q-bio0509006/qs.tex
1: % Template article for preprint document class `elsart'
2: % SP 2001/01/05
3: 
4: \documentclass{elsart}
5: 
6: \usepackage{isolatin1}
7: \usepackage[dvips]{graphicx}
8: \usepackage{color}
9: \usepackage{amssymb}
10: \usepackage{amsmath}
11: \usepackage{epsf}
12: \usepackage{subfigure}
13: %\usepackage{showlabels}
14: \setlength{\textwidth}{15cm} \setlength{\oddsidemargin}{-0cm}
15: \setlength{\textheight}{23cm} \setlength{\topmargin}{-.5cm}
16: \newcommand{\rmd}{\text{d}}
17: \newcommand{\dfracp}[2]{\dfrac{\partial #1}{\partial #2}}
18: \newcommand{\dfracd}[2]{\dfrac{\text{d} #1}{\text{d} #2}}
19: \newcommand{\average}[1]{\left\langle #1 \right\rangle}
20: 
21: %\graphicspath{}
22: 
23: \begin{document}
24: 
25: \begin{frontmatter}
26: 
27: 
28: \title{Modeling viral coevolution: HIV multi-clonal persistence and competition dynamics}
29: 
30: \author{F. Bagnoli\thanksref{a}\thanksref{c}\thanksref{1}},
31: \author{P. Li\`{o}\thanksref{b}\thanksref{2}},
32: \author{L. Sguanci\thanksref{a}\thanksref{c}\thanksref{3}}
33: 
34: \address[a]{Dipartimento di Energetica, Universit\`{a} di Firenze, Via S. Marta 3, 50139 Firenze, Italy.}
35: \address[b]{Computer Laboratory, University of Cambridge, CB3 0FD Cambridge, UK.}
36: \address[c]{INFN, sezione di Firenze and CSDC.}
37: 
38: \thanks[1]{franco.bagnoli@unifi.it}
39: \thanks[2]{pietro.lio@cl.cam.ac.uk}
40: \thanks[3]{luca.sguanci@unifi.it}
41: 
42: \begin{abstract}
43: % Text of abstract
44: The coexistence of different viral strains (quasispecies)
45: within the same host are nowadays observed for a growing number
46: of viruses, most notably HIV, Marburg and Ebola, but the 
47: conditions for the formation and survival of new strains have not
48: yet been understood. We present a model of HIV quasispecies competition,
49: that describes the conditions of viral quasispecies coexistence under
50: different immune system conditions. Our model incorporates both T and
51: B cells responses, and we show that the role of B cells is important
52: and additive to that of T cells. 
53: Simulations of coinfection (simultaneous infection) and
54: superinfection (delayed secondary infection) scenarios
55: in the early stages (days) and in the late stages of the infection
56: (years) are in agreement with emerging molecular biology findings. 
57: The immune response induces a competition among similar phenotypes,
58: leading to differentiation (quasi-speciation), escape dynamics and
59: complex oscillations of viral strain abundance. 
60: We found that the quasispecies dynamics
61: after superinfection or coinfection has time scales of
62: several months and becomes even slower when the immune system 
63: response is weak. Our model represents a general framework to study 
64: the speed and distribution of HIV quasispecies during disease progression, 
65: vaccination and therapy.
66: 
67: \end{abstract}
68: 
69: \begin{keyword}
70: % keywords here, in the form: keyword \sep keyword
71: HIV \sep viral dynamics \sep quasispecies \sep coinfection \sep superinfection
72: % PACS codes here, in the form: \PACS code \sep code
73: \PACS 
74: 87.23.Kg \sep
75: 87.16.-b \sep
76: 87.16.Ac 
77: \end{keyword}
78: \end{frontmatter}
79: 
80: % main text
81: \section{Introduction}
82: \label{sec:Introduction} The human immune system has the goal of
83: providing a basic defense against pathoge\-nic organisms
84: \cite{PC2001,PW1997}. The cells forming the immune system
85: communicate via direct contact and through
86: chemical signals. Using a very basic schematization, the immune
87: system protective action can be divided into an innate, aspecific
88: response, and an adaptive one.  This latter component  involves
89: two major types of cells: lymphocytes, (T and B lymphocyte cells)
90: and antigen presenting cells: monocytes, macrophages, B
91: cells and dendritic cells. Lymphocytes have the ability to react to
92: specific \emph{epitopes}, which  generally are portions of a
93: protein. The idea is to recognize those shapes that constitute a
94: fingerprint of the presence of a foreign agent (antigen), but not
95: to react to \emph{self} repertoire. Their response depends
96: on the T cell receptor (TCR) of the T cell, and immunoglobulins
97: (Ig) of the B cells. It is the great variety of cells
98: carrying TCRs or secerning Igs that allows the
99: recognition of all possible antigens.
100: 
101: Let us present a simplified description of the adaptive response,
102: see Ref.~\cite{PW1997} for a physicists-oriented introduction to
103: immune system modeling. The task of protecting the body implies
104: the detection and removal of free antigens, and the recognition
105: and suppression of infected (or cancer) cells. These
106: \emph{deviating} self cells are identified by a sophisticated
107: automatic mechanism, common to all cells, that presents on their
108: surface specific receptors along with epitopes. 
109: Both viral infection or cancer proliferation implies the
110: activation of \emph{novel} genes, and thus the production of
111: \emph{foreign} proteins that can thus be detected. However, due to
112: the danger (and cost) of an inflammation, the activation of the
113: immune adaptive response is subject to a \emph{double key}
114: protection, implemented by T helper cells.
115: 
116: T cells are divided into CD4+ (helper) and CD8+ (killer) types,
117: according with specific receptors present on their surface. CD4+ T
118: cells provide the regulatory mechanism that triggers the immune
119: response by finding and presenting antigens and releasing
120: stimulatory signals (cytokines). T cells are specific for a
121: given set of epitopes; simplifying, we can say that they are
122: specific for a given antigen. Once activated, CD8+ T cells search
123: for and kill other cells presenting the specific antigen.
124: Their action is triggered by CD4+ T cells recognizing that
125: antigen. This specificity is not absolute: there is a
126: certain variability of receptors allowing cross-recognition of similar
127: epitopes.
128: 
129: The coverage of all possible shapes is implemented by a unique mechanism of genetic  combinatorial sampling.
130: The protection against self recognition is due to the maturation phase.
131: Immature T lymphocytes are
132: generated in the bone marrow and then travel to the thymus where they
133: proceed through a selection process. In the thymus those CD4+ and CD8+ T
134: cells that initiate any reaction against \emph{self} shapes are selected out.
135: 
136: B cells are responsible of the removal of extracellular antigens.
137: Antigen specific CD4+ T cells are required to induce proliferation
138: of antigen specific B cells. The activated B cells then
139: proliferate and differentiate into plasma cells. These produce
140: soluble counterparts to the cell's receptors, called antibodies,
141: which circulate in the blood and fluids. Antibodies bind to
142: specific extracellular antigens and allow them to be removed by
143: other means.
144: 
145: An antigen selects a ``strain'' of lymphocytes which replicates in
146: response (clonal response). During this growth, the specificity of
147: receptors to a given antigen may be increased by an hypermutation
148: mechanism that slightly modifies the genes specifying the receptor
149: shapes. A few cells of this strain (memory cells) will persist for
150: a long time in the body, after viral eradication: should the
151: specific antigen reappear, the response will be faster and
152: stronger (vaccination).
153: 
154: Let's now describe the immune response in the framework of a HIV
155: infection. HIV infects cells that contain the CD4 receptor on
156: their cell membrane, through which the virus attaches and enters the cell. 
157: The CD4 molecules are found on T lymphocyte helper cells, 
158: monocytes and macrophages.
159: 
160: As HIV uses the CD4 molecule to establish the initial binding to a
161: cell, these cells are the main target for HIV infection. Since
162: CD4+ T cells are vital to the establishment of an immune response,
163: their progressive loss opens the way for AIDS-related
164: opportunistic infections~\cite{PF1993}.
165: 
166: The worldwide presence of several strains of the HIV virus and
167: their often simultaneous presence within a patient, due to the
168: increased frequency of multiple infections, are the new remarkable
169: features of HIV pandemia. For example, HIV-1, presently, exists as
170: three groups classified according to their degree of similarity.
171: Group M stands for main, Group O for outlier and Group N for
172: non-M, non-O. Group M is the predominant group and is divided into
173: 9 strains (subtypes or strains) that differ in the env gene~\cite{KM2004,YL2005}. 
174: Specific strains are predominant in different
175: geographical regions and their high mutation rates often allow
176: researchers to track the network of contacts. A growing number of
177: strains within each subtype is nowadays reported.
178: 
179: The origin of different HIV strains is related to the quasispecies
180: characteristics of HIV evolution~\cite{ES1989}. Quasispecies are the
181: combined result of mutations and recombination,
182: originating variability, and selection, that filters out this variability. 
183: The main source of mutations is the reverse transcription
184: phase~\cite{PC2001}, in which the viral RNA is converted into DNA by viral
185: proteins that are not very accurate. Other sources of variability
186: are the errors during the transcription phase, deletions and other
187: genetic rearrangements during the integration of viral DNA  into
188: the host chromosomes, including recombination with other viral DNA
189: or proviruses present in the cell. This mechanism thus increases
190: the viral diversity in the body. However, not all these variants
191: have the same capability of completing their life cycle and
192: multiplying their clone number (i.e.\ their fitness).
193: 
194: First of all, the infection capacity of mutants may vary, and also
195: their speed of replication~\cite{NP1998}. In some cases a single mutation is
196: sufficient to alter significatively the fitness~\cite{KH1997}, or
197: this may due to the accumulation of several mutations (smooth or
198: \emph{Fujiama} landscape~\cite{FP1997}). Since the number of
199: targets (the substrate) is limited, the strains that originate
200: more clones in the unit of time tend to eliminate less fitness
201: mutants, which are subsequently regenerated by the mutation
202: mechanism~\cite{ES1977}. While mutations are an essential
203: ingredient for exploring the genetic space in the search for the
204: fitness maximum, they also lowers the average fitness of the
205: strain, that generally is formed by a cloud of mutants around the
206: fitness maximum (quasispecies). For a given fitness landscape,
207: there is a maximum tolerable mutation rate above which the
208: quasispecies structure is lost (error catastrophe~\cite{ES1989}).
209: 
210: The HIV quasispecies seem able to compete and their persistence
211: affect AIDS progression ~\cite{CD2000,BL2005}. For
212: instance, some mutants may even be unable to colonize new cells,
213: but they may survive by parasitizing more invasive clones (this is
214: particularly observed when recombination of several strains is
215: present)~\cite{JD2000}. 
216: Finally, the immune response is specific, with a
217: certain tolerance, to a given antigen, and may vary from antigen
218: to antigen, due, for instance, to their similarities to self
219: shapes. Thus, the presence of an antigen originates an immune
220: response that suppresses also variants of the stimulating clone.
221: This results in a form of competition among strains or clones,
222: that prefer to stay \emph{away} in the genetic space in order to
223: escape the immune response and/or lower their intensity by dividing
224: the productive capacity of the immune system into several strains
225: (multi-infection). The effect of this competition may result in a
226: continuously varying dominant quasispecies that tries to escape
227: the immune response, or the splitting of the quasispecies into
228: multiple strains (quasi-speciation).  During the first phases of
229: HIV infection, when the immune response is vigorous, the first
230: phenomenon is observed until presumably the virus finds a
231: \emph{niche} in the genetic space, not sufficiently covered by the
232: immune system, originating AIDS~\cite{DSP1992,HM1995}. The
233: resistance to antiviral treatment may be ascribed to a similar
234: mechanism. When the immune system is depressed, also slightly less
235: adapted strains may survive, giving origin to multi-clonal
236: infections often observed in terminal patients~\cite{BM2002}.
237: 
238: Secondary HIV infections with closely related strains may induce
239: the response and clonal selection of a new subsets of CD4+ T cells
240: that recognize specifically the new HIV strain, even under
241: conditions of chronic infection by other strains. We can
242: distinguish two scenarios: strains coinfection (simultaneous
243: infection) and superinfection (delayed secondary infection)
244: scenarios (see for instance Refs.~\cite{R2002,L2003,HH2004,AW2002,GH2005}).
245: 
246: Our aim is to model viral strain coevolution and understand the
247: importance of factors such as mutation rates, strength of immune
248: response, cross-talks between T and B cells and competition among
249: strains in coinfection and superinfection.
250: 
251: The fate of an epidemic does not depend only on the evolution
252: inside a single host, but also on the transmissibility to other
253: hosts. The social factors involved in emergence of viral strains
254: are diverse and include global transportation, urban crowding and
255: poverty, changing behavioral patterns, human population growth,
256: etc. \cite{ZH1996}. We shall not deal with these aspects in the
257: present paper.
258: 
259: 
260: \begin{figure}[t]
261:   \centering
262:   \includegraphics[width=0.6\textwidth]{1.eps}
263:   \caption{Representation of the interactions between cells and/or cells and
264: viruses. HIV strains 
265: infect CD4+ T cells that become infected and produce new viruses.
266: At the same time HIV peptides are presented via APC cells to
267: T-Helper cells that become activated. Activated CD4+ T cells
268: trigger B and CD8+ cells reactions: the first release antibodies
269: that bind to the antigen while the latter directly remove infected
270: CD4+ T cells. '+' and '-' signs indicate cell/virus production or
271: removal.}
272:   \label{fig1}
273: \end{figure}
274: 
275: The article is organized in the following way: in the next
276: sections we introduce a model that combines T and B cells immune
277: response to HIV infection. Then we extend the model to include
278: quasispecies and we test the model in the scenarios of
279: coinfection and superinfection using parameters derived from
280: biological literature. Finally we discuss results and work in
281: progress.
282: 
283: In what follows, we shall use the terms strain and quasispecies,
284: according with the \emph{scale} at which one is examining the system.
285: In order to resolve the quasispecies details, one needs to use a
286: (genotypic or phenotypic) scale sufficiently fine to allow
287: differentiation inside a single quasispecies. On the
288: contrary, if one is interested in the \emph{ecological} relations
289: among quasispecies, it is more efficient to use a scale in which a
290: whole quasispecies is simply collapsed in a single value. In this
291: latter case we use the term strain to refer to the whole quasispecies.
292: 
293: \section{Models}
294: 
295: Mathematical models have proven valuable in understanding the
296: mechanisms of many of the observed features of HIV dynamics,
297: such as the positive and
298: negative regulation of T and B cell selection, the dynamics of
299: production of the TCR repertoire in the thymus, the progression to
300: and the latent phase of
301: AIDS~\cite{HM1995,CP2004,CS1996,DP1995,WN2002,PH1996,WP2004,WS1995}.
302: They have also been useful in forecasting the effects of multi drug therapy.
303: 
304: While the majority of HIV dynamics
305: models assume that each CD4+ T cell is infected by a single HIV
306: strain, Dixit and Perelson~\cite{DP2005} have shown that with the
307: progression of HIV infection multiple infections become very
308: likely.
309: 
310: Our goal was to investigate the conditions of the
311: multi-strains persistence within a single patient
312: and the effects of coinfection and superinfection. To set the scenarios of our
313: model, it is noteworthy that, in general, the immune response can
314: be seen as a prey/predator dynamics, in that the immune system is
315: the predator of the virus (the prey). The HIV dynamics is more
316: complex because the HIV is both the prey and the predator, i.e. it
317: attacks CD4+ T cells and is attacked by cells coordinated by CD4+
318: T cells.
319: 
320: Figure~\ref{fig1} illustrates the relationships among the
321: different quantities of our model. This schematization is valid
322: for the quasispecies case and for the
323: undifferentiated one. In this latter case, the
324: indication of strains (in parenthesis) should be
325: neglected. The
326: clonal amplification of naive T cells (of population $i$) depends on
327: the ability of ($i^{\text{th}}$ class) T cell to recognize all the infected
328: T cells (carrying an epitope from the $k^{\text{th}}$ class of viruses).
329: The rate of infection of naive T cells (of class $i$) depends on all
330: the viruses (containing the epitope $k$). In the same way, the
331: clonal expansion of infected T cells (of class $k$, meaning that
332: they have been infected by a virus of class $k$ irrespective of the
333: original T class) depends on the
334: interaction of these viruses with all the T cells (of any class $i$).
335: 
336: 
337: 
338: \subsection{Combining B and T cells responses}
339: 
340: \begin{figure}[t]
341:   \centering
342:   \includegraphics[width=0.6\textwidth]{2a.eps}
343:   \includegraphics[width=0.6\textwidth]{2b.eps}
344:   \caption{Typical time evolution of the single-species model; 
345:   diamonds represent
346: uninfected T cells, squares represent viruses and plus signs the
347: infected T cells. Plot (a) illustrates a
348:   scenario leading to a chronic infection
349:   ($\gamma^{(T)}=10^{-6}$, $\gamma^{(I)}=10^{-5}$, $\gamma^{(V)}=0$).
350:  In the presence of a further B cell response (plot (b)), the
351: infection is defeated by the immune system
352: ($\gamma^{(T)}=10^{-6}$, $\gamma^{(I)}=10^{-5}$ and
353: $\gamma^{(V)}=5\cdot10^{-6}$).}
354:   \label{fig2}
355: \end{figure}
356: 
357: 
358: 
359: Let us start from a widely accepted theoretical model of HIV
360: progression and immune response proposed by Perelson and
361: colleagues in 1995~\cite{HM1995}, recently extended to include CTL
362: response and stochastic components~\cite{CP2004}. There are
363: several works that discuss the role of B cells in the immune
364: response~\cite{DSP1992}. Since the progression to AIDS has been
365: found to correlate well with CD4+ T cells decrease, B cells are
366: thought to play a minor role in the immune response to HIV. Note
367: that B cells can act only as predator to the HIV, so their
368: coupling with HIV dynamics is different from that of T cells.
369: Our aim is to present a more general  model framework of
370: both T and B immune responses to HIV. We have first considered the
371: following system of differential equations describing the dynamics
372: of a single viral strain:
373: \begin{align}
374: \dot{T}  &= \left(  \lambda+\gamma^{(T)}I\, T\right)
375:   \left(  1-T/K\right)-\left(  \delta_{T}+\beta\, V\right)  T, \label{T}\\
376: \dot{I}  &= \beta\, V\,T-\left(  \delta_{I}+\gamma^{(I)}\,T\right)
377: I,\label{I}\\
378: \dot{V}  &= \pi\, I-\left(  c+\gamma^{(V)}\,T\right)  V.\label{V}
379: \end{align}
380: All quantities indicated by greek letters and $c$ have unit
381: $\text{days}^{-1}$.
382: 
383: This model considers the T-helper (CD4+) cells ($T$), and HIV
384: virus particles ($V$); the T cells can become infected ($I$). With
385: respect to Refs.~\cite{HM1995,CP2004}, in Eq.~\eqref{T}, we
386: describe how the number of naive T cells ($T$) which have passed
387: the thymus selection, depends on rate of formation in the bone
388: marrow ($\lambda$), and on clonal amplification upon stimulation
389: by infected cells, $I$ (term $I\, T$). They decrease with a rate
390: that is the sum of a natural clearance, $\left( \delta_{T}T\right)$,
391:  due to cell aging, and cell destruction upon virus infection,
392: $\left( VT\right)  $. The density of T cells is limited by a
393: saturating density/lymphonode capacity factor, $K$. Following
394: Ref.~\cite{WP2004} we have set $K=10^{12}$.
395: 
396: The second equation
397: describes the rate of infection, described by $\beta$, of naive T
398: cells upon the interaction with the virus (term $V\,T$). Infected T cells
399: are cleared out at a fixed rate, $\delta_{I}$, and due to the
400: action of natural killer cells, CD8+ (term $TI$).
401: 
402: The third equation
403: describes the budding of viruses from infected cells, $\pi$. Virus
404: particles are cleared out at rate $c$ (defective viruses) and
405: after immunoglobulin binding and subsequent engulfments by the
406: macrophages
407: (term $TV$). 
408: 
409: The $\gamma$ parameters have the same meaning of the
410: constant of association in chemistry, or can be thought as a
411: combination of both the probability of interaction and the
412: interaction strengths between cells ($\gamma^{(T)}$,$\gamma^{(I)}$) or
413: between cells and viruses ($\gamma^{(V)}$). Note that in the limit
414: $\gamma \rightarrow 0$ and $K\rightarrow\infty$, 
415: we recover the pattern of the standard
416: model~\cite{PH1996}.
417: 
418: The B cell response is  modeled using the
419: parameters corresponding to the T cells which activate them by
420: receptor recognition. Here, we have assumed the immunoglobulin
421: concentration, which represent the B cell response, are linearly
422: correlated to the concentration of activated B cells. These  in
423: turn  are supposed linearly correlated to the concentration of the
424: CD4+ T cells. Exploratory analysis with different
425: parameters, suggested us that Eq.~\eqref{V} allows to keep a
426: minimum number of parameters without loss of important details.
427: 
428: 
429: Figure~\ref{fig2} shows some typical behavior of this single-species
430: model, showing the case of asymptotic coexistence of immune system and
431: virus (chronical infection) and virus eradication. This latter
432: scenario may be triggered both by increasing the T killer
433: ($\gamma^{(I)}$) or B ($\gamma^{(V)}$) response. In no cases it is
434: possible to observe the defeat of the immune system, since a decrease
435: of the quantity of T cells, affects also the rate of viral production.
436: This reflects the observed fact that progression to AIDS and death is
437: caused by opportunistic infections and not directly by the HIV virus.
438: Model parameters are derived from medical
439: literature (see also Ref.~\cite{CP2004}).
440: 
441: 
442: \subsection{Modeling quasispecies competition and persistence}
443: 
444: Our actual model is obtained by extending Eqs~(\ref{T}-\ref{V}),
445: to include the effects of quasispecies persistence.
446: 
447: For the sake of simplicity, we
448: assume that each viral strain is characterized by just one epitope
449: $i$. We make use of coupled differential equations, one for each
450: viral quasispecies and T cell. Although this mean-field approach
451: disregards the effect of fluctuations and genetic drift in
452: quasispecies abundances, it is useful for understanding the
453: coarse-grain features of the behavior of the interplay between HIV and
454: the immune system.
455: 
456: The quasispecies model is described by the following set of equations:
457: \begin{align}
458: \dot{T}_{i}  &= \left(  \lambda_{i}+\sum_{k}\gamma_{ik}^{(T)}I_{k}T_{i}\right)
459:   \left(  1-\frac{1}{K}\sum_{i}T_{i}\right)  -
460:   \left(  \delta_{T}+\sum_{k}\beta_{k}V_{k}\right)T_{i}  ,\label{Ti}\\
461: \dot{I}_{k}  &= \left(  \sum_{k^{\prime}}\mu_{kk^{\prime}}
462: \beta_{k^{\prime}}V_{k^{\prime}}\right)\left(\sum_{i}T_{i}\right)
463:  -\left(  \delta_{I}+\sum_{i}\gamma_{ki}^{(I)}T_{i}\right)I_{k}
464:  ,\label{Ik}\\
465: \dot{V}_{k}  &= \pi
466: I_{k}-\left(c+\sum_{i}\gamma_{ki}^{(V)}T_{i}\right)V_{k}.\label{Vk}
467: \end{align}
468: 
469: The model considers the following cell types: T-helper (CD4+)
470: cells responding to virus strain $i$, ($T_{i}$); T cells (any strain)
471: infected by virus strain $k$, ($I_{k}$); abundance of viral strain
472: $k$, ($V_{k}$).
473: This means that viral strain
474: $k$ are identified by just one epitope, which is then displayed
475: on the surface of the
476: T cell of class $k$,  and that a T cell of class $i$ can be activated at
477: least by one CD4+ T cell carrying the epitope $k$, which is
478: specific of the viral strain $k$.
479: The indices $i$ ($k$) range from 1 to $N_i$ ($N_k$),
480: and in the following we have used $N_i=N_k=N$.
481: 
482: 
483: Equation~\eqref{Ti} describes the generation of T cells through two
484: mechanisms: the bone-marrow source (and selection in the thymus)
485: and the duplication of T cell strains activated upon the
486: recognition with an antigen carrying cell that may be even an
487: infected one. In order to take into account the limits of the
488: immune response, we have investigated both a logistic term and
489: different types of saturating and density-dependent functions~\cite{CP2004}.
490: The logistic term models the
491: global carrying capacity of immune system. In other
492: works~\cite{CP2004}, a saturating function or density dependent
493: function was used, aiming to
494: describe how the growth and development of the immune response is
495: limited by the timing of T cell division among other constraints
496: such as cell-cell recognition, signals diffusion, cell migration.
497: 
498: The death-rate term represents both a natural
499: death rate proportional to the population, and the infection rate of T
500: cells by any viral strain. The infection probability, reflected by
501: the term $\sum_{k}\beta_{k}V_{k}T_{i}$ and by the sum over $T_{i}$ in the
502: $I$ cell birth rate, is the same irrespective of the T class. We
503: assume that an infected T cell does no more contribute to the immune
504: response.
505: 
506: Equation~\eqref{Ik} describes the infection
507: dynamics. The incorporation of two death rate parameters reflects
508: the fact that the infected cells disappear due to cellular
509: death and after the action of natural killer cells (CD8+). There
510: are clear experimental evidences that CD4+ cells decrease during
511: the late HIV infection stages and in the AIDS state. In terms of
512: our models this mean that the parameter $\delta_{I}$
513: depends on the stage of the
514: infection. We have preferred to model the early stage of
515: infection, far from AIDS conditions and therefore we have set
516: $\delta_{I}$ to a constant, medical literature referred, value.
517: 
518: 
519: The term $\mu_{kk^{\prime}}$ represents a mutation process that
520: affect the phenotype. Mutations occur mainly during the reverse
521: transcriptase processing of the viral genomes. The mutation parameter 
522: is essential for the formation
523: of new quasispecies. Although RNA viruses have been reported to
524: have substitution rates of the order of $1 \cdot 10^{-3}$
525: substitution per site per replication~\cite{JH2002}.
526: Mansky and Temin~\cite{MT1995}
527: determined an in-vivo mutation rate of $3.5 \cdot 10^{-5}$.
528: The majority of mutations does not
529: change the amino acids. Moreover, most of amino acid substitutions are neutral
530: or quasi-neutral, since they do not change remarkably the protein
531: structure, and therefore the fitness of the species; effective
532: differences in fitness and behavior evolve through several amino
533: acid changes (see for instance \cite{LG2004}). Nevertheless, one
534: single event of recombination can often alter the fitness in a
535: substantial way. We take into consideration only those non-synonymous
536: mutations that alter the phenotype (protein structure),
537: and therefore we used a slight smaller
538: value for the mutation rate of the order of $10^{-5}$.
539: 
540: We have implemented the matrix $\mu_{kk^{\prime}}$ as a discrete
541: Laplacian (diffusion in phenotypic space) controlled by parameter
542: $\mu$ (without indices). 
543: 
544: 
545: 
546: Equation~\eqref{Vk} describes the virus replicative dynamics in terms of a
547: birth rate proportional to the virus "budding" numerosity.
548: The viral death rate parameters depend on the  rate of natural death and on the
549: recognition of virus by B cells.
550: 
551: As in the single-specie model, B-cells and T-killer cells are only
552: implicitly included in the model in order to reduce the
553: dimensionality without loosing too many details. We assume that
554: these responses are fast enough to be at equilibrium and they are
555: just proportional to the abundance of (cognate) $T$ helper cells.
556: 
557: While in the single-strain model, Eqs.~(\ref{T}-\ref{V}), 
558: the three $\gamma$ parameters are scalar values,
559: here $\gamma_{ik}^{(T)}$, $\gamma_{ki}^{(I)}$ and
560: $\gamma_{ki}^{(V)}$ are matrices describing the interactions
561: between viral strains and immune cells, i.e.\ who will interact
562: with whom. The $\gamma$ parameters represent the interaction
563: between cells and/or cells and viruses in terms of geometry and
564: strength of the interaction. It expresses in a single matrix all
565: the information about diffusion, epitope recognition and show some
566: analogies with a constant-in-time fitness, as described in Ref.~\cite{BB1997}.
567: 
568: When the $\gamma$ matrices are diagonal, each T cell
569: interacts only with one viral strain. The non-zero
570: elements of the $i^{\text{th}}$ row in the  $\gamma$ matrices
571: represent the strains of the virus
572: recognized by the immune cells of class $i$ while the value represents
573: the affinity maturation, i.e.\ the accuracy in recognition.
574:  
575: \begin{figure}[t]
576: \centering
577: \includegraphics[width=0.5\textwidth]{gamma.eps} 
578: \caption{Control parameters and shape of the $\gamma$-matrices}
579: \label{gamma}
580: \end{figure}
581:  
582: In order to control the shape of the $\gamma$ matrices with few
583: parameters, we choose the following functional form (see
584: Fig.~\ref{gamma})
585: \[
586:   \gamma_{ij} = 
587: A\left\{\frac{1}{2}\left[1+\cos\left(\frac{2\pi}{N}(i-j)\right)\right]\right\}^{\varepsilon_A}
588: +
589: B\left\{\frac{1}{2}\left[1+\cos\left(\frac{\pi}{N}(i+j)\right)\right]\right\}^{\varepsilon_B}.
590: \]
591: The parameters $A$ and $\varepsilon_A$ ($B$ and $\varepsilon_B$) 
592: control the shape of the
593: $\gamma$ matrices in the direction transverse (parallel) to the main
594: diagonal. 
595: The exponents $\varepsilon_A$ and $\varepsilon_B$ controls the smoothness of the
596: variation along the corresponding direction.  For 
597: $\varepsilon$ small the variation of the function
598: is very small, while for $\varepsilon\rightarrow \infty$  the
599: corresponding function is $\delta$-shape. So, a diagonal $\gamma$
600: matrix is obtained by setting $B=0$ and $\varepsilon_A$ large. 
601: For instance, differences in the fitness  of phenotypes may be
602: obtained by setting $B$ and $\varepsilon_B$ different from zero in
603: some of the $\gamma$-matrices, while the intra-species competition is
604: triggered by setting $A$ different from zero and $\varepsilon_A$ small.
605:  
606: 
607: It is worth noting that nowadays, 2-photons experiments allow to
608: measures recognition strength, movements and speed of the immune
609: cells~\cite{MC2004} that, in theory, can be used to derive estimates of the
610: $\gamma$ matrices. We think that this data is still in embryonal phase,
611: so we focus on the qualitatively features of the response.
612: 
613: 
614: Given the absence of a precise relationship
615: between genotype and phenotype changes, we have simply considered
616: the grouping together all quasispecies with the same fitness. Thus
617: the quasispecies constitute a linear space ordered in term of
618: phenotype similarity.
619: 
620: We have used the same values of model parameters by re-scaling
621: $T,I,V$, $\lambda_{i}$ and $K$ by the number $N$ of strains.
622: 
623: 
624: \section{Numerical results}
625: 
626: \begin{figure}[t]
627:   \centering
628:   \includegraphics[width=0.4\textwidth]{3a.eps}
629:   \includegraphics[width=0.4\textwidth]{3a-lt.eps}
630:   \caption{Typical time evolution of the T cell abundance
631:     in the multi-species model for short-term (a) and long-term
632:     behavior (b). We set $\mu=10^{-5}$,  
633:     $A = 5\cdot 10^{-6}$, $\varepsilon_A=10$, $B^{(T)} = 3\cdot
634:     10^{-6}$, $\varepsilon_B^{(T)}=10$ with $N=5$.
635:     In the inset, the y-axis on the left reports the asymptotic abundance of
636:     virus strains, the y-axis on the right shows the interaction
637:     strength (dashed line) between T cells and virus phenotypes (x-axis).}
638: \label{fig3}
639: \end{figure}
640: 
641: 
642: \begin{figure}[t]
643:   \centering
644:    \includegraphics[width=0.4\textwidth]{3b.eps} 
645:     \includegraphics[width=0.4\textwidth]{3c.eps} \\
646:     \includegraphics[width=0.4\textwidth]{3d.eps} \\
647:   \caption{ Phase space analysis of $A$s parameters of the $\gamma$
648:   matrices. 
649:   In plot (a) we fix $A^{(V)}$=$10^{-6}$ and we show
650:   $A^{(I)}$ vs. $A^{(T)}$. 
651:   In plot (b) we fix $A^{(I)}$=$10^{-6}$ and we show
652:   $A^{(V)}$ vs. $A^{(T)}$.
653:   In plot (c) we fix $A^{(T)}$=$10^{-6}$ and we show
654:   $A^{(V)}$ vs. $A^{(I)}$.      
655:   The other parameters are: number of quasi species $N=25$,
656:   $\varepsilon_A=10^3$, $B=0$ and $\mu=0$. Region I corresponds to the coexistence of immune system
657: and viruses (chronic infection) while region II corresponds to the
658: defeat of viral infection.}
659:   \label{fig3a}
660: \end{figure}
661: 
662: 
663: The evolution of T cells abundances in a scenario of quasispecies 
664: is shown in ~\ref{fig3}. Note that when the asymptotic state of our model is given by
665: a fixed point, the asymptotic distribution is insensitive of the
666: initial conditions, and the strains corresponding to higher fitness
667: are more abundant. However, one should consider that this asymptotic
668: state may be reached after such a long time that it may be outside any
669: practical scenario of the progression of a disease. The role of
670: mutations in the transitory regime is quite particular. First of all,
671: mutations are necessary to populate strains outside the first
672: inoculum, see also Fig.~\ref{fig4}. Mutations affect also the average
673: fitness and the width of the quasispecies and the Eigen's error threshold
674: ~\cite{BE2005} is a sort of extremal consequence of this effect. However, in
675: the presence of coupling among strains, due to competition or to a
676: global constraints (the $K$ parameter), the specific form of 
677: mutations does not play a fundamental role, see also
678: Ref.~\cite{BB1997}. 
679: 
680: It's noteworthy that a naive analogy between evolutionary systems and statistical
681: mechanics can be obtained by identifying the selection as the
682: (opposite) of energy and mutation as a sort of temperature
683: (\cite{T1992,L1987}). 
684: 
685: 
686: \subsection{Phase diagrams of $\gamma$ matrices}
687: 
688: We focused our attention on the parameter $A$ of the three $\gamma$ matrices 
689: which represent the major features of the interaction, Fig.~\ref{fig3a}.
690: The regions I and II in the three diagrams represent chronic infection
691: (coexistence of immune system and viral infection) and complete
692: recovery (virus defeat) from the disease, respectively.
693: For the sake of clarity instead of 3D diagrams, we choose
694: 2D projections, after setting the other dimension to a biologically
695: meaningful value.
696: 
697: The effect of B cells is shown in Fig.~\ref{fig3a}a : for a very low 
698: contribution of B cells to the immune response, there is a clear threshold 
699: effect on both $\gamma^{(T)}$ and $\gamma^{(I)}$. By considering a weak action of T killer cells,  the
700: threshold effect for $\gamma^{(T)}$ and $\gamma^{(V)}$ (B cells
701: effect) is somewhat decreased but still present, Fig.~\ref{fig3a}b.
702: Instead, Fig.~\ref{fig3a}c shows that the combined effects of B and T cells 
703: are almost additive, despite the parameter $K$ that limits the total 
704: population of T cells. For example, let's look at diagram (c). 
705: If we draw a line at
706: $A^{(I)}=2.56 \cdot 10^{-4}$, that is an extreme value for coexistence
707: in diagram (a), we can notice that both the two phases are present,
708: in particular there are values of $A^{(V)}$ leading to the defeat of
709: the infection.
710: 
711: \subsection{Coinfection dynamics}
712: 
713: \begin{figure}[t]
714:   \centering
715:   \includegraphics[width=7cm]{4a.eps} \includegraphics[width=7cm]{4c.eps} \\
716:   \includegraphics[width=7cm]{4b.eps} \includegraphics[width=7cm]{4d.eps} \\
717:     \caption{Time evolution of T (plots (a) and (c)) and V
718:     (plots (b) and (d)) populations after an inoculum with 
719:     two strains (phenotypes 0 and 2), $\mu=0$,  
720:     $A = 5\cdot 10^{-6}$, $\varepsilon_A=10$, $N=5$. Plots (a) and (b):  $B=0$;
721:     Plots (c) and (d): $B^{(T)} = 3\cdot 10^{-6}$,
722:     $\varepsilon_B^{(T)}=10$. In the inset the asymptotic distribution
723:     of T and V populations and the shape of the diagonal of $\gamma^{(T)}$ are shown. 
724:     }
725:   \label{fig4}
726: \end{figure}
727: 
728: The effect of competitive evolution observed after a single
729: inoculum of two different HIV strains at time $t=0$ is shown in
730: Fig.~\ref{fig4}.    
731: We have considered two different scenarios. The first one
732: (Fig.~\ref{fig4}a-b) represents the time evolution of T cells
733: and viruses when the epitopes of two different strains are subjected
734: to the same interaction strength.
735: The second scenario (Fig.~\ref{fig4}c-d) considers differences in the
736: recognition ability of viral antigens by T cells.
737: For example, variants of the
738: CCR5 receptors may induce partial resistance against HIV~\cite{GM2003}. 
739: De Boer and Perelson have shown~\cite{DP1995} that the phase space of the antigen-T cell
740: recognition is not homogeneous but it is patched with areas of
741: strong immune response and areas of lack of immune response. 
742: This difference in the strains targeted by the T cells generates changes in the 
743: fitness of the virus that turn into differences in their abundances. The immune system response is
744: almost the same in the two cases, and very similar to that obtained
745: without considering the phenotypic space, Fig.~\ref{fig2}. The asymptotic distribution of viruses reflects the
746: behavior of the fitness is induced by the stimulation
747: of the immune response, $\gamma_{ii}^{(T)}$, as in Fig.~\ref{fig3}. 
748: 
749: 
750: 
751: \subsection{Superinfection dynamics}
752: 
753: \begin{figure}[t]
754:   \centering
755:   \includegraphics[width=0.6\textwidth]{5a.eps} \\
756:   \includegraphics[width=0.6\textwidth]{5b.eps} \\
757:     \includegraphics[width=0.6\textwidth]{5c.eps} \\
758:     \caption{Short time behavior of uninfected T cells (a), 
759:       infected T cells (b) and
760:       viruses (c) under superinfection. Parameters as in 
761: Fig.~\ref{fig4}a-b.}
762:   \label{fig5}
763: \end{figure}
764: 
765: \begin{figure}[t]
766:   \centering
767:   \includegraphics[width=0.6\textwidth]{6a.eps} 
768:   \includegraphics[width=0.6\textwidth]{6b.eps} \\
769:     \caption{Long time behavior of virus abundances after superinfection
770:     at time $t=20$.  
771:     (a) healthy immune system ($\lambda=10^5$); (b)
772:     weak immune system ($\lambda=10^4$). 
773:     The other parameters used have  the same values 
774:     of  Fig.~\ref{fig4}a-b.
775:     Diamonds represent the first
776:     inoculum and squares the second inoculum.}
777:   \label{fig6}
778: \end{figure}
779: 
780: Results of short time and long term viral coevolution after superinfection are 
781: reported Figs.~\ref{fig5} and~\ref{fig6}, respectively. In both figures the first 
782: viral inoculum occurs at time $t=0 $ and the second infection at time $t=20$, 
783: when the immune response to the first inoculum has completed and the virus has 
784: established a chronic infection. After the second inoculum and a short transient, a
785: slow mounting of the second viral infection occurs. This low dynamics
786: continues on a scale of several months (Figure~\ref{fig6}) and 
787: eventually reaches the same level of the other quasispecies. This 
788: behavior represents another example of a slow relaxation to a fixed-point 
789: equilibrium. With the progression of the disease, when the immune system
790: is compromised (low number of T cells \emph{i.e.} low $\lambda$), 
791: a second inoculum strain requires a very long time to reach the same abundances 
792: of the first strain (Fig.~\ref{fig6}b).
793:  
794: Experimental evidences~\cite{JD2000} show that
795: the probability of observing new fitter recombinant strains 
796: increases with the number of already coevolving strains. This effect
797: has not been yet incorporated into our model.
798: 
799: 
800: \subsection{Coevolution and speciation}
801: 
802: \begin{figure}[t]
803:   \centering
804:   \includegraphics[width=6.2cm]{7a.eps} \includegraphics[width=7cm]{7b.eps} \\
805:   \includegraphics[width=7cm]{7c.eps} \includegraphics[width=7cm]{7d.eps} \\
806:     \caption{Speciation of virus quasispecies and uninfected T cells dynamics
807: after competitive superinfection at four different times: $t=0$
808: (a), $t=4.5$ (b), $t=5.25$ (c) and $t=5.75$ (d).
809: Virus strain 15 is present at time $t=0$, while strain 5 is
810: inoculated at time $t=1$. Mutation rate $\mu=10^{-4}$ and
811: non-uniform interaction strength as in Fig.~\ref{fig4}c-d. The dashed
812: line represents the abundances of T cells targeting
813: each viral phenotype.}
814:   \label{fig7}
815: \end{figure}
816: 
817: \begin{figure}[t]
818:   \centering
819:   \includegraphics[width=0.6\textwidth]{spec-lt.eps}
820:     \caption{Temporal evolution of speciation of virus quasispecies and uninfected T cells dynamics
821:     after superinfection; diamonds represent uninfected T cells,
822:     squares represent viruses. 
823:     Virus strain with phenotype~15 is present at time $t=0$ and the strain
824:     with phenotype~5 is inoculated at time $t=1$. 
825:     We set $A^{(T)} = 5\cdot 10^{-5}$, $A^{(I,V)} = 5\cdot 10^{-6}$, $\varepsilon_A=10$, $B^{(T)} = 3\cdot
826:     10^{-6}$, $\varepsilon_B^{(T)}=10$, with mutation rate $\mu=10^{-4}$.}
827:   \label{fig8}
828: \end{figure}
829: 
830: Finally, we have studied virus quasispecies formation in
831: more details. Figure~\ref{fig7}a shows the initial  
832: inoculum at phenotype 15 in a space of 25, followed by a delayed
833: inoculum at phenotype~5 at time $t=1$. The
834: non-uniform interaction strength would favor the central phenotypes.
835: However, since in this simulation the immune system does not discriminate
836: among similar phenotypes (modeled by $\varepsilon_A^{(T)}=10$),
837: there is an induced competition among neighboring strains. This
838: competition induces a separation
839: of the original quasi-species into two clusters (quasi-speciation), 
840: Fig.~\ref{fig7}b. However, as shown by Figs.~\ref{fig7}c-d
841: the immune system response continues to change in time, resulting into
842: a complex  coevolution with viral populations. 
843: Fig.~\ref{fig8} shows
844: the typical irregular evolution of a speciation dynamics.
845: 
846: 
847: \section{Discussion}
848: We presented a model of the within-patience persistence of HIV quasispecies. 
849: We have first extended the Perelson's standard model~\cite{PH1996} to
850: incorporate B cells response.
851: The B cells, once activated by specific T cells, can only
852: act as a predator, and not being directly targeted as a prey by
853: HIV. We found that this role represents a non negligible
854: contribution to the immune response in all cases where a virus or
855: bacterium is targeting (and being targeted by) T cells.
856: Interestingly, T cells have a role in both innate and adaptive
857: immune responses while B cells only in the adaptive system. It is
858: known that the innate immune response is present also in insects
859: while the adaptive is more recent, being present only in
860: vertebrate. Thus, B cells may have appeared also to fight back
861: viruses targeting specifically at T cells.
862: 
863: Recent works have shown that HIV quasispecies may compete
864: \cite{CD2000} and that persistence of the initial or ancestor
865: quasispecies is a good indicator for disease progression
866: \cite{BL2005}. 
867: 
868: Our model shows that the time evolution of the
869: competition between quasispecies is slow and has time scales of
870: several months. This provides a hint of why standard viral dynamics models,
871: which ignore multiple infections, are effective in describing
872: viral load evolution in HIV-infected individuals. 
873: 
874: Burch and Chao~\cite{BC2000} have stressed that the evolution
875: of an RNA virus is determined by its mutational neighborhood.
876: As the phenotype divergence among viral strains arises from differences 
877: in selection pressure, these differences may
878: lead, for instance, to a higher infection rate.
879: Since the competition is through the immune system response and given
880: that the phase space of antigen recognition is not homogeneously
881: covered~\cite{DSP1992}, the HIV high mutation rate allows the
882: quasispecies to find regions with weak immune response.
883: This competition may lead to speciation of viral
884: strains.
885: 
886: The introduction or modulation of a quasispecies may be used in therapy against an
887: already present aggressive strain. This would be particularly
888: effective during full AIDS stage when virus burden is particularly
889: high and more stringent conditions for competition.
890: It is noteworthy that Schnell and colleagues~\cite{SR1997} have
891: constructed a recombinant vesicular stomatitis virus that although
892: unable to infect normal cells, infected and
893: killed cells that were first infected with HIV causing a rapid
894: cytopathic infection. The authors showed that the introduction of
895: this engineered virus can achieve HIV load reduction of 92\% and
896: recovery of host cells to 17\% of their normal levels (see also
897: the mathematical model in Ref.~\cite{RG2003}).
898: 
899: Our model represents a general framework to investigate several
900: aspects of the evolution of HIV infections, for
901: example intermittency or switching dominance of strains and the
902: arising of new dominant strains during different phases of
903: therapy; how superinfection will evolve in case of replacement of
904: drug-resistant virus with a drug-sensitive virus and acquisition
905: of highly divergent viruses of different strains. It is also useful 
906: to investigate
907: whether antiviral treatment may increase susceptibility to
908: superinfection by decreasing antigen load. Different drug
909: treatments can alter the population of quasispecies. Quasispecies
910: may be the key for both understanding the emerging infectious
911: diseases and has implications for transmission, public health
912: counselling, treatment and vaccine development.
913: 
914: 
915: % The Appendices part is started with the command \appendix;
916: % appendix sections are then done as normal sections
917: % \appendix
918: 
919: % \section{}
920: % \label{}
921: \newpage
922: 
923: \begin{thebibliography}{00}
924: 
925: \bibitem{PC2001} J. Parkin and B. Cohen, The Lancet \textbf{357}, 1778
926: (2001).
927: 
928: \bibitem{PW1997} A.S. Perelson and G. Weisbuch, Rev. Mod. Phys. 
929: \textbf{576}, 51 (1997).
930: 
931: \bibitem{PF1993} G. Pantaleo, C. Graziosi and A.S. Fauci,
932:  N. Engl. J. Med. \textbf{328}, 327 (1993).
933: 
934: \bibitem{KM2004} G.H. Kijak, E. Sanders-Buell, N.D. Wolfe, E. Mpoudi-Ngole, B.
935: Kim, B. Brown, M. L. Robb, D. L. Birx, D. S. Burke, J. K. Carr and F.E. McCutchan
936: AIDS Research and Human Retroviruses \textbf{20}, 521 (2004).
937: 
938: \bibitem{YL2005} O.O. Yang, E.S. Daar, B.D. Jamieson, A. Balamurugan,
939: D.M. Smith, J.A. Pitt, C.J.
940: Petropoulos, D.D. Richman, S.J. Little and A.J.
941: Leigh-Brown,  J.  Virol. \textbf{79}, 860  (2005).
942: 
943: \bibitem{ES1989} M. Eigen, J. McCaskill and P. Schuster,
944: Advances in Chemical Physics \textbf{75}, 149 (1989).
945: 
946: \bibitem{NP1998} A.U. Neumann, N.P. Lam, H. Dahari, D.R. Gretch, T.E.
947: Wiley, T.J. Layden, and A.S. Perelson, Science \textbf{282}, 103
948: (1998).
949: 
950: \bibitem{KH1997} C. Krempl, B. Schultze, H. Laude and G. Herrler,  
951: J. Virol. \textbf{71}, 3285 (1997)
952: 
953: \bibitem{FP1997} S. Franz and L. Peliti, J. Phys. A \textbf{30}, 4481 (1997).
954: 
955: \bibitem{ES1977} M. Eigen and  P. Schuster, Naturwissenschaften
956: \textbf{64}, 541 (1977). 
957: 
958: \bibitem{CD2000} L. Costa, M. Janini, M. Pinto and R. Diaz,  
959: Int. Conf. AIDS. \textbf{13} MoPpA1005 (2000).
960: 
961: \bibitem{BL2005} G. Bello, C. Casado, V. Sandonis, M. Alonso-Nieto, J. L. Vicario,
962: S. Garc\'{i}a, V. Hernando, C. Rodr\'{i}guez, J. del Romero and C.
963: L\'{o}pez-Gal\'{i}ndez, J. Gen. Virol. \textbf{86}, 355 (2005).
964: 
965: \bibitem{JD2000} A.E. Jetzt, H. Yu, G.J. Klarmann, Y. Ron, B.D. Preston,
966:  and J.P. Dougherty, J. Virol. \textbf{74}, 1234 (2000). 
967: 
968: \bibitem{DSP1992} R.J. De Boer, L.A. Segel and A.S. Perelson, J. Theor. Biol.
969: \textbf{71}, 295 (1992).
970: 
971: \bibitem{HM1995} D.D. Ho, A.U. Neumann, A.S. Perelson, W. Chen, J.M.
972: Leonard and
973: M. Markowitz,  Nature \textbf{373}, 123 (1995).
974: 
975: \bibitem{BM2002} J.T. Blackard, D.E. Cohen and K.H. Mayer,  Clin.
976: Infect. Dis.
977: \textbf{34}, 1108 (2002).
978: 
979: \bibitem{R2002} A. Ramos, D.J. Hu, L. Nguyen,K.O. Phan, S. Vanichseni, 
980: N. Promadej, K. Choopanya, M. Callahan, N.L. Young, J. McNicholl, T.D.
981: Mastro, T.M. Folks and S. Subbarao, J. Virol., \textbf{76}, 7444
982: (2002).
983: 
984: \bibitem{L2003} J. A. Levy, The Lancet \textbf{361}, 98 (2003).
985: 
986: \bibitem{HH2004} P. T. Haaft, E. J. Verschoor, B. Verstrepen, H.
987: Niphuis, R. Dubbes, W. Koornstra, W. Bogers, B. Rosenwirth and J.
988: L. Heeney. J. of Gen. Virol. \textbf{85}, 3735 (2004).
989: 
990: \bibitem{AW2002} M. Altfeld, T.M. Allen, X.G. Yu, M.N. Johnston, 
991: D. Agrawal, B.T. Korber, D.C. Montefiori, D.H. O'Connor, 
992: B.T. Davis, P.K. Lee, E. Maier, J. Harlow, P.J.R. Goulder, 
993: C. Brander, E.S. Rosenberg and B.D.
994: Walker, Nature \textbf{420}, 434 (2002).
995: 
996: \bibitem{GH2005} M. Gerhardt, D. Mloka, S. Tovanabutra, E. Sanders-Buell, 
997: O. Hoffmann, L. Maboko, D. Mmbando, D.L. Birx, F.E. McCutchan 
998: and M. Hoelscher, J. Virol.\textbf{79}, 8249 (2005).
999: 
1000: \bibitem{ZH1996} P.M. de A. Zanotto, E.A. Gould, G.F. Gao, H.P.
1001: Harvey, and E.C. Holmes, Proc. Natl. Acad. Sci. USA \textbf{93},
1002: 548 (1996).
1003: 
1004: \bibitem{CP2004} D.L. Chao, M.P. Davenport, S. Forrest, and A.S.
1005: Perelson,
1006: J. Theor. Biol. \textbf{228}, 227 (2004).
1007: 
1008: \bibitem{CS1996} F. Celada and P. E. Seiden, Eur. J. Immunol.
1009: \textbf{26}, 1350 (1996).
1010: 
1011: \bibitem{DP1995}  R.J. De Boer and A.S. Perelson, J. Theor. Biol. \textbf{175}, 567
1012: (1995).
1013: 
1014: \bibitem{WN2002} D. Wodarz1 and M.A. Nowak, BioEssays \textbf{24}, 1178 (2002).
1015: 
1016: \bibitem{PH1996}  A.S. Perelson, A.U. Neumann, M. Markowitz, J.M.
1017: Leonard and D.D. Ho,  Science \textbf{271}, 1582 (1996).
1018: 
1019: \bibitem{WP2004} F.W. Wiegel and A.S. Perelson, Immunol. Cell Biol. 
1020: \textbf{82}, 127 (2004).
1021: 
1022: \bibitem{WS1995} X. Wei, S.K. Ghosh, M.E. Taylor, V.A. Johnson, E.A. Emini, P.
1023: Deutsch, J.D. Lifson, S. Bonhoeffer, M.A. Nowak, B.H. Hahn, M.S. Saag
1024: and G.M. Shaw, Nature \textbf{373}, 117 (1995).
1025: 
1026: \bibitem{DP2005} N. M. Dixit and A. S. Perelson  \textbf{78}, 8942 (2005).
1027: 
1028: \bibitem{JH2002} M. G. Jenkins, A. Rambaut, O. G. Pybus and E. C.
1029: Holmes, J Mol Evol \textbf{54}, 156 (2002).
1030: 
1031: \bibitem{MT1995} L.M. Mansky and H.M. Temin,
1032:  J. Virol. \textbf{69}, 5087 (1995).
1033: 
1034: \bibitem{LG2004} P. Lio and N. Goldman, Trends in Microbiology \textbf{12}, 106 (2004).
1035: 
1036: \bibitem{BB1997} F. Bagnoli and M. Bezzi. Phys. Rev. Lett. \textbf{79}, 3302 (1997).
1037: 
1038: \bibitem{MC2004} M.J. Miller, O. Safrina, I. Parker and M.D. Cahalan,
1039:  J. Cell Biol. \textbf{167}, i1 (2004).
1040: 
1041: \bibitem{BE2005} C. K. Biebricher and M. Eigen \textbf{107}, 117 (2005).
1042: 
1043: \bibitem{T1992} P. Tarazona, Phys. Rev. A \textbf{45}, 6038 (1992).
1044: 
1045: \bibitem{L1987} I. Leuthausser, J. Stat. Phys. \textbf{48}, 343 (1987).
1046: 
1047: \bibitem{GM2003} A.P. Galvani and M. Slatkin, 
1048: Proc. Natl. Acad.
1049: Sci. USA \textbf{100}, 15276 (2003).
1050: 
1051: \bibitem{BC2000} C. L. Burch and L. Chao,  Nature \textbf{406}, 625 (2000).
1052: 
1053: \bibitem{SR1997} M.J. Schnell, E. Johnson, L. Buonocore and  J.K. Rose, 
1054: Cell \textbf{90}, 849 (1997).
1055: 
1056: \bibitem{RG2003} T. Revilla and G. Garc\'{y}a-Ramos, 
1057:  Math. Biosc. \textbf{185}, 191 (2003).
1058: 
1059: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1060: % non-referenced bibitems
1061: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1062: \end{thebibliography}
1063: 
1064: \end{document}
1065: