Type 1 interferons such as interferon-alpha (IFNα) inhibit replication of Human immunodeficiency virus (HIV-1) by upregulating the expression of genes that interfere with specific steps in the viral life cycle. This pathway thus represents a potential target for immune-based therapies that can alter the dynamics of host-virus interactions to benefit the host. To obtain a deeper mechanistic understanding of how IFNα impacts spreading HIV-1 infection, we modeled the interaction of HIV-1 with CD4 T cells and IFNα as a dynamical system. This model was then tested using experimental data from a cell culture model of spreading HIV-1 infection. We found that a model in which IFNα induces reversible cellular states that block both early and late stages of HIV-1 infection, combined with a saturating rate of conversion to these states, was able to successfully fit the experimental dataset. Sensitivity analysis showed that the potency of inhibition by IFNα was particularly dependent on specific network parameters and rate constants. This model will be useful for designing new therapies targeting the IFNα network in HIV-1-infected individuals, as well as potentially serving as a template for understanding the interaction of IFNα with other viruses.
Citation: Browne EP, Letham B, Rudin C (2016) A Computational Model of Inhibition of HIV-1 by Interferon-Alpha. PLoS ONE 11(3): e0152316. https://doi.org/10.1371/journal.pone.0152316
Editor: Hans A. Kestler, University of Ulm, GERMANY
Received: October 9, 2015; Accepted: March 11, 2016; Published: March 24, 2016
Copyright: © 2016 Browne et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: All raw data files are available at the following URL: https://github.com/bletham/IFN.
Funding: This work was supported by an Army Research Office grant to Cynthia Rudin (W911NF-15-1-0155). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
Around 65 million people worldwide have been infected with Human immunodeficiency virus (HIV-1) . Although progress has been made in mitigating disease with antiviral chemotherapy, a protective vaccine has proved elusive, and other approaches are still needed. Furthermore, permanently eliminating virus from patients undergoing therapy has been difficult due to the existence of latently infected reservoirs that are resistant to standard antiviral therapy . Another possible approach to treating HIV-1 infection is to alter aspects of virus-host dynamics by targeting host pathways that inhibit or enhance infection. For this to be successful, a deep understanding of the dynamics underlying how specific host pathways interact with HIV-1 will likely be required.
The application of mathematical modeling to HIV-1 dynamics during acute and chronic infection has been highly successful at improving our understanding of the basic features of clinical infection. In particular, fundamental insights into the response to antiviral therapy, and the existence of multiple long-lived virus reservoirs have been revealed [3–6]. In early models, the extent of infection was typically limited by target cell abundance, although more recent HIV-1 models have also considered the impact of virus-specific CD8 T cells [7,8]. However these models have not yet included the impact of the host innate immune system.
A key component of the host innate response to HIV-1 infection is the type 1 interferon (IFN) system [9,10]. In humans, type 1 IFNs consist of a family of related cytokines including 13 subtypes of IFNα, and two subtypes of IFNβ, that are secreted in response to stimulation of microbe-sensing pattern-recognition receptors such as Toll-like receptors (TLRs), RIG-I-like receptors (RLRs) and NOD-like receptors (NLRs) . Type 1 IFNs bind the IFNα receptor (IFNAR) and activate phosphorylation of the signaling molecules STAT1 and STAT2, which then bind to Interferon regulatory factor 9 (IRF9) to form the Interferon-stimulated gene factor 3 (ISGF3) complex . ISGF3 then binds to conserved Interferon-sensitive response elements (ISREs) found upstream of interferon-sensitive genes (ISGs) . Dozens of ISGs are upregulated by IFNα, the function of which only a few are clearly understood . Overall, IFNs create an antiviral state that can either prevent de novo infections, or inhibit later stages of virus replication in cells, such as assembly and egress.
IFNα is detectable in the plasma during acute HIV-1 infection, and this cytokine is predominantly secreted by plasmacytoid dendritic cells (pDCs). pDCs detect HIV-1 via the single-stranded RNA sensor TLR7, and secrete high levels of IFNα due to constitutive expression of the Interferon regulatory factor 7 (IRF7) transcription factor [15,16]. A number of ISGs have been shown to have anti-HIV-1 activity . In particular, Myxovirus resistance protein 2 (MX2) inhibits early stages of infection , while other ISGs such as Apolipoprotein B mRNA editing enzyme, catalytic polypeptide-like 3G (APOBEC3G), Tripartite motif-containing 22 (TRIM22) and Tetherin target later stages of infection, such as virus release or the infectivity of virus particles [19–21]. Furthermore, IFNα inhibits HIV-1 replication in tissue culture and blockade of IFNα during Simian immunodeficiency virus (SIV) infection leads to higher virus levels in vivo .
Due to its HIV-1-inhibiting properties, IFNα has attracted interest as a therapeutic target for HIV-1 infection. However, treatment of HIV-1 patients with recombinant IFNα has produced inconsistent and disappointing results, with only modest effects on virus levels being observed [22–24]. However, since the structure of the IFNα inhibitory network, as well as the parameters that regulate its activity, are poorly understood, a better understanding of the underlying dynamics of this response may lead to improved and more effective IFNα-based therapies. For example, dynamical models may help pinpoint which molecular components of the IFNα inhibition network will achieve the most potent or most durable results. Furthermore, modeling approaches have previously been applied to the interaction of Hepatitis C virus (HCV) with pegylated IFNα, and have yielded valuable insights [25–27].
To achieve this goal, we simulated the inhibition of HIV-1 by IFNα using a dynamical system modeling approach, and tested this model in a well-defined experimental system. In our model, IFNα interacts with both HIV-1-infected and uninfected cells to induce a reversible state of blocked infection, and we demonstrate that this model makes testable predictions about how specific network parameters may be targeted to achieve maximal inhibition of HIV-1.
IFNα inhibits spreading HIV-1 infection
To generate an experimental dataset against which different models can be tested, we performed experiments with a tissue culture model of spreading HIV-1 infection. CEM-5.25 cells are a human CD4 T cell line that are susceptible to HIV-1 infection and contain an integrated HIV-1 Long terminal repeat-Green fluorescent protein (LTR-GFP) reporter cassette . Infected cells express GFP due to the viral Tat protein permitting transcription of GFP from the integrated LTR. This allows us to distinguish HIV-1-infected cells (GFP+) from uninfected cells (GFP-). CEM-5.25 cells were infected with HIV-1, and both cell and supernatant samples were then isolated at 24h intervals for 72h post infection. Cells samples were analyzed by flow cytometry to determine 1) absolute cell numbers and 2) the percentage of HIV-1-infected (% GFP+) cells. Infectious virus in the supernatant was quantified by focus-forming assay using GHOST-X4 reporter cells . In infected cell cultures, the percent GFP+ cells, as well as the concentration of infectious HIV-1 in the supernatant, rises exponentially until 3dpi (Fig 1B and 1C). By day 3, total cell levels decline due to infection-induced cytopathic effect (Fig 1A).
A human CD4 T cell line was infected with HIV-1 and the total concentration of cells (A), the proportion of infected (GFP+) cells (B), and the concentration of infectious virus in the supernatant (C), were monitored at 24h intervals. Different concentrations of IFNα were added to the cells at 6h prior to infection, and the effect on total cell concentration (D), the percent infected cells (E), and the concentration of infectious HIV-1 in the supernatant (F), were measured at 24h intervals (only 3dpi data shown). Measurements were taken in quadruplicate and data shown are representative of three independent experiments. Error bars represent the standard deviation of the dataset.
To examine the effect of IFNα on the replication of HIV-1 in this system, different concentrations of IFNα were included in the tissue culture media from 6h prior to infection and maintained throughout the course of the infection. Inclusion of IFNα in the media caused a clear and progressive decrease in the accumulation of infected cells (Fig 1E) and infectious HIV-1 in the supernatant (Fig 1F), as well and an increase in overall cell density at later timepoints (Fig 1D). Interestingly, inhibition of HIV-1 by IFNα exhibits two key features–firstly that the inhibition curve is very broad–with differential inhibition being observed for IFNα concentrations over several orders of magnitude (from 2pg/mL to 20ng/mL), and secondly, that inhibition of HIV-1 increased only minimally above 2ng/mL, suggesting that inhibition saturates at higher IFNα concentrations. 50% inhibition of infection, as measured by infectious HIV-1 concentration at 3dpi, was achieved between at 0.02ng/mL and 0.2ng/mL (Fig 1F).
Model of IFNα inhibitory network
We constructed a dynamical system that models the dynamics of CD4 T cells and their interaction with both HIV-1 and IFNα (Fig 2, Table 1). This model shares a number of features with a model previously used to analyze the IFNα response to influenza infection . In this system, the initial species are HIV-1 (H), CD4 T cells (C) and IFNα (I). HIV-1 can infect CD4 T cells to generate infected cells (CH) at a rate proportional to the concentrations of HIV-1 and CD4 T cells, via the infection rate constant (k5). Infected cells secrete infectious HIV-1 (H) at a constant rate (k6). IFNα can bind to uninfected (C) or infected cells (CH) and convert them into refractory cells (CI) that cannot be infected, or infected-inhibited cells (CHI) that no longer release virus, respectively. The rate of this conversion is determined by the concentrations of susceptible cells and IFNα, as well as the k2 and k8 rate constants. Since the effects of IFNα on cells are reversible, CI and CHI revert at an intrinsic rate (k3) to C or CH respectively. H and I each have their own intrinsic decay rates that were measured separately (k7 and k9) (S1 Fig). Similar to previous models, consumption of virus by infected and uninfected cells is considered to be negligible . For these studies, an immortalized CD4 T cell line (CEM 5.25) was used that divides continuously and has an intrinsic death rate that is negligible when maintained at subconfluency. By contrast, HIV-1-infected cells (CH and CHI) cells die at an intrinsic rate due to the cytopathic effect of HIV-1 infection (k4). All CD4 T cell subsets divide at an intrinsic rate to generate additional cells of the same type. The replenishment rate of all CD4 T cells was modeled with the same rate for all subspecies (k1), which was experimentally determined by parameter fitting.
CD4 T cells are replenished at a rate proportional to the number of cells, leading to exponential growth (k1). HIV-1 particles infect cells (k5), and infected cells die at an inherent rate (k4). IFNα inhibits infection by converting uninfected cells (C) and infected cells (CH) into states in which they cannot be infected (CI), or which no longer release infectious virus (CHI) (k2), and this rate is also governed by a saturating rate constant (k8). Both these states are reversible, and without continued IFNα exposure, CI and CHI revert to C and CH respectively (k3). Infected cells (CH) secrete HIV-1 at a continuous rate (k6). Both HIV-1 and IFNα have their own natural decay rates (k7 and k9 respectively).
One key element of our model was the inclusion of a Michaelis-Menten saturating rate constant (k8) for the conversion of C to CI and CH to CHI. As such, the conversion rate of C to CI and CH to CHI saturates for high I concentrations, due to k8. Previous models of IFNα interaction with virus infection do not include saturation for conversion to an inhibited state , and this therefore represents a novel feature of this model.
Derivation of parameter estimates
First, the biological decay rates for HIV-1 (k7) and IFNα (k9) were experimentally determined separately and assumed to be invariant in the different conditions (S1 Fig). Interestingly, IFNα was found to have a decay rate that was negligible over the timescale studied in these experiments, indicating that its biological activity is relatively stable in tissue culture media, both in the presence and absence of cells (S1 Fig). We then estimated all seven remaining parameters (k1 through k6 and k8) and the initial CD4 T cell count C(0) using the 72h timecourse data. In particular, we estimated these parameters by minimizing the weighted sum of squared distances between each observation and the corresponding model prediction. Using our experimental data, we cannot directly distinguish between individual uninfected subspecies (C and CI) or infected subspecies (CH and CHI). Thus, the observed cell species, GFP+ and GFP- cells, were assumed to represent the sum of CH and CHI, and the sum of C and CI, respectively.
We compared the basic model (Model 1) to several variants of the model to determine which could more accurately fit the dataset, and to identify the features that make important contributions to model accuracy. For Models 1, 3 and 4, the rate of conversion to inhibited cell states saturates due to the inclusion of the k8 rate constant, while for models 2, 5 and 6 we take the conversion rate of C to CI and CH to CHI as linearly proportional to I, as has been done in previous models of IFNα-virus interaction. For Model 3, cell division was restricted to uninfected CD4 T cells that had not been exposed to IFNα (C). For Model 4, we only considered the interaction of IFNα with uninfected CD4 T cells, and ignored inhibition of previously infected cells–meaning that the CHI species was eliminated from the model. For Model 5, the death rate (k4) for the CHI species was modeled as distinct from the death rate of CH, while for Model 6, the interconversion rate constants (k2 and k3) between uninhibited and inhibited states were distinct for infected and uninfected cells. Different model configurations are described graphically in S2 Fig.
The accuracy of the model fits was compared between these six different model configurations (Table 2, S3 Fig). Notably, we found that inclusion of the saturation constant for IFNα-mediated inhibition dramatically improved the accuracy of the fitting, with significantly lower fit error for all model variants with saturating inhibition (Models 1, 3 and 4). Model 1 most accurately simulated overall timecourse data, and became our default model for further analysis. This model was also significantly superior to another variant with saturating inhibition that lacks CHI (Model 4) suggesting that considering the effects of IFNα on infected cells is important. Model 1 also accurately simulated important features of the timecourse data for %GFP- and %GFP+ T cells, as well as the concentration of infectious HIV-1 in the supernatant (Fig 3). The better fit for the saturating models was not simply due to the inclusion of an extra rate constant (k8), because other variants of the linear inhibition model with additional parameters (models 5 and 6) did not exhibit a comparable improvement in fit quality.
Experimental data for the concentration of uninfected (GFP-) cells (A), infected (GFP+) cells (B), and infectious HIV-1 (C) was fitted with a saturating model (Model 1). Open circles represent independent experimental data points, while for each species, the solid black lines represent the model’s best fit. Data points with the same value are shown side by side.
Estimated parameter values for the model rate constants were determined for all six model variants (Table 3, Table A in S1 File). Estimates for many parameters were similar between Model 1 (saturating) and Model 2 (linear). However, compared to Model 1, Model 2 exhibited a significantly lower k2 estimate and a higher k3 estimate. The predicted concentrations for each individual CD4 T cell subspecies were calculated from the best fit for both Model 1 and Model 2. Notably, for most concentrations of IFNα, Model 1 predicts a higher proportion of cells in the inhibited states compared to Model 2, although both models predict that the majority of inhibited cells are uninfected over the range of IFNα concentrations tested (S4 Fig).
To further investigate the performance of our model, we re-plotted the data for HIV-1 levels, percent GFP+ cells, and total cell concentration, at 3dpi, against different IFNα concentrations, and compared to model predictions (Fig 4). Significantly, Model 1 was able to simulate key features of the HIV-1 inhibition curve accurately, including plateaus at high and low IFNα levels, and a broadly sloped inhibitory curve. Furthermore, Model 1 performed better than a linear inhibition model (Model 2) in two out of the three curves, with statistically significantly lower RMSE values for HIV-1 concentrations and the percentage of infected cells.
Dose-response data for the effect of different IFNα concentrations on the concentration of HIV-1 (left panel), the percentage of GFP+ cells (middle panel), and on total cell concentration (right panel) at 3dpi were compared to model generated estimates for Models 1 and 2. Root mean squared errors (RMSE) are shown on the figure panels. Circles represent individual data points. An asterisk besides RMSE values for Model 2 denotes a statistically significant difference in performance relative to Model 1.
To test whether our assumption of zero intrinsic decay for IFNα activity within our experimental system affected the outcome of model fitting, we refitted the data with Model 1 and Model 2 with the half-life of IFNα arbitrarily set to one day, and found that under these conditions, Model 1 still dramatically outperformed Model 2 (Table B in S1 File). Also, we examined whether consideration of HIV-1 consumption by cells affected model fitting, by including an additional term to the HIV-1 decay rate equation, and determined that this had little impact on model fit error (Table C in S1 File).
Differential effects of inhibitory parameters on the potency of IFNα
Using the saturating default model, we examined the sensitivity of biologically significant outcomes of the model to specific parameters—particularly the parameters that regulated the IFNα-related component of the network (k2, k3 and k8). To measure the outcome of parameter scanning, we calculated the HIV-1 concentration at 3dpi as a function of IFNα concentration with different parameter values over a range of 10 fold higher or lower than our estimated values. As expected, increasing the rate at which IFNα converts CD4 T cells to an inhibited state (k2), and decreasing the reversion rate (k3), both individually resulted in lower total HIV-1 amounts at most concentrations of IFNα (Fig 5A). Modulating the saturation rate constant (k8) had negligible effect on HIV-1 levels at higher IFNα concentrations, but modulated the IFNα threshold at which saturation is achieved. Thus, although the inclusion of a saturation constant greatly improved accuracy of the model, the outcome of infection is less sensitive to the value of this parameter than for k2 and k3 at higher IFNα concentrations.
(A) Experimentally measured values for k2 (left panel), k3 (middle panel) and k8 (right panel), were modulated up (blue) and down (red) by 10-fold, and the effect on the outcome of infection determined by running a simulation with these parameter values substituted for the experimentally determined ones, and calculating the concentration of infectious HIV-1 at 3dpi. (B) The effect of modulating values for k2, k3 and k8 on IC50 (concentration required for 50% inhibition) and Vres (the percent HIV-1 replication at maximal IFNα concentration) were determined by scanning parameter values over a 10 fold range above and below our measured value, and calculating the HIV-1 concentration at 3dpi.
To describe the effects of IFNα on HIV-1, two metrics have previously been used–IC50, that is, the IFNα concentration that results in 50% of maximal inhibition, and Vres, which is the percent residual HIV-1 replication at the maximal IFNα concentration . Based on our simulations, we observe a clear differential effect of k2, k3 and k8 on Vres, namely that k2 and k3 strongly affect Vres, while k8 does not (Fig 5B). Interestingly, increased k2 values and decreased k3 values both reduced the IC50 of IFNα, while scaling these parameters in the opposite direction had little effect. By contrast, both higher and lower k8 values affected the IC50, with lower k8 leading to a lower IC50, and higher k8 leading to a higher IC50. Overall, we conclude that Vres is determined by the balance of k2 and k3, while IC50 is determined by the combined effect of k2, k3 and k8. Thus our model makes testable predictions about how these specific network parameters affect IFNα inhibition of HIV-1.
In this manuscript we describe, for the first time, a dynamical model that simulates the interaction of HIV-1 with IFNα, and demonstrate that this model accurately fits an experimental dataset. Also, we have estimated rate constants for how IFNα interacts with CD4 T cells and HIV-1, thereby providing a realistic range in which to perform simulations of infection. A key novel feature of this model is a saturating rate constant for IFNα’s effects on target cells, which significantly increased the accuracy of the model’s fit to the experimental data. In the absence of this feature, the model performed poorly in fitting, and had a profoundly different balance of forward and reverse rate constants for the generation of inhibited cellular states (k2 and k3). Furthermore, we have found that a model in which IFNα affects both uninfected and infected cells fit experimental data significantly better that one in which IFNα only affected uninfected cells. Although the interaction of IFNα with viral infection has been previously studied using modeling approaches [30,32–34], a saturating rate constant for inhibition of infection has not been a standard assumption of these models. A similar study that considered inhibition of influenza by IFNα found that a linear inhibition model, in which IFNα affects uninfected cells only, was able to successfully fit experimental data, suggesting that models of IFNα interaction with viruses may require virus-specific dynamics .
This model could be useful for the design of novel therapies for HIV-1 that target the IFNα pathway. Notably, our model recapitulates the observation that IFNα has a broad HIV-1-inhibition curve, meaning that changes in IFNα concentration over several orders of magnitude have only partial effects on HIV-1 levels. However, sensitivity analysis of the network rate constants indicates that modulation of the k2 and k3 rate constants can lead to a dramatically more potent inhibition of HIV-1 for a given IFNα concentration. By contrast, the value of the saturation rate constant (k8) has a lesser effect on the course of infection for most IFNα concentrations. As such, our model predicts that clinical modulation of k2 and k3 could prove more beneficial than simply boosting total IFNα levels by a similar factor or by modulating k8.
In order to design therapies that can target the biological processes controlled by these rate constants, it will be critical to identify the molecules that govern them. The signaling pathway and transcriptional response to IFNα have been extensively characterized, and the identities of many of the participating molecules have been described . The rate of conversion to inhibited states (k2) could reflect features such as the abundance of the IFNAR receptor, the expression level of IFNα signaling factors such as STAT1/2 and IRF9, as well as the transcription/translation rate for antiviral effector proteins. The reversion rate (k3), by contrast, is likely determined by the off-rate for ISGF3 detachment from the IRSE sequences in the promoters of ISGs, and/or the degradation rate for antiviral effector proteins in the cytoplasm. The requirement for a saturating rate constant (k8) to accurately fit our dataset could reflect any rate-limiting process in the generation of an ‘inhibited’ cell. Factors that contribute to these rate constants could potentially be experimentally identified by manipulating expression of known components of this pathway, and examining the impact on the network behavior. Novel therapies designed to target these rate constants could involve boosting or reducing the expression/activity of these key factors.
The application of this model to clinical data from HIV-1 patients could yield valuable insights into HIV-1 immunity and pathogenesis. HIV-1 infection results in a wide range of outcomes in terms of viral loads, immune responses, and disease progression. If parameter values can be derived for individual patients, these parameters could be examined for correlation with any of these clinical outcomes. For example, do non-progressors and rapid progressors have different parameter sets that can potentially contribute differences in the course of clinical infection? Some evidence suggests that this may indeed be the case—females mount an intrinsically stronger IFNα response than males, and also exhibit stronger early control of HIV-1 [35–38]. The dynamics of early acute HIV-1 infection, and the resulting innate immune response, likely contains valuable information relating to the IFNα response of individual patients that could be analyzed by a version of this model, but unfortunately, data from this phase of infection have been difficult to obtain, since most patient are not diagnosed until they have entered the chronic phase of infection.
An important consideration for IFNα-related therapy is that IFNα may play different and potentially opposing roles at different times during HIV-1 infection. Recent data suggest that while IFNα and pDCs do indeed limit HIV-1/SIV levels during early infection, they may also promote CD4 T cell depletion at later times, possibly due to IFNα-induced upregulation of apoptosis . However, administration of high doses of IFNα to SIV-infected African green monkeys, which naturally tolerate infection, does not enhance CD4 T cell loss or pathogenesis . Furthermore, blockade of IFNα signaling in SIV-infected Rhesus Macaques leads to higher viral loads and more rapid disease progression . Therefore, it is unclear if future models of IFNα’s role in clinical infection will have to incorporate this feature of IFNα’s behavior in order to accurately model the disease process.
Our model should also be considered in the light of some inherent caveats. For example, the model relies on mass-action interactions to simulate the innate immune response to infection. While this may accurately model responses in well-mixed compartments such as the blood, it may not be applicable to immune responses in solid tissues such as lymph nodes or mucosal surfaces. IFNα may also have additional indirect mechanisms of influencing the course of infection, such as helping to promote immune responses by Natural killer (NK) cells or lymphocytes [41,42]. Also, the results of this manuscript rely on an immortalized cell line that may behave differently than primary human CD4 T cells in vivo. Furthermore, HIV-1 employs several countermeasures against inhibitory ISGs, and different virus strains may exhibit differential sensitivity to IFNα .
Currently this model does not consider the contribution of endogenous IFNα to the course of infection. In vivo, the endogenous IFNα response is driven by plasmacytoid DCs (pDCs), and future development of this model to analyze clinical data from HIV-1 patients will likely require the incorporation of pDCs and endogenous IFNα. Our model also makes several assumptions that may not fully reflect details of HIV-1 infection, such as assuming that infected and uninfected CD4 T cells have equivalent division rates. It is also likely that some interferon-induced genes such as SCHLAFEN11 interfere with viral gene expression, which is a mechanism not currently taken into account by our model . Inhibition of viral transcription or translation by IFNα would result in GFP- HIV-infected cells, which are indistinguishable from uninfected cells (C and CI) by the assays used in this study. As such our data may underestimate the total number of HIV-infected cells in the presence of IFNα. Further studies will be needed to determine the extent of this mechanism and its impact on modeling IFNα and HIV-1.
Nevertheless, this study represents the first attempt to analyze the interaction of the innate immune system with HIV-1 from a computational perspective, and demonstrates that quantitative estimates for the parameters that regulate IFNα’s potency can be derived from experimental data. As such, these findings should serve as a valuable starting point for future studies investigating the dynamics of the host innate immune response to HIV-1.
Materials and Methods
Cells and viruses
CEM 5.25 cells and GHOST-X4 cells were obtained from Dan Littman (NYU) [28,44]. These cells were maintained in RPMI or DMEM media respectively with 10% fetal calf serum, glutamine, and penicillin/streptomycin. An integrated LTR-GFP cassette present in the genomic DNA of these cells facilitates expression of Green Fluorescent Protein (GFP) upon HIV-1 infection.
The HIV-1 strain used was NL4-3, a CXCR4-tropic subtype B strain . The pNL4-3 plasmid was obtained from aidsreagent.org. To generate stocks of infectious virus, 10ug of plasmid was transfected into subconfluent 10cm plate of HEK-293FT cells (Invitrogen) using Mirus LT-1 reagent (Mirus Bio). At 4hrs post-transfection, medium was replaced. At 2 days post-transfection, the supernatant was harvested, clarified by low-speed centrifugation, then filtered through 0.45uM filter. The samples were then aliquoted and frozen at -80C. The titer of the stock was then determined by focus-forming assay.
For experimental infections, CEM-5.25 cultures were first exposed to IFNα for 6 hrs, then infected with NL4-3 at a multiplicity of 0.1 focus-forming unit per cell for 1hr. The virus inoculum was then removed and replaced with fresh media containing IFNα. The infected cells were then plated in 96 well plates at approximately 100 cells/uL. At 24h intervals a small fraction of the media (10%) was removed and replaced with media containing fresh IFNα.
Focus forming assays
GHOST-X4 cells were plated at 10000 cells per well in a 96 well plate. A ten-fold dilution series of each supernatant sample was made and 100ul of each dilution was used to innocculate GHOST-X4 cells. At 3dpi, the supernatant was removed and the cells were fixed in 4% paraformaldehyde (PFA). The plate was then analyzed by fluorescence microscopy and the number of GFP+ foci per well was counted. This was then used to calculate the concentration of focus-forming units (FFU) per mL in each of the original supernatant samples.
Cell samples from infected CEM-5.25 cultures were fixed in 2% PFA for 20 minutes, then diluted in phosphate buffered saline (PBS) and analyzed on an Accuri C6 flow cytometer. The total concentration of CD4 T cells, and the percentage of infected cells (% GFP+), was calculated for each original sample.
Model fitting, simulation, and other computational analyses were done using the Python programming language. The system of differential equations was integrated using SloppyCell . Model fitting was done by solving a weighted nonlinear least squares problem. Let xijlt represent the data collected for species i at interferon level j, in replicate l at time point t. Let yijt(k) be the corresponding model prediction for that species, interferon level, and time point, with parameters k. The weighted least squares problem is to minimize the fit error:
The measurement variance, σij2, was estimated by first measuring the variance across replicates at each time point, and then averaging these variances over time points. The nonlinear squares problem was solved using random-restart optimization. For each random-restart run, optimization was doing using the L-BFGS-B optimization routine [46,47]. Optimization was started from 25 randomly selected initial values, and then the final parameters were chosen as those that produced the lowest cost across all 25 runs.
Confidence intervals for the best-fit parameters were found using the profile likelihood method . The parameter confidence intervals reported here are simultaneous 95% confidence intervals.
Model comparison was done using two complementary methods: AIC and nonparametric confidence intervals on fit error. AIC provides a model comparison that incorporates the number of parameters, but requires assuming a normal noise model. Nonparametric confidence intervals were estimated using the percentile bootstrap, so do not require any assumption about the noise distribution. To determine if a difference in fit error between two models was statistically significant at the p<0.05 level, 95% confidence intervals were computed for the difference in fit errors. If this 95% confidence interval contained zero, it indicated that the difference in model fit was not statistically significant at the 0.05 level. If the confidence interval did not contain zero, it indicated that the difference in model fit was statistically significant.
S1 Fig. Calculation of decay rates for HIV-1 and IFNα.
S2 Fig. Alternative model configurations.
S3 Fig. Best fits for different model configurations.
S4 Fig. Estimated concentrations for individual cell species during timecourse infection.
Conceived and designed the experiments: EPB. Performed the experiments: EPB. Analyzed the data: EPB BL CR. Wrote the paper: EPB BL CR.
- 1. Bhutta ZA, Sommerfeld J, Lassi ZS, Salam RA, Das JK. Global burden, distribution, and interventions for infectious diseases of poverty. Infect Dis Poverty. 2014;3: 21. pmid:25110585
- 2. Archin NM, Sung JM, Garrido C, Soriano-Sarabia N, Margolis DM. Eradicating HIV-1 infection: seeking to clear a persistent pathogen. Nat Rev Microbiol. 2014;12: 750–764. pmid:25402363
- 3. Perelson AS, Ribeiro RM. Modeling the within-host dynamics of HIV infection. BMC Biol. 2013;11: 96. pmid:24020860
- 4. Althaus CL, Joos B, Perelson AS, Günthard HF. Quantifying the Turnover of Transcriptional Subclasses of HIV-1-Infected Cells. PLoS Comput Biol. 2014;10: e1003871. pmid:25340797
- 5. Ho DD, Neumann AU, Perelson AS, Chen W, Leonard JM, Markowitz M. Rapid turnover of plasma virions and CD4 lymphocytes in HIV-1 infection. Nature. 1995;373: 123–126. pmid:7816094
- 6. Perelson AS, Neumann AU, Markowitz M, Leonard JM, Ho DD. HIV-1 Dynamics in Vivo: Virion Clearance Rate, Infected Cell Life-Span, and Viral Generation Time. Science. 1996;271: 1582–1586. pmid:8599114
- 7. Stafford MA, Corey L, Cao Y, Daar ES, Ho DD, Perelson AS. Modeling plasma virus concentration during primary HIV infection. J Theor Biol. 2000;203: 285–301. pmid:10716909
- 8. Košmrlj A, Read EL, Qi Y, Allen TM, Altfeld M, Deeks SG, et al. Effects of thymic selection of the T-cell repertoire on HLA class I-associated control of HIV infection. Nature. 2010;465: 350–354. pmid:20445539
- 9. Hardy GAD, Sieg S, Rodriguez B, Anthony D, Asaad R, Jiang W, et al. Interferon-α Is the Primary Plasma Type-I IFN in HIV-1 Infection and Correlates with Immune Activation and Disease Markers. PLoS ONE. 2013;8: e56527. pmid:23437155
- 10. Sandler NG, Bosinger SE, Estes JD, Zhu RTR, Tharp GK, Boritz E, et al. Type I interferon responses in rhesus macaques prevent SIV infection and slow disease progression. Nature. 2014;511: 601–605. pmid:25043006
- 11. Iwasaki A. A Virological View of Innate Immune Recognition. Annu Rev Microbiol. 2012;66: 177–196. pmid:22994491
- 12. Fu XY, Kessler DS, Veals SA, Levy DE, Darnell JE. ISGF3, the transcriptional activator induced by interferon alpha, consists of multiple interacting polypeptide chains. Proc Natl Acad Sci U S A. 1990;87: 8555–8559. pmid:2236065
- 13. Levy DE, Kessler DS, Pine R, Reich N, Darnell JE. Interferon-induced nuclear factors that bind a shared promoter element correlate with positive and negative transcriptional control. Genes Dev. 1988;2: 383–393. pmid:3371658
- 14. Schneider WM, Chevillotte MD, Rice CM. Interferon-Stimulated Genes: A Complex Web of Host Defenses. Annu Rev Immunol. 2014;32: 513–545. pmid:24555472
- 15. Beignon A-S, McKenna K, Skoberne M, Manches O, DaSilva I, Kavanagh DG, et al. Endocytosis of HIV-1 activates plasmacytoid dendritic cells via Toll-like receptor–viral RNA interactions. J Clin Invest. 2005;115: 3265–3275. pmid:16224540
- 16. Prakash A, Smith E, Lee C, Levy DE. Tissue-specific Positive Feedback Requirements for Production of Type I Interferon following Virus Infection. J Biol Chem. 2005;280: 18651–18657. pmid:15767254
- 17. Schoggins JW, Wilson SJ, Panis M, Murphy MY, Jones CT, Bieniasz P, et al. A diverse array of gene products are effectors of the type I interferon antiviral response. Nature. 2011;472: 481–485. pmid:21478870
- 18. Kane M, Yadav SS, Bitzegeio J, Kutluay SB, Zang T, Wilson SJ, et al. Mx2 is an interferon induced inhibitor of HIV-1 infection. Nature. 2013;502: 563–566. pmid:24121441
- 19. Sheehy AM, Gaddis NC, Choi JD, Malim MH. Isolation of a human gene that inhibits HIV-1 infection and is suppressed by the viral Vif protein. Nature. 2002;418: 646–650. pmid:12167863
- 20. Neil SJD, Zang T, Bieniasz PD. Tetherin inhibits retrovirus release and is antagonized by HIV-1 Vpu. Nature. 2008;451: 425–430. pmid:18200009
- 21. Barr SD, Smiley JR, Bushman FD. The Interferon Response Inhibits HIV Particle Production by Induction of TRIM22. PLoS Pathog. 2008;4.
- 22. Asmuth DM, Murphy RL, Rosenkranz SL, Lertora JJL, Kottilil S, Cramer Y, et al. Safety, Tolerability and Mechanisms of Antiretroviral Activity of Peginterferon alfa-2a in HIV-1-Mono-infected Subjects: A Phase II Clinical Trial. J Infect Dis. 2010;201: 1686–1696. pmid:20420510
- 23. Manion M, Rodriguez B, Medvik K, Hardy G, Harding CV, Schooley RT, et al. Interferon-Alpha Administration Enhances CD8+ T Cell Activation in HIV Infection. PLoS ONE. 2012;7.
- 24. Azzoni L, Foulkes AS, Papasavvas E, Mexas AM, Lynn KM, Mounzer K, et al. Pegylated Interferon alfa-2a monotherapy results in suppression of HIV type 1 replication and decreased cell-associated HIV DNA integration. J Infect Dis. 2013;207: 213–222. pmid:23105144
- 25. Talal AH, Ribeiro RM, Powers KA, Grace M, Cullen C, Hussain M, et al. Pharmacodynamics of PEG-IFN alpha differentiate HIV/HCV coinfected sustained virological responders from nonresponders. Hepatol Baltim Md. 2006;43: 943–953.
- 26. Powers KA, Dixit NM, Ribeiro RM, Golia P, Talal AH, Perelson AS. Modeling viral and drug kinetics: hepatitis C virus treatment with pegylated interferon alfa-2b. Semin Liver Dis. 2003;23 Suppl 1: 13–18. pmid:12934163
- 27. Neumann AU, Lam NP, Dahari H, Gretch DR, Wiley TE, Layden TJ, et al. Hepatitis C viral dynamics in vivo and the antiviral efficacy of interferon-alpha therapy. Science. 1998;282: 103–107. pmid:9756471
- 28. Gervaix A, West D, Leoni LM, Richman DD, Wong-Staal F, Corbeil J. A new reporter cell line to monitor HIV infection and drug susceptibility in vitro. Proc Natl Acad Sci. 1997;94: 4653–4658. pmid:9114046
- 29. Cecilia D, KewalRamani VN, O’Leary J, Volsky B, Nyambi P, Burda S, et al. Neutralization profiles of primary human immunodeficiency virus type 1 isolates in the context of coreceptor usage. J Virol. 1998;72: 6988–6996. pmid:9696790
- 30. Pawelek KA, Huynh GT, Quinlivan M, Cullinane A, Rong L, Perelson AS. Modeling Within-Host Dynamics of Influenza Virus Infection Including Immune Responses. PLoS Comput Biol. 2012;8.
- 31. Fenton-May AE, Dibben O, Emmerich T, Ding H, Pfafferott K, Aasa-Chapman MM, et al. Relative resistance of HIV-1 founder viruses to control by interferon-alpha. Retrovirology. 2013;10: 146. pmid:24299076
- 32. Ciupe SM, Ribeiro RM, Nelson PW, Dusheiko G, Perelson AS. The role of cells refractory to productive infection in acute hepatitis B viral dynamics. Proc Natl Acad Sci U S A. 2007;104: 5050–5055. pmid:17360406
- 33. Saenz RA, Quinlivan M, Elton D, MacRae S, Blunden AS, Mumford JA, et al. Dynamics of Influenza Virus Infection and Pathology. J Virol. 2010;84: 3974–3983. pmid:20130053
- 34. Dahari H, Sainz B, Perelson AS, Uprichard SL. Modeling Subgenomic Hepatitis C Virus RNA Kinetics during Treatment with Alpha Interferon. J Virol. 2009;83: 6383–6390. pmid:19369346
- 35. Addo MM, Altfeld M. Sex-based differences in HIV type 1 pathogenesis. J Infect Dis. 2014;209 Suppl 3: S86–92. pmid:24966195
- 36. Meier A, Chang JJ, Chan ES, Pollard RB, Sidhu HK, Kulkarni S, et al. Sex differences in the TLR-mediated response of pDCs to HIV-1 are associated with higher immune activation in infected women. Nat Med. 2009;15: 955–959. pmid:19597505
- 37. Meier A, Chang JJ, Chan ES, Pollard RB, Sidhu HK, Kulkarni S, et al. Sex differences in the Toll-like receptor-mediated response of plasmacytoid dendritic cells to HIV-1. Nat Med. 2009;15: 955–959. pmid:19597505
- 38. Sterling TR, Vlahov D, Astemborski J, Hoover DR, Margolick JB, Quinn TC. Initial plasma HIV-1 RNA levels and progression to AIDS in women and men. N Engl J Med. 2001;344: 720–725. pmid:11236775
- 39. Fraietta JA, Mueller YM, Yang G, Boesteanu AC, Gracias DT, Do DH, et al. Type I Interferon Upregulates Bak and Contributes to T Cell Loss during Human Immunodeficiency Virus (HIV) Infection. PLoS Pathog. 2013;9.
- 40. Jacquelin B, Petitjean G, Kunkel D, Liovat A-S, Jochems SP, Rogers KA, et al. Innate Immune Responses and Rapid Control of Inflammation in African Green Monkeys Treated or Not with Interferon-Alpha during Primary SIVagm Infection. PLoS Pathog. 2014;10.
- 41. Huddlestone JR, Merigan TC, Oldstone MB. Induction and kinetics of natural killer cells in humans following interferon therapy. Nature. 1979;282: 417–419. pmid:503223
- 42. Crouse J, Kalinke U, Oxenius A. Regulation of antiviral T cell responses by type I interferons. Nat Rev Immunol. 2015;15: 231–242. pmid:25790790
- 43. Li M, Kao E, Gao X, Sandig H, Limmer K, Pavon-Eternod M, et al. Codon-usage-based inhibition of HIV protein synthesis by human schlafen 11. Nature. 2012;491: 125–128. pmid:23000900
- 44. Westervelt P, Gendelman HE, Ratner L. Identification of a determinant within the human immunodeficiency virus 1 surface envelope glycoprotein critical for productive infection of primary monocytes. Proc Natl Acad Sci U S A. 1991;88: 3097–3101. pmid:2014229
- 45. Gutenkunst RN, Waterfall JJ, Casey FP, Brown KS, Myers CR, Sethna JP. Universally Sloppy Parameter Sensitivities in Systems Biology Models. PLoS Comput Biol. 2007;3: e189.
- 46. Byrd R, Lu P, Nocedal J, Zhu C. A Limited Memory Algorithm for Bound Constrained Optimization. SIAM J Sci Comput. 1995;16: 1190–1208.
- 47. Zhu C, Byrd RH, Lu P, Nocedal J. Algorithm 778: L-BFGS-B: Fortran subroutines for large-scale bound-constrained optimization. ACM Trans Math Softw. 1997;23: 550–560.
- 48. Venzon DJ, Moolgavkar SH. A Method for Computing Profile-Likelihood-Based Confidence Intervals. J R Stat Soc Ser C Appl Stat. 1988;37: 87–94.