You are on page 1of 10

AIDS RESEARCH AND HUMAN RETROVIRUSES Volume 20, Number 12, 2004, pp.

13141323 Mary Ann Liebert, Inc.

Mutation, Fitness, Viral Diversity, and Predictive Markers of Disease Progression in a Computational Model of HIV Type 1 Infection

ABSTRACT The aim of this study was to develop a computational model of HIV infection able to simulate the natural history of the disease and to test predictive parameters of disease progression. We describe the results of a numerical simulation of the cellular and humoral immune response to HIV-1 infection as an adaptive pathway in a bit-string space. A total of 650 simulations of the HIV-1 dynamics were performed with a modified version of the CeladaSeiden immune system model. Statistics are in agreement with epidemiological studies showing a log normal distribution for the time span between infection and the development of AIDS. As predictive parameters of disease progression we found that HIV-1 accumulates bit mutations mainly in the peptide sequences recognized by cytotoxic CD8 T cells, indicating that cell-mediated immunity plays a major role in viral control. The viral load set point was closely correlated with the time from infection to development of AIDS. Viral divergence from the viral quasispecies that was present at the beginning of infection in long-term nonprogressors (LTNP) was found to be similar to that found in rapid progressors at the time CD4 T cells drop below the critical value of 200 cells/l. In contrast, the diversity indicated by the number of HIV strains present at the same time was higher for rapid and normal progressors compared to LTNP, suggesting that the early immune response can make the difference. This computational model may help to define the predictive parameters of HIV dynamics and disease progression, with potential applications in therapeutic and vaccine simulations.



ICAL FIELD, extensive work has been done by people in the mathematical biology community to understand the dynamics of HIV evolution within infected hosts. Cellular automata (CA) models of the HIV-1 infection were recently published,14 thus opening new perspectives for the application of mathematical biology to HIV disease. In particular, the viral load in HIV infection has been studied5 as a function of viral growth factor and mutation. CA modeling is based on a discrete simulation space with discrete entities interacting according to predefined rules. This allows us to simulate the complexity of biological

entities such as the immune cells and viral quasispecies in a computational model. In this study we report several simulations of HIV-1 dynamics performed with a modification of the CeladaSeiden immune system model (CS model). The original model was described elsewhere6,7 and we recently published a description of the enhancements to the model to include the HIV-1 infection.8 The statistical analysis of a set of simulations indicates that the model reproduces a number of clinical observations, and allows the predictive markers of HIV dynamics and disease progression to be interpreted. This computational model of HIV infection may provide information relevant to the settings of antiretroviral treatments and vaccine simulations, suggesting a potential application in testing different therapeutic regimens in machina.



Applicazioni del Calcolo (IAC) M. Picone, National Research Council (CNR), Rome, Italy. Institute for Infectious Diseases L. Spallanzani, IRCCS, Rome, Italy.




FIG. 1. Simulation space. The space of the simulation is a two-dimensional honeycomb grid. Cells and molecules interact locally (i.e., within each grid point) and diffuse randomly to adjacent sites. The behavior of each entity is briefly reported in Table 1 together with the set of other entities with which it interacts.

MATERIALS AND METHODS A stochastic cellular automata bit-string approach

Starting from the plain CS model,6 which already allows the simulation of viral infections, we implemented a number of specific features of HIV-1. Our model accounts for the major lymphocytes and myeloid cells in a microliter of peripheral blood.7 The model includes major classes of cells of the lymphoid lin-

eage [T helper lymphocytes (TH), cytotoxic T lymphocytes (CTL), B lymphocytes, and antibody-producer plasma cells (PLB)] and some of the myeloid lineage [macrophages (MA) and dendritic cells (DC); Fig. 1 and Table 1]. Cells and molecules interact according to rules expressed in the form of conditions (e.g., the state of the cell and/or the affinity between their receptors) and actions (e.g., change of state of the interacting cells, phagocytosis, etc.). The time resolution of the simulation is equivalent to 8 hr of real life. The simulation is car-

TABLE 1. Entity Lymphocyte B



THEIR FUNCTIONS Interacts with Antigen, CD4 T

Function Phagocytoses antigen upon recognition of one of its epitopes by means of the receptor (BCR); processes antigen peptides and presents peptides together with class II MHC molecule to TH; differentiates in plasma B cell upon stimulation by TH Produce antibodies of the IgG idiotype Induces B cell clone expansion upon recognition of the MHC II peptide complex on the B cell surface by one of its receptors (TCR); induces CTL clone expansion by secretion of interleukin-2 Kills infected cells upon recognition of the MHC I peptide complex on the infected cell surface by one of its receptors (TCR) Presents viral peptides to TC cells Phagocytoses antigen aspecifically; processes antigen peptides and presents peptides together with class II MHC molecule to TH Infects TH, MA, and DC; reproduces inside actively infected cells and causes rupture of the cell membrane with release of its viral contents

Lymphocyte plasma B (PLB) Lymphocyte T helper (CD4 T) Lymphocyte T cytotoxic (TC, CD8, T, CTL) Dendritic cell (DC) Macrophage (MA) HIV-1


1316 ried out in discrete steps that are repeated up to 20,000 times corresponding to about 18 years. Eventually the appearance of an emergent viral strain determines the death of the host (i.e., the simulation stops). This is arbitrarily declared when the viral load exceeds 108 copies/ml. In addition to representing both the cellular and the humoral immune response at one time, this model takes into account the concept of clones or repertoire of lymphocytes.9,10 The model gives us the chance to represent a vastalthough not so large number of different cell receptors. We use a sequence of bits (bit-string) to code for the receptor so that with l bits we represent a repertoire whose size is equal to 2l. Receptor binding is modeled through a probability, which is a function of the bitstring match.6,7 Nonperfect match allows cross-reactivity among recognized antigens and their mutants. The activation of the virus (beginning of transcription) happens with a certain probability pw. In such a way, we account for variants of the virus with a low value of pw, which may be interpreted either as strains having a poor adaptation (those having good chances to become extinct) or as strains that are activated very late. Newly assembled HIV-1 in productively infected cells accumulates inside the cell at a rate given by another parameter pr. With the same rate pr, a part of these virions bud from cell membranes. Hence, if pr is high, the accumulation of virions inside the cells causes cell rupture and consequent release of viral content into the plasma. Finally, HIV-1 mutates in productively infected cells. The mutation rate is given by pm, the last of the three parameters pw, pr, pm, which, together with the bit-strings for the viral phenotype, identify a viral strain. Reservoirs for HIV-1 are very important in understanding the dynamics of HIV-1, especially in correlation with the ability to survive highly active antiretroviral therapy.11,12 In our model we take into account the reservoirs provided by longlived memory CD4 T cells ignoring other potential anatomical sanctuaries (e.g., brain or eyes).13 To take into account the impairment of CD4 T cell production,14,15 we implemented a simple mechanism by which the probability of decreasing the number of CD4 T cell produced by the bone marrow by a factor of one cell every day is pr x2/(C 2 x2), where x is the number of CD4 T cells that are productively infected and C is a constant chosen proportionally to the original count of CD4 T cells in the simulated space. Lastly, it is worth mentioning that the growth of clones of lymphocytes is limited stochastically, since the probability for a cell that is in the mitotic cycle to create a copy of itself is multiplied by a factor that decreases exponentially with the total number of lymphocytes present in its neighborhood. In the context of CD4 T cell selective loss during HIV infection, this mechanism introduces a competition between CD4 and CD8 for replenishment and resembles, overall, the blind homeostasis hypothesis16 used in another mathematical model of HIV infection.17

CASTIGLIONE ET AL. virus in an individual at a given time (we will also be referring to them as escape mutants), or, in other words, the number of genetically different viruses.19 Hence, here, HIV-1 is represented by a collection of bit-strings coding for peptides and epitopes. The bit-string length used for the simulations presented herein is l 12, which gives a potential repertoire of 2l 4096 cell receptors, epitopes, and peptides (see Farmer et al.20 for the first use of a bit-string model in immunology). Moreover, since for simplicity HIV-1 is here represented as just one peptide (namely the T cell peptide) and one epitope (namely the B cell peptide), the potential number of viral strains is 224 108. With respect to our previous study,8 we have an additional level of description by which we can specify the functional properties of the simulated virus. We map the genotype to the phenotype by means of a simple formula that assigns to the activation, replication, and mutation rate, the triplet of numbers pw, pr, and pm between 0 and 1, respectively. The values pw, pr, and pm are calculated from different nonoverlapping zones of the binary string that describes the single epitope of the virus. Since the bit-mutation is completely random, it may flip any of the bits representing the peptide or the epitope. In either case a different outcome is obtained: (1) if the peptide is modified, the affinity with the molecules MHC class I or II changes. This accounts for the appearance of variants of the virus that are not subject to the cytotoxic activity of CD8 cells; (2) if the epitope is modified, then one of the three values of the triplet (pw, pr, pm) is modified. The range of possible values for pw, pr, and pm is chosen by looking at the results of a set of preliminary runs in such a way as to reproduce, on average, the three-phase dynamics of HIV infection. More to the point, 105 pw 104, 105 pr 104, and 104 pm 103 (recall that pm is a per-bit mutation rate). Note that each rate can assume only a subset of discrete values in these intervals. In particular, since we use four bits, there are 16 possible values for each rate.

RESULTS Disease progression and viral diversity

We aimed at reproducing the acute, chronic, and AIDS phase of disease progression in terms of viral load, viral diversity, and CD4 T cell count. In addition to virus lymphotropism, a suitable set of initial values for the triplet pw, pr, and pm, which determine the phenotype of the wild-type HIV-1, was required. The correct values had to give a reasonable acute phase no longer than a few weeks, with the virus load decreasing considerably.21 The results of 1 of 650 simulations are shown in Figure 2, which reproduces quite well the dynamics of the virus and CD4 T helper cell count as described by Fauci et al.21 In this representative case the latent phase of the infection lasts approximately 5.8 years. The acute phase is visible for the contemporary sharp peak of the viral load and a sudden decrease of the CD4 count (Fig. 2a). This is followed by a temporary rebound of CD4 cells, which eventually fall below the critical value of 200 cells/l, triggering a phase of severe immunodeficiency that results in AIDS and death. The viral load during the latent phase presents fluctuations called blips (Fig. 2b, in-

The HIV-1 phenotype: evolution in a bit-string space

In the present work, we follow the evolutionary hypothesis of Wolinsky et al.18 Since a model suitable for exploring such a hypothesis must entail the concept of viral diversity, we define viral diversity as the number of different variants of the



FIG. 2. Viral load, CD4 T counts, viral diversity, and viral divergence in a typical simulation. (a) The three-phase dynamics of HIV-1 infection.21 (b) The viral load (inset plot is in log scale to evidence the blips). (c) The total number and the count for memory CD4 T cells/l. (d) The total number of HIV strains and the number of escape mutants, i.e., the number of survivors. (e) The viral divergence, i.e., the distance from wild type in terms of bit-mutations (over 24 possible). Recall that a single time step of the simulation corresponds to 8 hr of real life.

1318 set), i.e., peaks of high viremia followed by temporary immune system reactions able to control them.22 Figure 2c shows the CD4 T memory cell count. A decrease of CD4 T memory cells is observed in conjunction with the global breakdown of the CD4 compartment at the beginning of the AIDS phase. Finally, upon emergence of uncontrolled viral strains, the number of CD4 T memory cells almost vanishes and the virus may be observed in large amounts in the blood compartment. About 80 different viral strains are produced by mutation from the original wild-type virus (which enters the system at time zero), but only about 25% of them elude immune surveillance (Fig. 2d). The average number of CD4 T cells that are productively infected at any time during the life of an infected individual is about 20/l, which is equivalent to a total of 20 5 106 108 in a normal adult with 5 liters of blood. This is in agreement with clinical data suggesting that productively infected CD4 T cells are estimated to be around 107108.23 The original wild-type viral strain evolves, by mutation, under the selective pressure of the immune response.19 The acute phase is followed by the silent one, during which the level of free virions is very low. In this phase the immune system struggles to keep the virus load at low levels but cannot avoid its mutation. Figure 2d shows the diversity defined as the current number of genetically different viruses. Within the set of viral strains, which includes those cleared by the immune system, we distinguish the escape mutants, those able to survive the immune system reaction. The HIV-1 divergence during the course of the disease, defined as genetic distance from the wild-type virus for the same simulation, is plotted in Figure 2e. Interestingly, we found a pattern of viral divergence and evolution of viral diversity that is in qualitative agreement with reported clinical results,19 showing that the divergence increases during disease progress and, in particular, during periods of strong viral rebound.


Estimated time to AIDS

In the absence of antiretroviral treatments, the time between infection in adults and the development of AIDS is approximately 712 years in normal progressors (median time is about 10 years).26 Rapid- and long-term progressors proceed to immunodeficiency status in a shorter and longer time period, respectively. We computed time to AIDS as the time the CD4 count falls below 200 in 1 l of blood. The distribution of results is shown in Figure 4a. Interestingly, such a distribution is well fitted by a log-normal distribution that, in general, can be theoretically expected under the assumption of a degradation process resulting from failures.27 Muoz and Xu28 evaluated a parametric model for the incubation of AIDS and found that the log-normal model provides a good fit to the data at their disposal. Log-normal distribution was confirmed by other studies as well.29 Moreover, clinical studies report that 12% of individuals do not develop AIDS within 20 years.30,31 We simulate about 18.3 years, and estimate that 12.3% of individuals do not develop AIDS within this time interval. Moreover, the literature shows that 5% of infected people develop AIDS within 3 years from infection.28,31,32 Accordingly, we found that 5% of our sample develops AIDS in 2.6 years.

Viral diversity threshold

Nowak and co-workers33 proposed that due to selection by the immune response, the diversity of the virus population would increase in time. A limit would be reached at which the immune system could not cope with the level of antigenic diversity. Our experimental setting is optimal for testing the antigenic diversity threshold since we can compare rates of evolution among individuals infected with the same source and having the same HLA types. The outcome is that our simulation agree with the theory of antigenic diversity threshold,33 in that rapidterm progressors succumb at a viral diversity threshold higher than long-term progressors, although we did not find a sharp transition. The plot in Figure 4b shows the ratio of the aforementioned quantities (i.e., escape/total HIV strains at the time of death). It appears that a large fraction (from 15% up to 50%) of strains survives in rapid progressors, whereas for long-term progressors the fraction of surviving strains remains close to the threshold value of 10%. The conclusion is that the number of escape mutants alone is a poor indicator of disease progression. The ratio of escape mutants to total viral diversity could, instead, be used as an indicator of disease progression.

Cytotoxicity and selection of mutants

Figure 3a shows the number of cells that are infected and those that present to the immune system the viral epitope bound on the class I MHC molecule in a typical simulation. This graph shows that the number of infected cells significantly increases as time goes by, while the number of infected cells that also present the antigen is rather stable. During the chronic phase the virus undergoes a rapid and massive turnover that, along with the high rate of errors accumulated by the error-prone reverse transcription mechanism, is able to produce up to 15% of differences compared with the original viral strain.24 Figure 3c shows the distribution, obtained in our numerical experiments, of the distance between the T cell epitope of the best-selected viral strain (the one with the largest number of copies in the simulated space) and the T cell epitope of the wild-type strain. It indicates that the cytotoxic immune response selects strains having low affinity with the class I MHC molecule. We can see that compared to Figure 3b, which shows the same distribution computed for the B cell epitope, HIV-specific CTLs are more active in selecting HIV strains with a low level of detection.25 We found that an average of 3.6 bit-mutations (out of 24) accounts for about 15% of the genetic diversity, very close to the value found for the viral envelope gene.24

Viral-load set point and CD4 T counts as markers of disease progression

Viral load measurement at the initial assessment is a very accurate predictor of prognosis. Recently, a set of interim recommendations has been made with respect to the use of viral load measurements to direct clinical care.34 Simulations in Figure 5a show a negative correlation between viral load (computed by averaging the number of free HIV-1 over the 624 month period after infection) and time to AIDS. However, the correlation shows at least two different regimes in correspondence with fast-progressors and non-fast progressors.



FIG. 3. Infected versus presenting cells and viral antigenic shift. (a) The number of productively infected cells and those that are presenting the viral epitope bound on the class I MHC molecule in a typical simulation. (c) The distribution of the distance between T cell epitopes of the best-selected viral strain and of the wild-type strain. It indicates that the cytotoxic immune response selects strains having low affinity to the class I MHC molecule. Compared to (b) showing the same distribution computed for the B cell epitope, we see that HIV-specific CTLs are more active in selecting HIV strains with a low level of detection.

A realistic correlation was observed between CD4 and CD8 count. It is interesting to observe how this relation changes as the disease progresses. In particular, shots of the ratio CD4/CD8 are taken at 1 (1.79 0.13), 3 (1.23 0.31), 5 (0.78 0.36), and 7 (0.47 0.37) years after infection and are significantly different for all the performed simulations (p 0.05, MannWhitney two-tailed test). The plots in Figure 5be show such correlations at different times of disease progression. Negative

correlation starts to be visible a few years after infection (the correlation is 0.0002 after 1 year, 0.62 after 3 years, 0.59 after 5 years, and 0.52 after 7 years). This can be explained by the cytotoxic activity of CD8 HIV-1-specific cells, which kill the infected CD4 T cells. The clinical significance of determining a CD4/CD8 ratio has been a matter of discussion.35 Our data clearly indicate that the CD4/CD8 ratio becomes significantly associated with HIV disease progression only a few years after infection.



FIG. 4. Natural course of HIV disease, viral diversity, and diversity threshold. (a) Histogram of the time elapsed between the primary infection and the onset of AIDS. The fit with a log normal by means of the MarquandatLevenberg nonlinear fitting algorithm is in quite good agreement with experimental data (mean 7.7 years, median 7.4 years). (b) The ratio between the number of escape HIV mutants and the total HIV strains versus the time of death (each dot represents one simulated individual). It appears that a large fraction (from 15% up to 50%) of strains survives in rapid progressors, whereas for long-term progressors the fraction of surviving strains remains close to the threshold value of 10%.

We used a model to study HIV-1 infection and, in particular, to reproduce the three-phase dynamics of its evolution. The model employed is well suited for a detailed description of the evolution of the viral strains in the bit-string space. Realistic correlations between CD4 and CD8 counts and viral load are found. Statistics computed over a large number of simulations representing identical initial conditions are in good agreement with corresponding clinical studies, recovering a log-normal

distribution for the time to AIDS. We showed how the immune system drives the evolution of the virus, which is related to the ability of HIV-1 to survive the pressure of the cytotoxic response while impairing immune cells production. Accordingly, we found that the cytotoxic immune response selects strains whose peptide sequences have low affinity with the class I MHC molecule. Nowak and co-workers33 proposed that due to selection by the immune response, the diversity of the virus population would increase in time. A limit would be reached at which the



FIG. 5. Viral set point and correlation between CD4 and CD8 in HIV disease progression. (a) The average value of the viral load computed over the latent phase versus time to AIDS (i.e., the time when the CD4 count falls below 200/l). Points are not well fitted by a straight line in a semilog plot but show at least two different regimes in correspondence to fast progressors and non-fast progressors. (b) CD4 correlates negatively with CD8. It is interesting to observe how this relation changes together with disease progression. The figures show the same correlation taken 1, 3, 5, and 7 years after infection (each dot represents a simulated individual).

1322 immune system could not cope with the level of antigenic diversity. Experimental evidence indicated that the opposite could be true, namely that rapid progressors could succumb to AIDS at a diversity threshold smaller than the threshold found for slow progressors.36,37 A plausible explanation for the correlation would be that in case of weak resistance, the virus quasispecies does not have to evolve new traits to overcome hurdles imposed by the immune system. Nowaks reply was that the diversity threshold is dependent on the nature of the CTL response itself, being a function of the HLA type of the infected patients. However, studies to check this hypothesis were not convincing. The results of our simulations also show that a larger HIV-1 diversity is produced in long-term progressors (not shown), probably because a stronger immune response is needed to reduce HIV-1 disease progression. However, looking at the ratio between escape HIV mutants (i.e., surviving strains) and total HIV strains vs. time to death it appears that a large fraction of strains survives in rapid-term progressors, whereas for long-term progressors the fraction of surviving strains remains below the threshold value of about 10%. The conclusion is that the number of escape HIV mutants alone is a poor indicator of disease progression. Differently, the ratio of escape HIV mutants to total HIV viral diversity could be used as an indicator of disease progression but, unfortunately, this is difficult to be determined with current clinical techniques. Another possible indicator of fast progression of the disease is the genetic diversity of HIV-1 in vivo together with a weak CTL response and high viral load.18 Overbaugh and Bangham38 state that It seems likely that both viral genetics/fitness and host selection pressure determine the extent of diversity and rate of progression. This result is very different from our simulation, which suggests that only genetics/fitness determines the extent of diversity because different runs (with the same selection pressure) generate different outcomes in antigenic diversity. Finally, a negative correlation between CD4 and CD8 counts starts to be visible a few years after infection probably due to the cytotoxic activity of CD8 HIV-1-specific cells that kill the infected CD4 T cells. From the viewpoint of antiretroviral HIV therapy, the computational model of HIV-1 infection offers issues for discussion concerning the role of the immune system and viral evolution. At present, randomized clinical trials provide strong evidence indicating that HIV-infected patients with a CD4 T cell count below 200/mm3 should be treated, but the golden moment at which to start antiretroviral therapy on patients with asymptomatic HIV infection and a CD4 T cell count above 200/mm3 and among patients in primary HIV infection remains unclear. Moreover, the reintroduction of therapy during structured treatment interruptions following early therapy of acute HIV infection or facilitating salvage after failed therapy should preferably be attempted in cohort studies or clinical trials. Thus, the identification of new parameters in disease progression could be useful in HIV therapy. To this aim, our model shows that CD8 T cell-mediated immunity plays a major role in viral control and that early immune response is crucial for viral control. A more focused analysis of these immunological parameters may provide additional information for a patient-tailored treatment.

CASTIGLIONE ET AL. In conclusion, the simulator can provide an effective and inexpensive tool to explore HIV-1 and to disclose interesting features related to its dynamics. The extrapolation of these data in clinical situations may help to improve treatment choices with the best expected outcome. In the contest of HIV vaccine design, the nature and type of immunity that should be induced are still a matter of debate. Antigen selection, delivery system, adjuvants, and different roots of administration create hundreds of possible combinations that can hardly be tested directly in humans or in suitable primate models. Therefore, this computational model of HIV infection might help to devise in vitro or in vivo experiments, as well as to test in machina various therapeutic regimes.

We wish to thank CASPUR for computing time. Alan S. Perelson and Avidan U. Newmann are kindly acknowledged for useful discussions. Luisa Santoro is kindly acknowledged for careful reading of the paper. F.P. and G.D. were supported by grants from the Italian Ministry of Health and from the Instituto Superiore di Sanit.

1. Pandey RB and Stauffer D: Metastability with probabilistic cellular automata in an HIV infection. J Stat Phys 1990;61:235240. 2. Pandey RB: Cellular automata approach to interacting cellular network models for the dynamics of cell population in an early HIV infection. Physica A 1991;179:442470. 3. Hershberg UR, Louzoun Y, Atlan H, and Solomon S: HIV time hierarchy: Winning the war while losing all the battles. Physica A 2001;289:178190. 4. Zorzenon dos Santos RM and Coutinho S: Dynamics of HIV infection: A cellular automata approach. Phys Rev Lett 2001;87: 168102.14. 5. Ruskin HJ, Pandey RB, and Liu Y: Viral load and stochastic mutation in a Monte Carlo simulation of HIV. Physica A 2002;311: 213220. 6. Celada F and Seiden P: A computer model of cellular interaction in the immune system. Immunol Today 1992;13:5662. 7. Bernaschi M and Castiglione F: Design and implementation of an immune system simulator. Computers Biol Med 2001;31:303331. 8. Bernaschi M and Castiglione F: Selection of escape mutants from immune recognition during HIV infection. Immunol Cell Biol 2002;80:307313. 9. Berek C and Milstein C: The dynamic nature of the antibody repertoire. Immunol Rev 1988;105:526. 10. Davis MM and Bjorkman PJ: T-cell antigen receptor genes and Tcell recognition. Nature 1988;334:395402. 11. Kuritzkes DR: HIV pathogenesis and viral markers. MEDSCAPE Portals, April 28, 2000. 12. Finzi D, Hermankova M, Pierson T, Carruth LM, Buck C, Chaisson RE, et al.: Identification of a reservoir for HIV-1 in patients on highly active antiretroviral therapy. Science 1997;278:1295 1300. 13. Graham BS: Infection with HIV-1. Br Med J 1998;317:12971301. 14. McCune JM: The dynamics of CD4 T-cell depletion in HIV disease. Nature 2001;410:974979.


15. Douek DC, Betts MR, Hill BJ, Little SJ, Lempicki R, Metcalf JA, et al.: Evidence for increased T cell turnover and decreased thymic output in HIV infection. J Immunol 2001;167:66636668. 16. Adleman LM and Wofsy D: T-cell homeostasis: Implications in HIV infection. J Acquir Immune Defic Syndr 1993;6:144152. 17. Mehr R and Perelson AS: Blind T-cell homeostasis and the CD4/CD8 ratio in the thymus and peripheral blood. J Acquir Immune Defic Syndr Hum Retrovirol 1997;14:387398. 18. Wolinsky SM, Korber BTM, Neumann AU, Daniels M, Kunstman KJ, Whetsell AJ, et al.: Adaptive evolution of human immunodeficiency virus-type 1 during the natural course of infection. Science 1996;272:537542. 19. Shankarappa R, Margolick JB, Gange SJ, Rodrigo AG, Upchurch D, Farzadegan H, et al.: Consistent viral evolutionary changes associated with the progression of human immunodeficiency virus type 1 infection. J Virol 1999;73:1048910502. 20. Farmer JD, Packard N, and Perelson A: The immune system, adaptation and machine learning. Physica D 1986;22:187204. 21. Fauci AS, Pantaleo G, Stanley S, and Weissman D: Immunopathogenic mechanisms of HIV infection. Ann Intern Med 1996;124: 654663. 22. Percus JK, Percus OE, Markowitz M, Ho DD, Di Mascio M, and Perelson A: The distribution of viral blips observed in HIV-1 infected patients treated with combination antiretroviral therapy. Bull Math Biol 2003;65(2):263277. 23. Haase AT, Henry K, Zupancic M, Sedgewick G, Faust RA, Melroe H, et al.: Quantitative image analysis of HIV-1 infection in lymphoid tissue. Science 1996;274:985989. 24. Learn GH Jr, Korber BT, Foley B, Hahn BH, Wolinsky SM, and Mullins JI: Maintaining the integrity of human immunodeficiency virus sequence databases. J Virol 1996;70:57205730. 25. Goulder PJR, Phillips RE, Colbert RA, Ogg G, Nowak MA, Giangrande P, et al.: Late escape from an immunodominant cytotoxic T lymphocyte response associated with progression to AIDS. Nat Med 1997;3:212217. 26. Lifson AR, Rutherford GW, and Jaffe HW: The natural history of human immunodeficiency virus infection. J Infect Dis 1988;158: 13601367. 27. Kolmogorov AN: Uber das logarithmisch normale Verteilungsgesetz der Dimensionen der Teilchen bei Zerstuckelung. Dockl Akad Nauk SSSR 1941;31:99101. 28. Muoz A and Xu J: Models for the incubation of AIDS and variations according to age and period. Stat Med 1996;15:24592473.

29. Tassie JM, Grabar S, Lancar R, Deloumeaux J, Bentata M, and Costagliola D: Time to AIDs from 1992 to 1999 in HIV-1-infected subjects with known date of infection. J Acquir Immune Defic Syndr 2002;30:8187. 30. Sheppard HW, Lang W, Ascher MS, Vittinghoff E, and Winkelstein W: The characterization of non-progressors: Long-term HIV-1 infection with stable CD4 T-cell levels. AIDS 1993;7:11591166. 31. Mellors JW, Muoz A, Giorgi JV, Margolick JB, Tassoni CJ, Gupta P, et al.: Plasma viral load and CD4 lymphocytes as prognostic markers of HIV-1 infection. Ann Intern Med 1997;126:946954. 32. Phair J, Jacobson L, Detels R, Rinaldo C, Saah A, Schrager L, et al.: Acquired immune deficiency syndrome occurring within 5 years of infection with human immunodeficiency virus type-1: The Multicenter AIDS Cohort Study. J Acquir Immune Defic Syndr 1992;5:490496. 33. Nowak MA, May RM, and Anderson RM: The evolutionary dynamics of HIV-1 quasispecies and the development of immunodeficiency. AIDS 1990;4:10951103. 34. The US Department of Health and Human Services (DHHS): Guidelines for the use of Antiretroviral agents in HIV-1 infected adults and adolescents, July 14, 2003. 35. Fahey JL, Taylor JM, Detels R, et al.: The prognostic value of cellular and serologic markers in infection with human immunodeficiency virus type 1. N Engl J Med 1990;322(3):166172. 36. Delwart EL, Pan H, Sheppard HW, Wolpert D, Neumann AU, Korber B, et al.: Slower evolution of human immunodeficiency virus type 1 quasispecies during progression to AIDS. J Virol 1997;71: 74987508. 37. Delwart EL, Sheppard HW, Walker BD, Goudsmit J, and Mullins JI: Human immunodeficiency virus type 1 evolution in vivo tracked by DNA heteroduplex mobility assays. J Virol 1994;68:66726683. 38. Overbaugh J and Bangham CR: Selection forces and constraints on retroviral sequence variation. Science 2001;292:11061109.

Address reprint requests to: Filippo Castiglione Istituto Applicazioni del Calcolo (IAC) M. Picone National Research Council (CNR) Viale del Policlinico 137-00161 Rome, Italy E-mail: