1: \documentclass[prl,aps,twocolumn,showpacs]{revtex4}
2: %\documentclass[pre,preprint,aps,showpacs]{revtex4}
3: \usepackage{epsfig}
4: \usepackage{graphicx}
5: \begin{document}
6: \input{epsf}
7: \title{Dynamics and pattern formation in invasive tumor growth}
8: \author{Evgeniy Khain and Leonard M. Sander}
9: \affiliation{Department of Physics and Michigan Center for
10: Theoretical Physics, The University of Michigan, Ann Arbor,
11: Michigan 48109}
12:
13: \begin{abstract}
14: In this work, we study the in-vitro dynamics of the most malignant
15: form of the primary brain tumor: Glioblastoma Multiforme.
16: Typically, the growing tumor consists of the inner dense
17: proliferating zone and the outer less dense invasive region.
18: Experiments with different types of cells show
19: qualitatively different behavior. Wild-type
20: cells invade a spherically symmetric manner, but mutant
21: cells are organized in tenuous branches. We formulate a model for
22: this sort of growth using two coupled reaction-diffusion equations
23: for the cell and nutrient concentrations. When
24: the ratio of the nutrient and cell diffusion coefficients exceeds
25: some critical value, the plane propagating front becomes unstable
26: with respect to transversal perturbations. The instability
27: threshold and the full phase-plane diagram in the parameter space
28: are determined. The results are in a good agreement with
29: experimental findings for the two types of cells.
30: \end{abstract}
31: \pacs{87.18.Ed, 87.18.Hf} \maketitle
32:
33: One of the most aggressive forms of primary
34: brain tumor is Glioblastoma Multiforme (GBM) \cite{Surawicz}.
35: Despite major advances in medical science the prognosis for victims of this disease is very poor \cite{Surawicz}: the median survival for patients with newly
36: diagnosed GBM is approximately 12 months. One of the main
37: reasons for such high mortality and poor success of
38: treatment is the fact that GBMs are highly invasive
39: \cite{Demuth2}. The growing tumor sheds invasive cells which run through the brain, see Fig.~\ref{tumor}. The invasive
40: nature of malignant gliomas makes treatment
41: difficult \cite{Demuth2}; secondary tumors are produced by the invasive cells even if the primary is removed. In this paper we introduce a reaction-diffusion model for invasion. By comparing two different cell lines we hope to get insight into invasion dynamics which is needs to be better understood.
42:
43: \begin{figure}[ht]
44: \vspace{-0.5cm}
45: \centerline{\includegraphics[width=5cm,clip=]{twotwo.eps}}
46: \caption{Growing tumors from in {\it
47: vitro} experiments \cite{Stein} in collagen gel for the wild-type
48: (a) and mutant (b) cells. These \emph{in vitro} tumors consist of an
49: inner proliferation zone with a very high density of cells and an
50: outer invasive zone, where the cell density is smaller. The structure
51: \emph{in vivo} is believed to be similar.
52: The radius of the inner zone here is about 250 $\mu$.
53: For wild-type cells a spherically symmetric pattern
54: is observed, (a). Mutant cells are organized in tenuous
55: branches, (b). Note also that the invasive region for the mutant
56: type cells grows slower than for wild-type cells. }
57: \label{tumor}
58: \end{figure}
59:
60: This work is inspired by recent \emph{in vitro} experiments \cite{Deisboeck,Stein} where
61: microscopic tumor spheroids (radius about 250 $\mu$) were placed
62: in collagen-I gel and allowed to grow. The cell lines used were U87 and
63: U87-$\Delta$EGFR. The first type is called `wild-type' in what follows.
64: The second is a mutant line \cite{Nagane} in which there is an amplification of the epidermal growth factor receptor (EGFR) gene. This amplification occurs
65: in approximately $40$ percent of cases of GBM \cite{Huncharek}.
66:
67: If we compare the growth of the two cell lines \emph{in vitro} we see two main differences; cf. Figure~\ref{tumor}. The invasive
68: region for the wild-type cells grows faster than for the mutant
69: cells, and the wild-type produces a
70: spherically symmetric pattern, whereas mutant
71: cells produce a branching pattern \cite{Stein}.
72: In the present work, we formulate a simple reaction-diffusion model that
73: is able to reproduce these experimental findings and may give insight
74: into the functional significance of the mutation.
75:
76: We interpret the branching shown in Fig.~\ref{tumor}b, as a
77: branching instability. Analogous instabilities were studied in the theory of combustion
78: \cite{Sivashinsky}, and in studies of the self-organization
79: of microorganisms \cite{Ben-Jacob}. One way of modeling this
80: is to assume that there is attraction between
81: cells \cite{Sander} due to the production of growth factors.
82: Another possibility is to assume nonlinear
83: diffusion, where the diffusion coefficient {\it increases} with
84: the density of the cells, as was proposed for bacterial
85: colonies \cite{Ben-Jacob}. We will introduce another mechanism
86: based on the known biology of GBM.
87:
88: In our continuum description, we will deal with the density of
89: cells $u(\mathbf{r},t)$ and the density of some growth factor
90: or nutrient (whichever controls the growth), $c(\mathbf{r},t).$
91: We assume that $c$ diffuses to the tumor from far away.
92: Each cancer cell is able to proliferate as well as to perform
93: random motion. It is known that within the inner region, cells
94: have quite a high proliferation rate whereas in the invasive region, cells have high motility, but low proliferation rate \cite{Giese}. This is an
95: indication of a dynamical switch between the cell phenotypes
96: \cite{Giese}. We model it by introducing a
97: \emph{density-dependent} proliferation term where the proliferation rate increases
98: with cell density. The simplest form for such a term is $\partial u/\partial t \propto u^2 c$. (The
99: extra power of $u$ compared to the usual $uc$ means that high density
100: gives rapid proliferation since the proliferation rate per cell, $(1/u)\partial u/\partial t \propto u$.) As we will see,
101: this term drives the instability and leads to branching.
102: We assume that in order to proliferate a cell needs to consume some
103: amount of $c$. The density $c$ obeys a diffusion equation with a
104: sink at the tumor cells.
105:
106: We encode these assumptions in the following equations:
107: \begin{eqnarray}
108: &&\frac{\partial u}{\partial t} = \mathbf{\nabla} \cdot ( D_u \,
109: \mathbf{\nabla} u) + \alpha \, u^2 c \,, \nonumber \\
110: &&\frac{\partial c}{\partial t} = \mathbf{\nabla} \cdot ( D_c \,
111: \mathbf{\nabla} c) - \beta \, u^2 c\,. \label{basic}
112: \end{eqnarray}
113: Here $D_u$ and $D_c$ are the diffusion coefficients of $u$ and
114: $c$, $\alpha$ is a proliferation
115: coefficient, and $\beta$ is the coefficient of nutrient
116: consumption. We assume that the density in the center of the tumor
117: is not very high compared to the density of closely packed cells
118: $u_c$. We suppose also that the nutrient concentration is kept
119: constant far from the tumor: $\lim_{r \rightarrow \infty} c(r) =
120: c_\infty$.
121:
122: In what follows we will measure cell
123: density in the units of some characteristic density $u_0$,
124: nutrient density in units of $c_\infty$, distance in units of
125: $[D_c/(\beta {u_0}^2)]^{1/2}$, and time in units of $(\beta
126: {u_0}^2)^{-1}$. This gives:
127: \begin{eqnarray}
128: &&\frac{\partial u}{\partial t} =
129: \frac{1}{\delta}\mathbf{\nabla}^2
130: u + \frac{1}{m}\, u^2\,c \,, \nonumber \\
131: &&\frac{\partial c}{\partial t} = \mathbf{\nabla}^2 c - u^2\,
132: c\,, \label{basicdim}
133: \end{eqnarray}
134: where $\delta = D_c/D_u$ is the ratio of the nutrient and cell
135: diffusion coefficients and $m = \beta u_0/(\alpha c_\infty)$ is
136: the ratio of consumption and proliferation rates.
137:
138: Typically, the nutrient or growth factor represented by $c$ is a small molecule. It
139: is expected to diffuse much faster than the cells. For example, the diffusion
140: coefficient of glucose in the brain is of the order of $10^{-7}
141: \mbox{cm}^2/\mbox{s}$, while the cell diffusion is of the order of
142: $10^{-9} \mbox{cm}^2/\mbox{s}$ \cite{Sander}, so that $\delta \sim 100$. A typical
143: nutrient consumption is $10^{-12} \mbox{g/cell/min}$ \cite{Li},
144: and a typical glucose concentration is of the order of $1
145: \mbox{g/l}$. Assuming that typical cell density within the
146: invasive region is of the order of $10^5 \mbox{cell}/\mbox{cm}^3$,
147: we estimate the consumption rate as $1.7\times 10^{-6}
148: \mbox{s}^{-1}$. The typical proliferation rate in experiments
149: \cite{Stein} is of the order of $1/\mbox{day}$, so that $m$ turns
150: out to be of the order of $0.1$.
151:
152: We work in a
153: two-dimensional channel geometry. Let $x$ be the direction of tumor growth, and
154: $y$ be the transverse direction, perpendicular to the direction of
155: the front propagation. In the $y$ direction we use periodic boundary conditions
156: with a finite channel width. In the $x$ direction, far ahead of the tumor, the cell concentration
157: is zero, $u(x=\infty) = 0$, and the scaled nutrient concentration
158: is unity, $c(x = \infty) = 1$. On the other hand, at $x = -\infty$
159: we demand $c = 0$. There is a conservation law in
160: Eqs.~(\ref{basicdim}): a volume integral over the system of $(m u
161: + c)$ is a conserved quantity. Its interpretation is that in our model a cell needs some amount of food to divide.
162: Therefore, at $x = -\infty$, $u = 1/m$ if there is a steady state. Our initial conditions are
163: $u=1/m$ for $x\le 0$; $u=0$ for $x >0$, and $c=0$, for $x\le 0$; $c=c_\infty$ for $x>0$. We will investigate the situation after transients have died away and a steady propagating state has been established.
164:
165: First, we consider the solutions of Eqs.~(\ref{basicdim}) in the
166: form of plane propagating fronts: $u = u_0(\xi); c = c_0(\xi)$,
167: $\xi = x-vt$. Substituting into Eqs.~(\ref{basicdim}) we arrive
168: at:
169: \begin{eqnarray}
170: && \frac{1}{\delta}{u_0}^{\prime\prime} + v {{u_0}^\prime}
171: + \frac{1}{m}\, {u_0}^2\,c_0 = 0 \,, \nonumber \\
172: &&{c_0}^{\prime\prime} + v {{c_0}^\prime} -{u_0}^2\,c_0 = 0\,.
173: \label{plane}
174: \end{eqnarray}
175: To obtain the profiles and the velocity of front propagation, we
176: performed (in Matlab) the following shooting procedure. First, we
177: write down Eqs.~(\ref{plane}) as four coupled first-order
178: differential equations in the form $(\vec{a})^{\prime} = M
179: (\vec{a})$, where $\vec{a}$ is the column of solutions with
180: elements ${u_0}, {u_0}^{\prime}, {c_0}, {c_0}^{\prime}$. Then we
181: find the eigenvalues and eigenvectors of $M$ at $\xi = \pm\infty$.
182: Starting with the solution $\vec{a}$ that is proportional to the
183: eigenvector belonging to the positive eigenvalue at $\xi = -
184: \infty$ (using the linearity of the problem, we choose the
185: constant of proportionality to be unity), we perform shooting by
186: the velocity of front propagation $v$. We find the profiles by
187: demanding that the solution $\vec{a}$ at $\xi = + \infty$ is a
188: linear combination of the two eigenvectors $\vec{\psi}_1$ and
189: $\vec{\psi}_2$ belonging to negative
190: eigenvalues $\lambda_1$ and $\lambda_2$ at $\xi = + \infty$:
191: $\vec{a} = b_1 \vec{\psi}_1 \exp(\lambda_1 \xi) + b_2 \vec{\psi}_2
192: \exp(\lambda_2 \xi)$. Figure~\ref{profile1} shows a typical
193: solution of Eqs.~(\ref{plane}) for $u=u(\xi)$ and $c=c(\xi)$.
194:
195: \begin{figure}[ht]
196: \vspace{-0.4cm}
197: \centerline{\includegraphics[width=6.5cm,clip=]{profile1.eps}}
198: \caption{Density profiles of cells and
199: nutrient from Eqs.~(\ref{plane}). The first
200: curve is the cell density $u=u(\xi)$ (solid line.)
201: The second curve (dashed) is $c=c(\xi)$ (inset).
202: The parameters are $m=1$, $\delta = 100$.}
203: \label{profile1}
204: \end{figure}
205:
206: We now perform a linear stability analysis. Consider perturbations in the transverse
207: direction
208: \begin{eqnarray}
209: &&u=u_{0}(\xi)+u_1(\xi)\exp(\gamma t +i k y), \nonumber
210: \\
211: &&c=c_{0}(\xi)+c_1(\xi)\exp(\gamma t +i k y)
212: \label{ansatz}
213: \end{eqnarray}
214: and substitute into Eqs.~(\ref{basicdim}). We have
215: \begin{eqnarray}
216: && \frac{1}{\delta}{u_1}^{\prime\prime} + v {{u_1}^\prime}
217: + \left(\frac{2}{m}\,c_0\,u_0 - \frac{k^2}{\delta} - \gamma\right) u_1 + \frac{1}{m}\,{u_0}^2\,c_1 = 0 \,, \nonumber \\
218: &&{c_1}^{\prime\prime} + v {{c_1}^\prime} - \left({u_0}^2 + k^2 +
219: \gamma\right) c_1 - 2 c_0\,u_0\,u_1 = 0\,. \label{linear}
220: \end{eqnarray}
221: For a fixed value of transverse wave number $k$, we should find
222: the perturbations $u_1(\xi)$ and $c_1(\xi)$, and the growth rate
223: $\gamma$. As before, we rewrite Eqs.~(\ref{linear}) as four
224: coupled first-order differential equations in the form
225: $(\vec{a}_{lin})^{\prime} = M_{lin} (\vec{a}_{lin})$. There are
226: two positive eigenvalues of the matrix $M_{lin}$ at $\xi = -
227: \infty$. We start at $\xi = - \infty$ from the linear combination
228: of the corresponding eigenvectors, $\vec{a}_{lin} = b_3
229: \vec{\psi}_3 \exp(\lambda_3 \xi) + b_4 \vec{\psi}_4 \exp(\lambda_4
230: \xi)$, where one can chose $b_4$ to be unity due to the linearity
231: of the problem. Performing shooting in two parameters, the growth
232: rate $\gamma$ and the constant $b_3$, we find the eigenfunctions
233: $u_1(\xi)$ and $c_1(\xi)$ by demanding that the solution
234: $\vec{a}_{lin}$ is given at $\xi = + \infty$ by a linear
235: combination of the two eigenvectors $\vec{\psi}_5$ and
236: $\vec{\psi}_6$ belonging to negative eigenvalues
237: $\lambda_5$ and $\lambda_6$, $\vec{a}_{lin} = b_5 \vec{\psi}_5
238: \exp(\lambda_5 \xi) + b_6 \vec{\psi}_6 \exp(\lambda_6 \xi)$.
239: Changing the value of transverse wave number $k$, for the fixed
240: $\delta$ and $m$, we calculate the dispersion curve $\gamma(k)$.
241:
242: \begin{figure}[ht]
243: \vspace{-0.4cm}
244: \centerline{\includegraphics[width=6.5cm,clip=]{dispesionnew1.eps}}
245: \caption{An example of the dispersion curve $\gamma(k)$. The
246: instability occurs if the ratio of diffusion coefficients $\delta$
247: exceeds a certain critical value. For $\delta > \delta_{cr}\approx
248: 2.300$, the growth rate $\gamma$ is positive for small $k$, while
249: for larger $k$, cell diffusion in the transverse direction
250: stabilizes the instability. The parameters are: $m=0.1$, $\delta =
251: 20$. The inset shows the dependence of the largest unstable wave
252: number $k_*$ on $\epsilon = (\delta -
253: \delta_{cr})/\delta_{cr}$. Numerical simulations are the
254: asterisks, the asymptote $k_* = (0.36/m)\epsilon^{1/2}$ is
255: the dotted line.}
256: \label{dispersion}
257: \end{figure}
258:
259: As was found previously in the context of chemical reactions for
260: $m=1$ \cite{Horvath}, plane fronts can become transversally
261: unstable if the ratio of diffusion coefficients $\delta$ exceeds a
262: certain critical value. Indeed, for $\delta > \delta_{cr}$, the
263: growth rate $\gamma$ is positive for small $k$, while for larger
264: $k$, cell diffusion in the transverse direction stabilizes the
265: instability. We checked that $\delta_{cr} \approx 2.300$, in
266: agreement with previous results \cite{Horvath,MALEVANETS}.
267: Figure~\ref{dispersion} shows an example of the dispersion curve for
268: $\delta > \delta_{cr}$. An inset shows the dependence of the
269: largest unstable wave number $k_*$ on
270: $\epsilon = (\delta - \delta_{cr})/\delta_{cr}$. Numerical
271: simulations are denoted by asterisks, the asymptote $k_* =
272: (0.36/m)\epsilon^{1/2}$ is shown by the dashed line.
273:
274: For a very wide system the instability threshold
275: $\delta_{cr}$ does not depend on $m$. To see this, introduce
276: new dimensionless variables $r = (m/\delta^{1/2})R$, $t =
277: m^2T$, and $u = U/m$. In this case, $m$ drops out of
278: the problem. A consequence of
279: the elimination of $m$ is that one can easily
280: find the dependence of the velocity,
281: $v$, and of the wave number, $k$, on $m$: $v =
282: m^{-1}\delta^{-1/2} V$, $k = m^{-1}\delta^{1/2} K$. Since
283: the scaled front velocity, $V$, and the scaled wave
284: number, $K$, must be independent of $m$, $v$
285: and $k$ are proportional to $m^{-1}$. We will not eliminate the
286: parameter $m$ from the problem and will work with
287: Eqs.~(\ref{basicdim}). Different types of cells have different
288: diffusion coefficients $D_u$ and different proliferation rates
289: $\alpha$. Therefore, it is convenient that the dependence of the
290: physical quantities $k_{phys} = k [D_c/(\beta {u_0}^2)]^{-1/2}$
291: and $v_{phys} = v (D_c \beta {u_0}^2)^{1/2}$ on $D_u$ and $\alpha$
292: enters only via the dimensionless variables $v$ and $k$.
293:
294: For a system of finite width we need to form a discrete set of modes from the
295: modes of the infinite-width system by imposing $kL=2n\pi$, where $n=1, 2, \dots$ and $L$ is the width of the system. For a small enough system we can `freeze out' the instability if $k_* < 2\pi /L.$ For a spherical tumor $L$ is of the order of the diameter.
296: For example, Figure~\ref{tumor}(a), could correspond to small $k_*$.
297:
298: \begin{figure}[ht]
299: \vspace{-0.4cm}
300: \centerline{\includegraphics[width=6.5cm,clip=]{phaseplanenew.eps}}
301: \caption{The phase plane $(\delta, m)$ with two
302: families of curves: $v = \mbox{const}$ and $k_* = \mbox{const}$,
303: calculated from Eqs.~(\ref{plane}) and (\ref{linear}). The
304: two curves marked by $\circ$'s show $v = \mbox{const}$; $v = 0.1$ (left), and
305: $v = 0.02$ (right) . The two curves marked by $*$'s show $k_* =\mbox{const}$ .
306: The left-hand curve corresponds
307: to a stronger instability, $k_* = 10$, while the right one
308: corresponds to a weaker instability, $k_* = 1$. Also shown are the
309: large $\delta$ asymptotes: $\delta = \mbox{const}/m$ for $v =
310: \mbox{const}$ curves, and $m = \mbox{const}$ for $k_* =
311: \mbox{const}$ curves, see text. The instability threshold for an
312: infinite stripe, $\delta \approx 2.300$, is plotted by the dotted
313: line. The larger-$\delta$ and smaller-$m$ region correspond to
314: wild-type cells, while mutant cells are in the region of
315: smaller $\delta$ and larger $m$. This suggests that wild-type
316: cells have a larger diffusion constant, but a smaller
317: proliferation rate, compared to mutant cells.}
318: \label{phaseplane}
319: \end{figure}
320:
321: We now consider a phase plane of parameters $(\delta,m)$, see
322: Fig.~\ref{phaseplane}. As mentioned above, the main differences
323: between the experiments with wild-type and mutant cells are in the
324: velocity of the front propagation and possible symmetry-breaking.
325: It means that the region of larger $v$ and smaller $k_*$ in this
326: phase plane corresponds to wild-type cells, while the region of
327: smaller $v$ and larger $k_*$ corresponds to mutant cells, see
328: Fig.~\ref{phaseplane}.
329:
330: We consider two families of
331: curves: the first is of constant front velocity $v =
332: \mbox{const}$ and the second is of constant largest unstable wave
333: numbers $k_* = \mbox{const}$. We focus first on $v = \mbox{const}$
334: curves. It was shown previously, that for $m=1$ and large
335: $\delta$, $v = 1.219 \delta^{-1}$ \cite{Needham}. Combining this
336: with the $m$ dependence, one can see that in the $(\delta,m)$
337: phase plane the curves of constant velocities for large $\delta$
338: are given by $\delta = \mbox{const}/m$. Then, we consider $k_* =
339: \mbox{const}$ curves. Our numerical calculations indicate that for
340: large values of $\delta$, the largest unstable wave number $k_*$
341: tends to some constant independent of $\delta$; the value of this
342: constant for $m=1$ is approximately $0.5$. Therefore, for large
343: $\delta$, $k_* = 0.5/m$, and the curves of constant $k_*$ are
344: given by $m = \mbox{const}$. Figure~\ref{phaseplane} shows also
345: these large $\delta$ asymptotes. A typical length scale is
346: $[D_c/(\beta {u_0}^2)]^{1/2}\sim 0.2 \mbox{cm}$ and a typical
347: velocity scale $(D_c \beta {u_0}^2)^{1/2} \sim 4\times 10^{-7}
348: \mbox{cm}/\mbox{s}$. Comparing this with experimental data
349: \cite{Stein}, one can see that the dimensionless wave number $k_*$
350: and front velocity $v$ should be of the order of $5$ and $0.1$,
351: correspondingly.
352:
353: As one can see on Fig.~\ref{phaseplane}, larger-$\delta$ and
354: smaller-$m$ region corresponds to wild-type cells, while mutant
355: type cells correspond to smaller-$\delta$ and larger-$m$ region.
356: Thus we predict that wild-type cells have a larger diffusion constant
357: but a smaller proliferation rate than mutant cells in the invasive zone.
358: Comparing these theoretical predictions with experiments
359: \cite{Stein}, two points should be taken into account. First,
360: experiments \cite{Stein} with tumor cells were performed in a
361: three-dimensional geometry, so the tumor growth can be described
362: by spherical propagating fronts for wild-type cells. In order to
363: explain branching patterns of mutant type cells, linear stability
364: analysis of spherical fronts, rather than plane fronts, should be
365: performed. In this case, the role of the initial tumor radius should
366: be analyzed. Second, the basic solutions of our model are plane
367: fronts which propagate with constant velocity. However, in
368: experiments, the more dense inner proliferative region grows
369: slower than the less dense outer invasive region \cite{Stein}.
370: Probably the tumors are in a transient regime, and the steady-state behavior will set in later \cite{Khain}. However, our
371: qualitative predictions should hold in three dimensions. Experimental verification of these
372: features of the growth should be possible.
373:
374: In summary, we have considered the growth of GBM tumors.
375: \emph{In vitro} experiments \cite{Stein}
376: showed that the dynamics of growth and resulting patterns are
377: quite different for wild-type and mutant cells. For the wild-type
378: the invasive region grows faster, and tumor remains
379: spherically symmetric. On the other hand, the invasive region
380: grows slower for the mutant cells, and there are indications
381: of symmetry-breaking of spherically symmetric growth. We
382: formulated a simple reaction diffusion model that captures these
383: experimental findings. Based on our model, we explain
384: different patterns by different diffusion constants and
385: proliferation rates of wild-type and mutant cells: wild-type cells
386: diffuse faster, but have a lower proliferation rate in the invasive zone.
387: We think that an attempt should be made to test these predictions and relate them to the microscopic biology of the two cell lines.
388:
389: \begin{acknowledgments} We would like to thank Andy Stein for many useful
390: conversations and T. Demuth and M. Berens for experimental results.
391: Supported by NIH Bioengineering Research Partnership grant R01 CA085139-01A2.
392: \end{acknowledgments}
393: \begin{thebibliography}{}
394: \bibitem{Surawicz} T.S. Surawicz, F. Davis, S. Freels, E.R. Laws,
395: and H.R. Menck, J Neurooncol. {\bf 40}, 151 (1998); F.G. Davis, S.
396: Freels, J. Grutsch, S. Barlas, and S. Brem, J. Neurosurg. {\bf
397: 88}, 1 (1998).
398: \bibitem{Demuth2} T. Demuth and M.E. Berens, J. Neurooncol. {\bf 70}, 217 (2004).
399: \bibitem{Huncharek} M. Huncharek and B. Kupelnick, Oncol. Res. {\bf 12}, 107 (2000).
400: \bibitem{Nagane} M. Nagane, H. Lin, W.K. Cavenee, H.-J. S. Huang,
401: Canc. Lett. {\bf 162}, S17 (2001).
402: \bibitem{Deisboeck} T.S. Deisboeck, M.E. Berens, A.R. Kansal, S. Torquato, A.O.
403: Stemmer-Rachamimov, E.A. Chiocca, Cell Prolif. {\bf 34}, 115
404: (2001).
405: \bibitem{Stein} A. Stein, D. Mobley, T. Demuth, M.E. Berens, and L.M. Sander,
406: unpublished.
407: \bibitem{Sivashinsky} G.I. Sivashinsky, Ann. Rev. Fluid Mech. {\bf 15}, 179 (1983).
408: \bibitem{Ben-Jacob} E. Ben-Jacob and I. Cohen, Adv. Phys. {\bf 49}, 395 (2000);
409: I. Golding, Y. Kozlovsky, I. Cohen, and E. Ben-Jacob, Physica A
410: {\bf 260}, 510 (1998).
411: \bibitem{Sander} L.M. Sander, T.S. Deisboeck, Phys. Rev. E {\bf 66}, 051901 (2002).
412: \bibitem{Giese} A. Giese, M.A. Loo, N. Tran, D. Haskett, S.W. Coons, and M.E. Berens,
413: Int. J. Cancer {\bf 67}, 275, (1996); A. Giese, R. Bjerkvig, M.E.
414: Berens, and M. Westphal, J. Clin. Oncol. {\bf 21}, 1624 (2003).
415: \bibitem{Li} C.K.N. Li, Cancer {\bf 50}, 2066 (1982).
416: %\bibitem{Murray} J.D. Murray, {\em Mathematical
417: %Biology} (Springer-Verlag, Berlin, 1989).
418: \bibitem{Horvath} D. Horwath, V. Petrov, S.K. Scott, and
419: K. Showalter, J. Chem. Phys. {\bf 98}, 6332 (1993); A. Toth, D.
420: Horvath, E. Jakab, J.H. Merkin, and S.K. Scott, J. Chem. Phys.
421: {\bf 114} 9947 (2001); E. Jakab, D. Horvath, A. Toth, J.H. Merkin,
422: and S.K. Scott, Chem. Phys. Lett. {\bf 342} 317 (2001).
423: \bibitem{MALEVANETS} A. Malevanets, A. Careta, and R. Kapral, Phys. Rev. E {\bf 52} 4724
424: (1995).
425: \bibitem{Needham} J. Billingham and D.J. Needham, Philos T. Roy. Soc.
426: {\bf 334}, 1 (1991).
427: \bibitem{Khain} E. Khain, L.M. Sander, and A. Stein,
428: unpublished.
429: %S. C. Ferreira, M. L. Martins, and M. J. Vilela, Phys. Rev. E 65,
430: %021907 (2002).
431: %
432: %S. C. Ferreira, M. L. Martins, and M. J. Vilela, Physica A 261,
433: %569 (1998).
434: \end{thebibliography}
435:
436: \end{document}
437: