While Mother Nature takes first prize in the race to develop new forms of CRISPR, biochemist David Liu is a close runner-up — and his CRISPR inventions have the potential to treat or prevent a long list of dreaded diseases, from progeria to Tay-Sachs. In 2016 Liu and his junior colleagues invented CRISPR “base editing,” which seamlessly changes a single DNA letter; that simplest of all edits may be all that’s required to repair mutations that cause thousands of inherited diseases. Last month he gave the world “prime editing,” which can delete long lengths of disease-causing DNA or insert DNA to repair dangerous mutations, all without triggering the chaotic (and possibly harmful) genome responses introduced by other forms of CRISPR.
In his spare time, Liu co-founded Editas Medicine (EDIT), Beam Therapeutics, and Prime Medicine. STAT invited readers to submit questions to Liu on the new technology. He also received some questions directly. From his perch at the Broad Institute of MIT and Harvard, Liu answered:
Anonymous: I saw the announcement regarding the deal between Prime Medicine & Beam Therapeutics. Will Prime Medicine develop its own therapeutic applications of prime editing?
As was announced, prime editing for human therapeutics will be jointly developed by both Prime Medicine and Beam Therapeutics, each focusing on different types of edits and distinct disease targets, which will help avoid redundancy and allow us to cover more disease territory overall. The companies will also share knowledge in prime editing as well as in accompanying technologies, such as delivery and manufacturing.
Anonymous: Why do another startup instead of putting the technology into Beam, where you’re a co-founder?
I believe patients will be best served by having multiple teams of people dedicated to transitioning each new technology into human therapeutics. The relationship between Beam Therapeutics and Prime Medicine was designed to maximize patients’ interests, so that as many patients as possible can benefit from base editing or prime editing technology. This cooperative arrangement also minimizes the unnecessary expenditure of time and resources on critical technologies such as delivery and manufacturing that are needed to bring any editing technology to patients. It also ensures that when base editing and prime editing both offer potential solutions to editing a disease target, both technologies can be explored non-competitively to maximize patient benefit.
Zoe A.: Can you please compare the pros and cons of prime editing versus base editing?
The first difference between base editing and prime editing is that base editing has been widely used for the past 3 1/2 years in organisms ranging from bacteria to plants to mice to primates. Addgene tells me that the DNA blueprints for base editors from our laboratory have been distributed more than 7,500 times to more than 1,000 researchers around the world, and more than 100 research papers from many different laboratories have been published using base editors to achieve desired gene edits for a wide variety of applications. While we are very excited about prime editing, it’s brand-new and there has only been one paper published thus far. So there’s much to do before we can know if prime editing will prove to be as general and robust as base editing has proven to be.
We directly compared prime editors and base editors in our study, and found that current base editors can offer higher editing efficiency and fewer indel byproducts than prime editors, while prime editors offer more targeting flexibility and greater editing precision. So when the desired edit is a transition point mutation (C to T, T to C, A to G, or G to A), and the target base is well-positioned for base editing (that is, a PAM sequence exists approximately 15 bases from the target site), then base editing can result in higher editing efficiencies and fewer byproducts. When the target base is not well-positioned for base editing, or when other “bystander” C or A bases are nearby that must not be edited, then prime editing offers major advantages since it does not require a precisely positioned PAM sequence and is a true “search-and-replace” editing capability, with no possibility of unwanted bystander editing at neighboring bases.
Of course, for classes of mutations other than the four types of point mutations that base editors can make, such as insertions, deletions, and the eight other kinds of point mutations, to our knowledge prime editing is currently the only approach that can make these mutations in human cells without requiring double-stranded DNA cuts or separate DNA templates.
Nucleases (such as the zinc-finger nucleases, TALE nucleases, and the original CRISPR-Cas9), base editors, and prime editors each have complementary strengths and weaknesses, just as scissors, pencils, and word processors each have unique and useful roles. All three classes of editing agents already have or will have roles in basic research and in applications such as human therapeutics and agriculture.
Arnav G.: Does this CRISPR technology hold the key to potentially change humanity forever, and if so can it be regulated safely enough to only cure diseases?
The scientific community and regulators must continue their efforts to ensure that all genome editing technologies are implemented in an ethical and responsible manner. This partnership will become increasingly important as genome editing technologies continue to advance in their capabilities and application scope. As we cautiously proceed, scientists, ethicists, and other members of society have an opportunity to engage in discussions about how to best use these technologies for good. Most people seem to agree that using genome editing to treat an otherwise untreatable disease that causes death or great suffering is ethical; indeed, it may be unethical to not move towards such applications as these technologies mature. Editing for disease prevention or other forms of what some view as human improvement poses more ethically and scientifically complicated questions. Many have advocated for a moratorium on clinical germline editing to create edited babies, which I support at this time.
Donna C: Tarlov cyst disease is thought to possibly be a genetic disease. Can you fix this problem with patients who already have this disease or perhaps in patients who have children and they’ve not yet been diagnosed with Tarlov cyst disease?
I’m not familiar with this disease, but many diseases with a genetic component also have non-genetic risk factors, making them less promising candidates for treatment with genome editing. The first diseases to be treated with genome editing technologies are likely diseases associated with highly penetrant mutations, such as sickle cell disease, for which having two copies of a certain mutation in one of your hemoglobin genes results in a very high probability of having the disease.
Jacqueline M.: How do you anticipate this impacting BRCA1 and BRCA2?
BRCA1 and BRCA2 variants that predispose individuals to cancer, and many other genetic variants like these, could in principle be addressed by genetic therapies. However, there are a number of challenges associated with using gene editing for this purpose. Effective treatment would likely require editing a large percentage of cells in the target tissue(s) in order to achieve a substantial decrease in the likelihood of cancer developing in that tissue, which in turn would require very efficient delivery, as well as high activity, of the agent making the desired edit in the relevant cells of patients. Much research will be required to study and explore these areas before gene editing techniques such as prime editing could be used in human patients.
Anonymous: What do you consider the key inventive step in the prime editing system?
There were a few key challenges to overcome, but I’ll highlight two. The big-picture idea behind prime editing is that it finds a target site in the genome and directly writes new strings of DNA letters that can replace the original DNA sequence. One of the earliest and greatest challenges was designing a way to make a new strand of DNA of our choosing at virtually any site in a vast genome. Ultimately, the development of the prime editing system was made possible by the scientific community’s collective knowledge about the biochemistry of both CRISPR-Cas9 and reverse transcriptases. Armed with this knowledge, many rounds of brainstorming and experiments about how to implement the big-picture idea with different designs led to a solution.
A second major challenge was achieving prime editing in mammalian cells. The initial development was done in a test tube, with an intermediate set of experiments done both in test tubes and in yeast cells. While the results of these early studies were encouraging, the inner workings of a mammalian cell are far more complex and less predictable. Our initial mammalian cell experiments yielded no editing at all, and it wasn’t clear if the problem was one we could overcome or if it instead was an intrinsic feature of mammalian cells. Strengthening the primer-template complex turned out to result in the first low-level but unequivocal prime editing in human cells, validating that a mammalian cell could support all the key steps.
Anonymous: The Nature paper mentions four cell lines in which the prime editing system was validated: HEK293T cells (in which most of the work was done), K562 cells, HeLa cells, and U2OS cells. Is that sufficient or do you think it needs to be demonstrated in other cell types?
Prime editing should be tested and optimized in as many cell types as researchers are interested in editing. Our initial study showed prime editing in four human cancer cell lines, as well as in post-mitotic primary mouse cortical neurons. The efficiency of prime editing varied quite a bit across these cell types, so illuminating the cell-type and cell-state determinants of prime editing outcomes is one focus of our current efforts.
Anonymous: Are there any cell types in which it may be more difficult to use prime editing?
The five types of cells in which we’ve performed prime editing thus far (four human cancer cell lines and post-mitotic primary mouse cortical neurons) are a small subset of the hundreds of types of cells researchers use in the biomedical sciences. Our understanding of the cellular factors that determine prime editing efficiency is incomplete at this point, so it’s difficult to predict which cells will be better or worse suited for prime editing. That said, the ability to deliver the prime editor construct (in DNA, RNA, and/or protein form) into cells is certainly one factor that contributes to the ease or difficulty of using any genome editing agent in a cell type.
Anonymous: Are there specific plant breeding applications for which prime editing is more promising than other genome editing approaches?
Some plant alleles of interest cannot be installed with base editing (because they are not transition point mutations) and cannot be efficiently accessed by nuclease-mediated approaches. In these cases, prime editing may offer a way to install desired changes.
Anonymous: What kind of changes will it not be useful for?
We haven’t tried using prime editing for very long (gene-sized) insertions or deletions, but based on its mechanism I suspect modifications to the basic prime editing scheme would be needed for such changes to be efficient.
Anonymous: In what kinds of mutations/conditions/genetic diseases does prime editing’s lack of a requirement for endogenous homology-directed repair (HDR) provide opportunities?
In cases when HDR is inefficient (that is, most cells that are not dividing) or when the indel, translocation, rearrangement, or p53 activation byproducts of making double-strand breaks is undesired, prime editing might offer attractive advantages and new opportunities over nuclease-based approaches.
Anonymous: Are unintended on- and off-target effects likely to be more or less of an issue with the new method compared with other methods?
We show in our study a side-by-side comparison of off-target editing between Cas9 nuclease and prime editing at known Cas9 off-target sites, with the result that prime editing is far less prone to off-target editing at known Cas9 off-target sites. Of course, these observations do not necessarily mean that prime editors will not have their own off-target activities, and additional studies are needed to study this possibility.
Anonymous: You note that “current base editors can offer higher editing efficiency and fewer indel byproducts than prime editors, while prime editors offer more targeting flexibility and greater editing precision.” What are the differences in efficiency?
We compared side-by-side current base editors with prime editors for making the same transition point mutations at the same target sites. For target bases that are well-positioned (relative to an existing PAM) for base editing, current base editors — which can be very efficient — offered higher editing efficiency (for example, double the yield of edited cells) than prime editors. For bases that were not well-positioned for base editing because a PAM was not present in an optimal position, prime editing resulted in higher editing efficiencies.
Anonymous: What are the differences in terms of generating indel products?
The current-generation cytosine base editor we tested in our comparison experiments resulted in similar or lower levels of indels than prime editors. Adenine base editors offer much lower levels of indels than prime editors; indeed, the very low level of indels induced by ABEs is a major strength of this class of base editors.
Anonymous: Are there specific technical/mechanistic reasons why current base editors can offer higher efficiency?
While the mechanisms of base editing and prime editing are very different, it’s worth noting that base editors have been refined through several generations of engineering, evolution, and/or empirical optimization in many cell types and organisms, which contributes to why current base editors tend to work efficiently and robustly, with relatively few byproducts compared to other editing methods.
Anonymous: Conversely, in what way can prime editors offer more targeting flexibility and greater editing precision than other editing methods?
Relative to base editors, prime editors can edit bases near or far from PAM sites, and thus are not strongly constrained by the presence or absence of PAMs near a target site, whereas base editors require that a PAM be located within a certain distance range from the target base. In terms of precision, base editors can edit multiple Cs or As within the base editing window (typically four to five nucleotides wide). In most cases, this “bystander” base editing does not introduce additional unwanted changes to the sequence of a protein encoded by a target gene, due to the degeneracy in the genetic code, but in other cases, bystander edits can be undesirable. With prime editing there is no possibility of inadvertently prime editing a nearby base because the edits are specified entirely by the RT template within the pegRNA.
Relative to nucleases such as zinc-finger nucleases, TALE nucleases, or Cas9, prime editors, like base editors, don’t require double-strand breaks, and thus ratios of desired edits to unwanted indels are typically high. In contrast, nucleases primarily result in an uncontrollable mixture of indels at the target site. For these reasons, prime editing is an unusually precise genome editing technology.
Anonymous: What are the 11% pathogenic alleles that are theoretically outside the scope of this technology?
To calculate the fraction of the 75,000+ pathogenic mutations in ClinVar that can in principle be corrected with prime editing, we summed each of the classes of changes that we performed in our study: all 12 possible base-to-base changes, insertions as large as 44 bp, deletions as large of 80 bp, and combinations of the above, yielding approximately 89% of the pathogenic mutations in ClinVar. It’s important to note that, contrary to some misquoting of our paper by a few Twitter analysts and media outlets, this is not the same as saying 89% of genetic diseases can be treated — treating a genetic disease typically requires many components, only one of which is correcting the mutation that causes the disease.
The remaining 11% of pathogenic alleles in ClinVar that we did not count in the 89% total are those classified as “copy number gain,” “copy number loss,” and “other.” While I suspect prime editing might be able to ameliorate some of the effects of these mutations, since we don’t know how large a segment of DNA prime editing can add or remove, to be conservative we did not include any mutations within these categories as potentially correctable with prime editing.
Ricardo S.: CNS looks like the area with less solutions. When do you expect to see gene editing for CNS monogenic diseases (Rett, Angelman, PWS, etc)? And for CNS polygenic diseases (autism, Parkinson’s, Alzheimer’s, dementia, etc.)?
There is a great deal of interest in applying gene editing methods such as prime editing to the treatment of diseases that currently have few or no effective therapeutic options. Diseases of the central nervous system are unfortunately common examples from this category. A number of challenges and questions must be addressed before a therapeutic gene editing system based on prime editing could be used in patients with CNS disorders, including delivery of the editing agent to the correct cells, and determining the extent of editing and the time window for editing that is required to achieve a therapeutic effect.
It is important to emphasize that it typically takes years to perform the necessary experiments on cells and animal models of a given genetic disease before a potential human therapy using prime editing, or any other genome editing technology, might be considered ready for clinical trials.
Ricardo S.: What are your thoughts on the rest of technologies, like ZFN and TALENs? Do they have their niche?
As I mentioned above, different genome editing methods provide complementary strengths and weaknesses, including those based on nucleases. Zinc-finger nucleases (ZFNs) and transcription activator-like effector nucleases (TALENs), like Cas9 nuclease, create double-strand breaks in DNA by cutting both DNA strands. There are applications such as gene or regulatory sequence disruption, or moving large segments of DNA, that are well-suited to genome editing strategies that proceed through double-strand breaks.
David P.: You have developed truly revolutionary technologies, elegantly addressing important challenges of biological engineering. What approach and which criteria do you use to decide on a specific strategy to address a scientific question? Please keep inspiring us!
Thank you for your kind words. Problem selection is a key criteria, in that there’s a large opportunity cost to spending time and effort to solve a problem or study a system that is unlikely to have a broader impact. Feasibility and risk are important considerations as well, but we are willing to pursue high-risk projects if the potential impact is also great. That’s not a license to fall in love with an idea that is fundamentally unsound or unrealistic, however! Often we will start by breaking down more ambitious projects such as DNA-templated synthesis, PACE, base editing, or prime editing into several smaller challenges, each of which seems at least on paper to be achievable. As we validate each of these smaller steps, we adjust our strategy and assumptions based on our successes or failures.
Ana Celia C.: My granddaughter Paulina was born with Tay-Sachs disease Sept. 10, 2017, she is now 2 years old, and thanks to palliative medicine I would say she is doing well given her age and alternatives. I am sure CRISPR-Cas9 is a very promising therapy. Is it possible for her (Paulina) to participate in a clinical trial for this treatment? How? I am the president of Cure Tay-Sachs Brasil and I know 61 patients with TS and allied diseases in Brazil.
I’ve very sorry to hear of Paulina’s Tay-Sachs disease. She is fortunate to have a grandmother who is so devoted to supporting the treatment of Tay-Sachs patients. While we showed in our paper that prime editing is capable of making the precise four-base deletion that corrects the most common cause of Tay-Sachs disease, demonstrating that an editing agent is able to fix a causative mutation is only one of many components towards developing a potential treatment that is suitable for a clinical trial. For example, establishing that the agent can be delivered in an animal model in a manner that reaches the target tissue, understanding to the extent possible any potential side-effects and how to best manage them, optimizing editing efficiency in the target tissue in animals, developing practical ways to manufacture the candidate therapeutic at the high quality and consistency level required for use in humans, and several other key requirements are also crucial to developing new therapies. Addressing these challenges takes years, but is important to maximizing the likelihood that patients benefit from a new therapy.
Michael D.: How do you get edits into a sufficient number of cells to effect a cure?
This is an important question, which has widely differing answers depending on the disease. For some diseases such as certain blood conditions, the relevant cells can be edited outside the body (ex vivo), then transplanted back into the patient, which makes editing the relevant cells easier in some respects. Genetic diseases also differ in the fraction of target cells that must be corrected to ameliorate or even cure the disease. Fortunately, for many (but not all) genetic diseases, correction of even a modest fraction of target cells is thought to offer benefits to patients. Determining the fraction of edited cells that will result in clinical benefit is one of the key requirements in undertaking the development of a gene editing therapy.
Emma S.: My boys are 6 months old and 3 years old and have a rare deletion early in the DMD gene sequence, 3-7. However, early-stage gene therapy and gene editing treatments generally focus on the smaller and more frequent deletions that happen later in the sequence, in the 40s-50s. Can the prime editor target a large deletion (adding back as many as five exons) and if so, can it be done that early in the sequence? That is to say, I see that you indicated it can “correct 89% of known disease-causing genetic variations in DNA,” so what do you see as the 11% that can’t be corrected, and why? Also, how can the gene editing affect some of the physical manifestations of any disease? Lastly, do you think this type of treatment will become available in their lifetime?
Our paper reports the insertion of up to 44 DNA letters at a target site using prime editing. We don’t know yet the limit of how large a targeted insertion can be made with prime editing, although I suspect that some additional variations of the method will be needed to make much larger insertions.
To calculate the fraction of the 75,000+ pathogenic mutations in ClinVar that can in principle be corrected with prime editing, we summed the categories of changes that we performed in our study: all 12 possible base-to-base changes, insertions as large as 44 bp, deletions as large of 80 bp, and combinations of the above, yielding approximately 89% of pathogenic mutations in ClinVar. It’s important to note that, contrary to some misquoting of our paper by some Twitter analysts and media outlets, this is not the same as saying 89% of genetic diseases can be treated — treating a genetic disease typically requires many components, only one of which is correcting the mutation that causes the disease, as I described above. In addition, many diseases with a genetic component are caused by multiple mutations, or a combination of mutations and environmental factors. So correcting mutations associated with diseases is a key step forward, but does not guarantee that the associated diseases will all be treatable.
Henry B.: Can prime editing be used in mitochondrial or bacterial genomes to act as a in vivo antibiotic? To treat sepsis after fast Nanopore sequencing?
CRISPR editing agents including Cas9 and base editors have already been used to edit bacterial DNA in several reports, and our lab has also begun to use prime editors to edit DNA in bacteria. Editing mitochondrial DNA with CRISPR is a longstanding goal of the field, but poses special challenges associated with coaxing the guide RNA (which programs CRISPR agents) to enter the mitochondria.
Namho K.: Always exciting to see new editing methods, but obviously delivery to target cells is a problem. When will the delivery field catch up with all these RNP discoveries, especially for non-virals?
This is an area of intense research and will be important for a number of therapeutic technologies, including prime editing. The past several decades have led to some substantial advances in both viral and non-viral delivery methods, several of which have been successfully used to deliver genome-editing agents in animal models of human genetic diseases. So as you point out, there’s a lot to be done in the field of delivering large molecular machines such as genome editing agents, although researchers have also made substantial progress that has yielded some clinically relevant delivery technologies.
Michael C.: Much has been made over the size of prime editing payloads, and whether they are too large for AAV delivery to be an option. What do you say? Are you simply betting the delivery landscape will change soon?
It’s true that more sophisticated molecular machines with more or larger components require the delivery of larger payloads, but some facts are helpful to maintain proper perspective. The instructions encoding the most commonly used form of CRISPR-Cas9 plus a guide RNA and associated regulatory sequences needed to make them in human cells totals around 5,000 base pairs of DNA sequence. Base editors plus a guide RNA and associated regulatory sequences are around 6,000 base pairs. And prime editors plus a guide RNA and associated regulatory sequences are around 7,000 base pairs. So while prime editors are larger than base editors, which are in turn larger than the original CRISPR Cas9, their size differences are not as dramatic as certain media reports have implied. (Fortunately, math, rather than clicks, determines the size of genetic cargoes!) A number of Cas9 variants, deaminase variants, and reverse transcriptase variants have already been described that could substantially shrink the size of programmable nucleases, base editors, and prime editors, and we and others have already begun to use these variants to minimize the size of genome editing agent instructions.
Anne B.: My question is related to this information in the article, “… The third component, an enzyme called reverse transcriptase that’s fused to Cas9, copies the RNA nucleotides carried by the pegRNA and transforms them into DNA nucleotides, which replace those at the target site.” How does the reverse transcriptase do this? Where does it get the nucleotides that become the fixed up/targeted part of DNA? Are the necessary nucleotides floating around in the nucleus or in the cytosol and if so, how do they hook up with the pegRNA?
The reverse transcriptase enzyme in prime editors uses the same DNA building blocks, called deoxynucleotide triphosphates, that the cell naturally for its own DNA replication and repair. These building blocks are therefore plentiful in the cell and available for incorporation into new DNA strands by reverse transcriptase. The reverse transcriptase enzyme uses the pegRNA sequence, a string of RNA nucleotides, to template the new string of DNA nucleotides, and adds the new DNA sequence, letter by letter, to the target DNA strand. During the process, the reverse transcriptase checks for the correct nucleotide pairing to ensure that it matches the pegRNA template.
Joanne K.: The materials from your paper became available on the day the paper came out. In only three days over 100 plasmid requests had been distributed by Addgene! What do you think this type of sharing will do for the field? What exciting things will scientists do to build on your work?
We deposited the basic prime editing constructs to Addgene far enough in advance of publication so the scientific community would immediately be able to test and use prime editing. It’s very exciting that the community moved so quickly; in the first 10 days since our paper was published, Addgene informed me that they sent about 450 prime editing constructs to more than 100 laboratories around the world. To realize the full potential of any new technology, it is essential for many researchers to test, validate, and improve upon the original method. Indeed, it will take a huge village to achieve a goal as ambitious as using an engineered molecular machine to replace disease-causing DNA sequences with normal ones to treat a genetic disease in a patient. We are eager to see diverse and creative applications of prime editing.
Anabella Maria G.: What advice do you have for young researchers looking to enter the complicated and, at times, controversial field of genetics?
Choose important problems, and treat your students with the care and respect they deserve.