Neurodevelopmental disease: Drug Discovery in Neuroscience

Top

00:00:06.13 So, my name is Ricardo Dolmetsch and I'm the Global Head of
00:00:11.04 Neuroscience at Novartis. And today I'm going to tell you
00:00:13.28 something about the efforts that we've been making to try to develop new
00:00:17.13 drugs for diseases of the central nervous system.
00:00:20.19 Before I start, I should tell you something about myself.
00:00:23.25 So there are maybe two things you should know, the first
00:00:26.05 is that I'm originally from Colombia. And so what that means is
00:00:29.12 my mom calls me twice a week. She's been doing this for the last
00:00:32.27 30 years. The second thing you should know is that I have a son
00:00:36.18 who has autism, and so I've spent most of my career trying
00:00:40.27 to understand the underlying cellular and molecular basis of
00:00:43.27 autism and other neurodevelopmental disorders. So when my
00:00:47.28 mom calls me, she often delivers what I describe as
00:00:50.19 damnation with faint praise. So a few years ago, she called me up
00:00:55.27 and she said, I'm amazed at how much progress you're making in
00:01:00.25 autism. Every day you discover something else that causes it.
00:01:03.24 And it didn't feel like success, in fact it felt a little like this dog
00:01:10.10 in this picture, which has the cookies but can't open the bag.
00:01:15.27 And so what I'm going to tell you about today is some of our
00:01:20.28 efforts to try and open that bag. So our mission is to develop
00:01:25.04 medicines that reduce suffering by treating disease of the
00:01:28.02 nervous system. And there is a great deal of suffering to go around.
00:01:32.07 So depression, 50,000 people a year die from suicide in the U.S.
00:01:38.25 Every year. Alzheimer's disease, there will be 115 million people
00:01:44.17 with Alzheimer's by the year 2050. This is an existential
00:01:50.15 threat to our medical system. We will not be able to pay
00:01:53.25 for those people. It will cost us more than a trillion dollars
00:01:56.25 to take care of them. Schizophrenia, there are about 3 million
00:02:02.12 people in the U.S. About 300,000 of those are in prisons and homeless
00:02:06.26 shelters, because that is how we take care of our schizophrenics.
00:02:09.19 Addiction, there are 30 million people in the US and in Europe.
00:02:14.12 In Massachusetts, where I live, there were more than 1,000
00:02:18.27 deaths last year. More than 100,000 visits to the emergency
00:02:22.05 room from overdoses. And then autism, which is very close
00:02:25.21 to my heart. About 1 in 150 children will require life long care.
00:02:30.12 And we have no treatments. In fact, we have no treatments
00:02:33.11 for any of these things, no cures. So the question is why?
00:02:40.00 And I should just say that we have a very proud history of
00:02:42.18 developing medicines that have improved people's lives
00:02:46.16 in neuroscience. But things have been rough over the last
00:02:50.15 few years. And I think this is illustrated in this study from
00:02:57.08 Michael Ringel and his colleagues, and what you see here
00:03:02.03 -- So they were trying to understand what it is that makes
00:03:05.00 a drug company successful. And what they found was
00:03:09.18 that in fact, among all the different disease areas, the only
00:03:13.04 thing that was negatively correlated with your success
00:03:15.14 as a drug company was if you were in neuroscience.
00:03:17.28 So of course, this is the last thing you want to hear if you're the
00:03:20.07 head of neuroscience at any major drug company.
00:03:22.12 And so one of the things we set out to understand
00:03:24.22 was why. Why is it that drug development in neuroscience
00:03:27.09 is so hard? The first problem is that we select diseases
00:03:32.23 that we just don't understand very well. So part of this comes from
00:03:37.20 this idea that we need to develop blockbuster drugs.
00:03:40.03 And so almost by definition, any disease that affects
00:03:44.16 a large number of people is not a single disease but
00:03:46.28 a whole set of diseases. And so for example, there are
00:03:49.21 many different kinds of depression. There are many different
00:03:51.29 kinds of schizophrenia. And we need to subdivide those
00:03:54.18 diseases into smaller indications so that we can get
00:03:57.28 at the underlying biology of the disease. The second
00:04:01.05 problem is that we just don't have very good preclinical
00:04:04.11 models. So our preclinical models just don't predict
00:04:08.10 whether something's going to work in humans. And this has
00:04:10.25 of course held us back for a long time. The third issue is
00:04:15.27 that drugs in neuroscience have to be very safe and they
00:04:21.02 have to get into the brain. And they have to be very safe
00:04:23.10 because we're mostly giving them to people who are
00:04:25.04 comparatively healthy, and we're often going to be giving it to them
00:04:28.02 for the rest of their lives. And of course they have to get into the
00:04:30.17 brain, which is just something you don't have to do for most
00:04:33.16 other drugs for most other diseases. And then the final problem is
00:04:40.17 that clinical trials in neuroscience are very hard. And they're
00:04:44.12 really hard for at least two reasons. First of all, the patients
00:04:47.01 are heterogeneous, as I mentioned before. But the
00:04:49.12 additional problem is that the endpoints are really hard to
00:04:52.18 measure. Because we're always measuring soft things like
00:04:55.02 cognition and mood, and measuring those things reliably
00:05:00.11 without big placebo effect is very difficult. So if we're going
00:05:03.11 to succeed in making new drugs for diseases of the brain,
00:05:06.15 we have to improve in all of these areas. So let's just start
00:05:10.15 from the very beginning. How should we select our disease
00:05:15.04 and our targets? Well, in neuroscience, there are essentially
00:05:18.08 two ways in which we've succeeded in the past. The first
00:05:20.17 is what I would describe as serendipity. So you may know
00:05:24.23 about the story of chlorpromazine, which is the first antipsychotic.
00:05:27.21 And chlorpromazine was never developed as an antipsychotic,
00:05:31.14 it was developed as an antihistamine. And only later did a
00:05:35.28 surgeon in France, Henri Laborit, first discover that it was
00:05:40.01 a sedative and started using it in his patients before surgery.
00:05:43.00 And then he had two colleagues in psychiatry, Pierre Deniker
00:05:46.21 and Jean Delay. And he noticed that they had very agitated
00:05:50.18 patients, and he suggested perhaps that they should use
00:05:53.04 chlorpromazine in these very agitated schizophrenics.
00:05:55.26 And it was found that once the sedation wore off,
00:05:59.18 they also had fewer psychotic episodes and fewer delusions.
00:06:03.11 So that was purely serendipitous, this was the first antipsychotic
00:06:09.13 and then this led to a whole series of antipsychotics, each
00:06:13.00 one of which was slightly better than the previous one.
00:06:15.00 The second approach was to start with human genetics.
00:06:18.01 And again, the idea here is that if there is a genetic mutation
00:06:23.11 that causes disease, we can use that mutation as a way of
00:06:27.05 trying to identify a target that we can use to develop
00:06:29.09 a drug. So one of the reasons of course for why
00:06:33.22 this has actually become possible is that there has been
00:06:36.10 a dramatic advance in sequencing DNA. And so the
00:06:41.07 cost of sequencing a human genome has declined
00:06:43.15 dramatically over the last 10 years. In fact, the cost
00:06:47.25 has gone down almost 1,000,000 fold. I don't know of
00:06:50.13 anything else that has gone down by a million fold. And so
00:06:53.03 this has made it possible then to get sequences
00:06:55.23 for many, many people. And has led to the identification
00:06:59.05 of many genes that either cause diseases or alternatively
00:07:02.18 that increase the susceptibility for a disease dramatically.
00:07:05.23 So there are essentially three types of mutations that cause
00:07:10.00 neuropsychiatric diseases. The first are very rare mutations
00:07:13.18 that are very penetrative. And so by definition, if you
00:07:17.05 have this mutation, you will have a particular disease.
00:07:19.19 Of course these are very unusual. Then somewhat
00:07:23.21 more commonly, there are these low frequency copy
00:07:26.23 number variants. And these are microdeletions or
00:07:29.25 microduplications of genes in the genome. And they
00:07:34.03 are often not causative, but they dramatically increase
00:07:38.20 susceptibility for a particular disease. And then finally,
00:07:42.04 at the other end, there are these common single nucleotide
00:07:44.12 polymorphisms. These are things that we all share.
00:07:47.08 And they can slightly increase our risk for a particular disease
00:07:50.27 and the thought is that combinations of these three things
00:07:53.17 can lead to a disease. So the model is sort of illustrated
00:07:57.16 here. Which is that everybody has these single nucleotide
00:08:01.04 polymorphisms that provide our genetic background, and
00:08:04.04 then superimposed on them are these copy number variants
00:08:07.07 and rare mutations, and they tilt the balance until we have
00:08:09.22 a disease. So this all sounds great, now the question is how
00:08:14.17 do we go from these genes to a behavior, and ultimately to
00:08:19.10 a drug that can change that behavior? So conceptually, it's
00:08:21.16 all very straightforward. You have a mutation in the gene
00:08:24.07 and that mutation in the gene in turn changes the structure
00:08:26.25 of a protein that then changes the function of a cell.
00:08:30.02 And that in turn, changes the function of a circuit that
00:08:33.14 changes behavior. Sounds easy. It turns out to be really
00:08:38.09 difficult to make a link between all these different levels.
00:08:42.03 So the standard approach, really over the last 20 years,
00:08:46.24 has been to make a mouse. You identify a mutation in a
00:08:49.13 human, and then you introduce that mutation into a
00:08:53.23 mouse. And you take that mouse and you study it.
00:08:56.14 And in fact, we've learned a great deal about
00:09:00.14 disease and about the underlying basis of many biological
00:09:05.20 processes by studying mutant mice. At the same time,
00:09:09.18 we've also learned that mouse models, especially of
00:09:15.04 neuropsychiatric diseases, are just not very predictive
00:09:17.27 of what's going to happen in humans. And the reason is
00:09:22.09 sort of illustrated here. It turns out that if you're looking at
00:09:27.27 a mouse, it's very difficult to distinguish between a
00:09:30.25 schizophrenic mouse from a bipolar mouse from a
00:09:32.07 depressed mouse from an anxious mouse from na
00:09:33.28 autistic mouse from a mouse with ADHD or cognitive
00:09:37.20 impairment, or from a high functioning mouse.
00:09:40.08 You know, all mice are sort of cognitively impaired, that's what
00:09:43.14 it needs to be a mouse. So as a consequence, it's
00:09:47.20 not surprising that our ability to model these diseases
00:09:50.27 preclinically in mice is very limited. So this has led us to
00:09:54.28 try and look for other approaches that we can use to try
00:09:58.00 and model diseases preclinically. So mice aren't great models
00:10:01.23 for many neurological and psychiatric diseases. And
00:10:05.05 of course, one of the reasons for that is we diverged about
00:10:09.15 60 million years ago, and of course many things have changed
00:10:12.00 between mice and humans. Now, that's not to say that they're
00:10:18.07 completely worthless. I think that, as I've said before, learned
00:10:22.09 a lot about the underlying basis of many physiological
00:10:26.16 processes by studying mice. However, what we can say
00:10:29.25 is this, the further away you get from a mutation, the
00:10:32.17 less predictive the model becomes. And sort of the
00:10:36.15 reason that that's true is that a mutation often changes
00:10:42.03 the biochemistry of a cell, and if you're looking for
00:10:45.01 the changes in biochemistry, it is much more likely to
00:10:48.09 work in a human. Because that is likely to be conserved.
00:10:50.20 As you go further away from the mutation and you now start
00:10:53.12 to study the functions of a circuit or behavior of a mouse,
00:10:56.04 that is much less likely to be predictive. And so, I think
00:10:59.26 many of the mistakes that we've made in the past in
00:11:02.04 developing drugs is that we've relied upon behavior
00:11:03.28 as a way of trying to develop a drug, and that has turned out
00:11:07.16 not to work. So what we'd like to do is we'd like to be able to
00:11:10.09 develop drugs for humans using humans. The problem
00:11:13.14 is that in neuroscience, we can't take biopsies of people's
00:11:16.06 brains. Most people don't want a hole drilled into their heads
00:11:19.19 so we could remove some of their neurons. I often
00:11:22.27 tell people that I'm very jealous of my colleagues in
00:11:25.23 cancer, because they have this ability to look at biopsies
00:11:30.07 from tumors that behave to a first approximation in a lab
00:11:33.06 the way they behave in a patient. We don't have that.
00:11:36.23 So the next best idea was if we couldn't get a biopsy,
00:11:40.27 then maybe we could make one. So for a long time,
00:11:43.14 this was a pipe dream. How are we going to make a
00:11:45.17 biopsy from a patient with some sort of genetic disease?
00:11:48.14 And what really changed was that in around 2006, a
00:11:54.29 Japanese group led by Shinya Yamanaka developed what was
00:11:58.04 truly revolutionary technology. And the underlying idea is sort of
00:12:02.07 illustrated here. So we start off by harvesting skin cells
00:12:05.22 or blood cells from a patient. And then -- and this is where the magic
00:12:09.12 comes in, we can take these cells, which are completely
00:12:11.21 100% specialized to be skin or blood, and we can convert
00:12:15.28 them into stem cells. And these are pluripotent stem cells
00:12:19.05 that have the capacity to make any other cell in your body.
00:12:21.18 And then we can convert those stem cells into neurons,
00:12:24.18 and then we can identify some defect in those cells
00:12:27.19 for a phenotype that then allows us to develop a drug
00:12:30.05 that changes that phenotype. And so the advantage that
00:12:33.28 this has is that first of all, we're studying human neurons
00:12:36.04 with all the things that make humans human. And in addition
00:12:41.15 to this, because it's a cellular phenotype it means that
00:12:43.25 the rate at which we can optimize a compound is much
00:12:47.23 faster than it would be if we had to keep injecting them into
00:12:50.10 a mouse or some other animal. So that's what we started
00:12:54.13 doing and we've been doing for about a decade.
00:12:58.05 So the first step in this is to start by reprogramming
00:13:02.07 somatic cells to generate stem cells. And so how do we do that?
00:13:07.03 And so the approach is illustrated here. We start with cells
00:13:10.26 that are sort of elongated like this, they're fibroblasts. And then
00:13:14.01 we introduce these factors that change the structure of the
00:13:18.00 DNA. And they essentially alter the folding in such a way that
00:13:22.26 the genome is no longer producing the proteins that are
00:13:26.27 required to make a fibroblast. They're now producing the proteins
00:13:29.00 that are required to make a stem cell. So this is called
00:13:32.22 reprogramming, and these cells are called induced pluripotent
00:13:35.02 stem cells, to distinguish them from the stem cells you would isolate
00:13:39.04 from an embryo. We have all kinds of ways of making sure
00:13:41.27 these cells really are stem cells. We can stain them for
00:13:44.10 markers like Nanog and SSEA-4 and TRA-2, which are these
00:13:49.19 proteins that are only produced in these stem cells. We can
00:13:52.12 also measure gene expression, which is actually what you see
00:13:55.05 here in the bottom right hand panel. And you can see
00:14:00.21 for example, that these iPSC cells, which are here, are actually
00:14:04.21 a very different from progenitors or neurons or fibroblasts.
00:14:09.13 So we can actually be very certain that these cells
00:14:12.20 really are stem cells. Now the next question is, can we
00:14:15.12 convert these stem cells into neurons, and how would we
00:14:17.22 do that? So it took us a very long time to actually try and
00:14:20.14 figure this out. And what we've done is we've actually
00:14:23.20 harnessed the very same pathways that are required
00:14:25.14 to build the nervous system. So we start with these
00:14:29.20 induced pluripotent stem cells, and then these
00:14:31.26 induced pluripotent stem cells over here are then converted
00:14:36.23 into these things called embryo bodies, which are these little
00:14:39.02 aggregates. And these aggregates contain cells that form
00:14:42.20 the initial three germ cells that are the basis of all mammals,
00:14:47.24 so this is ectoderm, endoderm, and mesoderm. It turns out
00:14:51.09 that the nervous system is derived from ectoderm, and so
00:14:55.19 we put them under conditions where you favor the
00:14:58.05 growth of ectoderm. And at this point, they start forming
00:15:01.05 these little circular structures called rosettes that look verymuch like the cross section of the developing neural tube.
00:15:06.29 And then we can take these rosettes and we can roll them
00:15:09.16 into little spheres, and these are called organoids. And
00:15:14.14 we can grow them for months in this way, and then
00:15:17.25 once these have reached a particular point in maturation,
00:15:21.22 we can do one of two things. We can either dissociate
00:15:23.07 them and have these sort of two dimensional cultures
00:15:26.15 that we can use to identify phenotypes. Or alternatively,
00:15:28.26 we can slice them and use them like little chunks of brain.
00:15:32.16 And so what we do is then we can generate cells that have
00:15:37.23 almost all the features of mature neurons. So one of the
00:15:41.05 things we can look at, for example, is we can look at their
00:15:43.17 ability to fire action potentials. And so in this case, we're actually
00:15:46.14 looking at the increases in calcium in these iPSC-derived neurons.
00:15:50.19 Which is what you see here. And every time you see a cell go
00:15:53.20 green, it's firing an action potential. And we do this just
00:15:55.16 simply to show that these really are neurons. We've taken
00:15:57.25 skin cell, converted it to a stem cell, taken that stem cell
00:16:00.20 and made it into a neuron, which is kind of amazing actually
00:16:03.22 that this can even be done. Of course the gold standard here
00:16:08.15 is not just simply the calcium, but actually to be able to
00:16:11.11 measure the electrical event inside the cell and see if these
00:16:15.03 cells actually fire action potentials. And in fact, you can do
00:16:17.27 this. We've developed ways of labeling specific classes
00:16:21.15 of neurons and actually recording from them, and they
00:16:24.05 have these beautiful action potentials which are the
00:16:27.09 signature of a neuron. We can also have them form
00:16:30.14 synapses, so that they form connections between different
00:16:33.17 neurons. And we can actually measure the synaptic transmission
00:16:38.15 between cells. And then finally, we can take these cells and we can
00:16:43.04 actually put them back into a mouse brain. Now why would
00:16:44.24 anybody want to do that? And the reason is that if you
00:16:47.27 have a cellular defect in a human neuron, sometimes you want
00:16:51.07 to make sure that when you give a drug to an organism, you can
00:16:53.27 actually change the cellular defect in situ, that is to say
00:16:56.14 in a mouse brain. So we have these chimeric mice that we can
00:17:00.06 use to test our drugs. So the other thing that we can do is
00:17:05.18 we can generate these brain organoids. And we spent a lot
00:17:07.27 of time trying to make sure these brain organoids look
00:17:11.07 a little bit like a developing human brain. And so after
00:17:15.13 about 60 days, they look approximately like an embryonic
00:17:19.23 12.5 human brain. So they're about the same size.
00:17:25.29 We can actually identify different structures. So in this
00:17:30.27 case, for example, you can see that this is a cross
00:17:33.01 section of one of these organoids. You can see that there
00:17:35.14 is a ventricle, which is very similar to a real brain.
00:17:38.23 There is this germinal zone, which are these green neurons
00:17:43.13 and these are the cells that give rise to neurons. There are
00:17:49.07 radial glial cells that then guide the neurons to the outer part
00:17:53.15 of the brain, which forms the cortex. And we can actually
00:17:56.12 see layers of cortex. We can even measure electrical
00:18:00.18 events in these cells in situ, much in the same way that
00:18:04.12 people do this in the brain. Now finally, we can use
00:18:10.14 these organoids to actually identify phenotypes that are
00:18:13.14 associated with disease. In this case, for example, we have
00:18:16.07 an organoid from a patient that has tuberous sclerosis. And
00:18:21.28 you can see that it has a very different kind of arrangement.
00:18:26.14 So we now have this ability then to generate human neurons
00:18:34.13 from people with disease. The final thing that is important
00:18:39.01 has been to automate this system, to make it actually possible
00:18:43.06 to use this for drug discovery. So when we first started doing this
00:18:46.18 about a decade ago, it would take a single individual 3 or 4
00:18:52.00 months to differentiate just one line. And oftentimes, these lines
00:18:58.02 would get contaminated, there was lots of variability, and
00:19:01.13 so in order to make this really viable, we had to develop
00:19:04.00 a robotic system that would allow us to differentiate many, many
00:19:07.06 lines at once. And to generate enough neurons so that our
00:19:11.02 scientists could actually use them for drug discovery.
00:19:12.24 So we have automated this iPSC platform using a robot that
00:19:17.07 looks like this, and this is a robot that essentially
00:19:21.27 differentiates the cells into neurons, it feeds the cells,
00:19:25.03 and you can do this for many, many lines at once. And
00:19:30.03 so this allows us to do two things, first we can generate
00:19:33.02 human neurons for a drug discovery project. We can identify
00:19:35.25 new phenotypes and targets, and then we can use them
00:19:38.08 to understand many things, including polygenic disease.
00:19:43.17 So for example, we can generate neurons from a whole
00:19:45.25 series of schizophrenics, and see if the neurons from somebody
00:19:48.21 with schizophrenia are on average different than neurons from
00:19:51.27 somebody without schizophrenia. We don't know if that's actually
00:19:54.29 true, but that's the kind of experiment that we can do.
00:19:56.08 So the final question then is, once you've made neurons
00:20:01.04 can you actually identify some sort of a phenotype
00:20:04.22 that would allow you to use this as a way to discover
00:20:09.01 new medicines for that disease. So can we actually do this?
00:20:14.05 So I'm going to give you just three examples of where we've
00:20:19.04 done this, and being able to use it to develop new medicines.
00:20:22.24 And the first example is a disease called spinal muscular
00:20:27.09 atrophy, which is the most common genetic cause of
00:20:29.24 infant death. It is an autosomal recessive disorder, which means
00:20:34.17 you need to have mutations in both copies of the gene that is
00:20:38.09 mutated, which in this case is the SMN1 gene. And it is
00:20:43.04 associated with this loss of alpha-motor neurons, which
00:20:45.06 are the motor neurons that control the movements of
00:20:48.03 essentially all your muscles. It leads to muscle atrophy,
00:20:52.01 disability, and eventually to death. And it is quite common.
00:20:57.07 So about 1 in 35 people in the west are carriers, about
00:21:01.09 1 in 6000 births result in SMA. And there are no therapies,
00:21:06.11 it is a truly devastating disease. And we have worked on
00:21:09.27 spinal muscular atrophy using both stem cells, as well as
00:21:13.28 mouse models. So SMA is, as I said, caused by this homozygous
00:21:19.28 mutation in SMN1 so that there is no functional SMN1
00:21:24.24 protein left. Now, fortunately, there's another gene called
00:21:27.19 the SMN2 gene, which encodes for a protein that is identical
00:21:31.12 to SMN1. So why don't we use SMN2? Well, we don't use SMN2
00:21:35.29 because it turns out that there's a point mutation in the SMN2
00:21:39.27 gene that means that it isn't properly spliced. So it's missing
00:21:44.10 an exon, which is exon 7, that doesn't get spliced properly.
00:21:49.09 So only about 10% of the SMN2 messenger RNAs make real
00:21:53.24 protein, the rest get degraded. So the question is, can we
00:21:56.13 develop a drug that changes SMN2 such that it now produces
00:22:00.17 SMN protein? So to do this, we had to develop an assay
00:22:05.02 that was amenable to high throughput screening.
00:22:07.09 And so the way this was done was to take the human SMN2
00:22:13.03 gene and at the very end, add a coding sequence for luciferase.
00:22:17.01 Which is, as you may know, is the light emitting enzyme
00:22:20.07 from fireflies. And this luciferase only gets produced properly if
00:22:27.23 exon 7 gets included in the gene. It doesn't get produced
00:22:33.15 if exon 7 isn't produced. So under normal circumstances, you don't
00:22:36.11 make very much luciferase. But if you have a compound that
00:22:39.11 corrects this, you suddenly start making luciferase. So we screened
00:22:41.11 1.5 million compounds. And we got three hits. Now normally this
00:22:46.08 is a failed screen, but the people who were working on this,
00:22:49.08 Rajeev Sivasankaran, who is a very talented biologist
00:22:53.19 in our group, he persevered. And together with a set of
00:22:59.12 chemists, they took one of these molecules, they did
00:23:03.13 a lot of medicinal chemistry, and then they eventually
00:23:06.21 developed a molecule that dramatically improves the splicing
00:23:10.08 of SMN2 so that it always includes exon 7. And so you
00:23:14.08 can see, more or less, what that looks like down here.
00:23:16.28 So again, just focus on the black line, you can see that as you add
00:23:24.24 more of the compound, the black line rises because you're getting
00:23:27.17 more and more of the full length product. The red line
00:23:32.06 is a control. And this is simply showing you that if you
00:23:36.20 do it in the other direction, so instead of having
00:23:38.18 a gene that produces luciferase if exon 7 is included, you have
00:23:42.05 a gene that produces luciferase if exon 7 is not included.
00:23:44.25 As you add more compound, you have less of the gene
00:23:48.10 that has the luciferase. So what does this actually do to the
00:23:54.10 SMN2 message in patient-derived neurons? And so this is
00:23:57.24 what you see here. As you add more and more of this
00:24:00.12 compound, we call LMI, you get more and more full length
00:24:03.26 SMN2 RNA. This is in patient-derived neurons. And you also
00:24:08.01 as you see over here, you get more and more of the protein.
00:24:11.28 You can actually do this even in the mouse brain, which is what
00:24:18.22 you'd like to see. So again, you start dosing a mouse with increasing
00:24:21.14 amounts of LMI, and the more you give, the more protein
00:24:24.26 you see. And this is why you see that these blue bars are getting
00:24:28.12 higher and higher, because as you give more of the drug,
00:24:30.18 you get more SMN protein in this mouse that has the
00:24:34.28 human SMN2 gene, it's called a delta 7 mouse.
00:24:39.10 Now these mice normally die, and again, you can see
00:24:42.17 that right here, there are these lines that are going all the way
00:24:45.23 to zero at around 20 days. But if you give them LMI, and you
00:24:52.14 can sort of see this year, at least half of the mice survive
00:24:54.19 for as long as we can measure. So this then, at least in the
00:24:58.22 mouse model, has the capacity to rescue the mice. And in the
00:25:03.07 human neuron model, it has the capacity to dramatically
00:25:06.04 increase the amount of SMN protein. So what does
00:25:11.24 LMI do? And again, this is very unusual. At first we thought
00:25:16.01 that it must bind to a protein the way that every other small
00:25:18.19 molecule is a drug does. But it turns out that it doesn't.
00:25:21.15 It turns out that LMI actually binds to RNA. It binds
00:25:25.19 to the pre-spliced message. And it also binds to an
00:25:31.00 RNA that is used as part of the splicing machinery, called the
00:25:34.26 U1 SNRNP. And it kind of bridges these two to allow splicing
00:25:38.24 to happen more efficiently. We've now progressed LMI into
00:25:44.28 patients, and these are babies with type 1 SMA. And what you
00:25:49.04 see here is something called the CHOP INTEND scale, which is
00:25:53.02 kind of the neurological exam for babies. And all these babies are
00:25:57.00 born with a very low CHOP INTEND score, and over time, it
00:26:00.24 just gets worse unless it's treated. What you see here is
00:26:04.25 that while it's not true in every patient, in a substantial fraction
00:26:07.29 of the patients, as we give them LMI, you see an increase in their
00:26:11.23 CHOP INTEND score. And we're sort of cautiously optimistic
00:26:16.10 that this might actually be beneficial for this population of patients.
00:26:20.15 Okay, I'm going to talk about another disease now, this is
00:26:23.03 something called Timothy Syndrome. Timothy syndrome is
00:26:26.29 associated with cutaneous syndactyly, hypoglycemia,
00:26:30.20 autism, and cardiac arrhythmia. We were interested in it because
00:26:35.10 it's associated with autism and cardiac arrhythmia, and we
00:26:37.20 don't know very much about these diseases. It is
00:26:41.05 caused by a mutation in a voltage gated calcium channel
00:26:43.29 called Cav1.2. And this is a protein that allows calcium into
00:26:49.13 cells. And so the mutation changes the way in which the channel
00:26:54.13 works. So again, for those of you who are not electrophysiologists,
00:26:58.14 you can see that normally, this channel opens and then it
00:27:05.12 closes over time. This is called an activation. And this mutation,
00:27:08.15 this glycine to arginine mutation, actually results in a channel
00:27:12.15 that stays open all the time. And so the question is how
00:27:16.14 does this lead to cardiac arrhythmia and to autism?
00:27:20.17 And so, there are two ways in which you might do this. So one
00:27:22.17 of the things we did was made neurons from these kids
00:27:25.21 and we started to see whether there were any defects in the
00:27:29.18 neurons. So one approach you could think is, okay what
00:27:32.05 does Cav1.2 normally do during development? And one
00:27:36.26 of the things that it does is it regulates the growth of
00:27:40.00 the dendritic arbor of a neuron. And it does it in an activity
00:27:44.28 dependent way. So we had to develop an assay for looking
00:27:47.02 at this real thin dendritic arbor. And so one way of doing this is
00:27:50.12 to introduce into the neuron a light-activated ion channel.
00:27:54.12 And to trigger electrical activity by stimulating the cells with
00:28:00.06 blue light. And so when you do this, every time you illuminate
00:28:04.06 the cell, it fires all these action potentials and then you can
00:28:06.14 generate these movies that allow you to look at the extension
00:28:10.28 or the retraction of every dendrite. And then we can use
00:28:14.16 automated software to analyze every one of these dendrites.
00:28:17.16 And so what we found was that if you looked at a cell
00:28:22.13 from healthy controls, there was an increase in the dendritic
00:28:27.11 arbor over time, so this is the blue curve. And if you looked at the
00:28:32.10 cells from Timothy patients, there was actually a decrease.
00:28:35.05 And so this told us then that this channel was somehow working
00:28:37.26 in a different way. So we've learned a lot and I won't go
00:28:42.16 through all of it about how exactly this happens, but one of the
00:28:45.14 things that this allowed us to do was it allowed us to use this
00:28:48.14 as a way of looking for small molecules that might actually
00:28:52.17 rescue this cellular phenotype. And so this is a graph where
00:28:58.10 we're showing the change in dendrites in cells treated
00:29:04.00 with various compounds. And so on the left is sort of the
00:29:07.00 positive control, so these are cells from healthy individuals,
00:29:10.06 and you can see that there's an increase in the dendrite length
00:29:13.28 over time. And the red dots are one time point, the blue dots
00:29:18.15 are another time point. And then in the next panel, you can see
00:29:23.05 that there is actually a decrease in the retraction of the
00:29:26.05 dendrite. And then we can treat the cells with a whole set
00:29:29.03 of compounds, a compound library, and then we can look
00:29:31.22 for things that rescue this defect in Timothy cells.
00:29:34.15 And so as you can see here, the vast majority of compounds
00:29:38.07 do nothing, but a small fraction of the compounds actually
00:29:41.12 restore dendritic growth. One of them is this compound called
00:29:44.26 roscovitine, and we've learned a lot about how roscovitine
00:29:48.10 works. And at least one of the ways in which it works is that
00:29:50.20 it seems to actually rescue the gating defect in that calcium
00:29:54.02 channel, which you remember doesn't inactivate very well, it
00:29:56.16 stays on. It actually allows that calcium channel to close.
00:29:59.21 So the other approach that you could use is an unbiased
00:30:03.23 approach. You can take these cells from patients and you can
00:30:06.26 compare them to cells from healthy controls and you can see
00:30:09.20 which sets of genes are regulated differentially. And so
00:30:13.04 one approach to that is to essentially measure the genes
00:30:20.03 using microarrays or RNA sequencing. So we've done this.
00:30:24.12 And so in red, you see all the genes that are upregulated
00:30:27.03 and in blue, you see all the genes that are downregulated.
00:30:29.13 Now, among these genes you can look for ones that
00:30:32.04 seem to be part of similar pathways. And so among the
00:30:36.05 little modules that we identified was a set of genes that
00:30:41.18 are important for the production of catecholamines. So
00:30:44.26 catecholamines are dopamine, epinephrine, and these are
00:30:47.03 really important neurotransmitters that regulate fight or
00:30:50.13 flight responses, that regulate reward, that regulate attention.
00:30:53.27 And so it seemed as if these genes were upregulated in
00:30:58.12 patients. And so we looked to see if those genes that were
00:31:02.09 upregulated also resulted in an increase in the activity
00:31:04.25 of the pathways that make these neurotransmitters.
00:31:07.28 And we did this by measuring tyrosine hydroxylase, which is
00:31:11.07 the rate-limiting enzyme in the production of these
00:31:14.02 neurotransmitters. And so what you see here is that
00:31:16.19 on the left, you have a healthy control, and on the right,
00:31:20.04 you have a patient with Timothy syndrome. And so you
00:31:23.24 see many, many more of these yellow cells, and these are
00:31:27.16 cells that have more of this tyrosine hydroxylase.
00:31:29.03 In fact, this is true not only of one patient with Timothy syndrome,
00:31:32.25 this is true of all the patients we looked at. And not only was it
00:31:36.27 true of all the patients, we could actually show that these cells
00:31:40.14 produce much more norepinephrine and a little bit more
00:31:42.25 dopamine. And then finally, we could show that this compound
00:31:48.11 that we had discovered that reversed the dendritic arborization
00:31:51.25 phenotype actually prevented retraction could also prevent the
00:31:56.11 cells from overproducing tyrosine hydroxylase. And this
00:31:58.28 was important because it told us that it wasn't as if these
00:32:02.20 cells had somehow become a different kind of cell, it was
00:32:06.08 simply that they were overproducing this set of genes that
00:32:09.00 are normally required to make these neurotransmitters.
00:32:12.00 So this is called ectopic expression. Okay, so finally we wanted
00:32:18.11 to understand how it was that this mutation, this Timothy mutation,
00:32:22.10 could cause cardiac arrhythmias. So we had a made a mouse,
00:32:26.25 actually, that had the Timothy mutation and we were a little
00:32:30.06 disappointed because the mouse had no cardiac phenotype.
00:32:32.19 And it turns out that that's not entirely surprising because
00:32:35.15 the mouse heartbeat is around 500 beats per minute, and
00:32:38.13 the human heartbeat is around 60 beats per minute.
00:32:40.28 And so there are many mutations that just don't affect
00:32:45.13 the mouse heartbeat in the same way that they affect the
00:32:48.16 human heartbeat. So we decided to make human cardiomyocytes
00:32:53.12 and the first thing that we did was we made some cardiomyocytes
00:32:56.00 from healthy controls, these are the wild types. And you can see
00:32:59.18 that these cardiomyocytes beat beautifully, in fact we were
00:33:02.23 greatly relieved that they beat at around 60 beats per minute.
00:33:05.08 And then, we made human cardiomyocytes from Timothy
00:33:09.25 patients. And when we did this, we found that they beat much
00:33:16.12 more slowly. And it's still going on, as you can see. And further
00:33:22.21 that they kind of stutter. And so you can kind of see this here.
00:33:27.05 Right? That in the controls, they're all beating nice and uniformly.
00:33:32.24 But in the Timothy patients sometimes they have little double
00:33:35.14 beats, sometimes they miss beats all together, like here for
00:33:38.09 example and there. So this then told us that there was
00:33:41.27 sort of an effect. In fact, we could also go in there and
00:33:44.12 actually measure the inward calcium current required for
00:33:47.24 the contraction and show that the patients in red here
00:33:51.02 have much bigger currents. And therefore, they have much
00:33:54.05 longer action potentials during each contraction cycle.
00:33:58.17 And this results in these little events called delayed
00:34:05.11 after depolarizations that people have associated with
00:34:08.05 cardiac arrhythmias. So again, this provided a cellular
00:34:12.14 phenotype for what is actually a phenotype in patients.
00:34:15.09 Okay, I'm going to finish by telling you a little bit about
00:34:19.15 the final disease we've worked on, this is something called
00:34:21.27 Phelan McDermid Syndrome. It's associated with
00:34:24.18 neonatal hypotonia, which is sort of floppiness, global
00:34:27.10 developmental delay, absent or severely delayed speech.
00:34:31.10 And it is caused by a mutation on chromosome 22, at the
00:34:36.03 very end of the chromosome, a region called Q13. And
00:34:40.04 it is normally associated with the deletion of a large number
00:34:43.10 of genes. Among the genes in that region is one gene called
00:34:47.11 Shank-3. And Shank-3 is a protein that acts as a glue
00:34:51.26 at synapses. It's usually found in the postsynaptic
00:34:55.21 events. And it's important because it binds to a large
00:34:59.21 number of ion channels and receptors that are found in that part of the
00:35:03.23 cell. And so we made neurons from these kids, we then had to
00:35:11.27 develop a way to see if there was a defect in the function
00:35:15.03 of synapses. And so that required two things, it required
00:35:18.12 development of a synaptic assay, and we rapidly
00:35:23.03 discovered for example, that if we plated control cells
00:35:25.20 and Phelan McDermid cells in different wells, it was very
00:35:28.22 difficult to get reliable data, so eventually we found that
00:35:31.26 we could plate them together and measure control cells
00:35:35.13 and patient cells in the same well at the same time, which
00:35:37.29 was important. And we could identify patient cells from control
00:35:42.01 cells by labeling them with different colored proteins. And
00:35:45.25 so the first thing we found was that when we measured
00:35:49.27 the synaptic events, and you can see this year. So over here you see
00:35:55.19 both control patients and you can see that there are these big
00:35:59.06 deflections. And these are currents associated with synaptic
00:36:02.28 transmission, and in Phelan McDermid patients, they're much
00:36:06.01 much smaller. And in fact, you can do this over many different
00:36:11.23 input voltages. And when you do that, you can see that
00:36:14.29 there is at every input voltage, the red, the little circles, are
00:36:19.12 smaller than the gray squares. And this tells you that there
00:36:23.18 is in fact less activation of one class of synaptic
00:36:29.10 neurotransmitter receptors, the AMPA receptor, and in fact
00:36:32.20 there is also a fairly severe defect in NMDA receptor
00:36:36.00 current as you see over here. So, the next question is
00:36:40.16 is this really due to a defect in Shank-3? Because remember
00:36:43.02 that I told you there are many genes that are deleted here.
00:36:44.27 And so to do that, we had to restore Shank-3 to these cells.
00:36:48.13 And so when we did this, we could actually show that there is
00:36:50.28 an increase in the postsynaptic events, that is to say the
00:36:57.10 synaptic transmission. And you can see in both of these
00:37:00.21 panels, again you see more of these single events as
00:37:04.10 well as a much bigger postsynaptic response when you
00:37:07.05 actually evoke it. And then the final question is, is there
00:37:12.05 some sort of molecule? Is there some growth factor?
00:37:14.27 And so we looked through the literature and we looked
00:37:17.09 to see what people had reported as important for
00:37:21.04 the formation of synapses. And we tested many things,
00:37:24.17 but one of the things that had been reported is that a
00:37:27.17 growth factor called insulin-like growth factor 1 could
00:37:31.27 increase the number of synapses in a number of different
00:37:36.03 mouse models. And so we looked to see if we added
00:37:39.16 IGF1, could we restore synaptic transmission. And so
00:37:43.10 in fact, what we showed was that if you treated the cells with
00:37:46.07 IGF1 for about 48 hours, you could see both an increase in
00:37:50.23 the number of these spontaneous events, and you could
00:37:53.00 also see an increase in the size of this kind of response
00:37:57.02 over here. So then that suggested that IGF1 was a possible
00:38:05.22 -- I wouldn't say treatment, but a way of reversing the
00:38:11.11 phenotype in the patient. The question is why? Why does IGF1
00:38:15.11 work at all? And so to do this, we actually did an experiment
00:38:19.24 where we looked at the activation of many different
00:38:22.20 signaling pathways by looking at which proteins were
00:38:26.16 phosphorylated. So many of you may know that phosphorylation
00:38:29.19 is a common way in which information is transduced from the
00:38:32.22 outside of the cell to the inside of the cell. It's a common
00:38:36.02 modification that is used for signal transduction, so we
00:38:41.02 looked at all the proteins that were phosphorylated
00:38:45.00 in patient cells compared to control cells. So what
00:38:49.00 you see here is a comparison of cells from healthy
00:38:55.00 controls and cells from Phelan McDermid patients.
00:38:58.10 So as the dots that are below the zero line represent peptides
00:39:09.01 that are under-phosphorylated in patients relative to controls.
00:39:13.01 And the ones that are above the 0 line represent
00:39:15.19 ones that are more phosphorylated. So what we noticed
00:39:19.14 was that there are a series of proteins that were mis-phosphorylated
00:39:24.18 that seemed to fit into a particular pathway. And specifically,
00:39:27.27 they fit into this pathway that is regulated by this kinase
00:39:32.02 called Akt. And that eventually leads to activation of another
00:39:38.11 kinase called mTor. And it turns out that these are downstream
00:39:42.27 of the insulin-like growth factor receptor. So this has been
00:39:47.06 extremely useful, because it tells us a lot about how this
00:39:51.20 insulin-like growth factor actually signals in cells and why
00:39:54.25 it might be effective in these patients. I should say that
00:39:59.01 the idea that insulin-like growth factor might be effective
00:40:03.14 in these patients has actually been tested clinically by
00:40:06.18 Joe Buxbaum and his colleagues. And they have found
00:40:10.06 that, and again this is in a very small trial, but they found
00:40:13.20 that when they measured, for example, social withdrawal,
00:40:15.27 which is one of the impairments of autism, there was
00:40:22.27 a decrease in social withdrawal, which you see in the blue
00:40:28.07 bars which are patients treated with insulin-like growth factor
00:40:33.23 relative to patients treated with a placebo. This is a small
00:40:39.11 study, and so hopefully it will reproduce when it is done
00:40:42.10 on a bigger scale. So let me just finish by touching on the
00:40:48.04 final thing that we need to improve, which is that I think that
00:40:51.13 in addition to improving our selection of targets and diseases,
00:40:55.26 and in addition to improving how we model these things
00:40:59.25 preclinically, we also have to think about how we improve
00:41:03.02 our clinical trials. And there are these three things that
00:41:06.16 we have to do. The first thing is that we have to think
00:41:11.06 about stratifying the patients according to their genetics and
00:41:14.06 pathophysiology, and not their behavior. And so this means that
00:41:18.14 we are probably going to be developing more effective
00:41:22.06 medicines, but for smaller numbers of people. The second
00:41:25.13 issue is that we need to think about how do we
00:41:29.08 develop new endpoints. Again, we're measuring these
00:41:33.16 very soft endpoints, like cognition and mood, and
00:41:37.24 we need to find ways to measure them on a continuous
00:41:40.15 basis in real life using phones and watches and other gadgets.
00:41:44.25 And I think this is well within our capacity to do this within the
00:41:47.29 next few years. And then finally, we need to think about how
00:41:51.26 we implement methods of collecting some of these endpoints
00:41:57.22 in an objective way to reduce the placebo response.
00:42:01.02 And again, there are a number of ways in which we can
00:42:05.00 evaluate individual improvement in ways that help us
00:42:08.28 eliminate the spurious effects of an observer or the spurious
00:42:14.06 effects of somebody coming into a facility to be evaluated.
00:42:18.07 So where are we now? So you may have seen this, this is
00:42:22.16 Gardner's cycle of innovation. And this is true of every new
00:42:25.04 technology, including the induced pluripotent stem cell
00:42:28.04 technology that I've talked about. And so normally, there's
00:42:31.11 technology trigger, and this is followed by this peak of inflated
00:42:34.01 expectation where whatever technology is going to solve
00:42:37.01 every problem, and then a few years after that, it becomes
00:42:40.26 clear that all those promises are not going to be met and
00:42:44.09 not as quickly as people thought, this is what is called the
00:42:47.14 trough of disillusionment. And then very gently, it emerges
00:42:50.15 from the trough of disillusionment with what is describe as the
00:42:53.06 slope of enlightenment. And I think that this is where we are
00:42:57.08 now. On the one hand, I think that we can make neurons, we
00:43:02.12 can make human neurons from individuals with neurological
00:43:04.06 and psychiatric diseases. We can identify defects in these
00:43:07.12 neurons. And we've come a long way in that sense.
00:43:10.08 We can identify compounds and reverse some of these
00:43:12.16 defects. At the same time, there are some clear limitations.
00:43:16.25 Human neurons are clearly imperfect models of human
00:43:19.22 diseases. This is best for things that have cellular defects
00:43:23.04 that are cell autonomous, not so good for things that
00:43:25.22 have defects in circuitry. And the cellular phenotypes
00:43:29.13 that we observe, may in fact not cause the disease. But
00:43:32.13 I've shown you a few examples where we're starting to get
00:43:36.08 inklings that the defects that we observe in these human
00:43:39.03 neurons actually reflect what is happening in a human brain.
00:43:43.08 And if we correct them, we have this ability to actually
00:43:47.10 treat patients with these serious disabling diseases.
00:43:50.17 So let me finish by thanking all the people who made
00:43:54.10 all of this work possible. I often say "we," and I mean
00:43:58.24 them. And this is them. None of this would be possible
00:44:02.05 without the fantastic group of people at Novartis, who make
00:44:06.18 all of this stuff happen. And of course, my lab at Stanford
00:44:09.15 who also made major contributions to this. So thank you.

This Talk

Speaker: Ricardo Dolmetsch

Audience:

Educators of H. School / Intro Undergrad
Educators of Adv. Undergrad / Grad
Researcher
Educators

Recorded: July 2016

Talk Overview

Drug discovery for diseases of the nervous system is difficult. Although mouse models are helpful to study many human diseases, they have serious limitations for understanding neurological and psychiatric disease. Dolmetsch describes a method developed by his group to produce induced pluripotent stem cell-derived neurons from patients. Using these iPSC neurons, they can identify molecular defects associated with a neurodevelopmental disease and then identify medicines to treat those diseases.

Speaker Bio

Ricardo Dolmetsch

Ricardo Dolmetsch moved from Colombia to the United States at age 17 to attend university. He received his BSc from Brown University, his PhD in Neuroscience from Stanford University and he was a postdoctoral fellow at Harvard. Dolmetsch is a faculty member in the Department of Neurobiology at Stanford University. He is also the Global… Continue Reading