Scientist Stories: David Schaffer, Engineering Viruses for Gene Therapy

I'm going to be talking to you about a project area or a scientific area which has really become half of my career, and the Hertz Foundation actually played a role in my first steps in this direction because I began to work in this general area as a graduate student with a fellowship. And although the direction has pivoted a couple of times, I've stayed within this area, and I'm pretty thrilled that the fellowship enabled me to first enter into gene therapy. So the motivation for this work is the fact that regardless of which tissue you look at inside the human body, there are a number of unsolved biomedical problems, significant unmet medical need, and specifically there are long-term chronic disorders which progress over the course of years or decades, progressively rob these tissues of function and progressively lead to a decline in, in human health and quality of life. And furthermore, these conditions also responsible for progressively driving up the cost of healthcare within our country. They don't have cures in many cases. They have treatments, and those treatments tend to be quite expensive. And as a result, the fraction of GDP that we've spent on healthcare over the past 2 to 3 decades has increased even faster than the rate of GDP growth. So from somebody who's— for somebody who's interested in developing novel treatments for human disease, This raises the question, can we actually develop better therapies to treat these conditions? And if you take a few steps back to the basic biology question of the central dogma of biology, where information is stored at the level of DNA, transcribed into RNA, and then translated into protein, to me, each one of these represents a potential set of drug targets. And the majority of drugs that are used in the clinic these days target at the level of protein. So small molecules like ibuprofen float around inside of a cell, typically find a hydrophobic pocket within an enzyme and block it as a competitive inhibitor. Or monoclonal antibodies roughly do the same thing on the outside of the cell, where they float around until they find a target, bind to it, and typically antagonize its function. However, it's possible to use RNA as a therapeutic target as well. So RNA interference is the basis of a couple of FDA-approved products now. Antisense oligonucleotides, which typically block the translation of RNA into protein is another potential modality. And furthermore, as we well know from Moderna and Pfizer these days, messenger RNAs themselves can actually be therapeutic molecules. However, the majority of my work has actually been back one further step, which is targeting drugs at the level of DNA. So if you're adding new DNA to a genome, this is gene therapy. If you're changing the sequence of existing DNA within a genome, that's genome editing. And if you're adding brand new genomes to a tissue, that's the addition of cells or cell therapy. And the advantage of working at the level of DNA is the fact that small molecules and proteins and RNA come and go. They have a finite pharmacokinetic half-life within the body. However, DNA uniquely has the potential to become a permanent part of an organ or a tissue. So you can begin to think about the idea of a single intervention at the level of DNA resulting in long-term therapeutic benefit or one and done. So just to further illustrate this idea, if you take a look at the average lifespan of a patient, once they become diagnosed with Parkinson's, it's on the order of 16 years or so. Congestive heart failure, it's on the order of 10 years. Age-related macular degeneration, around the same time. The pharmacokinetic half-life of a small molecule is on the order of hours. If you move to a monoclonal antibody, this is on the order of weeks. But if you're developing treatments based on small molecules or antibodies to treat these conditions, they are by definition treatments, not cures, because you have to continuously dose them throughout the rest of somebody's lifespan. However, these are two examples. I'll be presenting one of them in a bit more detail where patients were originally dosed with a DNA therapy, a gene therapy, back in the 2000s. And even so today, after 10 years plus, after that initial administration, they're still enjoying the clinical benefit of these treatments. So just to get everybody on the same page, gene therapy can be defined as the delivery of a DNA medicine, genetic material, to the cells of an individual for a therapeutic benefit. And this idea is not a new one. It was originally proposed back in the 1960s by a biologist by the name of Joshua Lederberg. Josh Lederberg won the Nobel Prize in Physiology and Medicine for discovering that bacteria can actually transfer a piece of DNA from one bacterial cell to another, process known as DNA conjugation. And he began to hypothesize that potentially you could transfer DNA into a person's cells as a therapy. But what has really held up the field in the intervening 50 years is this word right here, delivery, which is the fact that from the point of view of a cell, DNA is a really big molecule. It's charged, doesn't pass through membranes very well. And on its own, floating around in serum, DNA is not especially stable. So it's been exceedingly difficult to get enough DNA delivery into cells and tissues to begin to move the needle, to begin to modify the progression of underlying disease. So fortunately, over the past 2 to 3 decades, investigators have actually identified some very helpful delivery vehicles that come in the form of viruses. So the most successful of these to date is called adeno-associated virus, which I've been working on for nearly 25 years. And AAV is interesting, and it is among the smallest of all viruses, both in terms of its genome length—it only has around 4,700 nucleotides of information—and also in the dimensions of the physical particle. So it's around 25 nanometers in diameter or so. And the way that the virus works is it only contains two open reading frames, two genes. And the first of these, rep, encodes for some enzymes that mediate replication of the viral genome, the viral DNA. And the second has been the focus of around half of my career. And this cap gene encodes structural proteins, 60 copies of which will self-assemble to form this hollow soccer ball, this icosahedron. And it contains at its 5-fold axis of symmetry a pore. And one of these rep proteins will actually dock at the pore and will grab a single-stranded genome from elsewhere inside the cell and snake that genome inside of the particle to be able to load it with its payload and create an infectious variant. So these viruses have other interesting properties. They're non-pathogenic. They've never been associated with human disease, and as a result, most of the lay public hasn't heard of them because they're not SARS-CoV-2 or HIV or West Nile or Ebola. In addition, over 100 natural versions of this virus have been found or discovered within nature, and they contain diversity, especially on the peaks of these ridges coming off the surface of the soccer ball. And as a result of this diversity in the amino acid composition on the surface of the virus, they have slightly different delivery properties. But they all share the fact that they float around inside the body, they bind to specific receptors on the surface of a target cell, those receptors will endocytose or internalize the virus. It actually has an enzymatic activity that will enable it to chew its way through this lipid bilayer and escape to the inside of the cell. It becomes then translocated to the nucleus. The soccer ball opens up and releases its genome, and then that genome effectively forms a latent infection where the DNA can persist for years or even decades inside of our cells. So investigators back in the '90s began to exploit these properties to actually deliver DNA medicines rather than the virus's own genes. And the way that this works is you quite simply take the two pieces of viral DNA, the rep and the cap, cut them out of the viral genome, paste them onto a separate piece of DNA, and then this liberates space for the insertion of, in principle, any gene of interest, provided that it fits inside of these viral sequences. And once the soccer ball assembles inside of a producer or manufacturing cell, then this gene of interest, rather than the virus's own genes, becomes internalized into this virus, and it effectively turns into a Trojan horse that will then deliver the gene of interest rather than its own DNA. So this resulting recombinant viral delivery vehicle has a number of interesting properties. It's very safe. 90% of us on this call have already been exposed to one or more natural versions of AAV. And as a result, this indicates the high degree of safety of the parental virus. And the resulting particles based on the parent are also quite safe. It's been utilized in over 250 human clinical trials involving delivery of DNA medicine to muscle, liver, lung, brain, retina, and other tissues. Upon arriving in the nucleus of the target cell, now the gene of interest gets deposited inside that nucleus and can persist for very long periods of time, provided that that target cell is non-dividing, is non-mitotic. And as a result of these advantageous properties, AAV is beginning to enjoy success in a number of clinical trials. So shown here was the first for which this is the case. And this is, these are patients suffering from a rare genetic condition known as Leber's congenital amaurosis type 2, where they unfortunately inherited from both mom and dad a defective copy of a gene encoding an enzyme necessary for vision, for photoreceptor function. And in the absence of this gene, these patients progressively go blind. Investigators took the correct copy of that gene, loaded it into the virus, injected into the eye, And you can see 6 months later, the results very simply speak for themselves. This became the basis of a Phase 3 clinical trial that a company in Philadelphia ran. They submitted an application to the FDA for drug approval back in 2017, and the FDA approved it December of that year. And this became the very first gene therapy for a rare disease approved by the FDA and also the very first gene therapy based on this virus, AAV. There's since been one additional FDA approval of a gene therapy based on adeno-associated virus, or AAV, and this is for spinal muscular atrophy type 1. This is developed by Novartis, and this is the most common inherited cause of infant death, and these patients are now living beyond 10 years, even though the median lifespan on average for one of these patients is on the order of 10 months. There were also positive results for hemophilia B, where investigators delivered the correct copy of the DNA encoding the blood clotting factor that's mutated within hemophilia. And as a result, these patients have been effectively cured for life of hemophilia B. This will likely be the next FDA approval, likely later this year. So there have also been a number of other FDA approvals, or a couple of other FDA approvals for other conditions using different viruses, but by and large, AAV has been the most successful utilized in the body to date. So this is the good news. The not as good news is that these are quote-unquote low-hanging fruit for conditions that one might want to treat with a gene therapy. These are situations where some pretty savvy clinicians identified a disease condition where the natural versions of AAV were just good enough to start to get efficacy. And even so, they had to really brute force the delivery. They had to use massive doses of the virus. They had to, in some cases, use a very invasive route of administration to inject the virus right next to the the target cell, because otherwise it couldn't get through all the transport barriers it would have to surmount to be able to arrive next to the target. So unfortunately, as a result of these limitations in the delivery properties of the natural viruses, most disease targets are still beyond the capabilities of AAVs. In other words, we quite simply as a field need better viruses. So this is due to a no-brainer consideration to my mind, which is that viruses didn't evolve within nature to be used as human therapies. So as a result, it's up to us to improve the properties of these viruses to make them better for our purposes, rather than settling for what nature happens to have thrown over the fence at us. So they suffer from a number of challenges or a number of properties, including the fact that they biodistribute to the wrong tissues inside the body, the fact that, like I mentioned, many of us have already been infected with one or more natural versions of the AAV and we've raised antibodies against them that will neutralize particles based upon natural AAV versions. In addition, they completely lack the capacity to target delivery to one cell type versus another, and they're quite inefficient on many different cells, again, simply because evolution in nature didn't prepare them to serve as a DNA delivery, as a gene medicine delivery vehicle. So it's therefore incumbent upon us to re-engineer these viruses at the molecular level to make them better for our purposes. But this is really difficult to do because there are 4 million molecular weight products of tens of millions of years of evolution, and we quite simply don't have enough rational mechanistic information about how these viruses work in order to re-engineer them amino acid by amino acid to improve their properties. So fortunately, a very talented protein engineer in the field came up with a solution to this problem back in the early 1990s. Francis Arnold is a graduate of my department at UC Berkeley, a chemical engineer, who has since been a professor at Caltech. And back in the early '90s, she began to work on enzyme engineering. And rather than staring at a crystal structure and hypothesizing amino acid by amino acid how to make changes in order to alter the properties of enzymes, she began to launch this field of directed evolution where she instead made thousands of different variations of the protein and performed high-throughput selections to be able to identify the fittest. And for this innovation, she won the Nobel Prize in Chemistry in 2018. So I began to work in gene therapy a few years after Frances was doing her work, her seminal work with enzymes, and I began to see whether or not it'd be possible to adopt directed evolution and apply it to make better versions of viruses. So the rationale for this again is that if you take a look at an AAV over tens of millions of years, it's likely evolved to be highly optimal within its natural setting. But if we end up taking the virus out of its natural setting, it's a respiratory virus, and we start using it to deliver DNA to different cell types in the body, and we change the route of administration for the virus, then we're quite simply changing the rules as to what success means. And this soccer ball, this capsid sequence, for something that may have been highly optimized within nature may now be in a therapeutic setting completely suboptimal. So we began this process of starting out with this particular variant and creating a cloud of mutants around it and performing high-throughput selections round by round, step by step, so that we could potentially climb up this curve and be able to create a brand new virus that's now highly optimized as a therapeutic DNA delivery vehicle. So the way that this works is that we start out with a number of different input sequences As you may remember, I mentioned that there are a number of natural versions of this virus, each of which may be an interesting starting point for evolution. So we take these and then we perform genetic diversification using enzymology and molecular biology so that we create actually a huge amount of diversity within the sequence composition of these viruses. We've actually done this in 40 different ways to create 40 different viruses— or sorry, 40 different libraries. I'm not going to go through all these due to restrictions of time, but shown here are a couple where we used a computational biology approach guided by structural biology to identify positions where we could actually join together domains of the parent viruses and essentially breed or create random chimeras of brand new viruses that may have interesting properties that combine together features of the parents. In another approach, we use bioinformatics and computational biology to take all the existing AAV sequences in the world today, recreate a phylogenetic tree, a family tree, and effectively go back in time so that we could identify putative ancestral sequences that may have existed millions or tens of millions of years ago, and then created libraries around those putative ancestors. And then with a hypothesis that by definition, since those gave rise to all the existing AAVs today, they are evolvable. And if we apply different selective pressures in a Darwinian way, perhaps we could guide their evolution in other directions and towards other properties. But in sum total, we created 40 different such libraries, each of which had on the order of 10 to 100 million different sequences. So if you add them all together, we're talking about over a billion natural— or sorry, billion synthetic sequences of AAV that we've now made, which we think compares quite favorably to around the half dozen that are currently being used in human clinical trials. So once we've created a gene pool, the second step is to play Darwin and select the fittest. So we then take each one of these sequences that encodes a potential soccer ball that might be of use to us, and we'll package them to create libraries of viral particles, each of which is composed of a variant protein shell, variant soccer ball, that contains inside of it the DNA, the gene that encodes that protein shell. So we think of this as these particles being barcoded with their own genomes, with their own DNA. We can then perform a Darwinian selection. I'll give you a couple of examples and spin around this loop so that iteratively the billion-member library progressively converges down to just a small handful that are now specializing in having the properties that we would like them to have. And at any point along the way, we can pop open the soccer ball and sequence the DNA using next-generation sequencing to gain insights into this process of progressive convergence of the large library down to a few fit variants. So by doing this over the past 20-plus years, we've actually solved a number of these problems I mentioned to you earlier. We've created versions that are resistant to preexisting antibodies that exist within the human population. That biodistributes to the right tissues inside the body that can spread deep in those tissues, can target delivery towards particular cells, and it can very efficiently infect those target cells. And we've done this across a number of different tissues, including muscle, liver, lung, brain, and retina. But today I want to focus in on the retina because it's the work that's the furthest advanced. So this is the structure of your retina. We have our lens at the front. There's this vitreous gel or fluid on the inside of the eye. And in the back of the eye, we have painted a number of layers of neurons. Um, all the way at the back of the retina are the photoreceptor neurons, and when these get hit with a photon, they fire an action potential. That information gets processed by these inner neurons, and then finally, these retinal ganglion cells will fire an action potential along their axons, and their axons bundle together right here and form the optic nerve, and they relay that information into the brain for further processing of that visual data. So unfortunately, mutations in over 200 genes can actually result in the death of photoreceptor neurons, um, leading to progressive loss of vision. And there's one particularly important population of photoreceptor neurons known as the cones, and these reside in the middle of the eye in a structure known as the fovea or the macula, and they're important for high acuity or high accuracy vision. So if you're reading a page and your eye scans back and forth, That's so that the lens can keep this light focused in on the fovea, so you can exploit that region to be able to have the high accuracy vision necessary to read. This is a particularly important population of photoreceptors to protect. In addition to inherited genetic disorders or mutations that kill off these cells, including retinitis pigmentosa and Leber's congenital amaurosis, also more complex diseases such as age-related macular degeneration, will kill off these photoreceptors. And AMD, or age-related macular degeneration, affects millions of Americans. So the best place to deliver a drug into the eye would be into this vitreous fluid. It turns out that this is already done millions of times a year. It's around a 5-second injection, only involves eye drops, you know, local eye drop anesthesia. And this is done by retinal ophthalmologists to deliver drugs, protein drugs, to treat wet age-related macular degeneration. Those protein drugs only last about a month, so those patients have to come into the clinic to get injected around 10 times a year. But otherwise, you know, other than the extreme inconvenience of doing that, a single injection into the vitreous is actually very noninvasive and also gives a drug product the opportunity to come into contact with the full surface area of the retina. Unfortunately, however, natural AAVs, these 25-nanometer particles, lacked the capacity to make the trip from the vitreous fluid through a couple hundred microns of dense tissue and then reach the target cells, the photoreceptors. And as a result, um, ophthalmologists had to actually invent a new surgery, which they call a subretinal surgery, where they take the needle, puncture it all the way through the retina, and squeeze the virus-containing solution behind the photoreceptor layer which actually then separates the photoreceptors away from the underlying cellular layer, what is known as a retinal detachment. That's problematic for a couple of reasons. They have to do this. They did it for that FDA-approved product where I showed you that movie for earlier. But because this causes a retinal detachment, it can actually further damage the eye. In addition, this is what an optometrist sees if they're looking through the lens at the back of your eye., in this green region is the region that came into contact with the virus. After placing the needle here, you can't control how far the fluid spreads away laterally at the injection site, and you can't control the direction it goes. And typically it only protects around 1 to 5% of the surface area of the retina. So in that FDA-approved product, 95 to 99% of the retina is still being sacrificed to the disease. So we began to ask around 12 years ago or so, could we do this? Could we evolve a particle that has the capacity to transport, to overcome these transport barriers, and reach deep into this tissue so that the virus could actually come into contact with the target cell? So we began to select for this in Darwinian fashion in vivo, initially in mouse as a proof of concept, where we took our library—12 years ago was only around 10 million variants. We're now up to a billion, like I mentioned. But at that point, we took our 10 million variants injected it into the vitreous fluid of a mouse eye, waited a couple of weeks, and then specifically we had a mouse that was expressing a fluorescent protein, a green fluorescent protein, only within the photoreceptor neurons. So we could use a process called fluorescence-activated cell sorting to pull out the photoreceptor neurons after a couple of weeks, after the fittest variants had reached that cell type, and then we could use PCR, polymerase chain reaction, to pull out the genomes of the virus that made it to that target cell. We could then repackage that virus, readminister it, and go around this loop a half dozen times so the 10 million variants converge down to just a small handful. And as you can see in this pie chart, there was actually one dominant variant that really emerged from the pool. And it's shown in cartoon form here, where in green and red you can see that it experienced changes in the amino acid composition on the, on the surface of the particle locations that potentially would change the way that the virus interacted with the host cells, the host tissue. So we wanted to test how good this variant was, and so we loaded it with DNA encoding a fluorescent protein. This is the control. This is the, the parent version of the virus, which is called AAV2, and this is actually the version that's used in that FDA-approved product. And you can see why they had to do the subretinal surgery, because if you injected this virus into the vitreous fluid of the eye, this is a mouse eye, the photoreceptors are all the way back here, and you can see that the virus comes nowhere close to them. It's simply hitting a few neurons on the way. This is the variant that we evolved for being able to reach the photoreceptors, and you can see it's hitting every single cell population all the way from that first layer of neurons, the retinal ganglion cells, through the photoreceptor neurons. And furthermore, the delivery isn't limited to 1 to 5% of the retina. Surface area, we're actually hitting the entire retina. So this was done in a mouse as a proof of concept, and a company actually that I'm not affiliated with took this into two human clinical trials. But we were concerned about the potential of using this within a human for the reason that, as Francis Arnold likes to say, evolution teaches you, you get what you select for. We selected for a great rodent variant. It works really well in the rodent. Um, doesn't necessarily have to work within the primate eye or the human eye, however, and the rationale for why it may not translate is shown here. So if you take a look at the mouse eye, and then drawn to scale are the dog, the monkey, the cynomolgus macaque eye, and then the human eye. So the mouse eye only contains around 100 microliters of that vitreous fluid, the liquid inside the eye. The macaque has around 3,000, the human has 4,000. So there's quite simply a much larger region over which a virus must spread inside of a primate. There are further structural differences as well, that human beings and cynomacaques have this fovea, this high concentration of photoreceptors, this fovea otherwise known as a macula. Mice don't have a macula, so it's difficult to study a macular degeneration in organism that lacks that structure. And furthermore, right at the boundary between the vitreous fluid and the retinal tissue there's actually a layer of matrix protein called an interlimiting membrane. And this serves as a barrier for the transport of these viral particles into the retina. And this interlimiting membrane is quite simply a lot thicker within the human eye and the monkey eye. So when we tried to get this published, of course, the reviewers wanted to see how well our rodent variant would do in nonhuman primates. So we had to test it out. And so this is delivery using a number of different natural versions of AAV. You've already seen AAV2 before. We tested AAV5, AAV8, AAV9. These are just different natural isolates of AAV. And when you inject these into the vitreous of a nonhuman primate eye, typically all you get is this ring of delivery that corresponds to a superficial layer of neurons that surrounds the fovea of the eye. These become exposed and become an easier target for delivery, but unfortunately, you can see that it's hollow on the inside, which is where the cones reside, which indicates that none of these variants are able to reach the cone photoreceptors, this important population of cells. Our engineered variant, which we called 7MA, which was engineered in rodent, did better than all of these. If we jacked up the dosage quite high, we're able to start getting delivery to the cone photoreceptors and some, you know, punctate delivery outside of this region. However, this required a significantly high dosage. The animals got some inflammatory responses, and we quite simply felt we could do a lot better. So to begin to move beyond proof of concept and to start taking this towards human clinical development, around 7, 8 years ago, we started a company out of UC Berkeley called 4D Molecular Therapeutics, co-founded with a Berkeley alum by the name of David Kern. And we've developed a billion synthetic AAV capsid sequences, and we've begun to do selections in a number of different tissue targets. And then begun to advance these into human clinical trials. We went public last December. So we've done 14 different selections. I showed you an example of a retinal selection, but we've also done selections all in nonhuman primate for delivery to the brain, the lung, liver, heart, skeletal muscle, as well as a number of other tissues. The retina is the most advanced of these, so I want to describe that in a little bit more detail. So we now have 40 libraries, over a billion synthetic capsid sequences. We re-performed the selection as I described to you for mouse, but now within non-human primate. And as you can see here, this is actually Darwinian selection kind of in real time, step by step, where after each of the rounds 3 through 6, we took the pool that we isolated a virus from these different cell layers, especially the photoreceptor layer, sequenced it by DNA sequencing, and found that a progressively higher fraction of that pool was just a single dominant sequence, indicating that it was getting positively selected for in Darwinian fashion. And this particular variant was called R100, and it was distinct from that mouse variant that we had previously isolated. We then tested this one variant. We loaded it again with DNA encoding the green fluorescent protein. Here are a couple of negative controls where we did intravitreal administration of a parent sequence, AV2, just a natural AV. And again, you get that very narrow transduction pattern around the fovea. If you do subretinal surgery, you end up with this random shape of transduction where you can't really control the shape, but it's still only around 1% of the surface area of the retina. Whereas if we deliver inside the vitreous, the DNA encoding the green fluorescent protein using this engineered version R100, you can see within 1 week we're starting to get some gene expression. By 3 weeks it really turns on, and by 18 to 24 weeks we have blazingly high levels of GFP expression across the full surface area of the retina. We did a lot of histology to analyze the tissue, and you can see that across the entire retina, these regions, even if they don't appear to be white—this is simply due to autoexposure adjustment of the camera—, but across the retina we have high levels of expression of the green fluorescent protein within the photoreceptor layer and the retinal pigment epithelium, indicating this is a very efficient virus. And in addition, this persisted out through week 24, so 6 months in life, in the absence for much of that time of any sort of immunosuppression, indicating that this virus is very safe. It doesn't lead to an immune response or any sort of toxicological response. So based upon this, we began to initiate studies to be able to submit an investigational new drug application, or IND application, to the FDA, which is what you submit when you want to initiate a phase 1 clinical trial. So as part of the IND, you have to very extensively show the safety of this virus or of a new drug in a preclinical model, in this case a non-human primate, And we actually began to do that for two different disease targets. So you're probably familiar with retinitis pigmentosa. This is a collection of a couple of dozen different diseases where in each case a different gene is mutated. What they lead to is that unfortunately, you know, people have a full field of vision when they're really young. Progressively, they begin to lose peripheral vision so that they end up developing unfortunately into tunnel vision. At the very end of the disease, they're kind of viewing the world through a straw, and then finally it blinks out as they completely go blind. So, because our virus has the capacity to deliver DNA across the full span, the full surface area of the retina, it means that we can treat patients, young patients with retinitis pigmentosa, and potentially offer the opportunity to protect their full vision before it begins to degenerate. And we began to explore the possibility of treating a second unrelated condition called chorderemia, but that has a very similar progression of progressive loss of peripheral vision followed by central vision. So we performed extensive GMP manufacturing, good manufacturing practice manufacturing, to be able to generate material that could go into human beings. We performed extensive safety studies with the non-human primates, last year submitted two IND applications to the FDA, which cleared. So last May, it was slightly delayed as a result of COVID but we began to treat patients And to date, we've treated 6 patients for each one of these 2 diseases. And so far, you typically perform what's called a dose escalation study where you treat initially 3 patients at a lower dosage, then an independent data safety monitoring board will analyze the data, assess whether or not it's safe, and then give you authorization to go to a higher dose. And so in both of those cases, we successfully escalated to the high dose. Indicating that the treatment is safe. There hasn't been any evidence of toxicity within patients. I mentioned to you as well that this is a full platform technology, and we've also selected the virus for the capacity to be able to reach the heart after intravenous administration. Again, injected a billion variants into the bloodstream from a nonhuman primate, isolated cardiac tissue, and generated a completely different variant that now specializes at reaching the heart. Upon administration to the bloodstream. And we have a completely third different variant in which we aerosolized the library, applied it through aerosolization or inhalation into a nonhuman primate, and isolated a variant that was capable of delivering DNA to the lung epithelium. So we have a third clinical trial for the treatment of a cardiac condition called Fabry disease. We've treated one patient to date and hope to start treating additional patients this year. This is a rare condition, so enrollment tends to be a little bit slower. And then furthermore, we're filing two new INDs, investigational new drug applications, to the FDA later this year to initiate a trial for the treatment of wet age-related macular degeneration, which affects around three-quarters to a million Americans, and cystic fibrosis, which is the most common single gene disorder within the US and in Europe. So I'd like to very quickly close with talking about a new technology that we've developed to accelerate this process of genetically diversifying genomes to help out in the process of evolution. And this is a tool called Evolver. And because I come from UC Berkeley, I'm contractually obligated to use CRISPR-Cas9 in my lab. And so this is a CRISPR-based tool. So the way it works is that it contains a version of CRISPR, which, as you're familiar, Likely it's a double-stranded DNA nuclease. You can inactivate one, the nuclease cutting of one of those strands, so that it only nicks the DNA but doesn't make a double-stranded cut. And then we fused onto this a DNA polymerase, which is error-prone. So this makes mistakes as it's polymerizing DNA. So we can then use a guide RNA to direct this tool, which we called Evolver, to in principle any gene within any genome of an organism. The Cas9 nicks the DNA, and then this molecule dissociates, revolves 180 degrees, and then the DNA polymerase will find this nick, engage with it, and start polymerizing a track using strand displacement DNA polymerization, where it lays down a track of error-prone DNA. So it's basically randomly mutagenizing that DNA. And it can do so in a way in which you can control the processivity, the window of mutagenesis, the fidelity of the enzyme, in other words, the error frequency, as well as the AGTC misincorporation bias, simply by using different versions of that DNA polymerase. So we can see here, we initially did this work within E. coli, within bacteria as a model system. We've since translated this into yeast and then into human cells. But you can see that as a function of distance away from the nick, By next-generation sequencing, we have a very, very high error rate. And this error rate is around 8 million-fold higher than background mutagenesis. And we can end up doing this at window lengths that are around 350 nucleotides away from the nick site, from the site where the Cas9 binds. So we've begun to do this, to utilize this to do evolution inside of cells. And because this tool can mutagenize chromosomal DNA, we can actually do continuous live real-time mutagenesis and selection within cells, as opposed to doing those discrete rounds of evolution that I mentioned to you earlier. And this has the effect of actually dramatically increasing the library size that we can screen, as well as the speed of the process. So just to show you a couple of proof of principle examples of evolution, we took DNA encoding the green fluorescent protein but introduced into it a point mutation that disabled the GFP. So we could then take a guide RNA that would target this Evolver molecule right next to that point mutation, and it randomly mutagenized that DNA. In a small fraction of cases, it would make a mutation that would revert the DNA back to the normal sequence, and as a result, these cells would, would regain fluorescence. And then we could use fluorescence-activated cell sorting to pull out the cell population. So it shows as a proof of principle that we're able to get mutagenesis and then selection for a novel property. We also did this for a gene inside the bacterial genome. And specifically, there's a gene that encodes a ribosomal protein that is a target for an antibiotic called spectinomycin. And so if you end up mutagenizing that target gene, in a small fraction of cases, you'll make a point mutation that will confer resistance to this antibiotic, and that can be readily easily screened for. So you can see that for cells that are just normal cells, the frequency of spontaneous generation of a colony of cells that becomes resistant to this antibiotic is around 10^-10, which corresponds very well to kind of the background mutagenesis rate within— associated with natural cellular DNA replication. But when we use our engineered molecule and use a guide RNA that targets Evolver to the gene that is the target of this antibiotic, we see that there's a 4 to 5 order of magnitude increase in the frequency of emergence of resistance, of antibiotic resistance, indicating that we're very effectively mutagenizing this region within the genome. And the final piece of data I'll show is that we know as a field that you can use different guide RNAs and multiplex Cas9 to be able to target many different locations or target sites within a genome. So we actually combine two different antibiotics, spectinomycin and streptomycin. These two antibiotics target these two ribosomal proteins, and only when we use two different guide RNAs that will target our evolver molecule to the genes encoding these two proteins, only when we combine these two guide RNAs together do we end up with resistance, combinatorial resistance to these two antibiotics. Indicating that we're actually able to get mutagenesis and selection in real time for two different locations of the genome simultaneously. So we're currently utilizing this now as a tool to be able to mutagenize the sequences of viruses inside of mammalian cells so that we can effectively perform real-time accelerated mutagenesis and real-time selection to very rapidly evolve novel properties. So I'd like to close by leaving you with a couple of bullet points. Viruses are very potent gene delivery vehicles, but unfortunately, of course, nature never intended them for our use as a delivery vehicle for DNA medicine. So it's incumbent on us, again, to be able to make better versions of these viruses. So we've shown that directed evolution is a very powerful engine to create designer versions of AAVs. Where we can, in principle, optimize AAV for highly efficient targeted delivery to, in principle, any cell type through any single clinical route of administration that one might want to use to introduce the virus into the body, whether it be into the bloodstream through aerosolization or injection into the eye. We've used this to be able to fundamentally reprogram the virus to increase its efficiency, to target delivery towards particular cell types in the body, and in work I didn't show, actually, today, to enhance the capacity of the virus to evade the antibodies that lots of us are already carrying around in our bloodstream directed against the natural versions of AAV. This technology has now been translated into 5 human clinical trials, and the company anticipates launching 2 new human clinical trials by the end of the year. Hopefully within the next year, we'll start beginning to report out on efficacy of these therapeutic interventions. But at the same time, we're always very interested in further innovating and improving the technology platform such as by the development of this technology Evolver that can accelerate evolution. So I'd like to acknowledge the people who actually did the work. The work in the mouse retina, the proof of concept, was the product of Leo Byrne, Denise Alkara, and Tim Day in collaboration with John Flannery at UC Berkeley. The work in the non-human primate retina was conducted at 4D Molecular Therapeutics, a company in Emeryville. I'd like to call out my co-founder and our current CEO, David Curran, as well as our CSO Peter Francis. And then finally, the work on Evolver was almost the sole product of Shaked Halperin in collaboration with John Duber of bioengineering at UC Berkeley. So thanks so much for your time and attention. I'd be happy to answer any questions you might have.

Great, David. Thank you so much for the, uh, the great talk and such important work and introduction to the, the candidates that you've been developing. We'd like to turn it over to our participants, our audience here for questions. I didn't see any questions in the chat box, so I'll invite folks to feel free to go ahead and unmute your mic and jump in if you have a question for David.

I see a question in the chat box that says you indicated there are differences across species when you optimize. You know, how do these differences how do they pay attention to variations? Are they more reliant upon variations within a species, such as from one human to another? So, you know, I'll mention the first part, which is that there are species restrictions for viruses that have evolved in one species, you know, crossing over to another. And, you know, in another context with pathogenic viruses, that's actually a good thing. Otherwise, we, you know, find SARS-CoV-2 crossing from bat to human being, you know, much more often. However, you know, we found, we've definitely found that variants of AAV that we've evolved for mouse or for pig or for dog don't always translate well into non-human primate. But within non-human primate, we are doing the evolution within a number of different animal subjects, and they do vary from each other genetically. You know, typically a mouse is an inbred species where every laboratory mouse has an identical genome sequence to every other mouse. But nonhuman primates aren't bred in captivity, and so every subject has a different genome. So by going through multiple rounds of evolution, we're actually forcing the virus to be able to work successfully on a number of different individual primates, which we think is going to prepare it well for working on a number of different individual human primates.

Okay, David, looks like there's another question. Thank you for that reply.

Fantastic work.

And I'm not sure who MD Henry is, but I'll invite MD Henry to, if they'd like to unmute themselves and pose the question directly.

This is David Henry. Thank you very much for a great talk. I enjoyed it. Thank you. You started discussing about how AAV was really useful for going after hemophilia B. Would the mechanism for hemophilia A require a modification of AAV using directed evolution as well?

Yeah, there's a challenge with heme A. Hemophilia— so hemophilia B, one of the reasons it's been, it's been a low-hanging fruit is that you only need around 5% of normal levels of the protein in order to effectively cure the disease. So what investigators did was to pump in high concentrations of a natural version of AV into the bloodstream. It got soaked up into the liver, and the liver was sufficient enough to be a bioreactor to pump out around 5% of normal levels of that protein, which corresponds to like 50 nanograms per mL, and that, that's enough to cure a patient. So there are going to be challenges if you need higher concentrations of protein, And furthermore, the big challenge with hemophilia A is that the DNA is too big for AAV. So the average size of a gene within the human genome, or a protein coding sequence, is only on the order of around 450 nucleotides. AAV can fit 10 times that much. So you can actually fit a few, you know, several average-sized human proteins inside of AAV. Unfortunately, factor VIII, the factor that's defective within hemophilia A, is really big. It's one of those outliers. And so it's kind of hard to sandwich inside of AAV. Investigators have tried to trim it down by making some truncations to make a shorter version that's still active. And so that version is being pursued in the clinic, but it hasn't led to, to, uh, as strong or immediate success as investigators encountered with hemophilia B to date. So we can keep our fingers crossed though, but it, you know, may be possible to evolve a version of AAV that had a higher carrying capacity that could carry more DNA.

Great, thanks for the question, David. Looks like Constantine has a question. Constantine, if you'd like to unmute yourself, otherwise I can go ahead and summarize.

No, it's all good. Thanks so much for your talk, really enjoyed it. Can you talk about your thought processes or criteria for deciding like which diseases to tackle? It seems like there's an interesting like sweet spot to identify diseases that have specific genetic causes, but you can still fix them later in life. They aren't lethal up front, and you don't have irreversible changes that get locked in.

Yeah, that's right. So, I mean, there are a number of factors that go into choosing a particular disease target. One is what's the drug target? And the reason that the field started with monogenic disease, inherited disease, is that you can sequence the genome and you can pin the blame to a specific mutation within a specific gene. And if that was a mutation that disabled that gene, becomes a very simple hypothesis, which is gene X is broken in tissue Y. If I can deliver correct copies of gene X to that tissue, I can cure the disease. And in that case, the risk is not as much the disease biology, because you know what the faulty gene is. The risk is the delivery. That's been the big problem in the field for 50 years. So the way that we've been pursuing this within the company is that we start out with the rare disorder. Where the biology is very well characterized, where we know the faulty gene by definition. As a result, we know what the DNA medicine needs to look like. It's simply the correct version. And once we know that that delivery vehicle works for that tissue, we then start moving on to more complex, riskier diseases to treat where we may not know the drug target. So heart disease, congestive heart failure, you know, we have some hypotheses for drug targets for DNA medicines to carry in, but we're not quite so sure. You know, COPD, congestive pulmonary disease, you know, we know some potential drug targets, but we may not, we may not be right. So we want to make sure that we have a really good delivery vehicle before we start taking on additional risk with more complex diseases. The one that we feel a lot of confidence about is actually wet age-related macular degeneration in the eye, in that we have now two clinical trials ongoing where we've shown that the delivery vehicle is at least safe for delivery to these rare disorders. And, you know, fingers crossed, we'll be able to help those patients. There'll be efficacy. But if that delivery vehicle really works well in the eye, like it did in the nonhuman primate, wet age-related macular degeneration has a very well-defined drug target. Nobody knows exactly what caused wet AMD to begin with, but we know as a field that if you— I told you about those 10 injections a year of that protein therapy, What that protein therapy is, is it's a protein that binds to and inhibits vascular endothelial growth factor. And vascular endothelial growth factor, or VEGF, is a molecule that stimulates new blood vessel growth. And it's at elevated, you know, abnormally elevated levels in the eyes of wet AMD patients, which is why they end up with blood vessel growth into the retina, which eventually kills photoreceptors. So what we're doing is that we're taking DNA and coding one of those inhibitory proteins that binds to VEGF and blocks it. And we're taking that DNA, loading it into the virus, we're going to inject it into the eye, which we've already done in non-human primates and reported those results, very successful in primate. And we'll initiate that clinical trial later this year. And the thought or the hope is that now you've delivered the DNA that will lead that person to continuously secrete that anti-VEGF molecule for the rest of their lives, So they no longer have to go into the clinic 10 times a year to get poked in the retina with a needle. So the wet AMD is the first time we're going to progress towards a very large, more complex, more complicated disease target, but it's one for which there's a very well-defined drug target, which is why we're starting with it.

It's really insightful. Thanks so much.

So in the interest of time, there's 2 questions there. I'll read them back to you, David, then I'll encourage Doug Post and The other poster to jump in if they have any follow-ups. So given the uncertainty of what causes Alzheimer's disease, how would gene therapy be used to treat Alzheimer's?

Yeah, there's, um, the challenge there again is it's the drug target. Uh, there, you know, people thought it was beta amyloid, and you may remember that there were drug companies that were trying to develop antibodies against amyloid with the idea that if you, you know, if you could have an antibody, you would recruit the immune system to kind of clean up the amyloid. And unfortunately, you know, several of those clinical trials completely failed. So that meant that amyloid is not a good drug target. Now people are beginning to take a look at another target called tau that on the inside of the cell are these tangles of cytoskeletal protein called tau. And people think that these tangles, these abnormal form— formation of aggregates of this protein, are, you know, may drive Alzheimer's disease. And so now people are beginning to turn to that as a potential drug target. If that starts to work, then we'd have the basis of a gene therapy. We could deliver the DNA encoding some molecule that would attack these aggregates and eliminate them. But until there's some stronger indication within the field that tau is a good target, Alzheimer's is a pretty risky target for gene therapy. So fingers crossed, in the next 5 years we'll know whether or not tau is a good thing to target, and then I'd be a very strong proponent of beginning to develop a gene therapy for Alzheimer's.

Great. Looks like there's curiosity, interest in other areas. So is there any chance by which directed evolution in AAV can treat autoimmune disease or neurodegenerative diseases like multiple sclerosis?

Yes. There— so I'll give you an example. Rheumatoid arthritis, and also eczema and psoriasis, these are autoimmune disorders. And autoimmune disorders, as, as I think the, the person who asked the question well knows, those arise from a situation where the immune system abnormally begins to detect and to interpret one of our own proteins, one of our own molecules as non-self or as foreign. And so our immune system that gets, starts getting primed to attack that, that self-protein. And in the case of rheumatoid arthritis, that protein is inside of our joints. So you end up with joint inflammation. In the case of multiple sclerosis, it's in the myelin of our central nervous system. So you end up losing myelination, which leads to, to all the deficits you see in, in multiple sclerosis. So for some of those cases, people have identified drug targets that are responsible for the immune system's overreaction. In the case of rheumatoid arthritis, there's this inflammatory molecule called tumor necrosis factor alpha, TNF alpha. And there are drugs that are directed against TNF-alpha, drugs called Embrel or Humira that are developed by companies like AbbVie. And so unfortunately, these have to be injected at very high concentrations systemically, and they cause all kinds of side effects. And so a thought is that if you could load the DNA encoding one of those anti-TNF molecules into a virus, deliver it locally where the inflammation is occurring, you could have a little steady bioreactor secrete out small levels of this inhibitory molecule, which would turn off or dampen the immune response. So that's, you know, that, that, as at least as a thought experiment, it has been explored within the field. And again, what we need is a good delivery vehicle, and if we could generate a good delivery vehicle that could carry the virus to sites of myelination or sites of joint inflammation, then this could potentially be a viable approach for autoimmune disorders.