Introduction

Approximately 57 million people were estimated in 2020 to have chronic hepatitis C virus (HCV) infection, with ~ 1.5 million new infections and ~ 300,000 HCV-related deaths per year1. The availability of highly effective direct-acting antivirals (DAAs) can cure more than 95% of infected individuals, however access to diagnosis and treatment is low worldwide and there is no vaccine to prevent new HCV infections, about 80% of which will become chronic2,3, or re-infection in those who were cured by DAAs4. To accelerate the development of HCV vaccines, which is challenging5, clinical trials for a controlled human infection (CHI) model have been recently suggested6. The design of a CHI model requires careful consideration of the key elements including viral clearance rates, timing of blood sampling intervals for immunological evaluations, and susceptibility of the virus to DAAs. These elements must be fully understood for development of a final study protocol7. Analyzing early viral-host kinetics immediately after infection and developing theoretical modeling tools are important aspects for addressing the aforementioned key elements to successfully develop the CHI model for testing HCV vaccine candidates.

Studying early acute HCV kinetics in people is notoriously difficult because infections are usually asymptomatic8, although they may be recognized after iatrogenic exposures such as needle stick injuries, medical injection, needle sharing or other community exposures8. Our current understanding of early acute HCV-host dynamics is largely founded on data derived from infected chimpanzees using ordinary differential equations (ODEs)9. However, since June 2013, chimpanzees have no longer been available for biomedical research10 and appropriate immunocompetent animal models for preclinical vaccine testing are lacking. Nevertheless, a two-step clinical trial in CHI was recently proposed. The first step is to define the challenge virus and to establish the viral-host kinetic picture in the absence of vaccination7. Existing chimeric mice with humanized livers that lack an adaptive immune response have the potential to serve as a preclinical model for studying early HCV-host interactions11,12,13 for any candidate virus stock. In addition, the susceptibility of any inoculum to DAAs will need to be proven prior to administration to human volunteers14. Unless the donor of the challenge virus has a history of successful DAA treatment, this can only be performed in the humanized mouse model. However, detailed analysis of early virus kinetics in humanized mice from inoculation to steady state is lacking, since previous studies were performed during anti-HCV treatment after viral steady state was achieved11,12,13.

Recently, we analyzed and modeled, using agent-based modeling (ABM), early hepatitis B virus kinetics from inoculation to steady state in chimeric urokinase type plasminogen activator (uPA)/severe combined immunodeficiency (SCID) mice reconstituted with humanized livers in the absence of the adaptive immune response15,16. The aim of the current study was to examine the early HCV kinetics in uPA-SCID-chimeric mice and to use both ODE and ABM approaches to explain the observed kinetic patterns. The secondary aim was to compare the ODE and ABM approaches to provide insights into the utility of the two modeling tools for simulating early virus-host dynamics that could be used in the CHI model.

Materials and methods

Experiment setup

Five uPA+/-/SCID chimeric mice with humanized livers (PXB SCID-MhL, with hepatocyte donor: JFC [1 year, male Caucasian, purchased from In Vitro Technologies, Baltimore, MD, USA]) were produced as described previously17. Briefly, 1-2.5 × 105 cells were transplanted into 2–4 week old uPA+/−/SCID mice via intrasplenic injection. Replacement index of human hepatocytes in each mouse was estimated by blood human albumin levels measured by latex agglutination immunonephelometry (LX reagent “Eiken” Alb II; Eiken Chemical Co., Ltd., Tokyo, Japan) as described17. Ten male mice (5 PXB SCID-MhL, human albumin > 9 mg/mL (estimated replacement index > 80%)), and 5 SCID mice without humanized livers (SCID-M) were inoculated intravenously with HCV positive serum (genotype 1a, AB520610) at 1 × 106 RNA copies/animal. Serum HCV RNA titers were measured at 1 min, 30 min, 1 h and 6 hr, and days 1, 2, 3, 4, 6, 8, 10, 12, 14, 21, 28, and 35 post infection (p.i.) using quantitative real-time PCR as previously reported18. Mice were anesthetized with isoflurane using the MK-AT210 anesthesia machine for small animals as recently suggested19. A concentration of 4% was used for induction, 1.5% for blood collection, and 2% for sacrifice, followed by cervical dislocation. All methods were performed in accordance with the relevant guidelines and regulations of Hiroshima University. The experimental protocol was approved by the Ethics Review Committee for Animal Experimentation of the Graduate School of Biomedical Sciences, Hiroshima University (A14-195) in accordance with the ARRIVE guideline.

Ordinary differential equation model for HCV dynamics

We modified our previous model for acute HCV dynamics in chimpanzees9 by excluding adaptive immune response features of immune-related death of infected cells and noncytolytic clearance. The development of the proposed ODE model (Fig. 1) is described in detail in the Supplementary Material. The mathematical model is described by the following ODE model equations.

$$\begin{aligned}\frac{dT}{dt}&=-dT-{\beta}_{e}{V}_{e}T-{\beta}_{m}{V}_{m}T\left(1-\eta\left(t\right)\right) \\ \frac{dI}{dt}&={\beta}_{e}{V}_{e}T+{\beta}_{m}{V}_{m}T\left(1-\eta\left(t\right)\right)-{\delta}_{0}I \\ \frac{d{V}_{e}}{dt}&=-c{V}_{e} \\ \frac{d{V}_{m}}{dt}&=pI\left(1-\epsilon\left(t\right)\right)-c{V}_{m}\end{aligned}$$
(1)

T represents the number of uninfected human hepatocytes (i.e., target cells), I is the number of infected human hepatocytes, Ve is the inoculated viral load, and Vm represents the productive viral load produced by human hepatocytes. The model assumes that target cells have a death rate of d and are susceptible to de novo infection with rate constants \({\beta}_{e}\) and \({\beta}_{m}\) for the inoculated and productive viruses, respectively. Infected cell death/loss rate is represented by \({\delta}_{0}\). A time-dependent blocking of infection and blocking of production is represented by the step functions \(\eta\left(t\right)\) and \(\epsilon\left(t\right)\), respectively. The time when the step functions change from zero to \(\eta\) or \(\epsilon\) is denoted by \({t}_{ir}\). After this time, either blocking of production or blocking of infection occurs, both most likely representing innate immune response mechanisms. These terms capture how innate immunity limits viral spread, either by reducing viral production or preventing new infections in humanized mice lacking adaptive immune responses. Virions (Vm) are produced with rate \(p\), and are cleared with a rate constant \(c\). We assume that Vm and Ve are cleared with the same rate.

Fig. 1
figure 1

Schematic diagram of the ODE model (Eq. 1, two-viral population model). Ve denotes the inoculated virus population and c is the viral clearance rate constant from blood. Ve infects human hepatocytes (termed cells), T, at a rate \(\beta\)e. Uninfected cells die at a rate d. I represents HCV-infected cells, which produce virions at a production rate p, that die at a rate δ0. The virus Vm, produced by HCV-infected cells (I), infects uninfected cells at a rate \(\beta\)m. ε(t) and η(t) represent the possible time dependent efficacies in blocking viral production (0 ≤ ε ≤ 1) and infection (0 ≤ η ≤ 1), respectively, that begins at tir post infection. Uninfected and infected human hepatocyte death were set to 0 (i.e., d = δ0 = 0).

Full size image

Initial model parameter values

We used an initial human hepatocyte count of \({T}_{0}=3.0\times{10}^{8}\)/mL that was previously estimated based on the approximate human albumin (hAlb) level of each mouse and weight in a recent study for uPA-SCID-chimeric mice20. We assumed that the initial count of infected human hepatocytes is \({I}_{0}=0\) and therefore initial \({V}_{m0}=0\). Since there is no evidence of human hepatocyte death or significant proliferation during infection as can be reflected, in part, by stable hAlb levels21, we set d = δ0 = 0. The inoculated virus \({V}_{e0}\) for mouse 1, 2, 3, 4, and 5 was set to measured data at 1 min after inoculation, 1.62 × 106, 1.52 × 106, 0.97 × 106, 0.99 × 106, and 1.31 × 106 copies/mL, respectively. We assume that blocking of infection (\(\eta=0\)) and blocking of production (\(\epsilon=0)\) until time tir post-inoculation.

Parameter estimation for the ODE model

The inoculated virus (Ve) clearance rate \(c\) was computed using a linear regression model with time samples at 1 min, 30 min, 1 h and 6 h from the scikit-learn Python library (Version 0.24.1). We assumed that the sample at 6 h was below 6000 copies/mL, corresponding to the limit of detection for the quantitative real-time PCR, and was set to 6000. Therefore, the computed c is considered as a minimal estimate for \(c\). The infection rate of the inoculated virus could not be estimated with confidence, and therefore it was fixed at \({\beta}_{e}=1\times{10}^{-11}\) mL/virions/day (or \(4.17\times{10}^{-13}\) mL/virions/hour) as described in Supplementary Material C.

We applied the following two steps to estimate the remaining unknown parameters by fitting the ODE model to the HCV kinetics for each mouse. A first pass for fitting \({\beta}_{m},\varepsilon,\) and \(p\) was done using the curve fit feature Berkeley Madonna (Version 8.3.18) with the default values of all the above estimated parameters (initial model parameters and parameter c), with \({V}_{tot}={V}_{e}+{V}_{m}\) used to fit to the data. The best fit values obtained from Berkeley Madonna were used as initial guesses in a Python Jupyter Notebook (Version 6.3.0). It is important to note that during the Python analysis, we kept \({\beta}_{e},{t}_{ir}\) fixed in order to avoid identifiability issues. The remaining parameters \({\beta}_{m},\varepsilon,\) and \(p\) were fitted with the lmfit library (Version 1.1.0) using the least squares method. The lmfit Python library was utilized to determine the 95% confidence intervals for the final three fitted parameters (βm, p, ε).

Agent-based model (ABM) for HCV dynamics

We modified our recently developed ABM15 for acute hepatitis B virus infection in humanized mice to study acute HCV kinetics in these mice. To summarize, we created two types of agents to account for human hepatocytes and virus (V). Depending on the time of interacting with virus, individual human hepatocyte cells can be in one of the following three discrete states: uninfected susceptible target (T), infected cell in eclipse phase (IE) (i.e., not yet releasing progeny virus), or productively infected cell secreting progeny virus (IP) (Fig. 2). An eclipse phase parameter was assigned to individual in-silico human hepatocyte to account for the time duration before virus secretion begins after infection as previously reported in vitro22. The ABM execution is an iterative process where each iteration represents a discrete time step, where 1 step = 1 h. For each iteration, a portion of uninfected susceptible hepatocytes (T) are infected and then proceed to an eclipse phase (IE). After the eclipse phase, IE become productive (i.e., Ip) and start to release a new generation of progeny virus particles through its production cycle (Eq. 2 and Eq. 3).

$$P\left(\tau\right)=\left\{\begin{array}{c}vpaf\bullet\frac{{P}_{st}}{1+{e}^{-\gamma(\tau-\alpha)}}\quad t\le{t}_{ir}\\\frac{{P}_{st}}{1+{e}^{-\gamma(\tau-\alpha)}}\quad t>{t}_{ir}\end{array}\right.$$
(2)

where \(P\left(\tau\right)\) is number of virions produced by infected cells at a production cycle of \(\tau\). \({P}_{st}\) is steady state virus production, \(\alpha\) is number of cycles to reach to 50% of \({P}_{st}\), γ is the steepness of the production curve, and tir is the time when innate immunity starts to take effect (Table 2). To account for the biphasic HCV increase with (or without) a transient decline observed in HCV RNA kinetic data (Fig. 3), we multiplied \(P\left(\tau\right)\) with a constant parameter vpaf, representing a high production rate before tir, to simulate assumed blockage of viral production, i.e., parameter ε in Fig. 2A, or blockage of viral infection i.e., parameter η in Fig. 2A, reminiscent of our previous observation of such biphasic viral increase with (or without) a transient HCV decline seen in chimpanzees9.

$${l}_{\tau}=\delta{e}^{-\omega\tau}$$
(3)

where \({l}_{\tau}\)is interval between production cycles,\(\tau\) is the production cycle,\(\delta\) is scaling factor indicating the initial production cycle length, and \(\omega\) is decay constant. A schematic picture of the modified ABM is shown in Fig. 2.

Fig. 2
figure 2

Schematic diagrams of the ABM model. (A) The human hepatocytes can be only in one of the following three phases at a given time: (i) T, uninfected cells which are termed as target or susceptible cells, (ii) IE, HCV-infected cells in eclipse phase (i.e., not yet releasing virions), and (iii) IP = productively HCV-infected cells (i.e., actively releasing virions). The free virus in blood, V, is composed of infectious and non-infections virions. The parameter ρ represents the fraction of virions that are infectious, β represents the infection rate constant, Ω represents eclipse phase duration, c, represents viral clearance from blood, and P(τ) (Eq. 2) represents virion secretion from IP. Blockage of viral production and viral infection, starting at time tir post infection, are modeled with efficacy ε(tir) and η(tir), respectively. We assume (as in the ODE model, Fig. 1) that uninfected and infected human hepatocytes do not die during the 35 days of experiment. (B) Schematic diagram of viral production cycle for an individual human hepatocyte. P(τ) is the number of virions produced by an infected cell, and l (τ) is the time interval between production cycle (time unit: h). The virions were initially released by IP starting with a high production rate of P(1) virion per cell with a long production cycle of l(1) that gradually reaches a production of P(3) virions per cell with a shorten production cycle of l(3) and then proceeds to P(5) virions per cell per l(5) before virion production increases to reach to a steady state production rate of Pst virions per hour per cell.

Full size image
Fig. 3
figure 3

Kinetics of acute HCV infection in SCID mice. (A) Representative HCV kinetics in SCID-M mice, consistent with the absence of human hepatocytes (M6). Only Phase 1 was observed in all five SCID-M mice. (B) Representative HCV kinetics in 3 (out of 5 shown in Fig. 4) SCID-MhL mice, consistent with human hepatocytes (M1): Phase 1, rapid viral decline; Phase 2, rapid viral increase; Phase 3, transient decline; Phase 4, slower viral increase; Phase 5, steady state. A transient decline was observed between the two increase phases (Phases 2 and 4) around day 5. (C) Representative HCV kinetics in 2 (out of 5 in Fig. 4) SCID-MhL mice (M4) with 4 phases observed (without Phase 3: transient decline as shown in (B): Phase 1, rapid viral decline; Phase 2, rapid viral increase; Phase 4, slower viral increase; Phase 5, steady state. Solid black circles represent measured HCV RNA level and empty circles represent HCV RNA level lower than quantifiable limit (< 6000 copies/mL).

Full size image

ABM parameter estimation

The initial uninfected human hepatocyte number (T0), initial viral loads (Ve0) and HCV clearance rate constant from blood (c) were set as described above and in Table 1. The fraction of infectious HCV in blood is not known, therefore was arbitrary set as ρ = 0.5. Using predefined parameter ranges based on preliminary fits using AnyLogic, we applied a genetic algorithm (GA) to fit the ABM with the mice experimental data (Table 2). The objective of the GA was to find the parameter combination which can achieve the least j-score of mean of ten replicates with different random seeds. The GA was implemented using the DEAP23 evolutionary computation Python framework (specifically24: Chap. 7) and integrated into an EMEWS high-performance computing workflow using EMEWS queues for Python25. The use of high-performance computing resources enables the concurrent evaluation of large numbers of design points, reducing the time to solution. During each iteration of the GA, the best points from the currently evaluated population are selected using a tournament selection method to create a new population. Each of these points was then mated with another according to a crossover probability and, finally, each of the resulting points was mutated according to a mutation probability. At each GA algorithm iteration, the new population is evaluated in parallel, and the evaluation results were gathered. The GA population size was set to 204, the mutation probability to 0.2, the crossover probability to 0.5, and the number of iterations to 20. The runtime for a typical run was 4.8 h using full concurrency on 6 nodes (with 36 cores per node), or about 1037 core hours.

Table 1 Results of the ODE model fit parameters. Parameter \(c\)* was estimated using linear regression via scikit-learn’s least squares method (Version 1.3.0). A first pass for fitting βm, ε, p and tir† was done using the curve fit feature Berkeley Madonna (Version 8.3.18) using the values for the other parameters given in Sect. 2.2.1. Parameters βm, p, and ε were fit in Python with 95% confidence intervals in brackets. #Value for p for M1 was estimated without confidence. The upper bound for the confidence interval was infinity. tir† could not be fit with confidence, hence we had to base our estimates with Berkeley Madonna. \({\beta}_{e}=1\times{10}^{-11}\) mL/virions/day (\(4.17\times{10}^{-13}\) mL/virions/hour).
Full size table
Table 2 Best estimated ABM parameters using genetic algorithm (GA). *: The initial parameter ranges were obtained based on preliminary ABM fits done using AnyLogic. The objective function of GA is to minimize the difference between experimental data and simulated results. J-score represents the mean squared error (MSE) of ten replicates with different random seeds for a parameter combination. #: Best model fits, i.e., with the least J-scores, are shown in Fig. 4. **: The parameters min and max ranges of the GA fits represent top 50 J-scores from a total number of 2524, 2503, 2500, 2619, and 2597 simulation fits for mice 1 (M1), 2 (M2), 3 (M3), 4 (M4), and 5 (M5), respectively. NA, not applicable since minimum = maximum. ε representing efficacy in blocking viral production was calculated using the formula: ε = (vpaf - Pst)/vpaf.
Full size table

Results

HCV RNA kinetics

In both SCID-M (absence of human hepatocytes) and SCID-MhL (humanized hepatocytes) mice, HCV rapidly declined from ~ 6 log10 RNA copies/mL to below the assay quantifiable limit (< 3.8 log10 RNA copies/mL) during the first 6 h following inoculation (Fig. 3). There was no difference in HCV decline rate during the first 6 h p.i. between SCID-M and SCID-MhL mice. A productive infection was established in SCID-MhL mice (Fig. 3B and C, Fig. 4), but not in SCID-M mice, following the initial 6 h time point. Overall, the infection followed a consistent pattern of 5 distinct phases. After the initial decline in viral load (Phase 1) within the first 6 h, there was a rapid increase in HCV until day 4–5 p.i. (Phase 2). This was followed by a slower increase in viral load (Phase 4). Mice 1, 2, and 5 exhibited a transient decline (Phase 3) between the two periods of increase (Phases 2 and 4), occurring around days 4 to 5. The viral load stabilized at an average of 8.51 ± 0.17 log10 RNA copies/mL 10 to 20 days p.i., for all 5 mice (Phase 5), which was maintained throughout the remainder of the study period.

Fig. 4
figure 4

Comparison of model calibration between ODE approach (dashed lines) and ABM approach (solid lines) in all five liver-humanized (SCID-MhL) mice. Serum HCV RNA kinetics were depicted using solid black circles. Green curves represent the percentage of uninfected cell (T), yellow curves represent the percentage of cells in eclipse phase (IE), red curves represent the percentage of cells in the productive phase of infection (IP), and blue curves represent the HCV RNA level (V). The optimal parameter combination was incorporated into the ODE (Table 1) and ABM (Table 2) models to obtain the HCV kinetic trajectory for each mouse by assuming blocking viral production (denoted as ε). HCV RNA levels lower than the quantifiable limit (< 6000 copies/mL) are shown using empty circles.

Full size image

ODE modeling results

The basic model with one virus population (Supplementary Material A, Fig. S1A) or with an eclipse phase (Supplementary Material B, Fig. S1B) could not fit the observed initial rapid viral decline (Fig. 3B and C, Phase 1).​ To explain the initial rapid decline, a model with two virus populations (Eq. 1) was developed. This model reproduced the initial rapid viral decline well (Fig. 4and Fig. S1C), assuming that the inoculated virus was less infectious (60 to 260-fold, Table 1) compared to newly produced virions from infected human hepatocytes.

After the initial rapid viral decline that reached undetectable RNA levels (Fig. 3B and C, Phase 1), the virus resurged in a biphasic manner, i.e., a rapid phase during the first 2 to 6 days p.i. followed by a slower phase. A transient decline was observed in 3 mice in between these phases, and the virus eventually stabilized at high steady state levels in all 5 mice (Figs. 3 and 4). Modeling suggested that a transient decline that resulted in biphasic viral increase (Fig. 3B), can be best explained by assuming blockage in viral production (Fig. S2, bottom row) rather than blocking of infection (Fig. S2, top row) (see Supplementary Material D for more details). Therefore, we assumed that blocking viral production by an innate immune response was the main mechanism to explain the biphasic viral increase across all mice, reminiscent of the observed biphasic increase in chimpanzees that was attributed to an endogenous type I interferon response9. While we cannot completely exclude the possibility that blocking of infection contributed to the transient viral decline (Phase 3), blocking of infection was fixed to zero in the model because it cannot be estimated simultaneously with blocking of viral production.

The ODE model provided HCV kinetic parameters that are detailed in Fig. 4, blue dashed lines and Table 1. Modeling indicated that the median half-life of HCV in the serum of mice was 48 [min-max: 45–49] minutes. Modeling estimated a median efficacy ε = 92% [min-max: 91–95%] in blocking viral production that occurred at time tir = 2.8 [1.7–3.5] days p.i. The median virion production rate was estimated as 11.2 [8.1–27.1] virions/cell/hr and 0.9 [0.7–1.4] virions/cell/hr before and after tir, respectively.

ABM results

The ABM reproduced the multiphasic HCV kinetic patterns observed in humanized mice (Fig. 4, blue solid lines). The estimated ABM parameters for all mice are shown in Table 2. Interestingly, the median infection rate β = 18.5 × 10− 11 [min-max: 13.7 × 10− 11 – 22.2 × 10− 11] cell/h is similar for all 5 mice. The ABM predicts a variable eclipse phase (Ω) ranging from 1 to 52 h with a median minimal value Ωmin = 2 [1–2] h and a median maximal value Ωmax = 39 [38–52] h (Fig. S3, shaded box), reminiscent of the ~ 2-day delay before JFH-1 virus increased post infection in Huh-7.5.1 cells26.

The ABM predicts that after the eclipse phase, viral release from productively infected cells starts slowly with a long production cycle of 18 virions [14–150] per 11 h [9 − 12] that gradually decreases to 18 virions [14–150] per hour after ~ 1–2 days, resulting in increased viral production (Fig. S3). The eclipse phase durations did not affect the amount of virion production or the gradual decrease in the duration of the production cycle (Fig. S3, shaded box).

Two parameters were incorporated into the ABM (as in the ODE modeling approach) to account for blocking of viral production (ε) or infection (η), respectively, at time tir post infection to predict the observed biphasic viral expansion. Modeling results indicated that blocking of viral production provides a better goodness of fit (i.e., lower mean squared error as shown in Table S1) compared to solely blocking of infection (Fig. 4, and Fig. S2). While we cannot completely rule out an effect in slowing infection due to identifiability issues, to estimate simultaneously both ε and η, we set η = 0. Modeling results indicate a median efficacy ε = 94.3% [92.6–99.3%] in blocking viral production that started at a median tir = 4.1 [4.1–4.9] days p.i., after which the virion production cycle decreased to 1 virion/cell/h, where it remained until viral steady state was reached (Table 2).

Comparing the ODE and ABM approaches

A detailed comparison of the ODE and ABM approaches and predictions is summarized in Table S2. Differences in the two modeling approaches were particularly relevant regarding the eclipse phase and infected cell productivity. Based on an early study regarding replication of HCV in cell culture22, an eclipse phase was detected before virion production. The ABM was better able to capture the specific dynamics and timing of the eclipse phase and to simulate the decreasing viral production cycle as HCV production increased in Phase 2. The dynamics and timing of the eclipse phase were not well represented in the ODE due to the assumption of homogeneity and averaging. Therefore, the ABM was used to predict the eclipse phase and subsequent virion production cycles (Table 2 and Fig. S3). The ODE model suggested that the inoculated virus and the virus produced by infected human hepatocytes represented two different viral populations, where the former was 100-fold less infectious than the latter (Table S2). A potential advantage of the ODE is that it did not require that infected cells produce an integer value of virions per average unit time. In addition, the ABM required substantially greater computational cost than the ODE. The models yielded slightly different predictions such as a higher rate of infection of uninfected hepatocytes for the ODE and a higher efficacy of immune mediated blocking of viral production for the ABM. Both models indicated that immune mediated blocking of viral infection cannot explain the observed biphasic viral increase in all humanized mice (Fig. S2).

Comparison of analysis and modeling HCV kinetics in humanized mice and chimpanzees

Observed kinetics and model parameters for HCV replication in vivo were compared between the chimpanzee model9 and the humanized liver mouse model (Table 3). While the majority of chimpanzees were infected with ≤ 1170 HCV RNA copies, the humanized mice were inoculated with a higher viral load (~ 1 × 106 RNA copies) and had more frequent data sampling during the first 6 hr p.i. allowing us to capture the rapid viral decline and estimate HCV t1/2 of 48 min. An earlier (12–22 vs. 28–56 days) and higher viral plateau (8.5 vs. 6.1 log cp/mL) was reached in humanized mice compared to the chimpanzees.

Table 3 Comparison of parameter values obtained from chimpanzee9 and mice ODE models.
Full size table

Both in vivo models showed that the HCV increase p.i. was biphasic with a transient viral decline in between, in most cases. In silico modeling in both in vivo models predicted an efficacy in blocking viral production of about 90%, that started several days p.i. The chimpanzee and humanized mouse models were similar in terms of the observed HCV biphasic increase, the timing of the transient viral decline, and the predicted blocking of viral production and its efficacy in spite of the uncertainty regarding the number of susceptible cells (hepatocytes) in the chimpanzee model along with a ~ 3 order of magnitude lower estimated infection rate in the chimpanzee model compared to the humanized mouse model.

Discussion

To add valuable layers of understanding to the complex interplay between HCV and the host, we applied ODE and ABM approaches to examine the early HCV kinetics from inoculation to steady state in uPA-SCID-chimeric mice with humanized livers. While there are differences in the modeling approaches, both models can explain the observed viral kinetics and provide important insights into the dynamics of acute HCV infection. Modeling indicated that the median HCV half-life (t1/2) equals 48 min in humanized mice, consistent with the HCV t1/2 estimated in patients undergoing liver transplantation27. After a rapid viral decline, during which newly infected hepatocytes were not producing progeny virus, multiphasic HCV kinetic patterns were observed in the humanized mice, which were consistent with our previously reported kinetics in chimpanzees after HCV inoculation and before seroconversion9. Moreover, the transient decline resulted in a biphasic viral increase seen in the humanized mice can be explained by a partial blockage of virion production possibly due to an early stage innate immune response in the human liver cells, consistent with the observed biphasic increase in chimpanzees that was shown to correlate with the induction of intrahepatic 2OAS-1 mRNA expression, a type I interferon-induced gene28.

The ODE model was adapted from a prior chimpanzee model9. Qualitatively, both the humanized mouse and chimpanzee models exhibit similar features, although the speed of the kinetics differs. As seen in chimpanzees before seroconversion, the virus in the mice had a biphasic increase with a transient decline in between, before HCV levels reached a steady state consistent with chronic infection. Compared to the chimpanzees, the mice had a transient decline 3–12 days earlier and reached steady state 26–34 days earlier (Fig. 49). Additionally, both the mice and chimpanzee models suggest that the biphasic increase is due to blocking of HCV production, with a similar efficacy of 90–95% in mice and 85–95% in chimpanzees, rather than a blocking of infection. The mechanism for blocking of virion production in chimpanzees was suggested to be through an endogenous type I interferon response9 and was confirmed through analysis of intrahepatic cytokines28. The humanized mouse model provides a means to further explore the nature of blocking HCV production in human hepatocytes, as hepatocytes can mount an innate immune response and produce type 1 interferons but lack an adaptive response. Lastly, we found that the HCV steady state level was about 100-fold higher, at an average of 8.5 log10 copies/mL in mice compared to 6.1 log10 copies/mL in chimpanzees. Higher levels of HCV at steady state (chronic infection) could reflect the lack of an adaptive immune response in the mouse model. In chimpanzees, HCV titers decreased exponentially following ALT elevations and seroconversion to HCV-specific antibodies9, which coincide with T cell infiltration in the liver and increases in intrahepatic cytokine levels28,29, indicating that the adaptive immune response impacts viral replication in chronic infection. The decrease in HCV RNA titers coincident with ALT elevations in chimpanzees has also been observed in human infections8. Overall, the similarity between the two models suggests this mouse model is well-suited to simulate early acute HCV infection.

As the only readily available animal model of acute HCV infection, the liver-humanized mouse model will almost certainly play a critical role in the safe development of a CHI model. Any proposed candidate viruses will have to exhibit standard and reproducible acute phase kinetics and must be shown to be susceptible to DAAs. It is straightforward to acquire an HCV sample from an infected person before successful DAA treatment. However, obtaining acute phase kinetics from patients is impractical since recognition of acute HCV is uncommon except in immunosuppressed solid organ transplant recipients from HCV-infected donors30. In addition, it has been acknowledged that using virus isolated from a chronically infected individual will introduce further complications, such as the presence of immune complexes that markedly reduce the viral infectivity levels and thus require a larger inoculum to be used14. An alternative approach would consist of inoculating a human volunteer with RNA transcribed in vitro from an infectious clone14. Samples would then be collected from the volunteer to provide the final inoculum for subsequent studies in the CHI model. However, prior to this, studies would have to be performed in the humanized mouse model to verify the infectiousness of the in vitro transcribed RNA and the susceptibility of the virus to DAAs. Indeed, a previous mathematical modeling study in humanized mice during anti-HCV treatment evaluated treatment efficacy and predicted drug mode of action11. Our modeling data provide support for the use of the humanized mouse model to assess the characteristics of candidate challenge viruses, regardless of the source.

As recently proposed7, the first step in developing a CHI model is to characterize the kinetics of viral replication and the immune response in the absence of vaccination. A dose of > 1000 HCV RNA copies, similar to that used in chimpanzee infections, is anticipated to establish consistent HCV infection and subsequent immune responses in humans14. Our modeling data from the humanized mice indicate that there could be an effect of infectious dose on viral kinetics (Fig. S4). Both models predict a time delay in reaching steady state of early viral kinetics in the mouse model with low titer inoculum. Previous studies have shown that infectious dose does not impact the outcome or acute phase viral titers in chimpanzees28,31. However, the blood volume to viral particle ratio would be much lower in the mouse model than the chimpanzee model, which may influence the impact of lower titers in the mouse system as previously reported during acute HBV infection15,16. To define the infectious dose, a small-scale inoculation study in a few participants of the CHI model has been suggested14,32. In such a study one volunteer would be infected at a low dose, and after infection is demonstrated, a small number of additional volunteers (3–4) would be infected14,32. Using the collected viral kinetics in the inoculation study in volunteers, modeling approaches developed in the current study could be used to test the reproducibility of viral replication in the CHI model.

To characterize the viral kinetics in the inoculation study (i.e., sentinel cohorts), blood sampling is recommended two times within day 1 after HCV challenge and then at days 3 and 7 postinfection for the pre-viremia phase7,33. Subsequently the measurements of HCV RNA levels could be measured twice weekly for 4 weeks and then weekly until spontaneous clearance or treatment initiation7. Although the inoculation study mentioned above would provide further information on the frequency of blood sampling as more data are accumulated, our modeling results show that a transient reduction between two phases of viral increase occurred within the first week post-infection (4.6 ± 0.4 days), which suggests that more frequent sampling especially around this time (~ day 5 postinfection) probably is required to provide a more precise estimation regarding the early HCV RNA kinetics.

Our study has several limitations. First, HCV RNA levels were only measured in the serum of mice, and thus more experimental studies would be needed to investigate the effect of innate immune response, such as measuring the intrahepatic levels of interferon response genes such as 2OAS-1 mRNA expression as was previously done in the chimpanzee model28. Second, the data for the number of HCV-infected cells is not available because liver biopsies cannot be obtained during mouse infections without sacrificing the mice. If the number of infected cells could be measured at the initial stage of infection such as 6 h post infection, the fitting process of both ODE and ABM can be improved and therefore add another layer of model confidence. Third, we assumed in both ODE and ABM models no death of human hepatocytes nor proliferation based on previous reported stable hAlb levels during infection21. However, while the levels of hAlb indicate the overall extent of the mass of the human portion of the liver chimera and persistence of human hepatocytes, the presence of cytolytic clearance of human hepatocytes (and hepatocyte proliferation) cannot be excluded without other measured markers such as alanine transaminase (ALT). Lastly, the ODE model suggested two distinct virus populations with different infectivity rates, whereas the ABM model proposed a single virus population that releases virions after the eclipse phase, followed by subsequent virion production cycles. The ABM estimated an eclipse phase duration of 1–52 h, consistent with in vitro experiments22, suggesting that ABM may be a more biologically-relevant model, as the two-virus population observed in ODE (without an eclipse phase) could be an artifact of the mean-field ODE modeling approach. Further experiments are necessary to verify the viral-host predictions suggested by ODE and ABM.

In conclusion, our study demonstrates the power and utility of using modeling approaches to gain comprehensive insights into the complexities of HCV-host interactions and supports the use of the liver-humanized mouse model for the CHI model development. Detailed viral kinetic data from CHI studies will provide the basis to extend and refine the mathematical models to dissect the impact of the innate and adaptative immune responses on early HCV infection in humans and to provide a guide to vaccine development and assessment.