Scientist Stories: David Liu, Inventing CRISPR Base and Prime Editing

Hello everyone, and welcome to this webinar co-produced by The CRISPR Journal and Genetic Engineering and Biotechnology News. Our special guest today is Professor David Liu, who will be talking about precision genome editing without double-strand breaks. Our thanks to TriLink Biotechnologies for sponsoring today's webcast. I'm your host, Kevin Davies, Executive Editor for The CRISPR Journal., and the author of the book, Editing Humanity: The CRISPR Revolution and the New Era of Genome Editing, which includes a chapter on the cool technologies you'll be hearing about in today's program. Our guest speaker today is Professor David Liu, a chemist by training. David's lab is the driving force behind the development of base editing and more recently Prime Editing. In addition to his prestigious academic titles shown here, he's also the co-founder of several gene editing companies, including Editas Medicine, Beam Therapeutics, and Prime Medicine. Following David's presentation, which will run for about 45 minutes, we'll also hear some remarks from Jessica Madigan. Jessica, also a chemist by training, has risen up the ranks of TriLink over the past decade to become their Director of Business Development. Before I hand over to David, please enter your questions for the Q&A session at the end of today's webcast. Simply hit the Ask a Question button at the top of the screen and submit. All right, let's get today's show underway. David, thank you so much for joining us. We are eagerly looking forward to your presentation.

On behalf of a large team of students, postdocs, and collaborators, it's a pleasure to give this GEN webinar on precision gene editing without double-strand DNA breaks. I'll begin with a roadmap of 4 topics I'll cover in this talk. I'll start with an overview of base editing, a general and precise gene editing technology that does not require double-strand DNA breaks. Then I'll describe how a new class of base editors has enabled precision gene editing in an important part of the cell that cannot be edited with CRISPR. Third, I'll share newly published work that exemplifies how we can use base editing to treat a serious genetic disease in an animal. And finally, I'll present prime editing, a highly versatile new editing technology that can perform true search and replace gene editing in living systems. Human gene variants that cause genetic diseases come in many varieties. They include the transition point mutations, shown here in blue, which cause conditions such as progeria, the rapid aging disease. The transversion point mutations, represented by the red wedge of this pie chart, cause conditions including sickle cell disease. And small insertions and small deletions cause diseases such as Tay-Sachs disease and cystic fibrosis, respectively. A long-standing goal of the life sciences has therefore been to develop the ability to install or correct all types of pathogenic mutations so we can study or treat the broadest possible range of the resulting diseases. Programmable nucleases such as zinc finger nucleases, talons, and CRISPR-Cas9, recognized by last year's Nobel Prize in Chemistry to Emmanuelle Charpentier and Jennifer Doudna, initiated the modern era of gene editing by enabling targeted gene disruption. For most genetic diseases, however, precise target gene correction rather than gene disruption is needed to benefit patients. Using programmable nucleases to precisely correct mutated genes in most therapeutic settings remains very challenging because in mammalian cells the primary response to a double-strand DNA break is to initiate end joining repair processes that result in uncontrolled mixtures of insertions and deletions, collectively called indels, as well as other cellular responses such as chromosomal translocations, or p53 activation. In the presence of a donor DNA template, seminal work from Maria Jason and others showed that a double-strand break can also stimulate homology-directed repair, or HDR. Unfortunately, the efficiency of HDR in most cell types, as well as the challenges associated with donor DNA template delivery, have greatly limited the therapeutic relevance of HDR. As a result, as of February 2021, none of the 43 CRISPR clinical trials that were underway using programmable nucleases are trying to correct a pathogenic mutation. Instead, current therapeutic CRISPR clinical trials overwhelmingly are focused on the small fraction of targets for which gene disruption through indel formation can be therapeutic, such as the disruption of genes that silence fetal hemoglobin to treat blood disorders caused by mutations in adult hemoglobin genes. To provide one potential solution to this challenge, we developed base editors. Base editors use the targeting mechanism of programmable DNA binding proteins to engage a desired DNA sequence, but instead of cutting the DNA, they use deaminase enzymes from nature or those evolved in our laboratory to directly convert one target base into another, and then they guide the cell through DNA repair processes to make this conversion permanent on both DNA strands. Cytosine base editors convert C to T or G to A, while adenine base editors convert A to G or T to C. We reported the first base editors 4 years ago, which is ancient history by the standards of the genome editing field, so I'll only briefly summarize how we developed them. Cytosine base editors are fusions of 3 proteins. First, a catalytically impaired Cas9, shown in blue, which cannot cut the DNA double helix, but retains its ability to bind DNA in a guide RNA programmed manner. Second, a cytidine deaminase enzyme shown in red, which catalyzes the hydrolytic deamination of cytosine to uracil in single-stranded DNA. And third, a uracil glycosylase inhibitor in purple, which protects the uracil from removal by the cell. In the presence of target DNA, The Cas9 domain opens up the target DNA site and pairs one DNA strand with the guide RNA. The unpaired DNA forms a single-stranded bubble within which all cytosines within about a 5-base window are deaminated into uracil, creating a U-G mispair. The uracil glycosylase inhibitor domain protects the uracil from base excision repair. To trick the cell into replacing the G on the opposite strand with an A, the most efficient base editors also nick the unedited G-containing strand, which stimulates the cell into remaking that strand using the uracil-containing strand as a template. The result is conversion of a targeted CG base pair into a UA or TA base pair. In our original report in 2016, we showed that BE1, the simple fusion of a dead Cas9 with a cytidine deaminase, gave low but detectable editing efficiencies. BE2, which added the UGI domain, improved editing efficiency substantially, and BE3, which added the nicking feature to BE2, edited the most efficiently of the three original cytosine base editors. By converting targeted CG base pairs to TA, cytosine base editors can correct up to 14% of known pathogenic mutations shown here in blue. Correcting the lion's share of disease-associated point mutations, however, requires conversion of an AT base pair to a GC base pair, the opposite of what a cytosine base editor does. In 2017, we reported the development of an adenine base editor which accomplishes this conversion. ABEs catalyze the hydrolytic deamination of adenosine yielding inosine, which is read as G during transcription or DNA replication. So in principle, developing an ABE should be achievable simply by replacing the cytidine deaminase in BE3 with a deoxyadenosine deaminase. The ABE would open up target DNA, deaminate A to I, then nick the non-edited strand containing the T, tricking the cell's mismatch machinery into replacing it with a C as it remakes that strand. The result would be conversion of an AT base pair to a GC. The big problem with this idea, however, is that there is no known natural enzyme that deaminates adenine in DNA. To overcome this problem, we used laboratory evolution to evolve the first enzyme that deaminates deoxyadenosine. We created a Darwinian selection system that rewards the deamination of adenine in DNA with the restoration of antibiotic resistance in bacteria. Then made large libraries of mutant TadA tRNA deaminase genes. After 7 consecutive rounds of mutation selection and characterization in mammalian cells, we succeeded in evolving ABE7.10, which acquired 14 mutations that together allow it to deaminate DNA rather than tRNA, and that also generalize the enzyme's sequence tolerance to accept adenines in a wide variety of sequence contexts, rather than the UAC context favored by the wild-type tRNA deaminase. The result was the first editing agent capable of installing targeted AT to GC edits in human cells. Base editing has proven quite general and can function efficiently even to edit post-mitotic primary cells in vitro and in vivo. Shown here is the use of an ABE or a CBE to edit post-mitotic primary cells in the inner ear at the top, and primary cortical neurons in mice at the bottom. All possible types of off-target base editing have now been studied in depth and minimized for both CBEs and ABEs by labs including Keith Johns, Hui Yan's, Kaisha Gao's, Nicole Godelis', and our own. Cas-dependent off-target DNA editing arises from the fact that Cas proteins are not perfectly specific for their target DNA sequence, and will occasionally bind sequences that are very similar but not identical to the target. Cas-independent off-target DNA editing and off-target RNA editing arise from the opportunity of deaminase enzymes, which natively process accessible DNA and RNA sequences, to continue to do so at a low level even when tethered to a programmable DNA binding protein. For all of these types of off-target base editing, CBE and ABE variants and delivery methods have been developed that greatly reduce off-target of base editing, often down to background levels observed in untreated control cells. Importantly, the consequences of off-target base editing differ from those of off-target cleavage of DNA by nucleases. Off-target DNA cleavage in the same cell as on-target DNA cleavage results in multiple double-strand DNA breaks. Several laboratories, including those of Kathy Nayakin, David Pellman, Carl June, Paul Thomas, Zhanjun Li, and Alan Bradley have shown that cellular double-strand breaks, especially multiple DNA cuts, promote large deletions, translocations, chromothripsis, and other chromosomal abnormalities. Off-target base editing, in contrast, results in multiple point mutations and thus far has been shown by multiple labs including Zhanjun Li's group to not result in detected chromosomal rearrangements. Since their debut 5 years ago, base editors have been used by many laboratories in a variety of different organisms ranging from bacteria to insects to plants to primates, resulting in hundreds of publications. Addgene, the nonprofit library for DNA blueprints, has sent our base editor constructs more than 10,000 times to fulfill requests from more than 3,000 laboratories around the world. We and other labs have greatly expanded the capabilities of base editors over the past few years. As one recent example, we used our phage-assisted continuous evolution system, or PACE, to rapidly evolve new deoxyadenosine deaminases that greatly enhance ABE activity. From a PACE selection that requires an ABE to fix two premature stop codons in T7 polymerase in order to allow replication of phage encoding evolving ABE variants, we rapidly evolved ABE8e, a remarkably active base editor. In collaboration with Jennifer Doudna's group, we found that ABE8e is about 600 to 1,100 times faster than previously reported ABE variants such as ABE7.10. ABE8e's higher activity enables base editing levels in well-transfected cells to approach 100%. A three-way collaboration with Jennifer's lab and Peter Beale's group revealed the beautiful cryo-EM structure of ABE8e, showing in molecular detail how the enzyme has evolved to accept DNA and to acquire much faster kinetics. We have also used the PACE system to rapidly evolve the other domain of base editors, the CRISPR-Cas domain, to be able to target previously inaccessible PAM sequences, thereby expanding the targeting scope of base editors. Base editors are especially PAM dependent because their precision arises from a small editing window, canonically about 5 nucleotides wide, within which the target nucleotide must be positioned. This requirement means that for canonical base editors, a PAM must exist approximately 15 ± 2 bases from the target nucleotide. Limiting the PAMs for base editing to the NGG PAM of spCas9 means only about one quarter of all known pathogenic transition point mutations, can be connect— corrected by base editing. Using a PACE circuit that selects spCas9 mutants that can bind PAMs of our choosing, we evolved new Cas9 variants that can recognize PAM sequences other than NGG. Using this PACE system, we evolved 3 new families of Cas9 variants through hundreds of generations of evolution, that collectively recognize the vast majority of NR PAM space, where R is an A or a G. Together, these PAM variants in principle allow about 95% of known pathogenic transition point mutations to now be installed or corrected with an ABE or a CBE. Integrating ABE8e with the new NR PAM Cas variants enables editing efficiencies and precise allele access that were really not possible before. For example, fine-tuning the positioning of an ABEAD with an evolved CACC PAM variant enabled us, in collaboration with Mitch Weiss's lab, to achieve more than 80% correction of the hemoglobin mutation that is the most common cause of sickle cell disease into a non-pathogenic variant called hemoglobin Makassar in primary human CD34-positive cells from a sickle cell patient, and this editing was done with minimal indels. This level of editing performance has exciting clinical applications for the treatment of globinopathies. Hundreds of cytosine and adenine base editors have now been published that collectively use more than 70 deaminases, more than 40 Cas protein variants, and more than 25 different architectures. This dizzying array of base editor choices, coupled with the lack of a systematic and comprehensive understanding of base editor characteristics, can make it difficult to choose the best base editor to achieve the desired edit. To begin to address this issue, we recently reported a machine learning model trained on experimental data from editing each of 38,000 genetically integrated target sites with each of 11 different CBEs and ABEs. We used the resulting treasure trove of data, a tiny summary of which is shown on the left, to train machine learning models that accurately predict base editing efficiency and product distributions for the most commonly used base editors. We called the resulting suite of machine learning models Beehive, and here you can see predicted versus experimentally observed editing efficiencies of using 4 different base editors to correct to wild type 7,000 to 10,000 pathogenic point mutations, A public website that allows researchers to easily use BEHIVE, CRISPR-BEHIVE.design, is now live. It allows you to simply paste in a genomic DNA sequence, the guide RNA target sequence you propose to use, and one of 11 CBE or ABE variants, and then will output the predicted editing efficiencies with an estimated abundance for each genotype, and will also translate protein coding gene products so you can easily see how each base editor is predicted to change the sequence of your gene and protein. One of the few important regions of biological DNA space that's thus far untouched by precision genome editing is mitochondrial DNA. Mutations in mitochondrial DNA cause many genetic diseases and collectively affect about 1 in 5,000 people. The vast majority of pathogenic mitochondrial DNA mutations are transition point mutations that in principle could be installed or corrected with a cytosine or adenine base editor. However, CRISPR has not been successfully used in the mitochondria for the simple reason that no mechanisms to transport guide RNAs into mitochondria are known, perhaps because in mammalian cells the mitochondrial genome is thought to contain all of the small RNAs needed for mitochondrial DNA gene expression. As a result, the only form of targeted mitochondrial DNA manipulation reported to date is targeted mitochondrial DNA destruction, which can be achieved with RNA-free programmable nucleases such as zinc finger nucleases or TALENs. In a three-way collaboration with Joseph Mogus's lab and Vamsi Motha's lab, We recently developed the first precision editing capability for mitochondrial DNA. Joseph Mogus's lab discovered an interbacterial toxin called DddA that is predicted to be a cytidine deaminase, but when they characterized the toxin biochemically, they found that it does not deaminate single-stranded DNA or RNA like all previously described cytidine deaminases., but instead it deaminates cytosines in double-stranded DNA. This unique ability of DDDA deaminase to operate on double-stranded DNA raises the possibility that we might be able to develop CRISPR-free all-protein-based editors that could enable the first precision editing of mitochondrial DNA, whereas all previous CBEs and ABEs deaminate single-stranded DNA, and therefore require CRISPR to generate a single-stranded DNA bubble at the target site, the ability of DddA to deaminate double-stranded DNA raises the possibility of using programmable DNA binding proteins such as zinc finger arrays or TALE arrays to engage target DNA and recruit a fused DddA to the target site without requiring a CRISPR protein or any guide RNA. To summarize a long story, we made non-toxic versions of this highly toxic protein by splitting it into two pieces, then linked each half of split DddA to a tail repeat array programmed to bind adjacent DNA sequences either in the nucleus or in the mitochondria. After answering many questions that defined an optimal CRISPR-free base editor architecture, We identified the one shown here as enabling efficient targeted conversion of CG base pairs to TA base pairs in mitochondria or in nuclei. The optimized DdCBE, as we call it, consists of a mitochondrial targeting sequence followed by a tail repeat array, a short linker, one half of the DddA deaminase toxin, and one UGI domain.. A complete DdCBE consists of two such proteins, which together assemble on target DNA to bring together two inactive halves of Ddda into an active deaminase, triggered by the high effective molarity of each half when bound to adjacent DNA sites. DdCBEs can base edit nuclear DNA efficiently if you replace the mitochondrial targeting sequence with a nuclear localization sequence. In mitochondria, ddcBEs edit very cleanly with no detectable indels, probably due to a lack of double-strand break repair pathways in mitochondria. Treating human cells with these ddcBEs showed that we could efficiently use them to install CG to TA point mutations for the first time at a variety of mitochondrial DNA sites with 5 to 50% efficiency depending on the site, on the position of the C within the editing window between the tail binding sites, and on the orientation of the fusion proteins. Once again, we observed no detectable indel byproducts from base editing in the mitochondria at any of these sites. We assessed off-target editing by these new CRISPR-free mitochondrial base editors by sequencing at a very high depth the entire mitochondrial genome in treated or control cells. We found that off-target editing was quite low in general, with substantial off-target editing observed for only one variant shown in the red bar here, which used a known N-terminal tail domain that binds DNA promiscuously, but for the other DdCBEs, ratios of on-target editing to off-target editing were very high. DdCBEs can be used to model pathogenic mitochondrial DNA mutations in mammalian cells, which could be especially useful to generate or correct cell or animal models of mitochondrial genetic diseases. For example, we showed that these new base editors can install a known pathogenic mutation mutation in ND4, a mitochondrial gene that encodes part of Complex I, the first enzyme in the electron transport chain of aerobic respiration. Installing this mutation using DDCBE indeed lowers oxidative phosphorylation and respiration rates, and also selectively lowers Complex I abundance and activity, but not those of other complexes in the electron transport chain. Our lab and others have begun to use base editors to address animal models of human genetic diseases. In an exciting collaboration with the labs of Francis Collins, Mike Erdos, Jonathan Brown, Can Cao, Charles Lin, and Leslie Gordon, we sought to use an adenine base editor to directly reverse back to wild-type the mutation that causes the rapid aging disease progeria. In 90% of progeria patients, a single C-to-T mutation in one copy of lamin A creates an early splice site in an exon that results in the loss of 50 amino acids of coding sequence that includes a proteolytic cleavage site. The resulting progerin protein remains farnesylated, damages nuclear membranes, and causes early death of around 14 years of age, typically from cardiovascular failure. Despite a very passionate community of patients and researchers who have attacked this disease using many approaches, including a recently approved farnesyl transferase small molecule inhibitor, current therapies have not yet greatly extended patient lifespan. Because this mutation is dominant, gene complementation to add additional copies of wild-type lamin A does not rescue progeria, and even approaches to use nucleases to disrupt the mutated allele are challenged by the fact that the mutation differs from the healthy copy of the allele by only a single base pair, and occurs at a cryptic splice site that makes the effects of uncontrolled indel byproducts which result from double strand breaks difficult to predict. We envisioned using an adenine base editor to directly correct back to wild type the lamin A mutation that causes progeria. We used lentivirus to deliver ABE7.10 targeted with an engineered PAM Cas9 variant developed by Keith Johns' lab to correct fibroblasts from two progeria patients and observed about 90% correction back to wild-type lamin A with minimal indels.. Compared to untreated cells, we observed a significant reduction in the amount of mis-spliced progerin transcript among treated cells, as well as a large reduction in the presence of the toxic progerin protein in treated cells, which now show a lamin Western blot that resembles samples from their unaffected parent, and both the presence of progerin and the characteristic blebbing of the nuclear envelope are rescued in the treated patient fibroblasts. To assess potential off-target DNA editing from the guide RNA and Cas9 variant used to correct the causative mutation in the patient-derived cells, we used Keith John's CircleSeq method, which is a highly sensitive way of identifying the DNA sites that are engaged by a specific Cas9 protein and guide RNA. We amplified from the genomic DNA of ABE-treated cells the top 32 candidate off-target sites identified by the CircleSeq method, and found that none of them showed off-target editing levels above our detection threshold of 0.1%. We also performed transcriptome-wide off-target RNA editing analysis and observed no significant increase in average A-to-I editing levels among cellular RNAs. These results with cultured patient cells were sufficiently encouraging that we initiated a series of long-term animal studies in collaboration with Francis Collins and Mike Erdos at NIH and Jonathan Brown at Vanderbilt. Since base editors are currently too large to fit into a single AAV vector, we divided the base editors into two proteins, each fused to a protein splicing trans-intine, and packaged each half into separate AAV9 capsids. We then injected the mouse model of human progeria developed by Francis Collins's lab, which contains 2 copies of the mutated human lamin gene with 10^11 to 10^12 total viral genomes per mouse, that is 5 times 10^10 to 5 times 10^11 of each AAV9, either 3 or 14 days after birth. We were pleased to see that a single systemic retroorbital injection of ABE AAV resulted in substantial editing of a variety of organs including the aorta, which often fails as a proximal cause of death in progeria human patients. Indeed, we observed about 25% correction of the mutation in the aorta of people P14-injected mice at 6 months, and about half as much editing in the aorta for the P3-injected mice. This DNA correction was accompanied by few indel byproducts, an advantage of avoiding double-stranded DNA breaks. At the RNA and protein levels, we observed an apparent amplification of this correction, with a dramatic reduction of progerin transcripts and progerian protein levels in most tissues including the heart. That 25% correction at the DNA level could lead to more substantial correction at the RNA and protein levels, suggests that corrected cells may be more active or may be able to replace uncorrected cells in some tissues. We were especially excited when we performed histological studies on the treated mice. A wild-type mouse at 6 months of age shows a healthy aortic cross-section shown here with plenty of smooth muscle nuclei, which appear as purple dots in the upper image, and relatively little adventitia, the stiff tissue that shows up as beige, that accumulates around the aorta in progeria mice and in human patients. Now if you look at a saline-injected progeria mouse you see about a 10-fold average loss of smooth muscle cells and the formation of a thick adventitial layer surrounding the aorta, but the P14-injected ABE-treated progeria mice show almost wild-type-like smooth muscle cell counts and very little adventitia. The P3-injected mice, which were edited in the aorta half as efficiently,, as the p14-injected mice showed an intermediate phenotype. We used immunofluorescence to visualize both progerin, shown in green, and wild-type human lamin, shown in red, in aortal tissue from wild-type C57 black mice from control progeria mice injected with saline at p14., and from progeria mice injected with ABE AAV at P14. Wild-type C57 black mice show no human lamin protein and no progerin as expected, you can see that on the left. Tissue from progeria mice that are only 28 days old, before smooth muscle cells are lost, show smooth muscle nuclei with correctly spliced human lamin protein, and with progerin protein in green, as expected, because these young progeria mice contain two copies of the mutated human progerin gene. At 6 months of age, the saline-injected progeria mice have now lost the vast majority of their vascular smooth muscle cells, as expected, but remarkably, at the same 6 months of age, the ABE-AAV-injected progeria mice show plenty of vascular smooth muscle cells with human lamin, but very few of them contain progerin. These results demonstrate that in vivo correction of the causative human progeria mutation in mice greatly reduces the level of progerin in aortal vascular smooth muscle cells, and instead corrects the progeria mutation to allow these cells to express corrected human lamin. We also performed a longevity study on treated and controlled progeria mice, the lifespan of the less well-edited P3 injected cohort was 1.8-fold longer than their saline-injected counterparts, and remarkably the better edited P14 injected cohort reached a median lifespan of 510 days, which corresponds to the beginning of old age in wild-type normal C57 black mice and represents approximately a 2.5-fold longer lifespan. To our knowledge, this degree of median lifespan extension in an animal model of progeria has never before been achieved. Beyond Kaplan-Meier curves, the preservation of general animal vitality among the treated mice is remarkable. The untreated progeria mice live only about 7 months old, by which time the mice have thinned and whitened coats, pronounced kyphosis or outward curvature of the spine, and very low activity levels as you can see here. Now here is a video of 3 P14-injected ABE-treated females at 11 months old, so far older than any untreated progeria mice can survive. Kyphosis is not pronounced and activity levels are very high. And here are 3 P14-injected ABE-treated males plus 1 wild-type C57 black mouse for comparison, which is the slightly larger mouse, and again you can see that they have healthy coats, no obvious kyphosis, and high activity levels. We're hopeful that these results collectively provide a foundation to bring to patients the benefits from directly correcting the mutation that causes progeria, and we're currently working with our collaborators in coordination with scientists at NIH and Beam Therapeutics to do so. So base editors can correct pathogenic transition mutations in vitro and in vivo, in some cases with dramatic rescue of disease phenotype, but what about the other pathogenic mutations beyond transition point mutations, since these other mutations account for 70% of known pathogenic gene variants, we sought to develop other methods to directly install or correct transversions, deletions, or insertions without requiring double-stranded DNA breaks, or donor DNA templates. At the end of 2019, we reported prime editing, a new mammalian cell gene editing technology, to achieve these goals. Prime editors are fusions of Cas nickases with engineered reverse transcriptases. They use an engineered prime editing guide RNA, or pegRNA for short, which not only specifies the target site for editing, but also encodes the desired edit. Prime editors nick the target DNA site, then use the 3' end of the freshly nicked DNA strand to prime reverse transcription of an extension on the pegRNA that serves as a reverse transcriptase template. The engineered reverse transcriptase domain of the prime editor then copies the desired edit directly onto the target DNA strand, creating a 3' flap that the cell resolves into a heteroduplex containing one edited and one unedited strand. The PE3 system then nicks the non-edited strand to cause the cell to remake the strand using the edited strand as a template, thereby completing the editing of both DNA strands. Since the RT template of the pegRNA is specified entirely by the researcher, one can make virtually any small substitution, insertion, or deletion using this strategy. PE3 shows remarkable versatility in the types of edits that can now be directly installed in DNA in human cells. Shown here are all 24 possible single base substitutions that we made in a stretch of 8 consecutive DNA base pairs in the human genome. Indeed, we can use prime editing to perform targeted transversions, deletions, insertions, and combinations thereof for the first time without requiring double-strand breaks or donor DNA templates. While the data in the previous slide was from HEK293 T cells, the cells in which we developed prime editing, other cell types even including post-mitotic primary cortical neurons can also support prime editing albeit with varying efficiencies that are generally lower than that of HEK293T cells, in collaboration with Britt Adamson's lab at Princeton we recently performed a large-scale screen to identify cellular determinants of prime editing efficiencies with some exciting new insights and improved prime editing systems that we hope to report soon. Previous CRISPR-based editing methods derive DNA specificity primarily from the single DNA guide RNA hybridization event. In contrast, prime editing requires 3 such hybridization events in order for prime editing to take place productively. First, hybridization of the pegRNA to the target DNA site, Second, hybridization of the freshly nicked target DNA strand to the primer binding site of the pegRNA, and third, hybridization of the reverse transcribed 3' flap onto the unedited strand of DNA. Since each of these three hybridization events presents an opportunity for an off-target sequence to be rejected, we speculated that prime editing might offer lower editing levels at off-target DNA sites than Cas9 nuclease. To test this possibility, we examined 16 of the most notorious off-target sites in the human genome of 4 on-target guide RNAs, including the hex site 4 off-target number 3 that is consistently edited even more efficiently than the on-target site. A side-by-side comparison of editing efficiencies of Cas9 nuclease compared with the prime editors at these 16 off-target sites, using the same pegRNAs to program both the nuclease and the prime editors revealed that prime editing efficiencies at these off-target sites were indeed lower in every case than with Cas9 nuclease. In fact, only one of the 16 off-target sites we examined, the notorious hex site 4 off-target number 3, showed more than 1% off-target editing,, and in this case choosing a pegRNA that minimizes 3' flap homology with the off-target site greatly reduced off-target editing back down to only 0.2%. As you have probably gathered from all the data I've presented, base editing and prime editing offer complementary strengths. Base editors have been relentlessly studied and optimized for 4 years now, and have been used in hundreds of publications, they currently offer higher editing efficiencies and lower indel levels than prime editors, at least at target sites that are well suited for base editing. When the target nucleotide does not lie within the base editing window, however, such as C3 shown here, then prime editors offer higher efficiencies than base editors., and because prime editors are immune to bystander editing, they have little difficulty editing, for example, the first and third of a string of 3 consecutive Cs, whereas base editors will edit all the Cs or all the As within the roughly 4 or 5 nucleotide wide base editing window. So base editors offer higher efficiency and fewer indels, while prime editors offer greater versatility precision, and targeting scope. We've started to use prime editing to install or correct a variety of pathogenic mutations. We can efficiently generate homozygous sickle cell mutant cell lines using prime editing, and correct such a cell line by replacing the T that causes most sickle cell disease with a wild-type A. Likewise, we can generate with PE3 a homozygous cell line that contains the 4-base insertion that causes most cases of Tay-Sachs disease, then correct these cells back to wild-type using PE3. We can also install a protective transversion mutation that makes mice and humans resistant to prion disease, and we can insert epitope tags and recombination sites into targeted locations in the human genome for biotechnology applications. Even though we reported prime editing shortly before the pandemic slowed down or shut down many research groups including our own, it has been terrific to see prime editing already result in dozens of publications and preprints in a variety of cell types and organisms including crop plants, organoids, mouse embryos, and postnatal mice, Adgene has distributed our prime editor constructs to fulfill requests from more than 1,000 laboratories around the world. To summarize, base editing enables the 4 transition point mutations to be installed at targeted sites without double-strand breaks or donor DNA templates, and has been advanced through the work of many labs to include developments such as CRISPR-free base editors that enable the first precision editing of mitochondrial DNA in living cells. Base editing coupled with in vivo delivery systems such as the AAV system I described can correct animal models of human genetic diseases. A single injection of ABE AAV into progeria mice rescued the disease at the DNA, RNA, protein, nuclear morphology, aortal histology, lifespan, and animal vitality levels, and prime editing while much newer than base editing establishes how a nicked target DNA site can directly prime DNA polymerization to introduce a wide range of edits into human cells again without requiring double strand breaks or donor DNA, even just 5 years ago The prospect of engineering molecular machines that correct a specific DNA base pair in an animal that causes a devastating genetic disease to alleviate some of the consequences of the disease really seemed like science fiction, so I'm very grateful to a highly talented and dedicated group of students, postdocs, and collaborators who made this work possible. Andrew Anzalone, now at Prime Medicine, led the development of prime editing. Monda Arbab and Max Shen led the development of the Beehive machine learning model that predicts base editing outcomes. Jordan Doman and Aditya Raghuram led the study and development of base editors that show lower off-target editing. Nicole Gudelli, now at Beam Therapeutics, developed the first adenine base editor. Luke Koblan, in collaboration with the labs of Francis Collins, Mike Erdos, Jonathan Brown, Leslie Gordon, and others led our progeria study. Alexis Komor, now a professor at UCSD, developed the first base editor, the CBE. John Levy and Ariel Ye developed the base editing AAV system. Beverly Mock, in collaboration with researchers from Joseph Mogus's lab and Vamsi Motha's lab, led the mitochondrial base editing project. Shannon Miller and Tina Wan, now a professor at the University of Wisconsin, evolved the NR-PAM Cas9 variants. Michelle Richter and Kevin Zhao evolved ABE8e, and in collaboration with the labs of Jennifer Doudna and Peter Beale, characterized this new adenine base editor. Finally, I'm grateful for all of these agencies for their funding, for TriLink for providing reagents,, and I'd be happy to answer any questions.

Thank you. Thank you so much, David. That was a brilliant presentation. And thank you also for staying around for our Q&A discussion, which will begin shortly. A reminder to everyone, please submit your questions for David into the Q&A box, and we'll get to those in a minute.

Before that, I'd now like to welcome TriLink's Jessica Madigan, who will give a short presentation. Hi, Jessica. Thank you to Dr. Liu and to Jen for giving me the opportunity to present today. I'm Jessica Madigan, Director of Business Development at TriLink Biotechnologies. TriLink is a CDMO/CMO that specializes in manufacturing of modified nucleic acids. We offer research, preclinical, and clinical grades. I have about 5 minutes today to give you a quick overview of the advantages of mRNA for mediating efficient editing. These are our forward-looking statements. So what are the benefits of using mRNA to express base editors? By using mRNA, you can benefit from increased specificity due to transient expression. mRNA only needs to be delivered to the cytoplasm. While direct delivery of the RNP complex may sound like the most straightforward option, delivering a large protein across a cell membrane can be extremely challenging, and mRNA has been shown to be more efficient than plasmids. On the left is a visual representation of delivery of the three different approaches. While delivery of a protein has shown to result in rapid onset of gene editing, it must be delivered across two cellular barriers: the cell membrane and the nuclear membrane. Delivery of DNA must also cross two cellular barriers: Additionally, DNA gene must undergo transcription and translation before editing. mRNA only needs to be delivered to the cytoplasm. After translation occurs, the protein is shuttled into the nucleus by the nuclear localization signals and can now edit targeted DNA. On the right is the editing comparison of cells treated with base editor plasmid DNA, mRNA, or protein at one genomic locus in HEK 293 T cells and their on- and off-target editing ratios. In this set of experiments, mRNA yielded the best editing efficiency. Now, when planning to manufacture an mRNA, the important considerations for an ideal base editor include Cap1 structure, low immune stimulation, optimization of untranslated regions, and a poly(A) tail of 80 bases or longer. Cap1. At TriLink, we developed a co-transcriptional Cap analog called CleanCap. It offers improvements over legacy capping technologies as it yields a natural Cap1 structure. Traditional capping analogs yield the Cap0 structure. Cap structures are important in self/non-self recognition. Cap1 structures are found in higher eukaryotic messages. While Cap0 is sufficient for ribosomal recruitment, it can activate the immune system. Cap1 methylation reduces binding to pattern recognition receptors, and IFITZ recognizes non-methylated caps and sequesters them from the translational machinery. Additionally, Cap1 structures are more resistant to decapping enzymes, which results in longer persistence of mRNA in vivo and greater protein translation. CleanCap initiates transcription with a trimer nucleotide as shown here. The structure consists of an N7-methylguanosine followed by 2 bases that are complementary to the first 2 bases of of the mRNA transcript. Another benefit of CleanCap is increased yield from the manufacturing process. CleanCap yields more than 3 times that of ARCA or mCAP, as demonstrated by this capping efficiency assay. CleanCap results in greater than 95% capped product. A major contaminant of in vitro transcription reaction is double-stranded RNA, an unwanted byproduct and highly immune stimulatory. Modifications of mRNA with pseudoU, N1-methylpseudO, or 5-methoxyU have been shown to reduce binding to innate immune sensors in vitro. They also reduce toxicity and prolong expression. It is important to note that when using modified nucleotides, it is recommended to codon optimize the open reading frame to deplete for uridine content. Uridine-depleted 5-methoxyU modified RNAs give higher expression than wild type in cultured cells. In this experiment, adenine base editor mRNAs are compared: wild type without codon optimization, wild type with codon optimization, and 5-methoxy-U modified with codon optimization. Cas9 mRNA served as a control. 5 micrograms of mRNA was transfected. The adenine base editor mRNA with both 5-methoxyuridine modification and uridine depletion showed the highest protein expression in HEK293 T cells by Western blot. I wanted to spend a slide on considerations for guide RNA. TriLink's reputation is built on our expertise in synthesizing difficult, long, and unusual constructs containing extensive modifications. TriLink custom manufactures high-quality guide RNAs for research and GMP. We offer single guide RNAs commonly used for CRISPR gene editing, and we also offer prime editing guide RNAs, which are substantially longer than the standard CRISPR guides. The users design and customize their sequence for editing. TriLink can introduce various modifications through chemical synthesis, and our scales range from 5 micromoles to 5 millimoles. In conclusion, expression of base editors from mRNA is more efficient than plasmid or RNP. CleanCap, a co-transcriptional cap analog, mimics a Cap1 structure and is the superior capping technology Uridine depletion and modified RNAs are used for maximal base editing expression, and TriLink can be your single source for mRNA and guide strands. I'd like to acknowledge the contributors to these slides and thank you for your help, and thank you for your attention today.

Please contact our team to discuss your base editing manufacturing needs. Thank you very much, Jessica. Please stay Submit your questions for our Q&A session, which is about to start. A reminder also that we will be archiving today's webinar so you can view it on demand or share it with colleagues. All right, time to address a few questions.

Just give us a couple of moments to transition into our live Q&A session. All right, hi everybody, Kevin Davis back with you, joined by David Liu and Jessica Madigan is also with us. So we have about 20 minutes, I hope, for questions, and thank you for, for all your questions. They've been absolutely flooding in, and that's because we have more than 2,500 people who signed up for today's webinar and a good a fraction of that are on the call. As David noted before we began, that's 5 times the capacity crowd at a Cold Spring Harbor conference, so thank you everybody. David, first off, before we turn to the audience questions, it's been a great week for you. Congratulations on your election to the National Academy of Sciences.

What does that distinction mean to you? Thank you. Well, I think the National Academy is a special honor because of course it's election by scientific peers. So that's why it's especially meaningful. It's, you know, really like I think any recognition of our work, it just reminds me how fortunate I am to have the opportunity to work with so many incredibly hardworking and talented students and postdocs, collaborators, and colleagues.

So really it's an award that I think comments more on their achievements than on mine. Very generous of you. All right, let's dive in. We have more questions than we will possibly be able to get to, but let's do our best. Ven Gopal asked, does the chromatin structure or epigenetic changes such as methylation obviously interfere with base editing, and are the latest versions of your base editing plasmids available for research?

Uh, yes, so the last part is the easiest answer. We generally deposit everything that we publish that we think could be useful to people to Addgene, as long as Addgene allows that kind of material to be distributed. So they won't distribute bacteriophage, for example, but they, they will distribute plasmids. So we, we tried to deposit in advance so that by the time it works its way through Addgene's quality control process and resequencing, that, that it's roughly coincident with when the papers are released. Sometimes we get it right, and sometimes it's a little bit delayed, but everything should be available from Addgene that's been published that's sort of a primary use. And certainly if you're having a hard time getting materials, you can, you can let me know. And then with respect to chromatin, I think a universal feature of, or aspect of all gene editing agents, is that they can be sensitive to the chromatin state of the target sequence. So I'd say there are some both published and unpublished data showing that highly chromatinized targets can really suffer in terms of their editing efficiency. So this is a problem that's been noted by a number of labs in, including our own, and it's just something to keep an eye on. I'm actually somewhat surprised that at least certain gene editing agents, including many of the Cas9 variants that are commonly used, seem to have the ability to access even partially chromatinized target sites. That is, editing efficiencies are usually not too bad even if the target according to, say, ATAC-seq data is, you know, moderately chromatinized. But I think that is an opportunity for future improvement and development and understanding for all these classes of gene editing agents. And then in terms of methylation state, the third question, which is I guess the second question asked, different deaminases will accept, or that is, will tolerate or not tolerate methylated bases to different extents, and I think An overall survey of the dozens of different deaminases that have now been used in base editors suggests that the, the few studies that have explicitly looked at different rates of methylation versus methylated versus unmethylated base editing, that the methylation preferences of the deaminases seem to be reflected in the methylation tolerances of the resulting base editors.

Great, thank you, David. Let's move on to another question. John is from Editas, a company of course you know quite well. The field has now seen CRISPR editing, base editing, and prime editing reduced to practice in cell culture and animal models across many laboratories. What are your thoughts, David, on the potential for CRISPR-guided transposases and recombinases to be added to the gene editing toolbox in the near future?

Why do you think the but these have been harder to reduce to practice in eukaryotic cells to date? Yeah, that's a good question. John, in the interest of full disclosure, was a former postdoc in my lab. Good question, John. So, you know, our lab and others have tried to make CRISPR recombinases, and we were partially successful. We reported a system we called Rec-Cas9, where basically a dead Cas9 was fused to an engineered gene beta recombinase that Carlos Barbas's lab developed, and it, it showed gut RNA dependence and, and some programmability, but the recombinase module we used still retains a fair amount of sequence requirement., and I think that's one of the limitations behind efforts to make fully reprogrammable recombinases is there isn't a generic recombinase domain that in a modular manner, to my knowledge, can be redirected fully by a programmable DNA binding domain. The transphosphate systems are, are really interesting. Some of their performances in bacteria have been just awesome to see. These are systems that my colleague Fang Zhang developed and a different system that Sam Sternberg at Columbia developed, and they each have unique strengths. They, as the question implies, I think have not— it has not yet been reported, at least, data showing that those systems can work in mammalian cells. But I know many, many labs are interested in that possibility, and all I can say is, you know, it's hard to bet against this field.

So I would hope and expect that in the coming years we will see some translation of those systems into mammalian cells. Indeed.

Jason asks, have you validated the absence of double-stranded breaks using single-cell sequencing? Um, I'm not quite sure what the question is asking, but, uh, there are a number of metrics one can use to assess the presence of double-stranded breaks. There are methods, of course, that can intercept the double-stranded breaks either in vitro or in cells or even in vivo, and then sequence, attach tags to those double-stranded breaks and sequence them. A variety of researchers have used those methods to characterize CRISPR nucleases, TALENs, zinc finger nucleases, and also to check frequencies of double-stranded breaks from various forms of base editors. And I think in unpublished work, at least some of those methods have been used in to characterize prime editors with the expected outcomes. You know, double-stranded breaks are rare but measurable occurrences in native cells and organisms, so there isn't going to be a native biological system that is completely free of double-stranded breaks, at least so long as there's cosmic rays and, you know, rocks and other sources of ionizing radiation, as well as other sources of DNA damage that can lead to double-stranded breaks. But I think the, the main upside of these methods that don't require making double-stranded breaks is you can really minimize the frequency of additional double-strand breaks that you're adding to the genome, recognizing that the presence of double-stranded breaks leads to these uncontrolled mixtures of products. Namely indels, which can be very useful if you're trying to disrupt a gene, but can be undesired if you're trying to generate a homogeneous genotype instead of a mixture of products after editing, or if you're trying to do the kinds of precise corrections that, that I showed in my talk.

Right, well, there is— it's a bit of a tough crowd out there, David, because Ali asks, where do you see the next breakthrough in gene editing? As if it's a guessing so he is unimpressed with everything that you've presented over the last 50 minutes. Would it be a smaller Cas enzyme or a new delivery system perhaps?

Yeah, I mean, there are lots of directions that the field still needs to go to fully realize the, you know, incredible potential both for basic science and for therapeutics. Of gene editing. It is— Hollywood, you know, used to teach people incorrectly, I think, since the 1980s, that we can sort of make arbitrary changes in DNA in living organisms. Of course, that hasn't been true until somewhat recently, but even now we can't quite make every possible change we'd like to make in any type of cell, in any type of of organism, especially in vivo. So there are still, there are still many opportunities and, and I'd say important advances that are needed to really achieve that goal of being able to make any kind of change in any kind of, of tissue in a living animal or in a human patient, respectively. You know, certainly delivery is a very common answer to this type of question, for a good reason. It It is a bottleneck through which many gene editing applications must pass. And I'd say, as John's question already implied, the insertion of gene-sized fragments into targeted sites in the human genome would be incredibly useful both for basic science, for biotechnology, and for therapeutics.

That's in part why there's so much interest in these transposases. There was news this morning, David, you may have seen, of in a clinical trial using for traditional gene therapy, not, not a gene editing method, but an adverse, a serious adverse event resulting presumably, but I don't know if it's proven, from AAV delivery. AAV, you touched on delivery methods incorporating AAV, typically thought to be a very safe viral vector.

Do you, are you, or is the field actively looking at non-viral delivery systems for base and prime editors? Yeah, absolutely. And, you know, I'd say that mRNA delivery, RNP delivery, that is delivering the protein-RNA complexes, those are when possible always preferred because the DNA specificity is maximized that way. That is, the off-target editing is minimized. Simply because those agents don't last long in the cell and the on-targets tend to be edited much faster than the off-targets. Viral delivery can be very useful, AAV in particular, because it remains one of the most efficacious ways to access certain cell types, at least without using integrating viruses or viruses that are more prone to integration. But whenever you introduce DNA into cells, there is a possibility of DNA integration. That integration in rare events can occur at sites that might lead to adverse effects, you know, so certainly I think all of the things being equal, you would rather use a non-viral method than a viral method, but unfortunately the landscape is not equal. You can't, there are trade-offs with any delivery method, and therefore I think AAV is going to remain an important therapeutic in vivo delivery method for macromolecular agents, including gene editing agents, for some time. But the AAV space and the viral vector delivery method space in general is advancing, and in fact I'd say it's advancing at an accelerating pace, much like the gene editing field has been advancing recently. So, you know, the work of lots of researchers including Luke Vandenberghe and the folks at Voyager, Ben Deverman, are constantly improving and I'd say making more safe these AAV vectors. And then a number of creative twists on viral vectors including the novel viruses and novel ways of using viruses are also starting to surface that have me quite excited about the future of delivery.

Great, actually this plays in very nicely to our next question. We received a couple of questions about your progeria portion of the presentation. Patrick Harrison is in Ireland. Hi Patrick, hopefully he's nursing a Guinness this time of night. Can you get the same results in progeria mice with much lower doses of AAV? What's the lowest dose, David, that gives a substantial physiological benefit?

I suppose the real question is, what's the multiplicity— what multiplicity of infection is efficacious? Right, yeah, it's a great question, one we are working on answering now with the more finely graded dose response set of experiments, as well as treating mice of many different ages. We're doing that in parallel while we take steps to advance the work that I presented towards potential clinical application, simply because the need is so great and the rescue, even in the form that, that I presented, was, was so strong. The short answer is we don't know. Even the the tenfold lower dosing that, that I showed in the progeria slides isn't a perfectly controlled experiment because those mice received tenfold lower AAV dose but were dosed much, much younger at P3, which corresponds roughly to the first, you know, to a 1-year-old toddler in terms of development at least. So that's quite different than the the 10-fold higher AAV dosing, which we did at P14, which corresponds in maturation level roughly to a 5-year-old human. And the receptors that mediate AAV entry into cells may not have been expressed optimally at P3 in those mice. That's been an observation of certain kinds of viruses that show an age-dependent tropism or an age-dependent transduction efficiency. So I guess all of this is a long-winded way of saying I don't really know the answer to your question, but it's an important question and it's one that we're actively trying to answer now.

Yeah, you presented some work, David, just going back to base editing on sickle cell disease early on, and I did want to give a quick shout out to the COVID feature article on the latest issue of the CRISPR Journal, which is from your colleagues at Beam Therapeutics, in which they introduced the concept and the first designs of inlaid base editors, where the architecture of the base editor as you presented it has been somewhat rearranged to provide greater editing flexibility, which they use to introduce the same variant that you briefly touched on the HB macaca variant, which is best we can tell is a benign variant. So if you can't switch sickle cell mutation to the wild type allele, then this would seem to be a perfectly acceptable alternative.

How useful do you think these IBEs and other re-architected base editors will be in expanding the base editing toolbox? Right, that's a great question. The work that they presented in that recent paper showing how you could actually integrate a deaminase domain into the architecture of CRISPR-Cas9 was terrific work. It, it showed remarkably good performance. It offered advantages in terms of editing performance and size. And, and while a couple of their labs had had reported efforts to integrate deaminases into CRISPR-Cas proteins. I think the results they reported by far had really taken it the farthest and yielded the best results. And, you know, with respect to the, the, the actual edit you referred to, I'm quite excited about the prospects of directly converting this pathogenic mutation, this glutamate to valine A to T point mutation that underlies almost all cases of the most deadly genetic disease in the world, the most common deadly genetic disease in the world, sickle cell anemia, sickle cell disease, to convert that pathogenic mutation into a benign naturally occurring variant. As you pointed, it's called hemoglobin Makassar, or HBBG, and there are already humans naturally born with this mutation, even in hemizygous or homozygous form, and it appears to be benign. They don't appear to have blood abnormalities, and so the, the prospect of directly converting the sickle cell mutation into a benign hemoglobin form without making double-stranded breaks, without introducing a variety of indels, indel byproducts at that site, without depending on homology-directed repair, which can be challenging, I think really provides an exciting potential avenue for a future treatment of this very common devastating genetic disease.

Yeah, great. I think we'll have to make this the final question, uh, two-part question. Uh, we just simply haven't got the bandwidth here to accommodate everybody. Um, hopefully we can do more, do more on prime editing later in the year, um, because there's clearly the demand. Um, John asks two questions about prime editing, David. Uh, John, as did many others, asks what are the limitations on the insertion size that you can create using prime editing, and Alessandro was curious whether you are considering evolving the reverse transcriptase portion of the complex to enhance editing efficiency, maybe using your PACE system.

Yes, so the answer to the first question is using the version of prime editing that we've published, we put in inserts that are up to about 100 nucleotides long, but in unpublished work we have developed some creative prime editing systems that allow much larger insertions, and so hopefully we'll report them at some point this year, assuming we don't all shut down again, that's for good, and then with respect to the second question, Yes, we have and are continuing to do exactly that. So as the questioner asked, we have developed a PACE system for prime editing and are using it and have used it to successfully evolve new prime editors, including ones with new reverse transcriptase enzymes that are evolved to have new and in some cases better editing capabilities, so we hope to report that soon as well, and we're pretty excited by the growth in the prime editing area, which is a major focus of our lab right now.

Of course, of course. Well, that's all the time we have for today's webinar. Today's webcast will be available on demand for repeat viewing. There was so much in it, and sharing with your colleagues as well. If you want more information on base and prime editing, I do hope you'll check out the latest episode of Guideposts. That's the CRISPR Journal's podcast series, which features my in-depth interview with David recorded just a few weeks ago. Guideposts is available on Spotify, Apple Podcasts, and your favorite podcast platform. We covered much of the ground presented today and a few other topics as well. Our thanks, our sincere thanks to David Liu from Harvard, the Broad, HHMI, and Jessica Madigan from TriLink as well. For supporting and speaking on in today's webinar, the WordCast team for their great hosting services as always, but most of all thanks to you for watching in such huge numbers. For everyone at GEN, or G-E-N as David calls it, and the CRISPR Journal, I'm Kevin Davis.

Thank you so much and goodbye for now.

Thanks everybody. Thanks Kevin.