Hi. I’m Greg Petsko. I’m a professor of Neurology and Neuroscience at Weill Cornell Medical College in New York City, where I’m also the Director of the Alzheimer’s Disease research institute. And today I want to talk to you about a potential gene therapy for one of the worst diseases in the world. In fact, if there’s a worse disease I don’t know what it is and I don’t wanna know. It’s a disease called ALS — amyotrophic lateral sclerosis — or Lou Gehrig’s Disease, and we arrived at this gene therapy in a fascinating way, from studies that began in a model organism. So, let’s start by talking a little bit about this disease. During the 1920s, the New York Yankees baseball team was arguably the greatest team ever assembled, might still be the greatest team ever assembled. Virtually every member of that baseball team ended up in the Hall of Fame. And the lower of these curves, here, is the team batting average. And even in that collection of immortals, one player stood out. His average is the upper curve and his name was Lou Gehrig. And until 1927, he was the greatest ball player anybody had ever seen. And then, that year, he began to feel not quite himself. You can see that very rapidly, over a period of less than a couple of years, he ended up worse by far than all of his other teammates, and less than three years after he retired from baseball, because of his illness, he was dead. And the disease that killed him we now call Lou Gehrig’s Disease. Doctors call it amyotrophic lateral sclerosis or ALS. It involves the slow death of the motor neurons — those are the neurons that connect your central nervous system to your muscles, and thereby tell your muscles what to do. An ALS patient loses control of their muscles, eventually becomes unable to breathe or to swallow. Motor neurons from the upper part of the body to the lower part of the body are affected by this disease, a progressive, horrible wasting away that typically kills its patients in five years or less after diagnosis. It’s generally thought of as a rare disease, but it’s actually not that rare. There are about as many people who get ALS every year as people who get multiple sclerosis, but it’s considered an orphan disease because you don’t live very long with this disease, so at any given time there are only 30-50,000 people in the United States who actually have ALS, but there are about 500,000 people alive at any moment who will get it. As I said, if there’s a worse disease, I don’t want to know what it is. Like Parkinson’s Disease, like Alzheimer’s Disease, as I talked about in earlier videos in this series, 90% of cases of ALS are sporadic and idiopathic. They occur randomly and we don’t know what causes them. 10% are genetic, they run in families, and they are inherited in either a recessive or a dominant manner, depending on the gene and the mutation. There’s only a single drug that has been approved for this disorder and, to be honest with you, it doesn’t do much. We now know a number of genes that are responsible for the familial cases of the disease and the human genetics has been tremendously valuable, as you’ll see in a moment, in pointing the way towards what the molecular basis for the disease might well be. Discovery began in the 1990s with Bob Brown, who discovered that superoxide dismutase was mutated in a significant percentage of the familial ALS patients, and then rapidly after that more and more genes began to be discovered. We now have over a dozen of them and interestingly enough many of the ones that have been discovered lately, including the one shown in red here, FUS, about which I’ll be talking quite a bit… many of these genes are in fact RNA-binding proteins. And that points to a very interesting possible molecular mechanism for the disease. Here are, in fact, two of the ones that have been most important in helping us understand both the sporadic and many of the cases of the familial forms of the disease: TDP-43 and FUS. They’re both RNA-binding proteins, the RNA-binding domain is indicated, and the RNA-binding domains of these two proteins, interestingly enough, are not where most of the ALS-causing mutations occur. They mostly occur in either this peculiar domain, which is rich in glycine residues in both proteins, or in the nuclear localization signals that direct these proteins to remain in the nucleus. In fact, nearly all of the mutations in FUS, here, are in its nuclear localization signal, which is right here at the C-terminal end of the gene. If you look at patients who have mutations in these proteins, you find that they show a peculiar distribution of TDP-43 or FUS. Instead of being nuclear, as it’s supposed to be, the mutant proteins in these patients are found mostly in the cytoplasm, where it aggregates — you can see the aggregates there in those green dots in the upper right panel. And those aggregates are cytoplasmic and consists of both FUS, or TDP-43, depending on the patient, and RNAs to which it’s bound. So, one possible way that this disease might actually happen is by mislocalization of TDP-43 and FUS to the cytoplasm, where it sequesters in the cytoplasm RNAs that should normally be free to be translated or have other functions. And that’s I think not a bad explanation; it’s a toxic gain-of-function due to mislocalization. Now, how does this connect to the sporadic disease? It’s not entirely clear except that these aggregates, of TDP-43 especially, are in fact found in the sporadic patients, even though they don’t have the mutations in TDP-43. So, they’ve got something else that’s causing TDP-43 to mislocalize and aggregate, just like it does in the familial form of the disease. That might be environmental, it might just be age, it could be mutations and some other things that we haven’t identified. But whatever the reason, it seems like most cases of ALS are cases that involve RNA-binding protein mislocalization and dysfunction. Now, we were asked to study this and, in our lab, when we try to model a disease, we don’t do it in mice; we do it in yeast. Even though yeast doesn’t have a central nervous system, it’s been used successfully by us and by others to model a number of human neurodegenerative diseases because yeast, as a simple eukaryote, contains most of the important intracellular pathways and processes that might be messed up in diseases like ALS. Yeast does splicing of RNA, yeast does RNA trafficking and transport.. . many of the things that you might figure would happen in a human neurodegenerative disease, those pathways exist in yeast and could be probed by yeast genetics. So, we did this. We took yeast cells and we overexpressed either TDP-43 or FUS, mutant and wild type, and we found, to our delight, that when we did that, first of all, it was toxic to the cell. You can see the dead yeast cells in the model where we overexpressed these genes. But in addition, it mislocalized those proteins to the cytoplasm, where they aggregated in exactly the same way that they did in the human patients. So, even though yeast doesn’t have motor neurons, yeast is able to recapitulate the proteotoxicity of FUS and TDP-43 in a quite straightforward way. And in yeast you can do things that are very hard to do in some of these other organisms — mice and so forth. Namely, you can do full-blown genetic screens as easy as pie. And so we did a high-throughput screen asking very simply, if we took the 5,500 known yeast genes and, one by one, expressed them in yeast that already had FUS or TDP-43, and would therefore die from it, could we rescue the toxicity of FUS or TDP-43 by overexpressing these other genes. And typically when you do an experiment like that with 5,000 genes, you expect a few hundred hits and then you have to figure out what pathways they belong in and so forth. We got 5 hits. 5 genes out of 5,500. I’ve personally never seen a screen that tight; it’s really quite remarkable. And every single one of these genes that suppressed the toxicity of FUS when overexpressed, they all were RNA-binding proteins. And none of them, none of them, changed the expression of FUS. They weren’t genes that just turned off FUS expression, they were doing something else. Yet they could completely rescue the toxicity of FUS. This was tremendously exciting to us. What was even more exciting was that most of these genes have human homologues, and we rapidly looked at one of them. So, the best of the suppressors was a yeast gene called ECM32, and ECM32 has a human homologue called UPF1, which is involved in a pathway called nonsense-mediated decay that has been well studied by a good friend of mine Lynne Maquat at Rochester University. And Lynne kindly sent us human UPF1-expressing plasmids, which we put in yeast, and we found that just like ECM32, the yeast gene, could suppress toxicity, so too human UPF1 could suppress the toxicity of FUS in yeast. And by the way, it could also suppressed TDP-43. In fact, for the rest of the talk, pretty much everything I tell you about FUS will be also true of TDP-43, or if I talk about TDP-43 it will also be true of FUS. Interestingly enough, the protein had to be fully functional: if we deleted domains, it no longer rescued; if we put in a point mutation that knocked out its helicase activity, it no longer rescued; we needed active UPF1 in order to rescue the phenotype. Now, the fascinating thing about the rescue of FUS or TDP-43 toxicity in yeast by either ECM32 or UPF1 is that it doesn’t change the mislocalization or aggregation of those nuclear proteins. TDP-43 and FUS remain in the cytoplasm, remain aggregated… we are not in fact reversing the molecular phenotype of the disease… we seem to be bypassing the phenomenon, and that’s really exciting, and it suggested that what we found in yeast might also work in human neurons. So, we turned to a human neuron model of the disease, and this was developed by Anthony Batarase in the lab of my friend, Steve Finkbeiner, at UCSF. And basically, it’s a model in which you take primary neurons in culture and you follow each neuron, microscopically, other time — 150-200 hours — and you look at its survival. And if you do this with 100s-1000 neurons, you get really good survival statistics and you can then see how those are modified by drugs, by genes, by disease models and so forth. So, using this, one can construct a model of ALS. One takes those primary neurons in culture and one puts in mutant TDP-43 or mutant FUS and one sees what effects it has on the survival time of the neuron and, in fact, as you might imagine, these proteins compromise the survival of the neurons. They become more death-prone much more rapidly. Interestingly enough, in this model, also, the proteins, TDP-43 and FUS mislocalize to the cytoplasm and form these dense aggregates just like they did in the yeast model of the disease, and just like they do in human patients. So, this looks like a pretty faithful model. Given that fact, we then set about using it to look at UPF1 potential treatment of ALS. Now, I need to show you how we actually study that. We follow the survival time of hundreds of neurons and we look at average behavior, but we express it in terms of what is called a Kaplan-Meier curve. It’s a curve that the cancer people like to use. It plots the risk of death as a function of time. So, normal neurons, in green here, they have a low risk of death over time, but as you start doing bad things to them, like overexpressing TDP-43 or FUS, or various mutant forms, you see you greatly raise the risk of death. So that, in the worst cases, by 150 hours, instead of most of the cells being alive, most of them are dead. And what you’re looking for in a curve like this are treatments that take these upper curves and bring them back down to baseline — that’s what we hope to see with the things we’re trying to do. So, first of all, we asked the question, is UPF1 a general survival factor for cells? And the answer is, no, it has no effect one way or another on the survival of neurons in this kind of model… might as well not be there. Okay, fine. Now, what about when you use the vulnerable neurons that have TDP-43 or FUS expressed? Well, it turns out that if you overexpress UPF1, you go right back down to baseline. So you restore normal survival times to both TDP-43- and FUS-expressing neurons by overexpressing human UPF1, and you only have to overexpress UPF1 2-fold to see this effect. And it is, in fact, dose-dependent. You can show that if you don’t quite overexpress enough of it, you don’t do too well, but if you overexpress 2-fold of more, you get right back down to baseline. And that’s really very encouraging. What else can we see? Well, one of the thing we can see is that, in this neuronal cell model of ALS, when we overexpress UPF1 we don’t change the localization of FUS and TDP-43 and it still aggregates. So, exactly the same thing we saw in yeast. Remember, in yeast, we didn’t change the mislocalization, we didn’t change the aggregation, we bypassed those things. And we’re bypassing them here in a neuron as well. The yeast model turned out to hold up remarkably well. What about other forms of ALS that don’t involve RNA-binding proteins? Some of the familial forms of the disease don’t seem to be due to RNA, and we wondered what would happen if we tried to rescue one of those. So, we made a model of the superoxide dismutase form of ALS, the earliest genetic form that was discovered. That’s not an RNA-binding protein, and we can’t rescue it; overexpressing UPF1 doesn’t do a thing. So, whatever we’re doing, it’s specific to the RNA-binding form of the disease, which, happily for us, is the most prevalent form of the disease, but that also makes sense because, let’s face it, we’re dealing with an RNA-binding protein. And different diseases aren’t rescued either. Here we tried to rescue Huntington’s, which is not thought to involve these kinds of RNA-binding proteins, and we can’t rescue a Huntington’s Disease model with UPF1; it doesn’t do anything. Okay, so given these results, we started to ask ourselves what I think is a really interesting and exciting question. That is, could UPF1 be a drug? Not just in cells in culture, but eventually in people? And, to do that, we would have to find a way to overexpress UPF1 in motor neurons in people, and now we’re talking of course about gene therapy. And gene therapy has had a colorful history and it’s not always been a nice history. Early attempts at gene therapy frequently led to toxic events in people, but more recently a set of new viruses have been developed, the so-called adeno-associated viruses, or AAVs, and those viral vectors seem to be very well-tolerated in human beings, in fact one of them’s been approved for treating lipoprotein lipase deficiency in Europe. It’s a small virus, so you can’t pack big genes in it, but we had a friend, Ron Klein at LSU, who was able to figure out how to package UPF1 into AAV, and then the trick is let’s put that gene in the virus, put the virus to an animal model of the disease, and see if we can get anywhere. Well, first of all, the great thing about the particular strain of AAV that we used, which is AAV9, is that it’s highly tropic to the central nervous system and, in fact, it loves motor neurons. So, here you can see an experiment in which you simply package GFP and ask, by imaging, how well do you transfect the spinal cord when you actually inject into a vein? And the answer is that spinal cord distribution of this virus is terrific. So, this is a very encouraging result — this is in a neonatal rat — and it allowed us to proceed to looking at a neonatal rat model of ALS. That model was developed by Ron Klein and his associates down at LSU, and the way it works is you deliver into the temporal vein of a newborn rat TDP-43, toxic forms of TDP-43, and you see what happens. And what happens is virtually all the rats become paralyzed and they become paralyzed in a matter of a few weeks. You can see the paralysis here, especially forelimb paralysis, very rapidly developed. But what happens in this model if, also early, you inject AAV9-UPF1. And the answer is, when you do that, the paralysis is prevented. It’s really quite striking; the animals have full use of their forelimbs. Now, this is a young rat model of the disease and we’re giving the gene early. So, that’s not quite the same thing as the sporadic human disease, which develops when people reach their 50s and 60s. So, a more realistic model might be an adult animal model of the disease, and those are very hard to deal with because they’re very aggressive models of the disease, but we decided to try it anyway. And so, in a second experiment, we looked at a small number of rats and we did the same thing. These are adult rats, now, that are expressing TDP-43 and they rapidly die from paralysis that affects their ability to breathe. They’re typically all dead by 5 weeks after you trigger the onset of the disease… sorry, by 3 weeks after triggering the onset of the disease. But if you give UPF1 in an AAV vector, then the rats survive at least 5 weeks, and in fact some of them survive longer than that. The overall survival was at least 30%. And this is very encouraging to us. First of all, it’s very hard to see this in any kind of model of ALS, this is a big effect. In humans, this would amount to an increased life expectancy of 2-3 years, so that’s a lot. But this is also a really crude experiment. We’re giving the virus intravenously, we’re not paying any special attention to target, with really high doses of the virus, the exact regions of the body that we want to — we can do better than this, I think. We can do a lot better and we’re in the process of trying that now, but these results by themselves are so encouraging that they’ve caused us to do more than that; they’ve caused us to go on and contemplate a human clinical trial of AAV9-UPF1 for ALS. And that trial is being planned and I’m hoping that, within a year of the time I’m recording this video, it’ll actually be starting. So, the story I’ve tried to tell you is, I think, a fascinating one, because we start with a model organism, with yeast. And model organisms have gone somewhat out of fashion in biomedical research these days, and I think that’s a crime. Most fundamental discoveries of cell biology were made originally in model organisms, and even in complex human diseases, as I hope I’ve shown, they have a lot to teach us about how the disease really works. We would never have been able to find UPF1 as a disease-modifying gene without the yeast genetic experiments that started this project. So, we’ve been able to go from yeast genes all the way down to a potential gene therapy that seems to work in rats and that might well work in people, at least we’ll have a chance to try it in people and find out. And although this will come 75 years too late for Lou Gehrig, maybe it’s something that might help the 1000s of people who are afflicted by the disease that still bears his name. It takes an army to tackle a disease and we’ve had fabulous collaborators. I’ve mentioned them as we’ve gone along. They have done remarkable work. The discovery, the fundamental discovery, that UPF1 suppressed the discovery of TDP-43 and FUS was made by our postdoc, Shulin Ju, a joint postdoc with me and Dagmar Ringe, and without Ron Klein’s willingness to try to package an impossibly packaged gene in AAV, we wouldn’t be where we are today. It’s a great story; these have been great people to work with. The last chapter still has to be written. Stay tuned. Thank you.