(Guest Lecture): April 2024 - Precision Medicine in Myeloma

So thank you guys all for coming out today on a Saturday. It's a really nice thing to get together and talk about MyLoma all together. So really it's inspiring to me to hear all of your journeys. I first want to start out by thanking Jenny for putting this all together. It's really an inspiration. So I'm Ben Diamond. I work at the University of Miami. I was hired by Dr. Langren and I'm kind of a translational guy. So I see patients in clinic and I also work to further our research efforts and to work on genomics. And I have a tough act to follow. I have to follow lunch first of all so I'm sure you're all tired. And I'm also going to try and talk a little bit about some data. And I'm going to try and do it in a way that we can make it a little bit more relatable make it a little bit less scary and so we can try to understand exactly what we need to do to move this disease forward and what we're currently doing what progress has been made. So precision medicine. Hear my disclosures. I was actually joking. My friend out there Austin Zolo was there and saw me in the hallway and said so you're going to talk about precision medicine. That's basically anything you want. Yeah. So we can do whatever we want. So precision medicine is basically the understanding and the idea that everybody's disease is different. And so we shouldn't use a one size fits all approach for every single patient which is kind of currently what we do now. Sure we make some modifications here and there but most of the time everyone's getting treated with the same things. And so we want to change that because everybody's genes their environments their experiences their level of fitness they're all different and all of that interacts together to make everyone's disease unique. And so in myeloma currently the approaches are looking at disease biology. What makes the disease tick. Dr. Fonseca and many others have said that this is not multiple myeloma. This is the multiple myeloma. There are many different diseases within this diagnosis. And so we need to understand the biology of each of those diseases so that we can adequately treat it. Then there's a level of risk. How aggressive is somebody's disease. Do I need to give more therapy for somebody or can I peel back. How long do I need to give therapy. Again these are all questions that I think we need to answer and multiple people in the field are working on altogether. So being that we work on genomics I kind of will leverage that and I say we're going to talk a lot about genomics today. And I want to show you an example of where we were in the past and kind of currently are and what we're moving. Let's see. Can you guys see that laser pointer. OK. OK. All right. So this is fish. All right. This is what we currently deal with in the clinic. It's been in use for decades. And it's relatively imprecise. The way it works is that you basically have to have dividing cells in the bone marrow. They need to be active and then you can shoot a probe that binds to a specific region of interest and then it glows under the microscope. So as you can imagine if the target is slightly weird or if the cells aren't dividing or depending on who's looking under the microscope the test can vary. And there are tremendous problems with it. We rely heavily on it. We like it. It helps us a lot but it really can be better. And so this is an 11 14 translocation. This is what happens when the IGH locus this is a region on chromosome 14 can break apart and link itself with a region on a different chromosome chromosome 11. And this is what we call a driver event. This pushes the disease forward. It causes it to divide. And so we want to detect it. So fish can do that but we can also do it better. So we can look at whole genome sequencing. The idea behind whole genome sequencing is that you can look at every single base pair of the DNA of multiple myeloma down to the last pair. So this is a wealth of data. I mean there are basically mega bases millions upon millions of base pairs in every single myeloma. And so one of the challenges how do we look at it. And this is so much data so many A's and C's and G's and T's. How do we look at it. Multiple different ways. And in this sense what we're looking at here is one of our patients who had whole genome sequencing done. You have chromosome 11 at the top chromosome 14 at the bottom and the wealth of data that we get here is that we can see for an 11 14 case the same as the one on the left. Not only can I tell that the patient has it but I can tell exactly where the DNA has broken. I know exactly what number in that base pair sequence is broken. I know exactly where it went. And then on top of that I see that there's also another event that we didn't pick up on the fish. There's a translocation with chromosome 7 and we don't yet know what that means but we're getting much more data. And this is only two chromosomes. Right. So what we did with an expensive fish study can now be done with whole genome sequencing. We can look at every chromosome. We can detect these abnormalities and much more. And we're going to talk about that much more. So in precision medicine again we're talking about how we want to use drugs that make sense for each person's unique disease. And the reason I brought up 11 14 is because currently it's one of the best examples for precision medicine in myeloma. I want to make the point that this is currently not an FDA approved drug in this space but many people are using it. There's a drug called Venetoclax which is a targeted Bcl2 inhibitor that's been shown to work in general for people with 11 14 myeloma. Now here's the Bellini trial. And I know I'm showing you a curve here but let's look at how we look at these curves. On the top. Where's my pointer here. So looking at the y axis here what we're looking at is the percent of patients that are progression free. So everybody starts at 100 percent. When you start 100 percent here that means that this is where we start therapy. Everyone is doing well when they start the drug. And then as time goes on on the x axis. There it is. So on the x axis time goes on people will start to progress and the curves will get lower and lower. And so the mark of a good therapy is that people on the therapy arm are going to be at the higher curve point and people on the lower. So that's how we look at a progression free survival curve or an overall survival curve. This is Kaplan Meyer curves. And so what we see here is this is a trial where people were given Velcade and dexamethasone plus Venetoclax or placebo. And what we saw in this trial and again this was for all comers not just people with 1114 mile long but all comers. We saw was that generally speaking people that got Venetoclax did better on the progression free survival front. But then the troubling thing was that when you look at overall survival you might notice that these curves are reversed. All right. And people that were actually on the placebo were doing better. People that were on Venetoclax were dying more than the people. So this is exactly what you don't want to see. This is disastrous. All right. And so what I'm about to tell you is that this is a good drug in myeloma but how could that be if people are dying more. So when they looked on this trial at patients with the 1114 translocation you see the difference. OK. These are the people that need to get the drug on the top. You're looking at people with the translocation and clearly getting the drug helps. Those people have a longer progression free survival. And then when they looked at it a different way looking at something called BCL2 staining which is actually the target of the drug they see pretty much the same thing. Right. And so this is a way that precision medicine can help us figure out exactly what drug needs to be used for what disease. And this is only the beginning. All right. So because of the troubles this trial originally had the drugs not currently approved and I really have to make it a point of saying that again it's not an FDA approved regimen but you can see that there are need to be certain strategies that are going to be different for each patient. OK. So I guess in the current state of precision medicine right. So in general people are doing quite well with multiple myeloma for example recently published the Percy study that was looking at Dara with VRD so for drugs followed by a transplant at four years 85 percent of people were progression free. I mean that's that's historic. It's really great. But the problem is that there are 15 percent to 25 percent of people that are still progressing within that first four years. These are the people that have high risk disease and people have been trying to use different features to try and understand how we can figure out who has this high risk disease from the start because clearly even this aggressive awesome combination therapy that we have is not working. So historically they've used fish. Here are a few of the markers that define high risk disease and you can see that they're kind of the usual bad actors that we see oftentimes. These are some of the driver genes that we see on fish testing. OK. People that have multiple numbers of these bad risk features are said to have ultra high risk. All right. So if you have combinations of these things then it goes without saying that people are going to do even worse. So an example of using precision medicine in this patient population is something that the folks over in the UK did. This was a trial called the optimum study. I really like this study. There are a couple of caveats but given this the information that we have these days it was a really nice idea. So they took people that had very ultra high risk disease. All right. People with these features and they said OK we know that the standard therapy doesn't work that great. So how can we make it better. And what they did was they took four drugs. I mean you don't have to look exactly at what this is to see that it's a lot of therapy. They use four drugs. They did a transplant with drugs on top of that and they did more consolidation and more consolidation and then maintenance with two drugs. So they did a lot of stuff and the argument could be made that if you know that people with ultra high risk disease are going to do poorly is it really ethical to put them on a control arm. So the nice thing about this study was that they put all of the patients on this aggressive treatment and they matched it to historical controls treated within the same health system with similar disease features and sort of a synthetic control arm. So this is kind of an idea of how although a little bit immature in the way they did it we can start to use AI based approaches to simulate control arms and make sure that everybody is getting the right kind of therapy and the most promising kind of therapy so that nobody is really getting something we know doesn't really function well. And in the sense you see here that people who had high risk myeloma matched to historical controls versus people on this trial did much better. The trouble of course is that you can make the argument that if we know the drugs weren't really going to work that well in the first place why should we just give more of the same drugs. OK. That's probably the only argument I have against this trial. More does not necessarily mean better if you know that the drugs aren't super active but nevertheless a huge step forward. So those are kind of the current approaches. All of this is based on fish. And in the last 20 minutes I'm going to talk to you about how we can use whole genome sequencing to better understand and currently you know some of the precision medicine approaches that we're applying at Sylvester and trying to move forward. So with fish the issues I think we mentioned all of these things are that there are technical issues that people using the test their operator issues. This is actually expensive. Go figure. Right. And the resolution of the information that you get is pretty low. But there are a lot of different other elements of disease that we need to evaluate to make sure that we are capturing everything. Right. So we want to look at the tumor itself. We want to look at the myeloma what makes it tick. That's genomics. We want to look at the micro environment. In other words the tumor doesn't operate in a vacuum. There has to be a number of immune cells that support its growth. Right. They have to let the myeloma progress and support it. And so we need to understand what features of the micro environment are going to govern that. Then we need to look at M.R.D. You've heard from Ola. You've heard from a lot of people that it's extremely important. And we can start to use things like circulating tumor DNA and other approaches to try and understand it rather than doing invasive tests. And then in terms of toxicity you know a lot of the new therapies are extremely promising but they have their toxicities and we need to be able to predict who those people are so that we can best intervene and make sure we choose the right therapies appropriately. So why genomics. Ola had sort of mentioned this a little bit earlier so I will kind of rehash some of these features that you don't get on fish that you can get with genomics. So there's a concept that in all cells over time you will accrue mutations. No cell is immune from this. This is always going to happen. But the kind of mutations that are being accrued they really matter. OK. One kind of this mutation that cells can accrue over time is from an enzyme called Apobec and you will hear about this moving forward. But the idea is that this seems to be associated with high risk disease with more aggressive disease with immunosuppressive or low immune system states. And it's often seen in metastasis. When solid tumors spread you see a lot of Apobec signature. We've seen that this is actually quite common in multiple myeloma. In our myeloma genomic lab Francesco Mora has pioneered work on this and you'll hear a little bit of work moving forward on this as well. That basically all patients with myeloma have Apobec is very, very common. But we can highlight that there's actually a group of about 11 percent of people that have hyper Apobec. This red curve over here. These are the people that have even more of this. And it's important because these people that have high Apobec or hyper Apobec are actually performing in the poorest in data from this is almost 1500 patients. OK. So this matters. OK. We need to identify people with hyper Apobec because they need something different. OK. If we can identify from this from the start it's going to separate us from fish which does not pick this up. Why else. So Ola also mentioned this concept of chromatopsis. I showed you an example of a whole genome before. This is chromosome 16. So and what you can see here is it's nice and straight. What you're supposed to have is two copies of each chromosome. And you do hear they're nice and straight. And here's a little break in the center of the centromere. And then you have this example of a chromatopsis. And what this is is basically when the genome shatters. OK. So chromosome 16 has shattered here and it's tried to put itself back together in a completely haphazard way. And when I say haphazard I mean haphazard. There are multiple different translocations. In other words the the DNA here on chromosome 16 is now connected to chromosomes 11 to 12 to 1 in this giant jumble of mess. And the problem with this is that in the process multiple different genes can actually become dysregulated. And you basically have multiple different driver mutations accrued all at a single point in time. It's a horrible thing. And it's much more powerful than looking at say a single mutation. There's multiple things happening in this one event. Fish cannot pick this up. Targeted mutations cannot pick this up. And it matters because it's seen in about 20 20 percent of people with multiple myeloma. And it's been associated with poor outcomes. It's a sign of genomic instability. In other words this is an aggressive malignancy that mutates frequently. And it has the potential to evade lots of treatments through the through the lens of having this ability to mutate and become unstable. So another thing that we have to identify because these people are going to need different approaches. So in this context I want to just bring you through a little example of some work that we've done looking at high risk smoldering multiple myeloma. And I think this is a good place to look at it because this is an area of active debate. Right. This is technically not a disease. All myeloma goes through and goes since molding myeloma before it becomes multiple myeloma. The idea then is that because this is a disease this is actually a condition that's asymptomatic that happens when you have the earliest signs of a clonal plasma cell. Lots of people have this. It's very common. And we really need to understand who is going to be the people that progress to having the actual multiple myeloma. People have tried to figure this out in the past. You know there's a high rate of progression. People have tried to figure this out in the past by developing risk models which are up here on the right. Multiple different risk models looking at lots of different features to try and figure out who of these patients with with small during disease is going to progress. The idea though is that all of them are based on basically M protein and free light chains etc. They're based on these kind of indirect measures. And so by virtue of that you're never actually getting a direct measure of disease biology. You're kind of guessing. And so by our best guesses when people are treated with single agent Revlimid you can see that in historical controls people do better in terms of delaying the time until the disease progresses and also delaying the time until there's any end organ damage. But that's kind of not good enough right because we know that when we include people on all of these different models we are capturing a lot of people that may have more indolent states. You're capturing people that probably already have the potential to have multiple myeloma. It's a very heterogeneous population. So we need to be a little bit more accurate than this. We don't know if the benefits that we're seeing are because the disease at this current point in time is actually less aggressive. It's less evolved because the people are actually more fit. They don't have any organ damage or because again as we mentioned this criteria is just a little bit too broad and it's capturing lots of different people. So looking at a trial that the doctors Kazanjian and Dr. Langren ran over at the NCI and also looking at work that was done by Irene Gabriel we actually took genetic information from all the patients that were on these interventional trials. One was using KRD the other one was using e-Latuzumab with Revlimid Indexomethazone in high risk smoldering multiple myeloma defined by those clinical criteria. We took all of their genetic information compiled together compared it to the COMPAS data set from the MMRF and tried to figure out what are the genomic features. OK so going a step beyond those those clinical risk scores that are defining how these people do on these interventional studies. Now I told you I was going to show you data and I don't want you to get sort of intimidated by this because overall this is actually very simple and I'll show you exactly why. OK so what we're looking here on the X axis over here sorry on the Y axis over here is a different genetic feature. I told you already about APOBEC. I told you a little bit about chromothorpsis. They're over here. And multiple other features that we've associated with having high risk disease are over here. And every single patient here in these columns is one single patient. And if you have a red check that means you have the feature. And so what I want you to realize is that all the people over here they have a lot of these high risk features right. They have a lot of these red check marks. So this is genomic complexity. We see a lot of these high risk features all clustered together. And here on the right side over here you see people that have none of these features. OK. This is genomic simplicity. Now when we cluster all these patients together we took away their diagnosis. This is smoldering multiple myeloma mixed with multiple myeloma. And of course what are we seeing. We see over here the majority of patients that were on these trials on the high risk smoldering multiple myeloma trials. They all have all the simplicity and they all do really well. OK. And so this is part of the reason that we really need to look at this because even though they were considered to be high risk by their clinical risk scores looking at the genomic lens they are all genomically simple and maybe this is why they're doing so well. On the other hand though within these trials there were five patients that did not do well. OK. These people progressed despite really nice therapy. And so we need to understand why did only some of these patients with high risk smoldering multiple myeloma actually progress. And we see here that they actually are the ones that cluster together with people that have high risk genomic features. OK. So using genomics you can actually see these two groups. You can see the people that had simple genomics that did great and you see the people that complex genomics that did not do great. OK. So identifying those people is going to help us figure out what are the interventions we need to use for that specific high risk population. OK. I'll skip this because I think we made the point and I will talk to you in the last 10 minutes about the prediction model. Right. So the purpose of going through all this data was to show you that not all multiple myelomas myelomas are the same that using a genomic lens there are multiple different high risk features. We see that a lot of these high risk features like RAS pathway mutations tumor suppressors a little gene called myc that's very important in multiple cancers. We talked about the APOBEC. We talked about the chromothoripsis. When people have none of these features myeloma is going to do well and we need to identify that because perhaps people with none of these features maybe need less aggressive therapy whereas people with all of these features are going to need something drastically different. And that's going to be the whole idea behind trying to profile everybody with genomics to try to understand what genomics are going to be able to do towards predicting how people will do with one therapy versus another and how much we have to put our foot on the gas versus take our foot off. So to that end we'll talk about what we're working on to personalize medicine and multiple myeloma. Dr. Langerin touched on this a little bit earlier in terms of a prediction model that's been built. And what's really exciting about is it's a living model that we can keep building on. Right. So people have done better over time. We've already said that. But there are two scenarios. People do better over time. But there are some patients that do not do well. We talked about that 15 to 25 percent of people that have high risk or functional high risk disease that despite all of our progress are still doing horribly and we need to identify them. And then on the other hand there are people who even like years ago with a therapy like VTD which is Velcade and thalidomide for example there are people that 10 years out are still doing wonderfully. All right. So there are two groups of people and basically what ends up happening is that they all get lumped together and treated the same way. So we need a model to help understand how we can predict who's going to end up in what group and how we can modulate the therapy to make sure that we don't over treat some and under treat others. So in order to do this some very smart people in the lab including Francesco Mora Ola Langerin and some people sitting in the audience over there there's Arjun over there who worked very extensively on this model basically combined data from a lot of collaborators. This is a multi institution effort taking data from compass from the MMRF data from Moffett from NYU from Sloan Kettering from Arkansas combining over about 2000 patients together that had clinical information that had genomic information that had their treatment histories and then importantly had either had whole exome which is kind of smaller scale and whole genome sequencing which is the entire DNA of the patient's tumor. So using all of that put together there was a two step process. The first is that this because this is the largest genomic data set that's ever been assembled you can first look at the drivers. So because you have so much power in a cohort that big you can define new events that are considered to be important in multiple myeloma that drive the disease forward. You have more knowledge of that than because you have all of this increased resolution for treatment history. You can actually plug this into a machine learning model to predict how those therapies are going to do on an individual basis and the net result that everyone is trying to achieve and that we have made steps towards achieving is to try to be able to take one person with one specific set of features plug them into the model and determine how they're going to do based on their disease features and also what treatments they may want to get. So this can really help us personalize the therapy for each individual patient. It's basically yeah and Francesco will usually say that it's basically like chat GPT for multiple myeloma. So I'm not going to take you through basically this complicated heat map over here but bear in mind that there's multiple complexities to trying to do an analysis like this and a lot of analyses in the past have failed because they've not really taken these things into account. The idea is that you can't just take all the patients and put them together into one bucket and the reason is that multiple therapies in myeloma are time dependent. So for example if you try to factor in a stem cell transplant people actually are going to progress before they even make it to the transplant. So you can't compare people that have been transplanted to people that have not because you're already disregarding a bunch of people that were too sick to even make it there. And again I will make the point that this is data from a little bit earlier in time where a lot of people were getting transplanted. So it's mostly a transplanted population but you have to in other words take into account a lot of these time dependent features because if you ignore them your analysis will fail and you won't be able to pick up on that heterogeneity. And when you take all these features into account and build a model based on them what you can see here is that this little box over here this is the interval of confidence for how good the prediction model that we've called Irma is compared to the other staging systems. All right. So this is the C index which is a measure of accuracy here is much higher for both overall survival and event free survival which is very similar to progression free survival. So the model is working quite nicely. Now what the model does is it spits out a state. So instead of looking at a progression free survival curve the idea is that you can be in a different state depending on where you are in your journey. Right. So if you want to predict how well you're going to do following your therapy you're not just alive or dead. OK. After your therapy there's the possibility that you've either progressed or not progressed. There's the possibility that you made it to maintenance. There's the possibility that you made it past maintenance but then didn't make it a step further. So this model accounts for all of those different states. And basically what it does is based on your treatment history it will spit out an estimate of where you're going to be what state you're going to be in depending on a current treatment. So this is a great example. This is a person that was young. They are less than 65. They had an ISS. This is an earlier staging system that we don't really use a ton of very much anymore but they had a high stage disease. They had a back which we mentioned earlier. They had math translocation which we consider to be very high risk and they also had a one p deletion. All right. So this is a patient that was treated with cyborg D which is cytoxan and dexamethasone and maintenance. We don't really use it that much anymore but basically because we're able to have all this resolution and plug in this treatment we can basically see exactly what the chances are that they are going to be alive after X number of years. Now the model is great because we can keep adding more information to it as we add more patients as we have different treatments. These numbers will be able to change. We'll be able to control for things like daratumor which is not in the model or CAR T cells or by specific antibodies. The more we feed it the more it grows and that's because it's based on machine learning. So it will be able to tell us that based on a certain set of genetics based on a certain set of clinical features and based on a treatment what the chances are that a patient is going to be a remission after X number of years. So I show you one example of this to kind of close us out over here. And again another heat map that I don't want you to be intimidated by because at the core it's very simple. What I want you to focus on is this area over here. It's the same case where what we're looking at is that every line every row here is one different high risk feature. OK. They're all listed here. You can't read them but it's OK. They're the same things the chromothoripsis the April back etc. They're all the high risk features. And what you see is that there's a group here that is clustered together. They have a lot of high risk features altogether. Each column is a patient. You can see that you know these patients have like maybe 12 of these high risk features where a couple of these patients have basically one to two or none. And what's happening in this set of patients over here is that the outcomes are relatively poor. And because the model can predict how people are going to do by the different treatments they're going to get you can actually see here how they're going to do. So here you can offer them VRD transplant and maintenance. You can offer them VRD alone. You can offer them VRD with transplant and no maintenance and VRD or continuous therapy for example. So the idea is that the model can let you select different therapies and see how people are going to do. So for these patients you'll notice that all of these treatments are dark or hot colors. All right. And a hot color means that people are going to have a higher risk of progression. They're going to have a higher risk of the disease coming back sooner. OK. The colder or the bluer your color is the better your disease is going to be. So what you see here is that for people with complex genetics none of these treatments are really going to help that much. All right. There's very little that we can do at least with these therapies. That is going to change how things go for them. OK. And so that's depressing. But it also just means we need to innovate. And again this does not control for Dara. Right. So maybe Dara is going to help here on the flip side. You have these people with less complex disease and you look at the treatments here and something very interesting pops up. You see that if VRD is given here you still kind of do poorly. All right. There's still some yellow here. But if you add the transplant adding more therapy seems to help. All right. And so here people are more blue and disease is going to do better. And on the converse you have people over here who are very similar. And in general what you can see is there's actually a group of people here that do well regardless of whether or not you give them a transplant. And so that's kind of the whole point of this is that if we can predict how treatment is going to go how complex genomic features kind of tie into this we can actually tell who needs more therapy who needs less who needs a transplant who doesn't need a transplant and how we can kind of further modulate in the future. And if we see that a lot of people are going to do poorly no matter what we give them with our current therapies are these are going to be the people that we really need to innovate for. And we have all of these T cell redirecting therapies the car T cells the by specific antibodies that are coming. This is the perfect opportunity to see what we can do differently for these people with very high risk features. OK. So this model is still a work in progress as you can see because it needs to have more treatments added to it. But it's currently available. It's online. Arjun was able to get it online and you can play with it. So you can go to the Web site. You can plug in different disease features. It's a little bit limited obviously because you need to have or it would be better if you had the whole X number the whole genome sequencing data. But you can play with it and see what different outcomes are going to be. All right. So this is a work in progress. We're going to add more patients. We're going to add more genetic information more treatments and basically make it better. OK. This is kind of the future of what we're trying to envision here. So in summary the idea behind this precision medicine approach is that again as we've kind of made painstakingly clear not every disease is the same. OK. Everybody has different disease and for that reason everybody needs personalized therapy. Genomics can be the answer to tailoring our treatment plans for the individual person. And the good news is that it's getting easier to do. Our genomics are getting much cheaper and currently lots of groups are working on how we can integrate it into the clinical setting. So I think very soon in our lifetimes we're going to see this make its way into the clinic as for computational models. So we talked a lot about this and Jenny talked a lot about this too with health tree. You have a huge wealth of information that's coming from sixty five thousand individuals of multiple myeloma. We need to figure out how we can integrate all of their information into these models. Again the challenge is that our model needs genomics which is not currently available. But nevertheless we can figure out a way to integrate all of this information together. OK. The time for crowd sourcing information is here and we're going to be building towards achieving that. And finally in the last 20 seconds I will say that again I'm biased because I do genomics but a lot of people in our group don't do genomics. They do other things like the immune microenvironment. This is stuff that Dr. Coffey you heard from earlier is had mentioned basically saying that you know we also need to factor in the immune system. OK. This is work that he did showing that people that achieved MRD negativity had immune systems that looked closer to healthy individuals as compared to people that did not achieve MRD negativity. So we need to factor this information into the model as well. And with that thank you all for your attention. I know we went into the weeds. I hope we tried to demystify some of this stuff and you took away a little bit. And then obviously I think the lab for everything that they're doing for this field and to you all for coming today.