Você está na página 1de 32

Authors Accepted Manuscript

What can be learned from a chaotic cancer model? C. Letellier, F. Denis, L.A. Aguirre

PII: DOI: Reference:

S0022-5193(13)00022-2 http://dx.doi.org/10.1016/j.jtbi.2013.01.003 YJTBI7193

www.elsevier.com/locate/yjtbi

To appear in:

Journal of Theoretical Biology

Received date: Revised date: Accepted date:

17 March 2012 20 November 2012 5 January 2013

Cite this article as: C. Letellier, F. Denis and L.A. Aguirre, What can be learned from a chaotic cancer model?, Journal of Theoretical Biology, http://dx.doi.org/10.1016/ j.jtbi.2013.01.003 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting galley proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

What can be learned from a chaotic cancer model?


Christophe Letellier
CORIA Universit e de Rouen, Av. de lUniversit e, BP 12, F-76801 Saint-Etienne du Rouvray cedex, France

Fabrice Denis
Centre Jean Bernard, 9 rue Beauverger, 72000 Le Mans, France

Luis A. Aguirre
Universidade Federeal de Minas Gerais Av. Ant onio Carlos 6627, 31270-901 Belo Horizonte MG, Brazil

Abstract A simple model of three competing cell populations (host, immune and tumor cells) is revisited by using a topological analysis and computing observability coecients. Our aim is to show that a non conventional analysis might suggest new trends in understanding the interactions of some tumor cells and their environment. The action of some parameter values on the resulting dynamics is investigated. Our results are related to some clinical features, suggesting that this model thus captures relevant phenomena to cell interactions. Keywords: Cancer model, Chaos, Observability, Therapy 1. Introduction Cancer results from a proliferation of tumor cells which invade nearby parts of the body. Tumors are microcosms of evolution with a mosaic of mutant cells that compete for space and resources, evade predation by the immune system and can even cooperate to disperse and colonize new organs

Corresponding Author Email address: Christophe.Letellier@coria.fr (Christophe Letellier)

Preprint submitted to Journal of Theoretical Biology

January 9, 2013

[1]. Unfortunately, interactions between cell populations like host cells, effector immune cells, tumor cells, endothelial cells, or others, are not well understood and the key parameters remain to be identied. Interactions between dierent cell populations in neoplasm not only induce genetic unstability of tumor cells but also unpredictable tumor growth [2]. Due to this, carcinogenesis population-based models may be used to better understand tumor dynamics and responses to treatment. Competition and predation between dierent types of cells in tumor exist in the form of resource consumption or in the interaction of neoplastic cells with the immune system, for instance [1]. Neoplastic cells can indeed have direct negative eects on each other or on immune and stromal cells through soluble factors that may be targeted by non tumor cell treatment [3, 4]. Carcinogenesis models based on Lotka-Volterra competition equations dene conditions under which cancerous cells might be driven up to extinction [5]. By reducing the negative competitive eects of cancer cells on normal cells and increasing the negative competitive eects of normal cells on cancer cells, the number of cancer cells which can be supported in a tissue can be reduced [6]. Such models can help to identify the parameters which should be targeted by therapies and to design the most eective strategy for drug treatment regimes [7, 8]. Lotka-Volterra equations are also known to produce chaotic behaviors in the case of prey-predator model [9, 10] as well as for competing species [11, 12]. Among the models based on Lotka-Volterra equations, a model describing the interactions between three cell populations, one being tumor cells, produces chaotic behaviour [13]. Its parameter values were chosen to match with some biological evidences [14]. This model could be thus considered as being qualitatively validated with experimental data. Identifying accurately the parameter values of such a model remains an open problem since there are rather dicult to deduce from biological experiments. This paper investigates the model proposed in [14], which describes the interactions among three cell populations: host cells, eector immune cells and tumor cells. This model was chosen for its chaotic dynamics [13] which has numerous analogies to clinical evidences [15, 16]. It will be shown that an extended dynamical analysis can provide insights about cellular interactions in agreement with some experimental and clinical evidences. The subsequent part of this paper is organized as follows. Section 2 briey describes the cancer model and how the parameter values were chosen for the analysis. Fixed points of the model are also discussed for these parameter values. The determination of the best variable to measure for investigating the underlying 2

dynamics is discussed in Section 3. A dynamical analysis is performed in Section 4. Bifurcation diagrams are described and a template which synthetizes the topological structure of the chaotic attractor is provided. Section 5 gives a discussion and some conclusions. 2. Model equations and xed points Let us consider a simple model for tumor growth based on Lotka-Volterra equations (see [17, 14] for instance). This model describes the interactions that take place in a single tumor-site compartment, between host cells H , eector immune cells E and tumor cells T . This model is devoted to a site in the neighborhood of a homogeneous tumor. The interactions of the immune cells with the tumor cells are described as in [17]. Since the nature of the growth law terms does not aect signicantly the tumor growth, the logistic growth term was retained [14]. This model is therefore a qualitative model. In spite of this, this model can describe some important aspects of the interactions between the three cell populations: H , E and T . The model is [14]: H = H 1 H 1 13 T H 1 2 T E 2 E = 23 T E E (1) T + 2 T = 3 T 1 32 T E , 31 T H T 3 where the rst equation corresponds to the rate of change of host cells growing according to a logistic function with a biotic capacity 1 . It is here supposed that tumor cells proliferate more quickly than host cells, a feature obtained using 3 > 1 . Tumor cells inhibit host cells at the rate 13 . The second equation corresponds to the rate of change of eector immune cells where the rst term represents the stimulation of the immunatory system by tumor cells with antigens specic to tumor cells. The rate of recognition of tumor cells by the immune system depends on the antigenicity of tumor cells. Since this recognition process is very complex, it is supposed to keep the model quite simple, that the stimulation of the immune system directly depends on the number of tumor cells with positive constant 2 and 2 . Eector immune cells are inactivated by tumor cells at the rate 23 : they naturally 2 . The third equation provides the growth rate of tumor die at the rate 3

cells. The rst term has the form of a logistic function which governs the tumor cells when they are alone: the growth rate is thus 3 and the biotic capacity 3 . The competition between host cells and tumor cells is described by a degradation of tumor cells according to the term 31 T H . Tumor cells are killed by eector immune cells at the rate 32 . All parameters of the cancer model (1) are supposed to be positive. The cancer model (1) only diers from the one proposed by de Pillis & Radunskaya [14] by the fact that it does not have a constant inux of eector immune cells. This was justied by Itik and Banks [13] as follows. It was supposed that eector immune cells are cytotoxic T-cells produced as naive cells, that is, as not being able to produce any response to tumor cells unless they are activated by antigen presenting cells via some major histocompatibility complex MHC-I (found in all nucleated cells, that is, not red blood cells) and HC-II (only found in antigen presenting cells) pathways in the presence of tumor specic antigens. Neglecting the constant inux of eector immune cells was also proposed by Kirschner and Panetta [18, 19]. After renormalization of variables H , E , and T , system (1) can be rewritten in the form x = 1 x(1 x) 13 xz 2 yz (2) 23 yz 2 y y = 1 + z z = z (1 z ) xz 32 yz, where we have the correspondence (x, y, z ) (H, E, T ), that is, x designates the normalized population of host cells, y the population of the eector immune cells, and z the tumor cells. In this study, 3 and 31 are set to 1, as in [13]. The set of equations in (2) denes a model for cells in competition as can be seen on the uence graph shown in Fig. 1. Contrary to what is commonly observed in prey-predator models, there is no particular prey. None is promoting the growth of another one and all the interactions between the dierent populations are repressing. This dynamics thus corresponds to highly competitive species. There is no interaction between host and eector immune cells. Consequently, tumor cells can be viewed as generalist competitors while the two others are specialist competitors, to use a terminology often used in ecology.

host cells + x y z

+ + Effector Tumor cells immune cells

Figure 1: Fluence graph for the cancer model (2). Solid (dashed) lines represents linear (nonlinear) interactions. The sign describes if the interactions promote (+) or repress (-) the growth of the species.

System (2)

will be investigated for the following set of parameters: 1 = 0.6 13 = 1.5 2 = 4.5 23 = 0.2 growth rate of host cells host cell killing rate by tumor cells growth rate of eector immune cells eector immune cell inhibiting rate by tumor cells 2 = 0.5 eector immune cell mortality 32 = 2.5 eector immune cell killing rate by tumor cells ,

(3)

when not specied otherwise. These values were considered by Itik and Banks [13] for producing a chaotic attractor (Fig. 2) which is structured around ve of the seven real xed points S0 = x0 = 0 y0 = 0 z0 = 0 , x2 = 0 y2 = 0 z2 = 1 , S1 = x1 = 1 y1 = 0 z1 = 0 , x3 = 0 y3 = 0.347000 z3 = 0.132503 ,

S2 =

S3 =

S4 =

x4 = 0.668743 y4 = 0.079502 z4 = 0.132503 , and S6 =

x5 = 0 S5 = y5 = 7.147000 z5 = 18.867497 x6 = 46.168742 y6 = 11.320498 z6 = 18.867497 .

Two of them (S5 and S6 ) are irrelevant for the resulting dynamics since they have negative coordinates (negative populations are not dened and, consequently, the dynamics must take place in the positive octant). The xed point at the origin corresponds to a situtation where there is no cell at all. Its eigenvalues are 0 = (0.6, 0.5, 1), thus corresponding to a saddle. This means that when x = z = 0, the eector immune would quickly disappear according to the negative eigenvalue. With x = z = 0, the cancer model (2) reduces to =0 x y = 2 y (4) z = 0. As observed in dynamics of competing species, the boundaries of the positive octant cannot be crossed by the trajectory. Moreover, the attractor is conned in a triedron having the origin asapex. For an extended analysis of such constraints in another model for species in competition, see [20]. Fixed point S1 has eigenvalues 1 = (0.6, 0.5, 0): this is therefore a marginally stable node. When y = z = 0, that is, when there are only host cells, the dynamics is governed by a logistic equation (as introduced by Verhulst [21]) and the population converges to its maximum value (x = 1). Point S1 is not distinguished from S0 in the y -z plane projection shown in Fig. 2. Fixed point S2 corresponds to a situation where only tumor cells are present. The eigenvalues are 2 = (0.9, 1.55, 1). The point is thus of a saddle type. By setting x = y = 0 in system (2), the dynamics is reduced to =0 x y =0 (5) z = z (1 z ) , that is, tumor cells are governed by a logistic function and the population converges toward z = 1. 6

Effector immune cells y

0,8

0,6

0,4
S3

0,2
S4

S0-S1

S2

0,2

0,4

0,6

0,8

Tumor cells z

Figure 2: Chaotic attractor solution to the cancer model (2) plotted with the xed points and some of their eigenvectors. Parameter values as in Eq. (3).

The xed point S3 is associated with a situation where only eector immune and tumor cells co-exist. This xed point can be considered as being not surrounded by the attractor. Its eigenvalues 3 = (0.0660.613i, 0.401) correspond to a saddle-focus. The stable manifold of this xed point belongs to the boundary of the attraction basin of the chaotic attractor. Point S4 is associated with the co-existence of the three type of cells; its eigenvalues 4 = (0.0403 0.235i, 0.614) also dene a saddle-focus. The xed point S4 , which is surrounded by the attractor, has a 2D unstable manifold, in that respect, point S4 diers from point S3 which has a 1D unstable manifold. It is interesting to notice that the local dinamics around S4 display oscillations with natural frequency f = 0.235/2 = 0.037 Hz, which is very close to the pseudo-period of the underlying dynamics associated with a frequency f = 0.039 Hz. The two saddle-foci S3 and S4 structure the chaotic attractor as the R ossler attractor is structured by two saddle-foci [22]. Such a conguration has a deep impact on the dynamics which will be favourably compared to the R ossler attractor as described below. 3. Observability Coecients When a single scalar time series s = h(x) from a dynamical system = f(x) with x Rm is measured, it is not guaranteed that any state x x Rm can be reconstructed (observed) in the space Rm spanned by the 7

reconstructed coordinates X Rn (n m) which are typically delay or derivative coordinates [23]. The determination of whether the system is observable (in that case, any state x Rm can be unambigously reconstructed from Rn (X)) is based on the the observability matrix Os (x) = where L0 f h(x) x Lj f h(x)
1 dLn L1 f h(x) f ... (x) x ds T

(6)

1 Lj f h(x) f(x) , (7) = x T and where L0 f h(x) = h(x). The system is said to be fully observable if O O is full rank. The observability matrix corresponds to the Jacobian matrix of the coordinate transformation between the original state space Rm (x) and the reconstructed space Rn (X) when derivative coordinates are used [24]. Typically the singular set associated with states x Rm which cannot be T Os ] = 0. When this determinant is difreconstructed is dened by Det[Os m ferent from zero for all x R (x), the system is said to be observable and there is a global dieomorphism between the original state space Rm (x) and the reconstructed state space Rn (X). To avoid a yes-or-no classication of observability, the following coecient can be estimated for monovariable embeddings [24, 25] T Os , x(t)] | | min [Os s (x) = , (8) T | max [Os Os , x(t)] | T T Os , x(t)] indicates the maximum eigenvalue of matrix Os Os where max [Os T estimated at point x(t) (likewise for min) and () indicates the transpose. Then 0 s (x) 1, and the lower bound is reached when the system is unobservable at point x. The observability coecients can be computed and averaged along an orbit thus

s =

1 T

s x(t),
t=0

(9)

where T is the nal time considered. The case of multivariate embeddings was investigated in [26]. Some symbolic observability coecients were also introduced [27]. A procedure which only requires data has been put forward in [28]. 8

For instance, when applied to the R ossler system [29] x = y z y = x + ay z = b + z ( x c) ,

(10)

2 2 2 the symbolic observability coecients x = 0.88, y = 1.00, and z = 0.44. These coecients mean that variable-y provides a global dieomorphism between the original R3 (x, y, z ) and reconstructed R3 (X = y, Y = y, Z = y ) spaces, that is, the best observability which can be expected. Contrary 2 to this, variable z is a rather poor variable (z = 0.44) in what concerns observability and many diculties are encountered when dynamical analyses are performed using this variable [30]. Dierential embeddings induced by these three variables provide intuitive insight for such features (Fig. 3). Thus, the dierential embedding induced by the poorest variable z presents a small domain near the plane z = 0 where all the dynamics is squeezed and where distinguishing dierent states or dierent trajectories is rather dicult. There is no such a feature in the dierential embedding induced by the two other variables. A squeezed region in the reconstructed embedding is thus a signature of poor of observability [28].
4
4

4 2
2

-2
-2

-2 -4

-4

-6 -4
-2 0 2 4 6

-4

-6

-4

-2

X=x

X=y

X=z

(a) Variable x

(b) Variable y

(c) Variable z

Figure 3: Dierential embeddings induced by the three variables of the R ossler system. Parameter values: a = 0.398, b = 2, and c = 4.

In the case of the cancer model (2), the dierential embedding induced by each of the three variables are shown in Fig. 4. The embedding induced by variable x is the single one which does not show any strongly squeezed domain. For instance, in the lower left part of the y -induced embedding, 9

all revolutions on the attractor pass through a squeezed region where it is quite dicult to distinguish them from the others. There is a similar domain located in the upper left part of the z -induced embedding (Fig. 4c). From these embeddings, it appears that variable x presents the best observable of the dynamics.
0,4

0,1
0,3

0,1

0,05

0,2

dx / dt

0,1

dz / dt
0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8

dy /dt

-0,1

-0,05
-0,1

-0,2

-0,1
-0,2

-0,3

0,2

0,4

0,6

0,8

0,1

0,2

0,3

0,4

0,5

0,6

0,7

Host cells x

Cellules immunitaires effectrices y

Cellules tumorales z

(a) Host Cells

(b) Eector immune cells

(c) Tumor cells

Figure 4: Dierential embedding induced by each of the three variables of the cancer model (2). Parameter values: 1 = 0.518 and as in Eq. (3) otherwise.

Computing the observability coecients as dened in Eq. (9) where the underline has been omitted leads to x = 0.021 > z = 0.015 > y = 0.001 , thus conrming that variable x is the best observable. Variable y is the poorest. The main message of this analysis is that the dynamics of the cancer model is observed with a greater reliability from the population of host cells (variable x) rather than from eector immune or tumor cells (variable y and z , respectively). Tumor cells provide an observability of the dynamics that is only slightly less than that provided by host cells, but eector immune cells are associated with a very poor observability (y is smaller than x and z by one order of magnitude). From the therapy point of view, and according to the interplay between the observability coecients and the quality of synchronization [31] or of control [30], these results suggest to control the dynamics by acting on the population of host cells. Acting on the population of tumor cells by destroying them for instance is also ecient, but trying an action on the eector immune cells (e.g. cancer vaccine) would be rather inecient as conrmed by [32]. For instance, 82 patients with advanced prostate cancer 10

who received a vaccine did not dier much in terms of time-to-progression (p = 0.052) from 45 patients who received a placebo [33]. A slight benet in survival (25.9 months with vaccine versus 21.4 months with placebo, p = 0.010) was observed in that study. But this was not satistically signicant (19 months with vaccine versus 15.7 months with placebo) in another study [34]. The relative irrelevance of the contribution of immune cells in the dynamics of this model will be also conrmed in the dynamical study discussed in the next section. Let us now assume that we are able to observe simultaneously two populations among the three considered in this cancer model. We have thus three dierent possibilities: to measure i) x and y , ii) y and z , and iii) x and z . For each case, there are two possibilities to produce a three dimensional reconstructed space: adding the rst derivative of the rst i) or of the second population ii) observed. Using the procedure for multivariate observability analysis discussed in [26], we obtained R3 (x, x, y ): x2 y = 0.29; R3 (x, y, y ): y2 x = 0.08; R3 (x, z, z ): z 2 x = 0.08. R3 (y, z, z ): z 2 y = 0.01; R3 (x, x, z ): x2 z = 0.00 (non observable); z ): y2 z = 0.00 (non observable). R3 (y, y, According to these observability coecients, observing the tumor cells with another population whose rst derivative is used as a third variable does not provide a reliable reconstructed space (y2 z = x2 z = 0), that is, a space where any state is easily observed. When the evolution of tumor cells is derivated once and the population of eective immune cells is observed is added does not improve too much the observability (z 2 y = 0.01). Contrary to this, the ) is conobservability is slightly better when the reconstructed space R3 (x, z, z sidered (z 2 x = 0.08). The best situation is to observe host cells with eector immune cells and to derive variable x once. Roughly, such a derivation allows to get information about tumor cells and the information about eector immune cells is directly obtained by measurements. Derivating variable y is far less ecient due to the low observability associated with that variable. 11

Working with the reconstructed space R3 (x, x, y ) would be the best strategy when two variables are recorded. From that point of view, acting simultaneously on the host cells and the eector immune cells could be, at least from a dynamical point of view, a possible therapy to reduce the population of tumor cells. 4. A dynamical analysis 4.1. Bifurcation analysis Since only xed points were investigated in [14] and a single set of parameter values was considered in [13], we choose to start our analysis by varying one of the parameter of the cancer model. From the observability coecients analysis, acting on the host cells appears to be the best strategy to act on the cancer dynamics. We thus started by varying the growth rate 1 of the host cells. The so-obtained bifurcation diagram (Fig. 5a) starts with a period-1 limit cycle. It is followed by a period-doubling cascade. The chaotic attractor observed beyond the accumulation point (at which the orbit 2 occurs) must therefore be characterized by a smooth unimodal map since a period-doubling cascade is the universal route to chaos in such a map [35, 36]. Many periodic windows can then be observed. The uctuation range of the population of host cells (Fig. 5a) and tumor cells (Fig. 5b) increases when the host cell growth rate is increased up to 1 0.54: between the period-1 limit cycle and this largest attractor, many periodic orbits were created. Beyond this value, the pruning of periodic orbits dominates over the creation of periodic orbits and the uctuation range of the populations decreases. This is the signature of an antimonotonicity [37] which requires the occurence of at least a third branch in the rst-return map. A simple sequence of reverse bifurcations as observed in bubbling [38] would lead to a completely dierent bifurcation diagram. To be correctly interpreted from the population point of view, the bifurcation diagram has to be computed in a slightly dierent way than commonly done. Indeed, the required information about a population is twofold: we would like to know its range of variability, that is, its smallest and largest values and whether the extrema of the oscillations of the population studied can uctuate over this range. Consequently, for computing the bifurcation diagram, the minimal and maximal values at each oscillations were retained. Note that the dynamical regimes are the same in both original and modied bifurcation diagrams in spite of dierent ordinates used to compute them: the 12

1 0,9 Host cells xn 0,8 0,7 0,6 0,5 0,4 0,7 0,8 0,9 0,6 Host cell growth rate 1 (a) Common bifurcation diagram: host cells versus 1 0,5 1

1 0,8

Host cells xn

0,6 0,4 0,2 0 0,4 0,5 0,6 0,7 0,8 0,9 1 1,1 1,2 1,3 1,4

Host cell growth rate 1 (b) Modied bifurcation diagram: host cells versus 1

Tumor cells zn

0,8 0,6 0,4 0,2 0 0,4 0,5 0,6 0,7 0,8 0,9 1 1,1 1,2 1,3 1,4

Host cell growth rate 1 (c) Modied bifurcation diagram: tumor cells versus 1

Figure 5: Bifurcation diagrams of the cancer model (2) versus the host cell growth rate. Other parameter values as in Eq. (3).

13

modied bifurcation diagram is only easier to interpret from the uctuations of populations point of view. The modied bifurcation diagrams (Fig. 5b) reveals that when the growth rate of host cells is increased, the population of host cells can take larger maximal values (near to 1) but the corresponding minima becomes very low (near to 0). In fact, the cycle-to-cycle variability is reduced by increasing the growth rate of host cells in the sense that the population of host cells remains near its maximum value for a quite long period but can decrease very quickly to zero (Fig. 6b). Compare to what is observed for moderate growth rate 1 values (Fig. 6a), the oscillations are more rare (host cells remain at their maximum for about 1000 units of time, that is, far longer than with a moderate 1 -value for which each oscillation takes about 3 or 4 units of time). Thus, when the host cell growth rate is large enough, host cells become suciently strong to resist to the tumor cells for a certain period of time. Unfortunately, there is still a possibility that tumor cells proliferate: when this happens, this is very quickly and almost all host cells are destroyed, thus leading to a deleterious dynamics. A strategy which only stimulates the growth of host cells would turn rst to an apparent good status but would also put the patient into conditions for presenting a fast tumoral growth, as biologically observed when tumor and host cells expressed similar growth factor receptors such as EGFR (epithelial growth factor receptor). Acting preferentially on host cells rather than on tumor cells could be an interesting strategy as it can be observed sometimes in radiotherapy [39] and in interactions between cancer and host cells [40, 41]. Acting on host cells has also a natural interest because they mutate far less often than tumor cells. Nevertheless, another action should be applied to better control the growth of tumor cells. This would suggest a combined therapy. More commonly, the population of host cells increases with the corresponding growth rate. For relatively low growth rates, the development of the host cells contributes also to develop tumor cells, a fact which is known from biological data [40]. One can remark that tumor cells are developed with an increase of the amplitude of the oscillations presented by all the populations. These uctuations result from the strong competition between the two populations, when one is large, the other is small, and vice versa (Figs. 6). Tumor cells thus destabilize patients health by inducing some uctuations in the population of host cells. When the growth rate is large enough, host cells hinder the tumor cells from becoming too numerous for a quite long period of time. Fluctuations are thus more rare and the popula14

tion of tumor cells decreases (right part of the bifurcation diagrams shown in Figs. 5). Nevetheless, sometimes, tumor cells succeeds in proliferating and are thus able to kill most of host cells, inducing a rather deleterious dynamics. A second bifurcation diagram was computed by varying the rate 13 with which tumor cells kill host cells. Decreasing 13 has a rather similar eect as increasing the growth rate 1 of host cells, that is, to promote the growth of the host cells population. For each parameter which inuences the dynamics, the corresponding bifurcation diagram is roughly similar to those shown in Figs. 5. This has to be interpreted with the fact that the parameter space of a system can be viewed as made of shells of equivalent behaviours, more or less as an onion is made. Another bifurcation diagram was computed by varying the rate 32 with which eector immune cells kill tumor cells. No bifurcation was observed over the range of 32 -values (Fig. 7). This means that there is no impact on the dynamics played by this nonlinear term, that is, by the interaction between eector immune cells and tumor cells. We observed that setting 32 to zero leads to an ejection of the trajectory to innity. The chaotic attractor therefore exists only when 32 > 0 but does not depend on the value of that parameter. The nonlinear term 32 yz is thus of a very limited interest from the therapeutic point of view, as conrmed by many negative therapeutic protocols in immunotherapy [42, 43]. 4.2. Topological analysis The structure of the chaotic attractor produced by this cancer model will now be analyzed for parameter values corresponding to a unimodal attractor, that is, an attractor characterized by a unimodal map. Such a situation occurs slightly before the 1 -value for which the largest uctuations of the populations are observed in the bifurcation diagram (Fig. 5). There are many other parameter values for which an equivalent attractor is observed. In fact, there is a domain of the parameter space associated with such an attractor. Dierent sets of parameter values could be associated with dierent patients or with a patients status at dierent moment of his disease. The attractor here investigated is selected from topological criteria because it will help to identify the general representative structure of most of the attractors produced by this system, as discussed below. The third branch of the rst-return map to a Poincar e section occurs for 1 = 0.518. Variable x was chosen to perform the topological analysis for the reasons discussed in Sec. 3. 15

1 0,8

Host cells x

0,6 0,4 0,2 0 0 0,8 10 20 30 40 50

Tumor cells z

0,6 0,4 0,2 0 0 10 20 30 40 50

Adimensional time t (a) 1 = 0.54


1

Host cells x

0,8 0,6 0,4 0,2 0 0 1 2000 4000 6000 8000 10000 12000 14000

Tumor cells z

0,8 0,6 0,4 0,2 0 0 2000 4000 6000 8000 10000 12000 14000

Adimensional time t (b) 1 = 1.40


Figure 6: Time evolution of host and tumor cells for two values of the host cell growth rate 1 . Other parameter values as in Eq. (3).

16

0,2 0,175

Tumor cells zn

0,15 0,125 0,1 0,075 0,05 0,025 0 1 2 3 4 5

Tumor cell killing rate 32 by effector immune cells

Figure 7: Bifurcation diagram of the cancer model (2) versus the tumor cell killing rate by the eector immune cells. Other parameter values as in Eq. (3).

Using the space reconstructed from this variable (Fig. 8a) it is possible to get a regular plane projection of couples of periodic orbits (regular meaning that there are no more than two segments crossings at the same point). The differential embedding induced by variable x reveals a rather simple structure, which ressembles to the structure of the spiral R ossler attractor [22]. This is conrmed by a rst-return map (Fig. 8b) to the Poincar e section dened as P = (xn , x n ) R2 | x n = 0, x n < 0 . (11) There is a slightly layered structure in the decreasing branch which could induce a third branch in the template. Nevertheless these two decreasing branches will be approximated by a single one in order to yield a fair representation of the attractor structure using a template. The increasing branch is associated with an even local torsion (preserving order branch) and the decreasing branch with an odd local torsion (reversing order branch) [22]. This means that the increasing (decreasing) branch presents an even (odd) number of -twists. These properties are commonly observed after a period-doubling cascade necessarily induced by a smooth unimodal map [22]. From the bifurcation diagram, there is a strong consequence of the layered structure. Indeed, the second decreasing branch would cross the rst increasing branch around the maximum of the map (Fig. 8b), a feature that would break the determinism of the system. This is avoided by a series of 17

0,1

0,9

0,85

0,05
0,8

dx / dt

xn+1
0,75 0,7 0,65

-0,05

-0,1 0 0,2 0,4 0,6 0,8

0,65

0,7

0,75

Host cells x

xn

0,8

0,85

0,9

(a) Chaotic attractor

(b) First-return map

Figure 8: Chaotic behavior produced by the cancer model (2) characterized by a smooth unimodal map. Parameter values: 1 = 0.518 and as in Eq. (3) for others.

reverse bifurcations when 1 is increased beyond 0.518. These bifurcations induce a pruning of the unstable periodic orbits (similar features are induced by a homoclinic orbit in the R ossler system [22]). The pruning is associated with the existence of a third (increasing) monotonic branch in the rst-return map, that is, of a second critical point (see Fig. 11 as discussed later). Concomitant to this, new periodic orbits are also created. The dynamics is thus multimodal with an inevitable antimonotonicity [37]. A symbolic dynamics is said to be complete when all symbolic sequences are realized as periodic orbits within the attractor. Since the crisis which avoids the intersection between the two decreasing branches occurs quite far from the completeness of the symbolic dynamics built on the two symbols 0 and 1, many periodic sequences built with these two symbols are not found in the attractor shown in Fig. 8a. For instance, orbits with period less than seven are limited to (1) (10) (101111) (10111) (100) (100101) . (1011) (101110) (10110) (101)

This can be also viewed in the rst-return map (Fig. 8b) where the increasing branch remains far from the rst bisecting line (in smooth unimodal maps as the logistic function or the rst-return map to a Poincar e section of the 18

R ossler system, the completeness of the symbolic dynamics is reached when the increasing branch touches the rst bisecting line, see for instance [22]). In order to determine the branched manifold which can be associated with the chaotic attractor shown in Fig. 8a, we present (as examples) two knots made of the two orbits (1) and (10) (Fig. 9a), and of orbits (10) and (10110) (Fig. 9b), respectively. The oriented crossings are signed using the third coordinate (variable x ) of the dierential embedding induced by variable x. The linking numbers are thus given by the half-sum of the oriented crossings [44, 22]. In the present cases, we obtain lk(1, 10) = 1 (2) = 1, 2 1 and lk(10110, 10) = 2 (8) = 4.
0,1 0,1

0,05

0,05

dx /dt

dx /dt
(1) (10) Negative crossings

-0,05

-0,05
(10) (10110) Negative crossings

-0,1 0 0,1 0,2 0,3 0,4 0,5 0,6

-0,1 0 0,1 0,2 0,3 0,4 0,5 0,6

0,7

0,8

0,9

0,7

0,8

0,9

Host cells x

Host cells x

(a) Orbits (1) and (10)

(b) Orbits (10110) and (10)

Figure 9: Regular plane projections of two knots made of unstable periodic orbits extracted from the chaotic attractor solution to the cancer model (2).

These two knots present only negative (anticlockwise rotation) crossings. As we already mentioned, the attractor is indeed similar to the spiral R ossler attractor which is characterized by the linking matrix [44, 22] Mij = 0 1 1 1 . (12)

The diagonal elements Mii of such a matrix encodes the number of -twists in the ith branch of the template and the o-diagonal elements Mij corresponds to the number of permutation between the ith and the j th branches. The corresponding template is drawn in Fig. 10 with the three periodic orbits. 19

The oriented crossing can thus be counted from such a picture to check that the linking numbers obtained from a template construction are in agreement with those counted in plane projections of the corresponding orbits (Fig. 9). The chaotic attractor solution to the cancer model is thus topologically equivalent to the spiral R ossler attractor.

(10110) (10) (1)

Figure 10: Template (or branched manifold) scheming the chaotic attractor solution to the cancer model (2) with 1 = 0.518. Orbits (1), (10) and (10110) are also shown in order to count the linking numbers: lk(10, 1) = 1, lk(10110, 1) = 2, and lk(10110, 10) = 4.

When the growth rate of host cells is increased to 1 = 0.73, the rstreturn map presents seven monotonic branches (Fig. 11a), that is, it has six critical points. The layered structure layered for 1 = 0.518 (Fig. 8b) is no longer obvious (only a slight double structure can be seen in the rst increas20

ing branch). Such a feature justies a posteriori the used approximation according to which only two distinct branches for 1 = 0.518 were retained. The new branches are clearly less and less developed as it is similarly observed in multimodal chaotic attractors solution to the R ossler system as shown in Fig. 11b (see [22] for details). We observed that the pruning of periodic orbits is more important in the case of the cancer model than in the R ossler system. This was already identied in the unimodal case (Fig. 8). This dierence in the population of periodic orbits is the main departure between the two systems.
1

-7 -6 -5

0,95

0,9

0,85

yn+1

-4 -3 -2 -1 0

xn+1

0,8

0,75

0,75

0,8

0,85

xn

0,9

0,95

-1

-2

-3

yn

-4

-5

-6

-7

(a) Cancer model (1)

(b) R ossler system

Figure 11: Multimodal rst-return map to a Poincar e section for the cancer model (a) and the R ossler system (b). Parameter values: (a) 1 = 0.73 and others values as in Eq. (3). (b) a = 0.5505, b = 2, and c = 4.

The fact that the dynamics of the cancer model (2) produces chaotic attractors whose structures as topologically equivalent to the attractors solution to the R ossler system (even when they are characterized by multimodal maps) allows to be sure that any attractor produced by this cancer model has an equivalent produced by the R ossler system. From the extensive analysis performed in the R ossler system [22], this also certify that the attractor here investigated is rather representative to what can be observed for other parameter values. From the cancer point of view, this means that there is not a too large variety of dynamics which can be produced by this model.

21

5. Discussion and conclusion A simple model for competing cells published by de Pillis and Radunskaya, namely host, eector immune and tumor cells, was investigated using three lines of attack: observability, bifurcation and topological analysis. This approach might suggest new trends in understanding the growth of some kinds of tumors and, even in designing their treatments. We thus showed that using bifurcation diagrams, the action of some terms can be identied. In the present paper, we showed that increasing the growth rate of host cells was equivalent to decreasing the rate with which tumor cells kill host cells. Nevertheless, if promoting host cells helps to reduce the uctuations of the populations, and favor long period of time during which host cells are at their maximal population and tumor cells are very few, there are also strong inversion of populations for which tumor cells become the most important populatoin: such oscillations could correspond to fast growing cancer such as small lung cancer. From a topological point of view, a relevant similarity was found between this model for competing populations of cells and the R ossler system, since both produce chaotic attractors which are topologically equivalent. The bifurcation diagrams are quite similar for both systems. The populations of unstable periodic orbits is less developed in the cancer model, something which could be possibly explained by the fact that the attractor must take place in the positive octant, a restriction not observed in the R ossler system. The attractors are structured in both systems by two saddle-foci, one (S4 ) which has a two-dimensional unstable manifold and which is surrounded by the attractor, and one (S3 ) which has a one-dimensional unstable manifold and which is not surrounded by the attractor (although this may be not so obvious on the projection shown in Fig. 2). The latter is associated with the boundary of the attraction basin. It was thus showed that in order to investigate the underlying dynamics, the populations of the host cells, or possibly, the tumor cells should be measured. It is rather inecient to measure only the population of eector immune cells, due to the poor observability it provides on the dynamics. This result is interesting because clinicians never use immunological parameters to assess tumor behaviour, even during immunotherapy [45]. Moreover, at a larger scale, tumor dynamic is commonly assessed by its impact on weight and health condition (using performance or Karnofsky performance scale [46, 47, 48]), as well as tumor size measurements. 22

Since observability is related to controllability (one is the dual of the other), therapies acting on the host cell population could be eective. Although this seems a convenient way to address the problem, further investigations using controllability concepts are needed here. A possible approach could be to increase the growth rate of host cells, and/or to biologically inuence interactions between these cells and tumor cells. This possible action results from the fact that the population of tumor cells decreases when the growth rate of host cells increases. This cancer model also indicates that the growth rate of tumor cells implies large uctuations of all the populations, leading the patients to present oscillating health conditions (as observed in clinic). Such an approach is thus interesting because models for competing species deal with patients outcome rather than only with the tumor outcome, that is, with the interactions between the patient and the tumor, which are strongly correlated to patient outcome. The nonlinear behavior which is often observed in patients tumor dynamics contrasts with statistical or biological models, which are mostly based on single clones of tumor cell growth and on linear models, such as linear-quadratic or tumor control probability models of tumor radiosensitivity [50, 49, 51]. Although such models are widely used for clinical purposes, they present many known limitations because they do not take into account tumor cells heterogeneity and stromal cells interactions. Ecological models take into account the nonlinear interactions between tumoral and other cells and could be useful for therapeutic approaches such as non-tumoral-cell-targeted treatments. Indeed, there is no selected therapy that currently increases the growth rate of normal cells without also increasing the tumor cell proliferation. However, there are numerous ways to act on normal cells that lead to tumor shrinkage such as anti-angiogenic therapy, hormonotherapy and early antiperiostine therapy [40]. Moreover, this type of models is useful to better understand the dynamics underlying the growth of tumor cells. Although quite simple and possibly incomplete, these models for competing populations of cells capture features which are qualitatively in agreement with the daily experience of oncologists. Cancer cells are indeed not only responsible for tumor response to therapy or metastatic growth [2, 39, 40] but also for some eects on their environment, that is, on the host cells. Hence, the contribution of non-tumor cells in cancer dynamics appears to be very important in the global behaviour of the system. Cancel cell genomic instability leads to intra-tumoral cell heterogeneity and to biological advantages which can be taken into account by mathematical and biological models [52], in which the 23

interactions between tumor and non-tumor cells aect in a deterministic way tumor growth, resistance and response to treatment. Otherwise some tumor may be cured without treatment, a feature which is never observed in clinic. Consequently, population dynamical models must be developped to understand the population dynamics and the eect of evolutionary parameters of neoplasms, even if this remains at a qualitative level as is the case nowadays. Moreover, some available measurements provide biomarkers that can be used for risk stratication, intermediate endpoints and targets for new drugs [1]. Ecological models of interaction have to be considered for there global dynamic against or with cancer cells. One of the main advantages of the ecological models describing the interactions between tumor cells and their environment is that their qualitative dynamical analysis allows to get new insights on the resulting global dynamics although some details in the interactions between dierent types of tumor cells not only with host cells but also with each others are not considered. From that point of view, de Pullis and Radunskayas model is obviously not complete but it has the great interest to be generic enough to reproduce many dierent features which are daily observed in clinical practice as well as in biological experiments. Among others, this model suggests a chaotic dynamics for tumor growth in a host environment, that is, a nonlinear growth as observed in culture of cancer cells. The non conventional analysis we performed beyong investigating the xed points or providing a single set of parameter values for which there is a chaotic attractor, is hoped to be helpful for designing new diagnostic and control technique to limit the tumor growth. This would redene new strategy for dynamical therapies. Acknowledgements This work has been partially supported by CNPq (Brazil) and CNRS (France). [1] L. M. F. Merlo, J. W. Pepper, B. J. Reid & C. C. Maley, Cancer as an evolutionary and ecological process, Nature Reviews Cancer, 6, 924-935, 2006. [2] Y. Sun, J. Campisi, C. Higano, T. M. Beer, P. Porter, I. Coleman, L. True, & P. S. Nelson, Treatment-induced damage to the tumor microenvironment promotes prostate cancer therapy resistance through WNT16B, Nature Medicine, 18 (9), 1359-1368, 2012. 24

[3] J. R. Brahmer, S. S. Tykodi, L. Q. M. Chow, W.-J. Hwu, S. L. Topalian, P. Hwu, C. G. Drake, L. H. Camacho, J. Kauh, K. Odunsi, H. C. Pitot, O. Hamid, S. Bhatia, R. Martins, K. Eaton, S. Chen, T. M. Salay, S. Alaparthy, J. F. Grosso, A. J. Korman, S. M. Parker, S. Agrawal, S. M. Goldberg, D. M. Pardoll, A. Gupta, & J. M. Wigginton, Safety and activity of anti-PD-L1 antibody in patients with advanced cancer, New England Journal of Medicine, 366, 2455-2465, 2012. [4] E. S. Nakasone, H. A. Askautrud, T. Kees, J. H. Park, V. Plaks, A. J. Ewald, M. Fein, M. G. Rasch, Y. X. Tan, J. Qiu, J. Park, P. Sinha, M. J. Bissell, E. Frengen, Z. Werb & M. Egeblad, Imaging tumor-stroma interactions during chemotherapy reveals contributions of the microenvironment to resistance, Cancer Cell, 21 (4), 488-503, 2012. [5] U. Fory s, Multidimensional Lotka-Volterra systems for carcinogenesis mutation, Mathematical Methods in Applied Sciences, 33, 2287-2308, 2009. [6] R. A. Gatenby & T. L. Vincent, Application of quantitative models from population biology and evolutionary game theory to tumor therapeutic strategies, Molecular Cancer Therapeutics, 2, 919-927, 2003. [7] R. A. Gatenby & T. L. Vincent, An evolutionary model of carcinogenesis, Cancer Research, 63, 6212-6220, 2003. [8] J. D. Nagy, Competition and natural selection in a mathematical model of cancer, Bulletin of Mathematical Biology, 66, 663-687, 2004. [9] M. Gilpin, Spiral chaos in a predator prey model, The American Naturalist, 113, 306-308, 1979. [10] A. Arneodo, P. Coullet & C. Tresser, Occurrence of strange attractors in three-dimensional Volterra equations, Physics Letters A, 79, 259-263, 1980. [11] A. Arneodo, P. Coullet, J. Peyraud & C. Tresser, Strange attractors in Volterra equations for species in competition, Journal of Mathematical Biology, 14 (2), 153-157, 1982. 25

[12] J. A. Vano, J. C. Wildenberg, M. B. Anderson, J. K. Noel & J. C. Sprott, Chaos in low-dimensional Lotka8093Volterra models of competition, Nonlinearity, 19, 2391-2404, 2006. [13] M. Itik & S. P. Banks, Chaos in a three-dimensional cancer model, International Journal Bifurcation & Chaos, 20 (1), 71-79 (2010) [14] L. G. de Pillis & A. Radunskaya, The dynamics of an optimally controlled tumor model: A case study, Mathematical and Computer Modelling, 37, 1221-1244, 2003. [15] F. Denis & C. Letellier, Chaos therapy: a fascinating concept for oncologists, Cancer Radiothrapie, 16 (3), 230-236, 2012. [16] F. Denis & C. Letellier, Radiotherapy and chaos theory: The tit and the buttery, Cancer Radiothrapie, 16 (5-6), 404-409, 2012. [17] V. A. Kuznetsov & I. A. Makalkin, Nonlinear dynamics of immunogenic tumors: parameter estimation and global bifurcation analysis, Bulletin of Mathematical Biology, 56 (2), 295-321, 1994 [18] D. Kirschner & J. C. Panetta, odeling immunotherapy of the tumor-immune interaction, Journal of Mathemtical Biology, 37, 235252, 1998. [19] N. Kronik, Y. Kogan, V. Vainstein & Z. Agur, Improving alloreactive CTL immunotherapy for malignant gliomas using a simulation model of their interactive dynamics, Cancer Immunology & Immunotherapy, 57, 425-439, 2008. [20] J. Coste, J. Peyraud & P. Coullet, Asymptotic behaviors in the dynamics of competing species, SIAM Journal of Applied Mathematics, 36, 516-543, 1979. [21] P.-F. Verhulst, Notice sur la loi que la population suit dans son accroissement, Correspondance Math ematique et Physique, 10, 113-125, 1838. [22] C. Letellier, P. Dutertre & B. Maheu, Unstable periodic orbits and templates of the R ossler system: toward a systematic topological characterization, Chaos, 5 (1), 271-282, 1995. 26

[23] R. Hermann & A. Krener, Nonlinear controllability and observability, IEEE Transactions on Automatic Control, 22 (5), 728-740, 1977. [24] C. Letellier, L. A. Aguirre & J. Maquet, Relation between observability and dierential embeddings for nonlinear dynamics, Physical Review E, 71, 066213, 2005. [25] C. Letellier & L. A. Aguirre, Investigating nonlinear dynamics from time series: the inuence of symmetries and the choice of observables, Chaos, 12, 549-558, 2002. [26] L. A. Aguirre & C. Letellier, Observability of multivariate dierential embeddings, Journal of Physics A, 38, 6311-6326, 2005. [27] C. Letellier & L. A. Aguirre, Symbolic observability coecients for univariate and multivariate analysis, Physical Review E, 79, 066210, 2009. [28] L. A. Aguirre & C. Letellier, Investigating observability properties from data in nonlinear dynamics, Physical Review E, 83, 066209, 2011. ssler, An equation for continuous chaos, Physics Letters A, [29] O. E. Ro 57 (5), 397-398, 1976. [30] C. Letellier, L. A. Aguirre & J. Maquet, How the choice of the observable may inuence the analysis of nonlinear dynamical system, Communications in Nonlinear Science and Numerical Simulation, 11, 555-576, 2006. [31] C. Letellier & L. A. Aguirre, On the interplay among synchronization, observability and dynamics, Physical Review E, 82, 016204, 2010. [32] R. Kiessling & A. Choudhury, Cancer vaccines, British Journal of Surgery, 94, 1449-1450, 2007. [33] E. J. Small, P. F. Schellhammer, C. S. Higano, C. H. Redfern, J. J. Nemunaitis, F. H. Valone, S. S. Verjee, L. A. Jones & R. M. Hershberg, Placebo-controlled phase III trial of immunologic therapy with sipuleucel-T (APC8015) in patients with metastatic, 27

asymptomatic hormone refractory prostate cancer, Journal of Clinical Oncology, 24, 3089-3094, 2006. [34] C. Higano, P. A. Burch, E. J. Small, P. Schellhammer, R. Lemon, S. Verjee & R. Hershberg, Immunotherapy (APC8015) for androgen independent prostate cancer (AIPC): nal progression and survival data from a second Phase III trial, ECCO 13th European Conference (Paris), October 2005. [35] M. J. Feigenbaum, Quantitative universality for a class of nonlinear transformation, Journal of Statistical Physics, 19 (1), 25-52, 1978. [36] P. Coullet & C. Tresser, It erations dendomorphismes et groupe de renormalisation, Journal de Physique (Colloque C5 suppl ement), 8 (39), C5-25, 1978. [37] S. P. Dawson, C. Grebogi, I. Kan & H. Koc ak, Antimonotonicity: inevitable reversals of period-doubling cascades, Physics Letters A, 162, 249-254, 1992. [38] U. Parlitz, V. Englisch, C. Scheffczyk & W. Lauterborn, Bifurcation structure of bubble oscillators, Journal of the Acoustcal Society of America, 88 (2), 1061-1077, 1990. [39] M. Garcia-Barros, F. Paris, C. Cordon-Cardo, D. Lyden, S. Rafii, A. Haimovitz-Friedman, Z. Fuks & R. Kolesnick, Tumor response to radiotherapy regulated by endothelial cell apoptosis, Science, 300, 1155-1159 (2003) [40] I. Malanchi, A. Santamaria-Martnez, E. Susanto, H. Peng, H.-A. Lehr, J.-F. Delaloye & J. Huelsken, Interactions between cancer stem cells and their niche govern metastatic colonization, Nature, 481, 85-89, 2012. [41] K. J. Png, N. Halberg, M. Yoshida & S. F. Tavazoie, A microRNA regulon that mediates endothelial recruitment and metastasis by cancer cells, Nature, 481, 190-194, 2012. [42] A. Choudhury, S. Mosolits, P. Kokhaei, L. Hansson, M. Palma & H. Ellstedt, Clinical results of vaccine therapy for cancer: learning from history for improving the future, Advances in Cancer Research, 95, 147-202 (2006) 28

[43] M. Chi & A. Z. Dudek, Vaccine therapy for metastatic melanoma: systematic review and meta-analysis of clinical trials, Melanoma Research, 21 (3), 165-174 (2011) [44] N. B. Tufillaro, T. Abbott & J. Reilly, An Experimental Approach to Nonlinear Dynamics and Chaos, Addison-Wesley, New York, 1992. [45] S. L. Topalian, F. S. Hodi, J. R. Brahmer, S. N. Gettinger, D. C. Smith, D. F. McDermott, J. D. Powderly, R. D. Carvajal, J. A. Sosman, M. B. Atkins, P. D. Leming, D. R. Spigel, S. J. Antonia, L. Horn, C. G. Drake, D. M. Pardoll, L. Chen, W. H. Sharfman, R. A. Anders, J. M. Taube, T. L. McMiller, H. Xu, A. J. Korman, M. Jure-Kunkel, S. Agrawal, D. McDonald, G. D. Kollia, A. Gupta, J. M. Wigginton & M. Sznol, Safety, activity, and immune correlates of anti-PD-1 antibody in cancer, New England Journal of Medicine, 366, 2443-2454, 2012. [46] D. A. Karnofsky & J. H. Burchenal, The clinical evaluation of chemotherapeutic agents in cancer. In: Evaluation of chemotherapeutic agents, C. M. MacLeod (ed.), Columbia University Press (New York), pp. 191-205, 1949. [47] C. Conill, E. Verger & M. Salamero, Performance status assessment in cancer patients, Cancer, 65 (8), 1864-1866, 1990. [48] G. Buccheri, D. Ferrigno & M. Tamburini, Karnofsky and ECOG performance status scoring in lung cancer: A prospective, longitudinal study of 536 patients from a single institution , European Journal of Cancer, 32 (7), 1135-1141, 1996. [49] J. F. Fowler, The linear-quadratic formula and progress in fractionated radiotherapy, British Journal of Radiology, 62, 679-694, 1989. [50] H. V. Jaina, S. K. Clinton, A. Bhinder & A. Friedmana, Mathematical modeling of prostate cancer progression in response to androgen ablation therapy, Proceedings of the National Academy of Science, 108 (49), 19701-19706, 2011.

29

[51] I. El Naqa, P. Pater & J. Seuntjens, Monte Carlo role in radiobiological modelling of radiotherapy outcomes, Physics in Medicine and Biology, 57 (11), R75-R97, 2012. [52] F. Michor, Y. Iwasa & M. A. Nowak, Dynamics of cancer progression, Nature Reviews Cancer, 4 (3), 197-205, 2004.

30

Weperformedatopologicalandanobservabilityanalysisofacancermodel. Increasingthegrowthrateofhostcellsdevelopsthefluctuationsofpopulations. Increasingthegrowthrateofhostcellsinducesmorerarebutfastgrowingtumors. Thekillingrateoftumorcellsbytheimmunecellsdoesnotaffectthedynamics.

Você também pode gostar