1: % Include figure files
2:
3:
4: \documentclass[prl,twocolumn,floatfix,showpacs]{revtex4}
5: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
6: \usepackage{amsmath}
7: \usepackage{epsfig}
8: \usepackage{bm}
9:
10: \setcounter{MaxMatrixCols}{10}
11: %TCIDATA{OutputFilter=Latex.dll}
12: %TCIDATA{Version=5.50.0.2890}
13: %TCIDATA{<META NAME="SaveForMode" CONTENT="1">}
14: %TCIDATA{BibliographyScheme=BibTeX}
15: %TCIDATA{LastRevised=Tuesday, September 12, 2006 01:15:35}
16: %TCIDATA{<META NAME="GraphicsSave" CONTENT="32">}
17:
18:
19: \begin{document}
20:
21: \title{A Physical Theory of the Competition that Allows HIV to Escape from
22: the Immune System}
23: \author{Guanyu Wang$^{1,2}$ and Michael W. Deem$^1$}
24: \affiliation{\hbox{}$^1$Department of Bioengineering and Department of Physics \&
25: Astronomy, Rice University, Houston, TX 77005--1892, USA \\
26: \hbox{}$^2$Department of Physics, George Washington University, Washington,
27: D.C. 20052, USA}
28:
29: \begin{abstract}
30: %In contrast to an ideal immune system,
31: Competition within the immune system
32: may degrade immune control of viral infections. We formalize
33: the evolution that occurs in both HIV-1 and the immune system quasispecies.
34: %starting from the hypothesis of an ideal immune system.
35: Inclusion of
36: competition in the immune system leads to a novel balance between the
37: immune response and HIV-1, in which the eventual outcome is HIV-1 escape
38: rather than control.
39: %By inclusion of the immune system evolutionary
40: %processes, our
41: The analytical model reproduces the three stages of HIV-1 infection.
42: %, which was previously deemed difficult.
43: We propose a vaccine
44: regimen that may be able to reduce competition between T cells, potentially
45: eliminating the third stage of HIV-1.
46: \end{abstract}
47:
48: \pacs{87.10.+e, 87.15.Aa, 87.17.-d, 87.23.Kg}
49: \maketitle
50:
51: %\date{\today}
52:
53: Our immune system is highly effective in suppressing most viral infections,
54: due to the many different T cells that exist in the repertoire of one
55: person. While many different T cells can
56: recognize a virus, only
57: those of highest affinity respond in large numbers and participate in
58: eliminating the virus. One limitation of our immune system stems from the
59: competition among T cells of similar specificity for the virus. For a series
60: of discrete infections over time, competition in the immune system is
61: associated with the phenomenon of deceptive imprinting \cite{Kohler}, or
62: original antigenic sin, and has been characterized by a random energy model
63: \cite{Deem}. Original antigenic sin is the tendency for memory immune cells
64: produced in response to a first viral infection to suppress the creation of
65: new immune cells in response to a second infection with a related strain.
66: Moreover, while they are used, these memory immune cells may not be optimal
67: for control of this second, different viral strain. Another form of
68: competition in the immune system occurs when several viral strains
69: simultaneously infect one person. In this case, the T cells compete to
70: recognize the different strains, and recognition of all strains may not be
71: uniformly effective. This immunodominance of one strain over others means
72: that the immune response to multiple infections is not a simple
73: superposition of the responses to each individual infection
74: \cite{dengue}.
75:
76: For a chronic infection with a mutating virus, the effects of competition
77: are more subtle. Since the infection changes slowly over months or years,
78: there may be no apparent original antigenic sin, as the immune system slowly
79: updates itself in an attempt to clear the virus. If the virus evolves mainly
80: through point mutation, the mutants will be only slightly different from
81: their parents and can still be eliminated by the memory T cells. Moreover,
82: whenever a sufficiently new viral strain emerges through the accumulation of
83: point mutations that cannot immediately be cleared by the immune system,
84: there is enough time for the immune system to create new T cells against the
85: new strain since the infection is chronic. Therefore, competition might
86: represent a relatively small weakness of the immune system against chronic
87: infections. However, when facing HIV-1, any trivial weakness may become the
88: immune system's Achilles' heel. A typical HIV-1 infection is characterized
89: by three phases. See Fig. \ref{fig:timecourse}.
90: %
91: %
92: \begin{figure}[t!]
93: \begin{center}
94: \epsfig{file=timecourse.eps,width=0.9\columnwidth,clip=}
95: \end{center}
96: \caption{The entire time course of HIV-1 infection. The circles are clinical
97: data from \protect\cite{Pantaleo}. The solid line is generated by our model.}
98: \label{fig:timecourse}
99: \end{figure}
100: %
101: %
102: The viral load initially
103: increases dramatically in the first viremia, followed by a sharp decline due
104: to immunological control. Thereafter, a long period of clinical latency
105: lasts for 6--10 years with small viral load. Finally, the HIV-1 viral load
106: climbs up again in the final viremia, in the onset of AIDS. Most
107: modeling works focus on the first viremia (e.g., \cite{Perelson,
108: Rouzine}). In \cite{Kamp}, the second and third stages were
109: modeled but without the first stage. The entire time course has been
110: difficult to reproduce faithfully by a unified mathematical model, probably
111: due to the disparate dynamics of the three phases. To describe each phase, a
112: different set of parameters has been given. The three stage dynamics
113: is reproduced by computational models in \cite{Hershberg, Rita}, but
114: the mapping between parameters and physiology remains to be fully established
115: in those models \cite{strain}.
116:
117: In this letter, we develop an analytical, physical theory
118: of the immune response that
119: explains HIV-1 pathogenesis. A key concept in the model is a population of
120: viruses with different sequences that arise due to viral evolution, termed
121: the viral quasispecies. Since HIV-1 and the immune system are well-matched
122: competitors, the viral quasispecies and its dual, the lymphocyte
123: quasispecies, should be equally well matched. These considerations lead to a
124: theory with a large number of dynamical equations, which is reduced to a low
125: dimensional model by symmetry. The model fits well with the standard HIV-1
126: data of the entire infection time course, and the parameters are all
127: biologically reasonable. A novel equilibrium between the immune system and
128: virus quasispecies is found, which, instead of fixing the viral load, favors
129: viral escape.
130:
131: We let $V_{i}$ ($i=0,1,\cdots I-1$) denote the HIV-1 quasispecies $i$. We
132: let $v_{i}$ denote the concentration, or load, of $V_{i}$. We let $%
133: v=\sum_{i=0}^{I-1}v_{i}$ denote the total viral load. We let $X_{i}$ denote
134: the T cell, or effector T lymphocyte, that is specific for $V_{i}$. We let $%
135: x_{i}$ denote the concentration of this T cell, and we let $%
136: x=\sum_{i=0}^{I-1}x_{i}$ denote the total T cell concentration. We let $%
137: k_{ji}$ denote the affinity between the T cell $X_{j}$ and the virus $V_{i}$%
138: . As a fairly generic assumption, we take $k_{ij}=\exp ( -\beta
139: d_{ij}) $, where $d_{ij}=d_{ji}$ is the Hamming distance between $V_{i}
140: $ and $V_{j}$. Since roughly two amino acid substitutions still permit
141: recognition, we expect $\beta $ to be on the order of $1/2$.
142: % We thus find
143: %the matrix $\mathbf{K}=( k_{ij}) $ to be a $I\times I$ symmetric
144: %one. The diagonal elements represent the maximum affinity that is
145: %normalized: $k_{ii}=1$.
146: The $k_{ij}$ ($i\neq j$)
147: represent cross-reactivities, which are generally beneficial for the immune
148: system because they offer additional killing. However, the side effect of
149: cross-reactivity is that it induces competition when the T cells
150: expand in concentration during the activation phase of the immune response.
151:
152: We first hypothesize an ideal immune system that has the following
153: properties: (i) There is no competition among the different T cells. This
154: allows for the free development of effector T cells whose total
155: concentration will reach its maximum capacity, $x = \sum_{i=0}^{I-1} x_i = 1$%
156: . For a realistic system, we will find $x<1$ due to competition. (ii) There
157: is no bias toward any viral quasispecies in generating its corresponding
158: effector cells. Therefore one has $x_{i}/x=v_{i}/v$. Since $x=1$, one has $%
159: x_{i}=v_{i}/v$.
160:
161: The real immune system and the competition induced in the response to HIV-1
162: are modeled as follows. Compared to cytotoxic T cell lymphocytes (CTLs),
163: antibodies are much less effective in controlling HIV-1. Therefore, we focus
164: on the CTL responses, and a virus is accordingly identified by its epitope
165: that stimulates CTLs. The epitope is $N=$ 8--11 amino acids long, and in our
166: model we assume $N=9$. The initial strain of virus is denoted by $V_{0}$.
167: The assumption of an unique initial strain is based on the bottleneck
168: effect \cite{Manrubia} and the reversion effect due to lack of CTL pressure
169: at infection \cite{Kent}. We assume that at each of the $N$ sites on the
170: epitope there are $A$ possible amino acid substitutions. The quasispecies
171: space size is thus $I=( A+1)^{N}$. We use $A=2$ because each
172: site can accept few substitutions to keep the virus viable. We find that the
173: simulation results are not sensitive to $A$. The subscript of $V_{i}$ is in
174: the range $0\leq i\leq (1+A)^{N}-1$ and can be written in form of a $A+1$%
175: -radix number $i=(a_{N},a_{N-1},\cdots ,a_{1})$, with $0\leq a_{n}\leq A$.
176: The Hamming distance between epitopes is $d_{ij}=\sum_{n=1}^{N}(1-\delta
177: _{a_{n}^{i},a_{n}^{j}})$, where $\delta _{lm}$ is the Kronecker delta. The
178: model is
179: \begin{eqnarray}
180: \frac{dv_{i}}{dt}&=&r_{i}v_{i}-mNAv_{i}+m\Gamma _{i}-c_{1}f_{i}( \mathbf{x
181: }) v_{i} \label{eq:dvi}
182: \\
183: \frac{dx_{i}}{dt}&=&\lambda \left( \frac{v_{i}}{v}-x_{i}\right)
184: -c_{3}g_{i}( \mathbf{x}) x_{i}, \label{eq:dxi}
185: \end{eqnarray}
186: for $i=0,1,\cdots I-1$, with the initial condition $v_{0}(0)\neq 0,v_{i\neq
187: 0}(0)=0$, and \ $x_{i}(0)=0$. Previous models have assumed linear
188: suppression ($f_{i}=$ const) and not tracked both viral and T cell
189: quasispecies evolution \cite{Perelson}.
190:
191: In Eq.\ (\ref{eq:dvi}), $r_{i}=r_{0}\exp ( -\alpha d_{i0}) $ is
192: the replication rate of $V_{i}$. The wild type $V_{0}$ replicates the
193: fastest, and mutation away demands some fitness cost. The term $m$ is the
194: rate of mutation of the amino acids at any one of the $N$ sites into any one
195: of the other $A$ amino acids (per amino acid per day). The term $mNAv_{i}$
196: thus represents mutation away from quasispecies $i$ to its $NA$\ neighbors
197: (efflux). The term $m\Gamma _{i}$ represents the reverse process (influx),
198: where
199: $
200: \Gamma _{i}=\sum_{j=0}^{A}
201: \left(
202: v_{(j,a_{N-1},\ldots, a_{1})}+v_{(a_{N},j,\ldots, a_{1})}
203: +\ldots +v_{(a_{N},a_{N-1},\ldots, j)}
204: \right) -Nv_{i}
205: $
206: is the sum over all viral species a unit Hamming distance away from $i$. The
207: term
208: $
209: f_{i}( \mathbf{x}) =y_{i}( \mathbf{x}) /
210: [ c_{2}+y_{i}( \mathbf{x}) ]
211: $
212: represents the lysis ratio of targeted cells harboring $V_{i}$, where
213: $
214: y_{i}( \mathbf{x})
215: =\sum_{j=0}^{I-1}x_{j}k_{ji}=\sum_{j=0}^{I-1}x_{j}\exp ( -\beta
216: d_{ji})
217: $
218: The derivation of
219: $f_{i}( \mathbf{x})$
220: follows as in \cite{Park}
221: by treating the competitive binding process as a Langmuir adsorption
222: isotherm of the CTLs onto targeted cells that are infected with $V_{i}$. The
223: concentration and affinity here are dimensionless, while in \cite{Park}
224: their physical values $\{ x_{j}\} $ and $\{ k_{ji}\} $
225: are used.
226: %, with the units mol/l and l/mol, respectively.
227: That is,
228: $
229: f_{i}( \mathbf{x}) =\sum_{j}\{ x_{j}\} \{ k_{ji}\}
230: /[
231: 1+\sum_{j}\{ x_{j}\} \{ k_{ji}\} ]
232: $.
233: We let $\{ x \} _{\max }$ and $\{ k \} _{\max }$ denote the
234: maximum values of concentration and affinity, respectively.
235: %. We may define $%
236: %\left\{ x_{j}\right\} =x_{j}\left\{ x\right\} _{\max }$ and $\left\{
237: %k_{ji}\right\} =k_{ji}\left\{ k\right\} _{\max }$. By substitution into (\ref%
238: %{eq:ff}) and comparison with (\ref{eq:f}), one
239: We, thus, obtain $c_{2}=1/(
240: \{ x \} _{\max } \{ k \} _{\max }) $. Typical
241: concentrations are on the order of $10^{-6}$ mol/l, typical binding
242: constants are on the order of $10^{6}$ l/mol, and so we expect $c_{2}$ to be
243: on the order of unity. Indeed, to fit clinical data, we will find
244: $c_{2}=0.8$.
245: The coefficient $c_{1}$ is the rate of clearance of virus released from
246: targeted cells that are being lysed, and to fit clinical data, we will find $%
247: c_{1}=4.0$ day$^{-1}$, close to a previous estimation of 4.3 day$^{-1}$
248: \cite{Nelson}.
249:
250: The first part of the right hand side of Eq.\ (\ref{eq:dxi}) represents a
251: delay process with time constant $1/\lambda $, since the immune system
252: always lags in response to viral changes. The term $v_{i}/v$, which equals $%
253: x_{i}$ in an ideal immune system, now serves as the moving target for $x_{i}$
254: to catch. Moreover, the tracking of the virus by the immune system is
255: disturbed by the competition $c_{3}g_{i}( \mathbf{x}) x_{i}$,
256: where
257: $
258: g_{i}( \mathbf{x} ) =\sum_{j\neq i}x_{j}k_{ji}=y_{i}( \mathbf{%
259: x} ) -x_{i}
260: $
261: represents competition from the other T cells. This term has been used in
262: studying the competition within an ecosystem
263: %, where $\mathbf{K}$\ was called
264: %the competition matrix
265: \cite{Bastolla}. The T cells that develop into $X_{i}$%
266: , must be stimulated by $V_{i}$. However, since $X_{j}$ ($j\neq i$) can
267: occupy $V_{i}$, a delay in the maturation of the T cell precursors into $%
268: X_{i}$ can occur.
269:
270: The number of non-linear equations in the model is $2(A+1)^{N}=39366$, which
271: is difficult for analysis and computation. However, there is apparent
272: symmetry in these equations: $v_{i}=v_{j}$ and $x_{i}=x_{j}$ as long as $%
273: d_{i0}=d_{j0}$. That is, all the viruses that have an identical distance to
274: the wild type have the same dynamics and can be put into the same group. We
275: let $\tilde{v}_{n}$ ($\tilde{x}_{n}$) denote the number of elements in the
276: viral (CTL) group $n$, one has $\tilde{v}_{n}=Q_{N}^{n}v_{i}$ and \ $\tilde{x%
277: }_{n}=Q_{N}^{n}x_{i}$ for any $i$ satisfying $d_{i0}=n$, where $%
278: Q_{N}^{n}=C_{N}^{n}A^{n}$
279: and $C_{N}^{n}=N!/[ n!(N-n)!] $.
280: By
281: substituting $\tilde{v}_{n}$ and $\tilde{x}_{n}$ into Eqs.\ (\ref{eq:dvi}, %
282: \ref{eq:dxi}) and noting that
283: $
284: \Gamma _{i}=(n/Q_{N}^{n-1})\tilde{v}_{n-1}+
285: [n( A-1) /
286: Q_{N}^{n}]\tilde{v}_{n}+
287: [( N-n) A/Q_{N}^{n+1}]\tilde{v}_{n+1}
288: $,
289: one obtains
290: \begin{eqnarray}
291: \frac{d\tilde{v}_{n}}{dt} &=&mA\left( N-n+1\right) \tilde{v}_{n-1}+m\left(
292: n+1\right) \tilde{v}_{n+1} \notag \\
293: &&+\left( \gamma _{0}\exp \left( -\alpha n\right) -c_{1}\tilde{f}_{n}\left(
294: \mathbf{\tilde{x}}\right) \right) \tilde{v}_{n} \notag \\
295: &&-mA\left( N-n+n/A\right) \tilde{v}_{n} \label{eq:dvn}
296: \\
297: \frac{d\tilde{x}_{n}}{dt}&=&\lambda \left( \frac{\tilde{v}_{n}}{v}-\tilde{x}%
298: _{n}\right) -c_{3}\tilde{y}_{n}\left( \mathbf{\tilde{x}}\right) \tilde{x}%
299: _{n}+\frac{c_{3}}{Q_{N}^{n}}\tilde{x}_{n}^{2} \label{eq:dxn1}
300: \end{eqnarray}
301: for $n=0,1,\cdots ,N$, where $\tilde{f}_{n}( \mathbf{\tilde{x}}%
302: ) =\tilde{y}_{n}( \mathbf{\tilde{x}}) /[c_{2}+\tilde{y}%
303: _{n}( \mathbf{\tilde{x}}) ]$. To obtain $\tilde{y}_{n}(
304: \mathbf{\tilde{x}}) $ from $y_{i}( \mathbf{x}) $, the
305: following two steps are performed.\nolinebreak\ \nolinebreak The first step
306: is to rearrange
307: %Eq.- (\ref{eq:yi})
308: the equation for $y_i$
309: by summing up all the $x_{j}$ that have
310: the same $d_{ji}=n^{\prime }$, namely to obtain
311: $
312: \tilde{y}_{n}
313: ( \mathbf{\tilde{x}}) =\sum_{n^{\prime
314: }=0}^{N}q_{n^{\prime }}\exp ( -\beta n^{\prime })$.
315: To be specific, let us consider
316: $
317: i=\left( \underset{N-n}{\underbrace{0\cdots 0}}\underset{n}{\underbrace{%
318: 11\cdots 1}}\right)
319: $
320: with $d_{0i}=n$ and a specific $d_{ji}=n^{\prime }$, $j$ must satisfy $%
321: n-n^{\prime }\leq d_{j0}\leq n+n^{\prime }$. To obtain a $j$\ with $%
322: d_{ji}=n^{\prime }$\ and $d_{j0}=k$, some 0's or 1's in this
323: expression for $i$
324: need to be changed, and the number of ways to realize this is $%
325: \sum_{l=k-n}^{n^{\prime }}Q_{N-n}^{l}C_{n}^{l_{1}}P_{n-l_{1}}^{l_{2}}$,
326: where $Q_{N-n}^{l}=C_{N-n}^{l}A^{l}$ corresponds to changing $l$ of the 0's
327: into non-0; $C_{n}^{l_{1}}$ corresponds to flipping $l_{1}$\ of the 1's into
328: 0; $P_{n-l_{1}}^{l_{2}}=C_{n-l_{1}}^{l_{2}}( A-1) ^{l_{2}}$
329: corresponds to changing $l_{2}$\ of the remaining $(n-l_{1})$ 1's into
330: digits other than 0 and 1. The restrictions are $l+l_{1}+l_{2}=n^{\prime }$
331: and $n+l-l_{1}=k$, from which one solves $l_{1}=n+l-k$ and $l_{2}=n^{\prime
332: }-n+k-2l$. Therefore one obtains
333: $
334: q_{n^{\prime }}=\sum_{k=n-n^{\prime }}^{n+n^{\prime }}
335: (\tilde{x}_{k}/ Q_{N}^{k})\sum_{l=k-n}^{n^{\prime
336: }}Q_{N-n}^{l}C_{n}^{n+l-k}P_{k-l}^{n^{\prime }-n+k-2l}
337: $.
338: The second step is to sum up all terms that have
339: the same $\tilde{x}_{k}$. That is, to obtain the matrix $\Phi_{nk}$
340: that is defined by $\mathbf{\tilde{y}}=\Phi \mathbf{\tilde{x}}
341: $. By substituting the expression of $q_{n^{\prime }}$ into
342: the expression for $\tilde{y}_{n}$
343: one obtains
344: $
345: \Phi _{nk}=\sum_{n^{\prime }=0}^{N}
346: [\exp ( -\beta n^{\prime
347: }) /Q_{N}^{k}]\sum_{l=k-n}^{n^{\prime
348: }}Q_{N-n}^{l}C_{n}^{n+l-k}P_{k-l}^{n^{\prime }-n+k-2l}
349: $.
350:
351: The model parameters are determined as follows. The term $mA$ is the
352: mutation rate per amino acid site per day. The conventional HIV-1 mutation
353: rate $\mu \approx 2.2\times 10^{-5}$ is per nucleotide base per replication
354: \cite{Huang}. Given that one replication cycle is about one day, one amino
355: acid has 3 nucleotide bases, and about $3/4$ nucleotide mutations are
356: nonsynonymous, one has $m=3\mu \times 0.75/A\approx 2.475\times 10^{-5}$ per
357: amino acid per day. The other parameter values are estimated by fitting the
358: standard data given by \cite{Pantaleo}. The viral load data is in terms of
359: the plasma HIV-1 titer, which is two raised to the number of serial $\times
360: 1/2$ dilutions that result in undetectable virus. In the first viremia
361: phase, the mutation has not yet taken significant effect, Eqs.\ (\ref{eq:dvi},%
362: \ref{eq:dxi}) can be simplified as $\dot{v}_{0}=r_{0}v_{0}-c_{1}x_{0}v_{0}/%
363: ( c_{2}+x_{0}) $ and $\dot{x}_{0}=\lambda ( 1-x_{0})$.
364: Since the CTL response peaks (i.e., $x_{0}$ approaching 1) at \ 9 to 12
365: weeks after the initial infection \cite{Abbas}, we calculate from $\dot{x_{0}%
366: }=\lambda (1-x_{0})$ that $\lambda $ is 0.033 day$^{-1}$. By noting that the
367: first viremia peaks roughly at week 6 with plasma virus titer reaching 630
368: and declines by week 12, we find $v_{0}( 0) =3.7 \times 10^{-7}$, $%
369: r_{0}=1.88$ day$^{-1}$, $c_{1}=4.0$, and $c_{2}=0.8$. The fitting of the
370: latency and the final viremia phases requires $\alpha =0.41$, $\beta =0.58$,
371: and $c_{3}=9.56\times 10^{-2}$.
372: %Repeated parameter estimations
373: %uniformly lead to values close to those given here.
374: As discussed above, only
375: $\alpha$ and $c_3$ lack comparison to literature values.
376:
377: Eqs.\ (\ref{eq:dvn}, \ref{eq:dxn1}) are still difficult for analysis because $%
378: v$ shows up in the denominator of Eq.\ (\ref{eq:dxn1}). By summing
379: Eq.\ (\ref{eq:dvn}) over $n$, one obtains
380: \begin{equation}
381: \frac{dv}{dt}=\sum_{n=0}^{N}\left( \gamma _{0}\exp ( -\alpha n)
382: -c_{1}\tilde{f}_{n}( \mathbf{\tilde{x}}) \right) \tilde{v}_{n}.
383: \label{eq:dv}
384: \end{equation}%
385: By defining $\tilde{p}_{n}=\tilde{v}_{n}/v$ and plugging $d\tilde{v}%
386: _{n}/dt=vd\tilde{p}_{n}/dt+\tilde{p}_{n}dv/dt$ into Eq.\ (\ref{eq:dvn}), one
387: obtains
388:
389: \begin{eqnarray}
390: \frac{d\tilde{p}_{n}}{dt} &=&mA( N-n+1) \tilde{p}_{n-1}+ \notag
391: \\
392: &&\left( \gamma _{0}\exp ( -\alpha n) -c_{1}\tilde{f}_{n}
393: ( \mathbf{\tilde{x}}) \right) \tilde{p}_{n} \notag \\
394: &&-mA( N-n+n/A) \tilde{p}_{n}+m( n+1) \tilde{p}_{n+1}
395: \notag \\
396: &&-\tilde{p}_{n}\sum_{n^{\prime }=0}^{N}\tilde{p}_{n^{\prime }}
397: \left( \gamma _{0}\exp ( -\alpha n^{\prime }) -c_{1}\tilde{f}_{n^{\prime
398: }}( \mathbf{\tilde{x}}) \right) \label{eq:dpdt}
399: \\
400: \frac{d\tilde{x}_{n}}{dt}&=&\lambda ( \tilde{p}_{n}-\tilde{x}_{n})
401: -c_{3}\tilde{y}_{n}( \mathbf{\tilde{x}}) \tilde{x}_{n}+\frac{c_{3}%
402: }{Q_{N}^{n}}\tilde{x}_{n}^{2}, \label{eq:dxdt}
403: \end{eqnarray}%
404: which are regular nonlinear ODEs. By integrating the above equations we find
405: that $\tilde{p}_{n}( t) $ and $\tilde{x}_{n}( t) $
406: approach the fixed point $(\bar{p}_{n}$, $\bar{x}_{n})$
407: as early as day 600. The eigenvalues of the Jacobian matrix evaluated
408: at the fixed point are all negative, indicating that the fixed point is
409: stable. That is, an equilibrium has been reached between the HIV-1
410: quasispecies and the immune system T cell quasispecies. This result holds
411: even as the viral load itself increases exponentially. All quasispecies
412: escape at the same rate, with their relative proportions fixed. To calculate
413: the rate of escape, Eq.\ (\ref{eq:dv}) is converted into $\dot{v}
414: ( t) =\varepsilon ( t) v ( t) $, where
415: \begin{equation}
416: \varepsilon ( t) =\sum_{n=0}^{N}\tilde{p}_{n}\left( \gamma
417: _{0}\exp ( -\alpha n) -\frac{c_{1}\sum_{k}\Phi _{nk}\tilde{x}_{k}
418: }{c_{2}+\sum_{k}\Phi _{nk}\tilde{x}_{k}}\right) . \label{eq:epsn}
419: \end{equation}
420: Because the viral quasispecies converges, $\varepsilon ( t) $\
421: will eventually converge to a constant $\bar{\varepsilon}$. Replacing $%
422: ( \tilde{p}_{n},\tilde{x}_{n}) $ by the fixed value $( \bar{p%
423: }_{n},\bar{x}_{n}) $, one obtains $\bar{\varepsilon}=0.0066$ day$^{-1}$%
424: . Because the equilibrium begins early, we conclude that the majority of
425: latent phase, which was previously mysterious, follows the same simple
426: dynamics as the final viremia phase. This result explains the observation
427: \cite{Mellors}
428: %\cite{Stekler}
429: that the lifespan of HIV patients is inversely correlated
430: with the minimum viral load $%
431: v_{\min }$ that happens at about day 150 and is a direct result of the
432: primary immune response. Because the onset time to the final phase scales as
433: $t\propto -\ln v_{\min }/\bar{\varepsilon}$, the smallness of $\bar{%
434: \varepsilon}$ significantly amplifies the life span dependence on $v_{\min }$%
435: .
436:
437: %The disparate dynamics of the three phases of HIV-1 disease progression has
438: %been recapitulated by a unified analytical model. The success is
439: %largely ascribe to the incorporation of competition in the model,
440: %which is not obvious but very important in chronic HIV-1 infection. Immune
441: %competition was ignored by all previous models on HIV dynamics, and
442: By incorporating immune competition,
443: we now have a physical theory based upon an analytical model. Upon
444: infection, the virus proliferates with a large rate around $1.88$ day$^{-1}$,
445: because the immune response has not been fully established. At about week
446: 6, the immune system brings the virus down to a trace level.
447: However, the ever increasing viral genetic diversity induces competition
448: among CTLs, which degrades the quality of immune control. Both a viral and a
449: T cell quasispecies are established. The concentrations of different CTLs
450: become fixed, as well as the percentages of different viral quasispecies.
451: However, the virus escapes exponentially. Interestingly, it is this
452: underlying pair of quasispecies that drives the seemingly nonequilibrium
453: process. While small, $c_{3}$ has become the Achilles' heel of the immune
454: system. Competition among T cells may exert a small influence for many
455: diseases, but is fatal for HIV-1.
456:
457: Our analysis predicts that competition among CTLs has a pivotal role on
458: viral dynamics. The function $\bar{\varepsilon}=\bar{\varepsilon}
459: ( c_{3}) $ is shown in Fig. \ref{fig:epsilon}.
460: %
461: %
462: \begin{figure}[tbp]
463: \begin{center}
464: \epsfig{file=epsilon.eps,width=0.85\columnwidth,clip=}
465: \end{center}
466: \caption{The escape rate $\bar{\protect\varepsilon}$ increases as the
467: competition $c_3$ increases.}
468: \label{fig:epsilon}
469: \end{figure}
470: %
471: %
472: One sees that as $c_{3}$,
473: or the level of competition, increases, the escape rate $\bar{\varepsilon}$
474: increases significantly. For the ideal immune system, $c_{3}=0$ and $\bar{%
475: \varepsilon}=-0.0065$, suggesting that the ideal immune system can control
476: HIV-1 infection.
477: Even for the real immune
478: system, the smallness of $\vert \bar{\varepsilon}\vert $
479: implies that HIV-1 and the immune system are well-matched competitors.
480: When all antigens are simultaneously present, competition
481: among T cells for activation and division occurs, and $c_3$ is large.
482: By vaccinating to
483: different lymph nodes with the different HIV antigens likely to occur,
484: competition between T cells is reduced, and T cells that recognize
485: each of the different antigens may be produced at high concentration
486: \cite{dengue}, and so $c_3 \approx 0$.
487: In this case the virus never rises again
488: after week 12. The regimen should be designed such that its components are
489: given according to the fixed viral quasispecies. According to our
490: simulation, the final viral percentages other than $\tilde{p}_{0}$ and $%
491: \tilde{p}_{5}$ are negligible. As for sensitivity to the parameters, if $%
492: \alpha $ or $\beta $ are increased, in which case the data of Fig.\ \ref%
493: {fig:timecourse} are not fit so well, $\tilde{p}_{6}$ also becomes
494: non-negligible. We thus propose a vaccine regimen that includes not only the
495: wild type but also the $\tilde{p}_{5}$ quasispecies. This vaccine must be
496: given before the eventual evolution of the quasispecies viral strains to
497: avoid immunodominance. If the vaccine were given after evolution of the $%
498: \tilde{p}_{5}$ species, then competition would be unavoidable due to prior
499: exposure of the immune system to the multiple viral strains.
500: % The time scale
501: %at which equilibrium is set up is roughly 500 days, rather than $1/m$ as
502: %expected by purely mutational considerations, due to the differential
503: %replication and suppression rates of these two quasispecies.
504:
505: %\bibliographystyle{unsrt}
506: \bibliography{hiv}
507:
508: \end{document}
509: