and it is my pleasure to welcome Dr. Larry Vernetti and Mark Medel to the Hepatotoxicity SIG January 2025. We are kicking off the ASLD seminars with this talk. So I'm going to give you a little bit of background about the two speakers. Dr. Larry Vernetti is the Director of Early Drug Safety, Drug Discovery Institute and Associate Professor of Computational and Systems Biology. He obtained his PhD in Pharmacology and Toxicology at the University of Arizona and has spent his time in drug discovery and his early career actually, as he just told us, as a study monitor at AbbVie. And he basically conducted non-GLP and GLP compliant drug safety trials in mice, rats, dogs, and non-human primates. He's also worked on candidate drug screening. He's written IUD toxicology summaries for several compounds, including interestingly, ritonavir, which you're all familiar with, which became the first protease inhibitor approved for HIV. His focus has been high content screening, image analysis, and in vitro liver models. And he's worked at, implemented his work at biotech companies. And more recently, he has been at the Drug Discovery Institute at the University of Pittsburgh. And I think that's been since 2010. And there he has pioneered an early drug safety toxicology program, which includes the microphysiologic systems we will hear about today. Mark is an assistant professor at the University of Pittsburgh, Oregon at Pathophysiology and Therapeutics Institute and the Department of Pathology in the Pitt School of Medicine. He uses human biomimetic liver microphysiology systems to model both normal, yes, normal and disease states, liver physiology. And his interests are in coupling the use of quantitative systems, pharmacology, and liver NPS as a precision medicine platform to identify novel biomarkers and targets for drug discovery. Specifically, he is focused on Masold and Masch and steatotic liver disease. He's also interested in 3D bioprinting of complex liver models and using a microphysiologic systems to examine the impact of the liver tumor micro environment and different cancer types. So without further ado, I'm going to hand it over to Larry. Please take us away. Sure. Thanks, Lily, for that very kind introduction. I'll be addressing how we use our three-dimensional liver acinus microphysiology system. We call it LAMPS. You'll see that pop up in my slides for predicting hepatic intrinsic clearance and hepatotoxicity. And as mentioned, Mark will be addressing how the same model is used in a different way for precision medicine, the cohort selection and clinical trials for MasLD. And can I advance? Okay. All right. I'm going to assume that some of you are familiar with this technology and others are not, but the microphysiology systems have a variety of names you'll see in the literature. Organ on chip, liver on chip, body on chip, organotypic culture models. That's from the EPA. They always like wordy, non-fifty descriptors. I'll be talking about the use for ADMET, Mark for disease modeling, preclinical trials and precision medicine. There's been some effort in this for regenerative medicine and, of course, in multi-organ systems, especially in terms of the MasLD disease. Okay. What are these? Well, in our case, it's actually a miniaturized functional unit of the liver. It does have multiple cell types. I'll review those in just a moment. We do assemble these things with as close as we can get to the in vivo protein cell matrix. It does respond very nicely to our microenvironment under flow and also under semi-static conditions. It does maintain cell-to-cell contacts and signaling molecules. We can supervise the construction, but in the end term, it kind of self-assembles itself into a three-dimensional structure. But importantly, it not only responds to xenobiotics, but we do capture key liver functions, insulin-sensitive glucose, protein synthesis, metabolism, acid production, those kinds of things. Finally, I'll be talking about our LAMPS model. It's a really more complex type of MPS system compared to what you might be used to, which may be organoids or steroids. And we do organize it in a very particular way, which I'll look over right now. These are the cells that we use in it. This group probably doesn't need any introduction to any of them, but I thought I'd throw these in just to show what we do. Of course, we need primary hepatocytes. And Mark, I believe, will talk a little bit about maybe iPSC-derived hepatocytes. We do have stellate cells in there in our case. The model I'll be describing, we are not using primary stellate cells. We have a stellate cell line, but it functions very effectively as a fibrosis inducer. We do use primary liver sinusoidal endothelial cells in our model, and we have found that these cells do express their fenestration. So that was a plus for using them. Cooper cells, we do not use primary Cooper cells. We do have a cell line, but again, it's very responsive to LPS and various stimuli to give us a pro-inflammatory response. And what we have in development is the inclusion of cholangiocytes. And the reason why it's in development is there, to our knowledge, there are currently no published functioning liver model with an actual functional bile duct. So we're working to try and find a way of including the bile duct in our model. Okay, how do we construct this thing? It is a structured, organized, like the liver sinusoid you see on the very top center. That's an image adapted from the Fergert publication. I'm sure this group is well aware of the structural organization of the sinusoidal unit, but this elongated structure has a couple unique things that we try and capture in it. First of all, starting from the portal field labeled as B in that upper tract, that's a very high level of oxygenation that's typically split into zone one, whereas at the other end near the central vein, it's a low oxygen tension in the blood, that is zone three. We can capture this in our model by altering the flow that goes through the model. 15 microliters per hour gives us a zonation of approximately oxygen tension in zone one, whereas slower speed at five microliters per hour gives us a zone three. I will say up front, we do, I do pretty much all of my toxicity modeling in zone three, but there are times when we look at both zone three and zone one. And to be clear, we have disease in two different chips. We can't combine these chips to get a whole gradient of zonation. We go down to the bottom, it shows the current chip we're using. We use a polystyrene device that we get from a chip shop. And then we supervise the layer in of the cells, starting with hepatocytes. And then we start after the hepatocytes attach and seed, we'll start to add in different cell types and ECM proteins. And what you see off to the right of that bottom is an actual picture of our layered structure after it's been assembled using various dyes. And you'll see that we do get flat layers stacked on top of each other, which is what we're striving for in this particular model. And then what we use this for, of course, is by controlling the flow rate, we can start addressing various things like the MasLD that Mark will talk about, the Admetox here, and as a precision medicine platform. Okay. I threw this slide in at the last moment. I often get asked about what test concentrations do you do with your drugs? We've adapted the principles that were published in 2013 by Godoy, and that for orderly administered drugs, the maximum concentration we use in our MPS model is 100 times C-max. And that's based upon those assumptions that you see, one, two, three, the differences between the single therapeutic dose and multi-dose steady state, the differences due to first pass clearance to the liver. And finally, some just sort of a scaling factor to account for genetic diversity and toxicodynamic variability in a large population. So you multiply those together, you get approximately 100 times C-max to detect rare or dilly events. Our typical drug concentrations in our MPS model is a range. It can be anywhere from less than C-max to up to a maximum of 100 C-max. And if you were to go into the literature and look, you'd see that people have many, many different ways of calculating maximum dose. I've seen doses in the millimolar concentrations for some of these, and even higher and much higher than 100 times C-max. Okay, this is how we set up our liver model for a toxicology evaluation. It takes about two and a half to three days to actually do the cell seeding under static conditions to allow a model to become stabilized. So we then start treating the drug flow at what you see day zero there. And for toxicology, we take it out 10 days, a little less for the disease modeling. And during that 10 days, we actually start collecting secretome, or basically the efflux media. Every 48 hours, we used to do every 24 hours, but we found that 48 hours gave us the information we needed. On day 10, we do do several live readouts. You'll see what kinds we do in a moment. We most often will fix the chips and then stain for additional endpoint analysis. But we have in the past, when necessary, we'll sacrifice the chips to get the mRNA, get the samples that we need for transcriptomic analysis. Okay, what kind of assays do you do? Well, if you look over to the right, you'll see there's a big long list. We do most of these in our assays. Some of them we do not do very routinely, such as about two-thirds down the list of immune cell infiltration. That has some logistical issues. So we don't do that unless absolutely necessary. Pretty much everything else that we will do with the exception of also the RNA-seq data, because that does sacrifice a chip. And these chips are kind of expensive. So unless we have a reason we want to do omics-type information from transcriptomics, we don't do it. And we have not been doing that for toxicology. We have three kinds of assays that we monitor. One, the in-life assays, we can use selection of live dyes that are known not to be toxic. And we can monitor things like bioeflux, generation of reactive oxygen species, and mitochondrial function. Meanwhile, we're collecting the secretome. We monitor what you see in panel B, albumin, blood urea, nitrogen, lactate, dehydrogenase, or LDH. We also get our fibrosis markers, some of our fibrosis markers, namely TMP1 and collagen 1A1. And then we have our endpoint measurements. Most often, we do steatosis and fibrosis that you see there. We can change those when we need to. We've also done cytokeratin 18 for various reasons in the past, too. And from the secretome, we do do a cytokine profile. And that's our pro-inflammatory response measurement. Because it's toxicology, especially liver toxicology, there are some clinical indications that show that we can measure that have been demonstrated to actually happen in vivo in the clinic. For instance, we know that, like, our mitochondrial cytochrome C release assay is a good apoptosis marker. Bioeflux inhibition will show, and I have a really cool film that will show at the end on that. Mitochondrial inhibition, of course, is known, reactive oxygen generation, release of pro-inflammatory cytokines, and CIP induction. Typically, we do all of these when I do a toxicology assay. Okay. Now, that was what I just described is mostly relevant to small organic molecules, which these days, about half the drugs are small organic molecules. There's certainly been an increase in biologic drugs. The interesting thing about biologic drugs is one wouldn't normally associate them with hepatotoxicity. Actually, the most common toxicity effects with biologic drugs, mostly monoclonal antibodies, has been cardiovascular effects. But there is a non-rare event of hepatotoxicity. So, we collaborated with Simulations Plus, their computational group, to build a model that could predict biological-induced hepatotoxicity or BILI. If you look over on the right, Simulations Plus has built a computational model that will predict BILI through the intersection between what we do in the laboratory, the in vitro mechanistic BILI data, which you see on the bottom left there. They'll model exposure data using from clinical determinations, and they'll get as much clinical data as they can, and they'll put it into their biologic sim modeling and simulatory tool and make predictions on if a particular biologic drug possibly have hepatotoxicity risk. Over on the left, I don't expect anybody to spend any time with this because it's a huge thing. It just shows the results of some of our screening with tocilizumab, GGF2, methotrexate, and then a combination of tocilizumab and methotrexate. And pretty much all the assays that I've described with the inclusion of some additional ones are shown here. And the color coordinating, of course, red means that's the decreased function, green is an increased function. The purple in this case is things that were trending towards either an increase or a decrease, but they didn't quite get to statistical significance. So we marked them out in case that became important in the modeling. We'd want to maybe go back in and try and flesh that out some more. We did add a couple of additional assays for the biologic drug-induced hepatotoxicity. We added more mass spec determination and started including bioacid conjugations as well as some bioacid secretions. Okay, so I've gone through how we build the model and the kinds of multi-parametric assays that we collect from these models. What do you do with the data? How do you interpret it? This is still a little bit of a work in progress, but what I want to review now is what we've done in the past and how we've handled that kind of information up to this time. I'll show you a very simple way of doing it. The first one, that assay profiling, well basically we just added up the number of significant changes and scored them that way. We did for a while, we were using something called ToxPi, this priority index. And we did one publication out on that using weighted combinations of assay readouts. We don't do that anymore. We weren't all that happy with the way it turned out. But I will then go over how we now are combining that in vitro data with computational simulation tools under the biologic sim that we get from SimulationsPlus. OK, this is a very simple way of some of our initial scoring to try and determine how well we would deal with our liver MPS models. And I think in this case, it's 15 compounds. We selected them as those that are clinically accepted as hepatotoxins. Those are considered true DILI compounds. And then some non-liver toxins. And basically, we put them through our assay. We counted the cumulative adverse responses you see in the concordance chart. And we then charted that against the number of adverse events reported per 10,000 prescriptions in this case. And we found a pretty good correlation of 0.85. That's pretty good. And importantly, compounds such as troglitazone, tocopone, and trovofloxacin all scored very high on this. And the non-liver toxins, erythromycin, famonidine, buspirone, lower on it. So this is pretty good. But it's not very sophisticated, as you can appreciate. So we then moved on to trying to find a way of doing the weighted averaging and maybe finding a way of something that's a little more automated in our scoring system. So we didn't have to do the manual cumulative index, as I just showed. So we did try this ToxPi scoring. And it actually went over a very large number of compounds, over 130. Pretty simple concept. It was published, oh, I don't have the date down. But it was published in about 2014 by, I believe, the EPA. And it's that assay that you see on the bottom, the way it's calculated. And it would take all of this data that we had from all these different assays, and it would make it into a single score, that kind of a dimensional score that you see over on the right-hand column. And then you would arbitrarily draw a line someplace. In this case, we said that anything under 0.4 would be a toxic compound. Anything above that would be a safe compound. It was a way of taking all this data and reducing it down to one easily digestible number, very black and white. Unfortunately, when we went through all 134 compounds, compared to what we did, what I just showed of a pretty nice correlation to clinical concordance, it wasn't as good. So we didn't stay long with this, and we're moving on. Now we're going into or using the more sophisticated tools. I'm going to show it here as an example, or just be what we do in BiologicSim. In Tocilizumab, we've also done many other compounds, but this one is the cleanest, and this is one that we've actually published. The way this is working now is a combination of what we do in our LAMPS model, our MPS models, and then Delicim would simulate plasma concentrations, and they also simulate their interstitial concentrations, clinical values that are appreciated, that are gathered. And then they would model the steatosis and ROS, in this case. They also have some CYP3A4 inhibition, which they modeled, but I didn't show here. And then they would make a prediction on it, and this is their proprietary, so I can't quite show how they make these prediction models, but what I am wanting to show you is if you look over on the very right, you'll see that there are four colored dots. Each of those dots represents an individual patient, simulated patient, and it's divided amongst the in vitro assays. And what I want you to appreciate is that the pharmacology of tosylizumab, if you're not familiar with it, it's an anti-IL-6 receptor antagonist, so it should interfere with any IL-6 signaling, and that's the pharmacology of it. And you can see that the pharmacology is not affected when we look at just the IL-6 signaling only. None of the patients have a rise in the transaminases. When we look at steatosis and in ROS only, you'll see that at least one of those patients comes up to that they would predict that would have a increase in ALT in this case. So that's kind of what we're going through with these more sophisticated modelings and how we're using this to start to define. Now, what you're seeing here is what we've done with the biologic molecules. Some sort of a simulation tool would then have to be applied to the small organic molecules, and we are actually working on that to some extent with the biologic sim people. OK, PK, that's important, and I left that last for a couple of different reasons. Drug metabolism, of course, most of us just think of it in terms of phase 1 and phase 2, but it's becoming increasingly important that phase 3, which is transporter-mediated elimination, is very important in this area. So we're going to try and start modeling some of that as well. OK, there are many, many, I mean, maybe 20, 30 different packages that are available that you can purchase. Some of them are free for predicting clinical PK from in vitro data, and this is our version of it that we built over time. We're looking at both bolus infusions, and I'll say that bolus infusions are being done by what I would call the gold standard of PK testing, which is short-term, one to four-hour hepatocyte suspensions with drug. That's what we're going to talk about here. And in this case, that would be what you see over on the very left, and on to day 1. OK, then we also looked at these bolus injections at day 6, 12, and 16, and used that data to make our calculations. And of course, we can look for metabolites. That's what's on the bottom. And these compounds were chosen to be very specific for various SIPs so that we knew that we were getting full function, for instance, trofenadine, the SIP 3A4. So that tells us that's functional models. OK, so going over to the left is really where it is. We look at the hepatocyte suspensions over, in this case, four hours with one molecule, diclofenac, and we'll see a decrease in diclofenac, increase in the 4-hydroxy metabolite. And then we can compare that to what we call steady state in our liver MPS model over in the top right. We take that data, we put it into, like I said, we built our own predictor model. And when we do that, actually, at least in the case of diclofenac here, the suspension model is pretty high, but the MPS model is more within the clinical range that we'd expect. And you can see the other calculations that we can make from this. And the one thing I want to say about, oh, sorry about that. OK, let me go back one more if I can. Maybe not. OK, all right, while this is running, I'll just say that one of the problems with the PK predictors was published in 2022 by Bowman and Bennett that these predictors usually under-predict clinical situation. So that means that our predictive models probably will need some computational effort to make them clinically significant. OK, now let me go on to these movies. I really enjoy watching biology in action. And when you get into the liver, there's not much biology in action to really observe. But the bioeflux inhibition and bioeflux control is maybe an exception in this. And what's running over in the left field is a control. And you see those kind of worm-like structures popping up and disappearing. Those are fluorescent dye that's being uptake through the basolateral media, then the efflux proteins around the apical surfaces into these biocatalytic spaces that form between hepatocytes. And you can see the dye going in and out. That's what it should look like. You then treat it. And next to it is a 50-micromolecule glitazone dye on treatment. And you'll see that transport inhibition is interfered with. It's not being pumped into the conicular spaces. And the cells just get brighter and brighter and brighter. Our group is known. We do a lot of image analysis. So we can take these images that you see here. And we can transfer them into dye measurements and dye efflux measurements that you see over on the left. And the true glitazone dye on shows clear inhibition. This is important, as I said, because bioeflux inhibition is really coming into what's known as an increased risk for hepatotoxicity. So this is something that we do routinely, although not at a movie level. OK, one thing that's important and has been kind of an issue with this whole technology, not just ours, but everybody's that's working in the MPS, is that within the same laboratory, you can maybe get some pretty good reproduction, reproducibility of your assays if you do it time and time again. But that's not true if you take that assay and send it to an independent laboratory. So we did that. We sent our LAMPS model down to Texas A&M. And they looked at how our model performed, not only as a toxicology model, but also as a PK model. And we got pretty good correlation. And we published that between ourselves, between Laboratory 1 and 2. And I honestly don't remember who was Lab 1 and Lab 2 in that particular set of data. But in general, we had pretty good reproducibility between the two labs. If you look over, though, on the very top right, you'll see from that same paper, which shows that even though we had good reproducibility between our two labs, as we went up in clearance values from the actual clinic on the x-axis there, we did fall off. We were under-predicting. Again, this was reported out by Bowman and Bennett. And so we were able to confirm that our models indeed, even though they're reproducible, do under-predict. And of course, what that tells you is is that you'll need some computational tools to make those predictions fall into where you want them to be. And why is that important? A couple of reasons. One, the inter-lab reproducibility gave us the incentive, I guess you could say, to apply for a TRACE program. The TRACE program, if you're not familiar with it, it's funding to get your models accepted by the FDA as a qualification tool. And that comes with a whole bunch of caveats, which I'm not going to take the time to go into. Maybe that can be a follow-up seminar for this series on the qualification tools of the FDA. But we're going to use our model for two contexts of use. And they're very specific contexts of use. One, we're going to look at using this tool to be able to predict hepatic clearance differences between healthy drug candidates in phase one and maybe liver disease, Maslow disease, specifically disease patients in phase two. And kind of an offshoot of that is we'll also look at, are there any increased risk of liver toxicity of a drug going from phase one in healthy to phase two Maslow D patients? Anyway, it's pretty nice, a lot of administrative, as you can imagine, and especially working with regulatory agencies. That's been an eye-opener all along. And with that, I'll stop and let Mark take it from here. Sure. Thanks, Larry. Let me share my screen. Stop sharing, Larry. In the meantime, I'm just going to ask everyone to please insert your questions in the chat so that I can know in what order to call on people. OK. So is that good? I think I got my screen shared. I see it. Looks good. OK, great. So once again, my name is Mark Needle. I'm also an assistant professor at the opt-in and a colleague of Larry's. And I'm going to switch gears a little bit using the same model that Larry just nicely described. And I'm going to tell you a little bit about our Precision Medicine platform using the LAMPS model for Matt Mazzoli. And the example I'm going to take you through today is involving the PNPLA-3 high-risk variant liver cells constructed with either the wild type or I-148 version M of that variant and look at differences in drug response and Mazzoli progression within the model. So as we all know, metabolic dysfunction associated steatotic liver disease, or Mazzoli, is a major global health problem where over 25% of the global adult population is on the Mazzoli spectrum. And the real driver is that this is a very heterogeneous disease that has a complex pathogenesis on its own. And this is driven by multiple factors, including genetics and lifestyle, environment, and various comorbidities. And so because of this complexity, there is currently only one approved drug for the treatment of MASH. And that's resmetarone. It was approved last year. And despite this drug's approval, if you look through the data of it, only a fraction of the patients, about 25%, positively responded in this case. So really, the challenge that we want to harness our liver MPS platform for is to deal with this heterogeneity and use it as a tool to address the clinical problem of identifying or better predicting which patients are most likely to respond to any given potential therapy. And really, this is all driven by patient heterogeneity. As I said, in patient heterogeneity, especially in complex diseases like Mazzoli, is really complicates Admetox studies, like Larry alluded to, drug testing, and selecting for clinical trials in various patient treatments. So really, what we're trying to do is use our chip to deal with that heterogeneity and first identify patient cohorts that are most similar and most likely to respond to a potential therapeutic. And as we all know, I'm sure, that there's the therapeutic landscape for Mazzoli MASH is really bubbling over right now. And there are many candidate therapeutics that are in various clinical trial stages. And so keeping this in mind and dealing with the issue of heterogeneity, all of these drugs are most likely to be pretty expensive to the patient and for the payers. And so the idea is that we want to better select for who will respond and not respond and potentially, you know, which combination therapies are optimal in specific cohorts of patients. And so our focus is on the use of biomimetic structured MPS, as Larry told you. And really the use of MPS is a continuum and it's a range of in vitro models that involves varying levels of model type. And it's a balance between basically the throughput of the model and the overall biomimetic or biological complexity of the model. And so our focus is on using the structured layered biomimetics that Larry already introduced to create liver MPS, to generate both phenotypic and mechanistic data that can be used both individually, like in a liver alone model or in a coupled organ MPS system, which I won't have time to talk about today, but that's another area that we regularly do work in where we couple our liver to other organ MPS like a pancreatic islet. And so the two main models that I use are the LAMPS, which Larry has already nicely described and the VLAMPS model. And both of these model systems are structured biomimetics and they each have their own advantages. I won't really dwell on this, but for the LAMPS model, you can individually study liver zones, zones one and three, whereas in the VLAMPS model, because it is a dual channel model where you can continuously monitor a continuous oxygen gradient within the same chip. So you have zone one, zone two, zone three, all within the same chip. And another advantage of the VLAMPS model is that it's a two channel system, whereas the LAMPS model is a single channel system. So within the VLAMPS, you have a two chamber system that's separated by a porous membrane. So you can do things like Larry alluded to, like the immune cell infiltration in that model, where you flow in say macrophages or PBMCs from the lower vascular channel and under the right stimulus, they will bind to the membrane and migrate up through into the hepatic channel. So it just all depends on your specific context of use and what question you want to particularly answer. And so I'm going to be telling you about how we use our LAMPS model today to model and study a Masaldy disease progression. And really this was work that we've done probably about four years ago now, where the first thing we set out to do was to develop assay and media conditions that allow us to better study the progression of Masaldy. And we did this by developing and validating in our model system, different stages of Masaldy progression from the normal fasting state to the early metabolic syndrome state or the Masaldy state, and then the later metabolic syndrome state, which is more the pro-fibrotic MASH-like state. And we accomplished this by modulating some key media components within our perfusion media, such as glucose, insulin, free fatty acids, and the addition of pro-fibrotic drivers like LPS and TGF-beta. And we can use these media formulations in either the LAMPS or the VLAMPS model. We can monitor, maintain these models in the normal or diseased media for about two weeks. And we have a panel of disease relevant metrics that we can measure either in real time or as endpoint measurements. And so the example I'm gonna tell you today is really looking at the key drivers of Masaldy progression, including diet, which is modulated by our media, and also genetics. And as an example, we studied the PNPLA3 variant, which is a hepatic lipid droplet localized protein. And it is known to be associated with increased and more severe Masaldy susceptibility and progression. And so the overview of our precision medicine approach in using these chips is really to deal with the idea of heterogeneity that I described earlier. And our overall goal is to do this in two ways. So to address the issue of heterogeneity, we use either patient-specific primary cells that have been genotyped for specific variants or other genomic factors. And we construct models with primary cells and non-parenchymal cells that can either also be primary cells or a mixed cell model like Larry described, that's a combination of some primary cells and some immortalized cell lines. So that's one way to construct the model. The other way to construct the model, which is associated with our NCATS-funded clinical trials on a chip program is the use of patient-derived iPSC liver cells, which I'll touch on a little bit more later. And these models are constructed from patient-derived iPSCs, which the patient provides a blood sample and through the work of our collaborators, iPSC cells are produced and differentiated into liver-specific iPSCs to create a so-called patient biomimetic twin or PBT, which we can use either of these cell types then to construct our biomimetic MPS, in this case, the LAMPS. And then we're able to characterize disease progression and drug response using this model by using the different media conditions that I described in the previous slide and monitoring a variety of disease-relevant metrics like steatosis, inflammation, stellate cell activation, and the progression of fibrosis. And so really, then, we are going to be moving into the patient-derived liver iPSC cells, where we then can construct the patient-specific chips and use it for a variety of applications, including the drug discovery, ADME-TOC studies, to inform clinical trial cohort selection and ultimately test preclinically patient therapeutic strategies. And I should just point out quickly that an important feature of this overall approach is that the use of the primary cell model serves as a benchmark for the results that we obtain using our iPSC-based model because we have to constantly validate the use of our iPSCs in these models to compare how well they function to the fully mature primary cells. So that's sort of a built-in check within our system. And so the first example using the primary cells is that we looked at our LAMPS model that was constructed with primary cells that contained the PNPLA3 I148M or GG variant. And what we found was that this variant demonstrates genotype-specific differences in steatosis, stellate cell activation, and the secretion of pro-fibrotic markers. And if you can see in panel A, looking at the A and B, looking at the quantitation of lipid toxins, what is the dye we use to quantitate lipid accumulation, in that in each media condition, the normal fasting EMS or LMS that compared to the CC wild type background, the GG variant results in a significantly higher amount of lipid accumulation, which is consistent with experimental model literature as well as clinical data. And we also were able to see staining for alpha smooth muscle actin, which is a marker for stellate cell activation, an increased amount of alpha SMA expression in either of our three media conditions similar to the lipid accumulation. And furthermore, we're also able to see increases in the amount of secretion from our devices in measuring pro-collagen 1A1, which is secreted by stellate cells and is a marker of fibrosis progression. So we can sort of see right off the bat using these genotype cells, where we tested three CC variants and two GG variants that there were genotype specific differences there between the two. And this table just summarizes other metrics that we looked at, including the secretion of some other cytokines. And you can see pretty much across the board that the phenotypes of these disease progression markers were more severe in the GG high-risk variant background. And so the next logical question for us was then, we now see genotype specific differences in disease progression. Now, are there any PNPLA3 specific differences in response to drug treatment? And to look at that, we used resmetarone since it was an approved drug. And what we see here is that we did observe a greater reduction of steatosis in the PNPLA3 wild-type background compared to the high-risk variant, also demonstrating a genotype specific response to a therapeutic. And that's shown in the graph here and in the lipid tox stain pattern shown in the figure there next to it. And in addition, we also looked at the alpha SMA expression and collagen 1A1 secretion in our model. And again, we saw the same sort of pattern where we saw a higher efficacy of this drug in the CC background compared to the GG background. And so this is just sort of summarized here, but really I think the take-home point is that using our MPS model or liver MPS model with some genotyped primary cells that we're able to recapitulate some PNPLA3 specific disease progression phenotypes that are consistent with what is known in the clinic and with some other published results using liver MPS system. And also our data is consistent with the trial data that was published for the resmetarone studies, although it didn't elaborate on genotype specific differences in those studies, this drug did reduce lipid accumulation in that clinical trial. So again, this is just sort of a test study that we performed to sort of validate our model for its potential use as a precision medicine platform where we can sort of pick up on these differences based on genetics environment. And so just to let you know, is that we're also using this similar approach to look at other Maseldy specific genetic variants, both high risk and protective. And so we're starting to look at different combinations of variants, both protective and disease progression associated variants in various combinations using either the genotype primary cells or iPSC derived cells. And then, so finally, this just sort of puts into perspective the evolution of how we sort of see the precision medicine for the development of Maseldy clinical trials and potential therapeutics. And so it really integrates the use of first clinical data, it has to start in the clinic where we have access to not only the patient blood draw to produce the iPSCs and then perform experiments in the PBTs, but also the clinical data that's associated with these patients, whether it's health history data, various clinomic or genomic data, and sort of the onboarding of other multi-omics that we can begin to gather from these patients, including metabolomics and proteomics, and then sort of use this patient digital twin and patient biomimetic twin in tandem to generate computational models that allow for us to predict drug response or predict disease progression and then test those hypotheses in the PBT platform before we actually move forward with the clinical trial. So it's a way to sort of deconvolve the heterogeneous nature of this patient population. And so, as I indicated, we're moving into doing these studies now in patient-specific iPSCs. And in our current project, where we have over 250 patients enrolled who have agreed to donate blood so that iPSCs can be produced and built into our liver MPS platform. And so it's just a real integrative team science-like approach that requires clinicians, bench scientists, computational biologists, and engineering. And so from our cohort of around 250 patients, we've identified 10 subjects from each genotype in which to generate the iPSCs from, 10 high-risk variants, 10 heterozygous variants, and 10 low-risk variants. And this is a lot of work and a lot of characterization. And we have these cells all made now. We're going through the process of production and differentiation of them and cryopreservation of them so that they can be used in our models. And we're starting to use these in our models. Here's an example of one of the wild-type cells models constructed with all iPSCs in a PNPLA3 wild-type background. And you can see, similar to the primary cell data, that LAMPs constructed with these all-iPSC-derived liver cells, we can A, demonstrate some Maseldy-specific phenotypes on the left, where we can see things like increased lipid accumulation and increased stellate cell activation, depending on the media that they're maintained in, their normal iPSC media or disease-specific media that's supplemented with free fatty acids and TGF-beta. And also, and this is just very recent data that we've just, you know, we've spent a lot of time characterizing these cells and learning how to work with them in our model system, which is not an easy thing to do. There's a ton of considerations that are things different from primary cells, like media half-lives. So being able to, you know, introduce a constant supply of fresh media to our chips. So working all that out and actually being able to show some disease progression phenotypes is a huge step. And we've now just started to do some introductory drug testing studies where we can demonstrate using resmetarone, a dose-dependent reduction in lipid accumulation that is, you know, similar to what we observed in primary cells. So we're in the process of sort of validating these results, repeating these results, and then we are going to, you know, sort of move into the full-on stage of going through our entire iPSC patient cohort from which we have the iPSCs developed. And so, just as a reminder, is that, you know, we're working with a relatively small patient cohort of 10, 10, and 10, but really we want to apply this, you know, in a much larger scale. So, you know, working with the iPSCs is a process, and to sort of make this process more reproducible and more higher throughput, we have a project collaborating with the NCATS 3D tissue bioprinting laboratory, in which we're doing just this, where the overall goal of this project is to introduce some automation and industrialization in building our liver MPS models with either primary or iPSC derived cells, you know, using their automation and 3D bioprinting capabilities. So, this will facilitate, you know, sort of a more industrialized approach to generating these patient biomimetic twins. And so, finally, just to sort of summarize, we're completing the differentiation and the characterization of our patient cohort iPSCs, you know, each of the differentiated liver cell types, and we're, you know, beginning to use them to produce our patient biomimetic twins from that cohort, and we're testing the disease progression and response to drug within this cohort, and we're also implementing, you know, various automation technologies to improve the overall reproducibility of producing and differentiating our iPSC derived liver cells. And this is just to remind you that, you know, we're using this approach as a precision medicine platform, and it's really based on both a computational arm to develop patient digital twin, as well as our biological engineering arm to then create these PBTs, and sort of these two technologies interface with one another to more smartly inform clinical trial design and predict response. And so, with that, I think that was the last slide I had. I'll just throw up my acknowledgement slide, and I think I'm on time, and we can, you know, take any… Fantastic. Wow. I feel like I should have organized two separate talks, because there's just so much of them. I feel like we're going to have a lot of specific questions to each. Fantastic model. Really, really interesting data. Great concept. I know there's a lot of people in this audience that have been working on this problem for a long time, and I'll just tell you that as somebody who is in the drug toxicity field, I always see these talks, and I'm like, where are the adaptive immune cells? Why is it so hard to kind of like get those into the system? And I know it's a challenge, and I always ask that. So, I won't get into that, because there's no answer for that right now. Hi. Johnny has a question. Johnny, you want to get started? And then, Johnny, the Q&A may go on for a while. Do you have to leave, or can you continue moderating? I can continue moderating, for sure. Yeah. So, my question was about, for Mark, about the isodelate cells. Yeah. So, have you done characterization to investigate, before treatment, if they're truly quiescent? Are they star-shaped? The alpha-sma activation and collagen secretion is very convincing, and that's a real advancement in modeling liver injury in the context of Maslady and Mash. And so, there's a couple of things I've put in there, like do they contain vitamin A? Do they have lipid droplets? And then, when they activate, do you see more of them? So, that's a good question. And this points to a general problem, just in general, with using any sort of stellate cells in experimental models, where, basically, the processes by which we either maintain them in culture for use, or the process of cryopreservation, you know, creates a sort of activated state in itself. And that's for, yeah. Just like with primary stellates, as well. I mean, as soon as they hit plastic, they activate, right? Yes. So, what I can tell you is that there is, they're not perfectly quiescent when we first put them into the model, because, you know, it's a stress on those cells, just building the model, you know, putting it together. So, even if we look at, like, specific cytokines, there's a nice pattern, which took us a while to sort of get a handle on, is that when we first build the model, we would see things like higher amounts of IL-6 and TNF alpha in the first early days after the model, like the first day, day or day or two. And then, that would sort of taper off and remain low. But then, once we reintroduce our disease media, they would come back up. So, there's sort of, like, a settling-in period that these models have to go through when you build them. But to specifically answer your question, we have not yet looked at the vitamin A receptor, but that's something that's on our list to do. But what I can tell you is that just from the imaging that we do on these, we can tell that, yes, these do store lipid, because they do take up the dye, and we can see that in the microscope. And when we introduce the disease media, we can see that, compared to the normal media, that they lose that lipid accumulation in the disease condition. So… That's good enough for me. Yeah. Okay. Dr. Apti also has a question on slide 19. Dr. Apti, do you want to unmute and ask a question? Sure. Hey, Mark. Great talk. It is nice to see you again. Yep. You had shown some data on progression of disease on slide 19, and I'm just wondering if you could go in a bit more detail about the timeline and how much it takes, what do you start with, and what do you end up with? Yeah. So, I guess I should tell you first that, so, right now, the slide 19 you're referring to is in our iPSC-based model, okay? Okay. And so, compared to the primary cell model, we can maintain the iPSC-based model for six days under flow in culture, which is not quite as long as we can maintain the primary cell model, but these are, you know, this is a different animal working with these cells. So, but we do that for up to six days, and the images and the measurements I've shown you there are taken as endpoint measurements. So, you're looking at a day six measurement where we look at either the alpha smooth muscle actin staining or the lipid accumulation via lipitox labeling. So, those are endpoint measurements. Do you have something where you looked at day three and then day six? I just was wondering if there is an increase and change in alpha smooth, things like that. So, for these particular studies, these were actually just done, and this was our first, you know, attempt at doing this. We just used the endpoint measurements for this study because we would need to sacrifice those chips at earlier time points, you know, to do that staining, and, you know, we wanted to see first if we could maintain it for, you know, six or eight days, so that, but the follow-up studies, we're going to take a time course. Yes. Your single cell cultures you maintain for up to, what, three weeks, or? Yeah, the single cell cultures can be maintained for a much longer period, like up to, like, over two weeks for sure. Yes. Okay. Thank you. Just as a follow-up question, how do you determine that the model is done or is not viable longer than that amount of days? Is it just, you're seeing reductions in albumin secretion, or how do you determine suitability at time points? It's a pretty apparent process for us, and we do see those things where we will see, like, a decrease in albumin or a decrease in urea nitrogen production, and that, you know, is a pretty good indicator that our model is, you know, ending its usefulness to us, and also, you know, we can, you know, just sacrifice a chamber on a chip at any point along the way and do, like, a live cell die or a dead cell marker, and we can look at it that way, and it's pretty apparent to us just by looking bright field. We can see that the model is, you know, degrading and dying. We can see cells rounding up. We can see cells sloughing off, so it's, you know, it's not subtle. Yeah. Got it. Great. Are there any other questions for our speakers? Please unmute and ask your question, or if you want, you can put it in the chat. I guess I should also just say that matrix components are a huge factor in determining how long we can make these models last as well. So, we have a question in the chat. Is the oxygen content physiologic in the system? Very good question. Larry, you want to take that one? Sure. The answer is yes, but that comes, again, with a caveat. When we pump the media over the cells and we actually monitored the oxygen tension right on top of the cells using some oxygen sensitive beads, those are earlier experiments that we published, we found that in that first 20 microns to maybe 40 microns above the layer, it is physiologic. Okay? If you go higher in the media, and depending on the configuration of the kind of chip you're using, the oxygen does go more towards what it is at room temperature, but it's interesting. The oxygen diffusion through the media is much slower than oxygen consumption by the cells. So, we can maintain that gradient right on top of the cells over the course of our experiment. And a quick follow-up question. So, you mentioned that you can achieve either zone 3 or zone 1 type. How are you determining that? Is that through cytochrome P450 expression, or are there other metrics, and is that mainly controlled by the flow rate on the chip? Is that guided by oxygen tension alone? Yes. To our understanding, we have looked at many effects of zone 1 and zone 3. I should have put a slide in for that. Excellent question. We see increased steatosis in zone 3 compared to zone 1. We see increased mitochondrial respiration in zone 1 compared to zone 3. We have looked at CYP2E1, and we did get higher expression in zone 3 compared to zone 1. We also get higher albumin output out of zone 1 compared to zone 3. Am I forgetting any, Mark? We've done a variety of different ones. Yeah, Wnt protein expression profiles as well. Oh, that's right. Yes. So, yeah, we have done that. Now, that's done in the LAMPS chip. The model that Mark was talking about at the end was our V-LAMPS chip, and that's a continuous gradient. There's a zone 1 on one end and zone 3 on the other end, as you would expect. And we've done oxygen tension in those and have confirmed the oxygen gradient. Very nice. Do we have any other questions for our speakers? We're at 1211, so we're a little bit past time, but we can – I'm sure we can – if people can hang on, we can have another question or two. This is Skip. Can you hear me? Yeah. Hi, Skip. Hey. Hi, John. So, I had a question for Larry Bernetti. You had a slide that we – I'm interested in the biologics that you were working with in your model, and you had a slide where you – I think it was the biologic sim slide. Anyway, you went from measuring some analytes, and then by the end, moving across your slide, you had a population-based – like, you could predict that a couple – I think people would have ALT elevations. So, my question is, I sort of understand how that's done when you're talking about reactive metabolites, because you can kind of do a population-based variation in how people metabolize this way or that way. How did you do that with a biologic? Do you know what I mean? How is that – is that a black box kind of proprietary thing? You know what I'm saying? It's sort of this movement that I just didn't catch. I totally understand your thing. Just for everybody else, biologics, especially the monoclonal antibiotics and proteins, they don't go through drug metabolism like small organic molecules. So, there is definitely a lack of reactive intermediates, and of course, reactive intermediates are the most common cause of hepatotoxicity with small molecules. No, we do not have that, and we did appreciate that. No, this seems to be something related to the effect of those compounds or those biologics. In the complete system, when we tested otocilizumab-caused ROS in hepatocytes and an increase in steatosis, and when we test those just in hepatocyte cells or some of the other cell types, the effect is kind of there, but it's not as strong. So, definitely required all four cell types to pull that out, and yet it is a black box. We don't know exactly. We do not know the mechanism. We only know these mechanisms of toxicity that were enhanced, and we look at phenotypic markers, but we've not tried to trace down what is actually going on there. I will say of interest is one of the things clinically that are being used, especially in cancer treatment, is the use of two or more of these biologic drugs combination. Interestingly enough, even though I did not show the data here, when we test each of these drugs, two drugs individually, say nivolumab and ipilimumab, individually, we see a little flutter of some signals, but when we combine them at relevant concentrations, we can see a very exaggerated toxicity response. So, we're still following up on that to see how useful that is going to be in that biologic predictive model, and unfortunately, it is an IP-protected model, how they're doing this, so I don't even know what it is. Thanks. Do you see a heterogeneous response with ipinivo, like as observed clinically with iLisi in your IPS patient population? Maybe that's a question for Dr. Vernetti. Have you done this? Yes. Yeah, so do you see any individual differences in that synergy between ipinivo? We actually do, based upon imaging data. I would say that, you know, good responding cells, bad responding cells, I'm assuming that's what you mean. We haven't done it between different donors. We basically stuck with one donor in these experiments, so I can't answer that question I can't answer that question for diversity, but yes, it is definitely at least heterogeneity amongst the cell population that we see. So, that suggests a potential autonomous effect on hepatocytes themselves? I would agree. I would agree, yeah. Interesting. Yeah, great. Okay. Do we have any more questions? All right. If not, I would like to thank Dr. Vernetti and Dr. Meidel for giving two very interesting and compelling seminars on their 3D liver microphysiologic systems, and with that, I think this closes out this SIG. So, thanks to everyone, and thanks to Dr. Dara for moderating, and I'm happy to step in at the end here. So, everyone have a wonderful day. Thank you. Thanks. Thank you. Bye-bye. Bye. Thank you.

Hepatotoxicity

Liver Microphysiology Systems

LAMPS

Hepatic Intrinsic Clearance

Drug-Induced Liver Injury

ADMET Studies

Precision Medicine

Metabolic Dysfunction-Associated Steatotic Liver Disease

Induced Pluripotent Stem Cells

Patient-Specific Models

Multidisciplinary Approach