q-bio0404034/main.tex
1: \documentclass[12pt]{article}
2: 
3: \usepackage{graphicx}   \usepackage{jtb}
4: \addtolength{\topmargin}{-1in}   \addtolength{\textheight}{1.5in}
5: \addtolength{\oddsidemargin}{-.5in}
6: \addtolength{\evensidemargin}{-.5in}   \addtolength{\textwidth}{1in}
7: \renewcommand{\baselinestretch}{1.3}
8: 
9: \begin{document}       
10: 
11: 
12: 
13: \title{The Dynamics of Acute Inflammation} \author{Rukmini
14: Kumar$^{1}$, Gilles Clermont$^2$, Yoram Vodovotz$^{3}$ and Carson
15: C. Chow$^{4}$\footnote{Corresponding  
16: Author, Present Address: Laboratory for Biological Modeling, NIDDK,
17: NIH, Bethesda, MD 20892}\\
18: $^1$Departments of Physics and Astronomy \\
19: $^2$Department of Critical Care Medicine\\
20: $^3$Department of Surgery\\
21: $^4$ Department of Mathematics \\ 
22: University of Pittsburgh, Pittsburgh, PA 15260}
23: %\date{Type Date Here}
24: 
25: \maketitle
26: %
27: %\documentclass[12]{report}
28: %\usepackage{amsmath}
29: %\usepackage{latexsym}
30: %\usepackage{psfig}
31: %\usepackage{graphics}
32: %\usepackage[dvips]{graphicx}
33: %\hbadness=10000 \tolerance=1600
34: %\setlength{\textwidth}{6.5 in}
35: %\setlength{\oddsidemargin}{0.0 in}
36: %\setlength{\evensidemargin}{0.5 in}
37: %\setlength{\topmargin}{-0.5625in}
38: %\setlength{\textheight}{9.0 in}
39: %\setlength{\headheight}{0.25in}
40: %\setlength{\headsep}{0.4in}
41: %\renewcommand{\baselinestretch}{1.5}
42: %\pagestyle{plain} 
43: 
44: \abstract{When the body is infected, it mounts an
45: \emph{acute inflammatory response} to rid itself of the pathogens and
46: restore health. Uncontrolled acute inflammation due to infection is
47: defined clinically as \emph{sepsis} and can culminate in organ
48: failure and death. We consider a three dimensional ordinary
49: differential equation 
50: model of inflammation consisting of a pathogen, and two inflammatory
51: mediators. The model reproduces the healthy outcome and diverse
52: negative outcomes, depending on initial conditions and parameters.  
53: We analyze the various bifurcations between the different outcomes
54: when key parameters are changed and suggest various therapeutic
55: strategies.  We suggest 
56: that the clinical condition of sepsis can arise from several distinct
57: physiological states, each of which requires a different treatment
58: approach.}  \\
59:  \textbf{Keywords:}
60: Systemic Inflammation, Sepsis, Ordinary Differential Equation (ODE)
61: models, Phase-space and Bifurcation analysis
62: 
63: \newpage
64: \section{Introduction}
65:   The initial  response of  the body  to an  infection or  trauma is
66: called   the   acute  inflammatory   response.   This  response   is
67: non-specific and  is the first line  of defense of  the body against
68: danger~\cite{janeway}.  It  consists  of  a  coordinated  local  and
69: systemic   mobilization  of   immune,  endocrine   and  neurological
70: mediators. In a healthy  response, the inflammatory response becomes
71: activated, clears the pathogen (in the event of infection), begins a
72: repair process  and abates. However, inflammation  itself can damage
73: otherwise   healthy  cells  which   could  then   further  stimulate
74: inflammation.  This  runaway inflammation can lead  to organ failure
75: and death. Systemic inflammation  accompanied by infection, based on
76: its      clinical       manifestations,      is      defined      as
77: sepsis~\cite{sepsisdefnold}.   Sepsis   is   a   common   and 
78: frequently  fatal  condition, with  750,000  cases  annually in  the
79: United States alone in  1995~\cite{Angus}.  
80: 
81: 
82: Though much  has been learned about the molecular and physiological
83: pathways of the acute  inflammatory response, this knowledge has not
84: led  to  many  effective  therapies  against  sepsis.  The sole
85: approved drug therapy for severe sepsis is
86: activated  Protein-C, which only  reduced  mortality by  6\% compared  with
87: controls in clinical trials~\cite{APC,review}. 
88: One  reason  for  the lack  of  effective
89: treatments may be that the  complex nature of the inflammatory response
90: renders the  effect of  targeting isolated components  of inflammation
91: difficult  to  predict.   Thus,  mathematical modeling  may  provide
92: insights into  the global dynamics of the  inflammatory process from
93: which therapies may be developed.  We propose that simple models of the
94: acute  inflammatory  response can exhibit various outcomes  and
95: facilitate an understanding of  the complex interactions between the
96: various components of the response.
97: 
98: We   present  a   simple  3-dimensional  model   of  the
99: inflammatory  response  to  infection  that captures  the  following
100: clinically  relevant  scenarios:  a  \emph{healthy} response  where
101: pathogen   is  cleared   and  the   body  returns   to  homeostasis,
102: \emph{recurrent infection} where  inflammation is inadequate and the
103: pathogen   cannot   be   completely   eliminated,   \emph{persistent
104: infectious inflammation} where  the pathogen levels and inflammation
105: are   high,  \emph{persistent  non-infectious   inflammation}  where
106: pathogen  is  cleared  but  inflammation persists  and  \emph{severe
107: immuno-deficiency} where  pathogen has  grown to saturation  but the
108: inflammatory response  is very low.  The model suggests that sepsis
109: is a multifaceted disease and  narrowly targeted interventions are
110: unlikely to succeed.
111:  
112: \section{Reduced Model of Acute Inflammation}
113: 
114: Invading pathogens such as bacteria are rapidly detected by the body and
115: an acute inflammatory response ensues.
116: Among the first responders are
117: phagocytic immune cells such as neutrophils and macrophages.  These
118: immune cells detect the bacterial cell components, become ``activated"
119: and begin to release pro-inflammatory cytokines such as Tumor Necrosis
120: Factor-$\alpha$ (TNF-$\alpha$), 
121: Interleukins (IL)-1, IL-6, IL-8  and High Motility Group Box-1
122: (HMGB-1) that activate more immune cells and recruit them to the 
123: sites of the infection~\cite{review}. In addition, anti-inflammatory
124: mediators such 
125: as IL-10 and Transforming Growth Factor-$\beta$
126: (TGF-$\beta$) are also released which inhibit the 
127: production of the pro-inflammatory mediators.  The activated 
128: phagocytic cells kill bacteria directly by engulfment and secretion of
129: toxic chemicals such as oxygen free radicals~\cite{janeway}.  These
130: substances can damage otherwise healthy cells.  These damaged or
131: dysfunctional cells can then induce more inflammation~\cite{decide1}.
132:  
133: Ideally, the inflammatory response eliminates the pathogen and then
134: subsides. In some cases, the response might not be strong enough to
135: clear the pathogen. In other cases, a positive feedback loop could
136: arise between the early and late pro-inflammatory waves leading to a
137: non-abating inflammatory response.  Clinically, a sustained acute
138: inflammatory response is manifested as septic shock and could
139: culminate in organ failure and death.  Though this is a simplification
140: of the pathogenesis of sepsis, the idea is supported by the
141: persistence of high levels of pro-inflammatory cytokines in
142: non-survivors of sepsis~\cite{pinsky,abraham}.
143: 
144: A simplified picture of the acute inflammatory response is that
145: an infectious pathogen triggers early pro-inflammatory responders
146: which attempt to kill the pathogen. The early inflammatory mediators
147: then activate later inflammatory mediators which can further excite
148: the early mediators.  This is the basis of our model which consists of
149: three variables: 1) a pathogen $p$, which is an instigator of the
150: innate immune response; 2) an early pro-inflammatory mediator $m$,
151: which can be thought of as representing the combined effects of immune
152: cells such as macrophages and neutrophils together with early
153: pro-inflammatory mediators such as TNF-$\alpha$ and IL-1; 3) a late
154: pro-inflammatory mediator $l$ which represents a late pro-inflammatory
155: feedback.  This is a combined effect of cytokines such as IL-6, HMGB-1
156: and stimulatory effects of tissue damage and dysfunction.  Although
157: our model is extremely simple, it captures qualitatively all the
158: salient features of a more complicated and biologically faithful model
159: that is currently being developed~\cite{ccm}.
160: 
161: We consider mass-action type kinetics in a well-mixed volume.  The
162: dynamics obey:
163: \begin{eqnarray}
164: \frac{dp}{dt}&=&k_pp(1-p)-k_{pm}mp\label{model1}\\
165: \frac{dm}{dt}&=&(k_{mp}p+l)m(1-m)-m\label{model2}\\
166: \frac{dl}{dt}&=&k_{lm}f(m)-k_ll \label{model3}
167: \end{eqnarray}
168: where
169: \begin{equation}
170: f(m)=1+\tanh \left(\frac{m-\theta}{w}\right),
171: \label{coupling}
172: \end{equation}
173: $\theta$ is an activation threshold and $w$ is an activation width.
174: All the variables and parameters are non-negative.  The pathogen $p$
175: obeys logistic growth and is killed when it interacts with $m$.  The
176: presence of $p$ or $l$ and $m$ will stimulate the growth of $m$ which
177: also has an intrinsic death rate.  This growth saturates as $m$
178: increases towards unity mimicking the effects of cell depletion and
179: anti-inflammatory cytokines.  The late mediator $l$ is recruited by
180: $m$ through a sigmoidal coupling function (\ref{coupling}) and is
181: cleared with the rate of $k_l$. In summary, the dynamics
182: obey predator-prey  dynamics with a delayed response.
183: 
184: We note that the specific nature of the interactions of our model are
185: not essential for the  qualitative dynamics we find.  As an example,
186: the factor $1-m$ in (\ref{model2}) may be replaced by $1/(1+m)$ or the
187: form of the coupling function (\ref{coupling}) can be changed without affecting
188: our conclusions.  The important element is that there be an early and
189: late inflammatory mediator with saturating kinetics.
190: 
191: \section{Dynamics of the Model}
192: 
193: 
194: \subsection{Numerical examples}
195: 
196: The model (\ref{model1}) - (\ref{model3}) exhibits behavior reminiscent
197: of what is observed in clinical settings.  Given an initial condition
198: of $p>0$, $m>0$ and $l>0$, $p$ grows, inducing $m$ and $l$ to grow and
199: attempt  to eliminate $p$.  Depending on initial conditions and
200: parameters, the ensuing orbits either approach stable fixed points or
201: undergo oscillations, each having a physiological interpretation.  We
202: note from (\ref{model2}) that a non-zero positive initial value for
203: $m$ is necessary to generate an inflammatory response. The background
204: level of late mediators is given by $l^{0}=(k_{lm}/k_l)f(0)$ and is
205: non-zero for our choice of parameters.  Both these properties are
206: consequences of our particular choice for the form of the model and
207: not essential for the qualitative results we find.
208: 
209: We must adopt one crucial element before we can interpret our results.
210: In a healthy response to infection, the inflammatory response should
211: become activated, eliminate the bacteria and return to rest.  However,
212: in model (\ref{model1})-(\ref{model3}), while $p$ can diminish
213: exponentially fast, it can never become completely eliminated in a 
214: finite time.  Thus a consequence is that if $m$ returns to rest after
215: an inflammatory episode, $p$ can re-grow from an infinitesimal
216: quantity.  This is an artifact of the model that arises because the
217: approximation of $p$ as a continuous variable breaks down when the
218: population becomes small and the discreteness of the pathogen number
219: becomes important.  In this regime, a stochastic or agent based model
220: where the pathogen can be completely eliminated is more appropriate.
221: We finesse the discreteness problem by introducing a threshold for
222: pathogen level.  When $p$ falls below this threshold, we consider it
223: to be completely cleared.   We propose that there is an effective threshold
224: representing a single pathogen particle below which,  on average, the
225: pathogen population is eliminated.  We will show how to calculate this
226: threshold explicitly for various models in a future publication.
227: 
228: We consider numerical examples for various values of $k_{pm}$,
229: $k_{mp}$, and $k_{lm}$.  In Section \ref{fpandb}, we show that these are 
230: natural bifurcation parameters of the system.  
231: The other parameters are fixed at $k_{pg}=3$,
232: $k_{l}=1$, $\theta=1$ and $w=0.5$.   
233: We show the effects of varying 
234: $\theta$ and $w$ in Appendix \ref{couplingcurve}.  Numerical
235: simulations and bifurcation plots were generated with
236: XPPAUT~\cite{bardxpp}.
237: 
238: A {\em healthy response} to infection as seen in
239: Fig.~\ref{fig:healthy} corresponds to an orbit that spirals outwards
240: so that $p$ falls below threshold during the oscillation.  The
241: pathogen is then completely cleared which allows the inflammatory
242: response to relax back to rest. 
243: 
244: In Fig.~\ref{fig:shock}, the same
245: parameters are used but the initial pathogen load is higher so that
246: instead of returning to background levels, the inflammatory mediators
247: are over-excited and remain elevated.  We relate this situation to a
248: state of {\em persistent non-infectious inflammation} where even
249: though the pathogen is cleared, the inflammatory response does not
250: abate. In our model, this state is a fixed point but in a real
251: organism we expect that if this condition continued, it would
252: eventually lead to multiple organ failure and
253: death~\cite{noinfsepsis}.  
254: 
255: Starting again from a healthy situation, if the pathogen
256: susceptibility to the host's defenses ($k_pm$)  is decreased, we can  
257: enter a domain of {\em persistent infectious 
258: inflammation} where the inflammatory response is high but the pathogen
259: still cannot be cleared as seen in Fig.~\ref{fig:sepsis}.  We would
260: equate this condition with a severe septic state where both infection
261: and inflammation are uncontrolled.  In this case the damage caused by
262: both the pathogen and inflammation are disrupting body function and if
263: unabated death will result. Patients with systemic inflammation, with
264: and without documented infection, 
265: are observed in clinical settings~\cite{epidem02}.
266: 
267: If we reduce the recruitment rate of $l$ ($k_{lm}$), the healthy response
268: can be turned into one of recurrent infection as seen in
269: Fig.~\ref{fig:recurr}. (Changes in other parameters could also lead to
270: recurrent infection as we see in the next section).  In this case, low
271: levels of infection persist indefinitely.  This could be likened to
272: infection with tuberculosis, yeast infections or low-grade bacterial
273: infections that persist for long 
274: periods of time~\cite{lowgrade}.  Although, an organism could survive
275: this state for a 
276: long duration it may eventually succumb.
277: 
278: Finally, in Fig.~\ref{fig:suppress}, the activation rate of $m$ due to
279: $p$ ($k_{mp}$) is very  weak, and we have a state  of {\em
280:   immuno-suppression} or 
281: {\em  immuno-deficiency} where  the pathogen  grows to  saturation and
282: does not elicit any response from  the body.  This could happen if the
283: immune-system had  already been  compromised by previous  infection or
284: trauma and  then the organism is  reinfected.  Opportunistic bacterial
285: and fungal infections have  been observed in immuno-suppressed patient
286: populations such as HIV infected  patients, the elderly and those with
287: organ transplants~\cite{nosoc}.
288: These five scenarios are the only possible outcomes in the model.
289: 
290: 
291: 
292: 
293: \subsection{Fixed Points and Bifurcations}
294: \label{fpandb}
295: 
296: These various regimes and transitions are best understood by examining
297: the fixed points and associated bifurcations of the model
298: (\ref{model1})-(\ref{model3}).  The fixed points satisfy the following
299: conditions:
300: \begin{eqnarray}
301: 0&=& p\left[p-1+\left(\frac{k_{pm}}{k_p} \right) m\right]
302: \label{cond1}\\
303: 0&=&m\left[m-1+\left(\frac{1}{k_{mp}p+l}\right)\right]\label{cond2}\\
304: l&=&\frac{k_{lm}}{k_l}f(m)\label{cond3}
305: \end{eqnarray}
306: Substituting (\ref{cond3}) into (\ref{cond2}) and rearranging gives
307: \begin{equation}
308: 0=m\left[p-\frac{1}{k_{mp}}\left(\frac{1}{1-m}-\frac{k_{lm}}{k_l}f(m)\right)\right].
309: \label{cond4}
310: \end{equation}
311: We consider three natural parameters $a\equiv k_{pm}/k_p$, $k_{mp}$,
312: and $b\equiv k_{lm}/k_l$, that appear in the fixed point conditions
313: (\ref{cond1}) and (\ref{cond4}). These three parameter combinations
314: represent the pathogen susceptibility to $m$ compared to its growth
315: rate (i.e. inverse of the pathogen virulence), the activation rate
316: of early responders $m$ due to $p$, and the effective recruitment rate of $l$
317: due to $m$, respectively.
318: 
319: The intersections of conditions (\ref{cond1}) and (\ref{cond4}) give
320: the fixed points of the system.  The dependence of the fixed points
321: with the parameters is best observed as a projection in the $p-m$
322: plane as shown in   
323: Fig.~\ref{fig:nullclines} where the fixed point conditions
324: (\ref{cond1}) and (\ref{cond4}) are plotted. Lines $p=0$ and $m=0$ are
325: unaffected by changes in parameters. The line
326: \begin{equation}
327: p=1-a m \label{fp2}
328: \end{equation} 
329: is affected only by $a$ and sweeps across the $p-m$ plane as $a$ is decreased.
330: The curve
331: \begin{equation}
332: p=\frac{1}{k_{mp}}\left(\frac{1}{1-m}-b f(m)\right).
333: \label{fp6}
334: \end{equation} 
335: is affected only by the immune parameters $b$ and $k_{mp}$. It drops
336: below $p=0$ as $b$ is
337: increased. Changing $k_{mp}$ affects the height and angle at which
338: (\ref{fp6}) intersects (\ref{fp2}).
339: 
340: There are five fixed points which we have labeled from FP~1 to FP~5. 
341: FP~1 is given by $p=0$, $m=0$, and $l=(k_{lm}/k_l)f(0)$.  This
342: fixed point is always unstable because the pathogen is growing.  FP~2
343: is given by $p=1$, $m=0$, $l=(k_{lm}/k_l)f(0)$ and may be interpreted
344: as a severely \emph{immuno-deficient} state.  Here the pathogen has
345: grown to saturation, but there is no early immune response and the
346: late response remains at the background level.  This point is stable when the
347: early immune response is very weak as shown in Appendix~\ref{appfp2}.
348: 
349: FP 4 and FP 5 arise from a saddle node bifurcation when $p=0$ in
350: (\ref{fp6}). This can be achieved by increasing $b$ from zero as seen
351: in the bifurcation plot in Fig.~\ref{fig:varylt}.  FP~4 is always
352: unstable and never represents any physiologically relevant
353: scenario. When stable, FP~5 represents the {\em persistent
354: non-infectious inflammation} fixed point. FP~3 is given by the
355: intersection of the line (\ref{fp2}) with the curve (\ref{fp6}). FP~3
356: could represent {\em healthy}, {\em recurrent infection} or {\em
357: persistent infectious inflammation} states, depending on parameter
358: values. Below, we vary the three parameter combinations $a$, $k_{mp}$ 
359: and $b$ and examine the bifurcation plots.
360: 
361: In the healthy scenario, $a$ is large enough so that FP~3 is an
362: unstable spiral and FP 5 is the only stable node ($k_{pm}=20$ in
363: Fig.~\ref{fig:varyp}). In this case, initial conditions close to FP 3
364: undergo oscillations that take the pathogen below the elimination
365: threshold and are interpreted as healthy trajectories.  A higher
366: pathogen initial condition for the same set of parameters takes the
367: trajectories directly to FP 5 and this behavior is interpreted as
368: persistent non-infectious inflammation. When $a$ is decreased,  the
369: slope of line (\ref{fp2}) decreases and it sweeps through FP 4
370: rendering it unstable through a transcritical bifurcation.  However,
371: $p$ is negative and thus FP 3 becomes unphysical ($k_{pm}=10$ in
372: Fig.~\ref{fig:varyp}). Hence FP 5 is the only stable fixed point for
373: all initial conditions in this regime.  As $a$ is decreased further FP
374: 3 crosses FP 5 and they exchange stability through another
375: transcritical bifurcation.  FP 3 then becomes the only stable fixed
376: point and is the global attractor for all initial conditions.
377: 
378: As expected, starting from a healthy scenario,  the outcome of the
379: system becomes more and more severe with decreasing pathogen
380: susceptibility (i.e. increasing effective pathogen virulence). If we
381: were to increase 
382: susceptibility $a$ starting from the healthy scenario, FP 3 changes
383: into a stable spiral 
384: surrounded by an unstable limit cycle through a subcritical Hopf
385: bifurcation. Now, trajectories within the unstable limit cycle spiral
386: into the stable fixed point FP 3 resulting in {\em recurrent
387: infection} and unstable spiral trajectories outside the unstable limit
388: cycle will eventually dip below threshold and be interpreted as
389: healthy. As FP 5 is still the attractor when the initial $p$ or $m$ is
390: very high, this set of parameters could lead to three different
391: outcomes based on the initial conditions ($k_{pm}=40$ in
392: Fig.~\ref{fig:varyp}). Increasing $a$ further, increases the radius of
393: the unstable limit cycle until it undergoes a homoclinic bifurcation
394: when it collides with FP 1. Here FP 3 is a stable spiral and only
395: recurrent infection is supported. The model therefore predicts that
396: pathogen clearance is not possible for a range of high pathogen
397: susceptibility. However, as we increase susceptibility much further, FP
398: 3 gets closer to the origin so that oscillations around it take $p$
399: below threshold and may be interpreted as healthy.
400: 
401: Varying $k_{mp}$, the strength of the early response, does not affect the
402: stability or position of FP 5, the persistent non-infectious
403: inflammation point. Decreasing $k_{mp}$ takes a healthy state to a
404: state where healthy and recurrent infection co-exist and finally to
405: one of recurrent infection only as shown in
406: Fig.~\ref{fig:varyer}. Decreasing $k_{mp}$ below $1-(k_{lm}/k_l)f(0)$
407: makes FP 3  collide with FP 2 (the severely immuno-deficient state
408: where $p$ grows to saturation and $m$ and $l$ remain at background
409: values). In a transcritical bifurcation, FP 2 becomes the stable fixed
410: point of the system and FP 3 becomes unphysical since
411: $m<0$. Increasing $k_{mp}$ from the healthy value reduces the value of
412: $p$ in FP 3 - this reduces the range of initial conditions for healthy
413: behavior. However, for even very large values of $k_{mp}$, FP 3 is an
414: unstable spiral with complex eigenvalues supporting healthy behavior
415: for initial conditions starting close to FP 3.
416: 
417: Varying the strength of the late response, $b$ affects both FP 3 and
418: FP 5. As we decrease $b$, FP 3 undergoes a sub-critical Hopf followed
419: by the homoclinic bifurcation of the unstable limit cycle similar to
420: the above cases (Fig. \ref{fig:varylt}). FP 4 and FP 5 are created in
421: a saddle-node bifurcation when (\ref{fp6}) intersects $p=0$ by
422: increasing $b$. However, there is also an upper limit to $b$ beyond
423: which healthy behavior is not supported. The curve (\ref{fp6}) is
424: below zero when it intersects (\ref{fp2}) and FP 3 is unphysical. Thus
425: when the late response is too high, persistent non-infectious inflammation
426: is the only possible outcome.
427: 
428: In the preceding, although we have examined the behavior of the system by
429: varying one parameter at a time, it gives us a picture
430: of the global dynamics of the system. Given the strong nonlinear
431: saturation in the dynamics,  orbits can either approach a fixed point,
432: a limit cycle, or a strange attractor.  Given that FP~1 and FP~4 are
433: always unstable and FP 2 is stable only when $k_{mp}$ is very low,
434: this leaves FP 3 (or stable limit cycles around it) and FP 5 as the
435: only  candidates for global attractors. Ideally however, we would like
436: FP 3 to be an unstable spiral or be surrounded by an unstable limit
437: cycle, so that oscillations around it can be interpreted as healthy
438: behavior.
439: 
440: This gives an explanation for why healthy behavior always entails the
441: risk of uncontrolled inflammation.  In order for the pathogen to be
442: cleared, we require an unstable spiral that can take $p$ below
443: threshold.  However, if it were not for the (externally imposed)
444: elimination threshold the orbit would eventually end up at some
445: attractor and persistent non-infectious inflammation is the global
446: attractor. Conversely, if the curve (\ref{fp6}) has not yet
447: intersected $p=0$ to form FP 4 and FP 5 (no persistent inflammation
448: fixed point), FP 3 or a stable limit cycle around it (recurrent
449: infection) is the global attractor.
450: 
451: There is a possibility that FP 3 could undergo a supercritical Hopf
452: bifurcation (discussed in Appendix~\ref{appfp3}) which leads to a
453: stable limit cycle around it. However, this is generically interpreted
454: as recurrent infection because fine tuning would be required to ensure
455: that the limit cycle takes $p$ below threshold. Although, not possible
456: in our model, a strange attractor might occur in higher dimensional
457: models.  We believe that this is unlikely given the large amounts of
458: negative feedback and dissipation in the system.  However, even if it
459: were to exist, it may still not lead to a healthy situation as $p$ may
460: never be eliminated or the early and late inflammatory mediators could
461: stay elevated albeit in a chaotic manner.
462: 
463: Physiologically, this implies that in order to completely rid the body
464: of a pathogen, the inflammatory response must respond strongly and
465: remain elevated for a long enough time.  If it responds too weakly,
466: then the pathogen persists.  If it abates too quickly, then
467: there will be recurrent oscillatory infection.  However, if it
468: responds too strongly and too persistently then there is a risk that
469: it will be self-sustaining even after the pathogen is cleared. Thus,
470: there is a trade-off between being able to eliminate pathogens
471: completely and risking nonabating inflammation.   In
472: Appendix~\ref{appbzone}, we calculate the parameter ranges where a
473: healthy response is possible.
474: 
475: 
476: \section{Discussion}
477: 
478: Systemic inflammation and the ensuing organ damage
479: is a major cause of mortality today~\cite{tilney,eiseman}. This is also a disease
480: created by modern medicine.  Before the discovery of advanced
481: resuscitation techniques, patients
482: could not be kept alive long enough for the condition to fully
483: unfold. Patients often died from blood loss and severe infections
484: before uncontrolled inflammation arose. With the advent and
485: improvement of antibiotics and organ support therapy,
486: the condition has  become increasingly  relevant~\cite{bonerev96}. The
487: incidence of systemic  inflammation  is  also  expected to  increase
488: with  further advancement   of   medical   technology   and   the
489: aging of the population which are more susceptible~\cite{Angus}.
490: 
491: It  is  now  recognized  that  the  dysregulation  of  the  underlying
492: processes of  acute inflammation can lead to multiple organ  failure and
493: death and that  this is a common pathway  for diverse instigators such
494: as  trauma,  hemorrhage and  infection.   Our  simple  model of  acute
495: inflammatory response to infection shows the various negative outcomes
496: that  arise from  improper inflammatory  response. These  scenarios of
497: persistent     inflammation    (with    and     without    infection),
498: immuno-suppression and  recurrent infection have all  been observed in
499: critically  ill  patients~\cite{epidem02,bonerev96}.  Although  strict
500: correspondence with clinical reality is difficult to establish because
501: of the  simplicity of  our model,  we are able  to classify  our model
502: behavior  into  the same  broad  categories.   The  model has  a  rich
503: bifurcation  structure and exploring  it allows  us to  understand how
504: changing  parameters   can  take  the  system  from   one  outcome  to
505: another. The model  shows that in order to have  a healthy response to
506: infection the virulence of the pathogen cannot be too strong or too weak,
507: the early  pro-inflammatory response cannot  be too weak, and  the late
508: response cannot be too strong or  too weak.  Analysis of this model gives
509: clues  to  approach  the  problem  of treating  severe sepsis.  Vastly
510: different  therapeutic strategies  are  called for  to  deal with  the
511: diverse negative outcomes.
512: 
513: The strength  of the acute  inflammatory response varies  in different
514: individuals and may vary depending on age and environmental factors in
515: the same  individual.  Studies suggest  a strong genetic  influence on
516: the outcome of sepsis, and  genetics may explain the wide variation in
517: the individual response to infection~\cite{Genreview}. We examined the
518: effect of  the strength of early and  late pro-inflammatory responses.
519: The model suggests that only the strength of the late pro-inflammatory
520: wave  governs  predisposition  towards  a state  of  persistent
521: inflammation.  No  matter how  exuberant the early  wave may  be, only
522: controlling the feedback from the  late wave can determine whether the
523: outcome is  healthy recovery  or uncontrolled inflammation.   Thus any
524: therapy for persistent non-infectious inflammation must target the
525: slow pro-inflammatory mediators.  
526: 
527: Experiments have found that survival
528: was improved in infected mice when the moderately slow cytokine IL-6 was
529: reduced~\cite{riedemann}.  However, if too much IL-6 was removed then
530: there was a detrimental effect.  This result is consistent with our model in
531: that a small reduction lowered the possibility of a sustained
532: inflammatory response but lowering it too much precluded the
533: possibility of eliminating the bacteria.
534: Evidence also has suggested that down-regulating  HMGB-1, a  late 
535: acting  pro-inflammatory  mediator might  be  a  potential target  for
536: anti-sepsis therapies~\cite{HMGB}.  Activated Protein-C which has been
537: recently   approved  for  treatment,   is  also partially a  late
538: anti-inflammatory agent~\cite{APC}.    Previous therapeutic  attempts
539: have mostly   focused   on 
540: down-regulating the early  pro-inflammatory mediators and as predicted
541: by    the     model    have    not     shown    great    effectiveness
542: \cite{clintrial1,clintrial2}.  
543: 
544: On the other hand, if   the  patient  is
545: suffering  from persistent infectious  inflammation, then  therapies
546: must be  aimed at both  reducing  the  pathogen   load  and  the  late
547: pro-inflammatory response.  In this case, timing of the therapies may
548: be important.  It would be necessary to reduce the bacterial load
549: first before reducing the inflammation.
550: 
551: Conversely, low pathogen virulence or  a weak immune response can lead
552: to  low level  persistent or  recurrent infection.   Other theoretical
553: models  of infection  have  similar predictions.   Persistence of  the
554: tuberculosis  bacterium   {\it  Mycobacterium  tuberculosis}   at  low
555: densities for extended periods has  been suggested to be the result of
556: slow  growth rates~\cite{antia}.   Similarly, a  down-regulated immune
557: response to  {\it Helicobacter pylori} bacteria has  been suggested to
558: result    in    its    persistent    colonization   of    the    human
559: stomach~\cite{denise}.  In  the clinical setting, patients  in the ICU
560: with  decreased  host defenses  are  susceptible to  hospital-acquired
561: infections~\cite{nosoc}.   These   infections  which  may   be  easily
562: resolved  in   a  healthy   individual  might  result   in  unresolved
563: inflammation     and     prove      fatal     to     a     compromised
564: individual. Immuno-stimulatory  therapy might  be effective in  such a
565: situation.
566: 
567: The model suggests that a  healthy outcome is possible only when the
568: risk  of  persistent inflammation  is  also  present.   A strong  late
569: immune  response that  increases the  risk of  unabated inflammation
570: also ensures complete elimination of pathogen.  This could explain why
571: this risk has been retained by the evolution. 
572: 
573: 
574: In the clinical setting, inflammatory states are defined by
575: symptoms and a few biological markers~\cite{sepsisdefnold}.  For example,
576: Systemic Inflammatory Response Syndrome (SIRS) is a condition
577: characterized by -
578: elevated respiratory and heart rates, fever, and an abnormal
579: white blood cell count. The severity of sepsis is based on the
580: presence of circulatory shock (low blood perfusion) and organ
581: failure~\cite{ASCC,muckart}.
582: Similarly, treatment and 
583: support of these critically ill patients is largely based on
584: clinical signs and a few biochemical and hematological parameters.
585: However, the
586: various outcomes of our model are  based on the levels of the immune
587: responders  and  the  pathogen.   Diverse  physiological  states  as
588: exemplified in  the model could  be manifested by the  same clinical
589: symptoms.   Hence,  one  major  problem for  finding  an  effective
590: treatment  of  sepsis is  a  diagnostic  one.   The model  shows  that
591: sepsis  should be  considered  as a  set  of distinct  physiological
592: disorders that require separate  therapies even though they may have
593: overlapping symptoms.  Categorizing septic states based on levels of
594: bio-markers rather than clinical symptoms would be the first step in
595: addressing   this  problem   and   effort  is   ongoing  in   that
596: direction~\cite{sepsisdefnnew,marshall}.
597: 
598: 
599: \section{Acknowledgments}
600: This work was funded in part by NIGMS grant R01-GM-67240 and the
601: A.~P.~Sloan Foundation.
602: 
603: 
604: \appendix
605: \section{Appendix}
606: \subsection{Eigenvalues of Fixed Points}
607: The Jacobian matrix of the model (\ref{model1}) - (\ref{model3}) is
608: \begin{equation}
609: \left(
610: \begin{array}{cccc}
611: k_p(1-2p)-k_{pm}m & -k_{pm}p & 0 \\ k_{mp}m(1-m) & (k_{mp}p+l)(1-2m)-1
612: & m(1-m)\\ 0 & \frac{k_{lm}}{w}{\rm sech}^{2}(\frac{m-\theta}{w}) &
613: -k_{l}
614: \end{array}
615: \right)
616: \label{jacobian}
617: \end{equation}
618: 
619: 
620: \subsubsection{FP 1}
621: \label{appfp1}
622: For FP1: $p=0,m=0,l=\frac{k_{lm}}{k_l}(1+\tanh(\theta/w))(\equiv
623: l^{0}$),  the Jacobian is 
624: \begin{equation}
625: \left(
626: \begin{array}{cccc}
627: k_p & 0 & 0\\ 0 & l^{0}-1& 0\\ 0 & \frac{k_{lm}}{w}{\rm
628: sech}^{2}\frac{\theta}{w}& -k_{l}
629: \end{array}
630: \right).
631: \end{equation}
632: This matrix is lower-triangular and the eigenvalues are  $k_p,
633: l^{0}-1, -k_{l}$. The fixed point is never stable as one of the
634: eigenvalues $k_{p}$ is always positive.
635: 
636: \subsubsection{FP 2}
637: \label{appfp2}
638: For FP2: $p=1,m=0,l=l^{0}$,  the Jacobian is
639: \begin{equation}
640: \left(
641: \begin{array}{cccc}
642: -k_p & -k_{pm} & 0\\ 0 & (k_{mp}+l^{0})-1& 0\\ 0 &
643: \frac{k_{lm}}{w}{\rm sech}^{2}\frac{\theta}{w}& -k_{l}
644: \end{array}
645: \right).
646: \end{equation}
647: The characteristic polynomial for the eigenvalue $\lambda$ is
648: \[(-k_p-\lambda)(-k_l-\lambda)(k_{mp}+l^0-1-\lambda)=0\]
649: with roots: $\lambda=-k_p$, $\lambda=-k_l$ and $\lambda=k_{mp}+l^0-1$.
650: Thus the condition for stability of FP2 is $k_{mp}<1-l^{0}$ or
651: $k_{mp}<1-\frac{k_{lm}}{k_l}(1-\tanh(\theta/w))$.  The other two
652: eigenvalues are always negative.
653: 
654: \subsubsection{FP 3}
655: \label{appfp3}
656: FP3 is the solution of the following equations,
657: \begin{eqnarray}
658: p&=&1-am\\ m&=&1-\frac{1}{k_{mp}p+l}\\ l&=&bf(m)
659: \end{eqnarray}
660: This may be reduced to a single transcendental equation for $m$:
661: \begin{equation}
662: m=1-\left(\frac{1}{k_{mp}}\left(\frac{1}{1-am}\right)+ bf(m)\right).
663: \end{equation}
664: All the parameter combinations  affect the position and eigenvalues of
665: FP 3.  Parameter choices for which $p$  or $m$ of FP 3 is  negative are
666: unphysical.  FP 3  could also have real or  complex eigenvalues.  FP 3
667: has  complex eigenvalues  and  can support  oscillatory behavior  when
668: (\ref{fp2}) intersects (\ref{fp6}) such that  $m$ of FP 3 is less than
669: $m$  of  FP  4  (Fig.~\ref{fig:nullclines}).  FP 3  can  undergo  Hopf
670: bifurcations in  this part of  the phase-space - the  subcritical Hopf
671: bifurcations on varying the  three parameter combinations are shown in
672: Figs. \ref{fig:varyp}, \ref{fig:varyer}, and \ref{fig:varylt}.
673:  
674: A supercritical Hopf bifurcation occurs when we change the coupling
675: curve (\ref{coupling}) so that $\theta$ is small (low threshold for
676: activation) and $w$ is large (shallow coupling curve).  In this case,
677: for increasing $b$ while keeping $a$ and $k_{mp}$ fixed at moderate
678: values, a state of persistent non-infectious inflammation will follow
679: recurrent infection and a small zone of stable limit cycles.  (Varying
680: the other parameters, for small $\theta$ and large $w$, does not
681: result in Hopf bifurcations). In the case of small $w$ (sharp coupling
682: curve) and small $\theta$ (low threshold of activation), it is
683: possible that more than one fixed point satisfies the conditions for
684: FP 3.  In this case, curve~(\ref{fp6}) intersects line (\ref{fp2})
685: twice before dipping below zero.  On varying the parameter
686: combinations, recurrent infection is bistable with persistent
687: inflammation and healthy behavior is never supported. These cases are
688: discussed in greater detail in Appendix~\ref{couplingcurve}.  In our
689: numerical examples, we have considered moderate $w$ and large
690: $\theta$, which result in subcritical Hopf bifurcations.
691: 
692: 
693: \subsubsection{FP 4 and FP 5}
694: \label{appfp45} 
695: These fixed points are the solutions of the following equations,
696: \begin{eqnarray}
697: p&=&0 \\ m&=&1-(\frac{1}{k_{mp}k_{pm}p+l})\\ l&=&\frac{k_{lm}}{k_l}f(m)
698: \end{eqnarray}
699: This may be reduced to one transcendental equation
700: \begin{equation}
701: bf(m)=\frac{1}{1-m}\label{fp45}
702: \end{equation}   
703: Equation (\ref{fp45}) may have 0, 1 or 2 solutions depending on the
704: parameters. Approximate solutions can be found for (\ref{fp45}) by
705: substituting a piece-wise linear function such as $f(m)=0,
706: m<\theta-w$; $f(m)=1+\frac{(m-\theta)}{w}, \theta-w<m<\theta+w$;
707: $f(m)=2, m>\theta+w$. Using the above approximate function and
708: assuming that FP 4 and 5 occur in region $\theta-w<m<\theta+w$, we
709: arrive at the following expressions for FP 4 and 5.
710: \begin{eqnarray}
711: p&=&0\\ m&=&\frac{q\mp\sqrt{q^2-\frac{4w}{b}}}{2}\\ l&=&bf(m)
712: \end{eqnarray}
713: where $q\equiv 1-\theta+w$.
714: 
715: This gives $b>4w/q^2$ as a lower  limit on $b$ for the existence of FP
716: 4 and  5.  When  there are  two solutions, they  are formed  through a
717: saddle-node bifurcation as shown  in Fig.~\ref{fig:varylt}.  FP~4 is a
718: saddle point and FP~5 is a stable node.
719: 
720: The  eigenvalues of  FP  4 and  5  can be  found  by substituting  the
721: solutions     $m=m^0$    of     (\ref{fp45})    into     the    Jacobian
722: (\ref{jacobian}).       One      of      the       eigenvalues      is
723: $\lambda_1=k_{p}-k_{pm}m^0$  and the  other two  are functions  of the
724: parameters $k_l$,  $k_{lm}$, $\theta$ and  $w$. We note  that $k_{mp}$
725: does not appear in the  expressions for the position or eigenvalues of
726: FP 4 and 5.
727: 
728: 
729: \subsection{Regimes for Healthy Response}
730: \label{appbzone}
731: Trajectories can be interpreted as healthy when FP~3 supports unstable
732: oscillations as an unstable spiral or as a stable spiral surrounded by
733: an  unstable limit cycle.  The unstable  limit cycle  is lost  when it
734: undergoes a  non-generic homoclinic bifurcation when  it collides with
735: FP~1. The  regimes of the  various parameters (for fixed  $\theta$ and
736: $w$) which maintain healthy behavior are discussed below.
737: 
738: For healthy  behavior, $a$ should be  large enough so that  FP~3 is to
739: the     left    of     FP~5     in    Fig.~\ref{fig:nullclines}     or
740: $(1/a)<(q-\sqrt{q^2-4w/b})/2$. Increasing  $k_{mp}$ even over  a large
741: range, does not  make FP 3 unphysical, nor does  FP~3 lose its complex
742: eigenvalues - thus even  for very large $k_{mp}$, healthy oscillations
743: are  supported.  The  upper  limit for  $a$  and the  lower limit  for
744: $k_{mp}$  to maintain healthy  oscillations are  difficult to  find in
745: closed  form as these  variations result  in the  radius of  the limit
746: cycle  increasing and  oscillations being  lost through  a non-generic
747: homoclinic bifurcation.
748: 
749: The strength of the late immune response $b$ should not be too high or
750: too low.  If $b$ is too  high, then FP  3 has $p<0$ and  hence healthy
751: behavior cannot be supported. In order for the late immune response to
752: not  be too  high, we  require that  $p$ of  condition  (\ref{fp6}) be
753: positive so that FP 3  is not unphysical. Applying this to (\ref{fp6})
754: gives the condition
755: \begin{equation}
756: b<\frac{1}{1-\tanh(\theta/w)}.
757: \label{w1}
758: \end{equation}
759: If $b$ is  too low, FP 4 and  5 are not yet formed  and (\ref{fp6}) is
760: well above $p=0$ so FP 3  remains the global attractor and is a stable
761: spiral. To avoid  too low of a response,  we would like FP 4  and 5 to
762: exist so that unstable oscillations from FP 3 are eventually attracted
763: to  FP 5.   Thus  the  condition for  existence  for FP  4  and 5,  as
764: calculated above in \ref{appfp45}, is
765: \begin{equation}
766: b>\frac{4w}{(1-\theta+w)^2}.
767: \label{w2}
768: \end{equation}
769: 
770: 
771: \subsubsection{Effect of Varying the Coupling Curve}
772: \label{couplingcurve}
773: The  shape  of the  coupling  curve (\ref{coupling})  (i.e.~parameters
774: $\theta$  and $w$)  can alter  the  results.  We  used $\theta=1$  and
775: $w=0.5$ in  our analysis. Parameter $\theta$ sets  the threshold where
776: $l$  is activated  by $m$  and parameter  $w$ gives  the  steepness of
777: recruitment.
778: 
779: From numerical simulations,  we find that when $w$  is too small, FP~3
780: never undergoes a Hopf  bifurcation, and as $k_{lm}/k_l$ is increased,
781: recurrent  infection and  persistent inflammation  are  bistable (For
782: example, $w=0.005$ 
783: with 'healthy' parameter set in Fig.~\ref{fig:healthy}).  As discussed
784: in Appendix~\ref{appfp3},  small $w$ could result in a shallow enough
785: curve (\ref{fp6}) that intersects line  (\ref{fp2})  more than once.
786: The two fixed points replacing FP~3
787: arise in a saddle-node  bifurcation. The  stable  node is
788: bistable  with FP~5  implying that recurrent  infection  is bistable  with
789: persistent non-infectious inflammation. 
790: When $w$ is too large ($w=1$),
791: FP~3 is a stable spiral that becomes unphysical as $k_{lm}/k_l$ increases.
792: Thus the system supports recurrent infection followed by
793: persistent inflammation and healthy behavior does not exist.
794: In summary, for $w$ too small or large the subcritical Hopf
795: bifurcation of FP 3 (and hence healthy behavior) 
796: does not occur for any of the parameter combinations.
797: 
798: When $\theta$  is small ($\theta=0.1$), FP 3  undergoes a supercritical
799: Hopf  bifurcation as  $b$ is  varied  but this  cannot be  generically
800: interpreted as  healthy behavior as oscillations about  a stable limit
801: cycle  may  not always  take  $p$  below  threshold.  As  $\theta$  is
802: increased, subcritical  Hopf bifurcations  are possible. The  range of
803: parameter $b$,  given by (\ref{w1}) and  (\ref{w2}), where oscillatory
804: behavior is  observed also increases as $\theta$  is increased.  Thus,
805: large $\theta$ and moderate $w$  would maximize the region for healthy
806: response.
807: 
808: In  the   above  analysis,  we   have  looked  at   varying  parameter
809: combinations rather than individual  parameters for conciseness of the
810: discussion. Varying the parameters  such as $k_p$ and  $k_{pm}$ individually
811: does  not  change  the  bifurcation  diagrams  qualitatively  and  our
812: interpretations   of   the  effect   of   pathogen  susceptibility   remain
813: unchanged. Similarly, varying $k_{lm}$ and $k_l$ individually does not
814: change  our  conclusions about  the  strength  of  late response.  The
815: phase-space  analyzed is  restricted between  $0<m<1$  since condition
816: (\ref{fp6}) prevents  $m$ from increasing beyond one.  Using any other
817: function  such  as $1/(1+m)$  to  saturate  $m$  can change the upper
818: bound of $m$ but the same behavior is retained.
819: 
820: 
821: \bibliographystyle{jtb}  \bibliography{jtb}
822: 
823: 
824: \newpage
825: \begin{figure} 
826: \begin{center}
827: \includegraphics[angle=270,scale=0.4]{health}
828: \end{center}
829: \caption{Time courses of $p$ (black), $m$ (red) and $l$ (blue) for the {\it healthy} response.  We
830: show orbits  with the $p$ elimination threshold (solid line) and
831: without (dotted line). With the threshold (set at $p_0=0.0005$), when $p$ drops below $p_0$, $p$ is set to zero and $m$ and $l$ return to background values.  Without the threshold, the orbits spiral outwards.
832: Parameters used
833: are ${k_{pm}}=30, {k_p}=3, k_{mp}=25, {k_{lm}}=15, {k_l}=1$. Initial
834: conditions are $p(0)=0.01, m(0)=0.05, l(0)=0.539$.}
835: \label{fig:healthy}
836: \end{figure}
837: 
838: \begin{figure} 
839: \begin{center}
840: \includegraphics[angle=270,scale=0.4]{shock.ps}
841: \end{center}
842: \caption{In {\it persistent non-infectious inflammation},  $p$ (black) is
843: eliminated but $m$ (red) and $l$ (blue) remain elevated.
844: Parameters used are $k_{pm}=30, k_p=3, k_{mp}=25, k_{lm}=15,
845: k_l=1$. Initial conditions are $ p(0)=0.2, m(0)=0.05, l(0)=0.539$.}
846: \label{fig:shock}
847: \end{figure}
848: 
849: \begin{figure} 
850: \begin{center}
851: \includegraphics[angle=270,scale=0.4]{sevsepsis.ps}
852: \end{center}
853: \caption{In {\it persistent infectious inflammation}, $p$ (black) cannot be
854: eliminated and $m$ (red) and $l$ (blue) remain elevated (as in
855: severe sepsis). Parameters used are
856: $k_{pm}=3, k_p=3, k_{mp}=25, k_{lm}=15, k_l=1$. Initial conditions are
857: $p(0)=0.01, m(0)=0.05, l(0)=0.539$.}
858: \label{fig:sepsis}
859: \end{figure}
860: 
861: \begin{figure} 
862: \begin{center}
863: \includegraphics[angle=270,scale=0.4]{recurr.ps}
864: \end{center}
865: \caption{In {\it recurrent infection}, $p$ (black) is low and $m$ (red) and $l$ (blue) remain above background values but not very
866: high (as in a low-grade infection). Parameters used are
867: $k_{pm}=30, k_p=3, k_{mp}=25, k_{lm}=5, k_l=1$. Initial conditions are
868: $p(0)=0.01, m(0)=0.05, l(0)=0.179$.}
869: \label{fig:recurr}
870: \end{figure}
871: 
872: \begin{figure} 
873: \begin{center}
874: \includegraphics[angle=270,scale=0.4]{sevimmsupp.ps}
875: \end{center}
876: \caption{In {\it severe immuno-deficiency},  $p$ (black) grows to saturation
877: and $m$ (red) and $l$ (blue) remains low or absent.  Parameters are
878: $k_{pm}=30, k_p=3, k_{mp}=0.4, k_{lm}=15, k_l=1, p(0)=0.01, m(0)=0.05, l(0)=0.539$.}
879: \label{fig:suppress}
880: \end{figure}
881: 
882: \begin{figure} 
883: \begin{center}
884: \includegraphics[angle=270,scale=0.4]{phaseplane.ps}
885: \end{center}
886: \caption{Projection of fixed-point conditions onto the $p$-$m$ plane.
887: Projected directions of field-flow are also
888: marked. The red curves correspond to conditions $p=0$ and (\ref{fp2})
889: and the black curves correspond to conditions $m=0$ and 
890: (\ref{fp6}). The fixed points are at the intersections of black and
891: red curves  and the direction of flow along a curve reverses at  
892: fixed points. The projection of the phase-space for a typical set of
893: parameters is shown.  FP~1 and FP~4 are saddles. FP~3 is a spiral
894: which is stable or unstable based on the actual values of the
895: parameters. FP~5 is a stable fixed point. FP~2 (not shown) is also a
896: saddle in this case.}
897: \label{fig:nullclines}
898: \end{figure}
899: 
900: 
901: \begin{figure} 
902: \begin{center}
903: \includegraphics[angle=270,scale=0.4]{kpmauto.ps}
904: \end{center}
905: \caption{Bifurcation plot showing outcome as $k_{pm}$ and hence
906:   $a=k_{pm}/k_p$ is increased. For low $k_{pm}$, FP 3 interpreted as
907:   {\em persistent infectious 
908:   inflammation} is the only fixed 
909: point of the system. As $k_{pm}$ is increased to
910: $\approx 2.7$, FP 5 ({\em persistent infectious inflammation})
911: becomes the attractor of the system. FP 3 becomes unphysical and
912: moves leftward in the $p$-$m$ plane (see inset). At $k_{pm}
913: \approx 11$, FP 3 undergoes a transcritical bifurcation with FP 4,
914: becoming physical again. It is an unstable spiral, oscillations
915: around which are interpreted as {\em healthy}.  On further increasing
916: $k_{pm}$, FP 3 undergoes a subcritical Hopf
917: bifurcation and is surrounded by an unstable limit-cycle, oscillations
918: around which are interpreted as {\em healthy}. Trajectories within the
919: limit-cycle, which spiral into FP 3, are interpreted as {\em recurrent
920: infection}. {\it Healthy} behavior is lost to a homoclinic bifurcation at
921: $k_{pm} \approx 57$. For large $k_{pm}$, FP 3 and FP 5 are
922: bistable (i.e.~{\em recurrent infection} and {\em persistent
923: non-infectious inflammation} are possible outcomes of the
924: system). (Inset : In the $p$-$m$ plane, line (\ref{fp2}) sweeps across
925:   the plane causing the above bifurcations.) } 
926: \label{fig:varyp}
927: \end{figure} 
928: 
929: 
930: 
931: 
932: \begin{figure} 
933: \begin{center}
934: \includegraphics[angle=270,scale=0.4]{kmpauto.ps}
935: \end{center}
936: \caption{Bifurcation plot showing outcome as $k_{mp}$ is
937: increased. For very low $k_{mp}(<0.5)$, FP~2 is 
938: stable and is interpreted as the {\em severely immuno-deficient}
939: state.  As $k_{mp}$ is increased, FP~3 becomes stable through a
940: transcritical bifurcation and {\em recurrent infection} is a possible
941: outcome of the system.  As $k_{mp}$ is increased further FP~3 undergoes
942: a subcritical Hopf bifurcation ($k_{mp} \approx 22$). Oscillations
943: around the unstable limit cycle surrounding FP~3 are interpreted as
944: {\em healthy} and are lost through a homoclinic bifurcation
945: ($k_{mp} \approx 9$). Trajectories within the limit-cycle spiral into
946: FP~3 and are interpreted as {\em recurrent infection}.  Beyond
947: the Hopf bifurcation ($k_{mp} > 22$), FP~3 remains an unstable
948: spiral, up to very large $k_{mp}$ ($k_{mp} \approx  2000$) and
949: oscillations around FP 3 are interpreted as {\em healthy}. On
950: varying $k_{mp}$, only the position and eigenvalues of FP 3 are
951: changed. FP 5 remains unaffected which is interpreted as {\em
952:   persistent infectious 
953: inflammation} and remains a possible outcome of the system. If curve
954: (\ref{fp6}) is shallow enough ($b$ was small enough) so that FP 5 is
955: not yet created, then FP 3 remains the stable global attractor through
956: changes in $k_{mp}$.  (Inset :  In the $p$-$m$ plane, FP 3 varies in
957: position and  
958: stability as (\ref{fp6}) varies, causing the above
959: bifurcations. FP~5 remains unaffected.)}
960: \label{fig:varyer}
961: \end{figure}
962:   
963: 
964: \begin{figure} 
965: \begin{center}
966: \includegraphics[angle=270,scale=0.4]{klmauto.ps}
967: \end{center}
968: \caption{Bifurcation plot showing outcome as $k_{lm}$  or $b=k_{lm}/k_l$
969: is increased. For low $k_{lm}$, FP 3 is a stable
970: spiral representing {\em recurrent infection}. As $k_{lm}$ is increased, FP
971: 4 and 5 are created through a saddle-node bifurcation at $k_{lm} \approx
972: 7$ and the {\em persistent non-infectious inflammation} is now a
973: possible outcome. At  $k_{lm}\approx 14$, FP 3 undergoes a
974: subcritical Hopf bifurcation and is surrounded
975: by an unstable limit cycle. The oscillations around the unstable limit
976: cycle are interpreted as {\em healthy} behavior and trajectories
977: inside the limit cycle spiral into FP 3 and are interpreted as {\em
978: recurrent infection}. {\it Healthy} behavior is lost  at $k_{lm} \approx 10$
979: due to a homoclinic bifurcation. Beyond the Hopf bifurcation, {\em
980: healthy} oscillations are still supported by FP 3 which is now an
981: unstable spiral. At  $k_{lm} \approx 21$, FP 3 meets FP 4 and
982: becomes unphysical through a transcritical bifurcation (as its $p<0$). Beyond this point, FP 5 is the only stable attractor and {\em persistent
983: non-infectious inflammation} is the only possible outcome of the
984: system.  (Inset :  In $p$-$m$ plane,
985:  (\ref{fp6}) descends as $k_{lm}$
986: is increased causing the above bifurcations.)}
987: \label{fig:varylt}
988: \end{figure}
989: 
990: 
991: \end{document}
992: