtain a lot of oligos to target, simultaneously, many positions in the genome and create libraries of mutants. And so this is what they did in this paper. That is very interesting. Another interesting study is this work that was done by the group of Jason Chin, where they actually use a strategy to… a CRISPR strategy to introduce many, many mutations in the… in the chromosome of E. coli to basically change the codon usage in the… in the chromosome. And the way they do that is by providing the synthetic DNA that they have modified on a bacterial artificial chromosome. And then they exchange that piece of DNA with the chromosome by introducing four breaks: two breaks on the bacterial artificial chromosome and also two breaks in the Escherichia coli chromosome. And here, too, the phage recombinases are used to promote, then, the integration of this artificial piece of DNA into the chromosome, here. And they show that, like this, you can do very extensive modifications of the bacterial chromosome. So, now I want to talk to you about another application of this Cas9 cleavage, because, if you remember, I told you that Cas9 is actually very efficient at killing bacteria when it cuts in the chromosome. And so, a few years ago, we thought, what if we could use this as an antimicrobial strategy? Could we program Cas9 to target virulence factors, antibiotic resistance genes, and specifically kill the bacteria that carries these genes? And the thing is that, to be able to do that efficiently, you need a good way to deliver this CRISPR system to all the bacteria in a… in a target population. And do that, we decided to co-opt phages. We used the capsid of bacteriophages as vectors, what we call phagemid vectors, that we program to carry Cas9 and a CRISPR that will produce a guide RNA to target an antibiotic resistance gene in the chromosome, here. And you see that, when we do that, what you… you see here is a lawn of Staphylococcus aureus on an agar plate. And then we spot a few microliters of this CRISPR phagemid preparation. And what you see here is that only when the bacteria carries a kanamycin resistance gene in the chromosome, and when we program the CRISPR system to target this kanamycin resistance gene, then we kill the bacteria and prevent the growth of the lawn, here. So, you might wonder, why do we think it’s interesting to have sequence-specific antimicrobials? And we think that this can be very interesting because if you are able to specifically eliminate antibiotic resistant bacteria, or virulent bacteria, then you can take advantage of the competition that exists in a specific niche to occupy this niche and use a nutrient. And this is very different from what happens when you use a classical antibiotic, where you will basically kill most of the bacteria in the niche, and then resistant mutants have a lot of free space to grow and occupy the niche again. Well, to simulate this in a test tube, basically, we did this experiment where we make a co-culture between two different Staphylococci, one that is sensitive to kanamycin and one that is resistant to kanamycin. But we put a GFP plasmid in the resistant one so that we can easily follow both of them in the same culture. And so, then we follow the optical density of the culture — this is the plain curve — and we also follow the fluorescence of the culture — that’s the dashed curve — to know the proportion of kanamycin-resistant to kanamycin-sensitive in the population. So, that’s the control experiment, and you see that about half of the population is the resistant, the bad guys we want to eliminate. Now, if we do another control experiment where we treat the population with kanamycin, you see exactly what we would expect, which is that we kill the sensitive one and all the population is fluorescent and resistance to kanamycin. Now, let’s say we make a slightly better antibiotic choice. We treat with streptomycin. Then, what happens is that we actually kill both bacterial populations, but at some point in time we start to see resistant bacteria appearing, mutants that take over the population. But they do so at the same frequency in the kanamycin resistant and in the kanamycin-sensitive populations. And in the end, we still have half of the population being the bad guys. But if we treat with a CRISPR system that specifically targets the antibiotic resistance gene — this aph gene, this is the purple curve — we see that we don’t recover any fluorescence signal in the end, here. So, it doesn’t mean that we killed all of the target bacteria, but we killed maybe 99% of them. And in the meantime, the sensitive ones, you see that they were able to grow in the population… in the… in the medium, here, occupy the niche, use the nutrients, and they don’t leave any room for the bad guys to come back. So, you see that in a specific situation like this, where we can take advantage of the competition between strains, this strategy can be very efficient at eliminating specific genotypes. We also wondered what happens if we use this to target something that is not in the chromosome of the bacteria, but carried by plasmids. You might know that antibiotic resistant genes and variants factors are actually very frequently carried by plasmids. And what happens is that, if you cut a plasmid with Cas9, you actually don’t kill the bacteria, but you eliminate the plasmid — you cure it from the cells. And here, we did this experiments in this USA300 methicillin-resistant Staphylococci that carries these two plasmids, pUSA01 and pUSA02. And actually, pUSA02, here, carries a tetracycline-resistance gene. And what we can do is that we can program the CRISPR to target these plasmids, either one at a time or we can even put two spacers, here, to target them both simultaneously. And what we saw is that whenever we programmed to target pUSA02, here, or here, together with pUSA01, we basically resensitize the population to treatments with tetracycline. So, by eliminating these plasmids that carry resistance genes, we can resensitize the bacterial population to an antibiotic treatment, which can be very interesting. We also went to do animal experiments. So, this is a simple skin colonization model in the mice, where we shave the back of the mice and recolonize them with a mixture of kanamycin-resistant and kanamycin-sensitive Staphylococci And then we treat with this CRISPR phagemid mixture… preparation that we programmed to target the kanamycin-resistant bacteria. And in this more complex environment of the mice skin, we showed that we were also able to specifically reduce the colonization of the antibiotic-resistant bacteria. So, this is a… this was a promising result that we did also in collaboration with Chad Euler and Vince Fischetti at The Rockefeller University when I was still a postdoc in the lab of Luciano Marraffini at The Rockefeller University. And this work has now been taken up by a startup called the Eligo Bioscience and, as a disclosure of interest, I’m also a shareholder of this company. I want to now switch gears and talk about another very cool application of this Cas9 nuclease. And this is the use of a catalytically dead variant of Cas9, called dCas9. And it’s basically still able to bind target positions in the DNA of the… of the cell, but then doesn’t cut it. And it just sits there. And what we found, and others found, is that it actually binds strongly enough to block transcription of DNA, and can become a very useful tool to do gene silencing. And so this is a figure from a paper we published in 2013, where we programmed Cas9 to target many positions in the promoter or inside a reporter GFP gene. And what we see is that, if we target the promoter of the gene, regardless of the orientation — whether we target the coding strand or the template strand of the gene — we have a very strong repression of the target gene. However, if we target inside the open reading frame of that gene, what we see is that you need to target the coding strand to obtain a good repression. If you target the template strand, you don’t get a very good repression. And we can see what is happening here, really, by doing a simple northern blot to visualize the messenger RNA transcripts. So, this is the full-length transcript of this GFP in the control experiments. Here is what happens if you target the promoter of the… of the gene. And you see that you don’t see the transcript at all, anymore. So, what happens is that you block the initiation of transcription. Here is what happens if you target inside the gene, but in the wrong orientation. And what you see is that you still produce a full-length transcript, which explains why you don’t block the gene expression so well. But you do see that you start to form a shorter transcript. And the size of that transcript matches exactly what would happen if Cas9… if the RNA polymerase bumps into Cas9, falls off the DNA, and this leads to this interrupted transcript. And now, if you target inside the gene but in the right orientation, so, targeting the coding strand, here you don’t see the full-length transcript at all, anymore. You just see these shorter transcripts, also with the exact size that is expected. There is something else that you can do with this dCas9 repression, which is try to fine-tune, exactly, by how much you want, to reduce the expression of a gene. And this is recent… recent work that we’ve done in my group to try to really understand how this happens, and compare two ways of doing this. And one way is to keep a high concentration of dCas9, but then introduce mismatches between the guide RNA and the target sequence. And by introducing more and more mismatches, we can basically reduce the repression obtained of the target gene. Another way of doing it would be to play with the concentration of the Cas9 protein. So, here we have the cas9 gene under the control of a pTet promoter, and so by playing with the inducer concentration we can obtain different concentrations, that lead to different repression levels of the target gene. And what we saw that was very interesting to us is that these two mechanisms by which we can fine-tune repression behave very differently in terms of the noise they generate in the expression of the target gene. And this is what you see here. By doing basically flow cytometry or microscopy we can look at repression happening at the individual cell level, and look at how noisy, in a given population, the expression becomes. And what we see is that when we play with the concentration of dCas9 we get a very noisy repression of the target gene. But when we keep a high concentration but play with the number of mismatches between the guide and the target, we basically don’t introduce any further noise to the gene expression than its natural noise. So, this is a very interesting property to study genetic systems. We were quite interested by this effect and wondered, how does this happen? How can we explain that we have so little noise when we reduce the complementarity of the… between the guide and the target? And there are basically two models that could explain this. You could think that by reducing the complementarity you would basically have an effect on the attachment rate or the detachment rate of Cas9 to DNA. And when Cas9 is bound it could block, but if it’s not bound then it would let the RNA polymerase through. But there is another model, which we call the kick-out model, according to which maybe Cas9 is still always bound to the DNA, but the number of mismatches we’ve introduced actually will affect the probability that the RNA polymerase will kick out dCas9 from the gene. A way to tell these two hypotheses apart is we can make the prediction that, in this model, if we increase the concentration of the active dCas9 complex, then we should have a better repression of the target gene. But for this model, playing with the concentration of dCas9 should not have any effect on the target for gene repression, as long as dCas9 is saturating. And so we showed that, indeed, the concentration of dCas9 did not have any effect on the target gene expression, whether you have full complementarity or whether you only have 11 bases of complementarity between the guide and the target. And the way we did that was by using a decoy spacer. We had CRISPR arrays with two spacers. Either there were two identical spacers, or we had one spacer targeting the gene and a decoy spacer that doesn’t target anything. And this is a genetic trick we used to basically have only half of the dCas9 cell being active dCas9, able to recognize the target. And we see that when we do so we don’t change at all the repression level of the target gene. So this, to us, was a demonstration that this kick-out model is correct and is able to explain also why we don’t introduce noise when we block gene expression by specific amounts using this strategy. So, we’ve formulated all this into a biophysical model. I’m not going too much into the detail, here. This was done in collaboration with the group of Sven van Teeffelen at the Institut Pasteur. And encourage you to have a look at these bioRxiv preprints that we hope to also publish soon in a peer-reviewed journal. What’s interesting, also, with this dCas9 system is that you can easily multiplex it by putting several guide RNAs in the CRISPR array. And as a proof of concept, we introduced two reporter genes, a GFP and an mCherry, in the chromosome of E. coli, and then we made a collection of CRISPR plasmids that target these two genes but with different numbers of mismatches to be able to precisely control the level of expression of these genes. And what you see here is that each color represents a bacterial cell population with specific guide RNAs, here. And we can nicely obtain basically every level of repression of each gene, and every combination of these different levels. And we think that this could become a very useful tool to study stoichiometry in genetic systems. So finally, I want to talk about a last use of these CRISPR techniques, and it’s the ability to use them in high-throughput screens. And you can do that by making what we call an arrayed screen, where we have individual guide RNA sequences in different, for instance, wells on the plate. Or you could do pooled screens, where you have a population of cells that is all mixed together, but each cell in the population carries a different guide RNA. And so, very recently, there were two excellent papers showing arrayed CRISPR screens performed in Bacillus subtilis, for this paper by Peters et al, and also in a Streptococcus pneumoniae in this paper by Liu et al. And as this shows, these papers are both very good demonstrations of the power of these dCas9 tools to basically identify drug targets, describe genetic networks, look at cell morphology, etc. And it’s also interesting that with this you can start to study, in detail, essential genes, because you can induce the expression of dCas9 whenever you want, to start to block the expression of an essential gene and see what happens to the cells. So in my lab, we’ve been focused lately on pooled CRISPR screens, where basically we can synthesize very large libraries of guide RNAs, 10^5 guide RNAs, that will target many positions in the chromosome of the Escherichia coli bacterium. And we can then clone these guide RNAs on a collection of plasmids that we introduce into the cells. And the experiment, then, is just to induce the expression of this dCas9 gene. And the readout of the experiment is done by sequencing this collection of guide RNAs, both before the experiments and then after the experiment. And by looking at how the proportion of guide RNAs changes during the course of the experiment, we can compute a fitness effect of each guide RNA in the population on the target cell. And I believe that this can lead to very rich data sets that I think will be very useful in the future to decipher bacterial physiology and genetic interaction networks. And just to give you a small taste of the type of results that you get with it, this is just showing you very raw data of the number of reads that you obtain for each guide RNA in this pool population. And what you see is that this is a control experiment, and the number of reads you get before the experiments versus the number of reads you get after the experiment. And what you see here is that, basically, when you don’t induce dCas9 not much changes. You have the same number of reads before the experiment and after the experiment for each guide RNA. But now, if you induce the expression of dCas9, you see that you start to have a lot of guide RNAs that start to go down, that are depleted from the libraries, and other guide RNAs can be enriched in the libraries. And so what this tells us is that these guide RNAs, by blocking the expression of specific genes, they produce a fitness defect in these specific experimental conditions, while those will give a fitness advantage to the bacteria. And so, we are now describing the properties of the screens in more detail, and hopefully will publish these results soon. So, with this, I’d like to thank my team at the Pasteur Institut. And specifically, today, I’ve shown you the work of a great postdoc, Lun Cui, and a very great, also, PhD student, Antoine Vigouroux. I want to mention, also a lot of the work that I showed was done during my postdoc at The Rockefeller University with Luciano Marraffini. And on this dCas9 characterization, its behavior at the single-cell level, was done in collaboration with the lab of Sven van Teeffelen at the Institut Pasteur. And of course it’s important to acknowledge the people who pay the bills. Thank you very much.