Patients metabolize drugs differently. Personalized physiologically based pharmacokinetic (PBPK) models improve precision dosing
Finding the right dose can be vital, especially in medicine. Medical therapies need to produce a certain concentration of the drug in patients’ blood for optimal healing. However, because metabolism efficiency differs from patient to patient, recommended standardized drug doses are not always effective. Rebekka Fendt, a systems biologist in the LiSyM research network, has been working on the promising principle of model-informed precision dosing toward personalizing physiologically based pharmacokinetic (PBPK) models for her doctoral thesis. “These are able to predict the optimal dose for each individual more precisely,” she says.
Good PBPK models are detailed representations of the physical processes in the body
“The goal of physiologically based pharmacokinetic models is to reflect the entire organism,” says Fendt, who is a doctoral student in the research lab of Professor Lars Küpfer. PBPK models are mathematical models that integrate all organs as virtual containers connected by a modeled blood circulation system. A good PBPK model can work through the details of all bodily processes that occur when medication is taken orally, for example: from its absorption in the intestines, to its distribution via the blood stream, metabolism in the liver, transport to the area of treatment, all the way to its excretion via the kidneys.
According to Fendt, PBPK models are especially effective for conducting virtual studies in which researchers want to simulate the interaction between medications or focus on specific patient groups, like those who are already taking other medications or have a damaged liver or kidneys. To be able to do this, PBPK models need to be able to reflect individual characteristics. However, currently most work only with average figures – in other words, “with idealized average patients,” says Fendt.
Average height and weight apply to only a few patients
Most patients do not fit the average height and weight. In addition, metabolism efficiency varies from patient to patient. This is why the one-size-fits-all approach to standardized doses can never work for all patients, which can have serious consequences. A fast metabolism or interactions between drugs can cause the optimal drug concentration to be reached for only a short period of time or not at all, with the result that optimal and successful treatment cannot be guaranteed. On the other hand, too much of a drug can also have harmful consequences, especially when the liver or other organs are already damaged.
“In individualized PBPK models, doctors are able to enter patients’ personal data to determine the most precise dose for them,” says Fendt about what could be a model for the future. This method of precisely estimating individual doses for patients using a mathematic model is called model-informed precision dosing (MIPD). Fendt employed this method to personalize a PBPK model for caffeine in an exemplary project. Caffeine was an ideal substance for the model because the way it is metabolized and its effects on metabolism in the body have been thoroughly researched. Fendt’s new model produces much more precise personalized predictions for the metabolism of caffeine than the model version that uses only data for average individuals.
Developing models based on experience and careful deliberation
“PBPK models are mechanistic,” says Fendt. They show how thousands of bodily processes are connected and have an effect on one another. This makes the models extremely complex. The disadvantage of this is that they require a broad basis of data to answer questions such as: How water soluble are drugs? How well do they bind with plasma proteins, or are they metabolized? “If you don’t have the data, you have to review the literature,” says Fendt.
Although PBPK models are based primarily on measurement data, they also work with hypotheses in the form of estimates and adjustments. “We have to estimate unknown and unmeasurable parameters,” says Fendt. Many measured data, such as in vitro results, must also be “fitted,” as the estimation of parameters is called. In order to do this, established mathematical formulas, logic, and experience are required. Fendt employed a modeling software available online to conduct the parameter estimates, as well as to design the model and carry out the simulation.1 Other modelers should hopefully also be able to produce a model that plausibly reflects the metabolic process. “Ideally, clinical patient data are available for that,” says Fendt. Researchers can then use these data to evaluate the model by assessing the accuracy of the model’s predictions.
Virtual twins represent real study participants
“Determining the quality of my model was my greatest challenge,” says Fendt about her personalized PBPK model for the caffeine metabolism project. “I had the huge advantage of being able to use data from a study,” she adds, referring to a caffeine study conducted at the Institute of Clinical Pharmacology at the Robert Bosch Hospital in Stuttgart in which she and her doctoral advisor collaborated. In this study, healthy volunteers were given a cocktail of five low-dose medications together with 50 milligrams of caffeine. Over a period of eight hours, researchers measured how efficiently the participants metabolized the caffeine by determining the amounts of caffeine and paraxanthine metabolites in their blood.
After constructing a PBPK reference model based on data for average individuals and making all the necessary adjustments, Fendt created a virtual twin for every real participant in the caffeine study. She then used these twins to increasingly personalize her model in several steps. In the first step, the virtual twins differed from their live counterparts only in terms of demographic data: age, weight, height, and sex. She then modeled each twin. “The predictions became significantly better,” she says. One reason for this is that height and weight have an impact on metabolic efficiency, because they correlate with the size of the liver, for example. Sex is also important because the distribution of body fat and muscle mass is different in men than in women. Age was the only factor that did not affect the model. All volunteers were between 18 and 50 years old – an age span during which age-related limitations seldom occur.
The higher the level of personalization, the more precise predictions will be
In the second step, Fendt integrated additional physiological data from laboratory tests, such as blood flow in the liver, glomerular filtration rate (GFR) for measuring kidney function, and the hematocrit level. “Due to the pharmacology of caffeine, these data did not improve [the model],” says Fendt. In the dose in which caffeine was administered to the volunteers, the only limiting factor was CYP1A2 activity – in other words, the speed with which this liver enzyme transforms caffeine into paraxanthine.
In step three, she therefore not only looked at patients’ demography and physiology, she also concentrated on their CYP1A2 activity. But because this cannot be measured, Fendt indirectly estimated the level of CYP1A2 activity based on the ratio of paraxanthine to caffeine in the volunteers’ blood four hours after they had ingested the drug cocktail. “The improvement was even more pronounced than with the demographic data,” she says. Level-three personalization was thus able to provide the most precise predictions regarding the individual caffeine pharmacokinetics in 23 of the 48 volunteers.
Other PBPK models also stand to profit from personalization
Fendt was able to demonstrate that personalized PBPK models are ideal tools for improving precision dosing for patients. The results of the study have now been published.2 When Fendt began working on her PhD project at Bayer AG, her doctoral supervisor Lars Küpfer was still a senior researcher at Bayer Technology Services in Leverkusen. Now Küpfer is Professor for Applied Microbiology at the RWTH Aachen University and Fendt is in the process of completing her doctoral thesis.
Sooner or later, personalized PBPK models like hers will be a part of everyday clinical use. “How easily can our personalization be applied to other systems?” she wonders. How can we make predicting the metabolism of other substances, particularly those that are highly relevant to clinical practice, more precise? Whether Fendt will soon be working at a university, a university of applied sciences, or in industry, one thing is certain: She will continue to explore questions like these with new personalized PBPK models in the future.
1 www.open-systems-pharmacology.org
2 Fendt R, Hofmann U, Schneider ARP, Schaeffeler E, Burghaus R, Yilmaz A, Blank LM, Kerb R, Lippert J, Schlender JF, Schwab M, Kuepfer L. Data-Driven Personalization of a Physiologically Based Pharmacokinetic Model for Caffeine: A Systematic Assessment. CPT Pharmacometrics Syst Pharmacol. 2021 May 30. doi: 10.1002/psp4.12646. Epub ahead of print. PMID: 34053199.