You are on page 1of 17

DISCRETE AND CONTINUOUS doi:10.3934/dcds.2012.32.

3081
DYNAMICAL SYSTEMS
Volume 32, Number 9, September 2012 pp. 3081–3097

CONSERVATION LAWS IN MATHEMATICAL BIOLOGY

Avner Friedman
The Ohio State University
Department of Mathematics
Columbus, OH 43210, USA

Abstract. Many mathematical models in biology can be described by con-


servation laws of the form

∂u
+ div(Vu) = F(t, x, u) (x = (x1 , . . . , xn )) (0.1)
∂t
where u = u(t, x) is a vector (u1 , . . . , uk ), F is a vector (F1 , . . . , Fk ), V is
a matrix with elements Vij (t, x, u), and Fi (t, x, u), Vij (t, x, u) are nonlinear
and/or non-local functions of u. From a mathematical point of view one would
like to establish, first of all, the existence and uniqueness of solutions under
some prescribed initial (and possibly also boundary) conditions. However, the
more interesting questions relate to establishing properties of the solutions that
are of biological interest.
In this article we give examples of biological processes whose mathematical
models are represented in the form (0.1). We describe results and present open
problems.

1. Drug resistance with infection by health care workers. Antibiotic resis-


tant organisms (ARO) pose an increasing serious health threat in hospitals. Factors
which contribute to the spread of ARO in hospitals are poor immune system of most
patients, close living quarters, and the contact with health care workers (HCWs)
as, for example, in patients with intravenous drip or catheter. D’Agata et al. [11]
[12] and Webb et al [40] developed a mathematical model of the growth in drug
resistance in terms of a system of dynamical equations. The population of patients
is divided into compartments: colonized, uncolonized, contaminated, uncontami-
nated, with the drug-resistant, or with non-drug resistant bacteria. The proportion
of the population in each compartment is considered as a variable, xi . A system
of differential equations for the xi is then introduced to describe bacterial trans-
mission among the various compartments and to explore optimal strategies of drug
treatment.
More recently Friedman et al. [31] developed a different model based on a system
of two conservation laws of the form (0.1) with V which depends on the dynamics of
the bacteria. The model considers two bacterial strains: non-drug resistant bacteria
b1 and drug-resistant bacteria b2 . We assume that each patient carries bacterial
strains of some load (b1 , b2 ). Denote by P (t, b1 , b2 ) the number density of patients
with bacterial load (b1 , b2 ), that is, the number of patients at time t with bacterial
load between (b1 , b2 ) and (b1 + ∆b1 , b2 + ∆b2 ) is approximately P (t, b1 , b2 )∆b1 ∆b2 ,

2000 Mathematics Subject Classification. 35K57, 35L60, 35L65, 35R35, 92C17, 92C37, 92C50.
Key words and phrases. Hyperbolic equations, reaction-diffusion equations, parabolic equa-
tions, free boundary problems, drug resistance, cell differentiation, wound healing, tumor growth.

3081
3082 AVNER FRIEDMAN

provided ∆b1 and ∆b2 are small numbers. Similarly, the number density of HCWs
with bacterial load (b1 , b2 ) is denoted by H(t, b1 , b2 ). We introduce the dynamics
of the bacteria in a patient and in a HCW, respectively, by

dbi dbi
= Ai (t, b1 , b2 , H), = Bi (t, b1 , b2 , P ), (i = 1, 2). (1.1)
dt dt
The function Ai depends on the natural growth of bi , the bacterial transmission
from H which contributes to an increase of bi , the response of the immune system
of patients to bi , the effect of drug treatment, and the mutation rate from non-drug
resistant bacteria to the drug-resistant bacteria. The function Bi depends on the
growth of bi and on the bacterial transmission from P . We assume that HCWs
undergo sterilization by the end of each shift, and, therefore, we do not include
infection, drug treatment and immune response in the dynamics of the bacteria
(b1 , b2 ) within the HCWs.
We assume that visits by HCWs, with H(t, a1 , a2 ), to the patients with P (t, b1 , b2 )
result in transmission of bacteria to patients, proportional to
H(t, a1 , a2 )(ai − bi )+ for species i,
where s+ = max{s, 0}. The dynamics of the bacteria in the patients is described
by the equations

db1
= λ1 b1 − νM1 (t, b)b1 − σ1 (t)b1 − µ(t)b1 + (1.2)
dt |{z} | {z } | {z } | {z }
growth immune response drug response mutation
Z
η1 H(t, a)(a1 − b1 )+ da ≡ A1 ,

| {z }
infection of patients by HCWs
db2
= λ2 b2 − νM2 (t, b)b2 − σ2 (t)b2 + µ(t)b1 + (1.3)
dt |{z} | {z } | {z } | {z }
growth immune response drug response mutation
Z
η2 H(t, a)(a2 − b2 )+ da ≡ A2

| {z }
infection of patients by HCWs

where b = (b1 , b2 ) varies in a domain Ω.


Similarly, the dynamics of the bacteria in the HCWs is given by
Z
db1
= λ1 b1 + η1 P (t, a)(a1 − b1 )+ da ≡ B1 , (1.4)
dt |{z}
growth Ω
| {z }
contamination of HCWs by patients
Z
db2
= λ 2 b2 + η2 P (t, a)(a2 − b2 )+ da ≡ B2 . (1.5)
dt |{z}
growth Ω
| {z }
contamination of HCWs by patients

Since HCWs undergo sterilization by the end of their shift, they do not become
infected.
CONSERVATION LAWS IN MATHEMATICAL BIOLOGY 3083

If we assume that patients enter or leave the hospital with total balance F (F
may be positive or negative) then, by conservation law,
∂P
+ div(AP ) = F (1.6)
∂t
where A = (A1 , A2 ) is defined in (1.2)-(1.3). Similarly,
∂H
+ div(BH) = 0 (1.7)
∂t
if the number of HCWs remains unchanged.
Patients are observed for a 6 weeks period and are given a drug when the bacterial
load b1 + b2 exceeds a threshold TH . Both TH and the duration of the treatment
are viewed as control variables, but the total amount of drug given over the 6 weeks
period is fixed. Mutation from the non-resistant strain b1 to the resistant strain b2 ,
resulting from the use of the drug, are included in the mutation term in (1.2)-(1.3).
We are interested in evaluating the bacterial loads
Z
Qi (t) = bi P (t, b)db (i = 1, 2),

and, in particular, in reducing Q2 .


The model predicts that as soon as drug is administered, the average non-
resistant load Q1 (t) will decrease and eventually (i.e., after 6 weeks) will reach
a very low level. However, the average load Q2 (t) of the drug-resistant bacteria
will decrease, immediately after treatment, but then will bounce back and remain
at a high level, dropping off eventually only if the total amount of drug to kill the
bacteria (or, alternately, the strength of this drug) is large enough. This result
agrees qualitatively with experimental results [4] [33] that underdosing increases
the drug-resistant bacteria.
The model also predicts that lower values for Q2 (t) are obtained if the (total fixed
amount of) drug is administered for a period of two weeks instead of one week.

2. T cell differentiation. Lymphocytes are white blood cells that play important
roles in the immune system. T cells and B cells are two major types of lympho-
cytes. B cells produce antibodies against pathogens while T cells are involved in
autoimmunity. Th lymphocytes represent a subtype of T cells that are identified
by the presence of surface antigens called CD4; they are referred to as CD4+ T
cells. Other subtypes of T cells include cytotoxic T cells (CD8+ ) and regulatory
T cells. Th cells are the most numerous of the T cells in a healthy person. After
an initial antigenic stimulation, Th lymphocytes differentiate into either one of two
distinct types of cells called Th1 and Th2. Th1 cells produce IFNγ that combat
intracellular pathogens; and this immune response, if abnormal, is associated with
inflammatory and autoimmune diseases. Th2 cells produce cytokines that activate
B cells to produce antibodies against extracellular pathogens; this response, if ab-
normal, is associated with allergies such as asthma. Whether a precursor Th cell
becomes Th1 or Th2 depends on ‘polarizing’ signals. Yates et al. [43] developed a
model of Th differentiation based on the interaction and competition between two
transcription factors, T-bet and GATA-3. High protein level of T-bet or GATA-3
corresponds to the Th1 phenotype or the Th2 phenotype. We shall denote by S1
and S2 the Th1 and Th2 polarizing cytokines, and by x1 and x2 the concentrations
3084 AVNER FRIEDMAN

of T-bet and GATA-3, respectively, in a Th cell. Then, the dynamics of x1 and x2


is described by

xn
 
dx1 S1 1
= −µx1 + α1 n 1 n + σ1 · + (2.1)
dt k1 + x1 ρ1 + S1 1 + x2 /γ2

β1 ≡ f1 (x1 , x2 , S1 ),

xn2
 
dx2 S2 1
= −µx2 + α2 n n + σ2 · + (2.2)
dt k2 + x 2 ρ2 + S2 1 + x1 /γ1

β2 ≡ f2 (x1 , x2 , S2 ),

where n ≥ 2. The first term on the right-hand side of each equation represents the
rate of protein degradation. The last term βi is the constant basal rate of protein
synthesis. The second term on the right-hand side represents autoactivation of
the transcription factor xi plus exogeneous signal, and both are inhibited by the
competing transcription factor xj (j 6= i). Si is given by
R
Ci (t) + xi φ(t, x1 , x2 )dx1 dx2
Si (t) = R , i = 1, 2, (2.3)
φ(t, x1 , x2 )dx1 dx2
R
where Ci (t) is non-T cell signal whereas xi φ(t, x1 , x2 )dx1 dx2 is the component of
the signal produced by the population of Th cells weighted by the density of xi ;
here φ(t, x1 , x2 ) denotes the population density of CD4+ T cells with concentration
(x1 , x2 ) at time t; the above integrals are taken over a rectangular domain

Ω = {(x1 , x2 ), 0 ≤ x1 ≤ A1 , 0 ≤ x2 ≤ A2 }.
By conservation of mass,
∂φ ∂ ∂
+ (f1 φ) + (f2 φ) = gφ (2.4)
∂t ∂x1 ∂x2
in Ω, where f1 , f2 are defined in (2.1)-(2.2), and g is a growth term, and
φ|∂Ω = 0, φ(0, x1 , x2 ) is prescribed.
Friedman, Kao and Shih [26] proved that, with appropriate choice of A1 , A2 ,
this problem has a unique solution, and they proceeded to study the behavior of
the solution as t → ∞.
Under some conditions on the parameters of the system they proved: If g → 0
fast enough as t → ∞, and lim Ci (t) exists for i = 1, 2, then as t → ∞ the Th cells
t→∞
aggregate in one, two or four peaks about points ai = (ai1 , ai2 ). More precisely,
n
X
φ(t, x1 , x2 ) → µi δai where n = 1, 2 or 4, µi > 0, (2.5)
i=1

δa is the Dirac function at a, and the convergence in (2.5) is in measure.


If ai1 > ai2 the cell tends to differentiate into Th1, and when the expression of
ai1 is high enough it will indeed differentiate into Th1. Conversely, if ai2 > ai1
CONSERVATION LAWS IN MATHEMATICAL BIOLOGY 3085

the cell will tend to differentiate into Th2. It is thus important to determine the
coordinates of ai .

Proof of (2.5). A basic idea in proving (2.5) is the introduction of upper and
lower dynamical systems

dx̂i dx̌i
= fˆi (x̂1 , x̂2 , Ŝi ), = fˇi (x̌1 , x̌2 , Ši ) (2.6)
dt dt
where x̂i (0) = x̌i (0) = xi (0) and

fˇi (x1 , x2 , Ši ) < fi (x1 , x2 , Si ) < fˆi (x1 , x2 , Ŝi ).

Initially we take
!
xn Ŝi
fˆi (x1 , x2 , Ŝi ) = −µi xi + αi n i n + σi + βi
ki + xi ρi + Ŝi
xni
 
Ši 1
fˇi (x1 , x2 , Ši ) = −µi xi + αi n n + σi + βi
ki + xi ρi + Ši 1 + Aj /γj

where j 6= i, and
R
Ĉi + Ai φ(t, x1 , x2 )dx1 dx2
Ŝi = R
φ(t, x1 , x2 )dx1 dx2

Či
Ši = R ,
φ(t, x1 , x2 )dx1 dx2

Či ≤ Ci (t) ≤ Ĉi for all t ≥ 0 (Či , Ĉi constants).

By comparison one can show that

x̌i (t) ≤ xi (t) ≤ x̂i (t) for all t > 0,

and
lim x̌i (t) = b̌i , lim x̂i (t) = b̂i (0 < b̌i < b̂i < Ai )
t→∞ t→∞

exist. Then, for any  > 0 and t1 large enough,

b̌i −  < xi (t) < b̂i +  if t > t1 . (2.7)

We next use (2.7) to introduce sharper upper and lower dynamics for t > t1 ,
namely,
!
n
x Ŝi 1
fˆi (x1 , x2 , Ŝi ) = −µi xi + αi n i
+ σi + βi
ki + xni ρi + Ŝi 1 + ( b̌j − )/γj

xn
 
Ši 1
fˇi (x1 , x2 , Ši ) = −µi xi + αi n i n + σi + βi
ki + xi ρi + Ši 1 + (b̂j + )/γj
3086 AVNER FRIEDMAN

where j 6= i, and
R
Ĉi + (b̂i + )φ(t, x1 , x2 )dx1 dx2
Ŝi = R ,
φ(t, x1 , x2 )dx1 dx2

R
Či + (b̌i − )φ(t, x1 , x2 )dx1 dx2
Ši = R ,
φ(t, x1 , x2 )dx1 dx2

Či ≤ Ci (t) ≤ Ĉi for all t ≥ t1 .


Repeating the above process we get
či −  < xi (t) < ĉi +  for all t ≥ t2 .
where b̌i < či < ĉi < b̂i .
Proceeding step-by-step in the same manner, we obtain a decreasing sequence of
rectangles Rk , whose limit R is either a rectangle or a single point. Under some
parameters regime one can prove that R is a single point a1 = (a1 , a2 ), and this
proves the assertion (2.5) with n = 1.
Under other parameters regimes each of the systems (2.6) has either two (or four)
stable equilibrium points. In this case the limit R is again a rectangle. Furthermore,
as proved in [26], each of the rectangles Rk contains two (or four) subrectangles
Rki (Rmi ⊂ Rki if m > k) such that, for any initial condition (x1 (0), x2 (0)),
(x1 (t), x2 (t)) belongs to one of the rectangles Rki for all t > tk ,
and the k’s are the same for a given i, i.e., tranjectories do not switch from Rki to
Rk+1,j where j 6= i. Furthermore, as k → ∞ the rectangles Rki decrease to a point
ai , and this completes the proof of (2.5).
The above ideas extend to more complex models. Consider an extension of the
above model which includes the fact that the transmission factor xi is prescribed
by mRNA yi . Following Mariani et al. [32], the autocatalytic process which the xi
are undergoing is given by the equation
dxi
= νi yi − τi xi
dt
where νi , τi are constants, coupled with the equations for the mRNA
xni
 
dyi Si 1
= −µi yi + αi n + σi · + βi ,
dt ki + xni ρi + Si 1 + xj /γj
for (i, j) = (1, 2) and (i, j) = (2, 1).
Introducing the population density of cells with concentration (x1 , x2 , y1 , y2 ) at
time t, φ(t, x1 , x2 , y1 , y2 ), the mass conservation law then yields,
2   X 2  
∂φ X ∂ dxi ∂ dyi
+ φ + φ = gφ.
∂t i=1
∂xi dt i=1
∂yi dt

In this case again it was proved by Friedman, Kao and Shih [27] that, for some
parameters regimes, as t → ∞
n
X
φ(t, x1 , x2 , y1 , y2 ) → µi δai
i=1
CONSERVATION LAWS IN MATHEMATICAL BIOLOGY 3087

where ai = (ai1 , ai2 , ai3 , ai4 ) and n = 1, 2 or 4. However, the proof that trajectories
cannot wander from Rki to Rk+1,j (j 6= i) for all k large enough is far more com-
plicated. It requires very careful analysis of the phase portrait in four dimensions.

3. Tumors. We assume that the tumor includes three types of cells: proliferating
cells with (mass) density p(x, t), quiescent cells with density q(x, t), and dead cells
with density n(x, t).
Following [36], we assume that quiescent cells become proliferating at a rate
KP (c) which depends on the concentration of nutrients c, and they become necrotic
at death rate KD (c). We also assume that proliferating cells become quiescent at
a rate KQ (c) and their death rate is KA (c). The density of proliferating cells is
increasing at a rate KB (c). Finally, we assume that dead cells are removed from
the tumor (by macrophages) at a constant rate KR .
We also assume that all the cells are physically identical in volume and mass and
that their density is constant throughout the tumor. Then

p + q + n = const. = θ. (3.1)
Due to the proliferation and removal of cells, there is a continuous motion of cells
within the tumor. We shall represent this movement by a velocity field v. We can
then write the conservation of mass law for the densities of the proliferating cells
p, the quiescent cells q, and the dead cells n within the tumor region Ω(t) in the
following form:

∂p
+ div(pv) = [KB (c) − KQ (c) − KA (c)]p + KP (c)q, (3.2)
∂t
∂q
+ div(qv) = KQ (c)p − [KP (c) + KD (c)]q, (3.3)
∂t
∂n
+ div(nv) = KA (c)p + KD (c)q − KR n. (3.4)
∂t
If we add equations (3.2)-(3.4) and use (3.1), we get

θ div v = KB (c)p − KR n, (3.5)


and we may replace (3.4) by (3.5). The tumor tissue will be treated as a porous
medium and the moving cells as fluid flow. In a porous medium, the velocity of v
of fluid flow is related to the fluid pressure σ by means of Darcy’s law.

v = −∇σ (3.6)
We assume that the nutrient concentration satisfies the diffusion equation

∂c
0 = ∆c − λ(p + q) in Ω(t). (3.7)
∂t
Eliminating n from (3.5) by (3.1) and taking, for simplicity, θ = 1, and recalling
(3.6), we obtain, in addition to (3.7), the following equations:

∂p
− ∇σ · ∇p = f (c, p, q) in Ω(t), t > 0, (3.8)
∂t
∂q
− ∇σ · ∇q = g(c, p, q) in Ω(t), t > 0, (3.9)
∂t
∆σ = −h(c, p, q) in Ω(t), t > 0, (3.10)
3088 AVNER FRIEDMAN

where
f (c, p, q) = [KB (c) − KQ (c) − KA (c)]p + KP (c)q − h(c, p, q)p,
g(c, p, q) = KQ (c)p − [KP (c) + KD (c)]q − h(c, p, q)q,
h(c, p, q) = −KR + [KB (c) + KR ]p + KR q.
The system (3.2)-(3.3), or (3.8)-(3.9), is a conservation law with velocity v which
depends non-locally on the variables p,q through the solution of the diffusion equa-
tions (3.7) and (3.10). In order to close the system we must give boundary condi-
tions.
We denote the boundary of Ω(t) by Γ(t) and impose the following boundary
conditions:

c = c on Γ(t), t > 0, (3.11)


σ = γκ on Γ(t), t > 0, (3.12)
where c is a constant, γ is a surface tension coefficient representing the cell-to-cell
adhesion, and κ is the mean curvature (κ = R1 if Ω(t) is a ball of radius R). The
boundary of the tumor varies in time; it is a ‘free boundary’. We assume that its
normal velocity Vn is the same as the normal velocity v·n of the cells in the outward
normal direction n, i.e.,

∂σ
= −Vn on Γ(t), t > 0. (3.13)
∂n
Since the velocity Vn of the free boundary coincides with v · n where v is the
velocity in the conservation laws (3.2)-(3.3), we do not need to assign boundary
conditions for p and q.
The system (3.7)-(3.13) with initial conditions on c, p and q was investigated
mathematically. Local existence and uniqueness of a solution with prescribed initial
data was proved in [1] [7]. Under radially symmetric data, global existence of a
radially symmetric solution was proved and asymptotic estimates were derived on
the free boundary Γ(t) = {r = R(t)} as t → ∞ [10]. There are only partial results
on the existence and uniqueness of a radially symmetric stationary solution and on
its asymptotic stability [9] [6]. However, in the case of only one population of cells
(i.e., p ≡ 1, q ≡ n ≡ 0) it was proved that there exists a unique radially symmetric
stationary solution [28]; it is asymptotically stable for all γ < γ∗ (for some γ∗ > 0)
but not for γ > γ∗ [2] [17]; see also [41]. Furthermore, there are symmetry-breaking
bifurcation branches of solutions initiating from points γ2 , γ3 , . . . , γn , . . . , where
γ2 ≥ γ∗ [29] [13] [18] [22].
So far we have assumed that the velocity v in the conservation laws (3.2)-(3.4)
is given by Darcy’s law for porous media. In general, the tissue where the tumor
develops is very heterogeneous and complex, so the Darcy’s law is just one of several
possible approximations. For some tumors, for example those that originate in the
mammary gland, it is more realistic to assume that the tissue is fluid-like, and model
it by Stokes equation.
In this case the relation between the velocity v and the pressure σ is given by

− ν∇2 v + ∇σ = f , (3.14)
div v = g (3.15)
where f = − ν3 ∇g, and the boundary condition (3.12) is replaced by
CONSERVATION LAWS IN MATHEMATICAL BIOLOGY 3089

T (v, σ)n = −γκn (3.16)


where

T (v, σ) = ν(∇v + ∇vT ) − (σ + div v)I,
3
ν is the viscosity coefficient, I is the unit matrix, and g is the proliferation rate,
given by the right-hand side of (3.5). Local existence and uniqueness of a solution
with prescribed initial data was proved in [14]. In the case of one population of
cells (p ≡ 1, q ≡ n ≡ 0) the existence of a unique stationary solution with radially
symmetric free boundary was proved in [14] and its asymptotic stability if γ < γ∗∗
(for some γ∗∗ > 0), but not if γ > γ∗∗ , was proved in [21]. As in the case of
porous medium, here too there exists a sequence of symmetry-breaking bifurcation
branches of solutions [20].

4. Dermal wounds. Consider a cutaneous, or dermal, wound which occupies a


region

W (t) = {(x1 , x2 , x3 ); (x1 , x2 ) ∈ W0 (t), −h(x1 , x2 , t) ≤ x3 ≤ 0}


where W0 (t) is the surface of the wound, lying in {x3 = 0}. Healing occurs as W0 (t)
and h(x1 , x2 , t) decrease in t. Taking a fixed cylindrical domain
q
R = {(x1 , x2 , x3 ); x21 + x22 < L, −H < x3 < 0}
such that W (0) ⊂ R, we view the tissue undergoing healing, as occupying the region

Ω(t) = R\W (t).


Let ρ denote the density of the extracellular matrix (ECM). During the healing
process ρ is increasing until, with complete recovery, it reaches the density ρ0 of
healthy normal tissue.
Wound healing is a complex biological process which involves several overlapping
stages and several types of cells and signaling molecules. It includes interactions
among blood platelets, macrophages, endothelial cells that line up the inner layer of
blood capillaries, and fibroblasts. Fibroblasts secrete collagen, which is the primary
component of the ECM. Among the various mathematical models of wound healing,
the articles [34] [35] [5] [39] emphasize the critical role of oxygen, which is tightly
connected to angiogenesis, that is, to the formation of new blood capillaries that
move toward the wound.
Here we consider a more recent model developed by Xue et al. [42] and Friedman
and Xue [30]. Since healing entails a movement of newly formed collagen within
the healing tissue Ω(t), we introduce a velocity field v in the ECM and assume that
cells and signaling molecules are moving with velocity v. By conservation of the
ECM mass we then have,

∂ρ
+ div(ρv) = G (4.1)
∂t
where  
kw ρ
G= f 1− − λρ. (4.2)
w+K ρmax
Here w is the concentration of oxygen, and f is the concentration of fibroblasts.
3090 AVNER FRIEDMAN

As in Section 3, the choice of v is to be determined by a physical constitutive


law. To determine such a law we note that the ECM in Ω(t) is a growing collagen
matrix which is elastic on a short time scale and viscous on a long time scale. We
shall model it as upper convected Maxwell (viscoelastic) fluid with isotropic pressure
depending on its density.
For an upper convected Maxwell fluid, the stress-strain relationship is given by
 

λ − (∇v)τ − τ (∇v)T + τ = η(∇v + ∇vT ),
Dt
where η is the shear viscosity, and, as shown in Xue et al. [42], the first term on
the left-hand side is very small, so after dropping it we obtain,

τ = η(∇v + ∇vT ). (4.3)


We next consider the momentum equation

∂(ρv)
+ ∇ · (ρv ⊗ v) = ∇ · σ,
∂t
where σ is the total stress. We can write σ = −P I + τ where P is the isotropic
pressure and τ is the deviatoric stress. Since healing is a slow, or quasi-stationary
process with negligible inertia, the last equation may be approximated by ∇ · σ = 0,
or

− ∇P + ∇ · τ = 0. (4.4)
For compressible material the isotropic pressure is a function of the density, i.e.,
P = P (ρ), and we take
 
ρ
P = βF −1 , (4.5)
ρ0
where β, ρ0 are positive constants, and F is a smooth approximation to the Heavi-
side function.
Taking the gradient in (4.3) and using (4.4), (4.5), we find that v = (v1 , v2 , v3 )
satisfies the elliptic system
η∇ · (∇v + ∇vT ) − ∇P = 0 in Ω(t), (4.6)
or
3  
X ∂ ∂vj ∂vi ∂P
η + − = 0 in Ω(t), (j = 1, 2, 3). (4.7)
i=1
∂xi ∂xi ∂xj ∂xj
We denote by Γ(t) the part of the boundary of Ω(t) which lies in {x3 < 0}. It is
natural to assume that there is no stress at the free boundary Γ(t). Hence
3  
X ∂vj ∂vi
η + νi − P νj = 0 on Γ(t), (j = 1, 2, 3). (4.8)
i=1
∂xi ∂xj
The boundary conditions for v at the fixed boundaries are

v1 = v2 = v3 = 0 on {x3 = −H and on x21 + x22 = L2 }, (4.9)


∂v1 ∂v2
= / W0 (t), x21 + x22 < L2 }.
= 0, v3 = 0 on {x3 = 0, (x1 , x2 ) ∈
∂x3 ∂x3
CONSERVATION LAWS IN MATHEMATICAL BIOLOGY 3091

We assume that the free boundary moves in the normal direction with velocity
Vn = v · n. We can represent this relation in the form

ψt + v · ∇ψ = 0 on Γ(t) (4.10)
where Γ(t) is given by the zero set of ψ, i.e., by ψ(t, x) = 0 where x = (x1 , x2 , x3 ).
It was proved by Friedman, Hu and Xue [25] that, given a smooth Γ(0) and a
smooth function P (t, x), there exists a unique smooth solution to the free boundary
problem (4.7)-(4.10) for a small time interval.
Of course P is not a given function!; it depends on ρ which in turn depends on
w and f , and, in fact, the variables ρ, w, f and v are involved in a larger system of
drift-diffusion equations which includes several cell types and several growth factors.
The existence and uniqueness of a solution for the complete system was also proved
in [25] for a small time interval. Here one of the critical steps consists in proving
that for any smooth P there exists a solution (v, ψ) of the free boundary problem
(4.7)-(4.10) for a small time t, and

|v|C 2+α,α/2 + |ψ|C 2+α,1+α/2 ≤ c0 |P |C 1+α,α/2


x,t x,t x,t

where c0 is a constant independent of P.


In the 2-dimensional radially symmetric case with free boundary r = R(t), it
was proved by Friedman, Hu and Xue [24] that R(t) is strictly decreasing in t,
unless the oxygen supply from the healthy tissue is mostly blocked due to vascular
damage, which indeed is often the case in chronic wounds. In this latter case
R(t) = const. > 0 for all t > t0 , for some finite time t0 .
Nothing is known about the behavior of the free boundary in the 3-dimensional
case. In fact, already the following special problem is open: Find conditions on
the function P and on Γ(0) such that for the solution of (4.7)-(4.10) there holds:
Ω(0) ⊂ Ω(t) for sufficiently small t.

5. Diffusion approximation for reaction-hyperbolic equations. Consider a


system of hyperbolic equations
n
X
{∂t pi + ∂x (vi pi )} = kij pj , i = 1, 2, · · · , n (n ≥ 2), (5.1)
j=1

where  is a small positive parameter.


We assume that the kij are constants,
n
X
kij ≥ 0 if i 6= j, kij = 0 for j = 1, 2, · · · , n, (5.2)
i=1
and

for any i0 6= i1 , there is a sequence of indices j1 , j2 , · · · , j` such that (5.3)


i0 = j1 , i1 = j` and kjm jm+1 > 0 for 1 ≤ m ≤ ` − 1.

Under these conditions, the null space of the matrix (kij )ni,j=1 is one dimensional,
and there exists a unique vector λ = (λ1 , · · · , λn ) such that
3092 AVNER FRIEDMAN

n
X n
X
kij λj = 0, 0 < λi < 1 for i = 1, 2, · · · , n, and λj = 1. (5.4)
j=1 j=1

The system (5.1)-(5.3) arises in several models of biological processes. One ex-
ample, with vi = const., arises in modeling the transport of neurofilaments along
the axon of a neuron. The neurofilaments are needed to fill in the space in the
interior of the axon. These proteins are made near the nucleus of the neuron and
are then transported, as cargo, along microtubules. The cargo is “loaded onto”, or
attached to, motor proteins that carry it along microtubules. There are two kinds
of motor proteins, kinesin and dynein. Kinesin motor proteins move forward to-
ward the synaptic terminal (anterograde motion), and dynein motor proteins move
backward (retrograde motion). We assume that neurofilaments are only capable of
movement in the longitudinal dimension of the axon when they are on track along
a microtubule, but they can switch on and off track.
When off track, neurofilaments pause for long periods until they get back on
track. When on track, neurofilaments alternate between short bouts of movement
and short pauses. Thus we divide the on track population of neurofilaments into
those that are moving and those that are making a short pause. We designate
separately the population of anterograde moving neurofilaments and the population
of retrograde moving neurofilaments. In this manner, the neurofilaments are divided
into five different populations:

u2 : neurofilaments bound to anterograde motors,


moving anterogradely, on track,
u1 : neurofilaments bound to anterograde motors, pausing, on track,
u0 : neurofilaments bound to anterograde or retrograde,
motors, pausing, off track,
u−1 : neurofilaments bound to retrograde motors
pausing, on track,
u−2 : neurofilaments bound to retrograde motors,
moving retrogradely, on track.
For simplicity we shall also denote by u−2 , u−1 , u0 , u1 , u2 the concentrations of
these five populations of neurofilaments. Following Craciun et al. [8] we assume that
neurofilaments can reverse direction (i.e. switch motors) when they are pausing, off
track. Thus we obtain the following diagram of possible transitions between the
five neurofilament populations:

u−2  u−1  u0  u1  u2
If we denote by ki,j the rate of change from ui to uj , then we have

k−1,−2 k0,−1 k0,1 k1,2


u−2  u−1  u0  u1  u2 .
k−2,−1 k−1,0 k1,0 k2,1
CONSERVATION LAWS IN MATHEMATICAL BIOLOGY 3093

These relations, together with mass conservation laws yield the following dynamical
system for 0 < x < L, where L is the length of the axon:

∂u2 ∂u2
= −vA + k1,2 u1 − k2,1 u2 ,
∂t ∂x
∂u1
= k2,1 u2 + k0,1 u0 − (k1,2 + k1,0 )u1 ,
∂t
∂u0
= k1,0 u1 + k−1,0 u−1 − (k0,1 + k0,−1 )u0 , (5.5)
∂t
∂u−1
= k0,−1 u0 + k−2,−1 u−2 − (k−1,0 + k−1,−2 )u−1 ,
∂t
∂u−2 ∂u−2
= vR + k−1,−2 u1 − k−2 u−2 ,
∂t ∂x
where vA is the velocity of neurofilaments moving anterograde, and vR is the velocity
of neurofilaments moving retrograde.
The model parameters were obtained in vitro experiments, by Brown et al. [3],
and the model simulation were shown, in [8], to be in good agreement with in vivo
experiments.
It was discovered by Reed et al. [37] that formally, as  → 0, the densities pi (x, t)
in (5.1), behave like approximate traveling wave solutions in the following sense: If
we set
n
x − vt X
s= √ where v = λ i vi (5.6)
 i=1
then the functions
Qi (s, t) = pi (x, t) (5.7)
converge to λi Q0 (s, t) where Q0 satisfies a diffusion equation
∂Q0 ∂ 2 Q0
− σ2 = 0 in − ∞ < s < ∞, t > 0, (5.8)
∂t ∂s2
and σ 2 > 0.
In the special case of (5.5), this asymptotic behavior was rigorously proved by
Friedman and Craciun [15] [16].
The following more general result is due to Friedman and B. Hu [19].

Theorem 5.1. Consider the problem (5.1)-(5.4) for 0 < x < ∞, with
 
x
pi (x, 0) = λi q0 √ , 0 < x < ∞, j = 1, . . . , n,

pj (0, t) = λj for all j for which vj > 0,
and assume that not all the vj coincide. Then

Qi (s, t) → λi Q0 (s, t) uniformly as  → 0


where Q0 satisfies (5.8) with the initial condition
(
1 if − ∞ < s < 0
Q0 (s, 0) =
q0 (s) if 0 < s < ∞.
3094 AVNER FRIEDMAN

The diffusion coefficient σ 2 can be calculated explicitly in terms of the λi , vi ,


and the kij .
There are biological models for which the vi are functions of x, while −∞ < x <
∞. For example, suppose x is the position of a molecular motor along a microtubule,
which it is transporting a load. Let Si represent different i-states of the motor, and
vi = V (x, t, Si ) denote the velocity of the motor in state i. The states Si could stand
for different attachment/detachment states or phosphorylation states, for example,
and the velocities and transitions between states could be dependent on the local
(chemical) environment (for example, ATP concentration, or concentrations of other
signaling molecules).
A second example is that of a neuron whose transmembrane potential is x and
changes of which are driven by various transmembrane ionic currents, Ij . In the
usual Hodgkin-Huxley formulation

dx X
Cm =− Ij − Iapp ,
dt j
where Cm is membrane capacitance and Iapp is the applied current. If the membrane
patch is small, then the ionic currents fluctuate randomly with the opening and
closing of individual ion channels. Thus, the ionic currents should be written as

Ij = gj Sj (x − Xj ),
where gj is the single channel conductance, Xj is the Nernst potential, and Sj is
the number of open channels of type j, a randomly fluctuating integer. Markov
models of ion channels typically have voltage-dependent, hence non-autonomous,
transitions between open and closed states. A recent article by Friedman, B. Hu
and Keener [23] provides references for these two examples. In this article Theorem
5.1 was extended to the case where vi = vi (x, t) is a smooth function, and
 
x
pi (x, 0) = λi q0 √ for − ∞ < x < ∞.

Let g(x, t) be the solution of

gt + v(x, t)gx = 0, −∞ < x < ∞, t > 0, (5.9)


g(x, 0) = x, −∞ < x < ∞
where
n
X
v(x, t) = λi vi (x, t),
i=1
and set
1
Qi (s, t) = pi (x, t), s = √ g(x, t).

n
(vi (x, t) − v(x, t))2 ≥ const. > 0, then
P
It was proved in [23] that, if
i−1

Qi (s, t) → λi Q0 (s, t) uniformly as  → 0,


where
∂Q0 ∂ 2 Q0
− σ 2 (t) − b(t)Q0 = 0, −∞ < s < ∞, t > 0 (5.10)
∂t ∂s2
Q0 (s, 0) = q0 (s), −∞ < s < ∞.
CONSERVATION LAWS IN MATHEMATICAL BIOLOGY 3095

The proof of this result, as well as that of Theorem 5.1, involve estimating higher
order derivatives of the form
∂` ∂k
`+k/2−1 ` Q
∂t ∂sk i
and proving that such terms converge to zero as  → 0.
Extension of these results to the case where the kij are variable functions remains
an open problem. We mention here one special case of a system that arise in gas
kinetics:

∂p1 1 ∂p1 1
+ = 2 (p1 + p2 )α (p2 − p1 ), (5.11)
∂t  ∂x 

∂p2 1 ∂p2 1
− = 2 (p1 + p2 )α (p1 − p2 )
∂t  ∂x 
where 0 ≤ α ≤ 1; by scaling, this system can be rewritten in the form (5.1) with kij
multiplied by (p1 + p2 )α . In this model the pi (i = 1, 2) represent the concentration
of two gases. It was established (see [38] and the references therein) that, under
appropriate initial data, pi → u as  → 0 where u satisfies the diffusion equation
 
∂u ∂ 1 ∂u
= .
∂t ∂x uα ∂x

6. Conclusion. Biological processes are typically very complex. Thus, although


the main theme of the present article is conservation laws in mathematical biology,
these laws, as we have seen, are strongly coupled to other systems, such as ODEs
with nonlocal or nonlinear coefficients, elliptic equations, and parabolic equations,
sometimes with free boundary. The questions we explored include asymptotic be-
havior of solutions as t → ∞ or as  → 0, and the behavior of the free boundary.
Historically the physical sciences motivated many mathematical ideas and theories.
Today the biological sciences have become another major source for mathematical
explorations.

REFERENCES

[1] B. V. Bazaliy and A. Friedman, A free boundary problem for an elliptic-parabolic system:
Application to a model of tumor growth, Comm. in PDE, 28 (2003), 517–560.
[2] B. Bazaliy and A. Friedman, Global existence and stability for an elliptic-parabolic free bound-
ary problem: An application to a model with tumor growth, Indiana Univ. Math. J., 52 (2003),
1265–1304.
[3] A. Brown, L. Wang and P. Jung, Stochastic simulation of neurofilament transport in axon:
‘Stop and go’ hypothesis, Molec. Biol. Cell, 16 (2005), 4243–4255.
[4] D. S. Burgess, Pharmacodynamic principles of antimicrobial therapy in the prevention of
resistance, Chest, 115 (1999), 19S–23S.
[5] H. M. Byrne, M. A. J. Chaplain, D. L. Evans and I. Hopkinson, Mathematical modelling of
angiogenesis in wound healing: Comparison of theory and experiment, J. Theor. Med., 2
(2000), 175–197.
[6] X. Chen, S. Cui and A. Friedman, A hyperbolic free boundary problem modeling tumor growth:
Asymptotic behavior , Trans. Amer. Math. Soc., 357 (2005), 4771–4804.
[7] X. Chen and A. Friedman, A free boundary problem for elliptic-hyperbolic system: An appli-
cation to tumor growth, SIAM J. Math. Anal., 35 (2003), 974–986.
[8] G. Craciun, A. Brown and A. Friedman, A dynamical system model of neurofilament transport
in axon, J. Theor. Biol., 237 (2005), 316–322.
3096 AVNER FRIEDMAN

[9] S. Cui and A. Friedman, A free boundary problem for a singular system of differential equa-
tions: An application to a model of tumor growth, Trans. Amer. Math. Soc., 355 (2003),
3537–3590.
[10] S. Cui and A. Friedman, A hyperbolic free boundary problem modeling tumor growth, Inter-
faces & Free Boundaries, 5 (2003), 159–181.
[11] E. M. C. D’Agata, M. A. Horn and G. F. Webb, The impact of persistent gastrointestinal
colonization on the transmission dynamics of vancomycin-resistant enterococci, The Journal
of Infectious Diseases, 185 (2002), 766–773.
[12] E. M. C. D’Agata, G. F. Webb and M. A. Horn, A mathematical model quantifying the impact
of antibiotic exposure and other interventions on the endemic prevalence of vancomycin-
resistant enterococci, The Journal of Infectious Diseases, 192 (2005), 2004–2011.
[13] M. Fontelos and A. Friedman, Symmetry-breaking bifurcations of free boundary problems in
three dimensions, Asymptotic Anal., 35 (2003), 187–206.
[14] A. Friedman, A free boundary problem for a coupled system of elliptic, hyperbolic, and Stokes
equations modeling tumor growth, Interfaces & Free Boundaries, 8 (2006), 247–261.
[15] A. Friedman and G. Craciun, A model of intracellular transport of particles in an axon, J.
Math. Biol., 51 (2005), 217–246.
[16] A. Friedman and G. Craciun, Approximate traveling waves in linear reaction-hyperbolic equa-
tions, SIAM J. Math. Anal., 38 (2006), 741–758.
[17] A. Friedman and B. Hu, Asymptotic stability for a free boundary problem arising in a tumor
model, J. Diff. Eqs., 227 (2006), 598–639.
[18] A. Friedman and B. Hu, Bifurcation from stability to instability for a free boundary problem
arising in a tumor model, Arch Rat. Mech. Anal., 180 (2006), 293–330.
[19] A. Friedman and B. Hu, Uniform convergence for approximate traveling waves in linear
reaction-hyperbolic systems, Indiana Univ. Math. J., 56 (2007), 2133–2158.
[20] A. Friedman and B. Hu, Bifurcation for a free boundary problem modeling tumor growth by
Stokes equation, SIAM J. Math. Anal., 39 (2007), 174–194.
[21] A. Friedman and B. Hu, Bifurcation from stability to instability for a free boundary problem
modeling tumor growth by Stokes equation, J. Math. Anal. Appl., 327 (2007), 643–664.
[22] A. Friedman and B. Hu, Stability and instability of Liapounov-Schmidt and Hopf bifurcations
for a free boundary problem arising in a tumor model, Trans. Amer. Math. Soc., 360 (2008),
5291–5342.
[23] A. Friedman, B. Hu and J. P. Keener, The diffusion approximation for linear non-autonomous
reaction-hyperbolic equations, submitted.
[24] A. Friedman, B. Hu and C. Xue, Analysis of a mathematical model of ischemic cutaneous
wounds, SIAM J. Math. Anal., 42 (2010), 2013–2040.
[25] A. Friedman, B. Hu and C. Xue, A three dimensional model of wound healing: Analysis and
computation, Discrete and Continuous Dynamical Systems Series B, to appear.
[26] A. Friedman, C.-Y. Kao and C.-W. Shih, Asymptotic phases in a cell differentiation model,
J. Diff. Eqs., 247 (2009), 736–769.
[27] A. Friedman, C.-Y. Kao and C.-W. Shih, Transcriptional control in cell differentiation: As-
ymptotic limits, submitted.
[28] A. Friedman and F. Reitich, Analysis of a mathematical model for growth of tumors, J. Math.
Biol., 38 (1999), 262–284.
[29] A. Friedman and F. Reitich, Symmetry-breaking bifurcation of analytic solutions to free
boundary problems: An application to a model of tumor growth, Trans. Amer. Math. Soc.,
353 (2001), 1587–1634.
[30] A. Friedman and C. Xue, A mathematical model for chronic wounds, Mathematical Bio-
sciences and Engineering, 8 (2011), 253–261.
[31] A. Friedman, N. Ziyadi and K. Boushaba, A model of drug resistance with infection by health
care workers, Math. Biosciences and Engineering, 7 (2010), 779–792.
[32] L. Mariani, M. Lohning, A. Radbruch and T. Hofer, Transcriptional control networks of cell
differentiation: Insights from helper T lymphocytes, Biophys. Mol. Biol., 86 (2004), 45–76.
[33] L. R. Peterson, Squeezing the antibiotic balloon: The impact of antimicrobial classes on
emerging resistance, Clin. Microbiol. Infect. 11 Suppl., 5 (2005), 4–16.
[34] G. J. Pettet, H. M. Byrne, D. L. S. McElwain and J. Norbury, A model of wound-healing
angiogenesis in soft tissue, Mathematical Biosciences, 136 (1996), 35–63.
[35] G. Pettet, M. A. J. Chaplain, D. L. S. McElwain and H. M. Byrne, On the role of angiogenesis
in wound healing, Proc. R. Soc. Lond. B, 263 (1996), 1487–1493.
CONSERVATION LAWS IN MATHEMATICAL BIOLOGY 3097

[36] G. J. Pettet, C. P. Please, M. J. Tindall and D. L. S. McElwain, The migration of cells in


multicell tumor spheroids, Bull. Math. Biol., 63 (2001), 231–257.
[37] M. C. Reed, S. Venakides and J. J. Blum, Approximate traveling waves in linear reaction-
hyperbolic equations, SIAM J. Appl. Math., 50 (1990), 167–180.
[38] F. Salvarani and J. L. Vazquez, The diffusive limit for Carleman-type kinetic models, Non-
linearity, 18 (2005), 1223–1248.
[39] R. C. Schugart, A. Friedman, R. Zhao and C. K. Sen, Wound angiogenesis as a function of
tissue oxygen tension: A mathematical model, PNAS, 105 (2008), 2628–2633.
[40] G. F. Webb, E. M. C. D’Agata, P. Magal and S. Ruan, A model of antibiotic-resistant bacterial
epidemics in hospitals, PNAS, 102 (2005), 13343–13348.
[41] J. Wu and S. Cui, Asymptotic stability of stationary solutions of a free boundary problem
modeling the growth of tumors with fluid tissues, SIAM J. Math. Anal., 41 (2009), 391–414.
[42] C. Xue, A. Friedman and C. K. Sen, A mathematical model of ischemic cutaneous wounds,
PNAS, 106 (2009), 16782–16787.
[43] A. Yates, R. Callard and J. Stark, Combining cytokine signaling with T-bet and GATA-3
regulation in Th1 and Th2 differentiation: A model for cellular decision making, J. Theor.
Biol., 231 (2004), 181–196.

Received September 2011; revised March 2012.


E-mail address: afriedman@math.ohio-state.edu

You might also like