Abstract
Eradicating HIV-1 infection is difficult because of the reservoir of latently infected cells that gets established soon after infection, remains hidden from antiretroviral drugs and host immune responses, and retains the capacity to reignite infection following the cessation of treatment. Drugs called latency-reversing agents (LRAs) are being developed to reactivate latently infected cells and render them susceptible to viral cytopathicity or immune killing. Whereas individual LRAs have failed to induce adequate reactivation, pairs of LRAs have been identified recently that act synergistically and hugely increase reactivation levels compared to individual LRAs. The maximum synergy achievable with LRA pairs is of clinical importance, as it would allow latency-reversal with minimal drug exposure. Here, we employed stochastic simulations of HIV-1 transcription and translation in latently infected cells to estimate this maximum synergy. We incorporated the predominant mechanisms of action of the two most promising classes of LRAs, namely, protein kinase C agonists and histone deacetylase inhibitors, and quantified the activity of individual LRAs in the two classes by mapping our simulations to corresponding in vitro experiments. Without any adjustable parameters, our simulations then quantitatively captured experimental observations of latency-reversal when the LRAs were used in pairs. Performing simulations representing a wide range of drug concentrations, we estimated the maximum synergy achievable with these LRA pairs. Importantly, we found with all the LRA pairs we considered that concentrations yielding the maximum synergy did not yield the maximum latency-reversal. Increasing concentrations to increase latency-reversal compromised synergy, unravelling a trade-off between synergy and efficacy in LRA combinations. The maximum synergy realizable with LRA pairs would thus be restricted by the desired level of latency-reversal, a constrained optimum we elucidated with our simulations. We expect this trade-off to be important in defining optimal LRA combinations that would maximize synergy while ensuring adequate latency-reversal.
Author summary
HIV-1 infection typically requires lifelong treatment because a class of infected cells called latently infected cells remains hidden from drugs and host immune responses and can reignite infection when treatment is stopped. Massive efforts are ongoing to devise ways to eliminate latently infected cells. The most advanced of the strategies developed for this purpose involves using drugs called latency-reversing agents (LRAs), which reactivate latently infected cells, effectively bringing them out of their hiding. Multiple mechanisms are involved in the establishment of latency. Pairs of LRAs targeting distinct mechanisms have been found to synergize and induce significantly higher latency-reversal than individual LRAs. If this synergy can be maximized, latency-reversal can be achieved with minimal drug exposure. Using stochastic simulations of HIV-1 latency, we unraveled a trade-off between synergy and efficacy in LRA pairs. Drug concentrations that maximized synergy did not also maximize latency-reversal. Drug concentrations that yielded higher latency-reversal compromised synergy and vice versa. This trade-off would constrain the synergy realizable between LRAs and guide the identification of optimal LRA combinations that would maximize synergy while ensuring adequate latency-reversal.
Introduction
Combination antiretroviral therapy (cART) for HIV-1 infection can suppress the viral load in infected individuals to below the detection limit but is unable to eradicate the virus [1]. The key obstacle to achieving sterilizing cure is the presence of a reservoir of long-lived latently infected cells that cannot be eliminated by cART [2] and can reignite infection upon the cessation of therapy [3]. Latently infected cells harbor replication-competent integrated HIV-1 genomes that remain transcriptionally silent, escaping the action of antiretroviral drugs and immune recognition [4]. The reservoir is thought to be established soon after infection [5] and is estimated to have a half-life of many years [2, 6]. Cells in the reservoir are thought to be activated stochastically [7] and can reignite infection often within weeks [8] but sometimes years after the cessation of therapy [9], necessitating lifelong therapy. Significant efforts are ongoing, therefore, to quantify the size of the reservoir, define the type and location of cells constituting it, and devise ways of eliminating it [10, 11].
The most advanced of the strategies to eliminate the latent reservoir, called the “shock and kill” strategy, advocates the use of drugs called latency-reversing agents (LRAs) which stimulate HIV-1 transcription in latently infected cells, rendering them susceptible to viral cytopathicity or immune recognition and killing [11–13]. Several classes of LRAs have been developed, each targeting one or more of the mechanisms underlying viral latency [14]. Multiple cellular and viral mechanisms have been implicated in the establishment of latency, including cytoplasmic localization of the transcription factor NF-κB, sequestration of the protein complex P-TEFb involved in transcription, and epigenetic silencing due to acetylation and methylation [10]. Additional mechanisms involving possible hardwiring of latency in the HIV-1 genome [15] as well as the prevention of latency-reversal by the mTOR complex [16] have recently been identified. LRAs called PKC agonists stimulate the PKC pathway leading among other things to the upregulation and enhanced nuclear translocation of NF-κB [17]. Another class of LRAs called histone deacetylase inhibitors (HDACi’s) induces chromatin remodeling, accelerating HIV-1 transcriptional elongation [18]. Other classes of LRAs include histone methyltransferase inhibitors (HMTi’s), DNA methyltransferase inhibitors (DNMTi’s), and bromodomain and extraterminal (BET) domain inhibitors, which induce HIV-1 transcription via other proposed mechanisms [14].
Several LRAs have been tested extensively in vitro and ex vivo and some have reached clinical trials [19–24]. LRAs have induced transient viral production in infected individuals but have failed to lower the size of the latent reservoir [12]. Individual LRAs have been shown ex vivo to be grossly inadequate at reversing latency when compared to the maximal reversal achieved with agents inducing global T cell activation [25, 26]. Furthermore, a single pulse of even the highest activating dose of the latter agents seems to reactivate only a subset of the latently infected cells in vitro [4]. Serial stimulation does seem to increase the fraction of cells reactivated [27], indicating that multiple doses of an LRA may be necessary to achieve the desired latency-reversal in vivo. Recent studies have therefore examined latency-reversal with combinations of LRAs acting via distinct mechanisms and identified, promisingly, combinations that can reactivate cells ex vivo to extents comparable to that achieved with global T cell activation [28, 29]. With the use of multiple drugs, however, toxicities may become limiting. LRAs that act synergistically, reactivating more cells together than expected based on their individual potencies, are therefore of particular interest as they may achieve the desired latency-reversal with the least drug exposure. Indeed, several in vitro and ex vivo studies have identified LRA combinations that act synergistically [28–39]. An important goal that follows is to determine the concentrations of the LRAs in these promising combinations that would lead to the maximum synergy.
In the present study, we employed stochastic simulations of HIV-1 transcription and translation in infected cells, i.e., the so-called intracellular HIV-1 latency circuit, to quantify the activities of individual LRAs and estimate the maximum synergy realizable between pairs of LRAs. Stochastic simulations of the HIV-1 latency circuit have been employed extensively in previous studies and have provided key insights into the mechanisms underlying the establishment of latency and its reversal [7, 15, 40–43]. Importantly, they recapitulate the two distinct fates, viz., productive infection (or activation) and latency, realized by cells following their infection by HIV-1 [7, 15, 40–42]. Furthermore, they have elucidated the role of the viral protein Tat in inducing a stochastic switch from latency to activation [7, 43], identified a potential cause of the slow viral load decline during cART [40], and suggested the intriguing possibility of latency-reversal by enhancing noise in viral transcription [41]. Previous studies, however, have not explicitly considered the role of LRAs. An alternative class of models that considers populations of latently infected cells, often using sophisticated techniques from the theory of branching processes, has been constructed to estimate quantities of clinical interest such as the reduction in the latent reservoir that would allow a desired duration of cART interruption without viral rebound [28, 44–48]. These latter models do not consider intracellular events explicitly, precluding a description of drug synergy. Here, we built on the previous stochastic simulations of the HIV-1 latency circuit by explicitly incorporating steps affected by LRAs. We considered LRAs belonging to the two most promising classes, viz., HDACi’s, which have entered clinical trials [19, 20, 23], and PKC agonists, which have shown synergy with almost every other class of LRAs in vitro and ex vivo [22, 28, 31]. We quantified the activity of individual LRAs by altering the rates of the relevant steps in the HIV-1 latency circuit to match the extent of latency-reversal observed in vitro. Our simulations then captured quantitatively, without any adjustable parameters, the synergy observed when the LRAs were employed together, giving us confidence in our formalism. We then applied our simulations to predict the maximum synergy realizable with the combinations and the corresponding drug concentrations. We found, interestingly, that a trade-off exists between the synergy and the efficacy of LRA combinations that could introduce new limits on their usage. The maximum synergy realizable would have to be constrained by this trade-off in achieving the desired extent of latency-reversal, a consideration that may be important in defining the optimal usage of LRA combinations.
Results
Model of the HIV-1 latency circuit
We considered the following set of events involved in HIV-1 transcription and translation in an HIV-1-infected cell harboring an integrated, replication-competent HIV-1 genome (Fig 1). The host transcription factor NF-κB is produced in the cytoplasm and is translocated to the nucleus [49]. In the nucleus, it reversibly binds the long terminal repeat (LTR) region of the integrated HIV-1 genome to form a complex [50], denoted LTRNF. This complex formation triggers HIV-1 transcription at a low, basal rate [7]. The viral mRNA thus produced is translocated to the cytoplasm, where it is translated. Translation yields the viral protein Tat as well as other viral proteins, P. Tat is translocated back to the nucleus, where it upregulates HIV-1 transcription via multiple mechanisms [51]. We focused here on its role in enhancing HIV-1 transcription by binding to the HIV-1 LTR. The binding of both NF-κB and Tat to HIV-1 LTR has been argued to be essential for enhanced HIV-1 transcription [52]. We therefore let Tat bind LTRNF to form a complex, denoted LTR-Tatd. Several other factors, including PTEF-b, are essential for transcriptional elongation [53], which we did not consider explicitly for simplicity (see Discussion) and based on previous studies where PTEF-b and the other factors are argued to be in stoichiometric excess [7, 40]. (Previous models have denoted LTR-Tatd as PTEF-bd [40].) Tat is typically in its deacetylated form, which in this complex can get reversibly acetylated [7], and is then denoted LTR-Tata. The latter complex triggers HIV-1 transcription at an enhanced rate [7], following which it dissociates into its constituents. The resulting viral mRNA can produce more Tat, which in turn can further accelerate HIV-1 transcription. This positive feedback, which is triggered following a stochastic build-up of Tat beyond a threshold, drives a cell out of latency [7, 43].
LRAs facilitate this reactivation. We assumed that PKC agonists increase the production rate of NF-κB, kNFκB, by a factor ϕPKCa in a dose-dependent manner. HDACi’s increase the HIV-1 transcription rate in the absence and presence of Tat, denoted kBasal and kTransact, respectively, by a factor ϕHDACi, again in a dose-dependent manner. Proteins and mRNA are subject to degradation in the nucleus and the cytoplasm.
We performed stochastic simulations of the above events, constituting the HIV-1 latency circuit, using the Gillespie algorithm [54] (Methods). The parameter values employed are in Table 1.
Table 1. Model parameters and their typical values.
Rate constant | Event | Value | Source |
---|---|---|---|
kNFκB | Production of NF-κB | 9×10−5 molecules s-1 | Fit |
kImpNFκB | Import of NF-κB into the nucleus | 9×10−2 s-1 | [68] |
kOn | Association of NF-κB with HIV-1 LTR | 2.1×10−5 molecules-1 s-1 | [50] |
kOff | Dissociation of NF-κB–LTR complex | 9.9×10−3 s-1 | [50] |
kBasal | Basal transcription of HIV-1 | 6.1×10−3 s-1 | Fit |
kExpmRNA | Export of HIV-1 mRNA to cytoplasm | 7.2×10−4 s-1 | [40] |
kProtein | Translation of HIV-1 proteins | 10−2 s-1 | |
kTat | Translation of Tat | 1.3×10−3 s-1 | |
kImpTat | Import of Tat into the nucleus | 5.1×10−3 s-1 | |
kBind | Association of Tat with NF-κB bound LTR | 1.5×10−4 molecules-1 s-1 | |
kUnbind | Dissociation of Tat–LTR complex | 1.7×10−2 s-1 | |
kAcetyl | Acetylation of LTR associated Tat | 10−3 s-1 | |
kDeacetyl | Deacetylation of LTR associated Tat | 0.13 s-1 | |
kTransact | Tat-induced transcription of HIV-1 | 0.1 s-1 | |
δmRNA | Degradation of HIV-1 mRNA | 4.8×10−5 s-1 | |
δTat | Degradation of Tat | 4.3×10−5 s-1 | |
δProtein | Degradation of viral proteins | 5×10−5 s-1 | |
δNFκB | Degradation of NF-κB | 2.8×10−5 s-1 | [69] |
ϕPKCa | Fold increase in kNFκB due to a PKC agonist | ≥1 | |
ϕHDACi | Fold increase in kBasal and kTransact due to an HDACi | ≥1 |
Basal reactivation of latently infected cells
We first considered the spontaneous reactivation of latently infected cells in the absence of drugs. A cell was assumed to be activated if the level of viral proteins in it rose above a threshold. To define this threshold, we performed simulations in the absence of NF-κB and Tat, where no activation of cells is expected (Methods). The highest protein levels achieved in these simulations were ~130 copies/cell (S1 Fig). We therefore set the threshold at 500 copies/cell, which could only be achieved via NF-κB- and/or Tat-mediated increase in HIV-1 transcription, indicating activation. We now performed simulations of the complete HIV-1 latency circuit (Fig 1; Eqs (1)–(18) in Methods). We set ϕPKCa = 1 and ϕHDACi = 1 to mark the absence of drugs. The simulations captured the two distinct fates achieved by infected cells, latency and activation (Fig 2A). Most of the cells remained latently infected, in consonance with experiments [29, 33]. We performed simulations with different values of kNFκB and kBasal, which remain poorly estimated, to identify conditions that mimicked experimental observations of the fraction of cells activated, fon, in the absence of drugs. In the experimental data we considered (Methods), fon = 0.039 ± 0.008 in 24 h following the start of the experiment [33]. We found that with kNFκB = 9 × 10−5 molecules s-1 and kBasal = 6.14 × 10−3 s-1, our simulations yielded fon = 0.039 ± 0.003 (Fig 2A), in close agreement with the data. We employed these values for further simulations. Increasing kNFκB, kBasal or both increased fon (Fig 2B–2D). We recognized that the combination of parameter values we employed was not unique. Alternative combinations, however, did not alter our key findings (S2 Fig).
Dose-response curves of individual drugs
We next considered the influence of individual drugs by increasing either ϕPKCa or ϕHDACi while keeping all the other parameters fixed. To mimic the influence of PKC agonists, we performed simulations with different values of ϕPKCa > 1 while keeping ϕHDACi = 1. We found that fon increased monotonically with ϕPKCa (Fig 3A). For instance, fon = 0.1 with ϕPKCa = 1.76 and fon = 0.38 with ϕPKCa = 3.87 (Fig 3B and 3C). To map our simulations to experiments, we considered data of fon as a function of the concentration, [D], of the PKC agonist bryostatin-1 [33]. Corresponding to each value of [D] employed, we obtained a value of ϕPKCa from our simulations that yielded fon in agreement with the experiments, thus linking our simulations to the experiments (Fig 3D). We fit a Hill equation (Eq (22)) to this data linking [D] to ϕPKCa and obtained a dose-response curve for bryostatin-1, which allowed estimation of fon also at values of [D] not employed in the experiments. The equation fit the data well (Fig 3D). We found that ϕPKCa rose from 1 when [D] = 0 to ~4.5 when [D] = 10 nM. A further increase in [D] yielded only a marginal increase in ϕPKCa; ϕPKCa appeared to saturate at ~6 (Fig 3D).
To mimic HDACi’s, we performed simulations with ϕHDACi > 1 while keeping ϕPKCa = 1. Again, fon increased monotonically with ϕHDACi (Fig 4A). For instance, with ϕHDACi = 1.24, we found that fon = 0.066 and with ϕHDACi = 1.36, fon = 0.082 (Fig 4B and 4C). We now considered experimental data for the HDACi’s VPA, TSA, and NaBut [33]. Following the procedure above, we obtained dose-response curves for each of these drugs linking their concentrations [D] to ϕHDACi (Fig 4D–4F). The Hill equation again provided good fits to data of all the three drugs. We found interestingly that ϕHDACi saturated to different values for the three drugs. It saturated at ~1.5 for VPA, ~2.2 for NaBut, and ~3 for TSA. (Note that for TSA, ϕHDACi did not saturate in the concentration range studied, but the trend towards saturation was evident. The best-fit parameters of the Hill equation provided an estimate of the saturating value of ϕHDACi) Ignoring toxicities, these fits thus allowed rank-ordering of the HDACi’s in terms of the maximum activation levels they can achieve individually. Accordingly, TSA>NaBut>VPA.
Combinations of PKC agonists and HDACi’s
We next predicted the influence of using bryostatin-1 together with each of the HDACi’s above and compared our predictions with experiments. Two concentrations of bryostatin-1, viz., 1 and 10 nM, and three concentrations of the HDACi’s viz., 1, 2.5, and 5 mM for VPA and NaBut and 100, 200, and 400 nM for TSA, were employed in combination in the experiments [33]. Our dose-response curves above identified the values of ϕPKCa and ϕHDACi corresponding to each of these concentrations. We employed these latter values in our simulations and predicted fon. We found that our predictions were in good agreement with experiments (Fig 5). With 10 nM bryostatin-1 and 2.5 mM VPA, we predicted fon = 0.64, whereas the corresponding experimental observation was fon = 0.67 (Fig 5A). With 10 nM bryostatin-1, our simulations were similarly in excellent agreement with data at other concentrations of VPA (Fig 5A) and all concentrations of NaBut and TSA (Fig 5B and 5C). With 1 nM bryostatin-1, our simulations were again in good agreement with data in combination with VPA and TSA, but tended to under-predict data with NaBut, where the data lay at the edge of the 95% confidence interval of our predictions. Overall, this agreement, without any adjustable parameters, was remarkable and gave us confidence in our model and our predictions. We applied it next to estimate the maximum synergy achievable between bryostatin-1 and each of these HDACi’s.
Synergy between PKC agonists and HDACi’s
We performed simulations with a range of values of ϕPKCa and ϕHDACi and estimated fon and the extent of synergy, β (see Eq (21)), for each pair of values of ϕPKCa and ϕHDACi. To encompass the range in the dose-response curves above, we let ϕPKCa and ϕHDACi both go from 1 to 10. We found that fon rose from its basal value as either ϕPKCa or ϕHDACi increased and eventually reached a value of 1 at sufficiently high values of these parameters (Fig 6A). For instance, when both ϕPKCa and ϕHDACi were >5, fon was nearly 1, indicating 100% activation. (Note that such high values of ϕHDACi are not realizable using the drugs we considered even at the highest doses, precluding such high activation rates.) Further, because increasing either ϕPKCa or ϕHDACi can increase fon, a locus of points on a ϕPKCa versus ϕHDACi plot could be identified that corresponded to a desired fon. For instance, the combination ϕPKCa ≈ 8 and ϕHDACi = 1 yielded fon = 0.9 as did the combination ϕPKCa ≈ 2 and ϕHDACi = 10 and a number of other combinations in between, yielding the locus for fon = 0.9 (Fig 6A). We thus identified loci for several values of fon starting from 0.2 to 0.96 (Fig 6A).
Using the simulations above, we calculated β for each pair of values of ϕPKCa and ϕHDACi following Eqs (19)–(21). To calculate β, the increase in fon over the basal level relative to the increase possible with maximal stimulation (defined by fax in Eq (19)) must be determined. In experiments, maximal stimulation was achieved by exposing cells to agents such as PMA/I, which yielded fon = 0.853 [33]. The mechanisms that prevent 100% activation remain unclear. In our simulations, increasing ϕPKCa or ϕHDACi sufficiently yielded fon = 1. Indeed, under such circumstances, our simulations agreed with a deterministic model of the HIV-1 latency circuit, which predicts 100% activation (S1 Text, S3 Fig). Effects, including stochastic ones, that limit activation following PMA/I exposure may thus exist upstream of the HIV-1 latency circuit we considered (for instance, in the PKC pathway leading to NF-κB synthesis). Here, based on our simulations, we let fon corresponding to the maximal stimulation, termed fon (positive control) in Eq (19), be 1.
We found that β ranged from 1 to ~1.9 as we spanned the range of values of ϕPKCa and ϕHDACi from 1–10 (Fig 6B). Note that β>1 implies synergy, β<1 antagonism and β = 1 Bliss independence. We found that β ≈ 1 when either ϕPKCa or ϕHDACi was very low (~1) or very high (~10). We understood these results as follows. When both ϕPKCa and ϕHDACi were low, the effect of either drug was negligible. The chance that the drugs acted on the same cell, given the strong stochastic effects, was small, precluding any synergy. When at least one of ϕPKCa and ϕHDACi was high, the corresponding drug(s) achieved nearly maximal activation, leaving little room for the other drug to exert any influence. Synergy therefore was again not possible.
The drugs interacted synergistically when ϕPKCa and ϕHDACi were both increased moderately above 1. Here, the PKC agonist increased NF-κB synthesis and its LTR binding, leaving the integrated HIV-1 genome poised for transcription. The HDACi could then accelerate transcription, triggering the Tat-mediated positive feedback and inducing activation. Indeed, we found that β was maximum, at ~1.9, when ϕPKCa and ϕHDACi were both ~2.2 (Fig 6B). β decreased as ϕPKCa and ϕHDACi either increased or decreased from these optimal values, which we visualized using a heat map (Fig 6B).
Drug concentrations yielding maximum synergy
We next considered the specific combinations of bryostatin-1 with VPA, TSA, and NaBut. Using the dose-response curves above, we identified concentrations of bryostatin-1 corresponding to the ϕPKCa values in Fig 6B. Similarly, we identified concentrations of each of the HDACi's corresponding to the ϕHDACi values therein. We thus obtained the extent of synergy achieved at any given concentration of bryostatin-1 used in conjunction with any given concentration of any of the HDACi's. We visualized this dependence of synergy on the concentrations again using heat maps (Fig 7). We could now identify the concentrations that yielded the maximum synergy. We found that 1.83 nM of bryostatin-1 together with 12.5 mM of NaBut yielded the maximum synergy of β~1.9 (Fig 7B). Similarly, the same concentration of bryostatin-1 along with 353 nM of TSA yielded β~1.9 (Fig 7C). With VPA, however, the saturation in the dose-response curve occurred at ϕHDACi ~1.5, which was much lower than the value of ~2.2 corresponding to the maximum synergy. Thus, VPA could only achieve sub-maximal synergy compared to the other HDACi's. We found that the maximum synergy achieved between VPA and bryostatin-1 saturated at β~1.7, which occurred at concentrations of >5 mM and ~1.8 nM, respectively, of the two drugs (Fig 7A).
Synergy-efficacy trade-off
Maximum synergy implies that the "increase" in fon due to the interaction between drugs over that in the absence of any interaction is the maximum. However, the resulting fon despite the maximal increase may not be adequately high. Indeed, we found that the values of ϕPKCa and ϕHDACi that yielded the maximum synergy, β~1.9, corresponded to fon~0.5 (Fig 6B). Higher drug concentrations would thus be necessary to achieve higher fon. The extent of synergy, however, would then be compromised. We visualized this trade-off between β and fon by superimposing the loci of constant fon on the heat map of β (Fig 8A). As fon increased beyond ~0.5, β kept decreasing below 1.9.
Given this trade-off, it is of interest to determine the maximum synergy achievable while ensuring a desired fon. On any locus of fixed fon, we found that β was low when ϕPKCa was high and ϕHDACi was low and vice versa, whereas β was high at intermediate values of ϕPKCa and ϕHDACi. We denoted the maximum β on a locus as βopt, to signify the optimum achieved constrained by fon. With fon = 0.9, for instance, we found that βopt was 1.69, which occurred at ϕPKCa~3.7 and ϕHDACi ~3.2. We thus computed βopt as a function of fon (Fig 8B) and also obtained the corresponding values of ϕPKCa and ϕHDACi (Fig 8C). When fon was low, βopt increased with fon. It reached a maximum at intermediate values of fon and subsequently declined for higher values of fon (Fig 8B). With fon = 0.96, for instance, βopt was ~1.54. As expected, ϕPKCa and ϕHDACi corresponding to βopt rose with fon (Fig 8C). Using the dose-response curves for the individual drugs (Figs 3 and 4), we could now estimate the drug concentrations corresponding to βopt as a function of fon (Fig 8D). Note that drug concentrations could be estimated for values of ϕPKCa and ϕHDACi below the saturation limits in the dose-response curves; where drug effects saturate, the synergy realizable would be lower than βopt. We thus found that to achieve fon = 0.8, for instance, we could use bryostatin-1 at 4.6 nM and TSA at 688 nM, yielding βopt~1.8. Any other combination of concentrations of these drugs that would yield the same fon would have a lower synergy, resulting in greater drug exposure than necessary.
Discussion
Eliminating latency in HIV-1 infection has not been possible so far except possibly with the Berlin patient, who continues to be in remission many years after a successful bone marrow transplantation [55]. Yet, that the influence of the latent reservoir can be limited to a point where adaptive immune responses can prevent full-blown infection has been demonstrated by the post-treatment control achieved by a subset of patients in the VISCONTI trial, who following early cART initiation have maintained undetectable viral load long after cessation of therapy [56, 57]. Lowering the size of the latent cell reservoir thus presents promise of durable control of infection and disease progression if not sterilizing cure. Achieving this lowering of the size of the latent reservoir too has proven a challenge in vivo [11, 12]. Combinations of LRAs targeting distinct mechanisms underlying HIV-1 latency have shown synergy in vitro and ex vivo in reactivating latently infected cells [28–39], presenting a potential avenue for achieving the desired reduction in the latent reservoir in vivo. In the present study, using stochastic simulations that quantitatively captured published in vitro experiments, we unraveled a trade-off between the synergy and the efficacy of LRA combinations that may constrain their in vivo potency and may have to be accounted for in defining optimal LRA combinations.
Our simulations build on the framework for recapitulating HIV-1 latency in silico employed extensively in previous studies [7, 15, 40–42]. The advance our study makes is in incorporating into the framework the specific steps in the HIV-1 latency circuit affected by two prominent classes of LRAs and describing experimental observations of latency-reversal with the LRAs and the associated synergy quantitatively. Specifically, we incorporated NF-κB synthesis, its nuclear translocation, and its binding to the HIV-1 LTR for initiating HIV-1 transcription, so that the effects of PKC agonists like bryostatin-1, which upregulate the PKC pathway and increase NF-κB synthesis [17], could be described. HDACi’s, the other class of drugs we considered, were assumed to affect the rate of HIV-1 transcription [18]. We estimated the influence of individual LRAs on these processes by matching our simulations with corresponding in vitro experiments. Remarkably, without any adjustable parameters, our simulations then predicted quantitatively experimental observations of the extent of latent cell reactivation when the LRAs were used in pairs, giving us confidence in our simulations. We used these simulations to predict the level of synergy expected at a host of concentrations of the drugs not employed in the experiments and identified the maximum synergy achievable.
The trade-off between synergy and efficacy we unraveled implies that at the drug concentrations that yield the maximum synergy, the extent of latent cell reactivation may not be maximal. Increasing drug concentrations could improve reactivation levels, but would compromise synergy. Maximizing synergy would minimize drug levels and hence side effects and costs. The absolute extent of reactivation, however, is likely to be more important clinically, say, for achieving post-treatment control [57]. The synergy realizable would thus have to be constrained by the extent of reactivation desired. We showed that this constrained optimum can be realized for a specified extent of reactivation by the appropriate choice of drug concentrations. Future experiments, employing a range of concentrations of LRAs used in pairs, as has been done recently to elucidate the synergy between the noise enhancer V11 and the PKC agonist prostratin [41], could serve to validate our prediction of this synergy-efficacy trade-off.
Based on our simulations, the drugs synergized because the PKC agonists upregulated NF-κB synthesis, increasing NF-κB bound to LTR, which rendered the HIV-1 genome amenable to transcription, and the HDACi’s increased the rate of this transcription. The drugs together thus led to a much higher level of HIV-1 transcription and latent cell reactivation than achieved by the drugs independently. We recognize that additional mechanisms of synergy between PKC agonists and HDACi’s have been proposed. For instance, PKC agonists may increase the production of P-TEFb and HDACi’s may facilitate the release of P-TEFb from the repressive 7SK snRNP complex [58]. Synergy has also been proposed to arise when HDACi’s increase the stochastic noise in HIV-1 gene expression and PKC agonists increase the mean HIV-1 gene expression level [41]. Tat may also contribute to the observed synergy by helping recruit the PTEF-b complex and other transcriptional elongational factors to the LTR [53, 59] or by increasing the nuclear uptake of NF-κB by freeing it from the inhibitory molecule IκB [60]. We did not consider these latter phenomena explicitly in our simulations. Our goal was to construct a minimal model of the HIV-1 latency circuit that could describe the influence of LRA combinations. That our simulations captured experimental observations quantitatively, without any adjustable parameters, indicated that our formalism was robust; the influence of the phenomena we ignored was either small in the experiments we considered or was suitably subsumed in our simulations.
We recognize that toxicity can be a key limitation in defining optimal LRA combinations. Previous studies have suggested strategies to optimize treatments based on the trade-off between efficacy and toxicity (e.g., [61]). Our study presents an additional factor, the synergy-efficacy trade-off, that is expected to be important in optimizing LRA treatments. Where toxicity is not limiting, the synergy-efficacy trade-off may be the defining factor in optimizing treatments. For instance, in vitro studies report that bryostatin-1 is not toxic up to concentrations of 10 nM [33], well above the concentrations (~2 nM) at which our simulations predict the maximum synergy. Where toxicity is significant, a more comprehensive formalism that integrates the synergy-efficacy trade-off with the toxicity-efficacy trade-off would serve to identify optimal LRA combinations. Constructing such a formalism presents a promising avenue for future studies.
Although in vitro studies, on which our simulations were based, have been widely used to test drug effects and have led to the identification of LRAs and their synergistic combinations (e.g., [29, 41, 61]; reviewed in [62]), they do not recapitulate all the complexities of the scenario in vivo [25]. For instance, the extent of reactivation of latent cells appears much higher in vitro than ex vivo, the latter more akin to the scenario in vivo [25, 26]. Our findings thus remain to be established ex vivo. Using our simulations to mimic ex vivo data was precluded by the large variations in the data introduced possibly by inter-patient differences; even with exposure to global T cell activating stimuli, which should induce maximal latent cell reactivation, a variation of >2 logs in the fold-induction in intracellular HIV-1 RNA and >3 logs in supernatant HIV-1 RNA levels has been observed ex vivo [28, 29]. The origins of these large variations remain to be fully understood. Nonetheless, we applied our simulations to in vitro data of 3 pairs of LRAs and found the trade-off between synergy and efficacy to be relevant to all the pairs, suggesting that our findings are likely to be more widely applicable, extending to ex vivo and in vivo settings and to other classes of LRAs. We therefore expect the synergy-efficacy trade-off to become a potentially important factor in defining optimal LRA combinations.
Methods
Stochastic simulations of the HIV-1 latency circuit
The events constituting the HIV-1 latency circuit (Fig 1) are listed schematically as reactions along with their rate constants in the equations below (Eqs (1)–(18)). Events representing the degradation of the entities involved are also included.
(1) |
(2) |
(3) |
(4) |
(5) |
(6) |
(7) |
(8) |
(9) |
(10) |
(11) |
(12) |
(13) |
(14) |
(15) |
(16) |
(17) |
(18) |
We performed stochastic simulations of the above system of events using the Gillespie algorithm implemented in the tool StochKit [63]. The rate constants were obtained from the literature (Table 1). The production rate of NF-κB, kNFκB, and the basal transcription rate of HIV-1, kBasal, remained uncertain. We fixed them to mimic experiments (see Results). We set the initial copy numbers of all species to zero, except LTR, which we set to 1, recognizing that a vast majority of infected cells harbors single proviruses [64]. Mimicking experiments [33], we ran the simulations for 24 h. If the protein level, P, in a cell crossed a threshold (see below), we considered the cell to have been activated. We performed simulations with 2000 cells and computed the fraction of cells that thus got activated. We repeated the simulations 10 times and obtained the average fraction of cells activated, which we denoted fon. Increasing the number of cells or realizations did not affect our results (S4 Fig).
To define the activation threshold, we performed the above simulations in the absence of NF-κB and Tat. The positive feedback leading to enhanced HIV-1 transcription then does not occur and cells are bound to remain latent. We therefore set the activation threshold to be sufficiently higher than the maximum protein level achieved in these simulations. We ensured that variation in this threshold did not affect our results (S2 Fig).
The influence of LRAs and synergy
We performed simulations with ϕPKCa = 1 and ϕHDACi = 1 to mimic experiments without drugs and with ϕPKCa > 1 and ϕHDACi >1 to mimic the scenario in the presence of PKC agonists and HDACi’s, respectively. To quantify the influence of drugs, we followed previous formalisms [28, 33, 61] and defined fon(no drug) and fon (positive control) as the activation level without drugs and with maximal stimulation. The normalized fraction of cells activated by drug ‘x’ was then given by
(19) |
The expected activation level if the two drugs were to act independently was “predicted” by the Bliss independence model:
(20) |
Here, fax was obtained by setting ϕPKCa > 1 and fay by setting ϕHDACi > 1, respectively. To mimic the use of the drugs simultaneously, we set both ϕPKCa > 1 and ϕHDACi > 1 and obtained the “observed” activation level, faxy,O. To quantify the interactions between the drugs, we computed
(21) |
If β > 1, the drugs act synergistically, whereas if β < 1, they act antagonistically. β = 1 would imply Bliss independence, i.e., the absence of any interaction between the drugs.
Comparisons with data
We considered the previously reported in vitro data on the activation of latently infected cells exposed to the PKC agonist bryostatin-1 and the HDACi’s VPA, NaBut, and TSA [33]. To compare our simulations with the data, we first considered data with single drugs. For each drug concentration employed, we identified a value of ϕ (either ϕPKCa or ϕHDACi) that yielded fon in agreement with the corresponding experimental observation. For bryostatin-1, we thus obtained values of ϕPKCa for the various drug levels employed. Similarly, for each of the HDACi’s we obtained values of ϕHDACi corresponding to the various drug levels employed. We then used each pair of these ϕPKCa and ϕHDACi values in our simulations to predict fon when the drugs were used together.
Optimum synergy
We performed simulations with different combinations of ϕPKCa and ϕHDACi, each spanning the range from 1 to 10, and estimated β (Eq (21)). The combination of values yielding the maximum β yielded the maximum synergy. Simultaneously, the values of fon allowed elucidation of the synergy-efficacy trade-off. To estimate drug concentrations corresponding to any specified β and fon, we constructed dose-response curves for each drug as follows. We employed the estimates of ϕ we obtained above corresponding to the concentrations used in the experiments and fit the modified Hill equation,
(22) |
to the data, where [D] was the drug concentration, using ϕ0 and ϕM as adjustable parameters. Note that the modified Hill equation above is identical in form to the standard E-max model of the influence of drugs [65]. The resulting dose-response curves then allowed us to identify drug concentrations that would yield the desired β and fon, including the maximum synergy. We identified these concentrations for bryostatin-1 and each of the three HDACi’s.
We performed data analysis using Python 2.7 [66] and plotted using Matplotlib 1.5.1 [67].
Supporting information
Data Availability
All relevant data are within the paper and its Supporting Information files.
Funding Statement
This work was supported by the Wellcome Trust/DBT India Alliance Senior Fellowship IA/S/14/1/501307 (NMD). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
References
- 1.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;12:750–64. doi: 10.1038/nrmicro3352 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Finzi D, Blankson J, Siliciano JD, Margolick JB, Chadwick K, Pierson T, et al. Latent infection of CD4+ T cells provides a mechanism for lifelong persistence of HIV-1, even in patients on effective combination therapy. Nat Med. 1999;5:512–7. doi: 10.1038/8394 [DOI] [PubMed] [Google Scholar]
- 3.Joos B, Fischer M, Kuster H, Pillai SK, Wong JK, Boni J, et al. HIV rebounds from latently infected cells, rather than from continuing low-level replication. Proc Natl Acad Sci U S A. 2008;105:16725–30. doi: 10.1073/pnas.0804192105 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Ho YC, Shan L, Hosmane NN, Wang J, Laskey SB, Rosenbloom DI, et al. Replication-competent noninduced proviruses in the latent reservoir increase barrier to HIV-1 cure. Cell. 2013;155:540–51. doi: 10.1016/j.cell.2013.09.020 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Whitney JB, Hill AL, Sanisetty S, Penaloza-MacMaster P, Liu J, Shetty M, et al. Rapid seeding of the viral reservoir prior to SIV viraemia in rhesus monkeys. Nature. 2014;512:74–7. doi: 10.1038/nature13594 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Siliciano JD, Kajdas J, Finzi D, Quinn TC, Chadwick K, Margolick JB, et al. Long-term follow-up studies confirm the stability of the latent reservoir for HIV-1 in resting CD4+ T cells. Nat Med. 2003;9:727–8. doi: 10.1038/nm880 [DOI] [PubMed] [Google Scholar]
- 7.Weinberger LS, Burnett JC, Toettcher JE, Arkin AP, Schaffer DV. Stochastic gene expression in a lentiviral positive-feedback loop: HIV-1 Tat fluctuations drive phenotypic diversity. Cell. 2005;122:169–82. doi: 10.1016/j.cell.2005.06.006 [DOI] [PubMed] [Google Scholar]
- 8.Ruiz L, Martinez-Picado J, Romeu J, Paredes R, Zayat MK, Marfil S, et al. Structured treatment interruption in chronically HIV-1 infected patients after long-term viral suppression. AIDS. 2000;14:397–403. [DOI] [PubMed] [Google Scholar]
- 9.Luzuriaga K, Gay H, Ziemniak C, Sanborn KB, Somasundaran M, Rainwater-Lovett K, et al. Viremic relapse after HIV-1 remission in a perinatally infected child. N Engl J Med. 2015;372:786–8. doi: 10.1056/NEJMc1413931 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Ruelas Debbie S, Greene Warner C. An integrated overview of HIV-1 latency. Cell. 2013;155:519–29. doi: 10.1016/j.cell.2013.09.044 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Margolis DM, Garcia JV, Hazuda DJ, Haynes BF. Latency reversal and viral clearance to cure HIV-1. Science. 2016;353:aaf6517 doi: 10.1126/science.aaf6517 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Rasmussen TA, Tolstrup M, Sogaard OS. Reversal of latency as part of a cure for HIV-1. Trends Microbiol. 2016;24:90–7. doi: 10.1016/j.tim.2015.11.003 [DOI] [PubMed] [Google Scholar]
- 13.Deeks SG, Lewin SR, Ross AL, Ananworanich J, Benkirane M, Cannon P, et al. International AIDS Society global scientific strategy: towards an HIV cure 2016. Nat Med. 2016;22:839–50. doi: 10.1038/nm.4108 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Margolis DM, Archin NM. Proviral latency, persistent human immunodeficiency virus infection, and the development of latency reversing agents. J Infect Dis. 2017;215:S111–S8. doi: 10.1093/infdis/jiw618 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Razooky BS, Pai A, Aull K, Rouzine IM, Weinberger LS. A hardwired HIV latency program. Cell. 2015;160:990–1001. doi: 10.1016/j.cell.2015.02.009 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Besnard E, Hakre S, Kampmann M, Lim HW, Hosmane NN, Martin A, et al. The mTOR complex controls HIV latency. Cell Host Microbe. 2016;20:785–97. doi: 10.1016/j.chom.2016.11.001 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Jiang G, Dandekar S. Targeting NF-kappaB signaling with protein kinase C agonists as an emerging strategy for combating HIV latency. AIDS Res Hum Retroviruses. 2015;31:4–12. doi: 10.1089/AID.2014.0199 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Margolis DM. Histone deacetylase inhibitors and HIV latency. Curr Opin HIV AIDS. 2011;6:25–9. doi: 10.1097/COH.0b013e328341242d [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Archin NM, Liberty AL, Kashuba AD, Choudhary SK, Kuruc JD, Crooks AM, et al. Administration of vorinostat disrupts HIV-1 latency in patients on antiretroviral therapy. Nature. 2012;487:482–5. doi: 10.1038/nature11286 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Sogaard OS, Graversen ME, Leth S, Olesen R, Brinkmann CR, Nissen SK, et al. The depsipeptide romidepsin reverses HIV-1 latency in vivo. PLoS Pathog. 2015;11:e1005142 doi: 10.1371/journal.ppat.1005142 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Elliott JH, McMahon JH, Chang CC, Lee SA, Hartogensis W, Bumpus N, et al. Short-term administration of disulfiram for reversal of latent HIV infection: a phase 2 dose-escalation study. Lancet HIV. 2015;2:e520–9. doi: 10.1016/S2352-3018(15)00226-X [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Gutierrez C, Serrano-Villar S, Madrid-Elena N, Perez-Elias MJ, Martin ME, Barbas C, et al. Bryostatin-1 for latent virus reactivation in HIV-infected patients on antiretroviral therapy. AIDS. 2016;30:1385–92. doi: 10.1097/QAD.0000000000001064 [DOI] [PubMed] [Google Scholar]
- 23.Rasmussen TA, Tolstrup M, Brinkmann CR, Olesen R, Erikstrup C, Solomon A, et al. Panobinostat, a histone deacetylase inhibitor, for latent-virus reactivation in HIV-infected patients on suppressive antiretroviral therapy: a phase 1/2, single group, clinical trial. Lancet HIV. 2014;1:e13–21. doi: 10.1016/S2352-3018(14)70014-1 [DOI] [PubMed] [Google Scholar]
- 24.Delagreverie HM, Delaugerre C, Lewin SR, Deeks SG, Li JZ. Ongoing clinical trials of human immunodeficiency virus latency-reversing and immunomodulatory agents. Open Forum Infect Dis. 2016;3:ofw189 doi: 10.1093/ofid/ofw189 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Bullen CK, Laird GM, Durand CM, Siliciano JD, Siliciano RF. New ex vivo approaches distinguish effective and ineffective single agents for reversing HIV-1 latency in vivo. Nat Med. 2014;20:425–9. doi: 10.1038/nm.3489 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Cillo AR, Sobolewski MD, Bosch RJ, Fyne E, Piatak M Jr., Coffin JM, et al. Quantification of HIV-1 latency reversal in resting CD4+ T cells from patients on suppressive antiretroviral therapy. Proc Natl Acad Sci U S A. 2014;111:7078–83. doi: 10.1073/pnas.1402873111 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Hosmane NN, Kwon KJ, Bruner KM, Capoferri AA, Beg S, Rosenbloom DIS, et al. Proliferation of latently infected CD4(+) T cells carrying replication-competent HIV-1: Potential role in latent reservoir dynamics. J Exp Med. 2017;214:959–72. doi: 10.1084/jem.20170193 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Laird GM, Bullen CK, Rosenbloom DI, Martin AR, Hill AL, Durand CM, et al. Ex vivo analysis identifies effective HIV-1 latency-reversing drug combinations. J Clin Invest. 2015;125:1901–12. doi: 10.1172/JCI80142 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Darcis G, Kula A, Bouchat S, Fujinaga K, Corazza F, Ait-Ammar A, et al. An in-depth comparison of latency-reversing agent combinations in various in vitro and ex vivo HIV-1 latency models identified bryostatin-1+JQ1 and ingenol-B+JQ1 to potently reactivate viral gene expression. PLoS Pathog. 2015;11:e1005063 doi: 10.1371/journal.ppat.1005063 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Quivy V, Adam E, Collette Y, Demonte D, Chariot A, Vanhulle C, et al. Synergistic activation of human immunodeficiency virus type 1 promoter activity by NF-kappaB and inhibitors of deacetylases: potential perspectives for the development of therapeutic strategies. J Virol. 2002;76:11091–103. doi: 10.1128/JVI.76.21.11091-11103.2002 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Reuse S, Calao M, Kabeya K, Guiguen A, Gatot JS, Quivy V, et al. Synergistic activation of HIV-1 expression by deacetylase inhibitors and prostratin: implications for treatment of latent infection. PLoS One. 2009;4:e6093 doi: 10.1371/journal.pone.0006093 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Burnett JC, Lim KI, Calafi A, Rossi JJ, Schaffer DV, Arkin AP. Combinatorial latency reactivation for HIV-1 subtypes and variants. J Virol. 2010;84:5958–74. doi: 10.1128/JVI.00161-10 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.Perez M, de Vinuesa AG, Sanchez-Duffhues G, Marquez N, Bellido ML, Munoz-Fernandez MA, et al. Bryostatin-1 synergizes with histone deacetylase inhibitors to reactivate HIV-1 from latency. Curr HIV Res. 2010;8:418–29. [DOI] [PubMed] [Google Scholar]
- 34.Martinez-Bonet M, Clemente MI, Serramia MJ, Munoz E, Moreno S, Munoz-Fernandez MA. Synergistic activation of latent HIV-1 expression by novel histone deacetylase inhibitors and bryostatin-1. Sci Rep. 2015;5:16445 doi: 10.1038/srep16445 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Jiang G, Mendes EA, Kaiser P, Wong DP, Tang Y, Cai I, et al. Synergistic reactivation of latent HIV expression by ingenol-3-angelate, PEP005, targeted NF-kB signaling in combination with JQ1 induced p-TEFb activation. PLoS Pathog. 2015;11:e1005066 doi: 10.1371/journal.ppat.1005066 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36.Bouchat S, Delacourt N, Kula A, Darcis G, Van Driessche B, Corazza F, et al. Sequential treatment with 5-aza-2'-deoxycytidine and deacetylase inhibitors reactivates HIV-1. EMBO Mol Med. 2016;8:117–38. doi: 10.15252/emmm.201505557 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 37.Chen D, Wang H, Aweya JJ, Chen Y, Chen M, Wu X, et al. HMBA enhances prostratin-induced activation of latent HIV-1 via suppressing the expression of negative feedback regulator A20/TNFAIP3 in NF-kappaB signaling. Biomed Res Int. 2016;2016:5173205 doi: 10.1155/2016/5173205 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38.Cary DC, Fujinaga K, Peterlin BM. Euphorbia kansui reactivates latent HIV. PLoS One. 2016;11:e0168027 doi: 10.1371/journal.pone.0168027 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39.Chen HC, Martinez JP, Zorita E, Meyerhans A, Filion GJ. Position effects influence HIV latency reversal. Nat Struct Mol Biol. 2017;24:47–54. doi: 10.1038/nsmb.3328 [DOI] [PubMed] [Google Scholar]
- 40.Althaus CL, De Boer RJ. Intracellular transactivation of HIV can account for the decelerating decay of virus load during drug therapy. Mol Syst Biol. 2010;6:348 doi: 10.1038/msb.2010.4 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 41.Dar RD, Hosmane NN, Arkin MR, Siliciano RF, Weinberger LS. Screening for noise in gene expression identifies drug synergies. Science. 2014;344:1392–6. doi: 10.1126/science.1250220 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 42.Chavali AK, Wong VC, Miller-Jensen K. Distinct promoter activation mechanisms modulate noise-driven HIV gene expression. Sci Rep. 2015;5:17661 doi: 10.1038/srep17661 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 43.Weinberger LS, Shenk T. An HIV feedback resistor: auto-regulatory circuit deactivator and noise buffer. PLoS Biol. 2007;5:e9 doi: 10.1371/journal.pbio.0050009 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 44.Hill AL, Rosenbloom DI, Fu F, Nowak MA, Siliciano RF. Predicting the outcomes of treatment to eradicate the latent reservoir for HIV-1. Proc Natl Acad Sci U S A. 2014;111:13475–80. doi: 10.1073/pnas.1406663111 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 45.Hill AL, Rosenbloom DI, Goldstein E, Hanhauser E, Kuritzkes DR, Siliciano RF, et al. Real-time predictions of reservoir size and rebound time during antiretroviral therapy interruption trials for HIV. PLoS Pathog. 2016;12:e1005535 doi: 10.1371/journal.ppat.1005535 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 46.Conway JM, Coombs D. A stochastic model of latently infected cell reactivation and viral blip generation in treated HIV patients. PLoS Comput Biol. 2011;7:e1002033 doi: 10.1371/journal.pcbi.1002033 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 47.Rouzine IM, Weinberger AD, Weinberger LS. An evolutionary role for HIV latency in enhancing viral transmission. Cell. 2015;160:1002–12. doi: 10.1016/j.cell.2015.02.017 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 48.Pinkevych M, Cromer D, Tolstrup M, Grimm AJ, Cooper DA, Lewin SR, et al. HIV Reactivation from latency after treatment interruption occurs on average every 5–8 days—Implications for HIV remission. PLoS Pathog. 2015;11:e1005000 doi: 10.1371/journal.ppat.1005000 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 49.Li Q, Verma IM. NF-[kappa]B regulation in the immune system. Nat Rev Immunol. 2002;2:725–34. doi: 10.1038/nri910 [DOI] [PubMed] [Google Scholar]
- 50.Stroud JC, Oltman A, Han A, Bates DL, Chen L. Structural basis of HIV-1 activation by NF-kappaB—a higher-order complex of p50:RelA bound to the HIV-1 LTR. J Mol Biol. 2009;393:98–112. doi: 10.1016/j.jmb.2009.08.023 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 51.Van Lint C, Bouchat S, Marcello A. HIV-1 transcription and latency: an update. Retrovirology. 2013;10:67 doi: 10.1186/1742-4690-10-67 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 52.Doppler C, Schalasta G, Amtmann E, Sauer G. Binding of NF-kB to the HIV-1 LTR is not sufficient to induce HIV-1 LTR activity. AIDS Res Hum Retroviruses. 1992;8:245–52. doi: 10.1089/aid.1992.8.245 [DOI] [PubMed] [Google Scholar]
- 53.Karn J, Stoltzfus CM. Transcriptional and posttranscriptional regulation of HIV-1 gene expression. Cold Spring Harb Perspect Med. 2012;2:a006916 doi: 10.1101/cshperspect.a006916 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 54.Gillespie DT. Stochastic simulation of chemical kinetics. Annu Rev Phys Chem. 2007;58:35–55. doi: 10.1146/annurev.physchem.58.032806.104637 [DOI] [PubMed] [Google Scholar]
- 55.Brown TR. I am the Berlin patient: a personal reflection. AIDS Res Hum Retroviruses. 2015;31:2–3. doi: 10.1089/AID.2014.0224 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 56.Saez-Cirion A, Bacchus C, Hocqueloux L, Avettand-Fenoel V, Girault I, Lecuroux C, et al. Post-treatment HIV-1 controllers with a long-term virological remission after the interruption of early initiated antiretroviral therapy ANRS VISCONTI Study. PLoS Pathog. 2013;9:e1003211 doi: 10.1371/journal.ppat.1003211 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 57.Conway JM, Perelson AS. Post-treatment control of HIV infection. Proc Natl Acad Sci U S A. 2015;112:5467–72. doi: 10.1073/pnas.1419162112 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 58.Bartholomeeusen K, Fujinaga K, Xiang Y, Peterlin BM. Histone deacetylase inhibitors (HDACis) that release the positive transcription elongation factor b (P-TEFb) from its inhibitory complex also activate HIV transcription. J Biol Chem. 2013;288:14400–7. doi: 10.1074/jbc.M113.464834 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 59.Sobhian B, Laguette N, Yatim A, Nakamura M, Levy Y, Kiernan R, et al. HIV-1 Tat assembles a multifunctional transcription elongation complex and stably associates with the 7SK snRNP. Mol Cell. 2010;38:439–51. doi: 10.1016/j.molcel.2010.04.012 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 60.Fiume G, Vecchio E, De Laurentiis A, Trimboli F, Palmieri C, Pisano A, et al. Human immunodeficiency virus-1 Tat activates NF-κB via physical interaction with IκB-α and p65. Nucleic Acids Res. 2012;40:3548–62. doi: 10.1093/nar/gkr1224 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 61.Wong VC, Fong LE, Adams NM, Xue Q, Dey SS, Miller-Jensen K. Quantitative evaluation and optimization of co-drugging to improve anti-HIV latency therapy. Cell Mol Bioeng. 2014;7:320–33. doi: 10.1007/s12195-014-0336-9 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 62.Spina CA, Anderson J, Archin NM, Bosque A, Chan J, Famiglietti M, et al. An in-depth comparison of latent HIV-1 reactivation in multiple cell model systems and resting CD4+ T cells from aviremic patients. PLoS Pathog. 2013;9:e1003834 doi: 10.1371/journal.ppat.1003834 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 63.Sanft KR, Wu S, Roh M, Fu J, Lim RK, Petzold LR. StochKit2: software for discrete stochastic simulation of biochemical systems with events. Bioinformatics. 2011;27:2457–8. doi: 10.1093/bioinformatics/btr401 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 64.Josefsson L, King MS, Makitalo B, Brannstrom J, Shao W, Maldarelli F, et al. Majority of CD4+ T cells from peripheral blood of HIV-1-infected individuals contain only one HIV DNA molecule. Proc Natl Acad Sci USA. 2011;108:11199–204. doi: 10.1073/pnas.1107729108 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 65.Felmlee MA, Morris ME, Mager DE. Mechanism-based pharmacodynamic modeling. Methods Mol Biol. 2012;929:583–600. doi: 10.1007/978-1-62703-050-2_21 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 66.Sanner MF. Python: A programming language for software integration and development. J Mol Graph Model. 1999;17:57–61. [PubMed] [Google Scholar]
- 67.Hunter JD. Matplotlib: A 2D graphics environment. Comput Sci Eng. 2007;9:90–5. [Google Scholar]
- 68.Werner SL, Barken D, Hoffmann A. Stimulus specificity of gene expression programs determined by temporal control of IKK activity. Science. 2005;309:1857–61. doi: 10.1126/science.1113319 [DOI] [PubMed] [Google Scholar]
- 69.Eden E, Geva-Zatorsky N, Issaeva I, Cohen A, Dekel E, Danon T, et al. Proteome half-life dynamics in living human cells. Science. 2011;331:764–8. doi: 10.1126/science.1199784 [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
All relevant data are within the paper and its Supporting Information files.