Advertisement

Timing the Emergence of Resistance to Anti-HIV Drugs with Large Genetic Barriers

Timing the Emergence of Resistance to Anti-HIV Drugs with Large Genetic Barriers

  • Pankhuri Arora, 
  • Narendra M. Dixit
PLOS
x
  • Published: March 13, 2009
  • DOI: 10.1371/journal.pcbi.1000305

Abstract

New antiretroviral drugs that offer large genetic barriers to resistance, such as the recently approved inhibitors of HIV-1 protease, tipranavir and darunavir, present promising weapons to avert the failure of current therapies for HIV infection. Optimal treatment strategies with the new drugs, however, are yet to be established. A key limitation is the poor understanding of the process by which HIV surmounts large genetic barriers to resistance. Extant models of HIV dynamics are predicated on the predominance of deterministic forces underlying the emergence of resistant genomes. In contrast, stochastic forces may dominate, especially when the genetic barrier is large, and delay the emergence of resistant genomes. We develop a mathematical model of HIV dynamics under the influence of an antiretroviral drug to predict the waiting time for the emergence of genomes that carry the requisite mutations to overcome the genetic barrier of the drug. We apply our model to describe the development of resistance to tipranavir in in vitro serial passage experiments. Model predictions of the times of emergence of different mutant genomes with increasing resistance to tipranavir are in quantitative agreement with experiments, indicating that our model captures the dynamics of the development of resistance to antiretroviral drugs accurately. Further, model predictions provide insights into the influence of underlying evolutionary processes such as recombination on the development of resistance, and suggest guidelines for drug design: drugs that offer large genetic barriers to resistance with resistance sites tightly localized on the viral genome and exhibiting positive epistatic interactions maximally inhibit the emergence of resistant genomes.

Author Summary

The ability of HIV to rapidly acquire mutations responsible for resistance to administered drugs underlies the failure of current antiretroviral therapies for HIV infection. The recent advent of drugs that offer large genetic barriers to resistance, e.g., tipranavir and darunavir, presents a new opportunity to devise therapies that remain efficacious over extended durations. The large number of mutations that HIV must accumulate for resistance to drugs with large genetic barriers impedes the failure of therapy. Further, these drugs appear to exhibit activity against viral strains resistant to other drugs in the same drug class, thereby significantly improving options for therapy. Rational identification of treatment protocols that maximize the impact of these new drugs requires a quantitative understanding of the process whereby HIV overcomes large genetic barriers to resistance. We develop a model that describes HIV dynamics under the influence of a drug that offers a large genetic barrier to resistance and predict the time of emergence of viral strains that overcome the large barrier. Model predictions provide insights into the roles of various evolutionary forces underlying the development of resistance, quantitatively describe the development of resistance to tipranavir in vitro, and suggest guidelines for drug design.

Introduction

Current antiretroviral therapies for HIV infection often fail to elicit lasting virological responses in patients because of the emergence of multidrug resistant strains of HIV [1],[2]. The enormous replication rate and the high mutation and recombination rates of HIV [3][7] propel the acquisition of mutations that confer upon HIV resistance to administered drugs. The same mutations are often responsible for resistance to multiple drugs belonging to a given drug class [1],[2]. Consequently, treatment options for patients who experience failure of therapy are restricted [8],[9]. The newly approved protease inhibitors (PIs), tipranavir and darunavir, offer large genetic barriers to resistance [10],[11]. The genetic barrier of a drug, n, is the number of mutations that HIV must accumulate to gain high level resistance to the drug [12]. When n is small (e.g., n = 1 for 3TC [1]), drug resistant genomes are likely to exist in patients prior to the onset of therapy [13]. As n increases, the likelihood of the pre-existence of resistant genomes decreases considerably [13],[14]. Resistant genomes must then emerge during therapy through mutation and/or recombination of susceptible genomes. The replication of susceptible genomes, however, is suppressed during therapy. Besides, HIV must undergo a large number of replication cycles to accumulate all the mutations required for resistance to a drug with large n. Consequently, the development of resistance to a drug with large n may be significantly delayed. Indeed, up to 9 months were required for HIV to develop resistance to tipranavir in in vitro serial passage experiments [10].

Current treatment guidelines for HIV infection recommend a combination of 3, but at least 2, active drugs, (i.e., drugs for which resistance has not developed) in order partly to increase the overall genetic barrier of therapy [9]. For treatment naïve patients, a combination of 2 nucleoside/nucleotide reverse transcriptase inhibitors (NRTIs) is typically employed in combination with either a non-nucleoside reverse transcriptase inhibitor (NNRTI), usually efavirenz, or a ritonavir-boosted PI, usually lopinavir [9]. With ritonavir-boosted lopinavir monotherapy, fewer patients achieved plasma HIV RNA levels below detection and more patients witnessed emergence of PI resistance mutations than in patients receiving ritonavir-boosted lopinavir in combination with 2 NRTIs [15]. Similarly, despite comparable times to virological failure, patients receiving a 2 drug combination of efavirenz and lopinavir experienced more frequent emergence of resistance than patients receiving a 3 drug combination of efavirenz or lopinavir and 2 NRTIs [16]. Therapy with 4 NRTIs had a similar response to therapy with efavirenz and 2 NRTIs [17]. Consequently, a 3 drug combination is the current standard of care for treatment naïve patients. When failure did occur with a 3 drug combination, it was typically associated with NNRTI resistance in patients receiving efavirenz but not with PI resistance in patients receiving lopinavir [16], in accordance with the larger genetic barriers offered by PIs than by NNRTIs [18]. The large genetic barrier in conjunction with a superior pharmacokinetic profile may also underlie the high rates of viral suppression despite sub-optimal adherence in patients receiving ritonavir-boosted lopinavir-based therapy [19].

For second-line therapy, which follows the failure of the initial regimen, a drug from a new drug class is recommended in order to minimize the risk of cross-resistance [9]. Thus, among several newly available agents [20], the fusion inhibitor enfuvirtide and the recently approved integrase inhibitor raltegravir present potent options. Both enfuvirtide and raltegravir, however, offer small genetic barriers and are therefore recommended for use in conjunction with a supporting drug such as darunavir [9]. Remarkably, the new PIs, tipranavir and darunavir, elicit responses against viral strains resistant to other PIs [11],[21], increasing options for second-line therapy. The new PIs thus present promising weapons to avert the failure of antiretroviral therapy. Indeed, significant efforts are ongoing to identify treatment protocols that maximize the impact of the new PIs [8],[22]. Identification of improved protocols hinges on our understanding of HIV dynamics under the influence of drugs that offer large genetic barriers to resistance and of the process by which HIV surmounts these large genetic barriers.

Description of the development of resistance to a drug with a large n is complicated for several reasons. First, resistance to such a drug typically develops gradually, increasing progressively with the number of mutations accumulated [10],[23]. As a result, the emergence and the competitive dynamics of a large number of distinct viral genomes carrying different combinations of resistance mutations and possessing various intermediate levels of resistance must be described. For instance, the accumulation of mutations at 6 loci confers high level resistance to tipranavir [10]. Consequently, depending on whether each resistance locus carries a mutation or not, 26, or 64, distinct strains (see below) may emerge in the course of infection. Because HIV is diploid, the 64 strains yield 64 homozygous and 2016 different kinds of heterozygous virions, whose evolutionary dynamics must be followed to describe how the genetic barrier of tipranavir is overcome. Second, the population size of HIV in vivo may be small, especially under the influence of therapy, which implies that the emergence of resistant genomes is likely to be governed by stochastic rather than deterministic effects [24]. Third, in addition to mutation, recombination can play a significant role in the formation of drug resistant strains that carry multiple mutations [25],[26]. The influence of recombination, which is yet to be fully understood, depends on several factors, viz., the frequency of multiple infections of cells, the effective population size of HIV in vivo, and the nature of fitness interactions between resistance mutations, characterized by epistasis [27][33]. No models exist that describe HIV dynamics under the simultaneous influence of mutation, multiple infections of cells, recombination, epistatic interactions between multiple resistance mutations, and stochastic effects of finite population sizes. Consequently, timing the failure of antiretroviral drugs with large genetic barriers is currently not possible. Rational identification of improved treatment protocols is therefore precluded.

Here, we develop a model of HIV dynamics that quantitatively predicts the expected waiting time for the emergence of genomes that carry the requisite mutations for resistance to a drug with any given genetic barrier. Extant models of HIV dynamics assume that deterministic forces are predominant in the emergence of drug resistance [34][36]. Consequently, extant models predict that drug resistant genomes emerge immediately upon the initiation of therapy, albeit in small numbers. In contrast, especially when the genetic barrier is large, stochastic forces are expected to dictate the emergence of resistant genomes. A key consequence of the predominance of stochastic forces is a delay in the emergence of resistant genomes following the initiation of therapy. Our model accounts for this delay in a deterministic manner by predicting the expected waiting time for the emergence of resistant genomes. Model predictions capture the development of resistance to tipranavir in vitro quantitatively, indicating that our model captures the underlying dynamics of the development of resistance to antiretroviral drugs. Further, model predictions provide insights into the impact of underlying evolutionary forces on the development of drug resistance and suggest guidelines for drug design.

Results

Model Formulation

We consider uninfected cells, T, exposed in the presence of a PI with a genetic barrier n to a viral population, V, containing genomes highly susceptible to the PI. The highly susceptible, or wild-type, genomes are assumed to contain no resistance mutations. As infection proceeds, error-prone replication gives rise to mutant genomes. distinct mutant genomes can arise, each with at least one resistance mutation (Figure 1). Our aim is to determine the waiting time for the first formation of the genome that carries all the n resistance mutations and is therefore highly resistant to the drug. We number the different viral genomes 0, 1, 2, 3…S, where genome 0 represents the wild-type (Figure 1). We let Vjh denote the population of virions containing genomes j and h, where j, h Є {0, 1, 2…S}. Because virions V10, for instance, are indistinguishable from virions V01, we impose the constraint jh [37]. Following the infection of a cell by a virion Vjh, mutation and recombination give rise to a proviral genome i Є {0, 1, 2…S} with probability Qi(jh). We distinguish infected cells by the proviral genomes they contain: Cells Ti are infected by a single provirus i and cells Tij by proviruses i and j, where ij and i, j Є {0, 1, 2…S}. Infected cells produce progeny virions. Drug action causes some of the progeny virions to be non-infectious [3],[34],[35]; we denote the noninfectious virion population by . Cells Ti and Tii infected by a single kind of provirus produce homozygous virions Vii and . Cells Tij infected with distinct proviruses (ij) yield homozygous virions Vii, , Vjj and and heterozygous virions Vij and . The resulting infection network is shown in part in Figure 2.

thumbnail
Figure 1. Schematic representation of the S viral genomes carrying different combinations of resistance mutations (stars) that emerge during the development of resistance to a drug with a genetic barrier n.

doi:10.1371/journal.pcbi.1000305.g001

thumbnail
Figure 2. Schematic representation of the infection network indicating the various singly and doubly infected cells, Ti and Tij, and homozygous and heterozygous virions, Vii and Vij, respectively, that emerge during the development of drug resistance.

Non-infectious virions are crossed.

doi:10.1371/journal.pcbi.1000305.g002

We construct dynamical equations to predict the time-evolution of various cell and viral populations and estimate the average waiting times for the first production of each of the S mutant proviral genomes (Methods). We denote by W the waiting time for the emergence of the provirus that contains all the n resistance mutations and hence overcomes the genetic barrier of the drug.

Model Predictions

We solve model equations to describe the development of resistance in in vitro serial passage experiments (e.g., [10]). Here, T0 uninfected cells are exposed to viruses in the presence of a known concentration of the PI. Infection is allowed to progress until time tp (~3.5 days), the duration of a passage. The resulting viral population is employed to initiate infection of a fresh set of T0 uninfected cells in the next passage. At the start of the first passage, the viral population is assumed to consist of V00 wild type viruses, highly susceptible to the drug. Gradually, genomes with increasing levels of drug resistance emerge.

Cell and virus dynamics.

We perform calculations for a genetic barrier n = 5, representative of ritonavir-boosted PIs [38]. We let the separation between successive resistance mutations, l = 100 nucleotides, and choose the efficacy profile shown in Figure 3 with epistasis E = 0 (also see Methods). Here, the efficacy against the wild type, ε0, and against the strain with n mutations, εn, correspond to 400 nM of tipranavir [10]. We assume that the efficacy against intermediate mutants, εm, depends on the number of mutations, m (0≤mn), the genomes contain. In Figure 4A, we present the evolution of populations of uninfected cells, T, infected cells, , and infectious virions, , with time following the onset of the experiment. In the first passage, T rises due to the proliferation of uninfected cells (Figure 4A, inset). At the same time, T* rises due to the infection of T, and V rises sharply due to viral production from T*. In the second passage, the higher V enhances the infection of T. Here, the loss of T due to infection dominates cell proliferation and T declines. Consequently, following an initial rise of T* due to infection of T, target cell limitation lowers the formation of new infected cells and causes T* to decline. The resulting lower viral production causes V to decline as well. This two phase behavior within a passage–an initial rise and the subsequent fall of T*–is observed in experiments [7] and is explained by models [37],[39]. The same two phase behavior repeats in ensuing passages and an oscillatory pseudo steady state is attained. Gradually, V rises marking the emergence of drug resistant genomes.

thumbnail
Figure 3. Efficacy, εm, of a genome carrying m (0≤mn) resistance mutations, when the genetic barrier n = 5, ε0 = 0.85, εn = 0.25, and the epistasis, E = 0.005 (green), 0 (red), and −0.005 (blue).

The inset shows the corresponding fitness ( = 1−εm) profiles.

doi:10.1371/journal.pcbi.1000305.g003

thumbnail
Figure 4. Model predictions of cell and viral dynamics.

The time evolution of (A) the number of uninfected cells (red), infected cells (blue), and infectious virions (green) and (B) homozygous virions carrying wild-type genomes (pink) and single (blue), double (green), triple (orange), quadruple (red), and quintuple (black) mutants, obtained by solving Eqs. (1)–(9) with the parameters T0 = 106 cells, V00 = 5×105 virions, n = 5, l = 100 nucleotides, and εm from Figure 3 with E = 0. The remaining parameters are listed in Methods. The inset in (A) shows the evolution for the first two passages.

doi:10.1371/journal.pcbi.1000305.g004

Emergence of resistant genomes.

In Figure 4B, we present the time-evolution of populations of infectious homozygous virions containing genomes with different numbers of resistance mutations. Initially, the viral population contains the wild type genomes alone. Following the onset of infection, as V rises, the rate of formation of single mutants increases. Single mutants emerge here in the first passage. Because drug efficacy is lower against single mutants than against wild type genomes (ε10; Figure 3), single mutants begin to grow at the expense of the wild type. (We note that unlike the scenario in vivo, passage experiments are designed to allow the growth of even wild-type genomes in the initial passages.) As infections by single mutants become significant, the rate of formation of double mutants rises. Double mutants emerge in ~40 days. With n = 5, different double mutants are possible. They emerge at slightly different times because of the differential influence of recombination: A double mutant that contains the two mutations on adjacent resistance loci is less likely to be formed by recombination than a double mutant with mutations on well separated loci; the number of crossovers increases with the separation [40]. Again, because ε210 (Figure 3), double mutants begin to outgrow single mutants and the wild type. This process continues with the sequential emergence of higher mutants until by W~100 days quintuple mutants emerge, which possess high level resistance to the drug. W~100 days is thus the waiting time for the emergence of the genome that overcomes the genetic barrier of the drug. From this point on, quintuple mutants dominate the viral population.

Several characteristics of drugs, viz., the genetic barrier, n, epistasis, E, and the separation between adjacent resistance loci, l, influence W, which we examine next.

Effect of the genetic barrier.

To examine the influence of the genetic barrier, we vary n for fixed values of ε0, εn, and E, and predict W. We find that W increases dramatically with n. For instance, W increases from ~12 days when n = 2 to ~100 days when n = 5 (Figure 5A), underscoring the advantage of a drug with a large n. As n increases, the number of mutations necessary for resistance increases. The number of replication cycles required to accumulate the necessary mutations increases correspondingly and delays the emergence of resistant genomes. The development of resistance is inhibited further by the delayed emergence of intermediate mutants. For the same ε0 and εn, the incremental fitness advantage with each mutation decreases as n increases. The smaller this advantage, the longer it takes for the resolution of the competition between different mutants. Thus, following their emergence, double mutants take longer to outgrow single mutants when n = 4 than when n = 3. When the influence of recombination is weak, triple mutants emerge predominantly by mutation of double mutants. Consequently, the waiting time for the emergence of triple mutants is larger when n = 4 than when n = 3. Indeed, triple mutants emerge in ~40 days when n = 3 and ~55 days when n = 4 (Figure 5A). Thus, the increasing number of replication cycles required and the slower emergence of intermediate mutants together result in the dramatic increase of W with n.

thumbnail
Figure 5. Model predictions of emergence and fixation times.

The expected waiting time for the emergence of (A) genomes with different numbers of resistance mutations for different n when E = 0, (B) the corresponding nth mutants as a function of E, (C) quintuple mutants when n = 5 as a function of the crossover frequency (ρl), for E = 0.005 (green), 0 (red), −0.005 (blue). The inset in (C) shows the corresponding reduction in the time of emergence, 1−W(ρl)/W(ρl = 0). (D) Model predictions of emergence (filled symbols) and fixation (open symbols) times of double mutants when n = 2 and E = 0.05 (green), 0 (red), −0.05 (blue). In (A) to (C), we let ε0 = 0.85 and εn = 0.25, whereas in (D) ε0 = 0.1 and εn = 0. All the other parameters are identical to those in Figure 4.

doi:10.1371/journal.pcbi.1000305.g005

Effect of epistasis.

The increase of W with n is amplified when E>0. Whereas W increases from ~12 to ~100 days when E = 0, W increases from ~12 to ~205 days when E = 0.005 as n increases from 2 to 5 (Figure 5B). As E increases, the fitness of intermediate mutants decreases (Figure 3). Consequently, intermediate mutants emerge slower, increasing W (Figures S1 and 5B). In contrast, the fitness of intermediate mutants is higher (Figure 3) and hence the increase of W with n is suppressed when E<0 (Figures S1 and 5B).

Effect of recombination.

Interestingly, recombination decreases W regardless of E (Figure 5C). Recombination accelerates the accumulation of mutations and expedites the emergence of resistant strains. Thus, upon increasing the recombination rate, which we accomplish by increasing l, the separation between resistance loci, W drops by ~50% when E = 0.005 and by ~80% when E = −0.005 for n = 5 from that in the absence of recombination (Figure 5C, inset). The greater drop in W when E<0 is because of the increased fitness (Figure 3) and hence greater prevalence of intermediate mutants, which in turn enhances the likelihood of the formation of heterozygous virions and facilitates the accumulation of mutations by recombination. This influence of recombination on W is robust to changes in n (Figure S2).

That recombination invariably lowers W is intriguing given that several studies argue that recombination may inhibit the fixation of resistance when E>0 (e.g., see [27],[33]). We therefore compute the fixation time, F, defined as the time when 90% of the genomes in the viral population are nth mutants. We find interestingly that recombination increases F when E>0, consistent with current expectations (Figure 5D) [27],[33]. When n = 2, W marks the time when the first double mutant emerges in the viral population. For times smaller than W, the wild-type and the single mutants alone exist in the viral population. Recombination then brings the mutations on the two single mutants together and accelerates the emergence of the double mutant regardless of E. After the double mutant emerges, recombination influences the competitive dynamics of the different viral strains and alters F. In particular, not only do single mutants recombine to yield the double mutant, but the double mutant could also be lost by recombination with the wild type. When E>0, recombination tends to lower the prevalence of the double mutant [27], resulting in the observed increase in F. Thus, recombination may lower W and yet increase F. We note that the distinction between emergence and fixation has been recognized earlier [41]. When E<0, recombination enhances the prevalence of the double mutant and decreases F. We recognize here that the influence of recombination on F is determined not only by E but also by n, ε0, εn, and the population size of HIV and its variation, examining all of which is beyond the scope of the present study. Our aim here is to predict W, which marks the emergence of drug resistant genomes.

Comparison with Experiments

We apply our model to describe the development of resistance to tipranavir in in vitro serial passage experiments [10]. We let n = 6 because a genome with 6 resistance mutations exhibited >10 fold resistance to tipranavir in these experiments. We choose IC50 values for different intermediate mutants from the ranges determined experimentally (Table S1). Further, we employ actual distances between resistance sites to calculate the recombination probabilities and also assign fitness advantages to genomes containing specific combinations of mutations (Table S1). (In contrast, in our calculations above, the number of mutations and not their specific combinations was assumed to determine the fitness advantage.) We also vary the concentration of tipranavir as in the experiments (Table S2). Further, following the experimental protocol, we employ 90% of the viral population at the end of any passage to initiate infection in the succeeding passage when the drug concentration is maintained constant across the passages and 50% of the viral population when the drug concentration is increased in the succeeding passage. Genomes carrying 2, 3, 5 and 6 resistance mutations were first observed in the experiment in passages 16, 33, 39 and 49, respectively [10]. In close agreement, our model predicts the emergence of these genomes in passages 14, 29, 44 and 49, respectively (Figure 6). (Ignoring the concept of the waiting time, i.e., letting wi = 0 in our model, severely underpredicts the times of emergence of drug resistant genomes (Figure 6). The agreement between model predictions and experiments indicates that our model captures the underlying dynamics of the development of resistance to antiretroviral drugs accurately.

thumbnail
Figure 6. Comparison of model predictions (red) and the experimentally observed [10] (blue) times of emergence of different mutants resistant to tipranavir.

The different mutants and the corresponding IC50 values are listed in Table S1. Also shown are the times when the numbers of the different mutant proviruses first reach 1 (green) predicted by our model when, following current models [34],[35], we assume that the waiting times wi = 0.

doi:10.1371/journal.pcbi.1000305.g006

Discussion

Current models of HIV dynamics successfully predict short-term changes in the plasma viral load in patients undergoing therapy but fail to provide a quantitative description of the emergence of drug resistance [34][36]. A key limitation of current models is the underlying assumption that the emergence of resistant genomes is governed by deterministic effects. Deterministic effects predominate when the population of cells in an infected individual is large. In a finite cell population, because the probability of the formation of a resistant genome with many mutations can be small, resistant genomes emerge stochastically. The waiting time for the emergence of resistant genomes can therefore be substantial. In contrast, by assuming that deterministic effects predominate, current models predict that resistant genomes emerge, albeit in very small numbers, immediately upon the onset of therapy. Once resistant genomes emerge, their numbers grow due to viral production from the cells they infect leading to the rapid fixation of resistance. Current models thus underestimate the time for the development of drug resistance (Figure 6).

Simulations of viral evolution, based on models of population genetics, consider finite populations and present descriptions of the stochastic emergence of drug resistant genomes [28],[29],[32]. Importantly, the simulations also enable incorporation of recombination and fitness interactions between multiple loci, which are central to the development of drug resistance but are not easily incorporated in models of HIV dynamics. The simulations, however, make several simplifying assumptions, such as fixed population sizes and discrete generations, which approximate the dynamics of the development of drug resistance and introduce uncertainties in the influence of underlying processes, such as recombination [30],[33]. Besides, simulations are difficult to incorporate in mathematical formalisms for therapy optimization.

Here, we develop a model that employs the deterministic framework of models of HIV dynamics and at the same time captures the influence of stochastic effects associated with the emergence of drug resistant genomes. To accomplish this, we invoke the concept of the expected waiting time. We develop a detailed description of mutation and recombination between multiple loci, which enables calculation of the probability of the formation of resistant genomes in one replication event. Given the viral and cell populations and the efficacy of the drug, the frequency of replication events and hence the rate of formation of resistant genomes is determined. From the rate of formation, we estimate the expected waiting time for the first resistant genome to emerge. Different mutant genomes are assumed to appear first in the viral population at their respective expected waiting times. The limitation of current models of HIV dynamics, which predict the emergence of resistant genomes immediately upon the start of therapy, is thus overcome. Yet, by calculating the “expected” waiting time, our model captures the influence of stochastic effects associated with the emergence of resistant genomes in an averaged sense and retains the dynamical framework of current models. The limitations of population genetics based simulations are also thus overcome.

The waiting time for the emergence of a genome carrying a certain number of mutations depends on the times of emergence and the growth of subpopulations of genomes with fewer mutations. Our model assumes that the latter genomes emerge at their expected waiting times. Consequently, the variation in the waiting times for the emergence of higher mutants due to the variation in the times of emergence of lower mutants is suppressed in our model. Further, following emergence, particularly when the population size is small, the chance that stochastic forces cause the extinction of genomes may be significant. We assume, however, that following emergence, the growth of genomes is deterministic. The extent of the uncertainties introduced in our model predictions by these simplifying assumptions remains to be estimated. Semi-stochastic simulations, where the times of emergence of mutant genomes alone are determined stochastically, and fully stochastic simulations (see, e.g., [42]) of the emergence of mutant genomes would serve as tests of our model. Performing the simulations, however, is beyond the scope of the present study. Here, we compare model predictions with experiments and find that our predictions are in close agreement with experimental observations [10] of the times of emergence of various genomes possessing different degrees of resistance to tipranavir, suggesting that our model captures the underlying dynamics of the development of drug resistance by HIV.

Model predictions indicate that the waiting time, W, for the emergence of the strain that overcomes the genetic barrier of a drug depends on several factors that may be tuned during drug design. A large genetic barrier significantly enhances W. This enhancement of W with the genetic barrier is amplified when fitness interactions between resistance loci exhibit positive epistasis. Recombination, in contrast, lowers W regardless of epistasis or the genetic barrier. If the separation between resistance loci is small, however, the role of recombination is suppressed. Thus, for delaying the emergence of resistant genomes, drugs that offer large genetic barriers with resistance sites localized tightly on the viral genome and exhibiting positive epistatic interactions are desirable. These observations may serve as guidelines for structure-based drug design [43]. The fixation of resistant genomes following their emergence may depend differently on drug characteristics and remains to be fully elucidated.

When distinctions between different viral genomes are ignored, the expected waiting time vanishes and our model reduces to the basic model of HIV dynamics, which successfully captures viral load changes in patients undergoing therapy [3],[34]. Our model may thus be applied to predict drug failure in vivo. Several advances of our model are essential, however, to describe the in vivo scenario accurately. First, the higher frequency of multiple infections [44], possible cell-cell transmission of infection [45],[46], and the existence of resistance mutations prior to the onset of therapy [1] in vivo must be incorporated into our model. Second, during potent drug therapy, viral replication may be suppressed significantly, resulting in a small effective population size of HIV. The variation of the waiting time about the mean may then become large. Consequently, the assumption that mutant genomes emerge at their expected waiting times becomes less accurate. Our model must therefore be advanced to account for the variation of the emergence times of genomes in vivo. Third, our model must be extended to drugs from other drug classes to mimic current combination therapies. With these advances, our model would enable timing the emergence of resistance to drugs in vivo and facilitate the identification of treatment protocols that maximally impede the failure of current therapies.

Methods

Dynamical Equations

We present equations below that describe the in vitro dynamics of various cell and viral populations.

Uninfected cells.

(1)Uninfected cells, T, proliferate at rate λ and die at rate dT. The rate of formation of infected cells Ti containing genome i is (see below). Summation over i from 0 to S = 2n−1 yields the total rate of loss of T due to infection by free virions. At the beginning of each passage, T is set to T0 = 106.

Singly infected cells.

(2)where i Є {0, 1, 2…S}. Here, k0TVjh is the rate at which virions Vjh infect T. Following infection, the genomes j and h undergo reverse transcription to produce provirus i with the probability Qi(jh) (see below). Thus, k0TQi(jh)Vjh is the rate at which uninfected cells acquire genome i following infection by Vjh. Summation over j and h yields the total rate, , of the production of cells Ti.

The rate can be small, especially if genome i contains many mutations. Consequently, when the cell population is finite, the formation of the first cell Ti is stochastic. We define ti as the waiting time for the formation of the first cell Ti. ti may assume any value between 0 and ∞ with a probability density dependent on (see below). Here, we assume as a simplification that the first cell Ti emerges at the expected waiting time . In addition, we assume that following emergence, the growth of Ti is deterministic. We therefore multiply the rate in Eq. (2) with the Heaviside function,(3)so that Ti = 0 when t<wi, and cells Ti are produced at the rate when twi. By allowing Ti to grow at the rate from time t = 0, extant models of HIV dynamics underestimate the time of emergence of drug resistant genomes. We derive estimates of wi below. We solve Eq. (2) with the initial condition, Ti(wi) = 1, and reset Ti to zero at the start of every passage. The other two terms in Eq. (2) represent the loss of cells Ti due to death at rate δ and due to further infections, which convert Ti to doubly infected cells.

Doubly infected cells.

(4a)Here is the rate at which cells Ti acquire a second provirus i. wii is the expected waiting time for the emergence of the first cell Tii. After the first infection of a cell, down-modulation of cell surface CD4 receptors reduces the susceptibility of the cell to new infections [47],[48]. Here, we let k1(<k0) be the mean rate constant for the infection of singly infected cells [37].

For cells infected with two different kinds of proviruses, we write(4b)where i<j and i, j Є {0, 1, 2…S}. Here, the two terms in the brackets correspond to the two ways of forming a doubly infected cell Tij: a cell Ti can be infected by provirus j or a cell Tj by provirus i. Following earlier studies, we ignore more than two infections of cells [37]. We solve the above equations with the initial conditions Tii(wii) = 1 and Tij(wij) = 1, respectively, and reset Tii and Tij to zero at the start of every passage.

Waiting time: At any time t, the rate of formation of cells Ti,(5)Because individual infection events occur independently, the formation of Ti may be described as a Poisson process with the instantaneous rate . The probability that the waiting time, ti, for the first formation of a cell Ti is smaller than s is then , 0≤s<∞ [49]. It follows that the expected waiting time, , or(6)Similarly,(7)where and when i<j, .

We recognize that the evaluation of the waiting times requires knowledge of the rates, e.g., , at all times t. We therefore devise a numerical approximation to estimate wi based on the values of until a given time t, which allows explicit integration of the dynamical equations (Text S1).

Reverse transcription.

To evaluate the probability Qi(jh), we decouple mutation and recombination [29],[32],[50]. We let genomes j and h recombine to produce genome k with probability Rk(jh) and let genome k mutate to genome i with probability Pik [50]. The number of different recombinants k that can be produced is 2d, where dn is the number of sites at which j and h differ. Summing over all the recombinants k gives the total probability of producing genome i by reverse transcription of genomes j and h,(8)

Recombination.

To determine Rk(jh), we compare genomes j and h at each of the n drug resistance sites and identify the distances l1, l2, etc., between the d successive sites at which the genomes differ (Figure S3). We then compare genome k with the genomes j and h to determine on which genome, j or h, the enzyme reverse transcriptase (RT) must be at each of the d distinguishing sites in order to yield the genome k. Figure S3 illustrates the desired path of RT for given j, h, and k. If Pdes(m) is the probability that RT is on the desired genome at the mth distinctive site, then , which is readily evaluated based on the probabilities that RT undergoes odd and even crossovers in any length l (Figure S3).

Mutation.

To calculate the probability of mutation, Pik, we compare the two genomes i and k at the n drug resistance sites and identify the u sites where the two genomes differ. The probability that genome k mutates at these u sites alone is , where 0≤un and μ is the mutation rate.

Virions.

(9a)(9b)(9c)(9d)Here Vii and Vij denote infectious and and non-infectious virions. In the absence of the drug, cells Ti and Tii produce homozygous virions Vii and cells Tij produce both homozygous virions Vii and Vjj and heterozygous virions Vij in the proportion ¼, ¼, and ½, respectively [37]. All infected cells with at least one provirus of type i thus contribute to the production of Vii. is the viral burst size, δ is the death rate of infected cells and c is the clearance rate of free virions. The above equations are solved with the initial condition that the wild type virions V00 alone exist at the start of the first passage. For every subsequent passage, the free virions at the end of the previous passage are employed to initiate infection.

Drug efficacy.

The efficacy of a PI is the fraction of progeny virions that it renders non-infectious. We assume that the drug efficacy εii against the protease produced by genome i depends on the number of resistance mutations, m(i), the genome contains, i.e., εii = εm [23]. We fix the efficacy against the wild type, ε0, and that against the strain with n mutations, εn, and determine the efficacies against intermediate strains from the epistasis, E, which is assumed constant for every pair of loci differing by two resistance mutations [27]. If E = 0, then the mutations do not interact with each other and each additional mutation increases the log fitness, which we define as log(1−εm), by the same amount (Figure 3). When E<0 (>0), the mutations interact antagonistically (synergistically) in increasing log fitness. For cells Ti and Tii, the fraction of progeny virions rendered noninfectious is εii. For cells Tij containing distinct proviruses, phenotypic mixing [27] implies that proteases i and j are equally likely to be included in a budding virion so that, on average, the fraction of virions rendered noninfectious is εij = (εiijj)/2.

When the efficacy is determined as a function of the drug concentration, C, as in our calculations in Figure 6, we write εii = C/(IC50(i)+C), where IC50(i) is the value of C at which εii is 50% [51].

Equations (1) to (9) represent a model of HIV dynamics that describes the development of resistance to a PI with a genetic barrier n. We solve the equations using a computer program written in C.

Model Parameters

We employ the following parameter values based on earlier studies [4],[37],[39],[52]: the birth and death rate of uninfected T cells, λ = 0.624 day−1 and dT = 0.018 day−1; the death rate of infected cells, δ = 1.44 day−1; the viral burst size, = 103; the viral clearance rate, c = 0.35 day−1; the second order rate constants of the infection of uninfected and singly infected cells, k0 = 10−8 day−1 and k1 = 0.7k0; the mutation and recombination rates, μ = 3×10−5 per site per replication, and ρ = 8.3×10−4 crossovers per site per replication.

Supporting Information

Figure S1.

Model predictions of the times of emergence of various mutants as the genetic barrier varies from n = 2 to 4 and the epistasis E = −0.01 (A) and 0.01 (B). All the other parameters are identical to those in Figure 4.

doi:10.1371/journal.pcbi.1000305.s001

(0.06 MB DOC)

Figure S2.

Model predictions of the time of emergence of nth mutants for different epistatic interactions, E, and genetic barriers n = 3 (A) and n = 4 (B). All the other parameters are identical to those in Figure 4.

doi:10.1371/journal.pcbi.1000305.s002

(0.06 MB DOC)

Figure S3.

Schematic representation of the production of genome k by recombination of genomes j and h. Stars indicate mutations. The arrow marks the desired path of the enzyme reverse transcriptase (RT) and allows determination of the probability, Rk(jh), that genome k is formed. At the first site where j and h differ, the probability that RT is on the desired genome, Pdes(1), is 1/2, because reverse transcription can commence on either of the two genomes with equal likelihood. At the second site, if the desired genome is the same as that of the first site, then RT will be on the desired genome if it undergoes an even number of crossovers in the intervening distance l1, the probability of which we write as Pdes(2) = Peven(l1). If the desired genome is different from that at the first site, then the probability that RT will be on the desired genome is Pdes(2) = Podd(l1). It follows that Rk(jh) = ΠPdes(m), where m ranges from 1 to d and the probabilities that even and odd crossovers occur in length l are [37] Peven(l) = exp(−ρl)cosh(ρl) and Podd(l) = exp(−ρl)sinh(ρl), respectively, with ρ the per site recombination rate of HIV.

doi:10.1371/journal.pcbi.1000305.s003

(0.03 MB DOC)

Table S1.

Sequences resistant to tipranavir and their IC50 values. Sequences with different combinations of resistance mutations observed experimentally, corresponding binary sequences illustrating the specific locations of mutations, marked as 1, when n = 6, and the respective IC50 values employed in our model are listed. The experimental IC50 values [10] are in brackets. In our simulations (Figure 6), we assign IC50 values to genomes as follows. To each genome i, we assign an IC50 value equal to the IC50 of the genome below that has the maximum number of mutations in common with the genome i but has no mutations in addition to those contained in i. For instance, the genome 101001 is assigned an IC50 of 101 nM, whereas the genome 000110 is assigned an IC50 of 60 nM, equal to the wild-type.

doi:10.1371/journal.pcbi.1000305.s004

(0.04 MB DOC)

Table S2.

Drug concentrations employed in the experiments [10] and in our calculations of Figure 6.

doi:10.1371/journal.pcbi.1000305.s005

(0.03 MB DOC)

Text S1.

Estimates of waiting times

doi:10.1371/journal.pcbi.1000305.s006

(0.06 MB DOC)

Acknowledgments

We are grateful to Elizabeth Wardrop and Michael G. Cordingley for providing details of their experimental protocol.

Author Contributions

Conceived and designed the experiments: PA NMD. Performed the experiments: PA. Analyzed the data: PA NMD. Contributed reagents/materials/analysis tools: PA NMD. Wrote the paper: PA NMD.

References

  1. 1. Clavel F, Hance AJ (2004) HIV drug resistance. N Engl J Med 350: 1023–1035.
  2. 2. Johnson VA, Brun-Vezinet F, Clotet B, Gunthard HF, Kuritzkes DR, et al. (2008) Update of the drug resistance mutations in HIV-1: Spring 2008. Top HIV Med 16: 62–68.
  3. 3. Perelson AS, Neumann AU, Markowitz M, Leonard JM, Ho DD (1996) HIV-1 dynamics in vivo: virion clearance rate, infected cell life-span, and viral generation time. Science 271: 1582–1586.
  4. 4. Mansky LM, Temin HM (1995) Lower in-vivo mutation-rate of human-immunodeficiency-virus type-1 than that predicted from the fidelity of purified reverse-transcriptase. J Virol 69: 5087–5094.
  5. 5. Jetzt AE, Yu H, Klarmann GJ, Ron Y, Preston BD, et al. (2000) High rate of recombination throughout the human immunodeficiency virus type 1 genome. J Virol 74: 1234–1240.
  6. 6. Rhodes T, Wargo H, Hu W-S (2003) High rates of human immunodeficiency virus type 1 recombination: Near-random segregation of markers one kilobase apart in one round of viral replication. J Virol 77: 11193–11200.
  7. 7. Levy DN, Aldrovandi GM, Kutsch O, Shaw GM (2004) Dynamics of HIV-1 recombination in its natural target cells. Proc Natl Acad Sci U S A 101: 4204–4209.
  8. 8. Wilkin TJ, Taylor B, Olender S, Hammer SM (2008) Advances in antiretroviral therapy. Top HIV Med 16: 31–60.
  9. 9. Hammer SM, Eron JJ, Reiss P, Schooley RT, Thompson MA, et al. (2008) Antiretroviral treatment of adult HIV infection: 2008 recommendations of the International AIDS Society—USA panel. JAMA 300: 555–570.
  10. 10. Doyon L, Tremblay S, Bourgon L, Wardrop E, Cordingley MG (2005) Selection and characterization of HIV-1 showing reduced susceptibility to the non-peptidic protease inhibitor tipranavir. Antiviral Res 68: 27–35.
  11. 11. de Meyer S, Vangeneugden T, van Baelen B, de Paepe E, van Marck H, et al. (2008) Resistance profile of darunavir: Combined 24-week results from the POWER trials. AIDS Res Hum Retroviruses 24: 379–388.
  12. 12. Shafer RW, Schapiro JM (2005) Drug resistance and antiretroviral drug development. J Antimicrob Chemother 55: 817–820.
  13. 13. Lapadula G, Izzo I, Gargiulo F, Paraninfo G, Castelnuovo F, et al. (2008) Updated prevalence of genotypic resistance among HIV-1 positive patients naive to antiretroviral therapy: a single center analysis. J Med Virol 80: 747–753.
  14. 14. Ribeiro RM, Bonhoeffer S, Nowak MA (1998) The frequency of resistant mutant virus before antiviral therapy. AIDS 12: 461–465.
  15. 15. Delfraissy JFO, Flandre P, Delaugerre C, Ghosn J, Horban A, et al. (2008) Lopinavir/ritonavir monotherapy or plus zidovudine and lamivudine in antiretroviral-naive HIV-infected patients. AIDS 22: 385–393.
  16. 16. Riddler SA, Haubrich R, DiRienzo AG, Peeples L, Powderly WG, et al. (2008) Class-sparing regimens for initial treatment of HIV-1 infection. N Engl J Med 358: 2095–2106.
  17. 17. Moyle G, Higgs C, Teague A, Mandalia S, Nelson M, et al. (2006) An open-label, randomized comparative pilot study of a single-class quadruple therapy regimen versus a 2-class triple therapy regimen for individuals initiating antiretroviral therapy. Antivir Ther 11: 73–78.
  18. 18. Kuritzkes DR (2004) Preventing and managing antiretroviral drug resistance. AIDS Patient Care STDS 18: 259–273.
  19. 19. Shuter J, Sarlo JA, Kanmaz TJ, Rode RA, Zingman BS (2007) HIV-infected patients receiving lopinavir/ritonavir-based antiretroviral therapy achieve high rates of virologic suppression despite adherence rates less than 95%. J Acquir Immun Defic Syndr 45: 4–8.
  20. 20. Perno CF, Moyle G, Tsoukas C, Ratanasuwan W, Gatell J, et al. (2008) Overcoming resistance to existing therapies in HIV-infected patients: the role of new antiretroviral drugs. J Med Virol 80: 565–576.
  21. 21. Larder BA, Hertogs K, Bloor S, van den Eynde C, DeCian W, et al. (2000) Tipranavir inhibits broadly protease inhibitor-resistant HIV-1 clinical samples. AIDS 14: 1943–1948.
  22. 22. Martinez-Cajas JL, Wainberg MA (2008) Antiretroviral therapy: optimal sequencing of therapy to avoid resistance. Drugs 68: 43–72.
  23. 23. Baxter JD, Schapiro JM, Boucher CAB, Kohlbrenner VM, Hall DB, et al. (2006) Genotypic changes in human immunodeficiency virus type 1 protease associated with reduced susceptibility and virologic response to the protease inhibitor tipranavir. J Virol 80: 10794–10801.
  24. 24. Kouyos RD, Althaus CL, Bonhoeffer S (2006) Stochastic or deterministic: what is the effective population size of HIV-1? Trends Microbiol 14: 507–511.
  25. 25. Moutouh L, Corbeil J, Richman DD (1996) Recombination leads to the rapid emergence of HIV-1 dually resistant mutants under selective drug pressure. Proc Natl Acad Sci U S A 93: 6106–6111.
  26. 26. Nora T, Charpentier C, Tenaillon O, Hoede C, Clavel F, et al. (2007) Contribution of recombination to the evolution of human immunodeficiency viruses expressing resistance to antiretroviral treatment. J Virol 81: 7620–7628.
  27. 27. Bretscher MT, Althaus CL, Muller V, Bonhoeffer S (2004) Recombination in HIV and the evolution of drug resistance: for better or for worse? Bioessays 26: 180–188.
  28. 28. Althaus CL, Bonhoeffer S (2005) Stochastic interplay between mutation and recombination during the acquisition of drug resistance mutations in human immunodeficiency virus type 1. J Virol 79: 13572–13578.
  29. 29. Bocharov G, Ford NJ, Edwards J, Breinig T, Wain-Hobson S, et al. (2005) A genetic-algorithm approach to simulating human immunodeficiency virus evolution reveals the strong impact of multiply infected cells and recombination. J Gen Virol 86: 3109–3118.
  30. 30. Fraser C (2005) HIV recombination: what is the impact on antiretroviral therapy? J R Soc Interface 2: 489–503.
  31. 31. Gheorghiu-Svirschevski S, Rouzine IM, Coffin JM (2007) Increasing sequence correlation limits the efficiency of recombination in a multisite evolution model. Mol Biol Evol 24: 574–586.
  32. 32. Vijay NNV, Vasantika , Ajmani R, Perelson AS, Dixit NM (2008) Recombination increases HIV fitness but not necessarily diversity. J Gen Virol 89: 1467–1477.
  33. 33. Dixit NM (2008) Modelling HIV infection dynamics: The role of recombination in the development of drug resistance. Future HIV Ther 2: 375–388.
  34. 34. Nowak MA, May RM (2000) Virus Dynamics: Mathematical Principles of Immunology and Virology. New York: Oxford University Press.
  35. 35. Perelson AS (2002) Modelling viral and immune system dynamics. Nat Rev Immunol 2: 28–36.
  36. 36. Muller V, Bonhoeffer S (2003) Mathematical approaches in the study of viral kinetics and drug resistance in HIV-1 infection. Curr Drug Targets Infect Disord 3: 329–344.
  37. 37. Suryavanshi GW, Dixit NM (2007) Emergence of recombinant forms of HIV: dynamics and scaling. PLoS Comput Biol 3: e205. doi:10.1371/journal.pcbi.0030205.
  38. 38. Martinez-Cajas JL, Wainberg MA (2007) Protease inhibitor resistance in HIV-infected patients: molecular and clinical perspectives. Antiviral Res 76: 203–221.
  39. 39. Dixit NM, Perelson AS (2005) HIV dynamics with multiple infections of target cells. Proc Natl Acad Sci U S A 102: 8198–8203.
  40. 40. Rhodes TD, Nikolaitchik O, Chen JB, Powell D, Hu WS (2005) Genetic recombination of human immunodeficiency virus type 1 in one round of viral replication: effects of genetic distance, target cells, accessory genes, and lack of high negative interference in crossover events. J Virol 79: 1666–1677.
  41. 41. Christiansen FB, Otto SP, Bergman A, Feldman MW (1998) Waiting with and without recombination: the time to production of a double mutant. Theor Popul Biol 53: 199–215.
  42. 42. Ribeiro RM, Bonhoeffer S (2000) Production of resistant HIV mutants during antiretroviral therapy. Proc Natl Acad Sci U S A 97: 7681–7686.
  43. 43. Yin P, Das D, Mitsuya H (2006) Overcoming HIV drug resistance through rational drug design based on molecular, biochemical, and structural profiles of HIV resistance. Cell Mol Life Sci 63: 1706–1724.
  44. 44. Jung A, Maier R, Vartanian JP, Bocharov G, Jung V, et al. (2002) Multiply infected spleen cells in HIV patients. Nature 418: 144–144.
  45. 45. Sato H, Orenstein J, Dimitrov D, Martin M (1992) Cell-to-cell spread of HIV-1 occurs within minutes and may not involve the participation of virus-particles. Virology 186: 712–724.
  46. 46. Dixit NM, Perelson AS (2004) Multiplicity of human immunodeficiency virus infections in lymphoid tissue. J Virol 78: 8942–8945.
  47. 47. Chen BK, Gandhi RT, Baltimore D (1996) CD4 down-modulation during infection of human T cells with human immunodeficiency virus type 1 involves independent activities of vpu, env, and nef. J Virol 70: 6044–6053.
  48. 48. Piguet V, Gu F, Foti M, Demaurex N, Gruenberg J, et al. (1999) Nef-induced CD4 degradation: a diacidic-based motif in Nef functions as a lysosomal targeting signal through the binding of β-COP in endosomes. Cell 97: 63–73.
  49. 49. Ross SM (2007) Introduction to Probability Models. San Diego, California: Academic Press.
  50. 50. Boerlijst MC, Bonhoeffer S, Nowak MA (1996) Viral quasi-species and recombination. Proc R Soc Lond B Biol Sci 263: 1577–1584.
  51. 51. Gabrielson J, Weiner D (2000) Pharmacokinetic and Pharmacodynamic Data Analysis: Concepts and Applications. Stockholm, Sweden: Apotekarsocieteten.
  52. 52. Speirs C, van Nimwegen E, Bolton D, Zavolan M, Duvall M, et al. (2005) Analysis of human immunodeficiency virus cytopathicity by using a new method for quantitating viral dynamics in cell culture. J Virol 79: 4025–4032.