Mechanistic PBPK/PD Modeling of Free and Nano-liposomal Irinotecan Reveals Formulation-Specific Determinants of Disposition and Efficacy

Abstract

Irinotecan is widely used in cancer therapy but is limited by significant toxicities due to systemic and intestinal exposure to its active metabolite, SN38. To improve its therapeutic profile, irinotecan has been encapsulated in pegylated liposome as a nano-liposomal form (nal-IRI) to modify its pharmacokinetics (PK) and enhance tumor delivery via the enhanced permeability and retention effect. While nal-IRI has shown clinical benefits, the formulation-specific PK and pharmacodynamics (PD) underlying its efficacy and safety remain unknown.

This study aimed to develop a physiologically based pharmacokinetic/pharmacodynamic (PBPK/PD) model to compare the disposition and tumor response of irinotecan and SN-38 following administration of free irinotecan (free-IRI, Camptosar^®) and nal-IRI (Onivyde^®) in pediatric tumor xenografts. Plasma and tissue PK data (liver, spleen, kidney, brain, lung, and tumor) were collected from healthy and tumor-bearing mice treated with various intravenous doses of both formulations.

The model accurately described plasma, tissue, and tumor concentrations of irinotecan and SN-38. Key determinants of disposition included enterohepatic recycling, carboxylesterase-mediated conversion in liver/plasma, and clearance through biliary/metabolic pathways for irinotecan, and biliary/renal routes for SN-38. Nal-IRI exhibited formulation-specific characteristics, including phagocyte-mediated uptake, nonlinear plasma clearance, liposomal release and permeability-limited tissue distribution, that are major determinants of nal-IRI disposition.

PD modeling indicated intra-tumoral SN-38 exposure was the principal driver of antitumor efficacy. Nal-IRI achieved sustained and higher SN-38 tumor exposure, producing more rapid and durable tumor suppression than free-IRI. This integrated PBPK/PD framework provides mechanistic insights into the enhanced efficacy of nal-IRI and supports its optimized use in irinotecan-based cancer therapy.

1. Introduction

Irinotecan (CPT-11) is an antineoplastic agent whose efficacy arises from inhibition of topoisomerase I, an enzyme that transiently cleaves and re-ligates single strands of DNA to relieve supercoiling during replication. Irinotecan is a prodrug that requires enzymatic conversion by carboxylesterases (CES), predominantly located in the liver, intestinal mucosa, and plasma, to generate its active metabolite, SN-38. SN-38 is approximately 1000 times more potent than irinotecan but is also associated with severe, dose-limiting toxicities, particularly diarrhea and neutropenia[1].

Following activation, SN-38 undergoes glucuronidation in the liver, primarily by the UGT1A1 enzyme to form SN-38 glucuronide (SN-38G), which is excreted into the bile and undergoes enterohepatic recirculation (Figure S1). In the intestine, bacterial β-glucuronidases can hydrolyze SN-38G back to SN-38, thereby increasing local and systemic exposure to the active metabolite. This recycling contributes to the significant gastrointestinal toxicity associated with irinotecan and has led to a Black Box Warning in its clinical labeling[1, 2].

To enhance efficacy and reduce adverse effects, irinotecan has been reformulated as a nano-liposomal formulation (nal-IRI), in which the drug is encapsulated within ~110 nm PEGylated liposomes. This nano-encapsulation provides several advantages: it protects irinotecan from premature CES-mediated hydrolysis and protein binding in plasma, and it exploits the enhanced permeability and retention (EPR) effect observed in tumors. The leaky endothelial junctions and impaired lymphatic drainage in typical tumor tissues allow nanoparticles to preferentially accumulate and persist within the tumor microenvironment, resulting in prolonged intra-tumoral drug exposure and potentially improved therapeutic outcome[3, 4].

Despite its clinical approval and demonstrated benefit, the pharmacokinetic (PK) and pharmacodynamic (PD) characteristics of nal-IRI remain incompletely understood. The disposition of nano-formulations is influenced by complex mechanisms, including phagocytic uptake by the mononuclear phagocyte system (MPS), drug-release kinetics, and tumor-specific factors[5-7], all of which present significant challenges for mechanistic modeling and prediction. A clear understanding of irinotecan PK is also essential to delineate formulation-specific behaviors and therapeutic effects associated with nal-IRI.

To address these gaps, we employed a physiologically based pharmacokinetic/pharmacodynamic (PBPK/PD) modeling approach to mechanistically characterize the disposition of irinotecan and SN-38 in both free and nano-liposomal forms. This approach enables the integration of drug absorption, distribution, metabolism, and excretion (ADME) processes across plasma, tissue and tumor compartments. Specifically, this study aimed to: (i) quantify and compare the tissue and tumor disposition of irinotecan and SN-38 in healthy and tumor-bearing murine models following IV administration of free-IRI and nal-IRI, and (ii) develop an integrated PBPK/PD model to elucidate the relationship between drug exposure and antitumor efficacy. This framework provides a quantitative, mechanistic basis for understanding the formulation-specific dynamics that underline the safety and efficacy of irinotecan-based therapies.

2. Methods

2.1. PK/PD data used for model development

Preclinical PK/PD data for irinotecan and nal-IRI were obtained from previously reported studies conducted in compliance with Institutional Animal Care and Use Committee (IACUC)-approved protocols [8]. In these studies, nal-IRI (MM-398, Onivyde^®, 5 mg/mL) and free irinotecan (Camptosar^®, 20 mg/mL) were administered intravenously to healthy BALB/c mice at doses of 5, 10, or 20 mg/kg, and tumor-bearing athymic nu/nu mice at a dose of 20 mg/kg. To establish a tumor-bearing model, human neuroblastoma CHLA-136 cells (15 x 10⁶ cells in 100 μL) were subcutaneously injected into athymic nu/nu mice. Following drug administration, concentrations of irinotecan and SN-38 were measured in plasma, tissues, and tumors using a validated bioanalytical assay method using an Agilent 1200 HPLC system coupled with an Applied Biosystems 4000 QTRP LC-MS/MS system. Samples were collected at multiple sampling points post-dose from plasma and various tissues, including brain, lung, kidney, liver, spleen, and tumor, as summarized in Table 1.

Table 1.

Summary of PK/PD data used for model development.

Mouse Model (n = 3)	Delivery Formulation	Dose (mg/kg) Single IV dose	Sampling Time (hr)	Samples Collected for PK analysis
Healthy mice (Balb/c)	Non-Liposomal (free-IRI)	5, 10, 20	0.083, 0.17, 0.25, 0.5, 1, 2, 4, 8	Plasma Brain Lung Kidney Liver Spleen
	Liposomal (nal-IRI)		1, 4, 8, 12, 24, 36, 48
Tumor-bearing mice (Athymic nu/nu)	Non-Liposomal (free-IRI)	20	4, 24, 48	Plasma Brain Lung Kidney Liver Tumor
	Liposomal (nal-IRI)
Mouse Model (n = 3)	Delivery Formulation	Dose (mg/kg) IV dose	Cell Line	PD measurement
Tumor-bearing mice (Athymic nu/nu)	Non-Liposomal (free-IRI)	10 2 total doses (Days 1 & 4)	CHLA-79m (NB) CHLA-79m (NB) CHLA-258 (EFT) CB-NJ-Rm (RMS) Rh30m (RMS)	Tumor Volume
	Liposomal (nal-IRI)

n: number of animals per each data time-point

For PD evaluation, subcutaneous xenograft models were established in 6- to 8-week-old female athymic nu/nu mice using multiple pediatric cancer cell lines: neuroblastoma (CHLA-136), rhabdomyosarcoma (CB-NJRm and Rh30m), and Ewing family tumor (CHLA-258). Once tumors reached an average volume of approximately 150 mm³, mice received a total of two intravenous doses (administered on Day 1 and Day 4) of either free-IRI or nal-IRI at 10 mg/kg. Tumor growth was monitored until tumor volume exceeded 1,500 mm³ to assess therapeutic efficacy. For each xenograft model, to ensure robustness and avoid over-interpretation of spare data, only time points with at least three individual tumor measurements were retained, and mean tumor volumes were calculated to generate representative tumor growth profiles. Five out of nine xenograft models met these predefined data quality criteria and were included in the analysis.

2.2. PK data analysis

Mean plasma concentration-time profiles of irinotecan and SN-38 were analyzed using standard non-compartmental analysis (NCA) to obtain descriptive PK parameters, including volume of distribution ($V_{ss}$), total clearance ($C{L}_{total}$), half-life ($t_{1∕2}$), elimination rate constant (${\lambda }_z$), maximum concentration ($C_{max}$), and the area under the curve from zero to infinity (${\mathit{AUC}}_{0-\infty }$). NCA was performed using Phoenix WinNonlin Version 6.4, (Certara USA, Inc., Princeton, NJ). Tissue concentrations of irinotecan and SN-38 were converted from μg/g to μg/mL by assuming a tissue density of 1 g/mL. The extent of tissue distribution was assessed by calculating tissue-to-plasma partition coefficient (${\mathit{Kp}}_T$) for irinotecan, and SN-38 as the ratio of tissue ${\mathit{AUC}}_{0-\infty }$ to plasma ${\mathit{AUC}}_{0-\infty }$.

$$K{p}_T=\frac{TissueAU{C}_{0-\infty }}{PlasmaAU{C}_{0-\infty }}$$ Eq. 1

2.3. PBPK/PD Model Development: Overview

The development of PBPK/PD models for both free-IRI and nal-IRI was conducted in a systemic, stepwise manner, as illustrated in Figure 1. The modeling workflow consisted of six key steps. In Step 1, a PBPK model was constructed to describe the disposition of irinotecan and SN-38 in plasma and non-tumor tissues of healthy mice receiving free-IRI formulation. In Step 2, this model was expanded to incorporate the nal-IRI formulation by introducing compartments representing nano-liposomes and phagocytic cells (PC), which allowed for the characterization of formulation-specific features such as liposomal release kinetics and phagocyte-mediated uptake. In Steps 3 and 4, a tumor compartment was integrated into both nal-IRI and irinotecan PBPK models to describe drug distribution in plasma, non-tumor tissues, and tumor in tumor-bearing mice. Upon finalization of the PBPK models, Steps 5 and 6 involved the development and integration of a PD model. This extended the PBPK models into a unified PBPK/PD framework to quantitatively capture tumor growth dynamics and the exposure-response relationship, with intra-tumoral drug concentrations as the driver of antitumor efficacy.

Figure 1. Workflow of PBPK/PD model development for free and nano-liposomal Irinotecan in mice.

This schematic illustrates the six-step process used to develop and integrate physiologically based pharmacokinetic (PBPK) and pharmacodynamic (PD) models for both free irinotecan (free-IRI) and nano-liposomal irinotecan (nal-IRI) in mice. Steps 1 and 2 establish PBPK models in healthy Balb/c mice, capturing key disposition of free-IRI and nal-IRI formulations. Steps 3 and 4 extend these models by adding a tumor compartment (CHLA-136 xenograft) in athymic nu/nu mice, enabling the prediction of intratumoral drug concentrations. Finally, Steps 5 and 6 integrate a tumor growth inhibition model, linking intra-tumoral drug exposures to PD effects to quantitatively investigate formulation-dependent differences in tumor response dynamics.

2.4. PBPK Model Development: free irinotecan in healthy mice (Step 1)

A dual PBPK model was developed to quantitatively characterize the disposition of irinotecan and SN-38 following a single IV dose of 5, 10, or 20 mg/kg in healthy mice. The model structure (Figure 2A) was based on a whole-body PBPK framework composed of perfusion-limited compartments. These compartments were parameterized using tissue volumes ($V_T$), blood flows ($Q_T$), and tissue-to-plasma partition coefficients ($K{p}_T$) to represent the distribution kinetics of both compounds. The model included six tissue compartments, brain, lung, kidney, liver, spleen, and a lumped rest-of-body compartment (“rest organ”), along with venous and arterial blood compartments, all connected through physiological blood flow. To reflect physiological sequence in which venous blood returning to heart is first oxygenated in the lungs before being distributed to the systemic circulation, input amount in lung dynamically depends on venous concentrations. CES-mediated metabolism of irinotecan to SN-38 was incorporated into the model as a metabolic clearance process in both the venous (${\mathit{CL}}_{\mathit{CES},\mathit{PL}}$) and liver (${\mathit{CL}}_{\mathit{CES},\mathit{Li}}$) compartments. The CES activity in the liver (${\mathit{CL}}_{\mathit{CES},\mathit{Li}}$) was assumed to represent both hepatic and intestinal CES contributions.

Figure 2. Structure of the dual PBPK/PD model for free and nanoliposomal irinotecan in healthy and tumor-bearing mice.

This scheme illustrates the mechanistic whole-body PBPK/PD model developed for both free irinotecan (free-IRI) and nano-liposomal irinotecan (nal-IRI). (A) The PBPK model for irinotecan and its active metabolite SN-38 following free-IRI includes plasma (venous and arterial) and six perfusion-limited tissue compartments: liver, kidney, spleen, lung, brain, and a lumped “rest of body” compartment. These compartments are parameterized by tissue blood flow ($Q_T$), tissue volume ($V_T$) and compound-specific tissue:blood partition coefficients (${\mathit{Kp}}_T$). The IV dose of irinotecan enters into venous compartment and undergoes fecal ($F_{\mathit{Fe}}$), renal (${\mathit{CL}}_R$) elimination, and enterohepatic recirculation (${\mathit{CL}}_{\mathit{Bi}}$ and $k_{\mathit{EHR}}$). For tumor-bearing mice, a perfusion-limited tumor compartment is added, parameterized by tumor blood flow ($Q_{\mathit{Tu}}$) and volume ($V_{\mathit{Tu}}$). CES-mediated SN-38 production occurs in the liver, plasma, and tumor compartments. (B) The PBPK model for nal-IRI consists of plasma and five permeability-limited tissue compartments (brain, lung, kidney, liver, and spleen), as well as a perfusion-limited “rest” compartment. (C) Phagocytic cell (PC) sub-compartments are incorporated into the spleen, liver, lung, and venous compartments, parametrized by phagocytic uptake ($k_{\mathit{up}}$), liposomal release ($k_{\mathit{rel}}$), CES-mediated metabolism (${\mathit{CL}}_{\mathit{CES},\mathit{PC}}$), and exocytosis (${\mathit{PC}}_{\mathit{Out}}$). Nal-IRI is linked to the free-IRI PBPK model (A) via a zero-order liposomal release process ($k_{\mathit{rel}}$) to account the disposition of the released irinotecan. (D) PD model consists of a tumor compartment controlled by the combined exponential growth kinetics ($k_g$) in untreated groups and a series of compartments (T1, T2, T3) to capture the delayed tumor cell death, driven mainly by intra-tumoral SN-38 concentration (${\mathit{KC}}_{50,\mathit{SN}38}$).

The temporal changes in irinotecan and SN-38 concentrations across plasma and tissues after the free-IRI formulation were described by a series of ordinary differential equations (ODEs), representing mass balance across compartments and accounting for distribution, metabolism, and elimination processes (Supplemental Information, Eqs. S1-13). Physiological values for organ volume ($V_{\mathit{organ}}$) and regional blood flow ($Q_{\mathit{organ}}$) were sourced from the literature [9] and PK-Sim^® software and summarized in Table 2.

Table 2.

Physiological parameters (blood flow & organ weight) used for a healthy and tumor-bearing mouse.

Mouse Model	Organ	Symbol	% Cardiac Output (FQtissue)	% Body Weight (FVtissue)
			Cardiac Output (CO) = 448.8 mL/h	Body Weight (BW) = 17.0 g
Healthy mice (Balb/c)	Lung	Lu	100	0.730
	Artery	ART	100	2.244
	Venous Pool	VEN	100	4.556
	Liver	Li	25.56	5.490
	Spleen	Sp	1.340	0.350
	Brain	Br	6.720	1.650
	Kidney	Ki	18.81	1.670
	Rest Organ	Re	47.53	83.31
			Cardiac Output (CO) = 686.4 mL/h	Body Weight (BW) = 26.0 g
Tumor-bearing mice (Athymic nu/nu)	Tumor	Tu	79.53 (88)	0.274 (F)
	Rest Organ	Re	35.94% - FQTUM	82.26% - FVTUM

Balb/c mouse-specific cardiac output were obtained from Gjedde et al.¹¹

Fraction of blood flow (FQ_TUM) and organ weight (FV_TUM) for a tumor-bearing mouse are obtained from Sharma et al.¹⁹

Elimination pathways and enterohepatic circulation:

The primary elimination routes for irinotecan and SN-38 are hepatic and renal. In the liver, irinotecan is metabolized to SN-38 via CES-mediated clearance (${\mathit{CL}}_{\mathit{CES},\mathit{Li}}$), and both compounds undergo extensive biliary excretion (${\mathit{CL}}_{\mathit{Bi},\mathit{CPT}}$ and ${\mathit{CL}}_{\mathit{Bi},\mathit{SN}38}$) and enterohepatic circulation (EHC)[1, 10]. Evidence of EHC was observed in our PK data, indicated by secondary plasma and tissue peaks (Figure 3A). To mechanistically describe this, a gallbladder (GB) compartment was added to the PBPK model (Figure 2A), linking biliary excretion from the liver back into systemic circulation (Eqs. S7-10). Drug accumulation and delayed re-release from the gallbladder were modeled using a series of transit compartments (TC), parameterized by the number of compartment (N) and the mean transit time (MTT) [11]. To capture the timing of reabsorption peaks, the onset (${\mathit{EHC}}_{\mathit{start}}$) and end times (${\mathit{EHC}}_{\mathit{end}}$) of drug re-entery into the liver was incorporated, allowing estimation of reabsorption rate constants for irinotecan and SN-38 ($k_{\mathit{EHR},\mathit{CPT}}$ and $k_{\mathit{EHR},\mathit{SN}38}$). This structure captured the delayed return of drug from the gallbladder corresponding to observed secondary peaks. Fecal elimination of the non-reabsorbed fraction (i.e., drugs that do not undergo EHC) was described using a drug-specific fecal fraction ($F_{\mathit{CPT}}$ and $F_{\mathit{SN}38}$), while the reabsorbed portion, (1- $F_{\mathit{CPT}}$) and (1-$F_{\mathit{SN}38}$), re-entered the liver at a zero-order rate ($k_{\mathit{EHR},\mathit{CPT}}$ and $k_{\mathit{EHR},\mathit{SN}38}$). Renal elimination (${\mathit{CL}}_{R,\mathit{CPT}}$ and ${\mathit{CL}}_{R,\mathit{SN}38}$) was modeled through kidney compartments.

Figure 3. Plasma and tissue PK profiles of irinotecan and SN-38 following administration of free irinotecan and nano-liposomal in healthy mice.

Plasma and tissue concentration-time profiles of irinotecan and SN-38 after intravenous (IV) administration of either free irinotecan (free-IRI, triangle symbols) (Panel A) or nano-liposomal irinotecan (nal-IRI, square symbols) (Panel B) at 5 mg/kg (pink), 10 mg/kg (orange), and 20 mg/kg (dark brown). Observed data represents mean concentrations from three mice per time point (n = 3). Model-predicted concentrations are shown as solid lines (free-IRI) and dashed lines (nal-IRI). Experimental data were obtained from previously published studies (Kang et al., 2015)

To quantify how much of irinotecan and SN38 returns to systemic circulation after being excreted into bile, recycling index (RCI) was calculated as follows:

$$RCI=F_{a,b}\cdot F_B=(1-F_{fe})\cdot \left(\frac{C{L}_{Bi}}{C{L}_{total}}\right)$$ Eq. 2

where, $F_{a,b}$ is the fraction of drug or metabolite absorbed from the bile, $F_B$ is the fraction of drug or metabolite eliminated in the bile, and $F_{\mathit{fe}}$ is fraction of dose excreted throughout the lumen, ${\mathit{CL}}_{\mathit{total}}$ is total clearance and ${\mathit{CL}}_{\mathit{Bi}}$ is biliary clearance of drug or metabolite.

2.5. PBPK Model Development: nal-IRI in healthy mice (Step 2)

Following the development of the base PBPK model for free-IRI, the model was expanded to describe the disposition of nal-IRI in healthy mice. This extended model incorporated additional compartments representing nano-liposomes and phagocytic cells (PC) to capture formulation-specific processes, including liposomal release kinetics and phagocyte-mediated uptake (Figure 2B). The nal-IRI PBPK model retained the same physiological tissue and plasma compartments as the irinotecan model, using consistent parameters for organ volumes, blood flows, and partition coefficients. The distribution, metabolism, elimination of the released irinotecan, and its conversion to SN-38 followed the previously established structure and parameters (Step 1: Figure 2A).

For nal-IRI distribution, both convection- and diffusion-limited transport across capillary membranes were incorporated in tissue compartments, except for the rest-of-organ compartment, where perfusion-limited distribution was assumed. The transport of nal-IRI into tissues was modeled using Fick’s perfusion law, incorporating the permeability coefficient-surface area product (${\mathit{PS}}_T$), which represents the rate-limiting step at the capillary interface. Tissue concentration of nal-IRI ($C_{\mathit{nalIRI},T}$) was described by the following equation:

$$\frac{d{C}_{nalIRI,T}}{dt}=\left(Q_T\cdot \left(C_{ART}-\frac{C_T}{K{p}_T}\right)\cdot \left(\frac{P{S}_T}{1+P{S}_T}\right)-k_{up,T}\cdot C_T\cdot V_{T,IS}\right)∕(V_{T,IS}+V_{T,IC})$$ Eq. 3

where, ${\mathit{PS}}_T$ represents permeability coefficient-surface area product between capillary blood and tissue and regional blood flow, with assumption of surface area is proportional to the regional blood flow. Further details and derivation of this equation are provided in the Supplementary information (Eqs. S14-21).

Nal-IRI undergoes release of irinotecan ($k_{\mathit{rel},T}$) in both plasma and tissue compartments following uptake. Phagocytes, which are abundant in organs such as spleen, liver, and lung, play a significant role in the distribution and processing of liposomal formulations[12]. PC-mediated uptake was incorporated into the venous and tissue compartments of these organs. Once nal-IRI is internalized by PC at a tissue-specific uptake rate ($k_{\mathit{up},T}$), irinotecan is released from the liposome ($k_{\mathit{rel},T}$) and subsequently undergoes CES-mediated conversion (${\mathit{CL}}_{\mathit{CES},\mathit{PC}}$) to SN-38 within the phagocytic compartment (Figure 2C). Both irinotecan and SN-38 then exit the phagocytes into surrounding tissue (${\mathit{PC}}_{\mathit{Out},T}$). To capture saturation of uptake mechanisms observed in plasma, a non-linear phagocytic uptake process was modeled using Michaelis-Menten kinetics ($V_{\max ,\mathrm{PC}}$ and ${\mathit{Km}}_{\mathit{PC}}$). The ODEs for nal-IRI disposition in healthy mice are provided in the Supplementary information (Eqs. S22-54).

2.6. PBPK Model Development: free-IRI and nal-IRI in tumor-bearing mice (Steps 3 & 4)

After establishing the base PBPK models for both free-IRI and nal-IRI in healthy mice, the models were extended to describe drug disposition in tumor-bearing mice by incorporating a tumor compartment into each formulation-specific model (Steps 3 & 4). For the irinotecan model (Step 3), a perfusion-limited tumor compartment was added and parameterized with tumor-specific physiological values, including tumor blood flow ($Q_{\mathit{Tu}}$), tumor volume ($V_{\mathit{Tu}}$), and the tumor-to-plasma partition coefficient (${\mathit{Kp}}_{\mathit{Tu},\mathit{CPT}}$). These parameters were estimated during model fitting, while the PK parameters previously derived from the healthy mouse models were fixed. Within the tumor, irinotecan undergoes CES-mediated metabolic conversion to SN-38 (${\mathit{CL}}_{\mathit{CES},\mathit{Tu}}$), as evidenced in prior literature [13]. Additionally, renal clearance of irinotecan (${\mathit{CL}}_{R,\mathit{CPT}}$) and SN-38 (${\mathit{CL}}_{R,\mathit{SN}38}$) were incorporated into the kidney compartment to account for systemic elimination. Due to availability of only one single dose level (20 mg/kg) of PK data and unavailable spleen PK profile of tumor-bearing mice, same spleen PK and no additional non-linear PK in disposition of nal-IRI between healthy and tumor-bearing mice were assumed for spleen throughout model development.

The ODEs describing the disposition of the free formulation in tumor-bearing mice are presented in the Supplementary Information (Section D). Introduction of the tumor compartment necessitated modification of the original system to incorporate tumor-specific distribution and metabolism of irinotecan and SN-38 (Eqs. S57-62). The equations describing other tissues (lung, liver, gallbladder, spleen, brain, and rest of organ) remained identical to those in the base model (Eqs. S5-10 and S13).

For the nal-IRI model (Step 4), the tumor compartment was incorporated into the nal-IRI base PBPK model to describe the nal-IRI PK in tumor-bearing mice. The same tumor-specific physiological parameters used in Step 3, tumor blood flow ($Q_{\mathit{Tu}}$) and tumor volume ($V_{\mathit{Tu}}$), were applied. Additional parameters specific to the nal-IRI formulation in the tumor, including the tumor partition coefficient (${\mathit{Kp}}_{\mathit{Tu},\mathit{nalIRI}}$) and the intratumor liposomal release rate ($k_{\mathit{rel},\mathit{Tu}}$) were introduced. All other physiological and PK parameters were retained from Steps 1-3, with the exception of ${\mathit{CL}}_{\mathit{CES},\mathit{Tu}}$, ${\mathit{CL}}_{R,\mathit{CPT}}$, and ${\mathit{CL}}_{R,\mathit{SN}38}$, which were re-estimated to reflect tumor-specific disposition.

The ODEs for the nal-IRI formulation in tumor-bearing mice are presented in the Supplementary Information (Section E). Following introduction of tumor-specific processes for nal-IRI, irinotecan and SN-38 (Eqs. S63-71), the original system was modified accordingly, while equations governing other tissues (lung, liver, gallbladder, spleen, brain, kidney, and rest organ) remained unchanged from the nal-IRI base model (Eqs. S31-54).

2.7. PBPK/PD Model Development: Antitumor efficacy of free-IRI and nal-IRI in tumor-bearing mice (Steps 5 & 6)

Following the finalization of PBPK models in tumor-bearing mice, the next step involved extending these models into a PBPK/PD framework. This integrated model was developed to quantitatively describe CHLA-136 tumor growth dynamics and to characterize the exposure-response relationship, using intra-tumoral SN-38 concentrations as the driver of antitumor efficacy (Figure 2D). The PBPK/PD framework also enabled a comparative evaluation of the therapeutic outcome between the free and nano-liposomal formulations of irinotecan.

Baseline tumor growth was modeled using an exponential ($k_g$) growth kinetics from an initial tumor volume (${\omega }_0$). These parameters were estimated by fitting the model to tumor volume data from the untreated control group. The antitumor effects were modeled using intra-tumor SN-38 concentration yielding 50% of maximum efficacy (${\mathit{KC}}_{50,\mathit{SN}38}$), maximum killing rate ($K_{\mathit{max}}$), Hill coefficients ($\gamma$) to characterize non-linear cell-killing effect, driven by intra-tumoral concentration of SN-38 ($C_{{\mathit{Tumor}}_{\mathit{SN}38}}$) for both formulations. To account for the observed delay in therapeutic onset, three transit compartments (T1, T2, T3) were included, with same first-order transit rate ($k_1$) to reflect delays in drug action (Figure 2D). Additionally, fold-difference of maximal killing rate ($\alpha$) of nano-liposomal formulation was parameterized and estimated to quantify a difference of maximum killing rate depends on formulations. The equations describing tumor growth model are provided in the Supplementary Information (Eqs. S72-76)

To evaluate the driver for anti-tumor efficacy, intra-tumoral irinotecan and SN-38 concentrations were assessed as potential drivers in the PD model. To assess the generalizability of the PBPK/PD model, the framework was applied to four additional pediatric tumor xenograft models, including CHLA-258, Rh30m, CB-NJRm, and CHLA-79, using the same model structure. The same tumor PK were assumed across all xenografts due to the absence of tumor-specific tissue PK data. Physiological and PK parameters derived from the PBPK model of both formulations to be fixed during PD model development. Those PD parameters were estimated separately for each xenograft to capture tumor growth dynamics.

The tumor static concentration (TSC), defined as the drug concentration at which tumor neither grow nor shrinks, was calculated using Eq. 4 based on the tumor growth rate ($k_g$), maximum killing rate ($K_{\mathit{max}}$), and drug potency (${\mathit{KC}}_{50,\mathit{SN}38}$) estimated from mouse PD data:

$$TSC=\frac{(k_g\times K{C}_{50,SN38})}{(K_{max}-k_g)}$$ Eq. 4

The calculated TSC threshold for all pediatric cell lines were then overlaid with simulated mouse tumor PK profiles at 20 mg/kg for both nal-IRI and free irinotecan. This enabled assessment of the magnitude and duration of intra-tumoral SN-38 concentration above the TSC required for effective antitumor activity.

2.8. Model Parameterization

Physiological parameters, including organ volumes and blood flow, were fixed to literature-reported values. Structural parameters governing tumor physiology (i.e., tumor volume and perfusion) were estimated initially and subsequently fixed, whereas formulation-specific and tumor-specific PK parameters were estimated within biologically plausible, literature-informed bounds. For parameters that were poorly identifiable or associated with high uncertainty due to limited tissue-level data or the absence of supporting in vitro measurements (e.g., $k_{\mathit{rel},T},k_{\mathit{up},T},{\mathit{PC}}_{\mathit{Out},T}$), an initial estimation step was followed by fixing them to physiologically reasonable values to improve identifiability and precision of the remaining parameters. Phagocytic uptake rates were constrained within ranges reported for organ-specific macrophages uptake of pegylated liposomes (~80-200 nm), as summarized in nanomaterial PBPK modeling studies.

Free irinotecan and SN-38 were modeled using a perfusion-limited tissue distribution framework, consistent with the behavior of small, lipophilic compounds. In contrast, nano-liposomal irinotecan was described using a permeability-limited distribution model across capillary membranes in all organs except the lumped rest-of-body compartment. It was assumed that once released from the liposome, irinotecan follows the same disposition as the free formulation. For PD modeling, tumor PK was assumed to be consistent across all xenograft models.

Physiological and PK parameters were retained as much as possible between healthy and tumor-bearing mice, with exception of selected clearance parameters, which were re-estimated to account for tumor-specific alterations in drug disposition. To further evaluate the impact of these assumptions, a global sensitivity analysis was conducted to confirm that the fixed parameters were not dominant drivers of key model outputs, thereby supporting the robustness of the model despite limited parameter identifiability.

2.9. Model selection criteria and global sensitivity analysis

Model parameter estimation was performed using R studio and ADAPT5 (Biomedical Simulation Resource, CA) through the Maximum Likelihood Estimation (MLE) approach, optimizing the fit between observed and model-predicted data. Throughout model development, various model structures were evaluated to best represent the kinetics of tissue distribution (e.g., permeability-limited vs. perfusion-limited models), phagocytic uptake (e.g., first-order vs zero-order), and the inclusion or exclusion of enterohepatic circulation. Final model selection was based on a combination of criteria, including goodness-of-fit, Akaike Information Criterion (AIC), precision of parameter estimates (as measured by coefficient of variation, CV%), and physiological plausibility of the estimated parameters. Model predictive performance was evaluated using quantitative goodness-of-fit (GOF) diagnostics, including predicted-versus-observed concentration plots on a log-log scale and absolute average fold error (AAFE) metrics for irinotecan and SN-38 concentrations.

To evaluate the influence of individual parameters on model output and reduce model complexity, a global sensitivity analysis (GSA) was performed using the Partial Rank Correlation Coefficient (PRCC) method, implemented via the mrgsolve and sensitivity packages in R. Parameters were classified as strongly (∣PRCC∣ > 0.5), moderately (0.3-0.5), or weakly (0.2-0.3) influential. Parameters exhibiting low sensitivity (i.e., negligible influence on key PK and PD outcomes) were excluded from further optimization to enhance computational efficiency and model robustness.

2.10. Model Validation

To further assess external predictive performance, the model was evaluated against independent plasma and tumor PK data from Kalra et al. [13], who reported the disposition of free irinotecan and nal-IRI in HT-29-bearing NOD/SCID mice. Model predictions for plasma PK were evaluated at a dose of 20 mg/kg, while tumor PK predictions were generated at 40 mg/kg to match the dose level at which tumor concentration–time profiles were reported in the original study. Predicted concentration–time profiles were overlaid with observed data to assess the model’s ability to reproduce formulation-specific plasma and tumor exposure patterns.

2.11. Model Simulations: Toxicity Risk & Exploring Formulation Strategy

Upon PBPK/PD model development, we performed simulations to investigate the source of toxicity risk, specifically, elevated SN-38 concentration in off-target tissues, when comparing free versus nano-liposomal formulations. These simulations were designed to provide insights into formulation strategies that minimize systemic toxicity while enhancing targeted drug delivery to tumors.

To identify optimal formulation strategies, we developed a composite scoring system integrating three criteria: (i) tumor-to-liver AUC ratio of SN-38, (ii) duration of tumor SN-38 concentration above KC₅₀ ($T_{\mathit{above}>{\mathit{KC}}_{50}}$), (iii) inverse of plasma SN-38 AUC (${\mathit{AUC}}_{\mathit{plasma},\mathit{SN}38}$) as a surrogate for off-target toxicity risk. Each metric was normalized to a 0-1 range using min-max normalization to prevent unintended weighting arising from differences in numerical scale across component metrics, and then normalized metrics were then combined into a composite heatmap to visualize formulation performance. This approach was used to ensure that no single component dominates the overall score and that poor performance in any individual criterion appropriately penalizes that composite ranking. The overall composite score was calculated as follows (Eq. 5):

$$\begin{gathered} CompsiteScore \\ =\frac{AU{C}_{Tumor,SN38}}{AU{C}_{Liver,SN38}}\cdot T_{above>K{C}_{50}} \\ \cdot \frac{1}{AU{C}_{plasma,SN38}} \end{gathered}$$ Eq. 5

3. Results

3.1. Plasma and tissue PK data

Compared with the free drug formulation, nal-IRI administration resulted in markedly increased systemic exposure, as reflected by higher AUC_0-∞ and ${\mathrm{C}}_{\max }$ for both irinotecan and SN-38 (Figure 3, Table 3). The volume of distribution ($V_{ss}$) was smaller following nal-IRI, indicating more restricted tissue distribution. In addition, total systemic clearances ($C{L}_{total}$) for both irinotecan and SN-38 was consistently lower after nal-IRI administration, resulting in prolonged systemic exposure and extended terminal half-life ($t_{1∕2}$). Across the evaluated dose range (5, 10, 20 mg/kg), AUC increased more than propotionally with dose, whereas $C{L}_{total}$ decreased, demonstrating nonlinear PK of irinotecan and SN-38 following nal-IRI administration (Table 3).

Table 3.

Non-compartmental analysis (NCA) of plasma PK data

			Pharmacokinetic Parameters
Formulation	Compound	Dose mg/kg	λz hr−1	t1/2 hr	AUC0-∞ μg·hr/mL	Cmax μg/mL	CLTotal mL/hr	Vss mL
Free-IRI	irinotecan	5	0.48	1.44	0.35	0.31	246	267
		10	0.20	3.49	2.10	1.59	80.8	329
		20	0.68	1.03	2.99	2.78	114	90.2
	SN-38	5	0.21	2.06	0.31	0.57	275	284
		10	0.22	1.77	0.62	1.42	273	284
		20	0.20	2.17	0.87	1.37	391	543
Nal-IRI	irinotecan	5	0.26	2.69	287	50.1	0.30	1.14
		10	0.25	2.73	854	101	0.20	1.11
		20	0.16	4.43	2519	197	0.14	1.33
	SN-38	5	0.34	3.31	41.1	3.68	2.07	12.7
		10	0.39	3.18	105	7.19	1.62	15.0
		20	0.32	3.55	335	21.1	1.02	13.3

For the free formulation, irinotecan showed highest tissue distribution to kidney, with a tissue-to-plasma partition coefficient (${\mathit{Kp}}_{,T}$) of 128.8, followed by liver (74.31), spleen (63.14), lung (31.73), and brain (2.627) (Table 4). For SN-38, tissue distribution was greatest in liver (25.91), followed by kidney (10.86), spleen (3.646), lung (3.061), and brain (0.669). In contrast, nal-IRI exhibited distinct tissue distribution characteristics. Partition coefficients were highest in liver (27.20) and spleen (13.71), consistent with the known reticuloendothelial system uptake of liposomal formulations, but substantially lower than the free formulation. Distribution to kidney (3.209), lung (0.917), and brain (0.152) was also markedly reduced compared to the free drug, reflecting the altered biodistribution profile of the liposomal formulation.

Table 4.

The estimated tissue-to-plasma partition coefficients (${\mathit{Kp}}_T$) from the whole-body PBPK model in comparison to the calculated values based on the observed PK profiles.

Compound	Organs	Parameter	Calculated#	Formulation
				Free Irinotecan		Nal-IRI
				Healthy mice	Tumor mice	Healthy mice	Tumor mice
Irinotecan	Brain	${\mathit{Kp}}_{{\mathit{Br}}_{\mathit{CPT}}}$	2.627	2.679 (7)
	Lung	${\mathit{Kp}}_{{\mathit{Lu}}_{\mathit{CPT}}}$	31.73	32.96 (9)
	Kidney	${\mathit{Kp}}_{{\mathit{Ki}}_{\mathit{CPT}}}$	128.8	158.9 (6)
	Spleen	${\mathit{Kp}}_{{\mathit{Sp}}_{\mathit{CPT}}}$	63.14	51.05 (7)
	Liver	${\mathit{Kp}}_{{\mathit{Li}}_{\mathit{CPT}}}$	74.31	57.74 (13)
	Rest	${\mathit{Kp}}_{{\mathit{Re}}_{\mathit{CPT}}}$	--	0.123 (33)	30.04 (9)	0.123 (F)	31.59 (57)
	Tumor	${\mathit{Kp}}_{{\mathit{Tu}}_{\mathit{CPT}}}$	--	--	18.51 (45)	--	38.56 (30)
SN-38	Brain	${\mathit{Kp}}_{{\mathit{Br}}_{\mathit{SN}38}}$	0.669	0.370 (13)
	Lung	${\mathit{Kp}}_{{\mathit{Lu}}_{\mathit{SN}38}}$	3.061	1.490 (9)
	Kidney	${\mathit{Kp}}_{{\mathit{Ki}}_{\mathit{SN}38}}$	10.86	11.21 (27)
	Spleen	${\mathit{Kp}}_{{\mathit{Sp}}_{\mathit{SN}38}}$	3.646	2.024 (9)
	Liver	${\mathit{Kp}}_{{\mathit{Li}}_{\mathit{SN}38}}$	25.91	454.5 (92)
	Rest	${\mathit{Kp}}_{{\mathit{Re}}_{\mathit{SN}38}}$	--	1.291(13)	25.81 (40)	1.291 (13)	68.83 (69)
	Tumor	${\mathit{Kp}}_{{\mathit{Tu}}_{\mathit{SN}38}}$	3.955	--	8.646 (18)	--	1.816 (26)
Nal-IRI	Brain	${\mathit{Kp}}_{{\mathit{Br}}_{\mathit{nalIRI}}}$	0.152	--	--	0.776 (13)
	Lung	${\mathit{Kp}}_{{\mathit{Lu}}_{\mathit{nalIRI}}}$	0.917	--	--	0.411 (21)
	Kidney	${\mathit{Kp}}_{{\mathit{Ki}}_{\mathit{nalIRI}}}$	3.209	--	--	8.218 (17)
	Spleen	${\mathit{Kp}}_{{\mathit{Sp}}_{\mathit{nalIRI}}}$	13.71	--	--	0.249 (7)
	Liver	${\mathit{Kp}}_{{\mathit{Li}}_{\mathit{nalIRI}}}$	27.20	--	--	5.554 (23)
	Rest	${\mathit{Kp}}_{{\mathit{Re}}_{\mathit{nalIRI}}}$	--	--	--	0.919·10−4 (F)	0.722 ·10−2 (F)
	Tumor	${\mathit{Kp}}_{{\mathit{Tu}}_{\mathit{nalIRI}}}$	1.635	--	--	--	1.661 (69)

“--” not applicable

^#calculated as ${\mathit{Kp}}_T=\frac{\mathit{Tissue}{\mathit{AUC}}_{0-\infty }}{\mathit{Plasma}{\mathit{AUC}}_{0-\infty }}$

F: fixed

Rest: Rest organ

3.2. Dual PBPK model of free irinotecan in healthy mice

The PBPK model (Figure 2A) adequately captured the plasma and tissue PK profiles of irinotecan and SN-38 in healthy BALB/c mice following IV administration at 5, 10, and 20 mg/kg (Figure 3A). GOF plots showed acceptable agreement between observed and predicted concentration, with most predictions within a ± 2-fold range and AAFE below 3 across plasma and tissue compartments (Figures S2A & S3A). Tissue distribution for both compounds was best described by a perfusion-limited model, supported by their physicochemical properties (Log P: 2.78 for irinotecan and 1.87 for SN-38) [15-17] and extensive distribution into highly perfused organs, such as the liver and kidney. Although a permeability-limited model was initially evaluated, given the potential influence of ABC transporter substrates, model fits were comparable, and sensitive analysis showed the membrane permeability (${\mathit{PS}}_T$) had minimal impact. This supports the use of a perfusion-limited approach, which also provided more precise and physiologically relevant PK parameter estimates (Tables 4-5). Estimated tissue partition coefficients (${\mathit{Kp}}_{T,\mathit{CPT}},{\mathit{Kp}}_{T,\mathit{SN}38}$) were consistent with observed tissue exposures (Table 4).

Table 5.

The estimated PK parameters from the whole-body PBPK models

				Free irinotecan		Nal-IRI
				Healthy mice	Tumor-bearing mice	Healthy mice	Tumor-bearing mice
Compound	Parameter	Units	Parameter description	Value (CV%)
Irinotecan	${\mathit{CL}}_{\mathit{CES},\mathit{PL}}$	mL/hr	CES-mediated metabolic clearance in plasma	63.56 (13)
	${\mathit{CL}}_{\mathit{CES},\mathit{Li}}$	mL/hr	CES-mediated metabolic clearance in liver	0.631 (24)
	${\mathit{CL}}_{R,\mathit{CPT}}$	mL/hr	Renal clearance of irinotecan	--	0.944 (39)	--	0.007 (F)
	${\mathit{CL}}_{\mathit{CES},\mathit{Tu}}$	mL/hr	CES-mediated metabolic clearance in tumor	--	0.101 (71)	--	0.204 (F)
	${\mathit{PC}}_{\mathit{out},\mathit{Li},\mathit{CPT}}$	hr−1	irinotecan exocytosis rate constant from phagocytes in liver	-		0.124 (53)
	${\mathit{CL}}_{\mathit{CES},\mathit{PC},\mathit{PL}}$	hr−1	phagocytic CES-mediated metabolism in plasma	--		3.700 (F)
	${\mathit{CL}}_{\mathit{CES},\mathit{PC},\mathit{Lu}}$	hr−1	phagocytic CES-mediated metabolism in lung	--		1.450 (F)
	${\mathit{CL}}_{\mathit{CES},\mathit{PC},\mathit{Li}}$	hr−1	phagocytic CES-mediated metabolism in liver	--		0.013 (54)
	${\mathit{CL}}_{\mathit{CES},\mathit{PC},\mathit{Sp}}$	hr−1	phagocytic CES-mediated metabolism in spleen	--		15.92 (F)
	${\mathit{CL}}_{\mathit{Bi},\mathit{CPT}}$	mL/hr	Biliary clearance of irinotecan	0.547 (15)
	$k_{\mathit{EHR},\mathit{CPT}}$	hr−1	Enterohepatic reabsorption rate constant for irinotecan	0.889 (21)
	$F_{\mathit{CPT}}$	-	Fraction of irinotecan eliminated into intestinal lumen	0.139 (F)
	$N_{\mathit{CPT}}$	-	Number of transit compartments for irinotecan EHC	42.00 (F)
	${\mathit{MTT}}_{\mathit{CPT}}$	hr	Mean transit time for EHC	0.074 (F)
	${\mathit{Time}}_{\mathit{start}}$	hr	Time at which EHC begins	3.367 (6)
	${\mathit{Time}}_{\mathit{end}}$	hr	Time at which EHC ends	7.828 (6)
SN-38	${\mathit{CL}}_{R,\mathit{SN}38}$	mL/hr	Renal clearance of SN-38	2.662 (63)			10.88 (37)
	${\mathit{CL}}_{\mathit{Bi},\mathit{SN}38}$	mL/hr	Biliary clearance of SN-38	9.371 (9)
	$k_{\mathit{EHR},\mathit{SN}38}$	hr−1	Enterohepatic reabsorption rate constant for SN-38	0.906 (89)
	$F_{\mathit{SN}38}$	-	Fraction of SN-38 eliminated into intestinal lumen	0.640 (F)
	$N_{\mathit{SN}38}$	-	Number of transit compartments for SN-38 EHC	9.000 (F)
	${\mathit{MTT}}_{\mathit{SN}38}$	hr	Mean transit time for EHC	0.041 (F)
	${\mathit{PC}}_{\mathit{out},\mathit{PL},\mathit{SN}38}$	hr−1	SN-38 exocytosis rate constant from phagocytes in plasma	--		4.609 (14)
	${\mathit{PC}}_{\mathit{out},\mathit{Lu},\mathit{SN}38}$	hr−1	SN-38 exocytosis rate constant from phagocytes in lung	--		13.68 (F)
	${\mathit{PC}}_{\mathit{out},\mathit{Sp},\mathit{SN}38}$	hr−1	SN-38 exocytosis rate constant from phagocytes in spleen	--		5.612 (18)
	${\mathit{PC}}_{\mathit{out},\mathit{Li},\mathit{SN}38}$	hr−1	SN-38 exocytosis rate constant from phagocytes in liver	--		0.689 (F)
Nal-IRI	${\mathit{PS}}_{\mathit{Lu}}$	mL/hr	Permeability coefficient of lung capillary membrane	--		0.216 (33)
	${\mathit{PS}}_{\mathit{Sp}}$	mL/hr	Permeability coefficient of spleen capillary membrane	--		0.867 (47)
	${\mathit{PS}}_{\mathit{Li}}$	mL/hr	Permeability coefficient of liver capillary membrane	--		0.476 (49)
	${\mathit{PS}}_{\mathit{Ki}}$	mL/hr	Permeability coefficient of kidney capillary membrane	--		0.022 (76)
	${\mathit{PS}}_{\mathit{Br}}$	mL/hr	Permeability coefficient of brain capillary membrane	--		0.008 (39)
	${\mathit{Vmax}}_{\mathit{up},\mathit{PL}}$	μg/hr	Maximum phagocytic uptake rate in plasma	--		134.5 (20)
	${\mathit{Km}}_{\mathit{up},\mathit{PL}}$	μg/mL	Michaelis-Menten constant for phagocytic uptake in plasma	--		94.75 (21)
	$k_{\mathit{up},\mathit{Lu}}$	hr−1	Phagocytic uptake rate constant in lung	--		2.236 (62)
	$k_{\mathit{up},\mathit{Sp}}$	hr−1	Phagocytic uptake rate constant in spleen	--		110.1 (59)
	$k_{\mathit{up},\mathit{Li}}$	hr−1	Phagocytic uptake rate constant in liver	--		8.234 (40)
	$k_{\mathit{rel},\mathit{PL}}$	hr−1	Liposomal release rate constant in phagocytic compartment in plasma	--		4.773 (F)
	$k_{\mathit{rel},\mathit{Lu}}$	hr−1	Liposomal release rate constant in phagocytic compartment in lung	--		1.558 (54)
	$k_{\mathit{rel},\mathit{Li}}$	hr−1	Liposomal release rate constant in liver	--		0.065 (28)
	$k_{\mathit{rel},\mathit{Sp}}$	hr−1	Liposomal release rate constant in spleen	--		0.055 (10)
	$k_{\mathit{rel},\mathit{Tu}}$	hr−1	Liposomal release rate constant in tumor	--		--	1.204 (61)

F: fixed

The model successfully reproduced the secondary peak observed ~4 hr post-dose, attributed to EHC. The estimated EHC duration (${\mathit{EHC}}_{\mathit{start}}$ = 3.267 hr, ${\mathit{EHC}}_{\mathit{end}}$ = 7.828 hr) aligned well with plasma and tissue concentration profiles. EHC-related parameters (e.g., $N$ and $MTT$) were initially estimated and then fixed to improve stability. The percentage of irinotecan (13.9%) and SN-38 (67.4%) excreted into the intestinal lumen closely matched clinical observations (13.4% and 67.5%, respectively) [10], improving prediction accuracy. Reabsorption rates of irinotecan and SN-38 were similar ($k_{\mathit{EHR},\mathit{SN}38}$ = 0.906 $mL∕hr$; $k_{\mathit{EHR},\mathit{CPT}}$ = 0.889 $mL∕hr$). Biliary clearance (${\mathit{CL}}_{\mathit{Bi},\mathit{CPT}}$ = 0.547 mL/hr; ${\mathit{CL}}_{\mathit{Bi},\mathit{SN}38}$ = 9.371 mL/hr) was the dominant elimination pathway over renal clearance (${\mathit{CL}}_{R,\mathit{SN}38}$ = 2.662 mL/hr), consistent with pre-clinical and clinical data [1, 18, 19]. Renal clearance of irinotecan (${\mathit{CL}}_{R,\mathit{CPT}}$) was initially included but estimated to be negligible and therefore excluded from the final model. The higher biliary clearance of SN-38 likely reflects both its glucuronidation to SN-38G and subsequent reverse metabolism via intestinal β-glucuronidase.

Notably, the enterohepatic reabsorption index of SN-38 (RCI = 0.28) was substantially higher than that of irinotecan (0.007), likely contributing to SN-38’s prolonged terminal half-life, enhanced intestinal reabsorption, CES-mediated intestinal metabolism. Although CES enzymes are more abundant in the liver [20, 21], model estimates showed greater CES-mediated metabolism in plasma (${\mathit{CL}}_{\mathit{CES},\mathit{PL}}$ = 32.96 mL/hr) than liver (${\mathit{CL}}_{\mathit{CES},\mathit{Li}}$ = 0.631 mL/hr). This suggests that significant plasma CES-induced hydrolysis of irinotecan occurs rapidly after IV administration, which may contribute to the high plasma SN-38 exposure associated with neutropenia, underscoring the importance of protective role of liposomal encapsulation to limit systemic exposure.

In summary, the developed PBPK model accurately characterized the plasma and tissue PK of irinotecan and SN-38, incorporating key ADME processes, such as CES-mediated metabolism and EHC, with parameter estimates demonstrating strong physiological and biological plausibility (Tables 4-5).

3.3. Dual PBPK Model of Nano-Liposomal Irinotecan in Healthy Mice

To extend the base model (Figure 2A), nano-liposomal compartments were incorporated to describe the disposition of nal-IRI in healthy mice (Figure 2B). PC compartments were included in macrophage-rich tissues (i.e., kupffer cells in the liver, splenic macrophages in the spleen, alveolar macrophages in the lungs) reflecting their established role in nanoparticle sequestration and clearance [22, 23]. Key processes governing nal-IRI interactions with phagocytic cells included phagocytic uptake ($k_{\mathit{up},T}$), liposomal release ($k_{\mathit{rel},T}$), CES-mediated metabolism within phagocytes (${\mathit{CL}}_{\mathit{CES},\mathit{PC},T}$), and exocytosis (${\mathit{PC}}_{\mathit{out},T}$) (Figure 2C).

Among tissues, the highest uptake ($k_{\mathit{up},T}$) was in the spleen (110.1 mL/hr), followed by the liver (8.234 mL/hr), and lung (2.236 mL/hr), consistent with relative macrophage abundance (Table 5). Saturable phagocytic uptake kinetics in plasma (${\mathit{Vmax}}_{\mathit{Up},\mathit{PL}}$ = 134.5 mL/hr and ${\mathit{Km}}_{\mathit{up},\mathit{PL}}$ = 94.75 μcg/mL) was required to capture the non-linear plasma PK of nal-IRI (Figure 3B), in agreement with experimental evidence of capacity-limited uptake of pegylated nanoparticles [24].

The estimated release rate constants ($k_{\mathit{rel}}$) indicated minimal irinotecan release in plasma, with greater release in extravascular spaces of the liver (0.065 $h{r}^{-1}$) and spleen (0.055 $h{r}^{-1}$), and more rapid release within phagocytic compartments of plasma (4.773 $h{r}^{-1}$) and lung (1.558 $h{r}^{-1}$). These values represent effective, system-level parameters integrating liposomal stability, intracellular processing, and tissue residence time. In vitro studies of nal-IRI report measurable irinotecan release under physiologically relevant conditions, corresponding to rate constants of 0.096 - 0.5 hr⁻¹ [25, 26]. However, in vivo release kinetics are expected to differ due to additional determinants, including lysosomal processing within phagocytes, exposure to degradative enzymes (e.g., phopholipases, esterases, and lipases), pH gradients, temeprature effects, and protein corona formation [27]. These factors provide biological plausibility for divergenece between in vitro-derived and in vivo-estimated release rates.

Following release, irinotecan was further metabolized to SN-38 predominantly within splenic, plasma, and alveolar macrophages prior to exocytosis. Notably, CES activity in splenic phagocytic cells played a key role in metabolic conversion (${\mathit{CL}}_{\mathit{CES},\mathit{PC},\mathit{Sp}}$ = 15.92 mL/hr) after nal-IRI administration. CES-mediated metabolism in plasma (${\mathit{CL}}_{\mathit{CES},\mathit{PL}}$ = 63.56 mL/hr) exceeded that in phagocytes (${\mathit{CL}}_{\mathit{CES},\mathit{PC},\mathit{PL}}$ = 3.7 mL/hr), underscoring the protective effect of liposomal encapsulation in limiting premature metabolic conversion and thereby prolonging systemic exposure of irinotecan.

Unlike free irinotecan, the nal-IRI PK profile could not be captured using a perfusion-limited model. Therefore, permeability-limited transport across capillary membranes was modeled using modified Fick’s law, with the permeability coefficient (${\mathit{PS}}_T$) introduced to account for convection and diffusion-driven transport. The term $\frac{{\mathit{PS}}_T}{1+{\mathit{PS}}_T}$ effectively described nal-IRI’s restricted capillary passage, and GSA confirmed ${\mathit{PS}}_T$ as one of the most influential parameters (Figure S4). The highest capillary permeability was estimated in the spleen (${\mathit{PS}}_{\mathit{Sp}}$ = 0.868 mL/hr), followed by the lung (${\mathit{PS}}_{\mathit{Lu}}$ = 0.216 mL/hr), liver (${\mathit{PS}}_{\mathit{Li}}$ = 0.476 mL/hr), kidney (${\mathit{PS}}_{\mathit{Ki}}$ = 0.022 mL/hr), and brain (${\mathit{PS}}_{\mathit{Br}}$ = 0.008 mL/hr). Despite higher permeability in the kidney than lung, nal-IRI uptake was greater in the lung, likely due to higher local phagocytic activity and convection-driven transport. These findings suggest that both pressure gradients (i.e., convection) and local immune cell uptake may modulate nal-IRI distribution.

Overall, incorporating both the nano-liposomal and PC compartments allowed the model to accurately predict plasma and tissue distribution for both irinotecan and SN-38 in healthy mice at 5, 10, and 20 mg/kg IV doses (Figure 3B). Predicted-versus-observed plots and AAFE metrics both indicated acceptable predictive accuracy across plasma nad tissue for both irinotecan and SN-38 (Figures S2B & S3B). The model parameter estimates of nal-IRI PBPK model reflected known dispositional mechanisms of liposomal formulations and tissue physiology (Tables 4-5).

3.4. Dual PBPK Model of Free and Nano-Liposomal Irinotecan in Tumor-bearing Mice.

Building upon the base PBPK models developed for irinotecan and nal-IRI in healthy mice, a tumor compartment, modeled as perfusion-limited, was added to characterize drug disposition in CHLA-136 xenograft tumor-bearing mice (Figure 2). All physiological and non-tumor tissue PK parameters from the healthy mice model were retained. Tumor-specific parameters, including tumor volume ($V_{\mathit{Tu}}$), blood flow ($Q_{\mathit{Tu}}$), partition coefficient (${\mathit{Kp}}_{\mathit{Tu}}$), liposomal release ($k_{\mathit{rel},\mathit{Tu}}$), and CES-mediated metabolism (${\mathit{CL}}_{\mathit{CES},\mathit{Tu}}$) in the tumor [28], were estimated to account for tumor-specific disposition. This extended model adequately depicted the plasma, tissue, and tumor PK of irinotecan and SN-38 following a 20 mg/kg IV dose of free-IRI and nal-IRI (Figure 4). GOF analyses demonstrated acceptable agreement for tissue and tumor concentration across formulations (Figures S2C-D and S3C-D), with only modest over-prediction in plasma.

Figure 4. Plasma and tissue PK profiles of irinotecan and SN-38 following administration of free irinotecan and nano-liposomal irinotecan in tumor-bearing mice.

Plasma and tissue concentration-time profiles of irinotecan and SN-38 in tumor-bearing mice (CHLA-136 neuroblastoma xenograft) following intravenous (IV) administration of either 20 mg/kg free irinotecan (free-IRI, light blue) or 20 mg/kg of nano-liposomal irinotecan (nal-IRI, orange). Observed data represents mean concentrations from three mice per time point (n = 3). Model-predicted concentrations are shown as solid lines. Shaded regions indicate 95% confidence intervals of model prediction for free irinotecan (sky blue) nano-liposomal irinotecan (light orange). Experimental data were obtained from previously published studies (Kang et al., 2015)

Assuming tissue density of 1g/mL, the estimated tumor blood flow (79.53 mL/hr) and volume (0.274 mL or 274 mm³) for CHLA-136 neuroblastoma xenografts aligned with physiological values for mouse tumor models. Studies using Doppler ultrasound have reported that normal tissue perfusion typically ranges from 8 to 10 mL/100g/min, whereas tumor perfusion is highly variable and often approaching or exceeding (~ 100 mL/100 g/min) [29-31]. The estimated tumor blood flow in this study (79.53 mL/hr/0.274 g = 4.84 mL/g/min) falls within this expected range and consistent with notion that tumor perfusion can be much higher than normal tissue perfusion, supporting the plausibility of the derived tumor perfusion parameter. The tumor volume (274 mm³) is consistent with the size at which treatment began around 232 - 364 mm³ in PD study [8] (Figure 5), and was thus used to capture tumor PK.

Figure 5. Antitumor activity of free irinotecan and nano-liposomal irinotecan in pediatric xenograft models.

(A) Tumor growth response in CHLA-136, CHLA-79, CHLA-258, CB-NJRM, and Rh30m xenograft models following two intravenous doses (Day 1 & Day 4) of free irinotecan (free-IRI, 10 mg/kg) or nano-liposomal irinotecan (nal-IRI, 10 mg/kg). Treatment was initiated when tumor reached ~150 mm³, and mice were euthanized once tumors exceeded 1500 mm³. Symbols represent observed data (tumor volume [mm³]) vs. time post-treatment [days]) (n = 3 mice per group), while solid lines indicate model-predicted tumor growth for control, free-IRI, and nal-IRI groups. Experimental data were obtained from previously published studies (Kang et al., 2015)

(B) Tumor SN-38 concentrations following 20 mg/kg nal-IRI or free-IRI dosing in mice, compared against the estimated ${\mathit{KC}}_{50,\mathit{SN}38}$ for each pediatric xenograft (CB-NJRm, CHLA-258, CHLA-136, CHLA-79m, and Rh30m). Duration above ${\mathit{KC}}_{50,\mathit{SN}38}$ was consistently longer for nal-IRI compared with free-IRI.

CES-mediated metabolism in the tumor (${\mathit{CL}}_{\mathit{CES},\mathit{Tu}}$) was identified as a key determinant of intra-tumoral SN-38 exposure and was included in both the free (0.101 $mL∕hr$) and nano-liposomal (0.204 $mL∕hr$) models. This parameter proved particularly important for describing nal-IRI tumor PK, consistent with findings from literature[13], which demonstrated that intra-tumoral CES activity strongly influeces SN-38 exposure and antitumor efficacy. Their analysis across 13 tumor xenograft models found nal-IRI is more sensitive to CES-mediated activation than irinotecan, contributing to its prolonged tumor retention and higher therapeutic effect.

Interestingly, in contrast to the healthy mice model, where renal clearance of irinotecan was negligible, the tumor-bearing mice model required the inclusion of renal elimination to accurately predict kidney PK and improve overall parameter precision. Renal clearance values for irinotecan were estimated at 0.944 mL/hr for free-IRI and 0.007 mL/hr for liposomal formulations. Similarly, SN-38 showed increased renal clearance following nal-IRI administration in tumor-bearing mice (${\mathit{CL}}_{R,\mathit{SN}38}$ = 13.76 mL/hr) compared to healthy mice and free-IRI-treated group (2.662 mL/hr). Considering that irinotecan primarily binds to albumin, these changes are consistent with reports of tumor-induced hypoalbuminemia [32], particularly in neuroblastoma models like CHLA-136, thereby enhancing the renal filtration of the unbound fraction of irinotecan.

3.5. Dual PBPK/PD Model of free-IRI & nal-IRI for antitumor efficacy

Following development of PBPK models in tumor-bearing mice, an integrated PBPK/PD model (Figure 2D) was constructed to quantitatively describe CHLA-136 tumor growth dynamics, characterize the exposure-response relationship, and compare the antitumor efficacy of free irinotecan versus nano-liposomal irinotecan (Figure 5A). GSA and model fitting indicated that therapeutic response was primarily driven by intra-tumoral SN-38 PK rather than irinotecan levels. This aligns with the known pharmacology of irinotecan, where SN-38, its active metabolite, is 100-1000 times more potent and mainly responsible for cytotoxicity [33]. Accordingly, drug potency (${\mathit{KC}}_{50,\mathit{SN}38}$) in the PD model was linked solely to SN-38, underscoring the importance of accurately capturing intra-tumoral CES-mediated conversion of irinotecan to SN-38 and SN-38 tumor disposition (Figure 4).

The differences in efficacy between formulations ($\alpha$) were explored across PD parameters, with the maximum killing rate showing the greatest distinction (Table 6). Compared with the free-IRI, prolonged intra-tumoral SN-38 exposure with nal-IRI resulted in a ~15-fold higher maximal killing rate ($\alpha$ = 14.87), leading to superior tumor suppression.

TABLE 6.

The estimated pharmacodynamic parameters

Parameters units	Description	Values (CV%)
		CHLA-136	CHLA-79	CHLA-258	CB-NJRM	Rh30m
$\mathit{kg}(da{y}^{\text{-}1})$	Exponential growth	0.065 (4)	0.112 (6)	0.050 (6)	0.097 (11)	0.038 (6)
${\omega }_0(m{m}^3)$	Initial tumor volume	399.9 (4)	327.9 (7)	313.3 (8)	260.1 (12)	336.7 (5)
${\mathit{K}}_{\mathit{max}}(da{y}^{\text{-}1})$	Maximum killing rate	0.129 (25)	0.092 (32)	0.084 (28)	0.067 (F)	0.120 (F)
${\mathit{KC}}_{50,\mathit{SN}38}$ (μg/mL)	Intra-tumoral SN-38 concentration at 50% $K_{max}$	0.167 (38)	0.169 (44)	0.018 (13)	0.238 (35)	0.598 (24)
$\gamma$	Hill coefficient	1.997 (20)	1.998 (25)	1.683 (F)	1.998 (25)	1.822 (24)
$\alpha$	Difference of maximal killing rate	15.12 (59)	21.76 (F)	9.365 (65)	33.46 (55)	29.15 (53)
${\mathit{k}}_1(da{y}^{\text{-}1})$	Transit rate constant for therapeutic delay	0.351 (12)	0.463 (29)	0.344 (6)	0.482 (58)	0.463 (F)
$\text{TSC}(\mu ∕mL)$	Calculated tumor static concentration	0.013	0.018	0.002	0.037	0.048

To further validate the PBPK/PD framework, the same model was applied to four additional pediatric tumor xenograft models (Figure 5B-E): CHLA-258 (Ewing’s sarcoma), Rh30m and CB-NJRm (rhabdomyosarcoma), and CHLA-79 (neuroblastoma). Assuming consistent tumor PK across models, PD parameters, including natural tumor growth rate ($k_g$), maximum killing rate ($K_{\mathit{max}}$), and drug potentcy (${\mathit{KC}}_{50,\mathit{SN}38}$), and a transit rate constant ($k_1$) accounting for the delayed onset of antitumor activity, were separately estimated for each tumor type (Table 6). CHLA-258 was ~10 folds more sensitive to irinotecan (${\mathit{KC}}_{50,\mathit{SN}38}$= 0.018 μg/mL), compared with CHLA-136, CHLA-79, and CB-NJRM (0.16-0.24 μg/mL), whereas Rh30m was the least sensitive (0.598 μg/mL). Maximum killing rates were similar among all cell lines for free-IRI (0.084 – 0.12 day⁻¹), while nal-IRI consistently produced 8-23 fold greater maximum killing rate across all pediatric tumors.

Enhanced tumor retention of nal-IRI, together with higher intra-tumoral CES activity, sustained SN-38 concentrations above the TSCs for longer than free irinotecan, supporting its greater maximal killing effect across pediatric cell lines (Figure 5F). In contrast, free irinotecan maintained SN-38 above the TSCs only transiently.

Overall, the PBPK/PD model successfully captured tumor volume dynamics in CHLA-136, CHLA-79m, CB-NJRM, and Rh30m with good precision and low parameter variability. However, challenges in parameter estimation for CHLA-258 suggest potential differences in intra-tumoral SN-38 disposition, highlighting the need for tumor-specific refinement in future modeling efforts.

3.6. Gloabal Sensitivity Analysis (GSA) and Model Validation

A Partial Rank Correlation Coefficient (PRCC)-based global sensitivity analysis was performed to identify the most influential parameters governing the disposition of irinotecan and nal-IRI in plasma and tissues. This analysis was also used to evaluate the directionality of parameter influence (positive and negative correlations with compartment outputs) and to eliminate insensitive parameters, thereby streamlining the model structure.

Key parameters identified as most sensitive included tissue partition coefficients (${\mathit{Kp}}_T$), permeability coefficients (${\mathit{PS}}_T$), CES-mediated metabolic rates (${\mathit{CL}}_{\mathit{CES}}$), and clearance mechanisms such as biliary and renal elimination. Additionally, parameters associated with nal-IRI’s interaction with phagocytic cells, namely, phagocytic uptake ($k_{\mathit{up}}$), liposomal release ($k_{\mathit{rel}}$), CES activity within phagocytic cells (${\mathit{CL}}_{\mathit{CES},\mathit{PC}}$), and exocytosis (${\mathit{PC}}_{\mathit{out}}$), were found to play critical roles in describing the PK of both free irinotecan and nal-IRI.

Among these, permeability coefficients (${\mathit{PS}}_T$) exerted a greater and more sustained influence on nal-IRI’s tissue distribution than partition coefficient ($\mathit{Kp}$) in the nal-IRI PBPK model, as assessed by time-dependent PRCC magnitude and duration above sensitivity threshold (Figure S4), indicating that a permeability-limited framework is required to accurately describe the tissue distribution of nal-IRI. This is mechanistically consistent with the transport behavior of nanoparticles, which do not readily diffuse across membranes like small molecules (e.g., irinotecan) but instead undergo slow and restricted movement through capillary membranes governed by convection and diffusion limitations.

Although several estimated PK parameters exhibited high uncertainty (CV% > 50%) or were fixed during estimation due to limited identifiability, PRCC-based global sensitivity analysis demonstrated that model outputs were predominantly governed by a subset of influential parameters, particularly PSs that regulate nal-IRI tissue and tumor distribution. In contrast, parameters with high CV% consistently showed low PRCC magnitudes across time and therefore were not critical determinants of nal-IRI disposition. Importantly, fixing these low-identifiability parameters did not materially affect key model predictions. To further assess robustness, additional PRCC sensitivity analyses were conducted specifically for fixed parameters, confirming that they exerted minimal or context-limited influence on model outputs (Figures S5). Collectively, these results indicate that neither parameter uncertainty nor parameter fixation compromises the mechanistic conclusions of the model.

3.7. Model Validation

Model predictive performance was further evaluated using independent pharmacokinetic data reported by Karla and Kim et al. for nal-IRI and free irinotecan in HT-29 xenograft–bearing NOD/SCID mice [13]. Plasma concentration–time profiles following intravenous administration were directly compared with model simulations at matching doses, while tumor exposure was evaluated using simulations at the experimental dose reported for tumor PK.

The PBPK model adequately reproduced the plasma and tumor PK of free irinotecan, with only minor underprediction during the terminal phase (Figure S6). In contrast, larger discrepancies were observed for nal-IRI, including higher systemic exposure and more prolonged intratumoral retention than predicted by the base model, which were reproduced only after adjusting key parameters (reduced plasma phagocytic uptake and slower intratumoral release). These deviations were mechanistically consistent with biological differences between experimental systems, particularly the altered macrophage function and impaired innate immunity in NOD/SCID mice, which are known to reduce liposomal clearance and prolong circulation time [34].

Overall, this external evaluation supports the robustness of the PBPK framework for free irinotecan and highlights formulation- and host-dependent factors that influence nal-IRI disposition, providing important context for interpreting model predictions across experimental settings.

3.8. Model Simulations: Role of liposomal encapsulation and MPS in off-target toxicity

Because observed irinotecan concentrations represent both liposome-encapsulated drug (protected from CES-mediated metabolism) and released drug (readily converted to SN-38), we used model simulations to investigate how liposomal encapsulation and MPS interactions influence SN-38 distribution in tumor and normal tissues. Simulations showed that encapsulation reduced initial irinotecan levels in plasma, tumor, and off-target tissues while extending half-life through sustained release (Figure S7A). This prolonged exposure enhanced efficacy of nal-IRI but increased intestinal SN-38 exposure via biliary excretion (Figure S7B), consistent with the greater risk of diarrhea observed with nal-IRI.

Modulating formulation parameters highlighted potential strategies to improve the therapeutic index. Increasing intra-tumoral release ($k_{\mathit{rel},\mathit{Tu}}$) decreased tumor exposure while lowering off-target SN-38 levels (Figure S8A). Reducing hepatic phagocytic uptake ($k_{\mathit{up},\mathit{Li}}$) prolonged circulation and increased delivery to both tumor and non-tumor tissues (Figure S8B), Our model further suggested that MPS-rich organs exhibited high initial SN-38 concentrations, reflecting accumulation within phagocytes.

Finally, combined modulations of intra-tumoral release and hepatic phagocytic uptake were assessed using a composite scoring system integrating three criteria. This analysis indicated that faster intra-tumoral release together with phagocytic uptake saturation maintained tumor SN-38 concentrations above ${\mathit{KC}}_{50,\mathit{SN}38}$, improved tumor-to-liver AUC ratios, and achieved higher overall scores, suggesting improved on-target efficacy with reduced toxicity (Figure 7B).

4. Discussion

4.1. Clinical Context and Rationale

Irinotecan is widely used in metastatic colorectal cancer (mCRC), typically within FOLFIRI or FOLFOXIRI regimens, but its use is limited by dose-limiting diarrhea and neutropenia linked to systemic and intestine SN-38 exposure[1, 2]. Nanoliposomal irinotecan was developed to overcome the pharmacokinetic limitations of free irinotecan by extending plasma circulation, reducing rapid clearance, and enhancing tumor delivery[3]. Clinical trials such as NAPOLI-1 and NAPOLI-3 confirmed its benefit in metastatic pancreatic ductal adenocarcinoma (mPDAC), leading to FDA approval[35, 36]. However, broader adoption remains cautious due to cost, lack of head-to-head comparisons with free irinotecan, and concerns about GI toxicity[37]. In this context, PBPK modeling offers a mechanistic and translational framework to compare free irinotecan and nal-IRI, characterize formulation-specific disposition, and anticipate off-target risks[38, 39].

4.2. Mechanistic Insights from PBPK/PD Models

Our PBPK/PD models quantitatively captured key determinants of irinotecan and nal-IRI disposition across plasma, tumor, and tissues, including EHC, CES-mediated metabolism, and phagocytic-driven nanoparticle uptake. Rebound peaks of irinotecan and SN-38 from EHC were consistent with patient data[40], underscoring its role in interpatient variability and SN-38-induced diarrhea [2], particularly in UGT1A1 polymorphic populations [41, 42]. The higher estimated partition coefficient for the lumped “rest-of-body” compartment (${\mathit{Kp}}_{\mathit{Re},\mathit{CPT}}$ and ${\mathit{Kp}}_{\mathit{Re},\mathit{SN}38}$) in tumor-bearing mice and enhanced renal elimination possibly reflect tumor-related systemic changes such as hypoalbuminemia, VEGF-driven vascular remodeling, and tumor-induced inflammation[43, 44]. Tumor-induced inflammation and hypoalbuminemia are known to increase the unbound fraction of drug and change vascular permeability, which can cumulatively increase apparent tissue-to-plasma partitioning across multiple non-tumor organs. Experimental study has demonstrated that hypoalbuminemia can substantially modify tissue distribution of albumin-bound compounds, leading to increased uptake in peripheral organs [45]. In addition, extra-hepatic tumors have been shown to include systemic inflammatory response, such as elevated cytokines (e.g., IL-6), which downregulate drug transporters and metabolic enzymes and alter drug disposition systemically [46]. Because multiple tissues (e.g., muscle, adipose, heart, bone, pancreas) are lumped into the rest-of-body compartment, systemic physiological changes may manifest as large apparent shift of ${\mathit{Kp}}_{\mathit{Re},\mathit{CPT}}$ and ${\mathit{Kp}}_{\mathit{Re},\mathit{SN}38}$.

4.3. Nanoparticle-Specific Disposition

Compared to small molecules, nanoparticle PK is substantially more complex and governed by permeability-limited, convection-driven transport and phagocyte-mediated clearance. Our PBPK model confirmed these features, aligning with prior studies that define permeability-limited models as the standard for nanoparticle distribution [47]. Chen et al. showed that perfusion-limited models fail to capture delayed nanoparticle equilibration, requiring time-dependent partitioning [39]. In our model, we applied constant $\mathit{Kp}$ values with permeability-limited transport and phagocyte-mediated uptake, similar to Li et al [48], and proved essential to capture nal-IRI PK in both healthy and tumor-bearing mice.

Sensitivity analysis revealed a minimal impact of liposomal drug release in plasma, consistent with reports that ~ 95% of irinotecan remains encapsulated during circulation[49]. Li et al. recently developed a dialysis-based in-vitro release test (IVRT) for Onivyde^® and demonstrated liposomal irinotecan release kinetics are influenced by environmental pH, with higher pH accelerating drug release. The observed positive correlation between organ-specific pH and the estimated tissue-specific release rate constants ($k_{\mathit{rel},T}$) further supports the physiological plausibility of our model and justifies incorporating tissue-specific-release kinetics. Despite the relatively acidic tumor microenvironment, the estimated intra-tumoral release rate ($k_{\mathit{rel},\mathit{Tu}}$) was higher than those in liver ($k_{\mathit{rel},\mathit{Li}}$) and spleen ($k_{\mathit{rel},\mathit{Sp}}$). This finding suggests that effective in-vivo release is not governed by pH alone but also influenced by tumor-specific microenvironmental factors that may enhance liposomal destabilization and accelerate drug release. For instance, ammonia, which is byproduct of tumor cell metabolism, has been shown to accelerate release of liposomal doxorubicin independent of pH effects [50]. Moreover, differential formulation stability under various stress conditions provides additional evidence supporting for tissue-specific drug release [26]. Distribution was restricted by organ-specific barriers such as the blood-brain barrier (BBB) and renal filtration, limiting penetration into brain and kidney tissues. Although pegylation is known to enhance brain uptake, pegylated liposomes cannot passively cross the BBB [51-53]. The non-linear plasma PK of nal-IRI was adequately captured by modeling saturable phagocytic uptake at higher nal-IRI levels, consistent with the low abundance of circulating phagocytes in plasma compared to those in the spleen, liver, and lung.

Metabolism was tissue-specific. CES-mediated conversion of irinotecan was greater in liver and plasma than in phagocytic cells, consistent with differences in CES isoforms. CES1, abundant in macrophages, has lower catalytic efficiency than CES2, which predominates in liver and plasma [54]. CES2 exhibits ~12-fold higher affinity and ~5-fold greater hydrolysis rate than CES1 for irinotecan [55], explaining why hepatic CES activity and exocytosis from phagocytes dominated irinotecan conversion. Our GSA further confirmed that exocytosis-driven SN-38 release occurred mainly in hepatic phagocytes, reinforcing the central role of liver in metabolic conversion relative to other tissues. Taken together, these findings indicate that the disposition of nanoparticles and their released drug is determined not only by transport barriers but also by tissue-specific metabolic capacity.

4.4. Role of Encapsulation and MPS in Toxicity

The simulations highlight the dual effect of liposomal encapsulation: improved tumor exposure through enhanced uptake and sustained release but also increased intestinal SN-38 exposure through biliary excretion (Figure S7B), consistent with the diarrhea associated with nal-IRI.

This finding is consistent with meta-analysis, which compared severe (grade ≥3) dose-limiting toxicity between two different formulations, showed that there is no significant difference of diarrhea when 180 mg/m² of free irinotecan and 50 mg/m² of nal-IRI are given to the patients, consistent with that clinical approved dosing regimen is lower in nal-IRI (70 mg/m²) compared to free irinotecan (180 mg/m²) against metastatic pancreatic ductal adenocarcinoma (mPDAC) [56].

MPS uptake was a key determinant of off-target SN-38 levels. Kupffer cells, the dominant phagocytes in the liver, act as clearance barriers that restrict nanoparticle access to hepatocytes [57]. Our analysis shows that blockade or saturation of this pathway, as shown by Zelepukin et al [58], enhances nanoparticle delivery systemically. Likewise, formulation strategies enabling rapid intra-tumoral release, such as redox-responsive nanocarriers [59], may increase efficacy while reducing systemic exposure. Together, these insights emphasize the importance of balancing tumor-selective release kinetics and MPS interactions to optimize nal-IRI’s therapeutic index.

4.5. Translational Implications

The tumor growth inhibition model demonstrated that intra-tumoral SN-38, not irinotecan, drives tumor cell killing, with SN-38 levels exceeding the KC₅₀ for both formations. Given 100 – 1000-fold greater potency of SN-38, irinotecan concentration remains below effective cytotoxic thresholds. Enhanced tumor retention of nal-IRI and higher intra-tumoral CES activity sustained SN-38 exposure above the TSC for a longer duration than free irinotecan, resulting in a greater maximum killing rate, and explaining its superior cytotoxicity in all pediatric tumor models. In contrast, free irinotecan exhibited faster systemic clearance and lower tumor accumulation, leading to a shorter duration of effective intratumoral SN-38.

These findings highlight how PBPK/PD-informed analyses can mechanistically link formulation-dependent PK to PD outcomes, providing a quantitative framework for evaluating exposure-response relationships in preclinical models. Although direct clinical extrapolation was not pursued in the current study due to the lack of validated clinical PK data and known species differences in CES activity, tumor physiology, and drug disposition, the present framework establishes a foundation for future translational efforts to guide rational dose selection and regimen optimization, particularly valuable in pediatric oncology where clinical trial data are limited.

More broadly, this PBPK/PD approach provides a mechanistic insight into how nano-liposomal formulation changes systemic and intra-tumoral SN-38 exposure, which has direct implication for clinical dose optimization and toxicity mitigation. Sensitivity of MPS-related parameters and release kinetics highlight potential opportunities for rational nanoparticle design strategy to enhance efficacy but reduce dose-limiting toxicity of nanoparticle-based therapy.

4.6. Study Limitations and Future Directions

This study has several limitations. First, many tissue-specific processes, including phagocytic uptake, exocytosis, and liposomal release, were parameterized using literature-derived or estimated values [13], due to the lack of direct experimental measurements. Direct quantification of nanoparticle trafficking and clearance remains technically challenging [60, 61], which is a key barrier to refining these mechanistic parameters. In addition, LC-MS measures total drug without distinguishing active (ionized) from inactive (unionized) forms, potentially overestimating effective exposure [62]. Despite these limitations, the PBPK models generated physiologically meaningful insights into formulation-specific behaviors.

The optimized IVRT enables detection of partial drug leakage and provides a quantitative method to distinguish encapsulated from released (free) irinotean [26]. Integrating IVRT-derived data into the PBPK model would allow parameterization based on experimentally measured liposomal stability, thereby improving mechanistic accuracy and enhancing translational confidence. Coupling optimized IVRT data with PBPK/PD modeling may offer a regulatory science framework to establish in vitro-in vivo correlation for liposomal irinotecan, potentially facilitating the development and approval of generic liposomal formulations. Future efforts should refine nanoparticle uptake and release parameters with experimental measurements, incorporate patient-level variability in genetics and tumor physiology, and expand to integrate clinical trial and toxicity data. Addressing these gaps will strengthen translational predictions and enable rational optimization of nal-IRI–based regimens across patient populations.

The heterogeniety in the tumor microenvironment represents a key source of uncertainty for PD extrapolation, as tumor perfusion, blood flow, and intra-tumoral CES activity were not directly measured. Although these parameters were estimated within physiologically plausible ranges, variability in tumor physiology may contribute to variability in predicted responses. Future studies incorporating tumor-specific perfusion, blood flow, and metabolic measurements will further strengthen PD predictions.

Another limitation is the absence of an explicit intestinal lumen compartment to quantify SN-38-mediated GI toxicity. Because luminal PK data for SN-38/SN-38G remain limited, biliary excretion into the gallbladder was used as a surrogate for intestinal exposure. While this framework does not directly predict diarrhear risk, it provides mechanistic insight into formulation-dependent intestinal drug input. Incorporation of intestinal and luminal PK data in future will enable direct assessment of GI toxicity.

5. Conclusion

This study developed mechanistic PBPK/PD models for free irinotecan and its nano-liposomal formulations, providing a quantitative framework to characterize their disposition and efficacy. The irinotecan model captured CES-mediated metabolism and EHC, while the nal-IRI model incorporated nanoparticle-specific mechanisms including permeability-limited transport, phagocytic uptake, and liposomal release. Together, these processes explained the distinct disposition of each formulation. Importantly, our analysis showed that therapeutic activity is primarily driven by SN-38 exposures rather than total irinotecan levels, and prolonged SN-38 exposure leads to greater cell killing efficiency, underscoring differences in tumor response between nal-IRI and free irinotecan.

This modeling framework represents a significant advancement in addressing the complex disposition of nanoparticle-based therapeutics. By linking mechanistic drug disposition with PD outcomes, the models offer physiologically meaningful insights into both tumor-selective delivery and off-target toxicity. Beyond preclinical evaluation, this modeling framework can support translational applications such as dose scaling, pediatric projections, and comparative analyses in clinical regimens (NALIRIFOX vs. FOLFIRINOX). Overall, these PBPK/PD models establish a robust foundation for model-informed drug development and optimization of nanoparticle-based chemotherapies.

Figure 6. Model simulations of SN-38 exposure in tumor and non-tumor tissues to evaluate efficacy and toxicity risk.

(A) Simulated SN-38 concentration-time profiles in tumor, liver, and cumulative amount excreted into intestinal lumen following intravenous administration of free irinotecan (free-IRI, purple), nano-liposomal irinotecan (nal-IRI, green), and a modified formulation (nal-IRI with increased intra-tumoral release rate [×4.0] and reduced hepatic phagocytic uptake [×0.60], pink dashed line) in CHLA-136 tumor-bearing mice. Solid lines represent model-simulated concentrations. Note: Y-axis scales vary between panels.

(B) Composite scoring heatmap evaluating formulation strategies. Top panel: composite scores (0–1) across varying intra-tumoral release rates and hepatic phagocytic uptake values. Higher scores reflect enhanced efficacy (tumor exposure) and reduced off-target toxicity or distribution. Lower panels: individual contributing metric: tumor-to-liver AUC ratio (left), duration of SN-38 concentration above KC₅₀ (middle), and plasma SN-38 AUC (right).

Highlights.

• Physiologically based pharmacokinetic models developed for free and nano-liposomal irinotecan

• Models Identified key drivers of disposition for free and nano-liposomal irinotecan

• Pharmacodynamic model describes tumor dynamic of free and nano-liposomal irinotecan

• Highlight optimal formulation strategy that enhances efficacy and reduces toxicity

• Modeling framework supports translation from preclinical to clinical space

Glossary

ADME

Absorption, distribution, metabolism, and excretion

AIC (Akaike Information Criterion)

Statistical criterion used for model selection, quantifying relative goodness-of-fit among the models.

AUC (Area Under the Curve)

Integral of concentration versus time, representing drug exposure

Black Box Warning

Mandatory FDA label indicating life-threatening or severe adverse effects

Camptosar^® (free-IRI)

Commercial formulation of non-liposomal (free) irinotecan.

C_max

Maximum observed concentration following drug administration.

${\mathit{KC}}_{50,\mathit{SN}38}$

Intra-tumoral SN-38 concentration that achieves 50% of maximal drug effect.

K_max

Maximum tumor cell killing rate achievable with drug.

Partition Coefficient

Tissue-to-plasma concentration ratio, describing extent of tissue distribution.

MPS (Mononuclear Phagocyte System)

Network of immune cells specializing in phagocytosis, primarily composed of monocytes, macrophages, and dendric cells found throughout the body in organs (e.g., Kupffer cells, splenic macrophages)

NCA (Non-Compartmental Analysis)

Model-independent method used to calculate pharmacokinetic parameters directly from observed measure drug concentration-time data, without assumption about specific physiological compartments.

Nal-IRI (Onivyde^®)

FDA-approved nano-liposomal irinotecan formulation

PBPK (Physiologically Based Pharmacokinetic Model)

Mechanistic model integrating physiology and drug-specific parameters to predict concentration–time profiles across tissues.

PBPK/PD Model

Integrated pharmacokinetic–pharmacodynamic model linking drug disposition with therapeutic response.

PD (Pharmacodynamics)

Relationship between drug exposure and observed biological effect.

PEGylation

Surface modification of liposomes with polyethylene glycol (PEG) to increase circulation time and reduce clearance.

Perfusion-Limited Distribution

Tissue distribution determined by tissue blood flow

Permeability-Limited Distribution

Tissue distribution determined by capillary permeability

Phagocytic Uptake

Rate of nanoparticle uptake by phagocytic cells such as Kupffer cells or splenic macrophages.

PRCC (Partial Rank Correlation Coefficient)

Sensitivity analysis method quantifying influence and identifying positive and negative correlation of parameters on model outputs.

SN-38

Active metabolite of irinotecan

TGI (Tumor Growth Inhibition Model)

Pharmacodynamic model linking tumor growth dynamics with drug exposure.

TSC (Tumor Static Concentration)

SN-38 concentration at which neither tumor cell grows nor shrinks, resulting in stable tumor size.

UGT1A1

Enzyme responsible for glucuronidation of SN-38 to its inactive metabolite SN-38G.

V_ss (Volume of Distribution at Steady State)

Apparent volume describing the extent of drug distribution at equilibrium.

t_1/2 (Half-Life)

Time required for drug concentration to be decreased by half.

Appendix A. Supplementary Material

The supplementary material includes the document entitled “supplemental information” (Word file).