Integrating Pharmacokinetic–Pharmacodynamic Modeling and Physiologically Based Pharmacokinetic Modeling to Optimize Human Dose Predictions for Plasmodium falciparum Malaria: a Chloroquine Case Study

ABSTRACT

The translation of a preclinical antimalarial drug development candidate to the clinical phases should be supported by rational human dose selection. A model-informed strategy based on preclinical data, which incorporates pharmacokinetic–pharmacodynamic (PK-PD) properties with physiologically based pharmacokinetic (PBPK) modeling, is proposed to optimally predict an efficacious human dose and dosage regimen for the treatment of Plasmodium falciparum malaria. The viability of this approach was explored using chloroquine, which has an extensive clinical history for malaria treatment. First, the PK-PD parameters and the PK-PD driver of efficacy for chloroquine were determined through a dose fractionation study in the P. falciparum-infected humanized mouse model. A PBPK model for chloroquine was then developed for predicting the drug’s PK profiles in a human population, from which the human PK parameters were determined. Lastly, the PK-PD parameters estimated in the P. falciparum-infected mouse model and the human PK parameters derived from the PBPK model were integrated to simulate the human dose-response relationships against P. falciparum, which subsequently allowed the determination of an optimized treatment. The predicted efficacious human dose and dosage regimen for chloroquine were comparable to those recommended clinically for the treatment of uncomplicated, drug-sensitive malaria, which provided supportive evidence for the proposed model-based approach to antimalarial human dose predictions.

INTRODUCTION

Malaria is an infectious disease that is transmitted to humans through the bite of a female Anopheles mosquito harboring the Plasmodium parasite. Plasmodium falciparum was responsible for 99% of malaria-related deaths on the African continent, which incidentally accounted for 96% of the estimated 619,000 malaria fatalities that occurred globally in 2021 (1). Evidence of resistance to the current first-line artemisinin-based combination therapies, in the Greater Mekong Subregion and Northern Uganda, highlights the urgent need for new therapies (2, 3).

To facilitate a successful transition and progression of a nominated late-stage preclinical candidate to the clinical phases of drug development, human dose and dosage regimens should be optimized based on the PK-PD properties of the lead compound. In antibiotic drug development, human dose selection is informed by three PK parameters that govern efficacy, namely, the concentration-dependent drivers, which are the peak concentration (C_max) and the exposure, as determined by the area under the concentration-time curve (AUC), and the time-dependent driver, which is the duration in which the levels at the site of action are maintained above a predefined minimum effective concentration (4, 5). This approach has recently been adopted in preclinical antimalarial drug development, where dose fractionation–response studies are performed, by varying the dose and dosage intervals, to elucidate the PK requirement for efficacy (6).

Improved translational methods for drug discovery have been established for investigating antimalarial efficacy. Of note is the immunodeficient NOD SCID IL-2Rγ^null (NSG) mouse model, which supports the engraftment of human erythrocytes and subsequent infection and asexual blood-stage replication of the clinically relevant species of malaria, P. falciparum, which would otherwise be unable to grow in standard rodent models, due to the host-specificity of Plasmodium (7, 8). Dose fractionation–response relationships can thus be evaluated directly against the human malaria infection, which could provide a better prediction of the clinical PK-PD relationships. Therefore, based on the PK parameter associated with antimalarial efficacy, the human dose can be optimized to achieve a target concentration, exposure, or sustained duration of action, as indicated respectively by the C_max, AUC, and time above the minimum parasiticidal concentration (MPC), which is defined as the lowest concentration that results in the maximum effect.

PBPK modeling incorporates the physicochemical and pharmacokinetic properties of a drug with physiology to simulate its disposition in various tissues and organs of the whole body. This tool is especially useful for preclinical drug development as PBPK models can be built using in vitro absorption, distribution, metabolism, and elimination (ADME) data, thereby allowing human PK profiles to be predicted without the need for clinical data.

Herein we report a model-based strategy that can be used to rationally predict the efficacious human dose and dosage regimen for the treatment of P. falciparum malaria by incorporating PK-PD relationships with PBPK modeling. The well-established antimalarial chloroquine was chosen to investigate the suitability of our approach since there is clinical data available against which findings can be compared. The objectives of our study were to first determine the PK-PD parameters of chloroquine after dose fractionation in the P. falciparum-infected NSG mouse model, following which the PK-PD driver of efficacy for chloroquine and target magnitude of the PK parameter to which the minimum preliminary human dose was selected to achieve were identified. A PBPK model for chloroquine was then developed in a healthy human population, and PK profiles were simulated at various doses to determine the dose of chloroquine that would provide the target magnitude of the PK parameter driving efficacy. Lastly, the antimalarial effect of the minimum preliminary human dose was simulated and optimized by integrating the PBPK model-derived human PK parameters and the NSG PK-PD parameters of chloroquine against P. falciparum, from which an efficacious dose and dosage regimen were predicted.

RESULTS

Pharmacokinetic–pharmacodynamic modeling of experimental mouse data.

(i) Pharmacokinetic model evaluation. The diagnostic plots that were used to evaluate the final PK model displayed adequate agreement between the observed and the model-predicted chloroquine and the metabolite desethylcholoroquine concentration values, as shown in Fig. S1.1 and S1.2 in the supplemental material. Additionally, the PK parameters were estimated with reasonable precision, as indicated by the relative standard error (R.S.E.) values shown in Table 1.

TABLE 1.

Estimated population whole-blood PK parameters for chloroquine and desethylchloroquine following fractionated oral administrations of 12, 10, 8, and 6 mg/kg of chloroquine to NSG mice infected with P. falciparum^a

Model parameter	Value	R.S.E. (%)
PK parameter
ka (1/h)	1.19	22.1
Vc (L/kg)	0.41	23.4
Clp (L/h/kg)	0.59	47.3
Qp (L/h/kg)	104.7	49.1
Vp (L/kg)	21.8	3.8
Clm (L/h/kg)	3.8	9.57
Kpm (1/h)	4.81	17.5
Interindividual variability (SD)
ωka	0.39	18.5
ωVc	0.11	50.1
ωClp	1.12	35.6
ωClm	0.14	40.7
Error model
a1	0.024	8.67
b1	0.11	12.7
a2	0.013	8.36
b2	0.12	12.8

^aKa, absorption rate constant of parent; V_c, central compartment volume of distribution of parent; V_p, peripheral compartment volume of distribution of parent; Q_p, intercompartmental clearance of parent; Cl_p, clearance of parent; Cl_m, clearance of metabolite; K_pm, transformation rate constant of parent to metabolite; R.S.E., relative standard error.

(ii) Pharmacokinetic parameters. The estimated population PK parameters of the total whole-blood concentrations of chloroquine and desethylchloroquine in malaria-infected NSG mice are presented in Table 1. Chloroquine displayed a low clearance of 0.59 L/h/kg, which differed from its metabolite, which displayed a higher clearance of 3.8 L/h/kg. Chloroquine displayed a larger distribution in the peripheral compartment of 21.8 L/kg, compared to the central compartment, which displayed a volume of distribution of 0.41 L/kg.

(iii) Pharmacokinetic–pharmacodynamic model evaluation. The final direct-effect PK-PD model displayed an adequate fit to the observed and model-predicted PK-PD data for chloroquine against P. falciparum-infected NSG mice, as shown by the diagnostic plots in Fig. S1.3. The model-predicted PK-PD parameters were estimated with good precision, as demonstrated by their low R.S.E. values, as shown in Table 2.

TABLE 2.

Estimated population PK-PD parameters for chloroquine in NSG mice infected with Pf3D7^0087/N9 malaria^a

Model parameter	Value	R.S.E. (%)
PK-PD parameter
Kgrowth (1/h)	0.038	1.75
Kkill (1/h)	0.092	1.06
IC50 (μmol/L)	0.3	13.2
H	5 (fixed)
MPCb (μmol/L)	0.54
Interindividual variability (SD)
ωIC50	0.087	32.8
ωH	2.97	17.0
Error model
b3	0.3	9.13

^aK_growth, parasite growth rate constant; K_kill, parasite kill rate constant; IC₅₀, inhibitory concentration required to produce the half maximal effect; H, Hill coefficient; MPC, minimum parasiticidal concentration; R.S.E., relative standard error.

^bMPC was defined as the EC₉₅ and was calculated from the IC₅₀ and H.

(iv) Pharmacodynamic parameters. The predicted population PK-PD parameters for chloroquine in the P. falciparum-infected NSG mouse model are displayed in Table 2. The in vivo IC₅₀ and MPC for chloroquine against Pf3D7 were 0.3 and 0.54 μmol/L, respectively.

(v) Pharmacokinetic–pharmacodynamic driver of antimalarial efficacy. The PK-PD parameter driving the antimalarial activity of chloroquine in the P. falciparum-infected NSG mouse model was determined by fitting each PK-PD index, C_max/MPC, AUC_0–96/MPC, and percentage time above MPC, against the antimalarial response of each dose-fractionated treatment group on day 7 postinfection, as shown in Fig. 1. A nonlinear regression analysis could not be confidently performed for the PK-PD indices AUC_0–96/MPC and percentage time above MPC, with R² values of 0.26 and 0.37, respectively. The C_max/MPC index displayed the highest correlation with the reduction in parasitemia with an R² value of 0.71. This indicates that chloroquine exhibited concentration-dependent C_max/MPC killing in the P. falciparum-infected NSG mouse model.

FIG 1.

Relationship between the PK-PD indices C_max/MPC (a), AUC_0-96/MPC (b), and percentage time above MPC (c) and the corresponding antimalarial activity of chloroquine after a dose fractionation in P. falciparum-infected NSG mice. C_max, peak concentration; MPC, minimum parasiticidal concentration; AUC_0-96, area under concentration-time curve from 0 to 96 h; R², coefficient of determination.

The nonlinear regression model parameters used to describe the C_max/MPC relationship with the antimalarial activity of chloroquine are displayed in Table 3. The EC₅₀ and EC₉₅ values associated with the C_max/MPC index were 0.60 and 1.1 μmol/L, respectively, and the absolute C_max at the EC₉₅ was 0.59 μmol/L, which corresponded to the target C_max to which the minimum preliminary human dose was selected to attain.

TABLE 3.

Nonlinear regression model parameters for the C_max/MPC PK-PD index of chloroquine efficacy in the P. falciparum-infected NSG mouse model^a

Model parameter	Value (95% confidence interval)
Minimum response	0b
Maximum response	94.8 (78.9 to 110.8)
Hillslope	4.9 (0.6 to 9.2)
EC50c (μmol/L)	0.60 (0.50 to 0.72)
EC95c (μmol/L)	1.1 (0.64 to 1.9)
R2	0.71
Absolute Cmax at the EC95 (μmol/L)	0.59

^aEC₅₀, effective concentration required to produce the half maximal effect; EC₉₅, effective concentration required to produce 95% of the maximal effect; R², coefficient of determination.

^bThe bottom response value was constrained to a constant value of 0.

^cThe EC₅₀ and EC₉₅ values are equivalent to the C_max/MPC concentration.

Physiologically based pharmacokinetic model for chloroquine in humans.

The final PBPK model-predicted and clinically observed whole-blood concentration-time profiles of chloroquine after a single 300 mg oral administration to a healthy human population is displayed in Fig. 2. The observed concentration-time data displayed a reasonable fit within the 5th and 95th percentiles of the PBPK model-simulated chloroquine PK profile.

FIG 2.

Whole-blood concentration-time profiles of the clinically observed mean (red dots, n = 6 individuals), PBPK model-simulated (n = 100 individuals) median (blue line), and 5th and 95th percentile (blue shaded area) of chloroquine following a single 300 mg oral administration to healthy humans.

The PBPK model-estimated and clinically observed PK parameters for chloroquine are displayed in Table 4. The mean observed literature values for the C_max, T_½, and AUC_0-192 were all within the 5th to 95th percentile of the PBPK model-predicted values, except for the T_max, which was underestimated by the PBPK model, as shown by the predicted values, which ranged from 1.55 to 2.25 h compared to the observed value, which was 3.8 ± 0.74 h.

TABLE 4.

PBPK model estimated and clinically observed PK parameters for chloroquine after a single 300 mg oral administration to healthy individuals^a

PK parameter	Median (range) PBPK model predicted valueb	Mean (SE) clinically observed valuec (13)
Cmax (μmol/L)	1.14 (0.92 to 1.49)	0.996 (0.054)
Tmax (h)	1.95 (1.55 to 2.25)	3.8 (0.74)
T½ (h)	116.11 (65.48 to 242.09)	140 (10.5)
AUC0-192 (μmol.h/L)	41.51 (29.11 to 60.70)	50.97 (3)

^aC_max, peak concentration; T_max, time taken to reach peak concentration; T_½, half-life; AUC_0-192, area under concentration-time curve from 0 to 192 h; SE, standard error.

^bn = 100 individuals.

^cn = 6 individuals.

Pharmacokinetic modeling of simulated human data.

The PBPK model-simulated whole-blood concentration-time profiles for chloroquine were used to determine the primary PK parameters of chloroquine in healthy humans using nonlinear mixed-effects modeling. The PK model-predicted and PBPK model-simulated whole-blood concentration-time data for chloroquine were within reasonable agreement, as shown by the diagnostic plots in Fig. S3.1. The estimated population whole-blood PK parameters for chloroquine, as displayed in Table S3.1, were later used for the antimalarial dose-response simulations in humans.

Preliminary human dose prediction.

Concentration-time profiles for chloroquine were simulated using the PBPK model for a range of doses to determine the dose that provided the absolute concentration at the EC₉₅ of the PK-PD index associated with chloroquine activity, which was the C_max/MPC. From a linear regression analysis of the dose versus the corresponding C_max values, as shown in Fig. S2.1, it was determined that a dose of 150 mg was required to reach the target C_max of 0.59 μmol/L in humans.

Efficacious dose and dosage regimen optimization.

The population PK parameters for chloroquine in humans were linked to the population PK-PD parameters for chloroquine in the P. falciparum-infected NSG mouse model to simulate the antimalarial effect at various oral doses and dosage regimens of the drug in humans. The total treatment period was set at 48 h to cover at least one P. falciparum asexual life cycle, and the parasitemia levels were monitored for 144 h, to allow passage of three parasite life cycles. The first set of simulations investigated the dose-response for the first 24 h after chloroquine treatment, as displayed in Fig. 3. The first regimen that was simulated was the minimum preliminary human dose of a single administration of 150 mg of chloroquine, which was shown to be ineffective at suppressing parasite proliferation. This was attributed to the drug not having reached steady-state concentration, which necessitated further dose optimizations to determine the regimen that produces an efficacious steady-state profile. The PBPK model predicted a long half-life for chloroquine between 65.48 and 242.09 h, which suggested that a loading dose will be needed to achieve the steady-state concentration. A second regimen was simulated using a loading dose of 300 mg, and a third regimen was simulated using a loading dose of 600 mg, which was two and four times the preliminary dose, respectively. Although the loading dose of 300 mg was subtherapeutic, the loading dose of 600 mg resulted in a steady decline in parasitemia up until approximately 10 h, after which the parasite levels increased, which corresponded to decreasing chloroquine concentrations. To sustain the steady-state concentration of regimen 3, a maintenance dose was introduced in a fourth simulation, where a second dose of half the amount of the loading dose was administered at 8 h. Since the PBPK model predicted a T_max value for chloroquine of 1.95 h, the loading dose was administered 2 h before the parasite clearance began to plateau.

FIG 3.

Chloroquine human dose-response simulations, following oral administrations of regimen 1 (150 mg at 0 h, red line), regimen 2 (300 mg at 0 h, blue line), regimen 3 (600 mg at 0 h, green line), and regimen 4 (600 mg at 0 h, 300 mg at 8 h, purple line), using an integrated PK-PD model of the population PK parameters in humans and population PK-PD parameters in P. falciparum-infected NSG mice.

The second set of simulations investigated the dose-response for 144 h after chloroquine treatment, as displayed in Fig. 4. Regimen 4, which used a loading dose of 600 mg and maintenance dose of 300 mg at 8 h, displayed favorable efficacy, albeit parasites were not effectively cleared at 24 h. Therefore, a second maintenance dose was introduced at 24 h in regimen 5, which resulted in a sustained steady-state concentration and continued parasite clearance, as shown in Fig. 4. At approximately 48 h, the chloroquine concentration begins to fall below steady state, and since residual parasites were present at this time, a final maintenance dose of 300 mg was added in regimen 6, which now covered a total treatment period of 48 h, which is equivalent to one parasite life cycle. However, after treatment with regimen 6, there was evidence of recrudescence, as displayed by a plateau in the clearance of persisting parasites. Therefore, to maintain the 48 h treatment period, the initial loading and maintenance dose amounts were revisited.

FIG 4.

Chloroquine human dose-response simulations, following oral administrations of regimen 4 (600 mg at 0 h, 300 mg at 8 h, red line), regimen 5 (600 mg at 0 h, 300 mg at 8 h and 24 h, blue line), and regimen 6 (600 mg at 0 h, 300 mg at 8 h, 24 h, and 48 h, green line), using an integrated PK-PD model of the population PK parameters in humans and population PK-PD parameters in P. falciparum-infected NSG mice.

The third set of dose-response simulations for chloroquine describes parasite clearance at different loading and maintenance doses over a 144 h monitoring period, as presented in Fig. 5. For these simulations, the dosing intervals were kept at 0, 8, 24, and 48 h, and the ratio of the loading dose to the maintenance dose was kept at 2:1, since this regimen produced a steady-state concentration-time profile. Regimens 7, 8, and 9 were set to a loading dose of 700, 800, and 900 mg, respectively, and the corresponding maintenance doses were adjusted to half the amount of the loading dose. As shown in Fig. 5, the loading doses of 700, 800, and 900 mg and their respective maintenance doses were predicted to be effective since these treatments resulted in a favorable steady parasite clearance, with parasitemia levels at approximately 0.002% and below at 144 h, compared to the 600 mg loading-dose treatment, which appeared to show treatment failure by 144 h. The efficacy of the 800 and 900 mg loading dose regimens were similar; therefore, the proposed predicted efficacious dosage regimen for the treatment of P. falciparum in humans was 700 to 800 mg at 0 h, followed by 350 to 400 mg each at 8, 24, and 48 h.

FIG 5.

Chloroquine human dose-response simulations, following oral administrations of regimen 6 (600 mg at 0 h, 300 mg at 8 h, 24 h, and 48 h, red line), regimen 7 (700 mg at 0 h, 350 mg at 8 h, 24 h, and 48 h, blue line), regimen 8 (800 mg at 0 h, 400 mg at 8 h, 24 h, and 48 h, green line), and regimen 9 (900 mg at 0 h, 450 mg at 8 h, 24 h, and 48 h, purple line), using an integrated PK-PD model of the population PK parameters in humans and population PK-PD parameters in P. falciparum-infected NSG mice.

DISCUSSION

This study aimed to outline a model-based approach to rational antimalarial human dose and dosage regimen predictions using preclinical data of the investigational compound generated in the P. falciparum-infected NSG mouse model and PBPK modeling.

First, dose fractionation data revealed the PK-PD determinant for chloroquine efficacy in the P. falciparum-infected NSG mouse model was C_max/MPC. This can be attributed to the primary mechanism of action of chloroquine, which involves inhibition of the heme detoxification pathway. During its life cycle within the erythrocyte, P. falciparum degrades host hemoglobin to produce amino acids, which get incorporated into proteins synthesized by the parasite. A by-product of this reaction is the release of reactive heme, which is subsequently detoxified by the parasite through the generation of hemozoin, an inert crystal composed of heme dimers. Chloroquine acts by inhibiting the hemozoin formation pathway, which results in a fatal build-up of haem within the parasite. The in vitro cellular haem fractionation assay demonstrated that chloroquine causes a dose-dependent increase in free haem, which subsequently corresponds to decreased parasite survival, where at higher concentrations of chloroquine, when more drug is available to inhibit hemozoin formation, a larger amount of haem is present, which results in greater parasite death (9 to 11). Although these studies were performed in an in vitro setting where the chloroquine concentrations are constant, these observations can still be useful in postulating why the apparent PK-PD driver of efficacy for chloroquine was the concentration-dependent parameter C_max/MPC.

For a preclinical development candidate, PBPK models cannot be validated against clinical data; therefore, it is critical to generate precise data in the appropriate in vitro assay. For the purposes of this study, the PBPK model for chloroquine was validated against published clinical data to provide information on the requirements for building an accurately predictive in vivo PK model using in vitro data. Although they are compound specific, some of the essential in vitro parameters that influence in vivo drug disposition, which should be included in the PBPK model, are aqueous or biorelevant solubility, permeability, lipophilicity, intrinsic clearance, plasma protein binding, blood-to-plasma partitioning ratio, and the molecule’s physicochemical properties, such as the molecular weight and pKa. Correlation between in vitro ADME data and preclinical in vivo PK data should be performed, and evident disconnects should be understood and accounted for during PBPK model development.

The predicted efficacious dosage regimen for the treatment of P. falciparum in humans was 700 to 800 mg at 0 h, followed by 350 to 400 mg each at 8, 24, and 48 h. A loading dose and subsequent maintenance doses were needed to achieve steady-state concentration due to the predicted long half-life of chloroquine. The direct-effect PK-PD model used for the dose-response simulations in humans did not consider the different phases of the asexual intraerythrocytic parasite life cycle. Chloroquine has been shown to act predominately during the trophozoite stage of the parasite life cycle, when host hemoglobin degradation is at its peak; therefore, since a malaria infection is asynchronous, the total treatment period and therapeutic steady-state concentration were maintained for at least 48 h to ensure that all unsusceptible parasite forms undergo one life cycle where they will transition to the trophozoite phase and subsequently become sensitive toward chloroquine (11, 12). Since chloroquine and desethylchloroquine have similar in vitro activities against P. falciparum, the observed differences in the C_max values related to efficacy in the P. falciparum NSG mouse model and humans can be attributed, in part, to a contribution of the active major metabolite to the overall antimalarial activity of chloroquine exhibited in the NSG mouse model, which was not accounted for during the human dose-response simulations (13).

During the preclinical stages of drug development, extensive safety profiling is performed to highlight any potential toxicity liabilities of a molecule, which is used to inform the therapeutic window during dose selection. Retrospectively, chloroquine has been associated with cardiovascular effects related to human ether-a-go-go-related gene (hERG) channel inhibition, and has displayed an in vitro IC₅₀ against the hERG channel of 2.5 μmol/L in the embryonic kidney cell (HEK293) system (14, 15). From the chloroquine dose-response simulations in humans, the highest whole-blood C_max that was achieved at the loading dose of 800 mg was approximately 2.9 μmol/L (0.94 mg/L), which is estimated to equal a free plasma concentration of 0.331 μmol/L, given a whole-blood-to-plasma ratio of 3.5 and a fraction unbound in plasma of 0.4 (16, 17). Therefore, the free chloroquine plasma concentration reached at the highest dose of 800 mg was approximately 7-fold lower than the in vitro IC₅₀ value for hERG inhibition, which suggests that the predicted efficacious dosage range is safe and not within the boundary for potential cardiotoxic events.

The dosage regimen of chloroquine that is recommended clinically for the treatment of uncomplicated, drug-sensitive malaria is 600 mg at 0 h, followed by 300 mg each at 6, 24, and 48 h, which compares well to the predicted dose of 700 to 800 mg at 0 h, followed by 350 to 400 mg each at 8, 24, and 48 h (18). This demonstrates the suitability of the PK-PD and PBPK model-based approach, as outlined in Fig. 6, to predict an efficacious dose and dosage regimen for the treatment of P. falciparum malaria in humans.

FIG 6.

Proposed workflow for a model-based approach to antimalarial human dose predictions.

Although there will be translational differences between immunocompromised mice and humans with a functioning immune response, the proposed PK-PD strategy provides a data-driven point of departure to rationally guide clinical trial design for first-in-human studies for preclinical drug development candidates.

MATERIALS AND METHODS

In vivo efficacy and pharmacokinetics in malaria-infected humanized mice.

(i) Ethics. Animal studies were conducted at the Holistic Drug Discovery and Development (H3D) Centre Animal Research Facility, University of Cape Town (UCT). Ethical approval was granted by the UCT Animal Ethics Committee prior to study commencement (ethics approval reference number 021/015), and all procedures were performed in accordance with UCT’s animal ethics policies. Food and water were supplied ad libitum before and during the study.

(ii) Falciparum infection of humanized mice. The antimalarial activity of chloroquine was determined in the P. falciparum-infected NSG mouse model, in 6- to 10-week-old, male NSG mice, weighing between 25 and 35 g, using methods previously described (7, 8). Briefly, 28 NSG mice were intravenously engrafted daily with prepared human erythrocytes for 10 days, then the mice were intravenously injected in the tail vein with 2 × 10⁷ asynchronous Pf3D7^0087/N9-infected human erythrocytes (day 0). Pf3D7^0087/N9 is a chloroquine-sensitive strain that was developed and selected for infection in NSG mice at GlaxoSmithKline, Tres Cantos, Spain. The infection was left to establish for 3 days before commencement of treatment on day 3. The percentage of human erythrocytes was maintained above 50% with daily engraftments until the experimental endpoint on day 7 after infection.

(iii) Chloroquine administration and blood sampling. Chloroquine diphosphate was purchased from Sigma-Aldrich. The total dose of chloroquine base was administered as either 12, 10, 8, or 6 mg/kg over a consecutive 4-day treatment period, starting on the third day after infection with P. falciparum. Each total dose was given as either one dose on the third day, as two equally divided doses on the third and fifth day, or as four equally divided doses on days three to six. Chloroquine diphosphate was formulated in sterile phosphate buffered solution prior to treatment and administered by oral gavage, and the negative-control group was orally administered with sterile phosphate buffered solution only. Two mice were randomly assigned to each treatment group. Whole-blood PK and efficacy samples were collected via tail vein bleeding into lithium heparin-coated tubes. PK blood samples were collected for each chloroquine dosage group at 0.5, 1, 3, 5, and 24 h after the first administration on day 3, and at 0.5, 1, 3, and 24 h after administration on days 4, 5, and 6. PK samples were stored at −20°C until bioanalysis. Efficacy blood samples were collected before treatment for all experimental groups on days 3, 4, 5, 6, and 7. These samples were processed immediately after collection, and the percentage of infected human erythrocytes, or the parasitemia, and the percentage engraftment measurements were determined by fluorescence-activated cell sorting using an Accuri C6 Plus flow cytometer and FlowJo 10.8 software (Becton, Dickinson and Company), as previously described (19).

(iv) Quantification of chloroquine in whole blood. Detection of chloroquine and its major metabolite, desethylchloroquine, in whole blood was performed using an API 5500 triple quadrupole mass spectrometer (MS/MS) (SCIEX) coupled to an Agilent 1260 Infinity II high-performance liquid chromatography (HPLC) system (Agilent Technologies). Chloroquine, desethylchloroquine, and their respective deuterated internal standards were monitored at unit resolution in the positive multiple reaction monitoring mode (see Table S1.1 for MS/MS analytical parameters). Reversed-phase chromatographic separation of the analytes was achieved using a Waters Atlantis T3 3-μm (2.1 × 50 mm) analytical column, which was maintained at 30°C. The mobile phase was delivered at a constant flow rate of 0.45 mL/min, and a gradient elution method was used with 0.2% (vol/vol) formic acid in water and 0.2% (vol/vol) formic acid in methanol. Quantification of the analytes in the PK study samples was achieved using a calibration curve, which was prepared by parallel spiking 4% (vol/vol) acetonitrile working solutions, of pooled chloroquine and desethylchloroquine, into PK matrix-matched NSG mouse whole blood, to cover the concentration range from 2 to 1,000 ng/mL. Calibration standard and quality control samples were prepared in duplicate and processed concurrently with the experimental PK samples, to ensure acceptable accuracy and precision of the bioanalytical method. The analytes were isolated from 10 μL whole blood using protein precipitation with 75 μL 0.2% (vol/vol) formic acid in methanol, which was spiked with the deuterated internal standards chloroquine-d4 and desethylchloroquine-d4 for chloroquine and desethylchloroquine, respectively. Blood samples were vortexed for 1 min and then centrifuged at 5,590 × g for 5 min. Four microliters of the resulting supernatant was injected onto the analytical column for HPLC-MS/MS analysis. The unknown total concentrations of chloroquine and desethylchloroquine in whole blood were determined using Analyst 1.6.3 software (SCIEX).

(v) Pharmacokinetic–pharmacodynamic modeling. Nonlinear mixed effects modeling, in Monolix 2021R1 software (Lixoft), was used to develop a sequential PK-PD model for the total whole-blood concentration-time data for chloroquine after dose fractionation in the P. falciparum-infected NSG mouse model. Concentration values that were below the lower limit of quantification (LLOQ) of 2 ng/mL were censored in the PK analysis. The model parameters were estimated using stochastic approximation expectation maximization algorithm (20). A two-compartment model with first-order absorption and elimination and a one-compartment model with first-order absorption and elimination were used to describe the oral PK of chloroquine and desethylchloroquine, respectively. Chloroquine and desethylchloroquine PK were modeled jointly using a series of differential equations (equation 1):

$$\frac{\mathit{dAd}}{\mathit{dt}}\hspace{0.17em}=\hspace{0.17em}-\mathit{ka}\hspace{0.17em}*\hspace{0.17em}\mathit{Ad}$$ (1a)

$$\frac{\mathit{dAc}}{\mathit{dt}}\hspace{0.17em}=\hspace{0.17em}\mathit{ka}\hspace{0.17em}*\hspace{0.17em}\mathit{Ad}\hspace{0.17em}-\hspace{0.17em}k{e}_p\hspace{0.17em}*\hspace{0.17em}\mathit{Ac}\hspace{0.17em}-\hspace{0.17em}{k}_{12}\hspace{0.17em}*\hspace{0.17em}\mathit{Ac}\hspace{0.17em}+\hspace{0.17em}{k}_{21}\hspace{0.17em}*\hspace{0.17em}\mathit{Ap}\hspace{0.17em}-\hspace{0.17em}{K}_{\mathit{pm}}\hspace{0.17em}*\hspace{0.17em}\mathit{Am}$$ (1b)

$$\frac{\mathit{dAp}}{\mathit{dt}}\hspace{0.17em}=\hspace{0.17em}{k}_{12}\hspace{0.17em}*\hspace{0.17em}\mathit{Ac}\hspace{0.17em}-\hspace{0.17em}{k}_{21}\hspace{0.17em}*\hspace{0.17em}\mathit{Ap}$$ (1c)

$$\frac{\mathit{dAm}}{\mathit{dt}}\hspace{0.17em}=\hspace{0.17em}{k}_{\mathit{pm}}\hspace{0.17em}*\hspace{0.17em}\mathit{Ac}\hspace{0.17em}-\hspace{0.17em}k{e}_m\hspace{0.17em}*\hspace{0.17em}\mathit{Am}$$ (1d)

Since the volume of distribution for desethylchloroquine is not identifiable, it was assumed to be equal to the central compartment volume of distribution for chloroquine (V_c). Ad in equation 1a denotes the amount of chloroquine in the absorption compartment, whose absorption is described by the rate constant ka. Ac represents the amount of chloroquine in the central compartment, while Ap and Am represent the amount of chloroquine in the peripheral compartment and the amount of desethylchloroquine in the central compartment, respectively, as shown in equations 1b, 1c, and 1d. K_pm is the rate constant for the conversion of chloroquine to desethylchloroquine, and k₁₂ and k₂₁ are the distribution rate constants for chloroquine from the central to the peripheral compartment and from the peripheral to the central compartment, respectively. The elimination rate constant for chloroquine is represented by ke_p and by ke_m for desethylchloroquine. Chloroquine PK was parameterized in terms of V_c, the peripheral volume of distribution (V_p), clearance (Cl_p), and intercompartmental clearance (Q_p). The PK of desethylchloroquine was characterized in terms of clearance (Cl_m) and the formation rate constant, K_pm. The residual unexplained variability was modeled by a combined error model (additive and proportional). The PK model was linked to a direct-effect PD structural model, as shown in equation 2, and was used to describe the change in parasitemia (P) after treatment with chloroquine:

$$\frac{\mathit{dP}}{\mathit{dt}}\hspace{0.17em}=\hspace{0.17em}{K}_{\mathit{growth}}\hspace{0.17em}*\hspace{0.17em}P\hspace{0.17em}-\hspace{0.17em}{K}_{\mathit{kill}}\hspace{0.17em}*\hspace{0.17em}\frac{C_B^H}{C_B^H\hspace{0.17em}+\hspace{0.17em}I{C}_{50}^H}\hspace{0.17em}*\hspace{0.17em}P$$ (2)

where K_growth is the parasite growth rate, K_kill is the parasite kill rate, C_B is the chloroquine whole-blood concentration, H is the Hill coefficient, and IC₅₀ is the inhibitory concentration that produces the half maximal effect. The predicted individual initial parasitemia (P_i) was modeled as a function of the observed baseline parasitemia (P0) and random effects (η_i) (equation 3):

$$P_i\hspace{0.17em}=\hspace{0.17em}P0\hspace{0.17em}*\hspace{0.17em}{e}^{{\eta }_i}$$ (3)

The error model used was proportional, and the distribution was lognormal. The PD parameters were assumed to have lognormal distribution, except for the IC₅₀ and η_i, which were normally distributed. The Hill coefficient was fixed at different values of 0.25, 0.5, 1, 3, 5, 7, and 9, and the value that displayed the best fit was included in the final PK-PD model. For PK and PK-PD model development, the simplest model was first selected, and parameters were progressively added and retained if the inclusion of the parameter resulted in a statistically significant improved model fit, as indicated by a decrease in the objective function value (OFV) of at least 3.84. In addition to the OFV, the PK and PK-PD models were evaluated visually using the diagnostic plots of the observed and the individual and population model-predicted concentration and parasitemia values versus time; the individual-weighted residuals (IWRES) against time and predicted concentration and parasitemia; and the visual predictive checks. The MPC was defined as the effective concentration that produces 95% of the maximal response (EC₉₅), and was calculated using the PK-PD model estimated IC₅₀ and Hill coefficient, according to equation 4:

$$\mathit{LogE}{C}_{50}\hspace{0.17em}=\hspace{0.17em}\mathit{LogE}{C}_{95}\hspace{0.17em}-\hspace{0.17em}\frac{\mathrm{1}}{\mathit{Hillslope}}\hspace{0.17em}*\hspace{0.17em}\mathit{Log}\left(\frac{95}{\mathrm{100-95}}\right)$$ (4)

(vi) Pharmacokinetic–pharmacodynamic driver of antimalarial efficacy. In order to determine the PK-PD index that was most strongly associated with chloroquine antimalarial activity in the P. falciparum-infected NSG mouse model, the secondary PK parameters were estimated using the population PK model developed for chloroquine. The model predicted C_max, the area under the concentration-time curve from 0 to 96 h (AUC_0–96), which was calculated using the linear trapezoidal rule, and the percentage time in which the whole-blood concentration of chloroquine remained above the MPC was determined. The PK-PD index values for each dosage group were then calculated by determining the ratio of the PK parameter to the MPC, namely, C_max/MPC, AUC_0–96/MPC, and percentage time above MPC. Each of these PK-PD index values were correlated with their corresponding antimalarial activity, which was defined as the percentage reduction in parasitemia of the treatment group relative to the mean parasitemia of the untreated control group on day 7 postinfection. To numerically evaluate the relationship between the log₁₀-transformed PK-PD index and the antimalarial activity, the coefficient of determination (R²) was estimated using nonlinear regression analysis with a variable slope and four-parameter model, as shown in equation 5:

$$Y\hspace{0.17em}=\hspace{0.17em}\mathit{Bottom}\hspace{0.17em}+\hspace{0.17em}\frac{\left(\mathit{Top}\hspace{0.17em}-\hspace{0.17em}\mathit{Bottom}\right)}{\mathrm{1\hspace{0.17em}+\hspace{0.17em}1}{\mathrm{0}}^{\left(\mathit{LogE}{C}_{50}\hspace{0.17em}-\hspace{0.17em}X\right)\hspace{0.17em}*\hspace{0.17em}\mathit{Hillslope}}}$$ (5)

where Bottom is the Y value representing the minimum response, Top is the Y value representing the maximum response, and EC₅₀ is the effective concentration that produces a response halfway between the maximal and minimal effect. The minimum response was constrained to a 0% reduction in parasitemia of the treatment group with respect to the untreated control group. Nonlinear regression analysis was performed in Prism 8.4.1 software (GraphPad). The PK-PD index that displayed the highest correlation to antimalarial activity was identified as the PK-PD driver of efficacy, and the EC₅₀ and EC₉₅ were determined using equations 4 and 5. For the concentration-dependent PK-PD indices, the EC₉₅ value corresponded to the ratio of the C_max or AUC_0–96 to the MPC. The absolute C_max or AUC_0–96 that produced the response at the EC₉₅ was determined by multiplying the EC₉₅ value by the MPC. If the antimalarial activity of chloroquine was driven by C_max/MPC or AUC_0–96/MPC, then the absolute C_max or AUC_0–96 value at the EC₉₅ was used as the target concentration or exposure to which the preliminary human dose was set to achieve. For the time-dependent PK-PD index, the EC₉₅ value corresponded to the time in which the concentration of chloroquine remains above the MPC. If the antimalarial activity of chloroquine was time-dependent, then the preliminary human dose was selected to allow the concentration of chloroquine to remain above the MPC for the duration of time indicated by the EC₉₅ value.

Physiologically based pharmacokinetic model for chloroquine.

(i) Model development. A PBPK model was developed to describe the PK of chloroquine in humans using PK-Sim 11 software (Open Systems Pharmacology). Experimentally determined physicochemical and in vitro metabolism data for chloroquine were obtained from literature and used as input parameters in the PBPK model, as shown in Table S2.1. The partition coefficients were calculated using the Rodgers and Rowland method (21, 22). The whole-blood PK profiles for chloroquine were simulated in a healthy European population with 100 individuals and an equal distribution of males to females. The age, weight, height, and body mass index of the individuals in the virtual population ranged from 18 to 65 years, 55 to 90 kg, 150 to 180 cm, and 21 to 29 kg/m², respectively.

(ii) Model validation. The final PBPK model for chloroquine was validated using published clinical data from Neuvonen et al. (23). The trial design of the simulation was matched to the publication study design. Graph Digitizer 2.26 (GetData) was used to extract the chloroquine concentration-time data from the validation publication, which was then overlaid onto the simulated population PK profiles in PK-Sim. Additionally, the PBPK model-predicted PK parameters, namely, the C_max, the time taken to reach the peak concentration (T_max), the half-life (T_½), and the area under the concentration-time curve from 0 to 192 h (AUC_0–192), were compared to the clinically observed PK parameters. The accuracy of the PBPK model was based on whether the observed concentration-time data and PK parameters fell within the range of the simulated population PK data.

Human dose prediction.

(i) Preliminary human dose prediction. Whole blood concentration-time profiles for chloroquine were simulated using the final PBPK model for a series of single oral administrations of 50, 150, 400, and 750 mg of chloroquine in a healthy human population. The PK parameter, either C_max, AUC_0-96, or time above MPC, that was most strongly associated with chloroquine activity was determined at each of these dosages. With the assumption of dose proportionality, a linear regression analysis of the dose versus the corresponding predicted PK parameter was performed to interpolate the dose that produced the target value of the PK parameter at the ED₉₅ of the PK-PD driver of efficacy, as determined in the P. falciparum-infected NSG mouse model.

(ii) Efficacious human dose and dosage regimen prediction. Chloroquine population PK profiles were simulated using the final PBPK model after a single 150, 400, and 750 mg oral administration of chloroquine. Four individuals from each dose group were randomly selected to determine the primary PK parameters of chloroquine using nonlinear mixed effects modeling of the simulated whole blood chloroquine concentration-time data in Monolix. The model parameters were estimated using a stochastic approximation expectation maximization algorithm (20). A two-compartment model was chosen, with first-order absorption and elimination following oral administration, to describe the PK of chloroquine, and was modeled using a series of differential equations (equation 6):

$$\frac{\mathit{dAd}}{\mathit{dt}}\hspace{0.17em}=\hspace{0.17em}-\mathit{ka}\hspace{0.17em}*\hspace{0.17em}\mathit{Ad}$$ (6a)

$$\frac{\mathit{dAc}}{\mathit{dt}}\hspace{0.17em}=\hspace{0.17em}\mathit{ka}\hspace{0.17em}*\hspace{0.17em}\mathit{Ad}\hspace{0.17em}-\hspace{0.17em}k{e}_p\hspace{0.17em}*\hspace{0.17em}\mathit{Ac}\hspace{0.17em}-\hspace{0.17em}{k}_{12}\hspace{0.17em}*\hspace{0.17em}\mathit{Ac}\hspace{0.17em}+\hspace{0.17em}{k}_{21}\hspace{0.17em}*\hspace{0.17em}\mathit{Ap}$$ (6b)

$$\frac{\mathit{dAp}}{\mathit{dt}}\hspace{0.17em}=\hspace{0.17em}{k}_{12}\hspace{0.17em}*\hspace{0.17em}\mathit{Ac}\hspace{0.17em}-\hspace{0.17em}{k}_{21}\hspace{0.17em}*\hspace{0.17em}\mathit{Ap}$$ (6c)

Chloroquine PK was parameterized in terms of ka, Cl_p, V_c, Q_p, and V_p. Inclusion of model parameters was guided by the OFV and the diagnostic plots, as previously described. The error model and distribution were proportional and normal, respectively, and the PK parameters were assumed to have lognormal distribution. The final PK structural model and population PK parameters (ka, Cl_p, V_c, Q_p, and V_p) for chloroquine in humans, and the direct-effect PK-PD structural model and population PK-PD parameters (K_growth, K_kill, IC₅₀, H, and η_i) for chloroquine in NSG mice were integrated in Simulx 2021R1 software (Lixoft) to perform human dose-response simulations, with the assumption that the PD properties for chloroquine against P. falciparum are similar in humans and NSG mice. The initial parasitemia in humans was set at 1%. The preliminary human dose and various other doses and dosage regimens of chloroquine were simulated to determine the optimum treatment schedule that resulted in effective parasite clearance. For practicality, the simulated doses were rounded up or down to the nearest 50 mg.