Science in China Series A: Mathematics

Dec., 2008, Vol. 51, No. 12, 2315–2329

www.scichina.com math.scichina.com
www.springerlink.com

A mathematical model of combined therapies

against cancer using viruses and inhibitors
TAO YouShan1† & GUO Qian2
1 Department of Applied Mathematics, Dong Hua University, Shanghai 200051, China
2 Department of Mathematics, Shanghai Normal University, Shanghai 200234, China
(email: taoys@dhu.edu.cn, qguo@shnu.edu.cn)

Abstract This paper deals with a procedure for combined therapies against cancer using oncolytic
viruses and inhibitors. Replicating genetically modified adenoviruses infect cancer cells, reproduce
inside them and eventually cause their death (lysis). As infected cells die, the viruses inside them are
released and then proceed to infect other tumor cells. The successful entry of virus into cancer cells is
related to the presence of the coxsackie-adenovirus receptor (CAR). Mitogen-activated protein kinase
kinase (known as MEK) inhibitors can promote CAR expression, resulting in enhanced adenovirus entry
into cancer cells. However, MEK inhibitors can also cause G1 cell-cycle arrest, inhibiting reproduction of
the virus. To design an effective synergistic therapy, the promotion of virus infection must be optimally
balanced with inhibition of virus production. We introduce a mathematical model to describe the effects
of MEK inhibitors and viruses on tumor cells, and use it to explore the reduction of the tumor size
that can be achieved by the combined therapies. Furthermore, we find an optimal dose of inhibitor:
Poptimal = 1 − μ/δ for a certain initial density of cells (where μ is the removal rate of the dead cells
and δ is the death rate of the infected cells). The optimal timing of MEK inhibitors is also numerically
studied.
Keywords: tumors, oncolytic viruses, MEK inhibitors, mathematical modeling
MSC(2000): 35Q80, 35R35, 92A15, 92C50

1 Introduction
Over the past 30 years, a variety of mathematical models for tumor growth have been developed;
for example, cf. [1–9] and references therein.
Chemotherapy is widely used for the treatment of cancer. The effectiveness of cancer chemother-
apy is dependent on the concentration of drug reaching the tumor. However, one of the ob-
stacles in developing efficient chemotherapy against cancer is in the delivery process. The
macromolecules used as drug delivery carriers are too large to be transported into, and diffuse
within, the tumor (see [10]). Alternate approaches to therapy need to be developed.
Recently, oncolytic viruses which are genetically modified replication-competent viruses have
been proposed as an approach to bypassing the delivery problem. The virus is engineered to
selectively bind to receptors on the tumor cell surface (but not to the surface of normal healthy
cells). The virus particles then gain entry by endocytosis and proceed to reproduce within the

Received November 17, 2007; accepted January 22, 2008; published online August 30, 2008
DOI: 10.1007/s11425-008-0070-7
† Corresponding author

This work was supported by the National Natural Science Foundation of China (Grant No. 10571023)
2316 TAO YouShan & GUO Qian

tumor cell, eventually causing death (lysis). As an infected cell dies, the newly reproduced virus
particles burst out and then proceed to infect adjacent cancer cells. Mathematical models of
virotherapy have recently been developed and studied; for example, cf. [8, 9, 11].
ONYX-015 is one of oncolytic viruses tested in clinical trials and it is a genetically modified
adenovirus. Further studies in clinical trials have shown that the expression of the coxsackie-
adenovirus receptor (CAR) strongly influences the entry of virus into cancer cells (for example,
cf. [12]). Mitogen-activated protein kinase kinase (MEK, also known as MAP-kinase kinase)
inhibitors have been shown to promote CAR expression, and could result in increased ONYX-
015 entry into target cells (cf. [13] and references therein). This could lead to a novel combined
therapeutic approach to cancer, using ONYX-015 and MEK inhibitors. However, MEK in-
hibitors can cause temporary cell-cycle arrest, which inhibits the life-cycle of ONYX-015[13]
(the cell-cycle is generally considered to consist of four phases of proliferate growth: the growth
phase (G1 ), a phase of DNA synthesis (S), a period before cell division (G2 ) and mitosis (M);
and a single phase of quiescent behavior (G0 ). ONYX-015 needs to lock the cell in S-phase to
replicate and lyse the cell). So, MEK inhibitors may limit the replication of virus. To design an
effective protocol of combined therapies against cancer using ONYX-015 and MEK inhibitors,
the positive effect of MEK inhibitors should be optimally balanced with the negative effect of
MEK inhibitors. This complicates the dynamics of MEK inhibitors, viruses and tumor cells.
In this paper we develop and analyse a mathematical model of a spherical tumor treated
with ONYX-015 and MEK inhibitors. The model includes uninfected tumor cells, infected
tumor cells, necrotic tumor cells, free (extracellular) virus particles, coxsackie-adenovirus re-
ceptors, and MEK inhibitors. This model is a free boundary problem for a nonlinear system
of partial differential equations (PDEs), where the free boundary is the surface of the tumor.
The model is a generalization of Zurakowski and Wodarz’s model[13] . Their model provides
a simplified description of afore-mentioned combined therapies against cancer as spatial ho-
mogeneity assumption is supposed. However, the spatial structure of some solid tumors may
require spatially explicit PDE models. Modeling method in present paper is a combination of
modeling methods developed in [9, 11, 13]. In our PDE model, the tumor volume is modeled
as an incompressible fluid, through which cells travel via a convective field whose velocity is
u. Local changes in cell population lead to variations in the internal pressure, which in turn
induce motion of tumor cells. In this paper we find explicit parameter conditions of tumor
eradication for two different scenarios: strong inhibitors and weak inhibitors, respectively. We
also analytically find an optimal dose of inhibitor for a certain initial density of cells. The
model formulated in this paper can also be regarded as a generalization of the existing PDE
model[11] which is a model of virotherapy against cancer. The present paper deals with the
modeling of combined therapies against cancer using viruses and inhibitors, and the model al-
lows us to study possible optimal protocols of cancer therapies. Furthermore, in this paper we
remove two technical parameter assumptions (10.1) and (10.2) of Theorem 10.1 in [11]. This
extension is important, since the equilibrium point (x∗ , y∗ ) is most important in the four equilib-
rium points in [11] and the technical assumptions (10.1) and (10.2) in [11] are biologically strict.
The structure of this paper is as follows. In Section 2, we describe the model. In Section 3,
we non-dimensionalise the model and reduce the free (moving) boundary problem into a new
problem in a fixed region. In Section 4, we find the uniform equilibrium solutions of the model.
A mathematical model of combined therapies against cancer 2317

In Section 5, we study the stability for necrotic tumors and untreated tumors, respectively. In
Section 6, we analyse the stability of the uniform equilibrium solution corresponding to the
scenario of strong inhibitors, and we discuss the biological implications of this stability. In
Section 7, we study the stability of the uniform equilibrium solution corresponding to weak
inhibitors and give its biological implications. In Section 8, we analytically and numerically
study the model to explore possible optimal dose of inhibitors. In Section 9, we numerically
discuss the possible optimal timing of inhibitors. Finally, we close this article with a conclusion
section.

2 Model
The mathematical model includes the following variables:
x̂ = number density of uninfected tumor cells, i.e.,
the number of uninfected tumor cells in the unit volume (mm3 ),
ŷ = number density of infected tumor cells,
n̂ = number density of dead cells,
v̂ = number density of free virus, i.e., virus in the extracellular tissue,
q̂ = the average level of CAR on the surface of the cells,
u = the velocity field within the tumor.

The tumor volume is modeled as an incompressible fluid, through which cells travel via a
convective field whose velocity is u. The velocity field is a result of the spatio-temporal variation
due to the proliferation of uninfected cells and the removal of dead cells. Local changes in cell
population lead to variations in the internal pressure, which in turn induce motion of tumor
cells. The diameter of a virus particle is very small, compared with the typical diameter of a
tumor cell. Hence, free virus particles undergo diffusion rather than convection, as done in [11,
14, 15]. We consider a radially symmetrical tumor, so that all unknown functions depend only
on (r, t) where r is the distance from a point to the center of the tumor. The model was derived
by applying the principle of mass conservation to each of the variables and by considering
convection of cells and dispersion of virus particles. The model consists of the following system
of equations in the tumor region {r  R(t)}:
∂ x̂(r, t) 1 ∂
+ 2 [r2 u(r, t)x̂(r, t)] = λ[1 − P (t)]x̂(r, t) − β q̂(t)x̂(r, t)v̂(r, t), (1)
∂t r ∂r
∂ ŷ(r, t) 1 ∂
+ 2 [r2 u(r, t)ŷ(r, t)] = β q̂(t)x̂(r, t)v̂(r, t) − δ[1 − P (t)]ŷ(r, t), (2)
∂t r ∂r
∂ n̂(r, t) 1 ∂
+ 2 [r2 u(r, t)n̂(r, t)] = δ[1 − P (t)]ŷ(r, t) − μn̂(r, t), (3)
∂t r ∂r
 
∂v̂(r, t) 1 ∂ 2 ∂v̂(r, t)
−D 2 r = N δ[1 − P (t)]ŷ(r, t) − γv̂(r, t), (4)
∂t r ∂r ∂r
dq̂(t)
= ηP (t)[a − q̂(t)]. (5)
dt
The intensity of MEK inhibitor application is captured in the parameter P [13] . In order to
use the model to study the possible optimal timing of MEK inhibitor, we here assume that
2318 TAO YouShan & GUO Qian

P = P (t). It ranges from zero to one. We denote by q̂ the average expression level of CAR
on the surface of the cells, and it is modulated by MEK inhibitor application. So, q̂ = q̂(t),
which is assumed to be spatially homogeneous. In (1), λ denotes the net proliferation rate of
the uninfected cells in the absence of inhibitors, i.e., cell proliferation rate – cell death rate. We
neglect the effect of any externally supplied nutrients, assuming instead that, in the absence
of inhibitors, all cells proliferate at a constant rate λ (corresponding to the early stages of
avascular growth), as done in [6, 9, 11, 13, 14–21]. The proliferation of the uninfected cells can
be slowed down by the inhibitor (MEK inhibitors induce temporary cell-cycle arrest), captured
in the expression 1 − P (t) (cf. [13]). When the virus meets the uninfected cells, infection can
occur. The virus binds to CAR receptors on the uninfected cell surface. The infection rate is
thus proportional to the average number of receptors on the cell surface, q̂, the density of free
virus, v̂, and the density of uninfected cells, x̂. β is the rate constant of infection[9, 13] . In (2), δ
is the death rate of the infected cells by lysis[9, 21] . This parameter measures viral cytotoxicity.
However, MEK inhibitors induce temporary cell-cycle arrest, which inhibits the life-cycle of
virus and therefore inhibits the virus-induced cell death (virus needs to lock the cell in S-phase
to lyse the cell). Hence, the rate of virus-induced cell death is proportional to 1 − P (t) (cf. [13]).
That is, as the intensity of inhibitors is increased, the rate of virus-induced cell death decreases.
In (3), μ is the removal rate of the dead cells[9, 11] . In (4), γ is the removal (or clearance) rate
of virus[9, 11] , D is the diffusion coefficient of virus (cf. [11, 14–16, 20]) and N δ is the virus
release rate (N is the burst size of virus at the death of a cell) in the absence of inhibitors (cf.
[9, 11]). As afore-mentioned, MEK inhibitors induce temporary cell-cycle arrest, which inhibits
the life-cycle of virus (virus needs to lock the cell in S-phase to replicate). So, the virus release
rate is again diminished in the presence of the inhibitor (proportional to 1 − P (t), cf. [13]).
In (5), we assume that CAR is produced by cells with a rate η and production is proportional
to inhibitor activity, P (t). The expression a − q̂ represents the saturation of CAR expression.
That is, the cell cannot bear an infinite number of receptors (cf. [13]).
We assume that there are no voids within the tumor (cf. [6, 7, 9, 11, 14, 16, 21]), so that

x̂ + ŷ + n̂ ≡ const. ≡ θ. (6)

Summing equations (1)–(3) and invoking assumption (6) yield

θ ∂ 2
(r u(r, t)) = λ[1 − P (t)]x̂(r, t) − μn̂(r, t). (7)
r2 ∂r
We assume that the virus particles remain in the tumor, so that
∂ ∂
v̂(0, t) = 0, v̂(R(t), t) = 0, (8)
∂r ∂r
where the first equation in (8) is a consequence of the radial symmetry assumption.
Finally, the free boundary is subject to the kinematics condition
dR(t)
= u(R(t), t), (9)
dt
which is the equation of continuity: the velocity of the surface of the tumor is the same as the
velocity of cells at the surface.
A mathematical model of combined therapies against cancer 2319

We note that equation (3) is a consequence of equations (1), (2), (6) and (7), so that in the
sequel we may drop this equation and replace n̂ by θ − x̂ − ŷ in (7).
We also note that
u(0, t) = 0, (10)

which is also a consequence of the radial symmetry assumption.

To complete the statement of the free boundary problem, we impose the following initial
conditions:
R(0) is prescribed,
x̂(r, 0) = x̂0 (r), ŷ(r, 0) = ŷ0 (r), v̂(r, 0) = v̂0 (r), q̂(0) = q̂0  0,
(11)
where x̂0 (r), ŷ0 (r) and v̂0 (r) are nonnegative functions
with x̂0 (r) + ŷ0 (r)  θ, for 0  r  R(0).

Remark 2.1. If P (t) ≡ 0, then by (5) we have q̂(t) ≡ const. (> 0). This reduce the model
(1)–(11) to the model in [9, 11]. Hence, the model (1)–(11) can be regarded as an extension
of the model in [9, 11]. The model (1)–(11) is devoted to combined therapies against cancer
using viruses and inhibitors, while the model in [9, 11] deals with a therapy against cancer using
viruses alone. In fact, in Section 8 we will see that the effect of the above combined therapies
is better than the effect of the single-agent therapy using viruses.
In next section we will transform the free boundary problem (1)–(11) into a problem in a
fixed region.

3 Transformation
We introduce the variables
x̂ ŷ v̂
x̃ = , ỹ = , ṽ = , q̃ = q̂, ũ = u
θ θ θN
and the quantity p0 = βN θ
γ . The parameter p0 represents the mean number of virus particles
released by one virus (cf. [9]).
In terms of the new variables, system (1)–(11) takes the following form in {r < R(t), t > 0}:
dq̃
= ηP (t)(a − q̃), q̃(0) = q̃0  0, (12)
dt
∂ x̃ 1 ∂
= λ[1 − P (t)]x̃ − p0 γ q̃x̃ṽ − 2 (r2 ũx̃), (13)
∂t r ∂r
∂ ỹ 1 ∂
= p0 γ q̃x̃ṽ − δ[1 − P (t)]ỹ − 2 (r2 ũỹ), (14)
∂t r ∂r
x̃(r, 0) = x̃0 (r), ỹ(r, 0) = ỹ0 (r), x̃0 (r), ỹ0 (r)  0, x̃0 (r) + ỹ0 (r)  1, (15)
 
∂ṽ D ∂ 2 ∂ṽ
− 2 r = δ[1 − P (t)]ỹ − γṽ, (16)
∂t r ∂r ∂r
∂ ∂
ṽ(0, t) = 0, ṽ(R(t), t) = 0, (17)
∂r ∂r
ṽ(r, 0) = ṽ0 (r)  0, (18)
1 ∂ 2
(r ũ) = λ[1 − P (t)]x̃ − μ(1 − x̃ − ỹ), (19)
r2 ∂r
2320 TAO YouShan & GUO Qian

ũ(0, t) = 0, (20)
dR(t)
= ũ(R(t), t) for t > 0, R(0) is prescribed. (21)
dt
Maybe it is convenient to transform the region {0  r  R(t)} into the fixed region {0 
ρ  1} by ρ = ρ(r, t) = R(t)
r
. Introducing the change of variables:

x(ρ, t) = x̃(r, t), y(ρ, t) = ỹ(r, t), v(ρ, t) = ṽ(r, t),

q(t) = q̃(t), u(ρ, t) = ũ(r, t)/R(t),

system (12)–(21) takes the following forms in {0 < ρ < 1, t > 0}:
dq
= ηP (t)(a − q), q(0) = q0  0, (22)
dt
∂x ∂x
+ [u(ρ, t) − ρu(1, t)]
∂t ∂ρ
= λ[1 − P (t)]x − p0 γqxv − [−μ + (λ + μ − λP (t))x + μy]x, (23)
∂y ∂y
+ [u(ρ, t) − ρu(1, t)]
∂t ∂ρ
= p0 γqxv − δ[1 − P (t)]y − [−μ + (λ + μ − λP (t))x + μy]y, (24)

y(ρ, 0) = y0 (ρ), x0 (ρ), y0 (ρ)  0, x0 (ρ) + y0 (ρ)  1,

x(ρ, 0) = x0 (ρ), (25)
 
∂v D 1 ∂ 2 ∂v ∂v
− 2 ρ − ρu(1, t) = δ[1 − P (t)]y − γv, (26)
∂t R (t) ρ2 ∂ρ ∂ρ ∂ρ
vρ (0, t) = 0, vρ (1, t) = 0, (27)

v(ρ, 0) = v0 (ρ)  0, (28)

 ρ
1
u(ρ, t) = 2 s2 {−μ + [λ + μ − λP (t)]x(s, t) + μy(s, t)}ds, (29)
ρ 0
Ṙ(t) = R(t)u(1, t), R(0) is given. (30)

We shall also assume that

∂v0
x0 (ρ), y0 (ρ) and v0 (ρ) belong to C 1 [0, 1], and (1) = 0. (31)
∂ρ
Proceeding as in the proof of Theorem 2.1 in [11], we can obtain the following result on global
existence and uniqueness of solutions for system (22)–(31).
Theorem 3.1. System (22)–(31) has a unique solution (q(t), x(ρ, t), y(ρ, t), v(ρ, t), u(ρ, t),
R(t)) with x, ∂x/∂ρ, y, ∂y/∂ρ, v, ∂v/∂ρ, and u, ∂u/∂ρ in C[0  ρ  1, 0 < t < ∞], q(t) and
R(t) in C 1 [0, ∞), and R(0)e−αt < R(t) < R(0)eαt for some α > 0.
In Sections 4–7 we will find explicit parameter conditions for tumor eradication.

4 Uniform equilibrium solutions

Throughout this section we assume that P (t) ≡ const. = P. We then derive from (29) and (30)
that  1
dR(t)
= R(t) s2 [−μ + (λ + μ − λP )x(s, t) + μy(s, t)] ds. (32)
dt 0
A mathematical model of combined therapies against cancer 2321

Once x and y reach their uniform equilibrium states, the sign of dR(t) dt in (32) is invariant.
Therefore, the tumor’s long-term behavior can be predicted by simply analyzing the uniform
equilibrium solutions Es = (xs , ys , vs , qs ) of (22)–(24) and (26). These uniform equilibrium
solutions are obtained from the following equations:

ηP (a − qs ) = 0, (33)
λ(1 − P )xs − p0 γqs xs vs − [−μ + (λ + μ − λP )xs + μys ]xs = 0, (34)
p0 γqs xs vs − δ(1 − P )ys − [−μ + (λ + μ − λP )xs + μys ]ys = 0, (35)
δ(1 − P )ys − γvs = 0. (36)

Writing
p̄0 = p0 a, P̄ = 1 − P. (37)

We easily find that there are precisely four uniform equilibrium solutions:

E1 = (0, 0, 0, a), (38)

E2 = (1, 0, 0, a), (39)
   
δ δ δ
E3 = 0, 1 − P̄ , P̄ 1 − P̄ , a provided δ P̄ < μ, (40)
μ γ μ
 
δ
E4 = x∗ , y∗ , P̄ y∗ , a , (41)
γ

where
λμ − p̄0 δμ + p̄0 δ 2 P̄ + μδ (λP̄ + μ)(p̄0 δ − δ − λ)
x∗ = , y∗ = (42)
(p̄0 δ − λ)p̄0 δ P̄ (p̄0 δ − λ)p̄0 δ P̄
provided x∗  0, y∗  0.
By (33)–(36), we find that (x∗ , y∗ ) satisfies

λP̄ + μ − (λP̄ + μ)x∗ − (p̄0 δ P̄ + μ)y∗ = 0,
(43)
μ − δ P̄ + (p̄0 δ P̄ − λP̄ − μ)x∗ − μy∗ = 0,

which will be used later. In the analysis of the stability for the equilibrium solution E4 , we will
assume
λ λ
p̄0 > 1 + P̄ + , (44)
μ δ
which holds for the typical parameter values given in [9, 21] and 0  P̄  1.
Finally, we remark that if (44) holds and

δ P̄  μ, (45)

then by (42) we have x∗ > 0, y∗ > 0.

5 Stability for necrotic tumors and untreated tumors

For the equilibrium solution E1 , we have ns ≡ 1−xs −ys = 1, which corresponds to the necrotic
tumor; for the equilibrium solution E2 , we have xs = 1 and ys = vs = 0, which corresponds to
the untreated tumor. In this section we will discuss the stability for the equilibrium solutions
2322 TAO YouShan & GUO Qian

E1 and E2 . We denote the right-hand terms of equations (23) and (24) by f1 (x, y, v, q) and
f2 (x, y, v, q), respectively. By direct calculations we have

∂f1 
 = λP̄ + μ − 2(λP̄ + μ)xs − μys − p̄0 δ P̄ ys , (46)
∂x Es
∂f1 
 = −μxs , (47)
∂y Es
∂f2 
 = p̄0 δ P̄ ys − (λP̄ + μ)ys , (48)
∂x Es
∂f2 
 = −[δ P̄ − μ + (λP̄ + μ)xs + 2μys ]. (49)
∂y Es

We begin with considering the equilibrium E1 . By (46)–(49), the linearized coefficient matrix
A1 of the right-hand terms of hyperbolic system (23)–(24) at the equilibrium point E1 has the
form: ⎛ ⎞
∂f1 ∂f1 ⎛ ⎞
⎜ ∂x ∂y ⎟ ⎟ λ P̄ + μ 0
A1 = ⎜⎝ ∂f2 ∂f2 ⎠ =
⎝ ⎠.
0 μ − δ P̄
∂x ∂y E1

Clearly, A1 has at least one positive eigenvalue λP̄ + μ. Therefore, we have

Theorem 5.1. The equilibrium solution E1 is linearly unstable.

Remark 5.1. Mathematically, the instability of the equilibrium solution E1 may imply that
there exists an ε0 > 0, such that

x(·, t)C[0,1] + y(·, t)C[0,1]  ε0 as t → +∞,

where (x(ρ, t), y(ρ, t)), together with v(ρ, t) and q(t), is the solution to system (22)–(31) with
some initial condition (x(ρ, 0), y(ρ, 0)) satisfying x(·, 0)C[0,1] + y(·, 0)C[0,1] < δ (here δ > 0
is arbitrarily small). However, “ x(·, t)C[0,1] + y(·, t)C[0,1]  ε0 as t → +∞ ” biologically
means that the necrotic tumor revives as t large enough.
We next turn to study the stability of the equilibrium solution E2 . By (46)–(49), the lin-
earized coefficient matrix A2 of the right-hand terms of hyperbolic system (23)–(24) at the
equilibrium point E2 has the form:
⎛ ⎞
∂f1 ∂f1 ⎛ ⎞
⎜ ∂x ∂y ⎟⎟ −(λ P̄ + μ) −μ
A2 = ⎜
⎝ ∂f2 ∂f2 ⎠ =
⎝ ⎠.
0 −(λ + δ)P̄
∂x ∂y E2

We have
Theorem 5.2. Assume that
0  P < 1. (50)

Then the equilibrium solution E2 to system (22)–(31) is locally stable.

Proof. Under assumption (50), the two eigenvalues of matrix A2 are negative. Therefore, the
equilibrium solution E2 to system (22)–(31) is linearly stable. Furthermore, proceeding as in
A mathematical model of combined therapies against cancer 2323

the proofs of Lemma 6.2, Theorem 7.1 and Theorem 8.1 in [11], we can get the locally nonlinear
stability of the equilibrium solution E2 to system (22)–(31). We complete this proof.
Remark 5.2. Theorem 5.2 implies that if the MEK inhibitor is not too strong (i.e., (50)
holds), then the tumor without injection of virus will continually grows. In fact, by (29), (30)
and Theorem 5.2 we have
 1
Ṙ(t) λP̄
= s2 [−μ + (λP̄ + μ) + O(ε)] ds = + O(ε),
R(t) 0 3

where ε is sufficiently small. So,

λP̄
R(t) = R(0)e( 3 +O(ε))t
→ +∞ as t → +∞.

However, if the MEK inhibitor is sufficiently strong that P = 1, then the cell-cycle stops and
therefore the proliferation of tumor cells arrest, which leads to the stop of tumor growth. In
fact, if P = 1, then by (29), (30) and x(ρ, t) + y(ρ, t)  1 we have
 1
Ṙ(t)
= s2 [−μ + μ(x(s, t) + y(s, t))] ds  0,
R(t) 0

that is Ṙ(t)  0.

6 Strong inhibitor case: δ(1 − P ) < μ

For the typical parameter values: δ = 1
48 hr−1 and μ = 1
72 hr−1 given in [21], we have

δ > μ. (51)

Throughout this section we assume that the intensity of the inhibitor is so strong that

δ(1 − P ) < μ ⇐⇒ δ P̄ < μ. (52)

Under the assumption (52), the equilibrium solution E3 exists. In this section we will study
the stability of E3 . By (46)–(49), the linearized coefficient matrix A3 of the right-hand terms
of hyperbolic system (23)–(24) at the equilibrium point E3 has the form:
⎛ ⎞
∂f1 ∂f1 ⎛ ⎞
⎜ ∂x ∂y ⎟ (λ + δ)P̄ − p̄0 δ P̄ (1 − μδ P̄ ) 0
A3 = ⎜⎝ ∂f2 ∂f2 ⎠ =
⎟ ⎝ ⎠.
(p̄0 δ P̄ − λP̄ − μ)(1 − μδ P̄ ) −(μ − δ P̄ )
∂x ∂y E3

In addition to (52), throughout this section we also assume that

(λ + δ)μ
p̄0 > . (53)
δ(μ − δ P̄ )

Clearly, we conclude from the assumptions (52) and (53) that the two eigenvalues of matrix
A3 are negative, and therefore the equilibrium solution E3 is linearly stable. Furthermore,
proceeding as in the proofs of Lemma 6.2, Theorem 7.1 and Theorem 8.1 in [11], we can get the
locally nonlinear stability of the equilibrium solution E3 to system (22)–(31). More precisely,
we have (cf. Theorem 9.1 in [11])
2324 TAO YouShan & GUO Qian

Theorem 6.1. Assume that (52) and (53) hold, and let R(0) be arbitrary positive number.
If     
 
x0 (ρ), y0 (ρ) − 1 − δ P̄ , v0 (ρ) − δ P̄ 1 − δ P̄  (54)
 μ γ μ  1
C [0,1]

is sufficiently small, then Ṙ(t) < 0 for all t > 0 and

R(t)  R(0)e−αt as t → ∞, for some α > 0.

Theorem 6.1 suggests that no matter how large the initial tumor is, one may reduce its size
at exponential rate by injecting a dose of MEK inhibitor close to P and a dose of virus close to
γ P̄ (1 − μ P̄ ), provided initially the density of the infected cells is approximately (1 − μ P̄ ) and
δ δ δ

the density of uninfected cell is sufficiently small.

Remark 6.1. In Theorem 6.1 we assume that R(0) is arbitrary positive number. This
assumption is valid only in the sense of mathematics. Indeed, if the initial tumor is too large,
then the host cannot be alive in the practical world of biology. Therefore, the “arbitrary
positive number” in Theorem 6.1 should be regarded as “arbitrary positive number within
some reasonable range”.
Remark 6.2. For any q0  0 and P (t) ≡ P > 0, by (22) we have

q(t) = a + (q0 − a)e−ηP t → a as t → +∞.

Remark 6.3. To reduce the tumor, the parameter condition (53) implies that the requirement
of the replication ability of the virus strongly depends on the intensity of the MEK inhibitor.
That is, the stronger the inhibitor activity, the lower requirement of the replication ability of
the virus (this may be due to the higher rate of virus infection).

7 Weak inhibitor case: δ(1 − P )  μ

For the typical parameter δ and μ given in [21], the inequality (51) holds. Throughout this
section we assume that the intensity of the inhibitor is not sufficiently strong, so that

δ(1 − P )  μ ⇐⇒ δ P̄  μ. (55)

To study the stability of the equilibrium solution E4 , throughout this section we also assume
that
λ λ
p̄0 > 1 + P̄ + . (56)
μ δ
Combining (46)–(49) and (43), we find that the linearized coefficient matrix A4 of the right-hand
terms of hyperbolic system (23)–(24) at the equilibrium point E4 has the form:
⎛ ⎞
∂f1 ∂f1 ⎛ ⎞
⎜ ∂x ∂y ⎟ ⎟ −(λ P̄ + μ)x∗ −μx ∗
A4 = ⎜
⎝ ∂f2 ∂f2 ⎠ =
⎝ ⎠.
(p̄0 δ P̄ − λP̄ − μ)y∗ −p̄0 δ P̄ x∗ − μy∗
∂x ∂y E4

By (42), (55) and (56) we easily check that

x∗ > 0, y∗ > 0, (57)

A mathematical model of combined therapies against cancer 2325

and therefore

T rA4 = −(p̄0 δ P̄ + λP̄ + μ)x∗ − μy∗ < 0, (58)

|A4 | = (λP̄ + μ)x∗ (p̄0 δ P̄ x∗ + μy∗ ) + μx∗ (p̄0 δ P̄ − λP̄ − μ)y∗ (59)

= (λP̄ + μ)p̄0 δ P̄ x2∗ + μp̄0 δ P̄ x∗ y∗ > 0.

We conclude from (58) and (59) that the two eigenvalues of matrix A4 have negative real
parts, and therefore the equilibrium solution E4 is linearly stable. Furthermore, proceeding
as in the proofs of Lemma 6.2, Theorem 7.1 and Theorem 8.1 in [11], we can get the locally
nonlinear stability of the equilibrium solution E4 to system (22)–(31). More precisely, we have
(see Theorem 9.1 in [11])
Theorem 7.1. Assume that (55) and (56) hold, and let R(0) be arbitrary positive number.
If  
 
x0 (ρ) − x∗ , y0 (ρ) − y∗ , v0 (ρ) − δ P̄ y∗  (60)
 γ  1
C [0,1]

is sufficiently small, then Ṙ(t) < 0 for all t > 0 and

R(t)  R(0)e−αt as t → ∞, for some α > 0.

Proof. We derive from (29), (30) and the locally nonlinear stability of the equilibrium solution
E4 that
 1
Ṙ(t)
= s2 [−μ + (λP̄ + μ)x(s, t) + μy(s, t)] ds
R(t) 0
 1
= s2 [−μ + (λP̄ + μ)x∗ + μy∗ + ε] ds (ε is sufficiently small)
0
 
δμ λ λ 1
=− p̄0 − 1 − P̄ − + ε (for sufficiently large t)
3(p̄0 δ − λ) μ δ 3
 
δμ λ λ
− p̄0 − 1 − P̄ − := −α < 0 (by (56)).
6(p̄0 δ − λ) μ δ

We complete this proof.

Theorem 7.1 suggests that no matter how large the initial tumor is, one may reduce its size
at exponential rate by injecting a dose of MEK inhibitor close to P and a dose of virus close to
δ
γ P̄ y∗ , provided initially the density of the infected cells is approximately y∗ and the density of
uninfected cell is approximately x∗ .
Remark 7.1. Theorem 7.1 extends the results of Theorem 10.1 in [11] in the sense that this
new result includes the MEK inhibitor application. Furthermore, in the present paper we drop
two technical parameter assumptions (10.1) and (10.2) in [11]. This extension is important,
because the equilibrium point (x∗ , y∗ ) is most practical in the four equilibrium points in [11]
and the technical assumptions (10.1) and (10.2) in [11] are biologically strict.

8 Optimal dose of inhibitors

Let us first state an important analytical result on an optimal dose of inhibitors.
2326 TAO YouShan & GUO Qian

Theorem 8.1. Under assumptions (55) and (56),

μ
Poptimal := 1 − (61)
δ
is an optimal dose of inhibitor for tumor treatment using virus and MEK inhibitor, provided
initially the density of the infected cells is approximately y∗ , the density of uninfected cells is
approximately x∗ and the density of virus is approximately δ P̄ y∗ /γ (where (x∗ , y∗ ) is given by
(42) with P̄ = μ/δ).

Proof. Since Theorem 7.1 establishes the nonlinear locally stability of the equilibrium solution
E4 under assumptions (55) and (56), we only need to consider (x(ρ, t), y(ρ, t), v(ρ, t), q(t)) ≡ E4
for the proof of Theorem 8.1.
Define a set of the number P as follows:

Θ = {0  P  1| P satisfies assumptions (55) and (56)}.

For the typical parameter values: δ = 48 1

hr−1 , μ = 72
1
hr−1 , λ = 3.2 × 10−4 hr−1 , p0 = 3.73
given in [21] and a = 2 given in [13], we find that 0 ∈ Θ. Hence, Θ is non-empty.
We note that for any given P ∈ Θ, there exists a unique uniform equilibrium solution E4
given by (41) and (42), which corresponds to a successful therapy by Theorem 7.1. For this
uniform equilibrium solution E4 which varies with P ∈ Θ, we derive from (29), (30) and (42)
that  
Ṙ(t) 1 δμ λ λ
= [−μ + (λP̄ + μ)x∗ + μy∗ ] = − p̄0 − 1 − P̄ − . (62)
R(t) 3 3(p̄0 δ − λ) μ δ
By assumption (56), we find that
Ṙ(t)|P ∈Θ < 0. (63)
Since the effect of tumor treatment can be estimated by R(t) and R(t) varies with P ∈ Θ by
(62) (here we assume that other parameters in (62) are given), we raise the following interesting
question:
Is there a P∗ ∈ Θ such that

R(t)|P =P∗ = min R(t) (for large t) (64)

P ∈Θ

holds?
If the answer is “Yes”, we conclude from (63) and (64) that the effect of the therapy cor-
responding to P = P∗ is optimal compared with these effects of therapies corresponding to
P ∈ Θ. Therefore, we denote Poptimal := P∗ .
We easily derive from the formula (62), the optimal condition (64) and P ∈ Θ that
μ
P̄optimal ≡ 1 − Poptimal = ,
δ
where we assume that all the parameters except P in (62) are given. We complete this proof.
In the following we will numerically study the model (22)–(31). The typical parameter values
for our numerical work are (cf. [15, 21, 22])

η = 1.2, a = 2, q0 = 0.5, λ = 3.2 × 10−4 , μ = 1/72,

δ = 1/48, γ = 1, p0 = 3.73, D = 1, R(0) = 1.
A mathematical model of combined therapies against cancer 2327

For the above typical parameters, Poptimal and (x∗ , y∗ , v∗ ) in Theorem 8.1 take the following
approximate values:

Poptimal ≈ 0.333; (x∗ , y∗ , v∗ ) ≈ (0.0183, 0.1178, 0.0016). (65)

Theorem 8.1 finds an optimal dose of inhibitor if the initial density of (x, y, v) is near the
point (x∗ , y∗ , v∗ ) which varies with P ∈ Θ. We shall also ask the following question:
Given a general initial data (x0 , y0 , v0 ) which is independent of P (0  P  1), does there
exist a Poptimal such that

R(t)|P =Poptimal = min R(t) (for large t) (66)

0P 1

holds (here we assume that all the parameters except P in the model (22)–(31) are given)?
According to our numerical simulations (Figure 1 as an example), we have a partial answer
to this question when the general initial data (x0 , y0 , v0 ) stays within some range.
Figure 1 shows the effects of the different doses of MEK inhibitor on tumor growth for
(x0 , y0 , v0 ) = (0.02, 0.15, 0.002). Compared with the above analytical result (65), our numerical
simulations (Figure 1) show that the optimal dose of MEK inhibitor depends on the initial
densities of cells and virus, and that the dose of MEK inhibitor should be intelligently chosen
to achieve a balance between the promotion of virus infection and inhibition of virus production.

9 Optimal timing of inhibitors

In this section we will numerically study the effects of different timings of MEK inhibitor
application on tumor growth. Take (x0 , y0 , v0 ) close to (x∗ , y∗ , v∗ ), P (t) ≡ 0 for 0  t < tm and
P (t) ≡ 1 − μ/δ for tm  t  T in the system (22)–(31) (where tm is the start timing of MEK
inhibitor application after injection of virus). Our numerical simulation (see Figure 2) shows
that there exists an optimal timing topt
m = 0 (hours) for MEK inhibitor application. That is, to
achieve a better effect of tumor treatment with virus and inhibitor, the virus and the inhibitor
should be injected into the tumor at the same time.

Figure 1 The effects of the different doses of MEK inhibitor on tumor growth for (x0 , y0 , v0 ) = (0.02, 0.15, 0.002).
The simulations suggest that Poptimal ≈ 0.268.
2328 TAO YouShan & GUO Qian

Figure 2 The effects of the different timings of MEK inhibitor on tumor growth for P = 0.3333 and
μ
(x0 , y0 , v0 ) = (0.018, 0.118, 0.0016) which are close to Poptimal = 1 − δ
and (x∗ , y∗ , v∗ ), respectively. The
optimal timing topt
m = 0.

10 Conclusion
Many free boundary problems (FBPs) have arisen from physics (for example, cf. [22] and
references therein), however, nowadays some new FBPs come from biology. In this paper we
have developed an FBP model describing the effects of MEK inhibitors and viruses on tumor
cells, which is a generalization of the existing ODE model[13] . MEK inhibitor may promote
CAR expression, and then could increase the infection of virus. However, MEK inhibitor can
cause G1 cell-cycle arrest, and therefore inhibit the reproduction of virus. So, the concentration
of MEK inhibitor should be intelligently manipulated. Our mathematical model is formulated
as a moving boundary problem for a system of partial differential equations. The unknowns are
the radius of the tumor r = R(t), the evolving densities x(r, t), y(r, t), n(r, t), v(r, t) and q(t)
of the uninfected cells, the infected cells, the dead cells, the virus particles and the coxsackie-
adenovirus receptors, and the velocity field u(r, t). The model presented in this paper is also a
generalization of the existing PDE model in [11] which is a model of virotherapy. The present
paper deals with the modelling of a combined action of virus injection and inhibitor application.
For any concentration P (0  P < 1) of inhibitor, we can find a pair of initial densities
x(r, 0) ∼ const. = xs , y(r, 0) ∼ const. = ys

such that the following is true (see Theorems 6.1 and 7.1):
Given any initial radius R(0), by injecting virus particles of density
δ
v(r, 0) ∼ const. = vs where vs = (1 − P )ys
γ
and manipulating MEK inhibitor of concentration P , the radius R(t) will decrease monotoni-
cally and exponentially to zero as t increases to ∞.
We further study possible optimal dose of MEK inhibitor (in this paper we did not consider
possible adverse side effects of virus and MEK inhibitor). Our mathematical analysis (Theo-
rem 8.1) suggests that to achieve a better effect of tumor treatment with virus and inhibitor,
A mathematical model of combined therapies against cancer 2329

we may manipulate the initial densities x(r, 0) ∼ x∗ , y(r, 0) ∼ y∗ (where (x∗ , y∗ ) is given by
(42) with P̄ = μ/δ), then we manipulate MEK inhibitor of concentration P ∼ 1 − μ/δ and
injecting virus particles of density v(r, 0) ∼ δ P̄ y∗ /γ.
We also numerically study possible optimal timing of MEK inhibitor. Our simulation suggests
that to achieve a better effect of tumor treatment with virus and inhibitor, the virus and the
inhibitor should be injected into the tumor at the same time.
Since chemotherapy and radiotherapy are widely used for cancer treatment, there are some
model extensions that are of interest. Modeling a combined action of chemo- and viro- therapy
is an interesting problem. Modeling a combined action of chemo-, radio- and viro- therapy is
another interesting problem.
References
1 Adam J, Bellomo N. A Survey of Models for Tumor-Immune System Dynamics. Boston: Birkhäuser, 1997
2 Bellomo N, Forni G. Looking for new paradigms towards a biological-mathematical theory of complex
multicellular systems. Math Models Methods Appl Sci, 16: 1001–1029 (2006)
3 Bertuzzi A, Fasano A, Gandolfi A. A mathematical model for tumor cords incorporating the flow of inter-
stitial fluids. Math Models Methods Appl Sci, 15: 1735–1778 (2005)
4 Chaplain M A J, Lolas G. Mathematical modelling of cancer cell invasion of tissue: The role of the urokinase
plasminogen activation system. Math Models Methods Appl Sci, 15: 1685–1734 (2005)
5 Dingli D, Cascino M D, Josić K, et al. Mathematical modeling of cancer radiovirotherapy. Math Biosci,
199: 80–103 (2006)
6 Jackson T L, Byrne H M. A mathematical model to study the effects of drug resistance and vasculature on
the response of solid tumors to chemotherapy. Math Biosci, 164: 17–38 (2000)
7 Ward J P, King J R. Mathematical modelling of drug transport in tumour multicell spheroids and monolayer
cultures. Math Biosci, 181: 177–207 (2003)
8 Wodarz D. Viruses as antitumor weapons: defining conditions for tumor remission. Cancer Res, 61: 3501–
3507 (2001)
9 Wu J T, Byrne H M, Kirn D H, et al. Modeling and analysis of a virus that replicates selectively in tumor
cells. Bull Math Biol, 63: 731–768 (2001)
10 Jain R. Barriers to drug delivery in solid tumors. Sci Am, 271: 58–65 (1994)
11 Friedman A, Tao Y. Analysis of a model of a virus that replicates selectively in tumor cells. J Math Biol,
47: 391–423 (2003)
12 Bergelson J M, Cunningham J A, Droguett G, et al. Isolation of a common receptor for Coxsackie B viruses
and adenoviruses 2 and 5. Science, 275: 1320-1323 (1997)
13 Zurakowski R, Wodarz D. Model-driven approaches for in vitro combination therapy using ONYX-015
replicating oncolytic adenovirus. J Theoret Biol, 245: 1–8 (2007)
14 Friedman A, Tian J P, Fulci G, et al. Glioma virotherapy: Effects of innate immune suppression and
increased viral replication capacity. Cancer Res, 66: 2314–2319 (2006)
15 Tao Y, Guo Q. The competitive dynamics between tumor cells, a replication-competent virus and an immune
response. J Math Biol, 51: 37–74 (2005)
16 Norris E S, King J R, Byrne H M. Modelling the response of spatially structured tumours to chemotherapy:
Drug kinetics. Math Comput Modelling, 43: 820–837 (2006)
17 Tao Y, Chen M. An elliptic-hyperbolic free boundary problem modelling cancer therapy. Nonlinearity, 19:
419–440 (2006)
18 Tao Y, Guo Q. A free boundary problem modelling cancer radiovirotherapy. Math Models Methods Appl
Sci, 17: 1241–1259 (2007)
19 Tao Y, Yoshida N, Guo Q. Nonlinear analysis of a model of vascular tumour growth and treatment. Non-
linearity, 17: 867–895 (2004)
20 Tao Y, Zhang H. A parabolic-hyperbolic free boundary problem modelling tumor treatment with virus.
Math Models Methods Appl Sci, 17: 63–80 (2007)
21 Wu J T, Kirn D H, Wein L M. Analysis of a three-way race between tumor growth, a replication-competent
virus and an immune response. Bull Math Biol, 66: 605–625 (2004)
22 Yi F, Tao Y, Liu Z. Quasi-stationary Stefan problem as limit case of Mullins-Sekerka problem. Sci China
Ser A-Math, 40(2): 151–162 (1997)