Welcome to the Hans Popper state-of-the-art talk. To remind you who Hans Popper is, he was the driving force in founding this society together with Sheila Sherlock helped found hepatology and also a driving force behind establishing our journal. It's therefore a particular honor to get to give the Hans Popper state-of-the-art talk which this year will be given by Holger Willenbring. Dr. Willenbring has been a huge contributor to our field and rather than go over his pedigree, I'd like to advance my slide, there we go, and tell you just a couple things about some of his enormous contributions. He and his colleagues were the first to show definitive lineage tracing evidence that new hepatocytes derived from pre-existing hepatocytes in liver homeostasis and most models of liver regeneration in mice. That the mouse induced pluripotent stem cells can differentiate into hepatocytes that can repopulate the mouse liver and reverse liver failure. That human fibroblasts can be reprogrammed into hepatocytes that function and proliferate similar to primary human hepatocytes after transplantation into mice. And the first to perform in vivo reprogramming in the liver to generate therapeutically effective hepatocytes from pathogenic myofibroblasts. But wait, there's more. He defined the plasticity of hepatocytes by showing that they can build a functional biliary system from scratch in a mouse model of allergy syndrome. Uncovered the flip side of hepatocyte plasticity by showing that mouse hepatocytes can give rise to cholangiocarcinoma. Defined the extent and location of clonal expansion of hepatocytes in mouse liver showing mid zonal predominance in homeostasis and local predominance in injury. Many of these studies were done using AAV adenovirus associated viral vectors as tools. Culminating in why he is the perfect person to give today's talk about next generation liver gene therapy. With no further ado, Holger Willenbring. Thank you very much Nora Theroux for the invitation, it's definitely a great honor, feels a little bit undeserved. I think I have a little bit of a case of the imposter syndrome in particular because of the youthful picture that I submitted. And of course thank you Lori for the beautiful introduction, too much praise and I realize I've done way too much stuff in mice, so I need to shift gears a little bit and you'll see that today's talk is an effort in that direction. It's actually all preliminary data, so it's unpublished, I thought, somebody told me yesterday Hans Popper was a real progressive thinker, so I thought maybe I should push myself also a little bit. So you might have to take some of it with a grain of salt, it still has to withstand the peer review process to some extent. Okay, so I'll get started. So the title of my presentation is Next Generation Human Liver Gene Therapy, it's a pretty ambitious title, hopefully I can deliver partially to fulfill that promise a little bit. What I want to add first is that this is all, as Lori indicated, using adeno-associated viral vectors or short AAV vectors. You're all aware that AAV vectors have revitalized the gene therapy field, there's hundreds of clinical trials ongoing right now across all disciplines, there's five FDA approved therapies, including two in the liver, one for hemophilia A, one for hemophilia B. Before I get further into the AAV business, I want to maybe explain the title a little bit. What I'm trying to achieve, since I don't have clinical trial experience, what I'm trying to achieve is to discuss experimental tools and strategies that I think can address some of the limitations and challenges of current AAV gene therapy in the liver. One really big challenge in the clinical AAV space is the lack of reliable predictive models, and we'll address this in the first part of my talk, that's entitled To Maximize Efficacy and Safety of a Palsy-Targeted AAV Gene Therapy Through Improved Preclinical Prediction. And in the second part of my talk, I will tell you a little bit about our efforts to expand the reach of AAV liver gene therapy beyond the Palsy-specific diseases. Before I get to the actual data, just a few reminders about the basic biology of AAV. AAV is a naturally occurring virus with a single-stranded genome of about 5 KB in size. The genome resides within a protein shell that's called a capsid. And if I can get the mouse to work, that's called a capsid, and this capsid determines the cellular tropism of AAV vectors or AAVs in nature as well. And there are many different types of serotypes, and many of them differ very vastly in the tropism they display. In recombinant AAV vectors, the genes for the capsid and for the ability to replicate are moved from cis to trans to make space for an expression cassette. The capsids can be engineered to alter that tropism. There's numerous capsid engineering methods. Some prominent ones are listed here. One that's very developed and widely used is random shuffling of DNA fragments from different capsid serotypes. Another method is to display peptides on the outer surface of the capsid in an exposed region. Then there are also more rational approaches, for example, mutagenesis of certain nucleotides within the capsid gene or domain swapping between capsid serotypes. More recently, also, machine learning has been added to the portfolio to make the approach more efficient and comprehensive. I also want to briefly review how an AAV physically and functionally transduces a cell, that is, how it delivers a cargo to the cell and how that cargo is then expressed. The AAV attaches to the cell membrane. This is facilitated by primary glycan receptors and by secondary protein receptors. Then there's the internalization step via endocytosis. Then the capsid is packaged in endosomes from which it eventually escapes. And it makes its way to the perinucleus space where multiple viruses are accumulating. And then via active transport, the virus enters the nucleus. Only there, the genome of the virus is released from the capsid. And then this critical step of conversion of the single-stranded genome to the double-stranded genome happens. This is a rate-limiting step for transcription. And it can actually be bypassed by using double-stranded vectors instead of single-stranded vectors so that this second-strand synthesis step is not needed anymore. Eventually, the genome is formed into an episome, which is a critical step to achieve persistence of the AAV in the nucleus in the absence of genomic integration. I also attempted to summarize the clinical trial experience in the liver from the perspective of capsids. This is what this plot represents. So be patient with me as I guide you through it. So there have been 12 different capsids being trialed so far. They are listed here on the X-axis. Nine of them are wild-type capsids, and three of them, the ones on the right side here, are engineered capsids. Each of these little dots here is a clinical trial in which patients have actually been dosed with viruses. As you can see from the number of dots, the most commonly used capsids are the AV8, AV5, and AV6 capsids. In fact, the two FDA-approved therapies that I mentioned in the beginning, targeting hemophilia A and B, use the AV5 capsid. On the Y-axis is the maximum dose that was applied to patients in each of these clinical trials. And the question mark down here is just reflecting that the dose that was used in this particular trial using an AV3B capsid was not published. The most critical part of this plot is probably these triangles. These are actually not gene therapy trials for liver diseases. These are trials for myopathies. And the reason I'm adding them to this plot is because they used the highest doses ever used in humans for AV8 and AV9. And unfortunately, they were associated with four deaths of patients in the course. So if you will, one can use this tragic, lethal complication as an indication for what is the safe dose that is tolerated in humans when AV is applied. I try to indicate this here with this dotted line. So this really produces a challenge as to maximize the therapeutic effect within this narrow dose limit to not exceed the safety margin. This raises then the question of which is the best capsid to transduce the parasites in patients. Answering this question has been pretty difficult to achieve because there's a lot of heterogeneity in clinical trial design and preclinical models have proven to be fairly difficult, especially with regards to predictive validity. Normal mice are not very suitable for these kind of studies to test AV capsids with an eye towards clinical translation because there is a lot of species-specific differences between AV capsid transduction, so the results cannot be readily translated to humans. So what about human haplocytes in mice or monkeys themselves? So there's a lot of AV capsids that have been tested or compared against each other in either of these preclinical models. And I indicated them here with different colors in this legend. And looking at the relative transduction, haplocyte transduction results of these different capsids as indicated here with high and low transduction, so left side is high, right side is low, in both models. You can hopefully appreciate that these results are inconsistent not only between human haplocytes in mice and monkeys, but also within the respective preclinical models. So there's really confusion about which capsid would be best for use in human scenarios. Based on this confusion or prompted by this confusion, we decided to actually test AV capsid performance in the human liver itself. For this, we tapped into this technology that you're probably aware of, normothermic machine perfusion, and acquired a system to do this exclusively for research. Normothermic machine perfusion, for those of you who are not aware of it, is a method similar to exocorporeal membrane oxygenation or long-term heart-lung machine perfusion. So basically, there's a pump that pumps oxygenated blood through the organ. The machine we're using provides anatomically correct vascular connections, so it's the hepatic arteries and the portal vein are going into the liver, and the hepatic veins and the inferior cava are coming out of the liver. There's also bile drainage as in the normal body. As you can appreciate from this image, the liver is perfused with human red blood cells, albumin, nutrients, some drugs, and some metabolites. And here's a depiction of a liver before and after initiation of the perfusion. You can see that the perfusion is very homogeneous in this particular setting. We are also adding hemoconcentration to drag out the process of normothermic machine perfusion beyond what is done typically in a clinical setting. So we can perfuse the liver readily for five days with this moderate dialysis function provided by hemoconcentration. And importantly, during this perfusion time of five days, the liver is in a homeostatic state, so it really allows fairly bias-free experimentation. And this homeostatic state is indicated here by normal lactate and glucose metabolism, bile secretion, and really no sign of active, of ongoing liver injury in the liver. We also optimized our AV vectors. I indicated that self-complementary or SCAVs don't require second-strength synthesis, so they can express their cargo faster, and this is what we tapped into for this purpose. So this experiment shows you that there's almost no difference in the level of the parasite transduction indicated here by immunofluorescence in red or based on flow cytometry here on the right side between two days or 48 hours after virus injection into a mouse or seven days. So we have a pretty good system acting very quickly, which really allows us to minimize the NMP or normothermic machine perfusion time to two days, if you will, although we typically take it out to three days. We also multiplexed our approach to be able to do site-by-site comparison of different capsids. For this, we associated each capsid with a different fluorophore and with a different barcode. And then we focused, for the purpose of this talk, our initial attention on these capsids. The reason for that is the eight, five, and six capsids, as I showed you early on, are the most commonly used capsids in clinical trials. And then we focused on this engineered capsid, LK03, because it really shows a lot of promise in a recent clinical trial and is one of the emerging capsids in the field. We made viruses with these capsids and then co-injected them at the same dose into mice, A, normal mice, and B, mice humanized with human heparocytes. And we found that the tropism of all of these capsids, except for LK03, was very much directed towards mouse heparocytes, which is not surprising, but really nicely illustrates the point I made earlier, that there's substantial species-specific difference between capsid tropisms. So finally, this is the first piece of data from human liver. We initially, in the human liver, we focused on three capsids in side-by-side comparison. These are the AV8 and 5 capsid, and the challenger capsid, if you will, the LK03 capsid. We made viruses with these capsids and then co-injected them together into a normal liver at a very high dose, the same dose for each of these viruses. Because we expressed the unique fluorophore for each capsid, as well as the barcode for each capsid from a ubiquitously active promoter, we were able to perform comprehensive profiling of the transduction in the human liver using single-cell RNA sequencing, so this is single-cell RNA sequencing results. And the most important result is down here. We found that the LK03, so the challenger capsid, was most efficient at transducing hepatocytes in the human liver. It transduced almost 27 percent of all hepatocytes in the human liver. If you remember, there's 250 billion hepatocytes in the human liver. This is almost one-third of it, so it's a pretty high number achieved by this dose that we used in the human liver. The AV8 capsid was much less efficient at 6% transduction than the AV5 capsid. As you remember, the one that's in the clinic, FDA approved, was the least efficient at 0.4% transduction in the human liver. We also observed that the AV8 capsid has a tropism for sinusoidal endothelial cells in the liver, as well as macrophages. Next we validated these RNA-based results at the protein level using flow cytometry for the capsid-specific fluorophores that we used. Just very briefly, we found that indeed also at the protein level, the AVLK03 capsid was the strongest with regards to hypoxia transduction, almost 20% transduction based on this result. The distribution of the AV8 and AV5 capsid followed what we had seen in single-cell RNA-seq, and then we also looked at the AV8 tropism for endothelial cells and monocytes macrophages, and it was also preserved at the protein level. Next we wanted to validate these results and expand them a little by adding the AV6 capsid, which is another capsid that's commonly used in clinical trials. In this particular case, we reduced the dose by almost tenfold, and again co-injected these viruses into a normal human liver, and again used single-cell RNA sequencing to analyze the tropism of the respective capsids. And as before, we found that the LK03 capsid was the strongest with regards to hypoxia transduction, and the AV5 was the least efficient, and the AV8 and AV6 capsids were actually pretty much equal when it came to hypoxia transduction. We then broke down the hypoxites in this particular experiment based on functional zonation, distinguishing pericentral hypoxites in blue and peripartal hypoxites in yellow, with the idea that maybe there's capsid zonation in transduction. And indeed, we found that the AV8 capsid has a very mild peripartal preference, so a peripartal zonation of AV8 capsid transduction, which is quantified here, so it's not very dramatic, but the peripartal hypoxites are the orange bar, and the pericentral hypoxites are the blue bar. We found a little bit stronger zonation for the AV5 capsid, which prefers hypoxites around the central vein. This is quantified here, so the reverse tropism, if you will, when it comes to zonation. We validated these results, which are again at the RNA level, at the DNA level, using single molecule fish for these viruses, the AV8 virus in green, the AV5 virus in red, and we indeed found that there was more AV8 virus in the peripartal area, whereas there was more AV5 virus in the pericentral area. And finally, for this experiment, we also validated the RNA-based results for the four capsid comparison at the protein level, again using capsid-specific fluorophores as a readout and flow cytometry, and these results basically confirmed everything we'd seen so far, with AVLK03 dominating the pack, and AV8 and AV6 being pretty similar in transduction efficiency. We also looked at the non-parenchymal cells again, and indeed, we found again that AV8 has the ability to transduce endothelial cells, whereas the new player, AV6, seems to be best at transducing macrophages. We then combined three livers together, the results from three livers together, at the level of single-cell RNA sequencing, and we found that, interestingly, of all the endothelial cells analyzed, AV vectors only transduce sinusoidal endothelial cells, and of all the macrophages analyzed, AV vectors transduce preferentially cupra cells. And finally, we wanted to see whether steatosis has an effect on AV transduction in the human liver, because, as you can imagine, steatosis is likely a common comorbidity in patients considered for gene therapy trials. And we indeed found that, by breaking down the hepocytes again, this is a fatty liver or a steatotopic liver, by breaking down the hepocytes again into pericentral and periportal cells, we indeed found that the tropism of the AV5 capsids for pericentral hepocytes was enhanced, which is quantified here. So the blue is the pericentral hepocyte compartment, and the orange is the periportal hepocyte compartment. So the tropism of the AV8 capsid for periportal hepocytes was maintained, but not further enhanced. We then validated this again at the DNA level, using probes for the respective viruses, so AV8 in green, AV5 in blue. And interestingly, we found that most of this enhanced tropism of the AV capsid for the pericentral hepocytes is really due to uptake of the virus into the cells, as you can see from this massive accumulation of AV5 DNA in these pericentral hepocytes. We believe that, based on these results, and also based on the observation, which I'm not showing here, that the steatosis was limited to the pericentral area in this liver, that fat accumulation actually induces transduction of hepocytes by the AV5 capsid. So this leads me to the conclusions for the first part of my talk, and we can predict clinical performance of existing capsids, and then looking to the future, develop new capsids within the human liver that are exquisitely targeted to human cells, and are looking forward to doing that. As of now, with the results in hand, we can come up with a ranking for human hepocyte transduction in the human liver, which is LK03 being the most effective, followed by AAV8 and AV6, and then AV5. We observed a peripolar hepocyte tropism of the AAV8 capsids. This could be potentially exploited for therapy, so there are some diseases, obviously, that are zonated, like OTCD and PKU, so these might be good targets for AAV8 gene therapy. And then we observed a pericentral hepocyte tropism for the AV5 capsid, which is enhanced by steatosis, and this could be useful for AIP, and potentially also looking to the future, steatotic liver disease. We saw tropism of the AAV8 capsid for sinusoidal endothelial cells, which, as you know, are the source of factor VIII, which is defective in hemophilia, so this could actually be developed potentially into a more physiological gene therapy for hemophilia, and then potentially also to restore endothelial function in cirrhosis. And finally, we saw Kupfer cell tropism of AV6, which may play a role in the immune response of this capsid and vector. So for the second part of my talk, I want to talk mainly about non-hepocytes, so going away from hepocytes, and here is basically everything in mouse again, because the existing literature is really limited to mouse, so I summarized the existing literature here with regards to AV transduction of liver cells beyond hepocytes. In the left column are all the different capsids that have been tested, and in the columns next to it are hepocytes, then biliary epithelial cells or cholangiocytes, stelate cells, macrophages, and endothelial cells, and you can appreciate from this heat map style kind of display with red being high, pink being low, and blue being no transduction, and white being unknown that A, hepocytes are very efficiently transduced, and B, there's not much known really about any of the other cell types, and the little bit that is known is rather underwhelming with fairly low transduction efficiency. So we decided to do a screen, an unbiased screen for an AV capsid that can potentially transduce some of these non-hepocytes. This is the principal design of the capsid library. So there's a fluorophore that's expressed by all the vectors in the library. This is 100 vectors in this library, naturally occurring capsids as well as engineered capsids, but each capsid is associated with a unique barcode. And we injected this library into mice, into normal mice, and then we developed a fax protocol to isolate the different liver cell types from the same mouse, so from a single mouse, including hepocytes, biliary epithelial cells, stellate cells, macrophages, mesenchyme, and endothelial cells. And then within these populations, we sorted for cells expressing the vector-derived barcode, and then we found by RNA and DNA barcode sequencing the following pattern. Most of the vectors showed a pretty good relationship between DNA and RNA. There were a few exceptions, like AV5. AV5, as we've seen in the human liver, can deliver a lot of DNA, but doesn't really express very well, which translates into a negative RNA-DNA ratio here, and we were pleased to see that AV6 transduces stellate cells, as we had previously reported. But the most interesting finding was this one, the transduction at a very high level of biliary epithelial cells or cholangiocytes. And to just put this into perspective, these capsids, AV8 and Rhesus10, are capsids that have previously been reported to have some tropism for cholangiocytes, and you can appreciate that the tropism of AVBEC, as we call this new capsid, is much higher. We validated these results by barcode single-cell RNA sequencing. These are the cells we captured. We left the parasites out. The purple dots are all cells transduced by any of the vectors in the library, so pretty good distribution, and the green dots are all cells transduced only by the new capsid AVBEC, so it looks like it's covering much of the transduction of biliary epithelial cells by the library. This plot shows you that this new capsid AVBEC is pretty specific for biliary epithelial cells, and this again confirms what I showed you based on the next-generation sequencing data in the previous slide, that AVBEC transduces biliary cells much more than the previous contenders AV8 and Rhesus10. Of course, we validated this result at the protein level, and here are the results, both immunofluorescence as well as flow cytometry. This shows you that AVBEC—this is a single vector now, not the library anymore—AVBEC can transduce biliary cells very well, but it also transduces the parasites and macrophages. This is in the normal mouse, and we also tested this new capsid in mice with cholestasis and ductal reaction induced by DDC diet, and you can see that the vector can also transduce biliary epithelial cells within the ductal reaction, as you can see here, and interestingly, the tropism for the parasites as well as for macrophages is decreased under these conditions. Then we wanted to see whether we can get the off-targeting in normal mice under control. For this, we swapped out the ubiquitous promoter with a PROM1 promoter, so a biliary promoter, and added binding sites for microRNAs that are expressed in hepacites as well as in macrophages to suppress the message in these cells, and indeed, this worked very well. So the normal construct transduces a lot of hepacites on top of cholangiocytes, but this retargeted construct doesn't, so it's very specific for cholangiocytes. It may be a useful tool for the field going forward. And finally, we have an eye towards translation of this method, and I can't show you any data confirming that the tropism of AVBEC is conserved in the human liver yet, but we're working on it. But this is just an intermediate result with human biliary organoids derived from human cholangiocytes, and you can see that AVBEC is the most efficient at transducing these organoids, both at immunofluorescence as well as mRNA level. So in summary, I show you that barcode-based capsid screening is feasible to identify capsids with non-hepacite liver cell tropism. There's good correlation between next-generation sequencing and single-cell RNA sequencing. We found candidate capsids enriched in stellate cells, mesenchyme, and endothelial cells, and we found and validated a capsid that can efficiently transduce cholangiocytes in ductile reaction. And this, in my eyes, kind of raises now the possibility of biliary gene therapy, potentially for genetically encoded biliary diseases, but potentially also to reprogram the ductile reaction, if you will, which is derived from cholangiocytes as well as hepatocytes in diseases like alcoholic hepatitis by delivering hepatocyte transcription factors. And in my last slide, I want to thank all the people that contributed to this work. I highlighted the individual trainees on the slides that present the data. I want to further highlight our key collaborators, Garrett Roll, who's a transplant surgeon at UCSF, and then the UCSF Liver Center has been instrumental in making all of this work possible with all the different cores, and I want to specifically call out the core directors, Jackie Maher for the Cell Biology Core, Aris Matis for the Pathology Core, and Bruce Wang for the Genomics Core. I also should not forget our long-term collaborator and AAV pioneer, Dirk Grimm, and last but not least, our funding, dedicated funding by NIDDK and NIAAA, as well as general funding for the UCSF Liver Center from NIDDK. Thank you very much. That was a fabulous talk. Thank you so much. I will ask all of you to cast your eyes on the slide up here and to help me congratulate Dr. Wynne Arias, who's sitting here in the middle of the room. In the meantime, if the moderators for the Basic Science Symposium will please come to the stage.

Dr. Holger Willenbring

liver gene therapy

AAV vectors

transduction efficiency

hepatocytes

biliary epithelial cells

genetic liver diseases