Adeno-associated virus (AAV) mediated monoclonal antibody (mAb) expression in the brain represents an exciting one-and-done therapeutic option for many central nervous system (CNS) disorders that are hard to treat following systemic mAb administration. To facilitate the development of such AAV therapies, here we have developed an AAV-transgene-target quantitative systems pharmacology (QSP) model that can simultaneously characterize the disposition of AAV, expressed antibody, therapeutic target, and the mAb-target complex in plasma, tissues, and the brain. The model was built by integrating previously published platform physiologically based pharmacokinetic (PBPK) models for AAV, protein therapeutics, and brain, with special emphasis on transport pathways and transduction mechanisms within the brain. The developed QSP model was able to capture the pharmacokinetics of AAV and the expressed mAb in the systemic circulation and brain tissues following AAV administration via diverse routes in the rats. The model suggested that intra-cisternal magna injection of medium AAV dose or intrastriatal injection of low AAV dose would provide better therapeutic options compared to intravenous administration of high AAV dose. Locally administered AAV was predicted to lead to sufficient antibody exposure and substantial target engagement in cerebrospinal fluid (CSF) and interstitial fluid (ISF), while minimizing systemic AAV and antibody exposure, leading to maximal efficacy and minimal toxicity. Overall, the present work not only supports cross-species and cross-serotypes translational potential of previously published AAV PBPK model, but also highlights application of QSP modeling to guide the development of AAV therapies for neurological disorders.
Quantification of drug-cell interactions and subsequent cellular responses by using experimental data together with mathematical models of assumed binding and signalling schematics is vital to many research programmes; data fitting provides estimates for important pharmacological parameters including kinetic parameters controlling drug affinity and efficacy. Ordinary differential equation (ODE) models are a key component of many receptor theory studies used for this purpose. In using ODE simulations to fit experimental data and estimate these parameters, the theory of the identifiability properties of the system is often overlooked. Indeed, structural identifiability analysis (SIA) is often overlooked in many fields of bio-modelling. Building on recent SIA for linear ligand binding models in receptor theory, we present a new analysis of identifiability properties of nonlinear receptor theory models. We include models of ligand depletion in binding assays and ligand-induced dimerisation (LID). The classical SIA approaches of Taylor Series and similarity transformation are applied, using detailed step-by-step calculations to illustrate the complexity of the implementations. New results are obtained which show that the nonlinear ligand-depletion counterpart models of non-identifiable linear ligand excess models are globally identifiable from a single timecourse. Also, the LID model is shown to be globally identifiable if an experimental aparatus-dependent parameter is obtained. The analysis highlights issues of tractability of the methods for similar and higher-dimensional nonlinear models in receptor theory.
Adeno-associated virus (AAV) mediated monoclonal antibody (mAb) expression in the central nervous system (CNS) constitutes a promising approach for treating various CNS disorders. However, a comprehensive understanding of AAV biodistribution and antibody expression in the brain following viral vector administration via different local CNS routes and systemic administration is not well established. AAV9 and AAV5 vectors encoding antibody against a protein associated with neurodegeneration (i.e., anti-Target-X mAb), as well as AAV9 encoding a control antibody (Ctrl-mAb), were administered to rats via intravenous (IV), intra-cisterna magna (ICM), and intrastriatal (IST) routes. Effect of immunosuppression regimen was also investigated. Blood/plasma concentrations of the vector and mAb were measured at day 21 and day 56, the latter only for ICM group. AAV and mAb concentrations in different brain regions, interstitial fluid (ISF), cerebrospinal fluid (CSF), and some peripheral tissues were analyzed. Both the IST and ICM routes showed dose-proportional, and higher vector biodistribution and mAb expression in the brain compared to the IV route. Twenty-fold lower AAV dose administered via IST route produced similar transgene expression higher dose administered via ICM route. Non-immunosuppressed rats exhibited a sharp decline in mAb plasma concentration after 2nd week and lower brain and tissue homogenate concentrations at 3-week. IST and ICM administration of AAV results in higher and more sustained expression of antibody in the brain compared to IV delivery, particularly when immune responses are controlled using concomitant immunosuppression regimen.
The evolution of nonlinear mixed effects (NLME) modeling reflects a continuous cycle of innovation based on advances in numerical methods and computational power. This commentary outlines the evolution of NLME modeling that began with linearization-based approaches in the 1980s, progressed through sampling-based methods in the 2000s, and is now entering a new phase shaped by AI. Variational autoencoders bridge classical NLME modeling with AI-based methods allowing the development and application of AI-augmented PMX models. This opens the route for integrating multimodal data and addressing increasingly complex modeling challenges.
Target-mediated drug distribution (TMDD) is typically classified by accelerated removal of drug after serum concentration drops below a critical level. Susceptible drugs would be dosed to a schedule that maintains exposure above this level, where concentration drops monoexponentially. Out-dosing TMDD often implies that target binding is saturated or degraded. T-cell engagers and other biologics are often given with low dosages which make out-dosing TMDD unrealistic, yet are also typically observed to have monoexponential elimination. It can be challenging to determine whether such a concentration-time relationship is due to TMDD or standard non-specific (intrinsic) routes of clearance, and more so to scale this clearance from preclinical settings, as TMDD-driven clearance may not scale allometrically like intrinsic clearance. In this work, we derived an algebraic expression for the proportion of clearance that is target-mediated or intrinsic, which requires only the intrinsic half-life and the target receptor’s concentration, turnover rate and binding affinity for the drug. We further show that the typical transition from monoexponential decline of serum concentrations to accelerated, target-driven clearance occurs at concentrations given by two expressions in those same parameters. We apply the equations to published data for monoclonal antibodies and T-cell engagers, finding that we consistently characterize their clearance as expected. These expressions can be used to determine if a drug’s clearance would be expected to be driven by target-mediated or non-specific routes at pharmacologically active concentrations, and hence how to accurately translate pharmacokinetic properties from preclinical species to human, a process that is essential for drug development.
Therapeutic antibody dosing regimens are often constrained by formulation and administration factors that limit flexibility in dose and schedule. Subcutaneous (SC) delivery offers advantages over intravenous (IV) administration-such as improved convenience and potentially more favorable pharmacokinetic (PK) profiles-but is typically restricted by injection volume, requiring lower doses and more frequent dosing. Recombinant human hyaluronidase PH20 (rHuPH20) enables rapid administration of high-dose, high-volume therapeutics, thereby expanding the feasible SC delivery landscape. Despite these advances, biopharmaceutical development teams often lack tools to quantify the strategic impact of SC-enabling technologies and interpret PK simulations in a decision-ready format. This work introduces Operating Space Maps, a visual PK simulation framework that organizes multiple dosing scenarios into an intuitive landscape of dose-frequency options. Using rHuPH20 as a case study, a two-compartment PK model is applied to simulate representative antibody regimens for two illustrative scenarios: converting IV regimens to SC with rHuPH20 and extending SC dosing intervals beyond standard limits with rHuPH20. For example, relative to a 1000 mg IV benchmark regimen, a 1400 mg SC dose with rHuPH20 maintains equivalent average exposure while reducing Cmax by approximately 50%. The methodology is broadly applicable to other SC-enabling technologies for which sufficient foundational data exist to parameterize the model. By making trade-offs between Cmin, Cavg, and Cmax explicit and integrating SC delivery constraints, Operating Space Maps enable cross-functional development teams to evaluate regimen feasibility, optimize target product profiles, and guide strategic decisions early in development.
Concomitant use of intravenous immunoglobulin (IVIG) with monoclonal antibodies (mAbs) or bispecific T-cell engagers (TCEs) is routinely encountered across autoimmune diseases, B-cell malignancies, immunodeficiencies, and transplantation. These therapies share a key salvage pathway through the neonatal Fc receptor (FcRn), raising the potential for pharmacokinetic (PK) and pharmacodynamic (PD) interactions. A mechanistic model of IgG disposition originally developed in mice was extended and revalidated with an array of human datasets to investigate how IVIG dose, timing, and IgG pool dynamics influence therapeutic IgG behavior and PK/PD outcomes. The model captures endogenous and exogenous IgG kinetics via FcRn-mediated recycling in a peripheral (endosomal) compartment, coupled with first-order catabolism of unbound IgG. Simultaneous model-fitting against clinical data reproduced serum IgG kinetics and endogenous/pathogenic IgG reductions across autoimmune and transplant settings. It recapitulated the 36
BACKGROUND:Reliable population pharmacokinetic (PopPK) estimation is often compromised by outliers under Gaussian error models. While post hoc filtering using conditional weighted residuals (CWRES) is common, this approach is often insensitive due to model "masking" from variance inflation. METHODS:We implemented a one-compartment model in Monolix using a custom likelihood workaround to benchmark four distributions: Normal, Laplace, Generalized Error Distribution (GED), and Student's t. We assessed CWRES sensitivity under extreme contamination and compared estimation performance using theoretical tail-behavior analysis, controlled simulation studies spanning multiple contamination severities, and a real-world caffeine PK case study with influential terminal-phase deviations. RESULTS:Simulations revealed that CWRES-based diagnostics are unreliable; extreme outliers frequently produced |CWRES| < 6 because the Normal model inflated residual variance, masking the contamination. Exponential-tail models (Laplace, GED) improved robustness for moderate outliers but failed under extreme deviations due to insufficiently heavy tails. Conversely, the Student's t model, utilizing power-law tail behavior, maintained stable and minimally biased structural parameter estimates across the contamination settings examined. These patterns were confirmed in the caffeine case study. CONCLUSIONS:Reliance on CWRES-driven residual screening alone is methodologically fragile. Among the models evaluated, exponential-tail distributions are insufficient for extreme outliers, whereas the Student's t distribution provided the most consistent stability across the contamination settings examined here and showed the most robust overall performance among the residual-error models evaluated when influential outliers were present.
The TALAPRO-2 trial (NCT03395197) showed that the addition of talazoparib, a potent PARP inhibitor, to enzalutamide significantly improved radiographic progression-free survival in patients with mCRPC. Due to adverse events (AEs), approximately 62
Hemophilia A is a genetic bleeding disorder caused by a deficiency or absence of Factor VIII, leading to recurrent and spontaneous hemorrhages. Standard treatment typically involves regular prophylactic infusions of clotting factors to prevent bleeding episodes. However, individual variations in treatment response and reliance solely on plasma Factor VIII levels provide an imprecise assessment of bleeding risk. This study presents an intelligent dose-control system utilizing Deep Reinforcement Learning (DRL) integrated with a hybrid Pharmacokinetic-Pharmacodynamic Time-to-Event (PK-PD-TTE) environment. Within this framework, the control policy is learned using a Deep Q‑Network (DQN), enabling the agent to adapt treatment decisions through interaction with the simulated physiological environment. Unlike conventional methods, this framework allows the agent to observe continuous physiological states via Endogenous Thrombin Potential (ETP) and learn optimal dosing policies through trial-and-error. The proposed DQN agent was benchmarked against standard prophylaxis, a Fuzzy Logic controller, and a Bayesian Adaptive Model-Informed Precision Dosing (MIPD) strategy. Simulation results from a virtual cohort (N = 200) demonstrate that the DQN agent achieves a safety profile comparable to Bayesian MIPD while significantly improving factor utilization efficiency. Notably, in patients with low bleeding‑risk phenotypes, the DRL‑based approach achieved a similar bleeding rate to MIPD while reducing annual factor VIII consumption by approximately 39
Advanced in silico tools, such as physiologically based biopharmaceutics models (PBBM) and physiologically based pharmacokinetic models (PBPK), play a pivotal role in model-informed formulation development (MIFD). In the present study, these methodologies were utilized in the development of a novel rabeprazole modified-release (MR) formulation for improved patient compliance by reducing the dosing frequency relative to existing delayed-release (DR) formulation. A MR formulation containing combination of delayed release and pulsatile release components was designed based on the hypothetical dissolution targets. A PBBM was first established using literature-derived clinical data and used to simulate the performance of formulations with varied dissolution profiles. Initial MR formulation prototypes (once a day) were evaluated in pilot-1 bioequivalence (BE) study against reference formulation (twice a day) but results indicated infra bioequivalence outcome. After pilot-1 study, the formulation and the dissolution method were optimized and then model was refined to predict the pilot 1 outcomes. The validated model was then employed to select suitable pilot-2 prototypes and to prospectively predict their BE outcomes under fasted and fed conditions. The predicted BE ratios were in good agreement with the observed pilot-2 study results. Therefore, the same model was used to make prospective predictions with increased subjects to understand the in vivo performance of pivotal test formulation and lead to the successful bioequivalence outcome, product commercialization, ultimately aiding in increased patient compliance. This work highlights the value of a model-informed strategy in accelerating formulation optimization, reducing development timelines and costs, and strengthening decision-making within pharmaceutical development programs.
This study investigates the feasibility of adaptive optimization in pharmacokinetic/pharmacodynamic (PK/PD) models for predicting individual Bispectral Index (BIS) values during sedation and anesthesia by refining established models—specifically the Schnider, Marsh, Schuttler, Eleveld model for propofol and Minto model for remifentanil — using bounded optimization techniques. We adapt these population-based models through cumulative intraoperative BIS data to derive an effective set of PK/PD parameters for improved patient-specific BIS prediction. Comparative analyses using datasets from sedation and general anesthesia demonstrate that wider parameter bounding constraints improve model performance, as shown by higher Pearson correlations, and lower prediction errors, while maintaining computational efficiency. Our experiment results showed that Sequential Least Squares Quadratic Programming (SLSQP) and Least Squares methods excel in accuracy and computational efficiency, having the highest Pearson correlation of 0.56 and lowest root mean square error of 10.3. Real-time updates to PK/PD models using continuous BIS data may improve patient-specific BIS prediction during sedation and general anesthesia. These findings support the feasibility of adaptive calibration of population PK/PD models as a component of future model-based anesthetic management. Individualized dosing strategies could support closed-loop anesthesia systems and advance personalized medicine in perioperative care.
The historical origins of mechanistic pharmacokinetic–tumour growth inhibition (PK–TGI) modelling are usually located in the later twentieth century. However, many of the essential ideas were already present in the 1930s, although not brought together in a single formal framework. In 1937, Torsten Teorell published a physiological theory of drug absorption, distribution, and elimination based on transport across biological boundaries and exchange between anatomical spaces. Five years earlier, W. V. Mayneord showed that the growth of Jensen’s rat sarcoma was more naturally described on a linear dimension than on tumour volume and explained this behaviour mechanistically by assuming proliferation within a thin outer rim surrounding a necrotic core. Read together, these papers contain the key ingredients of a physiological PK–TGI model: Teorell provides the concentration–time function, C(t), and Mayneord provides a growth law expressed on tumour radius, dR/dt = g. The missing link is an effect term coupling exposure to inhibition of radial growth. In its simplest form, this yields dR/dt = g - ψ(C). This article is not intended as an exhaustive review. Rather, it argues that the 1930s should be recognised as a formative period in mechanistic biomedicine in which the conceptual foundations of physiological PK–TGI modelling were already available in print. Selected later examples are used only to show that, even when tumour growth is recast in exponential, Gompertzian, reaction-diffusion, agent-based, or modern semi-mechanistic PK–TGI language, the field repeatedly returns to the same central point: for a solid tumour, net growth is governed by the behaviour of a viable outer region rather than by the bulk alone.
The selection of a “good” model usually involves a combination of objective and subjective criteria. Although many aspects of model quality can be expressed numerically, certain desirable characteristics remain difficult—or even impossible—to quantify precisely. Multi-objective optimization (MOO) provides a systematic way to handle this challenge by explicitly incorporating and balancing both objective (measurable) and subjective (judgment-based) considerations when choosing among candidate solutions. The generated Pareto front represents a set of non-dominated models where no single solution can be improved in one objective without sacrificing the performance in another objective. Using the non-dominated sorting genetic algorithm II (NSGA-II), an implementation of MOO, we simultaneously considered objective function value and number of estimated parameters as competing criteria. Concentration measurements of 17-DMAG, quetiapine, clozapine and ziprasidone were applied to build population pharmacokinetic models through traditional stepwise search, machine learning based single-objective hybrid genetic algorithm (SOHGA) and MOO. Local downhill search with MOO was also assessed in this study. While both objectives improved, models with lower objective function value generally contained more estimated parameters. The number of non-dominated solutions for DMAG, ziprasidone, clozapine, and quetiapine was 17, 9, 9, and 13, respectively. The optimal model selected by SOHGA appeared on the Pareto front for DMAG, ziprasidone and clozapine datasets. Overall, MOO provides objective transparency to the cost of tradeoffs between competing model objectives, allowing researchers to better contextualize subjective criteria (e.g., biological plausibility, improvements in diagnostic plots) when aligning model selection with clinical context.
Medication nonadherence and antibiotic resistance are well-recognized threats to public health. In this paper, we use modeling and simulation to investigate how nonadherence affects the emergence of antibiotic resistance. Using a pharmacokinetic (PK) and pharmacodynamic (PD) model that tracks the stochastic bacteria population, we find that (a) missing just one or two doses may significantly increase the risk of resistance, and (b) this risk can be alleviated by taking double doses after missed doses. Furthermore, we quantify how such double dosing increases the antibiotic plasma concentration based on the antibiotic half-life and dosing schedule. Though the effects of missed doses depend on PKPD parameters, this study presents a framework to probe how adherence affects resistance for specific antibiotics. More generally, our results highlight the need to reexamine conventional recommendations for handling missed doses.
Pharmacokinetic (PK) models are widely used in drug development and dose planning, but they often face a trade-off between interpretability and flexibility. Traditional physiology-based PK models are transparent but may be too simple to capture complex or nonlinear behavior, while AI models are more flexible but can violate basic physical constraints such as nonnegativity or mass balance. In this work, we propose a constrained hybrid AI-PK approach in which small AI components are embedded within the PK equations as bounded corrections to mechanistic rates. The correction network is constructed to be internally conservative across compartments and to act as a controlled perturbation of the base model. We provide a simple mathematical analysis showing that this design preserves key qualitative properties, including boundedness and, under suitable conditions, positivity and stability. Numerical experiments on simulated and reported PK datasets show that the constrained hybrid model improves prediction accuracy and yields more stable extrapolation than both purely mechanistic and unconstrained hybrid models, while respecting the underlying physical structure.
Dermatomyositis (DM) is a rare and debilitating inflammatory disease associated with muscle weakness and skin manifestations. As with all rare diseases, clinical trials in DM are challenging due to the small number of patients available for study, and DM patients may even present with skin- or muscle-predominant disease. A Phase 2 study of an anti-interferon-beta monoclonal antibody (dazukibart) in DM was recently completed. Most trial data were collected in skin-predominant DM patients, measuring the Cutaneous Dermatomyositis Disease Area and Severity Index (CDASI); the Total Improvement Score (TIS) was only measured in a small subset of muscle-predominant DM patients. Because TIS is a more holistic endpoint, planning for further dazukibart development depended on a robust understanding of TIS response. This analysis aimed to develop an exposure-response model for dazukibart in DM patients. The model was intended to describe the timecourses of all relevant clinical responses, including CDASI and TIS, using the available data to collectively inform the exposure-response relationships. The model provided evidence for a TIS response in DM patients treated with dazukibart that was consistent with the observed data and supportive of further development. The model used here could be applied directly to model-based meta-analysis of other DM trials, and the general approach can be used in rare diseases with multiple endpoints. Trial registration numbers: NCT03181893, NCT02766621.
The purpose of this study was to elucidate the mechanisms underlying the wide variability in fetal-to-maternal (F/M) concentration ratios of digoxin by comprehensively characterizing maternal and fetal pharmacokinetics using a physiologically based pharmacokinetic (PBPK) model that explicitly incorporates pregnant women, the placenta, and the fetus, with particular emphasis on placental transfer processes and time-dependent concentration dynamics. Maternal–fetal digoxin pharmacokinetics were simulated using a maternal–placental–fetal PBPK model implemented in Simcyp™. Placental transfer was described by separating passive diffusion, informed by human placental perfusion data, and P-gp–mediated active efflux scaled from in vitro data using quantitative proteomics. Gestation-dependent fetal renal excretion and amniotic fluid pathways were incorporated. Model predictions were verified against reported clinical maternal pharmacokinetic data. Global sensitivity analysis and pathway contribution analysis were performed to identify key determinants of fetal exposure. Placental P-gp activity selectively influenced fetal digoxin exposure without affecting maternal pharmacokinetics, and model predictions were consistent with clinical maternal data. Fetal exposure was primarily governed by direct placental transfer, whereas amniotic fluid–mediated pathways contributed only minimally. Although F/M (AUC ratio) remained relatively stable, instantaneous F/M values exhibited marked time-dependent variability. These findings indicate that a single time-point F/M may not adequately reflect fetal exposure for digoxin. Time-resolved PBPK analyses provide a mechanistic framework to complement conventional F/M-based assessments of maternal–fetal drug exposure.
Polyarticular-course juvenile idiopathic arthritis (pcJIA) is a chronic condition that manifests before the age of 16 years, with pathology similar to that of adult rheumatoid arthritis (RA). Sarilumab is an interleukin-6 receptor inhibitor approved for RA and pcJIA. To obtain the approval for pcJIA, a single-arm, multiple-dose phase 2 study was conducted to determine the sarilumab dose for pcJIA. The population pharmacokinetics analysis of the phase 2 dose-finding portion data showed comparable pharmacokinetics; exposure–response analyses demonstrated similar or greater efficacy (JIA-ACR30/50/70), and consistent safety (reduction in absolute neutrophil count) in patients with pcJIA compared with adult patients with RA at similar sarilumab exposure. Based on the adult-to-pediatric extrapolation concept and justifications using modeling and simulation, the phase 2 sample size was increased to generate sufficient efficacy and safety data and evidence at selected doses and the requirement for a randomized controlled study in patients with pcJIA was waived. This novel approach enabled the pcJIA dose proposal by aligning with adult RA exposures, streamlined clinical development by removing the control arm requirement and minimized children participation while ensuring robustness of sarilumab efficacy and safety evidence for approval.
Bacteriophages follow complex replication dynamics in codependence with their host cells. The corresponding dynamics are usually described with three key parameters: the adsorption rate φ , the burst size β and the latent period τ . Single-cell experiments have shown cellular variability in latent period and burst size. Yet, those parameters are traditionally modelled as uniform values for the whole bacterial population, which may introduce bias in the predicted profiles. Here, we systematically assessed the relevance of modelling this cell-level variability in the context of bacteriophage dynamics. We developed a comprehensive modelling framework, aiming to: (i) identify which parameter distributions impact population dynamics, (ii) quantify how difference in distribution can influence predictions, and (iii) inform model selection by comparing relevant modelling approaches. A distributed delay differential equation model with randomly distributed latent period, burst size and adsorption rate within the cell population was developed. It also included logistic growth of the host cell, saturable adsorption rate and intracellular replication dynamics. This model was compared to classical models, both fixed delay differential equation and transit compartments models, over typical ranges of phage parameters. Only the distribution of the latent period and its influence on burst size impacted population dynamics. The two classical models proved to be good approximations of dynamics, but only at low cellular variability. As such, although the classical approaches can be preferred at lower variability, this distributed framework is warranted when high variability is expected, or when the other approaches fail to accurately capture the dynamics.
