In-host modeling of dengue virus and non-structural protein 1 and the effects of ivermectin in patients with acute dengue fever

Abstract

The increased incidence of dengue poses a substantially global public health challenge. There are no approved antiviral drugs to treat dengue infections. Ivermectin, an old anti-parasitic drug, had no effect on dengue viremia, but reduced the dengue non-structural protein 1 (NS1) in a clinical trial. This is potentially important, as NS1 may play a causal role in the pathogenesis of severe dengue. This study established an in-host model to characterize the plasma kinetics of dengue virus and NS1 with host immunity and evaluated the effects of ivermectin, using a population pharmacokinetic–pharmacodynamic (PK–PD) modeling approach, based on two studies in acute dengue fever: a placebo-controlled ivermectin study in 250 adult patients and an ivermectin PK–PD study in 24 pediatric patients. The proposed model described adequately the observed ivermectin pharmacokinetics, viral load, and NS1 data. Bodyweight was a significant covariate on ivermectin pharmacokinetics. We found that ivermectin reduced NS1 with an EC₅₀ of 67.5 μg/mL. In silico simulations suggested that ivermectin should be dosed within 48 h after fever onset, and that a daily dosage of 800 μg/kg could achieve substantial NS1 reduction. The in-host dengue model is useful to assess the drug effect in antiviral drug development for dengue fever.

Study Highlights

• WHAT IS THE CURRENT KNOWLEDGE ON THE TOPIC?

The global incidence of dengue has grown dramatically in recent decades. There are no approved antiviral drugs to treat dengue infections. It is proposed that the dengue non-structural protein 1 (NS1) has a direct role in the pathogenesis of severe dengue. Ivermectin inhibits dengue viral replication in vitro, but its role as a clinical therapy has not been evaluated.

• WHAT QUESTION DID THIS STUDY ADDRESS?

There is no detailed in-host model of dengue fever to describe dengue viremia and NS1 kinetics and their link to host-immunity, which is crucial to accurately assess potential drug effects of novel compounds. The relationship between ivermectin exposure and dengue viral and NS1 kinetics has not been quantitatively evaluated in patients with acute dengue fever.

• WHAT DOES THIS STUDY ADD TO OUR KNOWLEDGE?

An in-host model was established to describe dengue viremia and NS1 kinetics and their relationship to host immunity, based on observed data from two clinical studies in adult and pediatric patients with acute dengue fever. Furthermore, ivermectin showed a significant exposure–response relationship with NS1 reduction, but not with dengue viremia. In silico simulations suggested that ivermectin should be dosed within 48 h after fever onset, and that a daily dose of 800 μg/kg of ivermectin could achieve substantial NS1 reduction.

• OW MIGHT THIS CHANGE DRUG DISCOVERY, DEVELOPMENT, AND/OR THERAPEUTICS?

To the best of our knowledge, this is the first detailed in-host model of dengue infections reported in literature, and it could be a useful tool to assess novel and repurposed drugs in the treatment of dengue infections.

INTRODUCTION

Dengue is a mosquito-borne flavivirus disease that is now prevalent across more than 100 countries worldwide. There are four serotypes (dengue-1 to dengue-4) that can cause infections in human. The incidence of dengue has risen dramatically around the world in recent decades. In the first 10 months of 2023, over 4.5 million cases and over 4000 dengue-related deaths have been reported from 80 countries/territories globally.¹ The WHO estimates that 390 million dengue infections occur annually, of which 96 million manifest clinically.² Dengue symptoms last 2–7 days in most patients. Some patients, particularly children, can progress to life-threatening syndromes, such as dengue hemorrhagic fever (DHF) and dengue shock syndrome (DSS). It is estimated that 500,000 people with dengue require hospitalization due to warning signs or severe dengue, and that dengue causes about 20,000 deaths every year.² Secondary infections are subsequent new dengue infections (with any serotype) that occur following after a primary infection has been cleared, and can lead to more severe disease than the primary infection.

An integrated approach including vector control, prophylaxis with safe and effective vaccines, and an effective treatment in acute infections is suggested to tackle the increasing global number of dengue infections. The currently approved tetravalent dengue vaccines (CYD-TDV and TAK-003) reduce infections and hospitalizations, particularly in dengue-seropositive subjects.^3-5 As a key component of an integrated approach, antiviral treatment is still needed to address the high unmet medical need in patients with dengue infections to reduce the disease burden (i.e., morbidity, death, and hospitalization).

To date, there are no approved antiviral drugs to treat dengue infections. Several drugs have been proposed as repurposing candidates including chloroquine,⁶ balapiravir,⁷ celgosivir,^{8, 9} lovastatin,¹⁰ and ivermectin,¹¹ but randomized placebo-controlled trials have failed to demonstrate efficacy in terms of either reduction of viral load or improvement of other clinical outcomes. The treatment window for specific antivirals is relatively narrow as viremia declines rapidly around the fourth or fifth day of illness coincident with the rise in antibody concentrations. This is also the time when severe forms of the disease manifest.

Identification of biomarkers that correlate with subsequent progression to severe dengue would be very valuable for drug development. The dengue non-structural protein 1 (NS1) is a potential candidate. This glycoprotein is secreted in large amounts by infected cells and is detectable longer in patients' blood circulation than viremia. There is evidence suggesting that NS1 and anti-dengue NS1 antibodies are responsible for the pathogenesis of severe dengue (i.e., DHF/DSS).^{12, 13} Dengue NS1 antigen has a direct action on vascular endothelium, and is considered a major factor causing disruption of endothelial cell monolayer integrity. Meanwhile, NS1-mediated release of inflammatory cytokines from immune cells also contributes to endothelial hyperpermeability and vascular leak.^14-16 The major pathogenesis of anti-dengue NS1 antibodies in severe dengue is the uncontrolled release of cytokines. A recent review article summarizing 10 hospital-based studies from different countries found that the majority of these studies showed a positive correlation between high NS1 levels and dengue-associated severity (DHF/DSS).¹⁷

Ivermectin inhibits dengue replication in vitro.^{18, 19} In a randomized, placebo-controlled clinical trial that we conducted, ivermectin did not accelerate viral clearance, but did reduce circulating NS1 concentrations.¹¹ In addition, we found that ivermectin inhibits the nuclear transport of transcription factors required for the expression of chaperones, including GRp78 (78-kDa glucose-regulated protein), that support the folding and secretion of NS1.²⁰ In a subsequent pharmacometric clinical trial, the viral load and NS1 profiles were characterized in pediatric patients with dengue infection receiving different ivermectin dosing regimens. The objective of the current pharmacometric evaluation is to use data from both these studies to (1) characterize the pharmacokinetic (PK) properties of ivermectin in acute dengue, (2) characterize the kinetics of viral load and NS1 antigenemia and their determinants, and (3) assess quantitatively the relationship between ivermectin exposure and NS1 concentrations.

METHODS

Study design

The pharmacometric data came from an extended phase II/III randomized, placebo-controlled trial, which investigated the efficacy and safety of ivermectin in adult patients with acute dengue virus infection (NCT02045069, ESIDEN). It was conducted at Siriraj Hospital in Bangkok and Loei Hospital in Northern Thailand. The clinical results of this study have been reported elsewhere.¹¹ A total of 250 patients were enrolled in this trial resulting in 44, 101, and 105 patients being assigned randomly to once-daily ivermectin 400 μg/kg/day for 2 days with placebo on the third day, once-daily ivermectin 400 μg/kg/day for 3 days and once-daily placebo for 3 days, respectively.

Subsequently, we conducted a pharmacokinetic–pharmacodynamic (PK–PD) study of ivermectin in pediatric dengue patients (NCT03432442, PKIDEN) at Siriraj Hospital in Thailand. This was a two-phase clinical trial to assess the virological response to 400 and 600 μg/kg/day of once-daily ivermectin given for 3 days. A total of 24 pediatric patients were assigned to four arms, with six patients each stratified by bodyweight, to receive 400 μg/kg/day (first two arms) or 600 μg/kg/day (subsequent two arms) daily dose. These children were admitted and had an IV line for blood collection. Thus, the frequent number of blood draws did not translate to a high degree of pain or discomfort.

The details of study design, dose rationale, blood sampling and bioassay for ivermectin, viral load, NS1, and virion- and NS1-specific antibodies measurement in the adult and pediatric clinical studies are shown in Table S1 and Supplementary Material.

Ethic approval

The adult study was approved by the Siriraj Institutional Review Board, Faculty of Medicine Siriraj Hospital, Mahidol University (Protocol number: 553/2556) and the Ethical Review Committee for Research in Human Subjects, Ministry of Public Health, Thailand (Protocol number: 21/2556). The pediatric study was approved by the Siriraj Institutional Review Board, Faculty of Medicine Siriraj Hospital, Mahidol University (Protocol number: 728/2560) and the Oxford Tropical Research Ethics Committee, the University of Oxford, United Kingdom (Reference number: 39-17). All patients or their parents/guardians provided written informed consent.

Modeling analysis

The population PK–PD analysis was performed using nonlinear mixed-effects modeling in the NONMEM software (version 7.4, ICON Development Solutions, Ellicott City, MD, USA). Full details can be found in the Supplementary Material.

Ivermectin PK modeling

The concentration–time data of ivermectin were modeled using a population PK approach. Full details can be found in the Supplementary Material.

Virus kinetic modeling

A classical target cell-limited model, which includes target cell (T), infected cell (I), virus-producing cell (P) and viral load (V), was used to characterize dengue virus kinetics.^21-23 In the two clinical trials, the time of symptom onset (i.e., fever) was recorded, and from this we back-calculated the starting time of the infection, based on an average incubation period of 6 days from infection to symptom onset.^{24, 25} The model structure is shown in Figure 1, and the differential equations are given by Equations 1-6. Target cells are depleted by infection with free virus at rate βVT, and become infected cells. A flexible transit model with a stepwise addition of a fixed number of 1–10 transit-compartments was employed to describe the transition from infected cells to producing cells with a rate constant of k_tr1. Free virus is produced by producing cells at rate p and cleared at rate c. Producing cells die at rate δ.

$$\frac{\mathrm{d}T}{\mathrm{d}t}=-\mathit{\beta TV}$$(1)

$$\frac{\mathrm{d}{I}_1}{\mathrm{d}t}=\mathit{\beta TV}-k_{\mathrm{tr}1}{I}_1$$(2)

$$\frac{\mathrm{d}{I}_2}{\mathrm{d}t}=k_{\mathrm{tr}1}{I}_1-k_{\mathrm{tr}1}{I}_2$$(3)

$$\frac{\mathrm{d}{I}_n}{\mathrm{d}t}=k_{\mathrm{tr}1}{I}_{n-1}-k_{\mathrm{tr}1}{I}_n$$(4)

$$\frac{\mathrm{d}P}{\mathrm{d}t}=k_{\mathrm{tr}1}{I}_n-\mathrm{\delta }P$$(5)

$$\frac{\mathrm{d}V}{\mathrm{d}t}=\mathit{pP}-\mathrm{cV}$$(6)

FIGURE 1

Graphical overview of the dengue viral load, NS1 kinetic and ivermectin PK model. A classical target cell-limited model was used, which includes target cells (T), infected cells (I), production cells (P) and viral load (V; observations). Adaptive immune serology responses (Dengue-specific IgM and IgG) were incorporated into the virus elimination pathway, and Dengue NS1-specific IgG was included in the NS1 elimination pathway. β is the infection rate constant, ktr1 is the transition rate constant for infected cells, δ is the death rate constant of producing cells, p is the virus production rate constant, c is the virus elimination rate constant by innate immunity (IFN, NK cell), and k_A is the elimination (neutralization) rate of virus by dengue-specific IgG and IgM antibodies (A). The initial number of target cells (T₀) is fixed to the literature value of 3.5 × 10⁵ cells/mL. The concentration-time profile of IgM and IgG were fit using sigmoidal maximum effect models $\left(A_{\mathrm{IgM}}\left(t\right)=\frac{E_{\max \_\mathrm{IgM}}·T^{\gamma }}{{\mathit{ET}}_{50\_\mathrm{IgM}}^{\gamma }+T^{\gamma }},A_{\mathrm{IgG}}\left(t\right)=\frac{E_{\max \_\mathrm{IgG}}·T^{\gamma }}{{\mathit{ET}}_{50\_\mathrm{IgG}}^{\gamma }+T^{\gamma }}\right)$, where E_max is the maximum concentration of IgM or IgG, T is time after infection, ET₅₀ is the time to reach 50% maximum levels and γ is the shape parameter. The model-predicted individual IgM and IgG were normalized to a 0–1 scale by the equation of $A\left(t\right),i=\left(\frac{A_{\mathrm{IgM}}\left(t\right),i}{E_{\max \_\mathrm{IgM},i}}+\frac{A_{\mathrm{IgG}}\left(t\right),i}{E_{\max \_\mathrm{IgG},i}}\right)/2$. NS1 delayed secretion from viral-producing cells was assumed. k_tr2 is the transition rate constant for the NS1 secretion. k_p is the NS1 production rate, defined by $\frac{E_{\max ,\mathrm{NS}1}·T^{\gamma }}{{\mathit{ET}}_{50,\mathrm{NS}1}^{\gamma }+T^{\gamma }}$, where E_max,NS1 is the maximum production rate of NS1, ET_50,NS1 is the time to reach 50% E_max,NS1, T is the time after infection. k_B is the elimination rate constant of NS1, defined by a spline function, either equal to the initial rate constant (antibodies (Abs) level < a cutoff value) or the initial rate constant + slope· T·Abs (Abs ≥ a cutoff value). Ivermectin PK was described by a two- compartment disposition model with six transit absorption compartments. CL is the clearance. V_c is the central volume of distribution. V_P is the peripheral volume of distribution. Q is the inter-compartment clearance and k_tr3 is the transition rate constant for absorption. The inhibition of NS1 production by ivermectin was driven by an E_max function, where C_IVM is the model-predicted ivermectin concentration, E_max,IVM is the maximum inhibitory effect, EC_50,IVM is the ivermectin concentration to achieve half of maximum inhibitory effect.

The initial concentration of target cells (monocytes,²⁶ T₀) was fixed at a value of 3.5 × 10⁵ cells/mL,²² approximating normal peripheral blood monocyte counts and the initial value of the viral load (V₀) was assumed to be 1 dengue RNA copy number/mL.

The adaptive immune antibody response (dengue-specific IgM and IgG) was fitted separately using a linear, maximum effect (E_max) model, or sigmoidal E_max model (Equations 7 and 8), and the model with the best performance was selected. The model-predicted individual IgM and IgG concentrations were normalized to a 0–1 scale using Equation 9. With the assumption that the virus was neutralized by dengue-specific antibodies at the rate of k_A, Equation 6 becomes Equation 10.

$$A_{\mathrm{IgM}}\left(t\right)=\frac{E_{\max \_\mathrm{IgM}}·T^{\gamma }}{{\mathit{ET}}_{50\_\mathrm{IgM}}^{\gamma }+T^{\gamma }}$$(7)

$$A_{\mathrm{IgG}}\left(t\right)=\frac{E_{\max \_\mathrm{IgG}}·T^{\gamma }}{{\mathit{ET}}_{50\_\mathrm{IgG}}^{\gamma }+T^{\gamma }}$$(8)

$$A\left(t\right),i=\left(\frac{A_{\mathrm{IgM}}\left(t\right),i}{E_{\max \_\mathrm{IgM}},i}+\frac{A_{\mathrm{IgG}}\left(t\right),i}{E_{\max \_\mathrm{IgG}},i}\right)/2$$(9)

$$\frac{\mathrm{d}V}{\mathrm{d}t}=\mathit{pP}-\mathrm{cV}-k_A\mathrm{A}$$(10)

where $E_{\max \_\mathrm{IgM}}$ and $E_{\max \_\mathrm{IgG}}$ are the population estimates of the maximum concentrations of IgM and IgG. T is the time after infection. ET₅₀ is the time to reach 50% maximum levels and γ is the shape parameter. $A_{\mathrm{IgM}}\left(t\right),i,A_{\mathrm{IgG}}\left(t\right),i$ are the model-predicted individual IgM and IgG concentrations at time t. E_max,i is the model-predicted individual maximum concentration of IgM or IgG.

The covariates of interest affecting the virus kinetic model parameters, including serotype, age, bodyweight, infection type (primary or secondary infection), were evaluated using the forward inclusion and more strict backward elimination procedures (Supplementary Material).

NS1 kinetic modeling

The plasma NS1 concentration–time profile was modeled by implementing a NS1 compartment (N), linked to the producing cell compartment in a classical target cell-limited model, as shown in Figure 1. A flexible transit compartment (i.e., addition of 1–10 transit compartments) was used to describe the delayed process of NS1 secretion from the producing cell compartment, given by Equations 11-14, and the optimal number of compartments was determined by the lowest objective function value. NS1 secretion was evaluated using either first-order or time-dependent E_max processes (Equation 15). The elimination of NS1 was assumed to be a first-order process with rate constant k_B. The effects of NS1-specific IgM, IgA, and IgG (on day 5 after onset) on NS1 clearance were evaluated using either linear or spline functions (Equations 16 and 17). The effect of time on NS1 clearance was investigated further, based on Equation 17.

The NS1 kinetics were modeled sequentially by using the individual parameter estimates of the virus kinetic model described above as the input.

$$\frac{\mathrm{d}{P}_1}{\mathrm{d}t}=P-k_{\mathrm{tr}2}{P}_1$$(11)

$$\frac{\mathrm{d}{P}_2}{\mathrm{d}t}=k_{\mathrm{tr}2}{P}_1-k_{\mathrm{tr}2}{P}_2$$(12)

$$\frac{\mathrm{d}{P}_n}{\mathrm{d}t}=k_{\mathrm{tr}2}{P}_{n-1}-k_{\mathrm{tr}2}{P}_n$$(13)

$$\frac{\mathrm{d}N}{\mathrm{d}t}=k_p{P}_n-k_{B}N$$(14)

$$k_{\mathrm{p}}=\frac{E_{\max ,NS1}·T^{\gamma }}{{\mathrm{ET}}_{50,NS1}^{\gamma }+T^{\gamma }}$$(15)

$$k_{\mathrm{B}}=\text{Baseline If antibodies}\left(\mathrm{Abs}\right)<\mathrm{a}\text{cutoff value}$$(16)

$$k_{\mathrm{B}}=\text{Baseline}+\text{slope}·T·\mathrm{Abs}\text{If}\mathrm{Abs}\ge \mathrm{a}\text{cutoff value}$$(17)

where k_tr2 is the transition rate constant for NS1 secretion, k_p is the NS1 production rate, described by an E_max function, where, E_max,NS1 is the maximum production rate of NS1, ET_50,NS1 is the time to reach 50% E_max,NS1, T is the time after infection. k_B is the elimination rate constant of NS1, which is dependent on time and NS1-specific IgG.

PK–PD analysis

Since the distributions of viremia and NS1 levels were skewed, a natural logarithm transformation was conducted prior to the PK–PD analysis. Based on the developed virus and NS1 kinetic model, the effects of ivermectin (i.e., drug concentrations) on virus production rate and NS1 maximum production rate were investigated further using either linear or E_max models, respectively.

Model simulation

The viral load and NS1 profiles over time, based on a typical adult and pediatric patient, were simulated following different ivermectin dosing regimens to inform optimal ivermectin treatment. The area under the curve (AUC) of viral load and/or NS1 concentration profiles regarding each simulation scenario were summarized in the examination of dose optimization simulations.

RESULTS

The demographic characteristics of participants in two clinical studies are shown in Table S2.

Ivermectin pharmacokinetics

The final ivermectin PK model was a two-compartment disposition model with six transit compartments describing the absorption process (Figure 1). Bodyweight was the only significant covariate included in the model. Full details can be found in Supplementary Material.

Viral load kinetics

The viral kinetics were described adequately by the target cell-limited model with three transit compartments describing the delay from infected cell to production cell as shown in Figure 1.

In the model fitting, the concentration-time profile of both dengue-specific IgM and IgG was characterized adequately by sigmoidal E_max models (Figures S3 and S4). The individually predicted dengue-specific normalized IgM and IgG values (calculated from Equation 9) derived from the final model were implemented as determinants of viral clearance, which improved the model fit substantially (∆OFV = −43.067).

The final viral kinetic model included three significant covariates (age on infected cell death rate, serotype on infection rate and viral production rate). Full details can be found in Supplementary Material.

The final dengue virus kinetic model showed satisfactory goodness of fit (Figure S5) and good predictive performance for both ivermectin and placebo arms as shown in the VPC plots (Figure 2). The parameter estimates of final dengue viral kinetic model are presented in Table 1, showing good precision for estimates.

FIGURE 2

Visual predictive checks of the dengue viral kinetic model. (a) Adult patients receiving placebo treatment, (b) Adult patients receiving ivermectin treatment, (c) pediatric patients receiving ivermectin treatment. Based on 1000 stochastic simulations. Upper panel: Open circles represent the observations, solid and dash lines represent the 5th, 50th, and 95th percentiles of the observed data. The shaded areas represent the 95% prediction intervals around the simulated 5th, 50th, and 95th percentiles. Most of observed data fell within the 95% prediction interval. Lower panel: Open circles represent the observed fraction of censored data, and the shaded area represents the 95% prediction interval of the simulated fraction of censored data. LOQ is below the lower limit of quantification.

TABLE 1. Final parameter estimates of the in-host viral kinetic model of acute dengue infection.

Parameters	Estimates (RSE %)	SIR median (95% CI)	CV for IIV (RSE %)	SIR median (95% CI)	Shrinkage (%)
Viral kinetics
β (mL/day/copy)	10^−6.47 (2.0)	10^−6.45 (10^−6.61 to 10^−6.31)	50.5 (12.1)	51.5 (41.7–61.2)	10.7
T0 (cells/mL)	3.5 × 10^6 (fixed)	–	–		–
KTR (1/day)	6.73 (19.0)	6.53 (5.49–8.39)	–		–
Number of transit compartments	3 (fixed)	–	–		–
δ (1/day)	2.76 (4.2)	2.76 (2.58–2.98)	34.9 (8.9)	34.8 (30.2–40.0)	24.5
p (copy/mL/day)	10^3.3 (1.3)	10^3.3 (10^3.21 to 10^3.38)	–		–
V0 (copy/cell)	1 (fixed)	–	–		–
c (1/day)	3.04 (7.4)	3.03 (2.61–3.41)	–		–
kH (1/day)	4.58 (16.8)	4.63 (3.40–5.98)	–		–
Residual error	1.31 (3.1)	1.31 (1.23-1.38)	–		–
Covariates on viral kinetic parameters
Serotype 1 on β (%)	15.2 (6.2)	15.3 (12.8–17.3)
Serotype 3 on β (%)	−12.8 (8.3)	−12.8 (−15.2 to −10.3)
Serotype 1 on p (%)	35.6 (8.5)	35.9 (28.9–41.8)
Serotype 3 on p (%)	−29.0 (10.6)	−28.9 (−35.6 to −22.2)
Age on δ (%)	2.5 (51.2)	2.5 (0.96–3.8)

• Note: β is the infection rate constant, T₀ is the initial number of target cell, KTR is the transition rate constant for infected cells, δ is the death rate constant of producing cells, p is the virus production rate constant, V₀ is the initial viral copy number, c is the virus elimination rate constant by the innate immune (IFN, NK cell), and k_H is the virus elimination (neutralization) rate by dengue-specific IgG and IgM antibodies.
• Age was implemented on the parameter δ using a spline function with the age cutoff value of 20 years, $\delta ={\delta }_{\mathrm{TV}}·\left(1+{\theta }_{\mathrm{age}}·\left(\mathrm{age}-20\right)\right)$, ${\theta }_{\mathrm{age}}=0.025$ for patients below 20 years, otherwise, ${\theta }_{\mathrm{age}}=0$. Categorical covariates were implemented on a specific parameter using a proportional model, $\theta ={\theta }_{\mathrm{TV}}·\left(1+\theta \mathrm{cov}\right)$, where ${\theta }_{\mathrm{TV}}$ is the typical value of a given parameter, $\theta \mathrm{cov}$ is the categorical covariate effect with 0 for the reference population (serotype 2 or 4 infection).
• Abbreviation: SIR, sampling importance resampling approach.

NS1 antigenemia kinetics

The plasma NS1 concentration–time profile was best characterized by the proposed model with nine transit compartments to describe the delayed process from producing cell to NS1 secretion, NS1-specific IgG-dependent NS1 clearance, and a time-dependent E_max model describing the production rate of NS1 (Figure 1). The final dengue NS1 model included two significant covariates, infection type on ET₅₀ (the time to reach 50% NS1 production rate) and age on baseline NS1 clearance. Full details can be found in Supplementary Material.

The final dengue NS1 model showed satisfactory goodness of fit (Figure S5) and good predictive performance by treatment (ivermectin and placebo) and age (adults and children) as shown in VPC plots (Figure 3). The parameter estimates of the final dengue NS1 kinetic model are presented in Table 2, showing overall good precision for most of parameter estimates.

FIGURE 3

Prediction-corrected visual predictive checks of the dengue NS1 kinetic model. (a) Adult patients with placebo, (b) Adult patients receiving ivermectin treatment, (c) Pediatric patients receiving ivermectin treatment. Based on 1000 stochastic simulations. Upper panel: Open circles represent the observations, solid and dash lines represent the 5th, 50th, and 95th percentiles of the observed data. The shaded areas represent the 95% prediction intervals around the simulated 5th, 50th, and 95th percentiles. Most of observed data fell within the 95% prediction interval. Lower panel: Open circles represent the observed fraction of censored data, and the shaded area represent the 95% prediction interval of the simulated fraction of censored data. LOQ is below the lower limit of quantification.

TABLE 2. Final parameter estimates describing the Dengue NS1 kinetic model.

Parameters	Estimates (RSE %)	SIR (95% CI)	CV for IIV (RSE %)	SIR Median (95% CI)	Shrinkage (%)
NS1 kinetics
KTR (1/day)	6.76 (4.5)	6.79 (6.30–7.31)	37.9 (9.1)	37.6 (32.4–44.5)	26.9
Number of transit compartments	9 (fixed)	–	–		–
E max,NS1	1.04 (11.9)	1.03 (0.79–1.36)	–		–
ET50,NS1 (days)	8.3 (1.8)	8.3 (8.0–8.6)	13.4 (11.3)	13.7 (11.3–16.4)	34.8
γNS1	18.0 (10.1)	17.8 (15.1–20.7)	–		–
k base	0.479 (19.2)	0.470 (0.356–0.624)	90.3 (11.5)	91.4 (74.8–109.1)	27.9
Residual error	1.15 (2.5)	1.15 (1.07-1.22)
Covariates on NS1 kinetic parameters
NS1-specific IgG on kB	0.112 (13.2)	0.112 (0.091–0.136)	–		–
Primary infection on ET50,NS1 (%)	23.1 (24.1)	23.6 (15.4–33.2)
Age on kbase (%)	−30.1 (44.9)	−30.3 (−32.0 to −16.5)
Ivermectin effect on NS1
Maximum effect of ivermectin on Emax,NS1	0.99 (fixed)
EC50 of ivermectin concentration (ng/mL) on Emax,NS1	67.5 (31.0)	69.8 (34.6–127.7)	–		–

• Note: E_max,NS1 is maximum secretion rate constant of NS1, and ET_50,NS1 is the time reach 50% maximum NS1 secretion and γ_NS1 is the shape parameter. k_B is the NS1 elimination rate constant, which is incorporated in the model using a spline function $\left(k_{\text{B}}=k_{\text{base}}+{\theta }_{\mathrm{IgG}}·\text{Time}·\left(\text{logIgG}-4\right)\right)$, ${\theta }_{\mathrm{IgG}}$ = 0.108 for patients with NS1-specific IgG (natural log transformation) greater than 4, otherwise, ${\theta }_{\mathrm{IgG}}=0$.
• Age was implemented on the parameter k_base using a spline function with the age cutoff value of 20 years, $k_{\text{base}}=k_{\text{base},\mathrm{TV}}·\left(1+{\theta }_{\mathrm{age}}·\left(\mathrm{age}-20\right)\right)$, ${\theta }_{\mathrm{age}}$ = −0.301 for patients below 20 years, otherwise, ${\theta }_{\mathrm{age}}=0$. Categorical covariates were implemented on a specific parameter using a proportional model, $\theta ={\theta }_{\mathrm{TV}}·\left(1+\theta \mathrm{cov}\right)$, where ${\theta }_{\mathrm{TV}}$ is the typical value of a given parameter, $\theta \mathrm{cov}$ is the categorical covariate effect with 0 for reference population (seondary infection).
• The ivermectin concentration was incorporated on the parameter E_max,NS1 using an E_max function.
• Abbreviation: SIR, sampling importance resampling approach.

PK-PD analysis

Ivermectin effect on viral load

The effect of individually predicted concentrations of ivermectin was assessed on virus production rate in the PK-PD modeling analysis. Ivermectin did not significantly improve model fits, using either a linear model or E_max model (∆OFV = −0.102 and −0.132, respectively).

Ivermectin effect on NS1

Ivermectin concentrations showed a significant effect on NS1 production rate using both linear (∆OFV = −16.836) and E_max models (∆OFV = −16.658). The two candidate models showed similar predictive performances, although the E_max model was more stable compared to the linear model in terms of success in the covariance steps, and was therefore chosen as the final model. The final parameters were estimated with acceptable precision (Table 2).

In silico simulation

A typical adult patient is defined as an individual with the same bodyweight as the median bodyweight (60 kg), and the most prevalent dengue serotype (serotype 2 or 4) and infection type (secondary infection) observed in the study population. A typical pediatric patient has the same definition as above, except a bodyweight of 30 kg as the median bodyweight in the pediatric study population.

The final ivermectin PK, and dengue viral and NS1 kinetic models were used to generate virtual plasma NS1 concentration profiles in a typical adult patient and a pediatric patient in different dosing simulation scenarios. All patients were assumed to be infected on Day 0, followed by a 6-day incubation period before fever onset.

Assuming inhibition of NS1 production is beneficial, simulations of different dosing scenarios were conducted sequentially to inform timing, duration, and dose of proposed ivermectin treatment, as described below; (1) simulation of the first scenario was performed to optimize the timing of treatment (24, 48, and 72 h after fever onset (on day 6 after the infection)) at a certain dose level of ivermectin (400 μg/kg/day), (2) based on the results from the first scenario, simulation of the second scenario was conducted to optimize the duration of ivermectin treatment (1 day vs. 2 days vs. 3 days of 400 μg/kg/day), and (3) based on the results from the second scenario, simulation of the third scenario was conducted to optimize the dose of ivermectin treatment (400 vs. 600 vs. 800 μg/kg/day).

The simulated median NS1 plasma concentration-time profiles and percentage reduction in NS1 AUC compared to placebo at different dosing scenarios are shown in Figure 4 and Table 3. Overall, the results indicated that NS1 reduction was dosing time-dependent. Early dosing was necessary for maximal effects. The predicted NS1 profile with 3-day ivermectin treatment initiated on day 9 did not differ from that in the placebo group (Figure 4a). The proportional reduction in NS1 AUC_0-42 in patients receiving ivermectin 72 h after fever onset (day 9 after infection) was minimal, whereas substantial NS1 reduction was seen with early ivermectin treatment (e.g., 24 h after fever onset). This suggested that ivermectin should be dosed within 48 h after onset. According to the simulation of the second scenario, the NS1 profile (Figure 4b,c) and NS1 reduction (Table 3) of 3- and 2-day ivermectin treatment nearly overlapped, if the treatment was started on day 7 (day 1 after fever onset). The difference in NS1 reduction among 1-, 2- and 3-day ivermectin treatments was minimal, if the treatment started on day 8. This suggested the optimal treatment duration was 2-day treatment from day 7, and 1-day treatment from day 8, respectively. The simulation of the third scenario showed the greatest reduction in NS1 AUC (35.5% when dosing on days 7 and 8; 17.7% when dosing on day 8) with 800 μg/kg/day of ivermectin (Figure 4d–f; Table 3). The simulations conducted in a typical pediatric patient showed similar NS1 reduction pattern in terms of NS1 profile, proportional reduction in NS1 AUC as those in a typical adult (Refer to Figure S6 and Table S5).

FIGURE 4

Simulated median plasma NS1 profiles in different dosing scenarios using the final PK–PD model. The simulation was conducted in a typical adult patient (60 kg bodyweight, serotype 2 or 4 and secondary infection). The simulation scenarios assess the impact of ivermectin treatment timing (a), duration (b, c) and dose (d-f) on the NS1 profiles. The incubation period of dengue is assumed to be 6 days after the infection. The ivermectin treatment was initiated on 24 (day 7), 48 (day 8) or 72 (day 9) hours after illness onset.

TABLE 3. NS1 reduction relative to placebo for different simulated dosing scenarios in adult patients with dengue infection.

Scenarios	Ivermectin daily dose (μg/kg/day)	NS1 reduction in AUC0-42 over placebo (Median%, (95% CI))
Placebo	0	NA.
Inform optimal treatment starting time (same dose and treatment duration)
Dosing on Days 7, 8 and 9	400	27.8 (7.0–52.2)
Dosing on Days 8, 9 and 10	400	17.3 (0.7–45.6)
Dosing on Days 9, 10 and 11	400	4.8 (0.03–36.4)
Inform optimal treatment duration (same dose and treatment starting day)
Dosing on Day 7	400	14.1 (4.0–30.0)
Dosing on Days 7 and 8	400	23.3 (6.5–46.2)
Dosing on Days 7, 8, and 9	400	27.8 (7.0–52.2)
Dosing on Day 8	400	11.0 (0.6–29.2)
Dosing on Days 8 and 9	400	15.8 (0.7–40.5)
Dosing on Days 8, 9, and 10	400	17.3 (0.7–45.6)
Inform optimal dose amount (same treatment starting day and duration)
Dosing on Day 7	400	14.1 (4.0–30.0)
Dosing on Day 7	600	19.2 (5.6–37.5)
Dosing on Day 7	800	23.3 (7.2–43.2)
Dosing on Day 8	400	11.0 (0.6–29.2)
Dosing on Day 8	600	14.7 (0.8–36.0)
Dosing on Day 8	800	17.7 (0.9–41.5)
Dosing on Days 7 and 8	400	23.3 (6.5–46.2)
Dosing on Days 7 and 8	600	30.2 (8.5–54.8)
Dosing on Days 7 and 8	800	35.5 (9.9–60.8)

• Note: Simulations were conducted in a typical adult patient (60 kg body weight, serotype 2 or 4 and secondary infection). A total of 1000 virtual individual NS1 profiles were generated for each scenario. The patients were assumed to be infected on day 0, and had fever onset on day 6.

Ivermectin dosing regimen

Assuming that NS1 reduction is beneficial, then if treatment is initiated between 0 and 24 h after fever onset, the optimal dose is predicted to be 800 μg/kg/day for 2 days. If treatment is initiated between 24 and 48 h after fever onset, the optimal dose is 800 μg/kg administered as a single dose.

DISCUSSION

Ivermectin has been used extensively in humans for the treatment of onchocerciasis, intestinal helminths, lymphatic filariasis and scabies, and it is widely used as a veterinary antiparasitic and endectocide. Here, the ivermectin effect on dengue viral load and NS1 was evaluated in two clinical trials in Thai adults and children with acute dengue infection. The potential antiviral activity of ivermectin was assessed with PK–PD modeling. The in-host dynamics of dengue virus infection are complex.²³ To accommodate the host immune response and drug effects, a mathematical model was developed and fitted to the detailed clinical trial data. Circulating virus and NS1 kinetics were modeled sequentially, incorporating both viremia controlling host immune responses and the previously documented effects of ivermectin exposure on viral load and NS1. The pharmacometric analysis indicated that while ivermectin had no effect on dengue viremia, it did reduce NS1 exposures. This is potentially important as NS1 may play a causal role in the pathogenesis of severe dengue.¹²

The PK properties of ivermectin in dengue were largely similar to values reported previously in other diseases and healthy volunteers^{27, 28} Ivermectin is extensively metabolized by cytochrome P450 CYP3A to three major metabolites in humans.²⁹ In this study, the PK properties of ivermectin were characterized adequately in both adult and pediatric patients with acute dengue fever. Total bodyweight was the only significant covariate in the model. The apparent clearance (CL/F) was overall comparable to values reported previously in patients and healthy volunteers receiving ivermectin monotherapy.^30-33 The average terminal half-life was 94 h (3.91 days) in adult patients with dengue infection, which was slightly longer than that of previous reports: 33.2 h,³² 48.1 h,³³ and 80.7–90.6 h.³¹ This may be explained partially by inhibition of CYP450 function in acute febrile illness.³⁴ Another possible explanation of the longer half-life of ivermectin in this study could be an increased volume of distribution due to reduced plasma protein levels associated with acute dengue infection.³⁵ In children, the reported typical CL/F was 7.4–8.6 L/h in Trichuris trichiura-infected school-aged children (median bodyweight 22 kg) receiving 400–600 mg of ivermectin,^{36, 37} which was higher than that observed in the current study (4.7 L/h, when scaled to a 22 kg child).

Plasma viral densities have been correlated with the severity of dengue,^{38, 39} and the aim of directly acting antivirals is to attenuate illness by inhibiting virus replication. Peak viremia generally occurs around the time of symptom onset (e.g., fever), as demonstrated in influenza and COVID-19 infections.^{40, 41} Dengue peak viremia appears parallel with or very early after illness onset (within 1 day of illness).⁴² In this study, the viral load peak was not apparent in the majority of patients, and the most likely explanation was that patients were enrolled later after illness onset (i.e., approximately 60 h after fever onset). After the initial plateau, there was a rapid decline in viremia coincident with the rise in dengue-specific IgM and IgG. This pattern is similar to that reported previously.^{21, 22} Importantly, as shown in other self-limited viral infections (e.g., influenza and COVID-19), once viral load is declining rapidly, the effect of an antiviral drug is limited.⁴³ As peak viremia in dengue generally occurs around the time of symptom (fever) onset and rapid viremia reduction occurs around the fourth or fifth day as antibody rises, the window of opportunity for an effective antiviral is narrow. The recent dengue human infection models could provide insights to better understand virus growth patterns.⁴⁴

Previously published reports have shown that NS1 plays an important role in dengue severity.^{12, 13} Candidate drugs targeting NS1 may reduce the severity of dengue infection assuming that lowering NS1 reduction is beneficial. The NS1 profile we modeled in this study had a clear peak at an average of 46 h after the onset of fever, followed by a slow decay phase. Plasma clearance of NS1 was substantially slower than that of viremia. If NS1 is a genuine drug target, then these results suggested that treatment targeting NS1 production should be given within 2 days of illness onset. This PK–PD analysis demonstrated a significant relationship between ivermectin exposure and NS1 reduction. It is unclear how ivermectin works, and why the effect on NS1 differs from that on viremia. Regarding mechanism of action, ivermectin could affect the NS3 helicase activity in flavivirus replication in vitro.¹⁹ The action of ivermectin on NS3 and NS5, the two major enzymes required for virus replication,^{19, 45} could lead directly to a reduction in NS1 production and secretion. But the exact mechanism by which ivermectin reduces NS1 levels, and whether this is specifically through direct interference with NS1 maturation and secretion processes, independent of virus replication, remains unclear and requires further investigation. It remains possible that ivermectin does have weak direct antiviral activity in vivo but that the sensitivity to detect this from the rapidly declining viremia profile is less than that required to see drug-related changes in NS1, which is cleared more slowly. Clearly, further studies are required to confirm these various findings and hypotheses. According to the simulations, the proposed 800 mg/kg/day dose resulted in much higher NS1 reduction than the 400 mg/kg/day dose (on days 7, 8: 35.5% versus 23.3%; on day 8: 17.7% versus 11.0%). However, the high-dose ivermectin (800 mg/kg/day) was not evaluated in the two clinical trials. The safety of high-dose ivermectin (600 and 800 mg/kg/day) needs further evaluation in patients with dengue fever, although this dose has been demonstrated to be safe in some anti-parasitic adult clinical trials.^{46, 47} The safety profile of high-dose ivermectin (600 and 800 mg/kg/day) has not been adequately assessed in pediatric patients, although no significant safety signals associated with the 600 mg/kg/day dose were seen in this small pediatric clinical trial.

The four different dengue virus serotypes and the different patterns of previous exposure (and thus immunity) create substantial diversity in the in-host viral dynamics. In this study, serotype 1 infections had greater virus production rates, resulting in higher viral loads than with serotypes 2 and 4. In contrast, serotype 3 infections had relatively lower viral loads. Some previous studies have suggested that patients with primary infections have higher viral loads^{48, 49}; but we did not find this in the current study. Furthermore, we found that NS1 secretion and elimination are significantly slower during a primary infection. This was attributed to low NS1-specific IgG levels, resulting in flatter NS1 profiles, compared to secondary infections. This modeling result enables a mechanistic interpretation of the high NS1 levels in other clinical studies.⁴⁸

We observed higher viral loads and lower NS1 levels in pediatric patients in comparison with adult patients. The modeling analysis suggested that pediatric patients had lower infection cell death rates and higher NS1 baseline clearance. However, the mechanism of this was not clear. It might be relevant to the different host immunity pattern in pediatric patients, such as the IgM response in children, which was higher than that in adult patients as shown in the current study (e.g., IgM level on day 5 after onset, 877 vs. 54 U/L).

Our study has several key limitations: (1) although previously published reports have shown that high NS1 is relevant to dengue severity, this correlation does not necessarily imply that lowering NS1 will be therapeutically beneficial. Further studies are needed to fully understand the exact role of NS1 in severe dengue. (2) Few patients were enrolled within 24 h after fever onset, which might affect modeling the growth phase of viral load and NS1. A more mechanistic quantitative systems pharmacology (QSP) model might be helpful to further explore and characterize the viral load, host immunity profiles, and their interactions. (3) Since each patient only had one NS1-specific antibody measurement at day 5 after fever, the full NS1 concentration-time profile could not be modeled, and the dynamic effects of this antibody on NS1 clearance were not investigated further. (4) No placebo group was planned in the pediatric PK–PD study, so the effects of ivermectin on viral load and NS1 profiles in pediatric patients were assumed to be same as those of adult patients. (5) Differences in dengue viral loads have been reported between infection types and serotypes.⁵⁰ However, it would be difficult to conclude this in a covariate analysis, given the small number of primary dengue infection by each serotype, missing time of infection, or low viral load profiles. (6) The model is complex and several parameters are unidentifiable. It fitted the data well, but that does not mean it is correct. Further studies to confirm or refute these findings are necessary.

CONCLUSION

The kinetic properties of ivermectin, dengue viral load and NS1 were described adequately in adult and pediatric patients with acute dengue fever. Ivermectin was associated with concentration-dependent reduction in plasma NS1 concentrations. The in-host dengue model presented here could be a useful tool to assess the drug effects in antiviral drug development for dengue fever.

AUTHOR CONTRIBUTIONS

J.D. and J.T. wrote the manuscript. J.T., N.J.W., D.M., and P.A. designed the research. D.M., P.A., Y.S., N.A., K.L., D.P., and K.C. performed the research. J.D. and J.T. analyzed the data. C.P., T.P., P.A., A.S., S.N., N.K., N.T., and J.T. contributed new reagents/analytical tools.

ACKNOWLEDGMENTS

We thank the patients for their participation, the doctors and nurses at the Siriraj Hospital and Loei Hospital, Thailand, for caring for the patients and collecting and processing the pharmacokinetic samples, and the staff of the Department of Clinical Pharmacology, Mahidol-Oxford Tropical Medicine Research Unit for drug measurements.

FUNDING INFORMATION

This work was funded by Department of Disease Control, Ministry of Public Health (R015745005 and R015845009), the Office of the Higher Education Commission and Mahidol University under the National Research Universities Initiative, Thailand Research Fund (RDG59D0011), Health Systems Research Institute (R016141025), Faculty of Medicine Siriraj Hospital (R015936007, R016136003), and the Wellcome Trust (220211). Clinical trial registration: NCT02045069 (ESIDEN) and NCT03432442 (PKIDEN). The funders had no part in the study design, implementation, and analysis of the result, or the decision to publish this manuscript. For the purpose of open access, the author has applied for a CC BY public copyright license to any author-accepted manuscript version arising from this submission.

CONFLICT OF INTEREST STATEMENT

The authors declared no competing interests for this work.