Population Pharmacokinetic/Pharmacodynamic Modeling of Therapeutic Enzymes in Lysosomal Storage Diseases Authors: Ina Barzel, Ans T. van der Ploeg, W. W. M. Pim Pijnappel, P. Hugo M. van der Kuy, Johanna M. P. van den Hout, Tim Preijers Published: 2026 ABSTRACT BackgroundLysosomal storage diseases (LSDs) are rare genetic disorders treated with enzyme replacement therapy (ERT). However, treatment outcomes are highly variable, reflecting the complexities of therapeutic enzyme pharmacology, patient heterogeneity and therapy response. Population pharmacokinetic/pharmacodynamic (popPK/PD) modeling can help characterize this variability, identify covariates, and optimize dosing strategies.ObjectivesThis review aimed to provide a comprehensive overview of published popPK and popPD models of therapeutic enzymes in LSDs, summarize modeling strategies and study characteristics, and evaluate the quality of the available models.MethodsA systematic search of Medline, Embase, and Web of Science (inception-March 2025) identified studies reporting popPK and/or popPD models of therapeutic enzymes in patients with LSDs. Data on study characteristics, structural and statistical model choices, covariate analyses, and evaluation methods were extracted and compared.ResultsThis review included studies describing 6 popPK models and 3 popPD models of therapeutic enzymes in LSD. All models were developed using the nonlinear mixed-effects (NLME) modeling approach. The PK characteristics were adequately described using a two-compartment model in three studies, a three-compartment model in two studies, and a one-compartment model in one study. Three studies additionally assessed PK parameters in monocytes, leukocytes, and cerebrospinal fluid (CSF). Among all tested covariates, total body weight (TBW) was identified as a significant predictor for clearance (CL) and volume of distribution in the central compartment (Vc) in three studies. In one study evaluating PK parameters in CFS, age-based scaling was applied instead of weight-based allometric scaling to improve model fit. PopPD models were limited, with exposure-response relationships described either by an inhibitory maximal effect (Imax) model or by a longitudinal logistic regression model with a first-order Markov element.ConclusionsPopPK modeling of therapeutic enzymes in LSDs is relatively well established. However, progress in popPD modeling remains limited. Existing models support the use of indirect response and maximum effect (Emax) models to describe the delayed and saturable effects of ERT, and innovative approaches such as intracellular PK assessment and Markov modeling illustrate the potential of advanced pharmacometric methods. Nevertheless, there remains a need to further clarify the role of drug concentration measurements in target cells, to characterize CNS distribution in LSDs affecting the brain following intrathecal administration of ERTs and intravenous administration of a novel fusion protein, and to identify robust PD biomarkers in defining exposure-response relationships of therapeutic enzymes in LSDs.Supplementary InformationThe online version contains supplementary material available at 10.1007/s40262-026-01636-2. Key Points Population pharmacokinetic (popPK) modeling of therapeutic enzymes in lysosomal storage diseases (LSDs) is relatively well established despite the rarity of these disorders, with one-, two-, or three-compartment models describing the PK characteristics using the nonlinear mixed-effects (NLME) modeling approach and identifying total body weight (TBW) as a key covariate.In contrast, population pharmacodynamic (popPD) modeling remains sparse, although existing studies demonstrate the suitability of indirect response, Emax, and Markov models in capturing delayed and saturable treatment effects of enzyme replacement therapy (ERT).Overall, further research is needed to define the relevance of intracellular drug concentration measurements, characterizing central nervous system distribution in LSDs affecting the brain following intrathecal administration of ERTs and intravenous administration of a novel fusion protein, and robust PD biomarkers are warranted to better define exposure-response relationships and support individualized dosing strategies. Population pharmacokinetic (popPK) modeling of therapeutic enzymes in lysosomal storage diseases (LSDs) is relatively well established despite the rarity of these disorders, with one-, two-, or three-compartment models describing the PK characteristics using the nonlinear mixed-effects (NLME) modeling approach and identifying total body weight (TBW) as a key covariate.In contrast, population pharmacodynamic (popPD) modeling remains sparse, although existing studies demonstrate the suitability of indirect response, Emax, and Markov models in capturing delayed and saturable treatment effects of enzyme replacement therapy (ERT).Overall, further research is needed to define the relevance of intracellular drug concentration measurements, characterizing central nervous system distribution in LSDs affecting the brain following intrathecal administration of ERTs and intravenous administration of a novel fusion protein, and robust PD biomarkers are warranted to better define exposure-response relationships and support individualized dosing strategies. Introduction Lysosomal storage diseases (LSDs) are a heterogeneous group of rare genetic disorders caused by a deficiency of specific lysosomal enzymes required for the degradation of substrates such as glycogen, glycosaminoglycans, and glycolipids. The enzymatic impairment results in progressive intracellular accumulation of these substrates in various tissues, leading to organ dysfunction and clinical symptoms ranging from severe intellectual disabilities to cardiac abnormalities, and progressive muscle weakness, as observed in LSDs such as mucopolysaccharidoses (MPS), Pompe disease, and Gaucher disease (GD). The estimated incidence is 1 new case in 5000–10,000 newborns per year. Although LSDs typically occur in infancy and childhood, adult-onset forms also exist [1]. Enzyme replacement therapy (ERT) is the cornerstone of the treatment for LSDs to date [2, 3]. Intravenous (IV) administration of recombinant enzymes restores or augments deficient or lacking enzymatic activity, thereby reducing substrate accumulation and slowing or stabilizing disease progression. Despite the therapeutic benefits, clinical outcomes remain highly variable among patients. This variability is influenced by factors such as age at treatment initiation, the development of high-titer neutralizing anti-drug antibodies (ADAs), and intercurrent illnesses (for example, in Pompe disease, periods of reduced motor activity can accelerate functional decline). It also reflects the inherent complexities of enzyme pharmacology, including challenges in dosing optimization and tissue-specific uptake [4]. Overall, the variability in drug response between patients is a result of the variability in either pharmacokinetic (PK) or pharmacodynamic (PD) processes, or a combination of both. Pharmacokinetic and PD relationships of therapeutic enzymes are more complex than those of small-molecule drugs due to their large molecular size, complex tertiary structures, and folding [5, 6]. Therapeutic enzymes are eliminated primarily through proteolytic degradation or receptor-mediated uptake, while their physicochemical properties limit tissue penetration [7, 8]. A major challenge is their immunogenicity, as the formation of ADAs can accelerate drug clearance, reduce systemic exposure, and inhibit PD effects by preventing target engagement, ultimately leading to loss of efficacy [9, 10]. Most ERTs are given as IV infusions once weekly or every other week, with dosing based on total body weight (TBW). These regimens were established empirically through animal studies and pivotal trials, based on the rationale that weekly or biweekly infusions provide adequate tissue uptake and substrate clearance. However, pharmacological evidence supporting the clinical efficacy of current regimens remains limited, and blood pharmacokinetics provide little insight into enzyme uptake and sustained correction within target tissues and lysosomes. Consequently, optimal LSD management and ERT personalization continue to be explored. Population pharmacokinetic/pharmacodynamic (popPK/PD) modeling has emerged as an indispensable tool for characterizing variability in drug exposure and response [11, 12]. By integrating data from multiple individuals, popPK/PD models quantify inter- and intraindividual variability, identify covariate relationships (e.g., age, body weight, occurrence of ADAs) that may explain the variability of PK and PD parameters, and enable simulations of alternative dosing strategies [13, 14]. This population approach is particularly valuable in LSDs, where small sample sizes, sparse data, disease heterogeneity, and the potential enrollment of pediatric patients limit the feasibility of large clinical trials [15, 16]. Moreover, regulatory agencies like the European Medicines Agency (EMA) and the US Food and Drug Administration (FDA) acknowledge the use of popPK/PD modeling to guide dose selection, inform labeling, and support dosing strategies in special populations, including pediatrics and patients with hepatic impairment [17–19]. Although several popPK and popPD analyses of therapeutic enzymes in LSDs have been published, existing models vary in modeling strategies, covariate identification, and handling of immunogenicity. To date, no review has summarized and compared these approaches. Therefore, this systematic review aims to provide a comprehensive overview of all published popPK and popPD models for therapeutic enzymes in LSDs, discussing their similarities and differences. Furthermore, methods used to construct these popPK and popPD models, patient characteristics, identified covariate relationships, and the overall quality of the included studies are discussed in detail. Methods Search Strategy Studies on popPK and/or popPD modeling of therapeutic enzymes were systematically searched in the Medline, Embase, and Web of Science databases from inception to 31 March 2025 by a biomedical information specialist at the Medical Library, Erasmus University Medical Center, Rotterdam. The search queries applied are listed in the Supplementary Material. Data Collection Studies were included in the review if they met the following criteria: (1) human subjects were included, (2) therapeutic enzymes were used as the study drug, (3) the indication for the study drug was treatment of LSDs, and (4) a popPK and/or popPD model or PD analysis (exposure-response analysis) of the study drug was provided. Studies were excluded from this review if they: (1) were conducted in animals, (2) were published in languages other than English, Dutch, or German, (3) used non-compartmental analysis (NCA) to estimate PK parameters, (4) were review articles, and (5) were not available as full-text articles. Initial screening of title and abstract was performed independently by two investigators (TP and IB). Full-text screening of the selected publications was then conducted to ensure compliance with the inclusion criteria. From the eligible full-text articles, the following data were extracted: population characteristics, study drug, dosage regimen, sampling design, popPK model characteristics (e.g., modeling software, structural models, estimation method, statistical models, PK parameter estimates), covariate building processes (e.g., tested and included covariates, covariate selection method), and popPD model characteristics (e.g., modeling software, class of structural PD model, PD parameter estimates). Any discrepancies were resolved through discussion between the two investigators. Results Figure 1 depicts the study selection process, which was based on the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines [20]. A total of 456, 664, and 518 publications were identified from Medline, Embase, and Web of Science, respectively. After removal of duplicates, the titles and abstracts of 797 studies were screened, with 85 studies being considered potentially eligible. Ultimately, a total of 29 publications describing a popPK, popPD or popPK/PD analysis of therapeutic enzymes were selected. Two of these reported popPK results from the same study. The original article was retained for analysis of the popPK model, while the regulatory note was included for the exposure-response analysis, as this was not reported in the original publication. Of the remaining publications, seven presented popPK, popPD or popPK/PD models for therapeutic enzymes in LSDs and were selected for the present review.Fig. 1Flow chart of the study selection process (flow chart adapted from PRISMA). LSD lysosomal storage disease, NCA non-compartmental analysis, popPK/PD population pharmacokinetics/pharmacodynamics, PRISMA Preferred Reporting Items for Systematic Reviews and Meta-Analyses Flow chart of the study selection process (flow chart adapted from PRISMA). LSD lysosomal storage disease, NCA non-compartmental analysis, popPK/PD population pharmacokinetics/pharmacodynamics, PRISMA Preferred Reporting Items for Systematic Reviews and Meta-Analyses Study Characteristics A summary of the population characteristics, therapeutic enzymes studied, dosage regimen, and sampling designs of the included studies, which were published between 2018 and 2023, is presented in Table 1. As is common in rare diseases, the number of subjects included in the studies was small (< 100) and patient data were derived from Phase I/II and Phase III trials for the majority of the studies [21–24, 26]. Most of the studies employed an extensive blood sampling method for PK analysis with at least 10 plasma samples per patient [21–25], while one study used a sparse sampling technique for cerebrospinal fluid (CSF) data and rich sampling technique for plasma data [26]. Pre- and post-infusion blood samples were collected at multiple scheduled time points. The number of samples obtained per subject ranged from 8 to 31. Three studies additionally assessed PK parameters in monocytes [24], leukocytes [25], and CSF [26]. For a schematic overview of these popPK models, see supplemental Figure S1. Table 1General characteristics of the studies includedStudy, year [References]DrugIndicationDosage regimenRegistration (NCT) number of included clinical trialsPopulation characteristicsSubjects (N)Sex, female (%)Age, years (range)Body weight, kg (range) Qi et al., 2018 [23] Vestronidase alfaMPS type VII1, 2, 4 mg/kg IV EOW01856218, 02418455, 0223056623521.7–25.39.41–104.0 Berger et al., 2018 [24] ImigluceraseGD type 115–120 U/kg IV EOW, E3W or E4W019519893164.34–83NAGras-Colomer et al., 2021 [25]Imiglucerase/VelagluceraseGD type 111.0–67.9 U/kg IV EOWNA254420–7138.0–107.0Tuffal et al., 2023 [21]Avalglucosidase alfaLOPD5, 10, 20 mg/kg IV EOW01898364, 02032524, 02782741754846.0 (mean) ± 15.1 (SD)75.9 (mean) ± 20.1 (SD)Tiraboschi et al., 2023 [22]Avalglucosidase alfaLOPD/IOPD5, 10, 20, 40 mg/kg IV EOW01898364, 02032524, 02782741, 030194069146.2LOPD: 16.5–78.4IOPD: 1–11LOPD: 38–129IOPD: 9.9–63.5Troy et al., 2020 [26]Arylsulfatase A (TAK‐611)MLD10, 30 and 100 mg IT (2 different formulations of 100 mg) EOW0151002824NA19–107 months10.5–24.801856218: Phase I/II, open-label study to assess the safety, efficacy, and optimal dose of vestronidase alfa in subjects aged between 5 and 30 years with MPS VII02418455: Phase II, open-label study to assess safety, efficacy, and PK/PD in subjects aged < 5 years with MPS VII02230566: Phase III, randomized, placebo-controlled, blind-start (i.e., blinded crossover from placebo to active treatment) study to assess the efficacy and safety of vestronidase alfa in subjects aged between 5 and 35 years with MPS VII01951989: Phase II, open-label study to evaluate the pharmacokinetics of imiglucerase activity into target cellular compartment depending on dose and frequency of infusion01898364: Phase I, open-label study in treatment naïve and alglucosidase alfa treated LOPD patients02032524: Phase I/II, open-label study in LOPD patients02782741: Phase III, randomized, double-blind study in treatment naïve LOPD patients03019406: Phase II, randomized, open-label study in alglucosidase alfa treated IOPD patients01510028: Phase I/II, multicenter, open-label, dose-escalation studyE3W every 3 weeks, E4W every 4 weeks, EOW every other week, GD Gaucher disease, IOPD infantile-onset Pompe disease, IT intrathecal, IV intravenous, LOPD late-onset Pompe disease, MLD metachromatic leukodystrophy, MPS mucopolysaccharidoses, NA not available, NCT National Clinical Trial, SD standard deviation Table 1General characteristics of the studies includedStudy, year [References]DrugIndicationDosage regimenRegistration (NCT) number of included clinical trialsPopulation characteristicsSubjects (N)Sex, female (%)Age, years (range)Body weight, kg (range) Qi et al., 2018 [23] Vestronidase alfaMPS type VII1, 2, 4 mg/kg IV EOW01856218, 02418455, 0223056623521.7–25.39.41–104.0 Berger et al., 2018 [24] ImigluceraseGD type 115–120 U/kg IV EOW, E3W or E4W019519893164.34–83NAGras-Colomer et al., 2021 [25]Imiglucerase/VelagluceraseGD type 111.0–67.9 U/kg IV EOWNA254420–7138.0–107.0Tuffal et al., 2023 [21]Avalglucosidase alfaLOPD5, 10, 20 mg/kg IV EOW01898364, 02032524, 02782741754846.0 (mean) ± 15.1 (SD)75.9 (mean) ± 20.1 (SD)Tiraboschi et al., 2023 [22]Avalglucosidase alfaLOPD/IOPD5, 10, 20, 40 mg/kg IV EOW01898364, 02032524, 02782741, 030194069146.2LOPD: 16.5–78.4IOPD: 1–11LOPD: 38–129IOPD: 9.9–63.5Troy et al., 2020 [26]Arylsulfatase A (TAK‐611)MLD10, 30 and 100 mg IT (2 different formulations of 100 mg) EOW0151002824NA19–107 months10.5–24.801856218: Phase I/II, open-label study to assess the safety, efficacy, and optimal dose of vestronidase alfa in subjects aged between 5 and 30 years with MPS VII02418455: Phase II, open-label study to assess safety, efficacy, and PK/PD in subjects aged < 5 years with MPS VII02230566: Phase III, randomized, placebo-controlled, blind-start (i.e., blinded crossover from placebo to active treatment) study to assess the efficacy and safety of vestronidase alfa in subjects aged between 5 and 35 years with MPS VII01951989: Phase II, open-label study to evaluate the pharmacokinetics of imiglucerase activity into target cellular compartment depending on dose and frequency of infusion01898364: Phase I, open-label study in treatment naïve and alglucosidase alfa treated LOPD patients02032524: Phase I/II, open-label study in LOPD patients02782741: Phase III, randomized, double-blind study in treatment naïve LOPD patients03019406: Phase II, randomized, open-label study in alglucosidase alfa treated IOPD patients01510028: Phase I/II, multicenter, open-label, dose-escalation studyE3W every 3 weeks, E4W every 4 weeks, EOW every other week, GD Gaucher disease, IOPD infantile-onset Pompe disease, IT intrathecal, IV intravenous, LOPD late-onset Pompe disease, MLD metachromatic leukodystrophy, MPS mucopolysaccharidoses, NA not available, NCT National Clinical Trial, SD standard deviation General characteristics of the studies included LOPD: 16.5–78.4 IOPD: 1–11 LOPD: 38–129 IOPD: 9.9–63.5 01856218: Phase I/II, open-label study to assess the safety, efficacy, and optimal dose of vestronidase alfa in subjects aged between 5 and 30 years with MPS VII 02418455: Phase II, open-label study to assess safety, efficacy, and PK/PD in subjects aged < 5 years with MPS VII 02230566: Phase III, randomized, placebo-controlled, blind-start (i.e., blinded crossover from placebo to active treatment) study to assess the efficacy and safety of vestronidase alfa in subjects aged between 5 and 35 years with MPS VII 01951989: Phase II, open-label study to evaluate the pharmacokinetics of imiglucerase activity into target cellular compartment depending on dose and frequency of infusion 01898364: Phase I, open-label study in treatment naïve and alglucosidase alfa treated LOPD patients 02032524: Phase I/II, open-label study in LOPD patients 02782741: Phase III, randomized, double-blind study in treatment naïve LOPD patients 03019406: Phase II, randomized, open-label study in alglucosidase alfa treated IOPD patients 01510028: Phase I/II, multicenter, open-label, dose-escalation study E3W every 3 weeks, E4W every 4 weeks, EOW every other week, GD Gaucher disease, IOPD infantile-onset Pompe disease, IT intrathecal, IV intravenous, LOPD late-onset Pompe disease, MLD metachromatic leukodystrophy, MPS mucopolysaccharidoses, NA not available, NCT National Clinical Trial, SD standard deviation Population Pharmacokinetic Analyses A summary of the study characteristics for the included popPK models, covariate building processes, and the estimated PK model parameters is presented in Tables 1, 2 and 3, respectively. Table 2General description of population pharmacokinetic analyses, model parameters, and covariate relationshipsStudy, year [References]Sampling densitySample matrixSoftwareModelEstimation methodCovariate relationshipsTestedSelection method for statistical significance (p-value)Included in the final model (estimate, [95% CI])Relationship with PK parameterQi et al., 2018 [23]Rich (n = 15)Sparse (n = 8)PlasmaNONMEMa2-compartmentSAEMAge, body weight, sex, ethnicity, Nab, ADA, laboratory test resultsp = 0.001, i.e., its deletion was to increase the OFV by < 10.83 pointsTBW on CL and Q (0.587, [0.444; 0.730]), Vc and Vp (0.483, [0315; 0.651])CL = 1.97 × (TBW/20)0.587Q = 0.931 × (TBW/20)0.587Vc =1.52 × (TBW/20)0.483Vp = 3.11 × (TBW/20)0.483 Berger et al., 2018 [24] RichIntra-monocyteMONOLIX v4.32-compartmentSAEMNAGras-Colomer et al., 2021 [25]RichPlasma/leukocyteNONMEM v7.32-compartmentFOCE+ISex, race, Zdiag, age, body weight, and ActDNoneTuffal et al., 2023 [21]RichPlasmaNONMEM v7.4.13-compartmentFOCE+IBody weight, age, sex, ethnicity, renal function, CK, ALP, ALT, AST, bilirubin, albumin, occurrence of ADA formationΔOFV = 3.84, p < 0.05 (forward inclusion)ΔOFV = 10.8, p < 0.001 (backward elimination)NoneTiraboschi et al., 2023 [22]RichPlasmaNONMEM v7.4.13-compartmentFOCE+IAge, sex, body weight, albumin, ALP, ALT, AST, total bilirubin, CK, normalized creatinine clearance, occurrence of ADAs, the disease type (LOPD vs IOPD), previous treatment with alglucosidase alfaForward inclusion (alpha risk: 5%) and backward elimination (alpha risk: 0.1%)Time-varying TBW on CL (0.896, [0.754; 1.04]), V1 (0.661, [0.578; 0.7444]) and Vm (0.463, [0.352;0.574])CL = 0.808 × (TBW/70.5)0.896V1 = 3.37 × (TBW/70.5)0.661Vm = 12 × (TBW/70.5)0.463Troy et al., 2020 [26]Sparse: CSF dataRich: serum dataSerum/CSFNONMEM v7.3CNS: 2-compartmentSerum: 1-compartmentFOCE+IAge, continuous body weight, drug-manufacturing process, ADA titer, and presence of ADAs in serum and CSFAge, TBWCL = 1.69 × (TBW/15)0.75Vc = 0.183 × (TBW/15)ActD endogenous deficient glucocerebrosidase measured at diagnosis, ADA anti-drug antibody, ALP alkaline phosphatase, ALT alanine aminotransferase, AST aspartate aminotransferase, CI confidence interval, CK creatine kinase, CL clearance, CNS central nervous system, CSF cerebrospinal fluid, FOCE first-order estimation with interaction, IOPD infantile-onset Pompe disease, LOPD late-onset Pompe disease, NA not available, Nab Neutralizing antibodies, OFV objective function value, PK pharmacokinetic, Q intercompartmental clearance, SAEM stochastic approximation of the expectation-maximization algorithm, TBW total body weight, Vc/V1 volume of distribution in the central compartment, Vm nonlinear clearance, Vp volume of distribution in the peripheral compartment, Zdiag Zimran score indicating disease severityaVersion is not reportedTable 3Description of fixed- and random-effect model parameters for the population PK modelsCategoryParameterQi et al., 2018 [23]PlasmaBerger et al., 2018 [24]Intra-monocyteGras-Colomer et al., 2021 [25]Plasma/leukocyteθTuffal et al., 2023 [21]PlasmaTiraboschi et al., 2023 [22]PlasmaTroy et al., 2020 [26]Serum/CSF*Fixed effect parametersCL (L/h)/CL1 (L/d)1.973.85 × 10−20.8690.8083.01Vc/V1 (L)1.521.10 × 10−23.353.3769.1Q/Q2 (L/h)0.9310.2540.254Vp/V2 (L)3.11296296Cendo (103 nmol/h/mg)0.93V (103 U/kg/nmol × h × mg)4.46k (d−1)0.9k12 (d−1)0.94k21 (d−1)0.110.02060.0157Vm (mg/h)9.8212Km (µg/mL)0.4510.541Q3 (L/h)1.871.87V3 (L)1.311.31Qpc (L/h)0.02060.0157CL2 (L/d)1.25θV2 (L)2.57 × 10−1θBaseline plasma GCase (U/L)277∆CL1 (L/d)− 2.19 × 10−1kT (d−1)6.14 × 10−1VCSF (L)0.183*VCNS (L)1.69*QCNS (L/h)0.00280*Ktrans (h−1)0.581Inter-individual variability (CV%)CL / CL143.07130.230.80.0102VcV130.712.713.640.3Q86.9VpV215.4Cendo55.6V33.4k17.1k1265k21153Vm36.635.4Km43.152.4Qpc234156CL293θVCSF63.9*VCNS65.2*QCNS24.6*Ktrans42.0Residual unexplained variabilityAdditional errorProportional error (CV%)33.026Plasma: 29Leukocyte: 1135.334.6CSF: 98.1Serum: 67.5ΔCL1 maximal change from CL1 at time 0, Cendo endogenous glucocerebrosidase (GCase) activity, CL clearance, CL1 GCase plasma clearance at steady-state [25], CL2 GCase leukocyte clearance, CV coefficient of variation, k elimination rate constant from the central compartment, k12 and k21 transfer rate constants between compartments, Ktrans transit rate constant, Q and Q2 2-way inter-compartmental clearance between Vc and Vp, or V1 and V2, respectively, Q3 1-way inter-compartmental clearance between V2 and V3, Qpc low inter-compartmental clearance from V3 to V1, kT first-order rate constant, QCNS 2-way inter-compartmental clearance between VCNS and VCSF, V volume of distribution [24], Vc and V1 volume of distribution in the central compartment, V2 apparent volume of distribution in leukocyte [25], Vp, V2 and V3 volume of distribution in the peripheral compartments, VCNS volume of distribution in the central nervous system, VCSF volume of distribution in the cerebrospinal fluid, Vm and Km Michaelis–Menten parameters describing nonlinear clearance Table 2General description of population pharmacokinetic analyses, model parameters, and covariate relationshipsStudy, year [References]Sampling densitySample matrixSoftwareModelEstimation methodCovariate relationshipsTestedSelection method for statistical significance (p-value)Included in the final model (estimate, [95% CI])Relationship with PK parameterQi et al., 2018 [23]Rich (n = 15)Sparse (n = 8)PlasmaNONMEMa2-compartmentSAEMAge, body weight, sex, ethnicity, Nab, ADA, laboratory test resultsp = 0.001, i.e., its deletion was to increase the OFV by < 10.83 pointsTBW on CL and Q (0.587, [0.444; 0.730]), Vc and Vp (0.483, [0315; 0.651])CL = 1.97 × (TBW/20)0.587Q = 0.931 × (TBW/20)0.587Vc =1.52 × (TBW/20)0.483Vp = 3.11 × (TBW/20)0.483 Berger et al., 2018 [24] RichIntra-monocyteMONOLIX v4.32-compartmentSAEMNAGras-Colomer et al., 2021 [25]RichPlasma/leukocyteNONMEM v7.32-compartmentFOCE+ISex, race, Zdiag, age, body weight, and ActDNoneTuffal et al., 2023 [21]RichPlasmaNONMEM v7.4.13-compartmentFOCE+IBody weight, age, sex, ethnicity, renal function, CK, ALP, ALT, AST, bilirubin, albumin, occurrence of ADA formationΔOFV = 3.84, p < 0.05 (forward inclusion)ΔOFV = 10.8, p < 0.001 (backward elimination)NoneTiraboschi et al., 2023 [22]RichPlasmaNONMEM v7.4.13-compartmentFOCE+IAge, sex, body weight, albumin, ALP, ALT, AST, total bilirubin, CK, normalized creatinine clearance, occurrence of ADAs, the disease type (LOPD vs IOPD), previous treatment with alglucosidase alfaForward inclusion (alpha risk: 5%) and backward elimination (alpha risk: 0.1%)Time-varying TBW on CL (0.896, [0.754; 1.04]), V1 (0.661, [0.578; 0.7444]) and Vm (0.463, [0.352;0.574])CL = 0.808 × (TBW/70.5)0.896V1 = 3.37 × (TBW/70.5)0.661Vm = 12 × (TBW/70.5)0.463Troy et al., 2020 [26]Sparse: CSF dataRich: serum dataSerum/CSFNONMEM v7.3CNS: 2-compartmentSerum: 1-compartmentFOCE+IAge, continuous body weight, drug-manufacturing process, ADA titer, and presence of ADAs in serum and CSFAge, TBWCL = 1.69 × (TBW/15)0.75Vc = 0.183 × (TBW/15)ActD endogenous deficient glucocerebrosidase measured at diagnosis, ADA anti-drug antibody, ALP alkaline phosphatase, ALT alanine aminotransferase, AST aspartate aminotransferase, CI confidence interval, CK creatine kinase, CL clearance, CNS central nervous system, CSF cerebrospinal fluid, FOCE first-order estimation with interaction, IOPD infantile-onset Pompe disease, LOPD late-onset Pompe disease, NA not available, Nab Neutralizing antibodies, OFV objective function value, PK pharmacokinetic, Q intercompartmental clearance, SAEM stochastic approximation of the expectation-maximization algorithm, TBW total body weight, Vc/V1 volume of distribution in the central compartment, Vm nonlinear clearance, Vp volume of distribution in the peripheral compartment, Zdiag Zimran score indicating disease severityaVersion is not reported General description of population pharmacokinetic analyses, model parameters, and covariate relationships Rich (n = 15) Sparse (n = 8) CL = 1.97 × (TBW/20)0.587 Q = 0.931 × (TBW/20)0.587 Vc =1.52 × (TBW/20)0.483 Vp = 3.11 × (TBW/20)0.483 ΔOFV = 3.84, p < 0.05 (forward inclusion) ΔOFV = 10.8, p < 0.001 (backward elimination) CL = 0.808 × (TBW/70.5)0.896 V1 = 3.37 × (TBW/70.5)0.661 Vm = 12 × (TBW/70.5)0.463 Sparse: CSF data Rich: serum data CNS: 2-compartment Serum: 1-compartment CL = 1.69 × (TBW/15)0.75 Vc = 0.183 × (TBW/15) ActD endogenous deficient glucocerebrosidase measured at diagnosis, ADA anti-drug antibody, ALP alkaline phosphatase, ALT alanine aminotransferase, AST aspartate aminotransferase, CI confidence interval, CK creatine kinase, CL clearance, CNS central nervous system, CSF cerebrospinal fluid, FOCE first-order estimation with interaction, IOPD infantile-onset Pompe disease, LOPD late-onset Pompe disease, NA not available, Nab Neutralizing antibodies, OFV objective function value, PK pharmacokinetic, Q intercompartmental clearance, SAEM stochastic approximation of the expectation-maximization algorithm, TBW total body weight, Vc/V1 volume of distribution in the central compartment, Vm nonlinear clearance, Vp volume of distribution in the peripheral compartment, Zdiag Zimran score indicating disease severity aVersion is not reported Table 3Description of fixed- and random-effect model parameters for the population PK modelsCategoryParameterQi et al., 2018 [23]PlasmaBerger et al., 2018 [24]Intra-monocyteGras-Colomer et al., 2021 [25]Plasma/leukocyteθTuffal et al., 2023 [21]PlasmaTiraboschi et al., 2023 [22]PlasmaTroy et al., 2020 [26]Serum/CSF*Fixed effect parametersCL (L/h)/CL1 (L/d)1.973.85 × 10−20.8690.8083.01Vc/V1 (L)1.521.10 × 10−23.353.3769.1Q/Q2 (L/h)0.9310.2540.254Vp/V2 (L)3.11296296Cendo (103 nmol/h/mg)0.93V (103 U/kg/nmol × h × mg)4.46k (d−1)0.9k12 (d−1)0.94k21 (d−1)0.110.02060.0157Vm (mg/h)9.8212Km (µg/mL)0.4510.541Q3 (L/h)1.871.87V3 (L)1.311.31Qpc (L/h)0.02060.0157CL2 (L/d)1.25θV2 (L)2.57 × 10−1θBaseline plasma GCase (U/L)277∆CL1 (L/d)− 2.19 × 10−1kT (d−1)6.14 × 10−1VCSF (L)0.183*VCNS (L)1.69*QCNS (L/h)0.00280*Ktrans (h−1)0.581Inter-individual variability (CV%)CL / CL143.07130.230.80.0102VcV130.712.713.640.3Q86.9VpV215.4Cendo55.6V33.4k17.1k1265k21153Vm36.635.4Km43.152.4Qpc234156CL293θVCSF63.9*VCNS65.2*QCNS24.6*Ktrans42.0Residual unexplained variabilityAdditional errorProportional error (CV%)33.026Plasma: 29Leukocyte: 1135.334.6CSF: 98.1Serum: 67.5ΔCL1 maximal change from CL1 at time 0, Cendo endogenous glucocerebrosidase (GCase) activity, CL clearance, CL1 GCase plasma clearance at steady-state [25], CL2 GCase leukocyte clearance, CV coefficient of variation, k elimination rate constant from the central compartment, k12 and k21 transfer rate constants between compartments, Ktrans transit rate constant, Q and Q2 2-way inter-compartmental clearance between Vc and Vp, or V1 and V2, respectively, Q3 1-way inter-compartmental clearance between V2 and V3, Qpc low inter-compartmental clearance from V3 to V1, kT first-order rate constant, QCNS 2-way inter-compartmental clearance between VCNS and VCSF, V volume of distribution [24], Vc and V1 volume of distribution in the central compartment, V2 apparent volume of distribution in leukocyte [25], Vp, V2 and V3 volume of distribution in the peripheral compartments, VCNS volume of distribution in the central nervous system, VCSF volume of distribution in the cerebrospinal fluid, Vm and Km Michaelis–Menten parameters describing nonlinear clearance Description of fixed- and random-effect model parameters for the population PK models Plasma: 29 Leukocyte: 11 CSF: 98.1 Serum: 67.5 ΔCL1 maximal change from CL1 at time 0, Cendo endogenous glucocerebrosidase (GCase) activity, CL clearance, CL1 GCase plasma clearance at steady-state [25], CL2 GCase leukocyte clearance, CV coefficient of variation, k elimination rate constant from the central compartment, k12 and k21 transfer rate constants between compartments, Ktrans transit rate constant, Q and Q2 2-way inter-compartmental clearance between Vc and Vp, or V1 and V2, respectively, Q3 1-way inter-compartmental clearance between V2 and V3, Qpc low inter-compartmental clearance from V3 to V1, kT first-order rate constant, QCNS 2-way inter-compartmental clearance between VCNS and VCSF, V volume of distribution [24], Vc and V1 volume of distribution in the central compartment, V2 apparent volume of distribution in leukocyte [25], Vp, V2 and V3 volume of distribution in the peripheral compartments, VCNS volume of distribution in the central nervous system, VCSF volume of distribution in the cerebrospinal fluid, Vm and Km Michaelis–Menten parameters describing nonlinear clearance All popPK models were developed using the nonlinear mixed-effects (NLME) modeling approach. This method estimates the typical values of PK parameters for the whole population (fixed effects), while simultaneously quantifying the inter-individual variability (IIV) in PK parameters (random effects). The residual variability captures the remaining unexplained differences between observed and predicted drug concentrations. The most applied software was NONMEM and first-order estimation with interaction (FOCE+I) was applied in all cases for the estimation of popPK parameters [27]. In one case, MONOLIX was used, which uses the stochastic approximation of the expectation-maximization algorithm (SAEM) as a method for estimating the population parameters [28]. The aims of the individual studies varied depending on the therapeutic enzyme and patient population investigated. One study sought to justify dosage by linking enzyme activity thresholds of imiglucerase to disease severity of GD type 1 and establishing criteria for the initiation of ERT [24]. Another focused on evaluating alternative dosing regimens of avalglucosidase alfa in pediatric patients with late-onset Pompe disease (LOPD). The extension study by Tiraboschi et al. examined avalglucosidase alfa dosing in pediatric LOPD, using the structural model of Tuffal et al. as a starting point [21, 22]. The popPK dataset from adolescents and adults with LOPD was supplemented with pediatric infantile-onset Pompe disease (IOPD) data. Subsequently, PK simulations comparing exposures across different age and body weight groups were conducted to propose alternative dosing regimens based on a body weight cut-off [22]. The PK characteristics of therapeutic enzymes in LSDs were adequately described using a one-, two-, or three-compartment model. Three studies described the PK of the therapeutic enzyme as a two-compartment model [23–25], while one study fitted a one-compartment model [26]. In the latter study, CSF levels of arylsulfatase A in patients with metachromatic leukodystrophy (MLD) were well described by a two-compartment central nervous system (CNS) model (CFS and putative brain tissue) linked to plasma via a transit compartment. The model indicated rapid distribution from CSF into brain tissue (distributive half-life: 1 h) and slow return or persistence in CNS (terminal half-life: 477 h). Two studies used a three-compartment model to describe the PK of avalglucosidase alfa, a model with a central compartment with parallel linear and nonlinear elimination and two concatenated peripheral compartments with redistribution from the deep peripheral compartment to the central compartment [21, 22]. Graphical inspection of the concentration-time curves indicated the presence of two kinetic phases. The first phase, which accounted for ~ 99% of overall exposure, was adequately described by a linear clearance pathway and a saturable Michaelis–Menten clearance, characterized by the parameters Vm and Km. Notably, Michaelis–Menten clearance became effective during the initial infusion sequence, before the maximum plasma concentration (Cmax) was reached, reflecting the saturable nature of enzyme elimination at higher concentrations. The second phase contributed only marginally to total exposure, showing a flat profile that peaked around 72 h at low concentrations and persisted up to 336 h in two-thirds of the patients. The greatest variability was associated with inter-compartmental clearance (Qpc) from the second peripheral compartment to the central compartment (coefficient of variation [CV]: 234% and 154%, respectively), reflecting the difficulty of characterizing the deep compartment redistribution. Saturable plasma clearance, and consequently saturable cellular uptake, were also observed in model-based simulations of vestronidase alfa conducted to inform selection of the optimal dosing regimen for pediatric and adult patients with MPS type VII (MPS VII) [23]. These simulations demonstrated that a lower dose administered more frequently (4 mg/kg every 2 weeks vs 8 mg/kg every 4 weeks) maintains a longer duration of moderate plasma exposure above Kuptake (in vitro determined vestronidase alfa concentration corresponding to 50% maximum rate of cellular uptake). In addition, this regimen avoids high peak concentrations that can lead to saturation of cellular uptake. Two studies reported markedly low volumes of distribution in the central compartment (Vc) values, with 1.52 L in the study by Qi et al. [23], and 1.10 × 10−2 L (plasma) and 2.57 10 × 10−1 (leukocytes) in the study by Gras-Colomer et al. [25]. In three studies, TBW was the only morphometric variable evaluated and was found to impact the plasma PK of avalglucosidase alfa, vestronidase alfa, and arylsulfatase A in LSDs, respectively [22, 23, 26]. In the latter study evaluating PK parameters of arylsulfatase A in CFS, age-based scaling was applied instead of weight-based allometric scaling to improve model fit [26]. This approach was considered more appropriate as brain development follows a different trajectory than body weight. Two recent studies applied popPK modeling for therapeutic enzymes in GD1. Gras-Colomer et al. simultaneously characterized glucocerebrosidase (GCase) activity in plasma and leukocytes using a linked two-compartment model [25]. The analysis identified an exponential time-dependent effect on plasma clearance (CL), with a 2.32-fold increase after treatment initiation, consistent with the estimated negative typical population value for the maximal change from the clearance at steady-state (CL1) at time 0 (∆CL1). High IIV was observed for both CL1 (CV: 71%) and leukocyte clearance (CL2, CV: 93%), although no demographic or clinical covariates were obtained to explain these variabilities. The model-predicted change in clearance over time was calculated as follows:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathrm{CL}={\mathrm{CL}}_{1}+{\Delta \mathrm{CL}}_{1}\times {\mathrm{e}}^{\left(-{k}_{T }\times \mathrm{time}\right)}$$\end{document}CL=CL1+ΔCL1×e-kT×timewith CL being the actual clearance, CL1 the clearance at steady-state, ∆CL1 the maximal change from CL1 at time 0, and kT the first-order rate constant. Complementary insights were provided by Berger et al., who examined intra-monocyte PK of imiglucerase [24]. A two-compartment model adequately described intra-monocyte kinetics, revealing a biphasic decay after infusion and substantial IIV in distribution parameters. Endogenous intra-monocyte GCase activity (Cendo) was strongly associated with GD severity scores and could guide the decision to initiate treatment. Receiver operating characteristic (ROC) analysis confirmed the discriminative power of Cendo (area under the ROC curve [AUROC]: 0.912; 95% CI 0.80–1.00), suggesting the existence of a GCase activity threshold related to GD1 aggressiveness. In treated patients, GCase intra-monocyte activity was modeled as the sum of Cendo and treatment-related activity. Pharmacokinetic parameter estimates revealed large IIV in Cendo (CV: 56%) as well as in inter-compartmental transfer rates (k12: 65%; k21: 153%), with a residual error of 26%. These findings underline the clinical relevance of Cendo as a predictor of disease severity. Population Pharmacodynamic Analyses A summary of the popPD model characteristics and the estimated PD model parameters is presented in Tables 4 and 5. In two studies, popPD models were developed using the NLME modeling approach [23, 25], while one study explored exposure-response relationships using linear regression [29]. Table 4General description of population pharmacodynamic modelsStudy, year [References]DrugSubjects (N)Dosage regimenSoftwarePopulation characteristicsPD parameterPD modelAge, years (range)Body weight, kg (range)Qi et al., 2018 [23]Vestronidase alfa234 mg/kg IV EOWNONMEMaCreatinine-corrected uGAG CS and DSInhibitory maximal effect model1.7–25.39.41–104.0Li et al., 2023 [29]Avalglucosidase alfa5120 mg/kg IV EOWNRPredicted %FVC change from baseline and 6MWT change from baselineExploratory univariate E-R analyses using linear regression methodLOPD: 16.5–78.4IOPD: 1–11LOPD: 38–129IOPD: 9.9–63.5Gras-Colomer et al., 2021 [25]Imiglucerase/Velaglucerase1411.0–67.9 U/kg IV EOWNONMEM v7.3S-MRI scale in bone marrowLongitudinal logistic regression model with a first-order Markov element (discrete-time Markov model)20–7138.0–107.06MWT 6-minute walk test, CS chondroitin sulfate, DS dermatan sulfate, EOW every other week, E-R exposure-response, FVC forced vital capacity, NR not reported, S-MRI Spanish magnetic resonance imaging, uGAG urinary glycosaminoglycansaVersion is not reportedTable 5Description of fixed- and random-effect model parameters for the population pharmacodynamic modelsCategoryParameterQi et al., 2018 [23]Gras-Colomer et al., 2021 [25]Fixed effect parametersIC50 (µg×h/mL)CS: 8.6DS: 7.3Imax (%)CS: 82.0DS: 76.9A10− 4.29A21− 1.24A32− 2.34EC50 (U/L)15.73Emax2.33ActD (%)0.16Inter-individual variability (CV%)ImaxCS: 7.9DS: 8.7EC5092Residual unexplained variabilityAdditional errorCS: 6.1 µg×h/mLDS: 6.4 µg×h/mLActD endogenous deficient glucocerebrosidase measured at diagnosis and expressed as percentage with respect to the healthy individual activity, CV coefficient of variation, EC50 concentration producing 50% of the maximum effect, Emax maximum effect, IC50 drug concentration at which 50% of the maximum inhibitory effect is observed, Imax maximum inhibitory effect, A10, A21 and A32 transition probabilities from efficacy grade 1 to 0, 2 to 1 and 3 to 2, respectively Table 4General description of population pharmacodynamic modelsStudy, year [References]DrugSubjects (N)Dosage regimenSoftwarePopulation characteristicsPD parameterPD modelAge, years (range)Body weight, kg (range)Qi et al., 2018 [23]Vestronidase alfa234 mg/kg IV EOWNONMEMaCreatinine-corrected uGAG CS and DSInhibitory maximal effect model1.7–25.39.41–104.0Li et al., 2023 [29]Avalglucosidase alfa5120 mg/kg IV EOWNRPredicted %FVC change from baseline and 6MWT change from baselineExploratory univariate E-R analyses using linear regression methodLOPD: 16.5–78.4IOPD: 1–11LOPD: 38–129IOPD: 9.9–63.5Gras-Colomer et al., 2021 [25]Imiglucerase/Velaglucerase1411.0–67.9 U/kg IV EOWNONMEM v7.3S-MRI scale in bone marrowLongitudinal logistic regression model with a first-order Markov element (discrete-time Markov model)20–7138.0–107.06MWT 6-minute walk test, CS chondroitin sulfate, DS dermatan sulfate, EOW every other week, E-R exposure-response, FVC forced vital capacity, NR not reported, S-MRI Spanish magnetic resonance imaging, uGAG urinary glycosaminoglycansaVersion is not reported General description of population pharmacodynamic models LOPD: 16.5–78.4 IOPD: 1–11 LOPD: 38–129 IOPD: 9.9–63.5 6MWT 6-minute walk test, CS chondroitin sulfate, DS dermatan sulfate, EOW every other week, E-R exposure-response, FVC forced vital capacity, NR not reported, S-MRI Spanish magnetic resonance imaging, uGAG urinary glycosaminoglycans aVersion is not reported Table 5Description of fixed- and random-effect model parameters for the population pharmacodynamic modelsCategoryParameterQi et al., 2018 [23]Gras-Colomer et al., 2021 [25]Fixed effect parametersIC50 (µg×h/mL)CS: 8.6DS: 7.3Imax (%)CS: 82.0DS: 76.9A10− 4.29A21− 1.24A32− 2.34EC50 (U/L)15.73Emax2.33ActD (%)0.16Inter-individual variability (CV%)ImaxCS: 7.9DS: 8.7EC5092Residual unexplained variabilityAdditional errorCS: 6.1 µg×h/mLDS: 6.4 µg×h/mLActD endogenous deficient glucocerebrosidase measured at diagnosis and expressed as percentage with respect to the healthy individual activity, CV coefficient of variation, EC50 concentration producing 50% of the maximum effect, Emax maximum effect, IC50 drug concentration at which 50% of the maximum inhibitory effect is observed, Imax maximum inhibitory effect, A10, A21 and A32 transition probabilities from efficacy grade 1 to 0, 2 to 1 and 3 to 2, respectively Description of fixed- and random-effect model parameters for the population pharmacodynamic models CS: 8.6 DS: 7.3 CS: 82.0 DS: 76.9 CS: 7.9 DS: 8.7 CS: 6.1 µg×h/mL DS: 6.4 µg×h/mL ActD endogenous deficient glucocerebrosidase measured at diagnosis and expressed as percentage with respect to the healthy individual activity, CV coefficient of variation, EC50 concentration producing 50% of the maximum effect, Emax maximum effect, IC50 drug concentration at which 50% of the maximum inhibitory effect is observed, Imax maximum inhibitory effect, A10, A21 and A32 transition probabilities from efficacy grade 1 to 0, 2 to 1 and 3 to 2, respectively An inhibitory maximal effect (Imax) model was developed for vestronidase alfa and adequately described the exposure-response relationship, with estimated half-maximal inhibitory AUC values (IC50) values of 8.6 µg×h/mL and 7.3 µg×h/mL and maximum reductions (Imax) of 82% and 77% for urinary chondroitin sulfate (uCS) and urinary dermatan sulfate (uDS), respectively, which served as the primary efficacy endpoints [23]. The relationship between the individual vestronidase alfa area under the curve (AUC) values and the corresponding percentage change of PD endpoints from baseline at the same visit was described as follows:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I={I}_{0}- \left(\frac{{I}_{\mathrm{max}} \times \mathrm{AUC}}{\mathrm{AUC} + {IC}_{50}}\right)$$\end{document}I=I0-Imax×AUCAUC+IC50with I being the drug inhibitory effect (percentage reduction in uCS or uDS) at a certain vestronidase alfa serum exposure (AUC0–t), I0 the no-drug response and fixed to 0 in the model, Imax the maximal drug effect in percentage decrease in uCS or uDS, and IC50 the AUC of vestronidase alfa that achieves 50% of the maximum inhibitory drug effect. The exposure-response curve reached a plateau at the 4 mg/kg every other week (EOW) dose, indicating near-maximal efficacy at the recommended clinical regimen. For imiglucerase, the final exposure-efficacy relationship was best described by a longitudinal logistic regression model with a first-order Markov element, which significantly improved the fit compared with an ordered categorical model [25]. In the ordered categorical model, each observation was treated independently, without considering how a patient’s efficacy grade had changed over time. By contrast, the Markov element allowed the model to account for the patient’s previous efficacy grade when predicting the next one, thereby capturing the observed transition patterns between grades (e.g., improvement, worsening, or stability). This resulted in a better description of the number and direction of transitions between efficacy states compared with the ordered categorical approach. An Emax function linked steady-state concentrations of GCase in leukocytes to the probability of transition across the different Spanish magnetic resonance imaging (S-MRI) stages of bone marrow infiltration, an important parameter for evaluating disease severity. The endogenous deficient GCase activity (ActD) measured at diagnosis was found to be a predictive covariate on the transition probability. A discrete-time Markov model allowed proper prediction of the exposure-efficacy relationship using S-MRI scale in bone. Exploratory exposure-response analyses using linear regression evaluated the relationship between avalglucosidase alfa exposure (area under the concentration-time curve over 2 weeks [AUC2W] and Cmax) and changes in percentage forced vital capacity (FVC) and distance covered in the 6-minute walk test (6MWT) in LOPD patients. No apparent relationship between exposure and efficacy endpoints was identified [29]. Population Pharmacokinetic Model Evaluation Across all six studies, the quality of the popPK models was evaluated using both graphical and numerical diagnostic approaches [21–26]. Goodness-of-fit (GOF) plots (observed vs predicted values) consistently indicated that the models adequately described the observed data without evidence of systematic deviations or major bias. In the studies by Tuffal et al. and Tiraboschi et al., parameters of the second and third compartments were fixed to avoid model over-parameterization and non-identifiability [21, 22]. Predictive performance of the final models was further confirmed by visual predictive checks (VPCs), which demonstrated good agreement between simulated and observed concentrations. Model robustness and parameter stability were evaluated with nonparametric bootstrap analyses, typically based on 1000 simulation replicates. Median bootstrap estimates were consistent with the final parameter values and confidence intervals (CIs) were reasonably narrow, supporting parameter reliability. Convergence rates were high (approximately 85%–87%), representing a good stability of the final popPK models. Final popPK parameters were estimated with acceptable precision, as indicated by relative standard errors (RSEs) generally below 30%. An exception was observed for the IIV estimates in the study by Troy et al., where precision was poor, likely reflecting the small sample size [26]. The residual variability estimates also indicated a high degree of variability from other sources. As reported by Gras-Colomer et al. [25] and Troy et al. [26], shrinkage values were calculated to evaluate the reliability of Bayesian individual parameter estimates. In the study by Troy et al., these were generally high for each of the IIV random effects, with the exception of Vc, transit rate constant (Ktrans), and volume of distribution in the cerebrospinal fluid (VCSF), suggesting that only a subset of parameters was well informed by the available data. However, no studies reported a formal external validation, which is understandable given the limited availability of independent datasets in rare diseases such as LSDs. Discussion In this review, popPK and popPD studies of therapeutic enzyme therapy in LSDs were summarized. Six popPK studies covering four different LSDs were selected [21–26], two of which also contained popPD analyses [23, 25], and one study investigated exposure-response relationships [29]. From all available popPK analyses, four studies also included data from pediatric patients. As summarized in Table 1, all included studies included fewer than 100 subjects and relied on data obtained from Phase I/II and Phase III clinical trials, which inherently limit IIV and the ability to robustly characterize exposure-response relationships. Nevertheless, the majority of studies employed extensive plasma sampling schemes, with at least 8 samples and up to 31 samples per subject, enabling reliable popPK model development. Furthermore, all studies included in this review applied the NLME modeling approach for the analysis of PK and PD data. Nonlinear mixed-effects modeling is widely regarded as the gold standard in popPK/PD analyses, as it allows for the characterization of both PK/PD parameters in the population, quantification of variability between and within patients, and identification of covariates explaining the variability in PK/PD parameters [13, 30, 31]. The advantage of NLME modeling is its ability to reliably obtain PK and PD models using sparse clinical data, which is particularly relevant in rare diseases such as LSDs, where patient numbers are limited and rich sampling is often not feasible or unethical (e.g., pediatric patients) [30]. The NLME modeling approach is also used in simulating alternative dosing regimens, which can support dose optimization and trial design. Nevertheless, one should not forget that NLME modeling requires assumptions about structural and statistical models, which may influence parameter estimates and predictions. Model evaluation and external validation are, therefore, critical steps to ensure robustness of the model and reliability of the results. The estimated Vcs reported by Qi et al. (1.52 L) and Gras-Colomer et al. (1.10 × 10−2 L in plasma and 2.57 × 10−1 L in leukocytes) are markedly low, particularly when compared with the values reported in the other included studies (approximately 3.5 L). Qi et al. determined the Vc for an MPS VII subject with a baseline body weight of 20 kg [23]. The predicted apparent Vc of 1.52 L in a 20 kg pediatric patient corresponds to approximately 76 mL/kg, which is within the expected range of blood volume in children. For a large hydrophilic enzyme such as vestronidase alfa, largely confined to the intravascular and extracellular space, this estimate is pharmacokinetically plausible. Gras-Colomer et al., on the other hand, suggest in their discussion that the low predicted apparent Vc values could reflect a high distribution into target organs, where leukocytes are transformed into tissue macrophages [25]. They further note that the volume of distribution of peptides is generally small and limited to the extracellular space, which is consistent with the findings of the present analysis. Despite this explanation, these values seem unrealistically small. In four popPK studies, the occurrence of ADAs was assessed as a covariate for explaining IIV in estimated PK parameters [21–23, 26]. The impact was considered typically through graphical evaluation of ADA titers and neutralizing activity in relation to drug concentration. In some cases, high ADA titers were associated with reduced systemic concentrations of the therapeutic enzyme, and neutralizing ADAs corresponded with concentrations below the limit of quantification, suggesting a potential influence on drug disposition. However, the presence of ADAs was not included as a covariate in the final popPK models, often due to small sample sizes, sparse ADA data, or lack of consistent trends across patients. Anti-drug antibodies can alter pharmacokinetics by increasing clearance, formation of immune complexes, or neutralizing drug activity, which may diminish exposure and thus efficacy [9, 10]. In LSDs specifically, immunogenicity to ERT frequently occurs, particularly in patients lacking residual endogenous enzyme activity [32]. Although the impact of ADAs is often explored descriptively, for example, by comparing individual PK parameters or exposure metrics according to ADA status, their effects are rarely quantified within popPK/PD models. This is likely due to limited sample sizes, sparse longitudinal ADA data, and uncertainty regarding neutralizing capacity. While incorporating ADA measurements as time-varying covariates could improve characterization of exposure-response relationships, practical implementation is often challenging given the constraints of typical clinical trial datasets in rare diseases. In three of the included studies, allometric scaling was applied to scale the population PK parameters [22, 23, 26]. Allometric scaling is a widely used method that provides a simple and empirical approach for extrapolating PK parameters from adults to pediatric patients by accounting for differences in body weight in drug disposition. Clearance is commonly scaled to body weight using an exponent of 0.75, while volume of distribution is scaled with an exponent of 1.0, reflecting established physiological principles across species and age groups. However, these fixed exponents may not always hold in pediatrics, particularly in neonates and infants, as maturation of metabolic pathways, organ function, and drug-specific properties contribute substantially to clearance and volume of distribution beyond body weight alone [33, 34]. In one study, a time-varying value for body weight was considered for the allometric scaling process since the body weight of the pediatric population increased during the clinical study [22]. Troy et al. developed a two-compartment model for CSF data and a one-compartment model for plasma data. Weight-based allometric scaling was applied to the systemic components of the model, whereas a non-allometric age-based scaling method was used for the brain components to improve model fit, since brain maturation correlates more reliably with age than with TBW in children with MLD [26]. The studies included in this review identify TBW as a significant covariate for both CL and Vc. These findings suggest that the distribution and elimination of ERTs are influenced by TBW and may also be affected by body composition, underscoring the importance of body weight as an explanatory factor for IIV in CL and Vc in patients treated with ERT. As ERTs comprise large, hydrophilic proteins administered IV, their distribution is largely confined to the vascular and interstitial spaces. Therefore, IIV associated with their population PK parameters may be more correlated to LBW or IBW than to TBW. However, this has not been studied to date. Therefore, we would like to emphasize the importance of evaluating additional morphometric parameters beyond TBW. Dosing based solely on TBW may result in relatively high exposure in individuals with a high body fat percentage (high body mass index), in case that Vc and CL scale with morphometric parameters other than TBW. The pharmacokinetics of therapeutic enzymes are often nonlinear, and extensive mathematical modeling may be required for resolving complexities associated with their disposition [35]. Therapeutic enzymes are cleared from the bloodstream via both receptor-mediated processes, such as mannose 6-phosphate (M6P) receptor-mediated endocytosis, and non–receptor-mediated pathways, including non-specific proteolytic degradation [57, 58]. At lower concentrations, clearance can be rapid due to high receptor affinity, whereas at higher concentrations these receptors may become saturated, leading to a nonlinear relationship between dose and systemic exposure, as reported in the study conducted by Qi et al. [23]. Tuffal et al. observed a transient saturation of the nonlinear (Michaelis–Menten) clearance pathway during the infusion period of avalglucosidase alfa, when plasma concentrations were increasing toward Cmax [21]. Since this saturable process was limited to the high plasma concentrations achieved during infusion, the duration of saturation was prolonged at higher dose levels. However, Michaelis–Menten clearance contributed minimally to overall systemic clearance. While nonlinear clearance is usually saturated at concentrations above Km, simulations demonstrated that linear clearance predominated across the plasma concentration range observed at the approved dose of 20 mg/kg every 2 weeks. Overall, saturable uptake contributes to variability in drug disposition across patients, especially when variability in receptor expression, disease severity, or body weight affects enzyme distribution and clearance [58]. In a popPK analysis, these processes can be characterized and applied to guide dose optimization and individualization [22]. However, in popPK analyses, drug disposition is primarily data-driven and does not take the full physiological information into account. To evaluate and predict the potential efficacy of ERT in LSDs, there is a need to understand the physiological distribution of the drug and variables that may affect this. In contrast, physiologically based pharmacokinetic (PBPK) modeling provides a mechanistic approach to understand how drug properties and physiological processes affect pharmacokinetics. Physiologically based pharmacokinetic models make use of a larger number of compartments, each representing different organs and tissues connected by the circulatory system. This approach incorporates detailed physiological parameters, such as organ blood flow and tissue composition, allowing for better extrapolation across populations or species. While PBPK modeling can be used to estimate common PK parameters, it is particularly valuable for understanding drug behavior at a specific tissue level, which popPK cannot fully capture mechanistically [36]. The PBPK of therapeutic enzymes enables prediction of tissue concentration profiles that are relevant for analyzing on-target PD effects [37–39]. Despite its potential to capture complex processes such as tissue distribution, PBPK modeling remains largely unexplored for therapeutic enzymes in LSDs, with only a few exploratory studies reported to date [4, 40]. This represents an important opportunity for future research, as this approach could provide added value in optimizing dosing strategies for ERT in LSDs, where the most appropriate dosage regimen remains controversial. The effectiveness of ERT is strongly influenced by enzyme half-life, which differs markedly between plasma and intracellular compartments. Following IV administration, therapeutic enzymes circulate in plasma, where they typically exhibit short half-lives (minutes to 2.5 h). This is largely due to rapid receptor-mediated uptake, primarily via the M6P receptor. Consequently, plasma exposure declines rapidly after infusion. For ERT to be effective, the enzyme must be internalized by target cells and trafficked to lysosomes, where a prolonged half-life (between 2 and 9 days) allows sustained enzymatic activity. These contrasting half-lives have important therapeutic implications. While the short plasma half-life limits circulating enzyme availability, intracellular persistence largely determines the duration of the PD effect and clinical benefit. This distinction helps to explain why many ERTs are administered intermittently, despite rapid plasma clearance. For most ERTs, intracellular half-lives are sufficiently prolonged to support dosing intervals of 1–2 weeks or longer rather than daily administration. Despite their short plasma half-life, most therapeutic enzymes demonstrate clinical efficacy that extends well beyond baseline levels [41]. This demonstrates that intracellular and plasma PK might differ. In two studies, the intracellular PK was evaluated in plasma leukocytes and monocytes. In LSDs, therapeutic enzymes are predominantly internalized by receptor-mediated endocytosis into target cells. Moreover, the extent of this cellular uptake is a key determinant of ERT efficacy. In contrast, plasma concentrations of therapeutic enzymes decline quickly due to the short half-life of the enzymes and rapid tissue distribution. Thus, serum PK profiles primarily reflect distribution and clearance rather than intracellular availability. It is hypothesized that cellular enzyme activity correlates better with substrate reduction and clinical improvement than serum enzyme activity [24]. Leukocytes can be readily obtained from blood and are an accessible surrogate for leukocytes residing in tissue [24, 25]. Consistent with that, Gras-Colomer et al. developed a popPK/PD model in GD that linked leukocyte GCase activity to improvement in bone-marrow infiltration by macrophages (Gaucher cells), an important parameter for evaluating disease severity [25]. In this analysis, intracellular exposure is supposed to best predict clinical response, underscoring intracellular PK as the relevant exposure metric for ERT. These findings support using leukocyte enzyme activity to guide dose optimization and individualization. Accordingly, intracellular PK may better reflect actual PD effect and could be more relevant for understanding ERT responses [42]. Intracellular enzymatic activity is considered a PK endpoint for predicting individual enzymatic activity [25]. Therefore, intra-leukocyte activity is a possible marker for therapeutic individualization in patients receiving ERT in LSDs [43]. However, it is important to note that Qi et al. successfully developed a popPK/PD model for vestronidase alfa based on plasma concentrations, demonstrating that systemic PK parameters can also adequately describe exposure-response relationships in LSDs [23]. In the study by Troy et al., the PK of a therapeutic enzyme in CSF was evaluated [26]. Intravenously administered enzymes are not expected to cross the blood–brain barrier (BBB) [44, 45], whereas the majority of LSDs affect the brain [44]. Therefore, targeting the CNS in LSDs remains a major challenge. Beyond dose adjustments and alternative routes of administration, enzyme modifications such as fusion proteins and chemical changes that alter receptor affinity can increase blood concentrations of therapeutic enzymes and help maintain higher levels at the BBB for a prolonged period [46, 47]. Recently, Muenzer et al. reported the first results of a Phase 1–2, open-label study evaluating the safety and efficacy of tividenofusp alfa in children with MPS type II (MPS II) [59]. Tividenofusp alfa, a novel fusion protein using the transferrin receptor to cross the BBB, is administered IV while targeting the CNS. However, plasma PK provides limited insight into drug exposure at the intended site of action. Characterization of PK in CSF, and ideally in brain tissue, may offer more meaningful information on enzyme distribution, persistence, and pharmacologically relevant concentrations in the target organ. The popPK analysis of intrathecal-administered arylsulfatase A in patients with MLD by Troy et al., illustrates this principle: CSF concentrations were described by a two-compartment CNS model with rapid distribution into a brain tissue compartment and a remarkably long terminal half-life in the CNS (ca. 480 h), suggesting sustained drug persistence in the brain between two doses. These findings underscore the importance of assessing tissue concentration rather than relying solely on plasma exposure to link enzyme disposition to clinical outcomes and to guide optimization of dosing strategies. The relationship between the PK of therapeutic enzymes and their efficacy in patients with LSDs has hardly been investigated despite it being an important step for evaluating and predicting treatment outcomes. Linking PD data to the existing popPK model provides a more robust framework for predicting both enzyme exposure and its corresponding therapeutic effects, enhancing the clinical relevance of the model [48]. A popPK/PD model for therapeutic enzymes would provide insights into the concentration-effect relationship, allowing for a deeper understanding of how enzyme concentrations correlate with clinical efficacy and safety outcomes in patients with LSDs. Furthermore, such a model may support individualized dosing strategies, optimizing therapeutic outcomes by accounting for both IIV in drug exposure and its corresponding PD response. This also highlights the potential importance of intracellular drug concentrations in determining efficacy and in understanding and predicting inter-subject variability in drug response. Berger et al. selected intracellular concentrations for the development of a popPK model for imiglucerase based on the observation that the biweekly dosing regimen does not correspond with the very short plasma half-life of imiglucerase of just a few minutes. In the absence of pharmacological evidence to explain the clinical efficacy of this regimen, they hypothesized that intracellular PK may be more relevant than plasma PK for understanding ERT response. Given this apparent paradox between the very short plasma half-life and the clinical efficacy of imiglucerase, they proposed that intracellular and plasma PK might differ [24]. Similarly, Gras-Colomer et al. developed a popPK model incorporating both plasma and leukocyte enzymatic activities. For the exposure-response analysis, leukocyte enzymatic activity was selected, given that the monocyte/macrophage system represents the primary target in GD1. In addition, plasma enzymatic activity has shown higher IIV than leukocyte enzymatic activity, possibly due to the short plasma half-life of the therapeutic enzyme [25]. However, quantifying intracellular drug concentrations directly in humans is challenging. In the included studies intracellular concentration was assessed using circulating plasma monocytes and leukocytes. In GD1, uptake of imiglucerase via the mannose receptor on circulating blood monocytes has been shown to serve as an accessible surrogate for tissue macrophages [54, 55]. Moreover, pharmacokinetic-biomarker analyses based on GCase activity in leukocytes have demonstrated a good correlation with clinical response to ERT [53, 56]. However, in a previous study, Gras-Colomer et al. carried out an analysis using samples containing proportions of different leukocyte subsets [53], while enzyme activity has been shown to be 40- to 50-times higher in monocytes compared with lymphocytes and neutrophils in patients with GD1 [54]. In Pompe disease, the primary target tissues of recombinant human acid alpha-glucosidase (rhGAA) are cardiac and skeletal muscles, where the enzyme is transported to lysosomes to break down glycogen. Plasma drug concentrations are often used as a surrogate for tissue exposure, based on the assumption that systemic drug concentrations reflect intracellular drug concentrations at the site of action. However, previous research in Fabry and Pompe disease has demonstrated that this assumption does not fully apply to alglucosidase alfa and agalsidase beta and is likely also not applicable to other ERTs internalized via the M6P receptor due to the so-called 'sink effect' [49]. High expression of M6P and mannose receptors in organs like the liver and spleen causes significant uptake of the enzyme early after infusion, preventing it from reaching the main target tissues and thus acting as a sink. In GD, this phenomenon is advantageous as the macrophages in the liver and spleen are the target tissues. In the less favorable cases, increasing the enzyme dose may help to overcome this effect, although it may also increase the risk of adverse effects or antibody formation [50]. Further research on intracellular exposure is essential to determine therapeutic efficacy and to improve the understanding and prediction of IIV in ERT response. Among the included studies, only a few popPD analyses have been published whereas popPK analyses are more abundant. In the study from Qi et al., an indirect response model linking vestronidase alfa exposure to reduction in urinary glycosaminoglycans (uGAGs) in patients with MPS was established [23], whereas Gras-Colomer et al. applied a discrete-time Markov model to capture the relationship between predicted imiglucerase exposure in leukocytes and the probability of transition between different stages of bone marrow infiltration by Gaucher cells in GD [25]. In contrast, all other studies focused primarily on PK characterization, with PD evaluation limited to exploratory exposure-response analysis [29]. No apparent relationship between avalglucosidase alfa exposure (AUC2W and Cmax) and efficacy endpoints (changes in percentage predicted FVC and distance covered in the 6MWT) was identified. There were several reasons for this. First, the exposure-response analysis was based on data generated at a unique dose level (20 mg/kg every 2 weeks), reducing the ability to detect correlations between systemic exposure and clinical outcomes. Second, the pharmacological effect of avalglucosidase alfa depends on receptor-mediated cellular uptake and subsequent lysosomal trafficking. Therefore, plasma exposure metrics such as AUC and Cmax do not adequately reflect drug concentrations at the site of action. Third, the clinical endpoints evaluated are influenced by various patient- and disease-related factors, including disease heterogeneity, age, sex, height, weight, and fatigue, which may further affect potential effect-response relationships. The limited number of published popPD models underscores the need for further development of quantitative PK/PD frameworks that capture both biomarker and clinical endpoints in LSDs. For therapeutic enzymes in LSDs, indirect response models and Emax models appear particularly appropriate, as ERT does not directly cause an instantaneous change in clinical outcomes. Moreover, incorporating disease progression models would be valuable to link exposure to long-term functional outcomes, such as the 6MWT or FVC in Pompe disease [51]. However, the inclusion of placebo arms in life-threatening diseases is often unethical and, therefore, not feasible for modeling disease progression. As a realistic alternative, disease progression can be characterized using external and historical patient data rather than placebo controls. Historical cohort data collected prior to treatment initiation provide valuable information on untreated disease progression. In addition, longitudinal data from patients before the start of therapy can be used to estimate baseline progression rates and IIV within the same population. Furthermore, observational studies may serve as external comparators, provided that differences in patient characteristics and data collection are appropriately accounted for within the modeling framework. In some cases, data from clinical trials of therapies that were ultimately shown to be ineffective may also contribute to understanding disease progression, following careful evaluation of potential residual treatment effects and bias. Finally, Markov models may be useful when disease progression is described by a limited number of clinically defined categories (e.g., disease severity grade) and when patients are observed at relatively few time points over time, as is the case in GD [52]. Together, these approaches could enhance the ability of popPK/PD models to support dose optimization and individualized treatment strategies in LSDs. Conclusion This review highlights the currently published popPK and popPD models describing the disposition and effect of therapeutic enzymes in LSDs. To date, six popPK models and three popPD models of exposure-response relationships are available in the literature. All models were established using the NLME modeling approach. Progress in popPD modeling has been more limited, with only a few studies linking exposure to PD biomarkers or, more importantly, clinical outcomes. Nevertheless, these analyses provide valuable insights into the exposure-effect relationships, supporting the use of indirect response and Emax models to capture the delayed and saturable nature of ERT effects. Innovative approaches, such as intracellular PK measurements in plasma leukocytes and monocytes, age-based scaling for CNS distribution in children following IT administration of ERT, and the application of Markov models for categorical outcomes, illustrate the potential of advanced pharmacometric methods to address the unique complexities of LSDs. Nevertheless, there remains a need to further clarify the role of drug concentration measurements in target cells and to better define their relevance for therapeutic response, to characterize CNS distribution in LSDs affecting the brain following IT administration of ERTs and IV administration of a novel fusion protein, and to identify robust PD biomarkers in defining exposure-response relationships of therapeutic enzymes in LSDs. Supplementary Information Below is the link to the electronic supplementary material.Supplementary file1 (PDF 210 KB) Supplementary file1 (PDF 210 KB)