Immune control of HIV-1 infection after therapy interruption: immediate versus deferred antiretroviral therapy

We started from the analysis of the cohort data of naïve patients that received HAART (the therapeutic regimen included protease inhibitors, nucleoside reverse transcriptase inhibitors and non-nucleoside reverse transcriptase inhibitors) within six months from primary infection (i.e., including both the inflammatory phase and after seroconversion). Table 1 reports clinical information about the cohort data. In Table 2 patients have been classified according to the staging method proposed by Fiebig et al [9].

After a variable period of therapy (average: 3 ± 1 years), all subjects were enrolled in Structured Therapy Interruption (STI). For the majority of the subjects, plasma viremia reverts to detectable levels in a median time of ten days following the initial HAART interruption. Despite this rapid viral rebound, five subjects managed spontaneously to control viremia maintaining their plasma virus load below the detection limit for a longer time (see Figure 1 and 2). However, after 1 month, the virus was again detectable and the patients were enrolled in STI protocols. The CD4⁺ T cell response declined in absence of treatment though remaining well above the level of 200 cells/μl defined as the onset of full-blown disease. However, with re-institution of HAART the CD4⁺ T cell response rose again. Interestingly, in all five subjects exposure to the virus is associated with a substantial increase of the cytotoxic response.

We used the cohort data, in particular the time window before patients were enrolled in the STI protocol, to set the initial conditions of the model. First of all, we performed computer simulations in which we started a long course of HAART (2.5 years) within six months from primary infection (i.e., including both the inflammatory phase and after seroconversion). Then, we fixed immunological parameters at therapy start time on the basis of the average values measured in in vivo patients: (5.8 ± 0.2) RNA copies/ml (in logarithmic scale), (870 ± 50) CD4 cells/μl and (430 ± 50) CD8 cells/μl. Finally, the drug control effect of HAART is to suppress viremia, during the first ten weeks, below the detection level. This occurs in three phases with a specific time table [10, 11]. We have implemented the model to take into account this dynamics by considering the probability that a virus infects a cell during HAART, follows a characteristic power low decay. The average viremia of hundreds of silico patients has been compared with those obtained from in vivo patients whose viral load was known in the same time window (see Figure 3).

Due to the inability to eliminate virus infected cells, antiretroviral treatment fails to eradicate the infection and cessation of suppressive therapy will, most likely, be accompanied by viral rebound. We have analyzed the pace of viral rebound and the ability of early-treated in silico patients to spontaneously control HIV-1 replication, in relation to the outcomes of in vivo studies. We found that during the first month after treatment interruption, despite viral rebound, 16% of in silico patients are able to achieve, at least, a transient steady state off therapy, with viral load below the limit of detection (Figure 4 panel a, green rectangle in shadowed area). Our results are in good agreement with cohort data that give about 22% of successful cases of immune control (panel a, blue rectangle in shadowed area). After that, the virus becomes sparser but still detectable and only 2% of in silico patients manage spontaneously to control viremia six months after the initial discontinuation (shadowed area in panel b). Finally, the virus rebounds in one year off therapy, to reach the value at which it would level off in the absence of treatment (green rectangle in panel c). These findings are confirmed by in vivo data in which the median viremia one year after stopping HAART is about 10,000 RNA copies/ml (i.e., log(RNA/ml) = 4, blue rectangles in panel c). Despite considerable advances in the understanding of the interplay between HIV-1 and its host, the effect of antiretroviral treatment on the developing immune response remains elusive. During the acute stage of HIV-1 infection in untreated subjects, viral load expands exponentially and antiviral immune defenses are still developing [12]. Once the HIV-specific immune response has been established, viral load usually decreases until a viral set point is reached. It has been shown that initiation of HAART before the attainment of the natural equilibrium of virus load may blunt the natural anti-HIV immune activity [12]. We have investigated this issue and compared the results with those obtained in silico when HAART is delayed. In particular, in order to determine the optimal time to initiate HAART, we performed simulations in which we started a long course of HAART (2.5 years) either during seroconversion (Figure 5, green lines in panels a, b and c) or during the chronic stage of the disease (red lines of the same panels). In deferred treatment, we decided to start therapy when viral load was (4.0 ± 0.2) copies/ml (in logarithmic scale) and the CD4⁺ T cell count was (400 ± 50) cells/μl. Moreover, we compared CD4⁺ T cell count and viremia between treated and untreated in silico patients (blue lines in panels a, b and c).

Although the impact of a therapeutic intervention on plasma viremia is well documented [13‐15], much less is known about the rapidity of decline with respect to timing of therapy. Simulations reveal that with late treatment plasma viremia declines more rapidly, approaching the limit of detection within the first ten weeks, whereas viral load reaches the undetectable level only after six months when therapy is started at an early stage of disease (see Figure 3). Moreover, the virus rises more slowly in deferred treated patients. After one year, a small percentage (1%) of in silico patients still manage to spontaneously control viremia, whereas the virus rebounds completely one year after therapy discontinuation in early treated infected persons (Figure 5, shadowed area in panel f). The decresing number of in silico patients that spontaneously control viremia, correlates in time strictly with the restoration of the viral set point reached within one year of the termination of HAART. By that time, the efficacy of antiretroviral drugs on viremia is almost vanished and the virus reverts to the value at which it would level off in the absence of treatment (blue rectangle in panel d, e and f).

The evolution of the immune response and, in particular, the kinetics of CD4⁺ and CD8⁺ T cells in patients treated with HAART, are subject of heated debate [12, 16]. Our results confirm the expectation that initiation of HAART during seroconversion leads to the preservation of HIV-specific T helper cells with the consequent maintenance of specific immunity (Figure 5, green line panel b). In addition, we suggest that in silico patients following such an antiviral therapy regimen show a modest loss of CD8⁺ T cell numbers coincident with a decline in viremia (green line in panel c and a). When therapy ends, the viral load rises again within a year, but in in silico patients undergoing early treatment the natural immune activity remains impaired. It thus takes several years to restore the natural equilibrium between the host and its uninvited guest. On the contrary, prior to administration of HAART in chronically infected patients, HIV-1 infection is associated with a failure in T cell homeostasis [17], resulting in a gradual decline in CD4⁺ T cell numbers (red line panel b), whereas the cytotoxic activity is well developed (red line in panel c). Once therapy starts, CD4⁺ T lymphocyte counts quickly increase, bringing the host almost to immune restoration.

Simulation of delayed therapy produces enhancements of both functional immune response and immune control of infection. Our interpretation is that, during the period prior to initiation of therapy, the virus elicits a cytotoxic response. This, by contrast, is almost absent in early treatment, as the magnitude of the infection is considerably reduced by the action of the antiretroviral drugs.

Twenty-two patients (21 male and 1 female) with primary HIV-1 infection were selected at the Clinical Department of the National Institute for Infectious Disease "L. Spallanzani" in Rome. Criteria for diagnosis were: documented seronegative HIV-1 antibody test within the previous 6 months; acute symptomatic seroconversion illness; evolving HIV-specific antibody response by ELISA; positive HIV-DNA PCR in PBMC, or positive plasma HIV-RNA quantification in the absence of an antibody response. Fourteen subjects were enrolled into a pilot study of Structured Therapy Interruption (STI). Two different therapy interruption protocols were used: arm A (seven patients), 4 weeks off/8 weeks on HAART; arm B (seven patients), intermittent therapy guided by plasma HIV-RNA levels, according with pro tempore guidelines. Specifically, patients in arm B resumed treatment only when necessary (plasma HIV-RNA > 50,000 copies/ml) and stopped HAART when HIV-RNA became again undetectable. This on/off treatment regime was undertaken for the first 12 months. At this time point, all patients suspended HAART and were monitored for viral and immunological parameters up to month 24. Eligibility criteria to participate in the study were as follows: CD4⁺ T cell count ≥ 500 cells/μl and HIV viremia in the last two years below the detection limit, to have initiated HAART during HIV primary infection. In addition, eight patients followed unplanned treatment interruption mandating re-initiation of therapy according to medical approval (see Table 1).

The ethical committee of the institute (Clinical Department of the National Institute for Infectious Disease "L. Spallanzani" in Rome) approved this study and the patients gave a written informed consent to the blood sampling for the study.

The model of immune system response we employ has been quite extensively described in [8, 22‐25]. In short, it belongs to a class of models that resort to bit strings to represent "binding sites" of cells and molecules, as for example lymphocyte receptors (T lymphocytes receptor TCR, B lymphocytes receptor BCR), Major Histocompatibility Complexes MHC, antigen peptides and epitopes, immunocomplexes, etc. [26]. The affinity among the different biological entities of the model varies according to the Hamming distance between these bit strings, i.e., according to their complementarity fit. The model includes the major classes of cells of the lymphoid lineage (T helper lymphocytes or TH, cytotoxic T lymphocytes or CTL, B lymphocytes and antibody-producer plasma cells, PLB) and some of the myeloid lineage (macrophages, MA, and dendritic cells, DC). All these entities behave as finite state machines following a set of "rules" that describe the different phases of the recognition and response of the immune system against a pathogen.

The model represents a single lymph node of a vertebrate animal that is mapped onto a three-dimensional ellipsoid lattice with periodic boundary conditions (Figure 6). The primary lymphoid organs thymus and bone marrow are modelled apart: the thymus is implicitly represented by the positive and negative selection of immature thymocytes before they get into the lymphatic system, whereas the bone marrow generates already mature B lymphocytes. Hence, on the lattice there are only immunocompetent lymphocytes. In order to represent the special features of the HIV infection, the model has been enhanced with the description of the i) HIV replication inside infected lymphocytes; ii) T production impairment; iii) specific response against HIV strains; iv) HIV mutation and evolution. The evolution of HIV in untreated patients has been studied and described by means of this model in [27].

The virus is represented by two bit strings (each l bits long); the first one corresponds to the epitope (i.e., the BCR's binding site) and the second one to the peptide (i.e., the MHC class I and II's binding site). The simulator allows the definition of an arbitrary number of epitopes and peptides. In the simulations the bit string length l is equal to 16 corresponding to a potential repertoire of 65536 distinct receptors and molecules. Actually, since the virus is represented by one epitope and a single peptide, each strain is identified by 2 l = 32 bits. So, the potential number of different virus strains becomes equal to 2³². Each time step of the simulation corresponds to eight hours of "real life".

While the virus and the antibodies are uniquely represented (i.e., they are agents like the cells), molecules with small molecular weight like interleukins or chemokines are represented in terms of concentrations and their dynamics described by the following parabolic partial differential equations plus a degradation term accounting for the finite half-life of molecules:

where c = c(x) is the concentration of chemokines, D is the diffusion coefficient and λ is the half-life. We assume D = 3000 μm ²/min and λ = 3 hrs [28, 29]. Differences in cells mobility are taken into account as well. TH cells are the fastest ones, with an average velocity of 11 μm/min, followed by B cells with 6 μm/min and DC with a velocity of 3 μm/min [30].