DMPK Insights #14: Evolution of CNS Drug Disposition Models

Scott Summerfield: Hello and welcome to this Pharmaron podcast, part of our DMPK Insights podcast series. My name is Scott Summerfield and I lead the Metabolism Group in our UK based integrated DMPK discovery and development platform.

Today I’m delighted to be joined by Professor Elizabeth de Lange to discuss the evolution of CNS drug disposition models, taking a brief tour of the past, current state, and what the future might hold. Welcome, Elizabeth.

Elizabeth de Lange: Yeah, thank you so much. I’m very happy to have this podcast with you. That is very nice.

Scott Summerfield: By way of introduction, Elizabeth is a professor in predictive pharmacology and principal investigator at the research division of Systems Pharmacology and Pharmacy at Leiden Academic Center for Drug Research, abbreviated to LACDR for the rest of the podcast. With her team, she’s helping to unravel the rates and extent of mechanisms that govern central nervous system, target site pharmacokinetics, and the related pharmacodynamics with special emphasis on physiologically-based (PB) translation between species and conditions.

Ultimately this is to have a mathematical model that supports CNS drug development, and that includes reduction and replacement of nonclinical studies and can predict the best possible treatment for CNS conditions in individual patients. Elizabeth has contributed more than 160 publications to the peer reviewed literature, over 170 invited lectures and organized in excess of a hundred conferences, symposia, workshops, and courses. She has multiple leadership positions in scientific projects, also at the LACDR, nationally with the NVF, in large international organizations such as AAPS, and is currently coordinating the EU Consortium QSPainRelief.

She has a number of advisory roles, both advice and consultancy. And alongside all of this, Elizabeth contributes to education in the biopharmaceutical sciences, Bachelor and Master’s courses. Among other honors, Elizabeth received the AAPS Fellow Award in 2013, an honorary doctorate in pharmacy from Uppsala University in 2020, and the prestigious Sheiner Lecture Lifetime Achievement Award from ISOP in 2020.

Also with note, here for this podcast, is physiologically based models such as LeiCNS-PK3.0 that predicts pharmacokinetics in multiple locations within the CNS of mice, rats and humans, and is an important an example of the topic that we’re gonna cover today.

So again, Elizabeth, thank you so much for the chance to talk with you about CNS, DMPK and PD. It’s fascinating because the unique properties of the blood-brain barrier and the highly privileged position the brain holds in health and disease. Recently I was reminded of an article published in 1997, co-authored by yourself, Margareta, and Lennart Paalzow where there was a simple two compartment model for the passage of unbound drug across the blood-brain barrier from blood to brain. And actually probably the first mention of what would become known as Kpuu, the unbound concentration of drug.

Fast forward to today. The modeling efforts are far more sophisticated. And again, a great example being LeiCNS-PK3.0 which describes multiple compartments and fluid flows during the passage of drug along its journey from the peripheral circulation across the blood-brain barrier through the brain. So over that period of about 30 years, multiple scientific advancements have collaborated, leading to the emergence of PBPK-PKPD tools, and hence the topic for this podcast.

But first off, can you describe what kindled your interest in modeling and helping to answer key questions around the drug disposition into the brain?

Elizabeth de Lange: Yeah, sure I can. Yes. So, of course there is a driving interest in trying to understand everything that governs blood-brain barrier transport, and also what goes beyond in terms of pharmacokinetics in the CNS itself, in the central nervous system itself. And looking at that we know that is, it is rather complex and there are many processes ongoing that are kind of interrelated.

That means that in one location, PK may differ from the other ones because local pharmacokinetic processes might differ as well, and we just cannot put that in a graph in a 2D way. Or in a 3D way, because it’s actually multidimensional and to the end, you kind of need advanced models, mathematical models to put these all together.

And, I started with microdialysis as a tool to investigate the blood brain barrier transport and beyond. And I was in a group where there was a lot of modeling, so I also kind of learned aside from the modeling and with that, I kind of integrated it into the microdialysis work and ultimately with the modeling, I visualized the CNS pharmacokinetic model that would be like a Tomtom navigator because.

So this navigator you put in what kind of traveler do you have? Are you a bike? Are you a car? Are you a boat or a train? and, where do you want to go and what will be there on your route? What traffic will you go? So what you need is an infrastructure. You need to know what is the traveler and you need, and the outcome is the resulting traffic.

So that kind of relates to the CNS infrastructure, the drug that is traveling and the traffic is actually the pharmacokinetics. So I thought, okay, so how did they work? So of course they needed a lot of input on the development of that model, and they needed time and location specific traffic. And with that, kind of the prediction came. So that was kind of how things got sparked.

Scott Summerfield: Well, it’s a nice analogy actually, the traffic one. I’ve not heard that before. But it’s useful to kind of contextualize. Thank you.

Elizabeth de Lange: So it’s, but everyone can easily understand because they use it, right?

Scott Summerfield: Well, yeah, absolutely. And the thing is, it strikes a point about mathematical models. In a sense they are quite abstract, you know? So being able to contextualize that as an introduction’s obviously really valuable and that’s great. Thank you.

So in terms of understanding the relationship between unbound concentration and effect, brain Kpuu is now a well established parameter, and like I said, that 1997 paper was one of the was one of the first kind of mentions of unbound concentration gradient across the blood-brain barrier. But there’s another important part of the story, which I took from your bio sketch and namely that’s contextualizing both rate and extent of brain passage and transport. And I know having been a strong advocate for considering both rate extent by yourself, I wondered if you could share your thoughts on this and the importance of modeling the relationship between concentration and effect.

Elizabeth de Lange: So indeed has, so I think also with Margareta, we teamed up so nicely in so many ways. And also that paper was a kind of very good start off of our good collaboration and insights that we ,on the way, got it together, this piece by her, that piece by me. And we kind of, used it over, crossover way, but indeed, so the Kpuu brain, or maybe mechanistically we should call it Kpuu blood-brain barrier because it focuses more really on the blood-brain barrier itself.

It is actually, the ratio of unbound drug concentrations on both sides of the blood-brain barrier, but at steady state. And actually, it is not the ratio at any point in time. So it means that the pharmacokinetic profile of the brain extracellular fluid depends on both, and that is what the receptor sees. So if we have a very simple case of a compound just passively distributes into the brain, and it has a Kpuu blood-brain barrier of one. So this is also in the, 1997 paper. it could be a compound that is slowly crossing the blood-brain barrier or fast, or everything in between, and it really makes a difference in how the relationship is between the unbound concentrations in the brain extracellular fluid relatively to the unbound concentrations in plasma.

So it means that yeah, so the receptor or the target does not see a Kpuu. It sees the concentration as a function of time. So that is what we need to take into account. And that is why I believe that also we have to address rate as a very important one here. So the 1997 paper was actually the thinking beyond. So it’s rate and constant.

Scott Summerfield: Thank you Elizabeth. And I know there’s papers such as the Mastermind one that you did subsequently, which elaborates a lot of this and they’re fantastic read for those listeners out there.

So one approach that I’ve seen to describe how a model breaks down into four categories, and it’s kind of a loose thing and maybe we can just talk around it. Today is firstly whether a model is descriptive, so trying to understand what’s already happened. Diagnostic, trying to explain why we’ve seen it. Predictive. What could happen and then prescriptive what should happen.

And just going back to that 1997 article and I’ll mention the name here so you know, people can, can go look at it. So it’s drug equilibration across the blood-brain barrier, pharmacokinetic considerations based on the microdialysis method. And that was published in Pharmaceutical Research Journal, At the time that was arguably a descriptive model focusing, and essentially on assessing brain pharmacokinetics at the level of the BBB.

So where do you feel that we are now in terms of modern models like LeiCNS-PK3.0, I guess with respect to being descriptive, diagnostic, predictive or prescriptive? And then before that, the kind of key developments that took us from those very simple models to where you think we are now.

Elizabeth de Lange: So I like very much your kind of categories of what a model could be, or how to describe the model. Actually for the 1997 paper, it was not actually descriptive because it did not fit data. It was actually more a theoretical approach, to understand how rate and extent would of blood-brain barrier transport would influence the relationship between brain extracellular, fluid pharmacokinetics and plasma pharmacokinetics.

Of course, the unbound should always be the unbound, and it also included the so-called what if scenarios. So if a drug is passive with regard to blood-brain barrier transport, if it’s passive only, active influx transport, active efflux transport or combinations of that, what could we see? And actually with that, it gave kind of insights if what could happen. And so also, yeah, and of course the model was elicited by a discussion between Margareta and myself on the microdialysis data that I produced because that was new, because CSF was the side of sampling before, and so it provided also new information on how things could work and with the theoretical approach actually, it could also kind of explain what we have seen with microdialysis and also a little bit what could happen but not yet what should happen because it was too simple at the time. So it was really revolutionary at the time, of course.

But, then we got more insights, et cetera. And with the LeiCNS model, we actually also included what, not only what happens at the level of the blood-brain barrier, but also beyond in terms of what is the bulk flow of the extracellular fluid, what is the intra extracellular distribution, what is the CSF flow and all these kind of things. And also in terms of the physiology as it is.

And with that, we could actually have information on a pharmacokinetics that is indeed location, CNS location dependent, dependent on the relative contributions of the pharmacokinetic processes around that location. It can clearly show drug dependency and also a drug property can be put in to see what drug property would lead to what kind of pharmacokinetic profiles. And also it clearly shows the species dependency because it’s ultimately based on physiology that you can scale. You can scale the physiology and the drug remains the drug, whatever subject you give it to or whatever situation you give it to. But the physiology, so the context is the different one.

And what was very important to that end is, in developing is also that you have data that shows the interaction of the processes within. The same single subject there. So now I touched a little bit upon the mastermind, approach published in Fluids and Barriers of CNS in 2000, what was it, 10 or so? But, that we had so-called smart data, measuring with microdialysis not only in one spot in the brain extracellular fluid, but also in different locations in the CSF and kind of combine all that information to tease out with mathematical modeling what should be the interrelationships, and then do that for multiple drugs. And then in a very structured way, have kind of the general principles of drug distribution across the blood-brain barrier and into, and in the CNS itself.

Scott Summerfield: Thank you. Obviously there are two in any model, you’ve kind of got drug specific parameters. Yeah. And then you’ve kind of got system specific parameters and you kind of mentioned all the flows and, and things and they’re obviously really important in the, in those differential equations that are really describing what happens. I guess there’s quite a lot of fundamental research besides the computing capability to kind of build that system specific data set up across those three species. And I guess did, you draw on literature and maybe experiments as well to build that information up.

Elizabeth de Lange: So the key technique was actually microdialysis and we did it in rats, so that was the species that we build up all the information. We also did microdialysis experiments in mice, in earlier stage and also in a little later stage. And, of course in human we do not have, or only scarcely have microdialysis data. So that was also the Margareta group that actually had the results of morphine, for instance, in TBI, traumatic brain injury patients.

But, microdialysis was key, but also for all the physiological information, you need to have a picture of what the brain is. So the brain or the CNS infrastructure. I made it kind of a drawing of that and very kind of a block here for a cell and a blood-brain barrier here. Blood-CSF barrier there and then flows, et cetera, just to visualize the map of the CNS and then collecting indeed information on surfaces, on flow, and then in different species. So now we have mice, human, and rat. We are working on the pig as well at this moment in time.

So, trying to also have this bridging possibility, because if you know the physiology of a certain species, you can bridge it to the physiology of another species, and then with that bridge, the pharmacokinetics, and that makes it ultimately a very strong way, to be used also across species, across drugs, across conditions, et cetera.

Scott Summerfield: I guess back in 1997, I can’t even remember what computing was like back then.

Elizabeth de Lange: It was rather simple actually. And if you also see the differential equation, it was advanced at that time,? And it was really, oh wow! So at that time of feeling okay, we can kind of start to understand how things might work. And then it was so interesting, and that was with, Berkeley Madonna, that was kind of a software where we did these things in and also explore many other things. It was yeah much simpler at that time. And now of course it is circles of differential equations and all kinds of everything but it’s step by step. And with that and also we are kind of very deep into all aspects of the CNS as far as we can get with that.

Scott Summerfield: Just reflect on a bit of the stuff that we’ve just covered. I thinking of on modern CNS models and those advancements, what do you see as good examples of major achievements of what’s now possible in models like Lei-CNS PK 3.0 and others that are trying to do the same thing in terms of CNS PK and PKPD.

Elizabeth de Lange: Yeah, so it actually, of course it is the scientific community with CNS PBPK modeling and PD modeling that altogether makes going to the next stage in the last years and well, there is so much more possible. So if I look at the LeiCNS model, because I can talk on that’s easily, of course, we can kind of see or we were able to adequately predict the very scarce data from human. That is also the reason because we cannot just sample it. We can predict and, that makes it much stronger. So if that is adequately done even if it is CSF that is not the target site. So we can see if CSF is okay, we gain trust in also the predictions of brain extracellular fluid. And then with that we come to a next step.

And of course it is the concentration at the target site that drives the pharmacokinetics. So I think with, with now, so what we did for instance, was see, okay, what would change for small molecules in terms of drug distribution for instance,for Alzheimer’s patients. And the interesting thing of that was that actually we predicted for the three drugs that we investigated, that we did not predict changes so much. And that is also an interesting thing. And then there were also experimental data in mice, in Alzheimer mice that showed the same thing. So there was for small molecules, the ones that have been investigated, there was no change.

So with that, for instance, we learn what has been published about, all kind of changes in Alzheimer’s disease at the blood-brain barrier should merely reside on other mechanisms, not governing the small molecules. And so not so much the efflux and the influx and the passive things, but merely the vesical based transport modes. And if you look back, for instance, horse radish peroxidase, for instance, is also not just blood-brain barrier as it, we call it all blood-brain barrier permeability. But of course we have different aspects of that. And horseradish peroxidase is also actually vesicle based transported, so that is interesting. So with that, you can also kind of tease out what is going on.

So let’s see, what was your question remaining. So what is now possible? Yes, what is possible? So we can in a way predict for any drug molecule where we do have, of course the pharmacokinetic input. So pharmacokinetic in terms of plasma. So what kind of profile would we expect there? And that is also where we can use other PBPK models to see if you have this dose regimen, what would be the resulting plasma pharmacokinetics? And that is used as an input to expose the CNS because we have a CNS-PBPK model. And then we could see what would we expect for a certain drug molecule to, how would we expect its pharmacokinetic behavior? And that could be related to the need for, what concentrations or the information for what concentrations would be needed to stimulate a receptor or to have a proper interaction with the target, et cetera. So that could still be done in vitro, but also in vivo we can do a lot of things.

So, you mentioned in the introduction indeed, I’m chairing the European Horizon 2020 QS Pain Relief consortium. And in that we really try to integrate, different models in sequence. So we have the CNS-PBPK model, then we have a CNS binding kinetic model, then we have neuro circuit models. So it’s of course with the consortium partners and we have pharmacodynamic data at the level of in vitro cellular aspects. We have in vivo animal work, we have human participants, the healthy ones and the ones for QSPainRelief. Of course, the pain patients, and we have the pharmacodynamic outcomes.

So we try to with actual data as well as model approach, we try to integrate and with that, have picture on what is the pharmacokinetic that drives ultimately the pharmacodynamics, for instance, at the level of analgesia, but also on cognitive impairment, abuse liability and all these aspects that come for instance, with the use of opioids that are kind of central in this study. And with that we play a kind of Sudoku game. And so I mentioned Mastermind before, but now it is much more complex. So now we’re playing the Sudoku game and we try really to tease out what. And so for each part we try to peel off from dosing to the ultimate effect, all the mechanisms in between. And we have come to the receptor occupancy. We want to get beyond that of course, to really see what is between receptor occupancy and the ultimate effect. So for instance, cyclic AMP changes and all these kind of things that can be derived from in vitro. So with that kind of constantly progressing in our understanding and in our predictive power of ultimately. And that is my ultimate aim to be able to be able to predict the pharmacodynamics in human ultimately.

Scott Summerfield: That’s really nice applied research really. Because obviously, I mean, Paul Morgan and colleagues at Pfizer put forward the three pillars of drug survival, which I mean. Some people now refer to kind of as the four pillars. ’cause you’ve got the downstream stuff. And actually the more you know, the better you can predict, right?

Elizabeth de Lange: Yea, of course. And the attrition and CNS drug development is really high. It’s very hard and in many circumstances to kind of derive an unbound concentration in a human brain without, resorting to, well PET gives you total CSF, there’s the lag and the meaning of transporters flow and everything. And then the downstream pharmacology as well that you mentioned.

Scott Summerfield: So back in the day when I, just before 1997, I studied biological mass spectrometry, which now is kind of referred to as proteomics. And part of one of the things back in the day I would never have imagined was this sort of pharmaco-proteomics where you could look at the levels of transporters in the blood-brain barrier and other regions organelles and kind of begin to use that. Is that something that’s, how is that integrated into kind of the modeling in CNS now?

Elizabeth de Lange: We definitely do that and indeed thanks to the progress in this mass spectrometry field, it’s really intriguing. We use expression ratios between species for certain transporter, when it’s known as one of the translators as well. And for instance, if we have a Kpuu in the rat and we don’t have it in human, how can we make a translation on that? But also we dive deeper, so we do multiple new things. So there are things that have not been addressed explicitly so far.

So that is metabolism for drug metabolizing enzymes in the brain. So that is what we currently work on and we have data that we now can really capture. So that is nice. Also, CSF flow, what would happen if you drug a dose intravenously but also intra cerebral ventricularly because then it should be the same flow, right? If the drug doesn’t change the physiology, that is also what we have to take into account, of course, but that’s assume it does not. And now we have for the rat, we also have adapted CSF flows according to what would happen after intra-cerebral ventricular administration and it should of course also hold for the intravenously administered drugs. And it does. So that is also an improvement. Then also brain tumor.

But coming back a little bit more on your starting in the question with regard to, pharmaco-spectrometry, pharmaco-mass spectrometry, we are also diving into, or we have already made steps, with that, if we have blood-brain barrier, and we want to know also on that level, and also of course for the rest of the brain, but what is there as a transporter and what is the expression of the transporter and the protein expression, but also what is the relationship between protein expression and functionality for certain drug?

And now we have a kind of a sneak preview. We have work that will be published, in the coming time where, we found that there’s a drug dependent relationship between, the change in expression of a certain P-glycoprotein in this case that we took. And the resulting change in activity for that drug. Because we had at one point in time studies that were done in about a little bit more than 10 years ago in epileptic animals where we found in the same animals and are always a very important emphasis because then the data are connected.

We found after induction of a seizure. We did experiments with animals with quinidine and p-glycoprotein substrate. And so we had the profiles with microdialysis and then we added Tariquidar to block PGP, and then we found, then we just at different times after the seizure. So different animals were kind of used for one day, for two days, three day, four days, up till 30 days. We also measured in histology PGP expression, blood-brain barrier expression of PGP, and related that to the activity of PGP, and there was no relationship.

So there was a huge change in time in the expression. So it went up and it went down, but it was not related to a change in the pharmacokinetics of quinidine being a PGP substrate. So that is also what kind of spark, okay, what is going on? And we now have a handle on that. So that is what the future will bring, and with that also if we know expression of transporters and we have a little bit more information on yeah, what drug and what is that relationship we can make next steps. What would we expect in that case? For instance, in diseases, like Alzheimer’s disease, P-glycoprotein changes, does it make a difference for a certain drug? If we talk about small molecules, large molecules is a different part, of course.

Scott Summerfield: Elizabeth, thank you for sharing that. I will wait with baited breath. It’s an area the CNS is a very dear to me from my time in pharma. So that sounds fantastic.

So, we’ve talked a bit about sort of that as like an emerging opportunity as well. The development in CNS PK and PD modeling. What do you think would be valuable next from a research perspective? Obviously you’ve got sort of the very heavy science that happens in academia and also in industry. What do you think are the kind of important things that will really help the next generation of modeling?

Elizabeth de Lange: So what I think if it comes to the role of pharmaceutical industry, I think there should be a fourth and back. Because pharmaceutical industries have lots of data that could be shared in a way and ultimately up till. So what has been done in vitro setting, in preclinical animals, what were the first in human studies to really see everything. And also with what successes that is what we know of. And then there’s a little bit published about things, but not all the data are there. It is just kind of what we know, but also why things did not work. Because we can learn so much of all the things that did not work. And there’s so much of data and I think, we can team up so academia can team up with pharmaceutical industries, just to get a handle on why is something successful and why is something else not. And that is also the big Sudoku game again. If you have the people that like doing games like that to defragment information to really order it and to put it into a mathematical model that really can explain everything. Then I think we are really in the next leap.

Scott Summerfield: Elizabeth, thank you so much for sharing your thoughts on the current state of the art with models such as LeiCNS-PK3.0 and others, and what the future might hold. Thank you very much.

Elizabeth de Lange: Yeah, thank you too. It was really nice to have this conversation with you.

Scott Summerfield: And, also to our listeners, thank you for joining us for today’s episode of Pharmaron’s DMPK Insights series. We would like to remind you that our DMPK webinar series is also available on demand, covering a variety of key questions related to DMPK science in drug discovery and development. Stay tuned for more podcasts in our DMPK Insight series. Thank you very much and bye for now.