Abstract
Bacteria may change their behavior depending on the population density. Here we study a dynamical model in which cells of radius \(R\) within a diffusive medium communicate with each other via diffusion of a signalling substance produced by the cells. The model consists of an initial boundary value problem for a parabolic PDE describing the exterior concentration \(u\) of the signalling substance, coupled with \(N\) ODEs for the masses \(a_i\) of the substance within each cell. We show that for small \(R\) the model can be approximated by a hierarchy of models, namely first a system of \(N\) coupled delay ODEs, and in a second step by \(N\) coupled ODEs. We give some illustrations of the dynamics of the approximate model.
Similar content being viewed by others
References
Alberghini S, Polone E, Corich V, Carlot M, Seno F, Trovato A, Squartini A (2009) Consequences of relative cellular positioning on quorum sensing and bacterial cell-to-cell communication. FEMS Microbiol Lett 292:149–161
Ciorausecu D, Donato P (1999) An introduction to homogenization. Oxford University Press, Oxford
de Vries G, Hillen Th, Lewis M, Müller J, Schönfisch B (2006) A course in mathematical biology. SIAM, Philadelphia
Dockery JD, Keener JP (2001) A mathematical model for quorum sensing in Pseudomonas aeruginosa. Bull Math Biol 63:95–116
Evans LC (1998) Partial differential equations. American Mathematical Society, Providence
Fekete A, Kuttler C, Rothballer M, Fischer D, Buddrus-Schiemann K, Hense BA, Lucio M, Müller J, Schmitt-Kopplin P, Hartmann A (2010) Dynamic regulation of N-acyl-homoserine lactone production and degradation in Pseudomonas putida IsoF. FEMS Microbiol Ecol 72:22–34
Fuqua WC, Winans SC, Greenberg EP (1994) Quorum sensing in bacteria: the LuxR-LuxI family of cell density-responsive transcriptional regulators. J Bacteriol 176:269–275
Hense GH, Kuttler C, Müller J, Rothballer M, Hartmann A, Kreft J-U (2007) Does efficiency sensing unify diffusion and quorum sensing? Nat Rev Microbiol 5:230–239
Horswill AR, Stoodley P, Stewart PS, Parsek MR (2007) The effect of the chemical, biological, and physical environment on quorum sensing in structured microbial communities. Anal Bioanal Chem 387:371–380
Kaplan HB, Greenberg EP (1985) Diffusion of autoinducers is involved in regulation of the Vibrio fischeri luminescence system. J Bacteriol 163:1210–1214
Müller J, Kuttler C, Hense B, Rothballer M, Hartmann A (2006) Cell-cell communication by quorum sensing and dimension-reduction. J Math Biol 53:672–702
Müller J, Kuttler C, Hense B (2008) Sensitivity of the quorum sensing system is achieved by lowpass filtering. BioSystems 92:76–81
Redfield RJ (2002) Is quorum sensing a side effect of diffusion sensing? Trends Microbiol 10:365–370
Salsa S (2008) Partial differential equations in action. Springer, Milan
Slodicka M, Van Keer R (2002) Determination of a Robin coefficient in semilinear parabolic problems by means of boundary measurements. Inv Probl 18:139–152
Ward JP, King JR, Koerber AJ, Croft JM, Socket RE, Williams P (2004) Cell-signalling repression in bacterial quorum sensing. Math Med Biol 21:169–204
Acknowledgments
For fruitful discussions the authors thank B.A. Hense and A. Hutzenthaler, Helmholtz Zentrum München.
Author information
Authors and Affiliations
Corresponding author
Appendices
Appendix. Proof of approximation results: one cell
1.1 The idea
We want to approximate a solution \(a\) of (1) with \(N=1\), i.e.,
by a solution \(b\) of (4), i.e., \(b^{\prime }=f(b)-Mb, b(0)=a_0\). For \(u_0\) we assume the compatibility conditions (12). The main idea is to consider the auxiliary problem
and to calculate a suitable \(\tilde{b}\). Since (40a) can be explicitly solved to leading order, the ODE (40b) for \(b\) can then be written as
with \(\widehat{T}_R\) defined in (16), and small \(r\) due to the initial conditions in (40a). The main step is to derive a priori estimates for the difference between \(T_Ra\) and \(\widehat{T}_R\tilde{b}\), which, to lowest order, lead to the optimal choice \(\tilde{b}=Mb\). With the notation \(b_R\) for the solution of (41) we then show that \(b_R\rightarrow b\) where \(b\) solves (4) and thus prove Corollary 5.
1.2 The auxiliary problem
We compare solutions of (39) with solutions of (40). As already said, it turns out that in lowest order \(\tilde{b}(t)=Mb(t)\) is optimal, but first we keep \(\tilde{b}\) free. To simplify notation we introduce
and define
Lemma 13
Let \(w(x,t) = u(x,t)-p(x,t)|_{\Omega }\). Then \(w(x,t)\) satisfies
where
and \(\Vert g_2\Vert _{\infty }={\mathcal O}(1)\).
Proof
We have \( Bw=Bu-Bp=\frac{d_2}{R^2}a-Bp=:-g(x,t)\). The leading order terms of \(p\) can be calculated explicitly. For \(\tilde{p}=p-\alpha _0\psi _c\), where we recall \(\alpha (t)=\frac{d_2}{D+d_1}a(t)=\frac{M}{4\pi D}a(t)\), we find
Here \(\hat{c}=\tilde{b}-4\pi D\alpha _0\), using \(D\Delta \psi _c=-4\pi D\delta _0+h\) with \(\Vert h\Vert _{H^2}\le C\), cf. (7). Letting
we obtain, since \(\tilde{p}(0)=0\),
with \(\Vert p_2(\cdot ,t)\Vert _{H^2}\le C\alpha _0\), and in particular \(\Vert p_2\Vert _{L^\infty (\partial \Omega )}\le C\alpha _0\). Thus
Therefore,
with \(\Vert g_2(\cdot ,t)\Vert _{L^\infty (\partial \Omega )}\le C\alpha _0\). The integral kernels read
and thus
\(\square \)
Lemma 14
For solutions \(w\) of (42) we have
Proof
We have the a priori estimate
Integrating over time we find (45), and integrating a second time w.r.t. time yields (46). \(\square \)
By (43), the key step to minimize the right hand sides in (45), (46) for given \(a\), is to approximately solve for \(f\) the integral equation
with \(\hat{a}(t) = a(t)-a_0\), and \(f = \tilde{b} +4\pi D \alpha _0\). Formally, for \(\tau >0\)
With \(\Gamma (1/2)=\sqrt{\pi }\) and \(\Gamma (3/2) = \sqrt{\pi }/2\), we may write
In order to establish this equation not only formally we define the residual
Lemma 15
Let \(f\in C^{0,1/2}(\mathbb R _+), f(0)=0\). Then for all \(t_1>0\) there exists a \(C>0\) such that
Proof
We have
for \(\alpha \in (0,1)\). Thus, with \(A = \pi ^{-3/2} d_1\) and \(B = 2 \pi ^{-3/2} D\),
Select \(\alpha _1\in (0,1/2)\). We find by partial integration
In \(I_2\) we may directly choose \(\alpha _2=1/2\), and obtain \(I_2(\tau ) \le \rho ^{1/2}\, B\,\Vert f\Vert _{C^{0,1/2}}\Gamma (3/2-1/2)\). Last,
where we used in the last step that \(f(0)=0\). \(\square \)
Lemma 16
For all \(t_1>0\) there exists a \(C>0\) such that if \( \tilde{b}(t) = M\, a(t)\) then
Proof
First note that without suitable choice of \(\tilde{b}\), e.g., for \(\tilde{b}\equiv 0\) and \(a\not \equiv 0\), we obviously have \(\int _0^t \int _{\partial \Omega } g^2(x,\tau )\mathrm{d}o\mathrm{d}\tau \sim R^{-2}\). Denote
such that \(\hat{c}(t):=\tilde{b}-4\pi D\alpha _0=\hat{\tilde{a}}\). Then, since \(\tilde{a}(0)=0\), we have, for \(|x|=R\), and using \(\rho = R^2/(4D)\),
Thus, choosing \(\hat{\tilde{a}}=M\tilde{a}(t)\) the first integrand can be estimated, for each \(\tau \), as
Moreover, \(\frac{1}{R^2}\int _0^t g_2^2(x,\tau )\mathrm{d}\tau \le CtR^{-2}\), thus
and integrating over \(\partial \Omega \) now yields the result. \(\square \)
To compare solutions of the full problem (39) and the auxiliary problem (40) we finally need to estimate the differences of the traces of \(u\) and \(p\) on \(\partial \Omega \). For this we start with some explicit calculations involving \(\widehat{T}_R\). For \(b\in C^1(\mathbb R _+)\), let \(p=p_1+p_2\) be the solution of
i.e., \(p_1(x,t)=\int _0^t k(x,t-\tau )(Mb(\tau )-4\pi D\alpha _0)\mathrm{d}\tau +\alpha _0\psi _c\), cf. (44), such that
We find
Moreover, using
and \(M b(0)=4\pi D\alpha _0\) we obtain
Finally, for \(a,b\in C^0\) we have
Setting \(\zeta = a-b\) and assuming w.l.o.g. \(a_0=b_0\) this follows from
Lemma 17
For \(a,b\) as above we have \( \Vert T_R a-\widehat{T}_RMa\Vert _{L^1(0,t_1)} \le CRt_1(1+\Vert a\Vert _{C^{0,1/2}}).\)
Proof
We have
with \(\Vert p_2\Vert _{L^\infty (\partial \Omega )}\le C\), cf. Lemma 13. Next, reasoning like in Lemma 14 we have
and hence
Thus
and the result now follows from Lemma 16. \(\square \)
1.3 The full problem
Proof of Theorem 4
Let \(a,b\) be the solutions of (14), (15), respectively. We write the ODEs for \(a\) resp. \(b\) as
where we know that
For \(\zeta =a-b\) we obtain
where \(\eta (\tau )=\sup _{0\le \sigma \le \tau }\zeta (\sigma )\). In particular, \(\eta (t)\le CRt(1+\Vert a\Vert _{C^{0,1/2}})+C\int _0^t\eta (\tau )\mathrm{d}\tau \), and Gronwall’s inequality yields the result. \(\square \)
Proof of Corollary 5
The solutions \(a,b\) of (14), (15) depend on \(R\) and thus henceforth in particular \(b\) is denoted by \(b_R\). From Theorem 4 we have \(b_R(t)=a(t)+Rc_R(t)\) with \(\Vert c_R(t)\Vert _{C^0}\le C\) and \(a\in C^0([0,t_1])\) uniformly bounded by Theorem 1 and also equicontinuous since also \(\Vert a\Vert _{C^1}\le C\) independent of \(R\). Therefore, \((b_R)_{0<R<R_0}\) is also uniformly bounded and equicontinuous, and thus by Arzela–Ascoli we have \(b_R\rightarrow b\in C^0([0,t_1])\) as \(R\rightarrow 0\), at least for a subsequence. It remains to show that \(b\) fulfills (4).
Equation (15) for \(b_R\) is equivalent to
with \(h_R(b_R)(\tau )=f(b_R(\tau ))-4\pi d_2 b_R(\tau )+ d_1(\widehat{T}_RMb_R)(\tau )\). We show that
with \(h(b(\tau ))=f(b(\tau ))-Mb=f(b(\tau ))-4\pi d_2 b(\tau ) +d_1 \frac{M}{D} b(\tau )\). For the first two terms in \(h_R(b_R)\) we have uniform convergence \(|f(b_R(\tau ))-4\pi d_2 b_R(\tau )-f(b(\tau ))-4\pi d_2 b(\tau )| \le C\Vert b_R-b\Vert _{C^0}\le CR\). Finally, using (49), and setting \(b_R=b+Re_R\) with \(\Vert e_R\Vert _{C^0}\le C\) and \(e_R(0)=0\) we also find
uniformly in \(0\le t\le t_1\). \(\square \)
1.4 Improved approximation with delay
In Lemma 15 we showed that, for \(f(0)=0\),
with the residual \(R_\rho f\) defined by
where we recall \((K_\rho f)(t)= d_1\pi ^{-3/2} (F_{\rho , 3} f)(t) + 2 D \pi ^{-3/2}(F_{\rho , 5} f)(t), K_0f(t)=\pi ^{-1}(d_1+D)f(t)\). This was used in Lemma 16 to construct and estimate the approximation \(\tilde{b}(t)=Ma(t)\) of the solution \(b\) of the integral equation
The purpose of this appendix is to find the improved approximation \(b_{\mathrm{del}}\) of (53), i.e., to prove Theorem 6. From \(K_\rho \rightarrow K_0\) we may use, for \(\rho \) sufficiently small, a formal Neumann’s series
The problem with this formula is the loss of regularity in (52). To iterate, i.e., to estimate the second order terms \(K_0^{-2}(K_0-K_\rho )^2\) in (54), we need a \(C^{0,1/2}\)-bound for \({\mathcal R}_{\rho }f\).
Lemma 18
Let \(f\in C^{1,1/2}(\mathbb R _+), f(0)=0\). Then
We postpone the proof to the end of the section and first show that we obtain an improved approximation of \(a\).
Corollary 19
Let \(a\in C^{1,1/2}\) with \(a(0)=0\). Defining
we find
Proof
This follows from combining Lemma 15 and Lemma 18, i.e.,
\(\square \)
Proof of Theorem 6
We now compare the solution \(a\) of (14) for \(a_0=0\) and \(u_0=0\) with the solution of the delayed ODE
Since \(a\in C^{1,1/2}([0,t_1])\), see Remark 2, and as
we find
and the remainder of the proof works as the one of Theorem 4. \(\square \)
It remains to give the somewhat lengthy
Proof of Lemma 18
We claim that \(\Vert {\mathcal R}_\rho f\Vert _{C^{0,1/2}}=\Vert K_\rho f-K_0f\Vert _{C^{0,1/2}} \le C\rho ^{1/2} \Vert f\Vert _{C^{1,1/2}}\) where
For \(k=3,5\) we split
where \(0<\tau _1<\tau _2<t_1\) and
In the following we estimate term by term, always assuming \(0<\tau _1<\tau _2<t_1\). The critical terms are \(T_{1,k}\) which yield \(\Vert f\Vert _{C^{1,1/2}}\) on the right hand side of (55), while the estimates for all other \(T_{i,k}\) involve only \(\Vert f\Vert _{C^{1}}\).
-
a)
ad \(T_{1,k}\). We have
$$\begin{aligned} T_{1,k}&= \int _{\rho /\tau _1}^\infty \left(\int _0^{\rho /\zeta } \frac{|f^{\prime }(\tau _2-x)-f^{\prime }(\tau _1-x)|}{\sqrt{|\tau _2-\tau _1|}}\mathrm{d}x\right)\zeta ^{k/2-2}\mathrm{e}^{-\zeta }\mathrm{d}\zeta \\&\le \rho \int _{\rho /\tau _1}^\infty \Vert f^{\prime }\Vert _{C^{0,1/2}}\zeta ^{k/2-3}\mathrm{e}^{-\zeta } \mathrm{d}\zeta \le C\rho ^{1/2}\Vert f\Vert _{C^{1,1/2}}, \end{aligned}$$where for \(k=3\) we used
$$\begin{aligned} \int _{\rho /\tau _1}^\infty \zeta ^{3/2-3}\mathrm{e}^{-\zeta }\mathrm{d}\zeta =\int _{\rho /\tau _1}^1 \zeta ^{3/2-3}\mathrm{e}^{-\zeta }\mathrm{d}\zeta +\int _{1}^\infty \zeta ^{3/2-3}\mathrm{e}^{-\zeta }\mathrm{d}\zeta \le 2\rho ^{-1/2}\tau _1^{1/2}+C. \end{aligned}$$while for \(k=5\) we integrate by parts.
-
b)
ad \(T_{3,k}\). For \(k=3\) we have
$$\begin{aligned} |T_{3,3}|&= \frac{|f(\tau _1)|}{\sqrt{\tau _2-\tau _1}} \int _{\rho /\tau _2}^{\rho /\tau _1} \zeta ^{-1/2}e^{-\zeta } d\zeta \le \frac{|f(\tau _1)|}{\sqrt{\tau _2-\tau _1}} \int _{\rho /\tau _2}^{\rho /\tau _1} \zeta ^{-1/2} d\zeta \\&= |f(\tau _1)| 2\,\rho ^{1/2} \,\left( \frac{1}{\sqrt{\tau _1}} - \frac{1}{\sqrt{\tau _2}} \right) \frac{1}{\sqrt{\tau _2-\tau _1}} = 2\rho ^{1/2} \frac{|f(\tau _1)|}{\sqrt{\tau _1\tau _2}} \frac{\sqrt{\tau _2} - \sqrt{\tau _1}}{\sqrt{\tau _2-\tau _1}}\\&\le 2\rho ^{1/2} \frac{|f(\tau _1)|}{\tau _1} \frac{(\sqrt{\tau _2} - \sqrt{\tau _1})(\sqrt{\tau _1} + \sqrt{\tau _2})}{\sqrt{\tau _2-\tau _1}(\sqrt{\tau _1} + \sqrt{\tau _2})} = 2\rho ^{1/2} \frac{|f(\tau _1)|}{\tau _1} \frac{\sqrt{\tau _2-\tau _1}}{\sqrt{\tau _1} + \sqrt{\tau _2}}. \end{aligned}$$Now, \(f(\tau _1)/\tau _1 = f^{\prime }(\theta )\) for some \(\theta \in (0,\tau _1)\) as \(f(0)=0\), and thus \(|f(\tau _1)/\tau _1|\le \Vert f\Vert _{C^1}\). The second factor \(\sqrt{\tau _2-\tau _1}/(\sqrt{\tau _1} + \sqrt{\tau _2})=1\) for \(\tau _1=0\) and is monotonously decreasing in \(\tau _1\), thus bounded on \(0<\tau _1<\tau _2\). Hence \(|T_{3,3}|\le C\rho ^{1/2}\Vert f\Vert _{C^1}\) For \(k=5\) we use integration by parts to find
$$\begin{aligned} |T_{3,5}|&= \frac{|f(\tau _1)|}{\sqrt{\tau _2-\tau _1}} \int _{\rho /\tau _2}^{\rho /\tau _1} \zeta ^{1/2}e^{-\zeta } d\zeta \\&\le \frac{|f(\tau _1)|}{\sqrt{\tau _2-\tau _1}} \frac{1}{2} \left[\!-\!\left(\frac{\rho }{\tau _1}\right)^{1/2} e^{-\rho /\tau _1} \!+\! \left(\frac{\rho }{\tau _2}\right)^{1/2}e^{-\rho /\tau _2} \!+\! \int _{\rho /\tau _2}^{\rho /\tau _1} \zeta ^{-1/2}e^{-\zeta } d\zeta \right]\!. \end{aligned}$$From \(k=3\) we already know that \(|f(\tau _1)| \int _{\rho /\tau _2}^{\rho /\tau _1} \zeta ^{-1/2}e^{-\zeta } d\zeta /\sqrt{\tau _2-\tau _1} \le C \sqrt{\rho }\Vert f\Vert _{C^1}\), and it remains to estimate the remaining part
$$\begin{aligned}&\frac{|f(\tau _1)|}{\sqrt{\tau _2-\tau _1}} \frac{1}{2} \left[-\left(\frac{\rho }{\tau _1}\right)^{1/2} e^{-\rho /\tau _1} + \left(\frac{\rho }{\tau _2}\right)^{1/2}e^{-\rho /\tau _2} \right]\\&\quad =\frac{\sqrt{\rho }}{2} \frac{|f(\tau _1)|}{\sqrt{\tau _1\tau _2}} \frac{-\sqrt{\tau _2}e^{-\rho /\tau _1} -\sqrt{\tau _1}e^{-\rho /\tau _2}}{\sqrt{\tau _2-\tau _1}} \\&\quad = \frac{\sqrt{\rho }}{2} \frac{|f(\tau _1)|}{\sqrt{\tau _1\tau _2}} \frac{\sqrt{\tau _1}-\sqrt{\tau _2}}{\sqrt{\tau _2-\tau _1}} e^{-\rho /\tau _1} + \frac{\sqrt{\rho }}{2} \frac{|f(\tau _1)|}{\sqrt{\tau _1\tau _2}} \sqrt{\tau _1}\frac{e^{-\rho /\tau _2} - e^{-\rho /\tau _1}}{\sqrt{\tau _2-\tau _1}}. \end{aligned}$$The first term in the last sum is again known from case \(k=3\). Concerning the second we have \(|f(\tau _1)|/\sqrt{\tau _1\tau _2}\le \Vert f\Vert _{C^1}\) as before, and it remains to show that \(\sqrt{\tau _1}\frac{e^{-\rho /\tau _2} - e^{-\rho /\tau _1}}{\sqrt{\tau _2-\tau _1}}\) is bounded. With \(x=\tau _1/\tau _2\in (0,1)\) and \(z=\rho /\tau _2\in \mathbb R _+\) we have
$$\begin{aligned} \sqrt{\tau _1}\frac{e^{-\rho /\tau _2} - e^{-\rho /\tau _1}}{\sqrt{\tau _2-\tau _1}}&= \sqrt{\tau _1/\tau _2} \frac{e^{-\rho /\tau _2} - e^{-(\rho /\tau _2)(\tau _2/\tau _1)}}{\sqrt{1-\tau _1/\tau _2}}\\&= \sqrt{x}\frac{e^{-z}-e^{-z/x}}{\sqrt{1-x}}=h_1(x,z). \end{aligned}$$We fix \(x_0\in (0,1)\) and compute \(z_0=z_0(x_0)= -x_0\ln (x_0)/(1-x_0)\) which maximizes \(h_1(x_0,\cdot )\). This defines
$$\begin{aligned} h_2(x)&= \sqrt{x}\frac{e^{\frac{x\ln (x)}{1-x}} - e^{\frac{x\ln (x)}{x(1-x)}}}{\sqrt{1-x}} = \sqrt{x} e^{\frac{\ln (x)}{1-x}} \frac{1/x-1}{\sqrt{1-x}}\\&= \frac{e^{\frac{\ln (x)}{1-x}} }{ \sqrt{x}}\sqrt{1-x} = x^{1/(1-x)-1/2} \sqrt{1-x}, \end{aligned}$$which on \([0,1]\) is bounded by \(x^{1/2}(1-x)^{1/2}\le 1/2\) Alltogether, \(|T_{3,k}|\le C\rho ^{1/2}\Vert f\Vert _{C^1}\).
-
c)
ad \(T_{2,k}\). For \(k=3\) we have
$$\begin{aligned} |T_{2,3}|\le \frac{\rho \Vert f^{\prime }\Vert _{C^0}}{\sqrt{\tau _2-\tau _1}} \int _{\rho /\tau _2}^{\rho /\tau _1} \zeta ^{3/2-3} d\zeta \le \rho ^{1/2} \Vert f^{\prime }\Vert _{C^0} \frac{\sqrt{\tau _2}-\sqrt{\tau _1}}{\sqrt{\tau _2-\tau _1}}. \end{aligned}$$The last factor equals \( \frac{\sqrt{\tau _2-\tau _1}}{\sqrt{\tau _1}+\sqrt{\tau _2}}\) and is therefore bounded as in b). Thus \(|T_{2,3}|\le C\rho ^{1/2}\Vert f\Vert _{C^1}\). Also \(|T_{2,5}|\le C\rho ^{1/2}\Vert f\Vert _{C^1}\) by a similar estimate.
-
d)
ad \(T_{4,k}\). Here \(|T_{4,3}|\le \Vert f^{\prime }\Vert _{C^0}\sqrt{\tau _2-\tau _1}\int _0^{\rho /\tau _2} \zeta ^{3/2-2}\mathrm{d}\zeta \le C\rho ^{1/2}\Vert f\Vert _{C^1}\), and integrating by parts for \(k=5\) yields a similar result.\(\square \)
Appendix. Proof of approximation results: several cells
The basic idea for \(N\ge 2\) cells is to introduce a delta source for each cell, i.e., to consider
with \( \psi _c(x,t)=\sum _{i=1}^N \frac{d_2a_i(t)}{d_1{+}D} \left.\frac{\chi (\Vert x-x_i\Vert )}{\Vert x-x_i\Vert }\right|_{\Omega }\), cf. (24). The ODEs (58b) can then be rewritten as
with \(r\) due to the initial conditions, i.e., \(r\equiv 0\) if \(\psi _c|_{t=0}=0\). The proof Theorem 10 again consists of two parts: first, given \(a := (a_1(t),..,a_N(t))\), we need a good choice of \(\tilde{b} := (\tilde{b}_1,..,\tilde{b}_N)\) to control the difference between the solution \(p\) of (58) and the outer field \(u(x,t)\) on the boundary \(\partial \Omega \); see Lemma 23. Second, the communication terms \(\int _{\partial \Omega _i}\frac{d_1}{R} p\, do\) are to be replaced by functionals of \(\tilde{b}\); see Lemma 25. The proofs parallel that for one cell; most computations are straight forward (though often tedious) generalization of the one-cell-case. We only sketch the differences, and start with the scaled case \(\Vert x_i-x_j\Vert =\delta _{ij}=R^{2\eta }\tilde{\delta }_{ij}\).
Using explicit heat kernel calculations we first obtain the following generalization of Lemma 13.
Lemma 20
Let \(w(x,t) = u(x,t)-p(x,t)|_{\Omega }\). Then \(w(x,t)\) satisfies
where
with
and \(\Vert g_2^i\Vert _{\infty }={\mathcal O}(1)\).
Remark 21
The reason for splitting off \(\rho ^{-1/2-\eta }\) respectively \(\rho ^{-\eta /2}\) from \(H_{1,\rho }^jf\) and \(H_{2,\rho }^{i,j}f\) is that this way both terms are of order \(\rho ^0\). For \(H_{1,\rho }^jf\) we show this explicitly in Lemma 26 below, while for \(H_{2,\rho }^{i,j}f\) we may estimate, using \(\Vert x-x_i\Vert ={\mathcal O}(\rho ^{1/2})\) and \(\Vert x_i-x_j\Vert ={\mathcal O}(\rho ^{\eta /2})\),
Thus, the second interaction term \(H_{2,\rho }^{i,j} \tilde{b}_j\) can be neglected, while the first interaction term has to be taken into account, also if we scale the distances of cells \(\Vert x_i-x_j\Vert \) by \(\rho ^\eta , \eta \in (0,1/2)\).
1.1 Low order approximation
Let \(\delta _{ij}=R^{2\eta }\tilde{\delta }_{ij}\). Similarly to \(K_\rho \) in Lemma 15, the interaction delay terms \(H^j_{1,\rho }f\) may be approximated by undelayed terms as \(\rho \rightarrow 0\). Also the proof parallels that of Lemma 15.
Lemma 22
Assume \(\eta \in (0,1/2)\) and \(f\in C^{0,1/2}[0,t]\). There exists a \(C>0\) such that for all \(x\in \partial \Omega _i\) and \(j\ne i\)
We now transfer Lemma 16 and explain the (to this order) optimal choice \(\tilde{b}_i\).
Lemma 23
Let \(\eta \in (0,1/2)\). For all \(t_1>0\) there exists a \(C>0\) such that if
then for \(i=i,\ldots ,N\) and \(t\le t_1\)
Proof
We know that
with
The aim is to select \(\tilde{b}_i\) in such a way that \(I_{i,\rho } = {\mathcal O}(\rho )\). We know that, for \(x\in \partial \Omega _i\),
where \( K_0 f = \frac{d_1+D}{\pi }\, f \) and \( H_{1,0}^j f = \frac{(4 D)^{1/2-\eta }}{\pi \tilde{\delta }_{i,j}}\, f\). Thus, if we define \(I_0\) by
then \(| I_\rho -I_0| \le C \rho \Vert \tilde{b}_i\Vert _{C^{0,1/2}}\), using \(\alpha _{i0}=\frac{M}{4\pi D}a_{i0}\). It is sufficient to choose \(\tilde{b}_i\), such that \(I_0 = {\mathcal O}(\rho )\). Defining \(\tilde{b}_i = Ma_i+\rho ^{1/2-\eta } B_i\) and solving for \(B_i\) at \({\mathcal O}(\rho ^{1/2-\eta })\) yields (62). \(\square \)
Given \(a_i(t)\) we have an approximation \(p\) of \(u\) such that \(g_i=(Bu-Bp)|_{\Omega _i}\) is \({\mathcal O}(1)\). Next we control the inflow into the cell by a lemma paralleling Lemma 17; we skip the proof.
Lemma 24
Given \(a=(a_1,\ldots ,a_N)\), let \(\tilde{b}_i\) be defined by (62). Let \(T_{i,R} a = \frac{1}{R}\int _{\partial \Omega _i} u(t,x)\) and \( \widehat{T}_{i,R} \tilde{b} = \frac{1}{R}\int _{\partial \Omega _i} p_1(t,x)\) where \(p_1\) is the solution of (58) with zero initial data, as in (44) for \(N=1\). There exists a \(C>0\) such that
The Lipschitz-continuity of \(\widehat{T}_R^i\) can be shown similarly like that of \(\widehat{T}_R\), and the proof of Theorem 10, i.e., the justification of the delayed ODE system (32), now follows from Gronwalls inequality, exactly as in the proof of Theorem 4, replacing \(\rho \) by \(R^2/(4D)\) to obtain \( \tilde{b}_i(t) = M a_i(t) - R^{1-2\eta } \frac{d_1\, M}{d_1+D}\sum _{j\not = i}\frac{a_j(t)}{\tilde{\delta }_{i,j}}. \)
In order to obtain an ODE from (32) we approximate the delays
Lemma 25
There exists a \(C>0\) such that
Proof
The first estimate has been already derived in (49). Recalling \(\Vert x_i-x_j\Vert =\delta _{ij}=\tilde{\delta }_{ij}R^{2\eta }\) we obtain for \(j\not = i\)
as in (49). \(\square \)
Similar to (51), this last estimate is used in the proof of Corollary 11, that is, the justification of the approximate system (31), i.e.,
where in the second equality we dropped the \({\mathcal O}(R)\) resp. \({\mathcal O}(R^{2-2\eta })\) terms from Lemma 25.
1.2 Improved approximation
We now do not scale the distances between cells (\(\eta =0\)), and aim at an error bound of order \(R^2\). For simplicity we again assume zero initial conditions, s.t. \(\alpha _{i0}=a_{i0}=0\) and consequently \(g_{i,2}=0\) in Lemma 20. The analysis proceeds similar to that before; the pertinent approximation of \(H_{1,\rho }^j\), however, is different.
Lemma 26
There exists a \(C>0\) such that \( \Vert H_{1,\rho }^jf -I_{1,0}^{i,j}f\Vert _{C^0} \le C \rho ^{1/2}\Vert f\Vert _{C^{0}[0,t]}\) for \(x\in \partial \Omega _i\), where
Proof
For \(x\in \partial \Omega _i\),
Now, as \(|\Vert x-x_j\Vert -\Vert x_i-x_j\Vert |\le C \rho \),
by l’Hospital’s rule. Similarly,
using \( \left|\left( \frac{\Vert x-x_j\Vert ^2}{\Vert x_i-x_j\Vert ^2} -1 \right) \zeta \right| \le C \rho \rho ^{-1/2} = C \rho ^{1/2}\) for \(0\le \zeta \le 1/\sqrt{\rho }\). \(\square \)
Lemma 23 is slightly modified in order to define the appropriate approximation.
Lemma 27
For all \(t_1>0\) there exists a \(C>0\) such that if
then, replacing \(\tilde{b}_j\) by \(\tilde{c}_{j,del}\) in the definition of \(g(x,t)\), for \(i=i,\ldots ,N\) and \(t\le t_1\)
Proof
We start off as in the proof of Lemma 23, and find that (\(a_{i,0}=0, u_0=0\))
where \(I_{i,\rho }\) assumes the form (\(\eta =0, a_{i,0}=0, u_0=0\) and replacing \(\tilde{b}_j\) by \(\tilde{c}_j\))
We aim at a choice of \(\tilde{c}_i\) that leads to \(I_{i,\rho } = {\mathcal O}(\rho ^{2})\). As we know from Corollary 19 that the integral equation \(K_\rho \tilde{c}_i = 4 D\,d_2 a_i\) is solved up to order \(\rho \) by the choice \( \tilde{c}_{i,del}^0(t) = M ( 2I - K_0^{-1}K_\rho ) (a_i)(t)\), we plug in the ansatz \( \tilde{c}_{i,del}(t) = c_{i,del}^0(t) - \rho ^{1/2} B_i(t)\) to obtain
The natural choice for \(B_i\) that guarantees the necessary approximation order reads
As for \(x\in \partial \Omega _i\) we know \(\Vert (H_{1,\rho }^j-I_{1,0}^{i,j}) f \Vert _{C^0} \le C \rho ^{1/2}\Vert f\Vert _{C^{0}}\), we find \(I_{i,\rho } = {\mathcal O}(\rho ^{2})\).
\(\square \)
This lemma implies
and as \(\widehat{T}_R^i\) is Lipschitz-continuous, Gronwall’s lemma yields the approximation theorem as before, i.e., the solution \(\tilde{c}_{i,del}\) of
approximates \(a(t)\) up to an error of \({\mathcal O}(\rho ) = {\mathcal O}(R^2)\). In order to finish the proof of Theorem 12 it only remains to computate explicitly an approximation of
where, of course we take \(\eta =0\) in the definition of \(H_{1,\rho }^{j}\). Lemma 26 indicates that in case \(j\not = i\) and \(x\in \partial \Omega _i\) we have the estimate \(\Vert (H_{1,\rho }^{j} f)(t,x)-(I_{1,0}^{i,j}f)(t)\Vert _{C^0} \le C \rho ^{1/2}\Vert f\Vert _{C^0}\). Hence,
If we take into account that \(\tilde{c}_{i,del}(t) = \tilde{c}_{i,del}^0(t)-\rho ^{1/2}\tilde{c}_{i,del}^1(t)\), we find
Thus we may replace \(\widehat{T}_R^i \tilde{c}_{i,del}\) by this expression in (65) without increasing the approximation error and this completes the proof of Theorem 12.
Remark 28
In order to find stationary solutions of (35) we consider a constant function \(c(t) \equiv f_0\) as input into the pertinent delays and consider the limit \(t\rightarrow \infty \). We either already proved, or it is possible to prove with similar methods, that
Therefore, the stationary solutions of the DDE (35) and the ODE (31) agree (up to the scaling of the cell distances in the ODE case).
Rights and permissions
About this article
Cite this article
Müller, J., Uecker, H. Approximating the dynamics of communicating cells in a diffusive medium by ODEs—homogenization with localization. J. Math. Biol. 67, 1023–1065 (2013). https://doi.org/10.1007/s00285-012-0569-y
Received:
Revised:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00285-012-0569-y