Mathematical Biology

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 22

Mathematical Biology

J. Math. Biol. (2014) 69:817–838


DOI 10.1007/s00285-013-0718-y

A partial differential equation model and its reduction


to an ordinary differential equation model for prostate
tumor growth under intermittent hormone therapy

Youshan Tao · Qian Guo · Kazuyuki Aihara

Received: 7 June 2013 / Revised: 6 August 2013 / Published online: 28 August 2013
© Springer-Verlag Berlin Heidelberg 2013

Abstract Hormonal therapy with androgen suppression is a common treatment for


advanced prostate tumors. The emergence of androgen-independent cells, however,
leads to a tumor relapse under a condition of long-term androgen deprivation. Clinical
trials suggest that intermittent androgen suppression (IAS) with alternating on- and
off-treatment periods can delay the relapse when compared with continuous androgen
suppression (CAS). In this paper, we propose a mathematical model for prostate tumor
growth under IAS therapy. The model elucidates initial hormone sensitivity, an even-
tual relapse of a tumor under CAS therapy, and a delay of a relapse under IAS therapy,
which are due to the coexistence of androgen-dependent cells, androgen-independent
cells resulting from reversible changes by adaptation, and androgen-independent cells
resulting from irreversible changes by genetic mutations. The model is formulated as
a free boundary problem of partial differential equations that describe the evolution
of populations of the abovementioned three types of cells during on-treatment periods
and off-treatment periods. Moreover, the model can be transformed into a piecewise
linear ordinary differential equation model by introducing three new volume variables,
and the study of the resulting model may help to devise optimal IAS schedules.

Y. Tao (B)
Department of Applied Mathematics, Dong Hua University, Shanghai 200051,
People’s Republic of China
e-mail: taoys@dhu.edu.cn

Q. Guo
Department of Mathematics, Shanghai Normal University, Shanghai 200234,
People’s Republic of China
e-mail: qguo@shnu.edu.cn

K. Aihara
Institute of Industrial Science, The University of Tokyo, Tokyo 153-8505, Japan
e-mail: aihara@sat.t.u.-tokyo.ac.jp

123
818 Y. Tao et al.

Keywords Prostate cancer · Androgen independence · Intermittent androgen


suppression · Partial differential equations · Free boundary problem · Hybrid system

Mathametics Subject Classification 35B40 · 35Q92 · 35R35 · 92C60

1 Introduction

Hormone therapy with androgen deprivation is a major treatment for advanced prostate
cancer. Androgen suppression usually induces apoptosis of androgen-dependent (AD)
cancer cells, inhibits the increase of AD cells, and thereby results in temporal regression
of a prostate tumor. However, if continuous androgen suppression(CAS) is overly pro-
longed, the AD cells often change to androgen-independent (AI) cancer cells through
some biological mechanisms, such as genetic mutation (see Feldman and Feldman
2001) and adaptation (see Rennie et al. 2005). The emergence of AI cells leads to
a tumor relapse, because these AI cells are likely to increase even under androgen-
deprived conditions (see Rennie et al. 1990; Bladou et al. 1996). To understand the
mechanism of a relapse and examine the effect of CAS therapy, Jackson (2004a) ini-
tially proposed a mathematical model that describes the growth of a prostate tumor
under CAS. Her model reproduced well the three experimentally observed phases of
tumor progression; that is, the growth prior to treatments, the regression of a tumor
immediately following androgen deprivation, and the eventual AI relapse of the tumor.
However, the analysis of the model shows that CAS can only be effective in avoiding
a relapse for a small range of biological parameters (see Jackson 2004b). Jackson’s
model was recently extended to include genetic mutation and mutation inhibitors (see
Tanaka et al. 2010a; Tao et al. 2010b), and the analysis suggests that controlling
mutation may prevent a relapse.
Since androgen deprivation can inhibit the increase of AD cells in a tumor and
the proliferation rate of AI cells under androgen-rich conditions can be smaller than
that under androgen-poor conditions (see Jackson 2004a) intermittent androgen sup-
pression (IAS) can be comparable to CAS. Actually, Bruchovsky et al. (1990) first
proposed IAS in animal models, and since then the clinical efficacy of IAS for human
patients has been explored by many researchers (see Abrahamsson 2010; Bruchovsky
et al. 2000, 2006, 2007). Nevertheless, how optimally to schedule the on- and off-
treatment periods in IAS therapy remains controversial. To compare the effects of
CAS and IAS therapies, Ideta et al. (2008) proposed a mathematical model describing
the growth of a prostate tumor under IAS. The model is a hybrid ordinary differential
equation (ODE) system with two kinds of cancer cells, namely AD cells and AI ones.
Depending on the parameter values, the model can reflect three different kinds of
dynamics, i.e., a relapse under CAS, a delay of a relapse under IAS, and prevention
of a relapse under IAS. Ideta et al’s model (Ideta et al. 2008) can be extended into a
switched partial differential equation (PDE) model by considering the volume of the
tumor (see Guo et al. 2008). The numerical analysis in the article by (Guo et al. 2008)
suggests some possibly optimal protocols for IAS therapy.
A good biomarker for prostate cancer is serum prostate-specific antigen (PSA),
which can be related to the total number of cancer cells, or the size of a tumor. An
initiation or suspension of treatment in the IAS therapy is based on the serum PSA

123
A partial differential equation model and its reduction 819

level. On the one hand, although Ideta et al’s model can describe well the dynamics
of a prostate tumor under IAS, it is unable to reproduce the biphasic decline of serum
PSA during the on-treatment periods. On the other hand, from a mathematical per-
spective, the question of well-posedness for the hybrid ODE model Ideta et al. (Ideta
et al. 2008) or the switched PDE model Guo et al. (2008) largely remains open due
to the discontinuity of the switching function. For the two reasons above, Ideta et al’s
model needs improvement. To describe the biphasic decline of serum PSA during
on-treatment periods, Hitara et al. (Hirata et al. 2010a) proposed a piecewise linear
ODE model for IAS therapy. According to some biological mechanisms, the AI cell
population is divided into two sub-populations, namely, irreversible AI cells result-
ing from changes due to genetic mutations (see Feldman and Feldman 2001), and
reversible AI cells resulting from changes due to adaptation (see Rennie et al. 2005).
The reversible AI cells may revert to AD cells during off-treatment periods. The coex-
istence of reversible and irreversible changes possibly provides the best explanation
of early progression to androgen independence.
The present work aims to establish a PDE model including reversible and irre-
versible AI cells under IAS therapy by considering the volume of a tumor. We will
prove rigorously that the new PDE model proposed can be reduced to a hybrid dynam-
ical ODE model Hirata et al. (2010a) by introducing three new volume variables. By
solving the hybrid system for the on-treatment periods and the off-treatment periods,
respectively, we obtain an explicit formula for the tumor volume at any time, which
should be useful to fit the model to clinical PSA data (see Hirata et al. 2010a; Tanaka
et al. 2010a; Tao et al. 2010b) and devise possibly optimal IAS schedules (see Hirata
et al. 2010a; Suzuki et al. 2010).
Before concluding this introduction section, we remark that there is now mounting
agreement that mathematical modeling offers a helpful approach in the study of the
growth of a cancer, since it can be used not only to generate and categorize hypotheses
but also to design therapeutic strategies (see Byrne et al. 2006; Bellomo et al. 2008).
Moreover, the dynamical behavior of tumor growth is very complex and mathemati-
cally challenging (e.g., see Friedman 2007; Friedman and Hu 2006; Tello 2013).

2 Mathematical model

To predict or approximate the dynamical behavior of a tumor, we propose a continuum


model that can reflect a tumor relapse under CAS therapy, explain a delay of a relapse
under IAS therapy, and thus possibly suggest an optimal protocol for IAS therapy.
This approach for modeling solid tumor growth uses partial differential equations to
describe the spatiotemporal evolution of cell populations and is well documented in
the literature (e.g., see Adam 1986; Greenspan 1972; Jackson 2004a; Ward and King
2003; Friedman 2007; Tao et al. 2009).
To reduce the dimension of the system, we follow Hirata et al. (2010a) and Tanaka
et al. (2010a, 2010b) by assuming that the androgen level instantaneously switches
between the minimum level and the normal level at the moments of resumption and
suspension of treatment. The epigenetic mechanisms (see Rennie et al. 2005) and
mutation mechanisms (see Feldman and Feldman 2001) may provide a good explana-

123
820 Y. Tao et al.

(a) (b)

Fig. 1 States 1, 2, and 3 correspond to AD cells, reversible AI cells, and irreversible AI cells, respectively
(see Hirata et al. 2010a). The cancer cells of state 1 may change into cells of states 2 and 3, and the cancer
cells of state 2 may change into cells of state 3 during an on-treatment period. However, during an off-
treatment period, the cancer cells of state 2 may revert to cells of state 1, but the mutated cancer cells of
state 3 changes into cells of neither state 1 nor state 2

tion of the emergence of androgen independence in prostate cancers under hormone


therapy. Based on these mechanisms, we consider two types of changes for AI cells,
namely, reversible changes and irreversible changes.
As illustrated by Figure 1, a prostate tumor consists of three types of cells: AD cells
(state 1), reversible AI cells (state 2), and irreversible AI cells (state 3) (see Hirata et al.
2010a). During an on-treatment period, cells of state 1 may change into cells of state 2
through adaptation or into cells of state 3 through genetic mutations, and cells of state
2 may change into cells of state 3 through genetic mutations. During an off-treatment
period, cells of state 2 may revert to cells of state 1 due to the reversibility, whereas
the irreversible AI cells of state 3 can revert to neither cells of state 1 nor ones of 2
because some genes are mutated.
Following Jackson (Jackson 2004a,b) to formulate the mathematical model of
prostate tumor growth under hormone therapy, the prostate tumor is regarded as a
densely packed radially symmetric sphere of radius R(t) centered at the origin. We
exploit the spherical symmetry of the problem, letting x1 (r, t), x2 (r, t), and x3 (r, t)
be the volume fractions of the populations of AD cells, reversible AI cells, and irre-
versible AI cells, respectively. We further assume that tumor cells migrate through
a single convective velocity field u created by local volume changes driven by cell
proliferation and death (see Jackson 2004a; Ward and King 2003).
Based on the assumption that the system is meant to obey the no-void condition
(e.g., see Jackson 2004a; Ward and King 2003), and by the definitions of x1 , x2 , and
x3 , we have

x1 (r, t) + x2 (r, t) + x3 (r, t) ≡ 1 for all 0 ≤ r ≤ R(t), t ≥ 0. (2.1)

The cell-population dynamics for an on-treatment period is given as


⎧  
⎪ ∂ x1 1 ∂
⎪ ∂t (r, t) + r 2 u(r, t)x1 (r, t) = (a1,1
1 − a 1 − a 1 )x (r, t),
⎪ r 2 ∂r 3,1 1

⎨  
2,1
∂ x2 1 ∂
∂t (r, t) + r 2 u(r, t)x2 (r, t) = a2,1
1 x (r, t) + (a 1 − a 1 )x (r, t), (2.2)
r 2 ∂r 1 3,2 2

⎪  
2,2


⎩ ∂ x3 (r, t) + 1 ∂
r 2 u(r, t)x3 (r, t) = a3,1
1 x (r, t) + a 1 x (r, t) + a 1 x (r, t),
1
∂t r 2 ∂r 3,2 2 3,3 3

123
A partial differential equation model and its reduction 821

for 0 < r < R(t), t > 0, where u(r, t) := u(r, t)·r/|r|. Similarly, the cell population
dynamics for an off-treatment period is given as
⎧  
⎪ ∂ x1 1 ∂
∂t (r, t) + r 2 u(r, t)x1 (r, t) = a1,1
0 x (r, t) + a 0 x (r, t),

⎪ r 2 ∂r 1 1,2 2

⎨  
∂ x2 1 ∂
∂t (r, t) + r 2 u(r, t)x2 (r, t) = (a2,2
0 − a 0 )x (r, t), (2.3)
⎪ r 2 ∂r 1,2 2

⎪  
⎪ ∂ x3
⎩ 1 ∂
∂t (r, t) + r 2 u(r, t)x3 (r, t) = a3,3
0 x (r, t),
r 2 ∂r 3

for 0 < r < R(t), t > 0. The parameters a1,1 1 , a 1 , and a 1 are the growth rates of
2,2 3,3
the populations in states 1, 2, and 3 under an on-treatment condition. The parameters
1 , a 1 , and a 1 are the rates of change from state 1 to state 2, from state 1 to state
a2,1 3,1 3,2
3, and from state 2 to state 3, respectively, under an on-treatment condition. Follow-
ing Jackson (Jackson 2004a,b), we assume that AD cells are sensitive to androgen
suppression and decrease in androgen-deprived circumstances. Accordingly, we may
1 < 0. However, AI cells are not generally sensitive to androgen sup-
assume that a1,1
pression but are likely to increase even in androgen-deprived circumstances, so we
1 > 0 and a 1 > 0. According to the different sensitivities of the three
assume that a2,2 3,3
types of cells with respect to androgen suppression, it seems biologically reasonable
1 − a 1 − a 1 < a 1 − a 1 < a 1 . The parameters a 0 , a 0 , and
to assume that a1,1 2,1 3,1 2,2 3,2 3,3 1,1 2,2
0
a3,3 are the growth rates of the populations in states 1, 2, and 3 under an off-treatment
condition, and the parameter a1,20 is the rate of change from state 2 to state 1 under an

off-treatment condition. All four parameters (a1,1 0 , a 0 , a 0 , and a 0 ) are assumed


2,2 3,3 1,2
to be positive.
By the radial symmetry assumption of the problem, we have

∂ x1 ∂ x2 ∂ x3
(0, t) = (0, t) = (0, t) = u(0, t) = 0 for all t > 0. (2.4)
∂r ∂r ∂r

On the outer boundary of the tumor, we impose no-flux boundary conditions as


follows:
⎧

⎪ x1 (r, t) d R(t) − x (r, t)u(r, t) = 0,

⎪ dt 1
⎨ r =R(t)
d R(t)
x2 (r, t) dt − x2 (r, t)u(r, t) = 0, (2.5)

⎪  r =R(t)


⎩ x3 (r, t) dt − x3 (r, t)u(r, t)
d R(t)
= 0.
r =R(t)

To complete the system, we impose the following initial condition:



⎨ R(0) = R0 ,
x1 (r, 0) = x10 (r ), x2 (r, 0) = x20 (r ), x3 (r, 0) = x30 (r := 1 − x10 (r ) − x20 (r ),

x10 (r ) ≥ 0, x20 (r ) ≥ 0 and x10 (r ) + x20 (r ) ≤ 1 for all r ∈ [0, R0 ].
(2.6)

123
822 Y. Tao et al.

The system (2.1, 2.2, 2.4–2.6) describes the cell population dynamics for an on-
treatment period and is used in Section 3 to study the dynamics of tumor growth
under CAS therapy. Conversely, the system (2.1, 2.3–2.6) describes the cell population
dynamics for an off-treatment period and is used in Section 4 to predict the growth
dynamics of an untreated tumor. Importantly, the combination of these two systems
provides a hybrid system that is used in Section 5 to study the effects of IAS therapy.
Our study shows that CAS therapy may lead to a relapse of a tumor and that IAS
therapy may postpone the relapse.

3 Tumor relapse under CAS therapy

In this section we consider the system (2.1, 2.2, 2.4–2.6). We first note that adding the
three equations in (2.2) and using (2.1) yields an equation for the cellular velocity as
follows:

1 ∂  2 
r u(r, t) = a1,1
1
x1 (r, t) + a2,2
1
x2 (r, t) + a3,3
1
x3 (r, t). (3.1)
r 2 ∂r

We also note that adding the three equations in (2.5) and using (2.1) entails that

d R(t)
= u(R(t), t). (3.2)
dt

From the viewpoint of mathematical analysis, it may be convenient to transform


the moving domain {r ≤ R(t)} into a fixed domain. As was done in the paper
by Tao and Chen (Tao and Chen 2006) we introduce a transformation of variables
(r, t, x1 , x2 , x3 , u, R) → (ρ, t, x̃1 , x̃2 , x̃3 , ũ, R) as follows:

r u(ρ R(t), t)
ρ= , x̃i (ρ, t) = xi (ρ R(t), t) (i = 1, 2, 3), and ũ(ρ, t) = .
R(t) R(t)
(3.3)

In terms of the new variables, after dropping the tildes for notational convenience, a
straightforward derivation reveals that the system (2.1, 2.2, 2.4–2.6) takes the following
form in {0 < ρ < 1, t > 0}:

⎪ x1 (ρ, t) + x2 (ρ, t) + x3 (ρ, t) = 1,


⎪ ∂ x1 ∂ x1
∂t (ρ, t) + [u(ρ, t) − ρu(1, t)] ∂ρ (ρ, t) = (a1,1 − a2,1 − a3,1 )x 1 (ρ, t) − f 1 (x 1 , x 2 , x 3 ),
1 1 1



⎪ ∂ x2 (ρ, t) + [u(ρ, t) − ρu(1, t)] ∂ x2 (ρ, t) = a 1 x (ρ, t) + (a 1 − a 1 )x (ρ, t)



⎪ ∂t ∂ρ 2,1 1 2,2 3,2 2

⎪ − f 2 (x1 , x2 , x3 ),


⎨ ∂ x3 (ρ, t) + [u(ρ, t) − ρu(1, t)] ∂ x3 (ρ, t) = a 1 x (ρ, t) + a 1 x (ρ, t) + a 1 x (ρ, t)
∂t ∂ρ 3,1 1 3,2 2 3,3 3
(3.4)

⎪ − f 3 (x1 , x2 , x3 ),



⎪ x1 (ρ, 0) = x10 (ρ), x2 (ρ, 0) = x20 (ρ), x3 (ρ, 0) = x30 (ρ) := 1 − x10 (ρ) − x20 (ρ),

⎪ ∂ x1 ∂ x2

⎪ ∂ρ (0, t) = ∂ρ
⎪ (0, t) = ∂∂ρ
x3
(0, t) = 0,


ρ 1

⎪ u(ρ, t) = ρ 2 0 a1,1 x1 (s, t) + a2,2
1 1 x (s, t) + a 1 x (s, t)]s 2 ds,


2 3,3 3
d R(t)
dt = R(t)u(1, t), R(0) = R 0 ,

123
A partial differential equation model and its reduction 823

where f i (x1 , x2 , x3 ) := [a1,1


1 x (ρ, t) + a 1 x (ρ, t) + a 1 x (ρ, t)]x (ρ, t) for i =
1 2,2 2 3,3 3 i
1, 2, 3.
We first claim that the problem (3.4) is well posed.

Lemma 3.1 Assume that R0 > 0, (x10 (ρ), x20 (ρ), x30 (ρ)) ∈ (C 1 [0, 1])3 , x10 (ρ) ≥
0, x20 (ρ) ≥ 0, and x10 (ρ) + x20 (ρ) ≤ 1. Then, the problem (3.4) admits a unique
C 1 -smooth solution for all t > 0. Moreover,

0 ≤ x1 (ρ, t) ≤ 1, 0 ≤ x2 (ρ, t) ≤ 1, 0 ≤ x3 (ρ, t) = 1 − x1 (ρ, t) − x2 (ρ, t) ≤ 1


(3.5)

for all 0 ≤ ρ ≤ 1 and t > 0. Finally,

R0 e−βt ≤ R(t) ≤ R0 eβt for some β > 0 and all t > 0. (3.6)

Proof The proof is based on a straightforward fixed-point argument and the charac-
teristic theory of hyperbolic partial differential equations. It is technically similar to
that of Lemma 6.2 and Theorem 9.1 of (Friedman and Tao 2003), so we refrain from
repeating the details here.


Let us consider the dynamics of the system (3.4). To this end, we introduce the
following three volume variables:

1
Vi (t) := 4π R (t)
3
xi (ρ, t)ρ 2 dρ, i = 1, 2, 3. (3.7)
0

In conjunction with the first equation in (3.4), this yields

4π 3
V1 (t) + V2 (t) + V3 (t) ≡ R (t) := V (t), (3.8)
3
which is the tumor volume at time t. Note that the nonlinear partial differential equa-
tion (PDE) in the free boundary problem (3.4) can be reduced to a linear ordinary
differential equation (ODE) system with variables V1 (t), V2 (t), and V3 (t).

Lemma 3.2 The triple (V1 (t), V2 (t), V3 (t)) satisfies the following linear ODE sys-
tem:
⎧ d V1 (t)

⎪ = (a1,1
1 − a 1 − a 1 )V (t),
3,1 1 t > 0,


dt 2,1
⎨ d V2 (t) = a 1 V (t) + (a 1 − a 1 )V (t), t > 0,
dt 2,1 1 2,2 3,2 2 (3.9)

⎪ d V (t)
= 1 (t) + 1 (t) + 1 (t), >


3
a V1
3,1
a 3,2 V2 a 3,3 V3 t 0,
⎩ dt 1
Vi (0) = 4π R03 0 xi0 (ρ)ρ 2 dρ, i = 1, 2, 3.

Proof A straightforward derivation with the second, seventh, and eighth of the equa-
tions in (3.4) entails that

123
824 Y. Tao et al.

1 1
d V1 (t) d R(t) ∂ x1 2
= 12π R 2 (t) x1 ρ 2 dρ + 4π R 3 (t) ρ dρ
dt dt ∂t
0 0
1
= 12π R 3 (t)u(1, t) x1 ρ 2 dρ
0
1
+4π R (t)
3
[(a1,1
1
− a2,1
1
− a3,1
1
)x1 − (a1,1
1
x1 + a2,2
1
x2 + a3,3
1
x3 )x1 ]ρ 2 dρ
0
1
∂ x1
−4π R 3 (t) [ρ 2 u(ρ, t) − ρ 3 u(1, t)] dρ
∂ρ
0
1
= 12π R 3 (t)u(1, t) x1 ρ 2 dρ
0
1
+4π R (t)
3
[(a1,1
1
− a2,1
1
− a3,1
1
)x1 − (a1,1
1
x1 + a2,2
1
x2 + a3,3
1
x3 )x1 ]ρ 2 dρ
0
−4π R (t)[ρ 2 u(ρ, t) − ρ 3 u(1, t)]x1 (ρ, t)|1ρ=0
3

1
∂  2 
+4π R (t)
3
x1 ρ u(ρ, t) dρ
∂ρ
0
1
−12π R 3 (t)u(1, t) x1 ρ 2 dρ
0
1
= 4π R 3 (t) [(a1,1
1
− a2,1
1
− a3,1
1
)x1 − (a1,1
1
x1 + a2,2
1
x2 + a3,3
1
x3 )x1 ]ρ 2 dρ
0
1
∂  2 
+4π R (t)
3
x1 ρ u(ρ, t) dρ
∂ρ
0
1
= 4π R 3 (t) [(a1,1
1
− a2,1
1
− a3,1
1
)x1 − (a1,1
1
x1 + a2,2
1
x2 + a3,3
1
x3 )x1 ]ρ 2 dρ
0
1
+4π R (t)
3
x1 (a1,1
1
x1 + a2,2
1
x2 + a3,3
1
x3 )ρ 2 dρ
0
1
= 4π R 3 (t) (a1,1
1
− a2,1
1
− a3,1
1
)x1 ρ 2 dρ
0
= (a1,1
1
− a2,1
1
− a3,1
1
)V1 (t) for all t > 0.

123
A partial differential equation model and its reduction 825

Similarly, by straightforward derivations we obtain

d V2 (t)
= a2,1
1
V1 (t) + (a2,2
1
− a3,2
1
)V2 (t) for all t > 0,
dt

and

d V3 (t)
= a3,1
1
V1 (t) + a3,2
1
V2 (t) + a3,3
1
V3 (t) for all t > 0.
dt

This completes the proof.




Furthermore, we can solve the linear ODE system (3.9) to obtain explicit formu-
lae for V1 (t), V2 (t), and V3 (t) at any time t. Therefore, according to (3.8), we can
explicitly obtain the dynamics of tumor growth.

Theorem 3.3 Assume that

1
a1,1 < 0, (3.10)
1
a2,2 > 0, 1
a3,3 > 0, 1
a2,1 > 0, 1
a3,1 > 0, 1
a3,2 >0 (3.11)

and

1
a1,1 − a2,1
1
− a3,1
1

= a2,2
1
− a3,2
1
, a2,2
1
− a3,2
1

= a3,3
1
. (3.12)

Then, the tumor will eventually relapse; that is,

V (t) → +∞ as t → +∞. (3.13)

Moreover, the tumor volume V (t) at any time t can be predicted with the following
explicit formula:

V (t) = V1 (0)e(a1,1 −a2,1 −a3,1 )t


1 1 1

1 V (0)
a2,1 1
 1
(a2,2 −a3,2
1 )t (a1,1
1 −a 1 −a 1 )t
+ 1 e − e 2,1 3,1
a2,2 − a3,2
1 − a1 + a1 + a1
1,1 2,1 3,1
a3,1 V1 (0)
1 
1 t
a3,3 (a1,1
1 −a 1 −a 1 )t
+ 1 e − e 2,1 3,1
−a1,1 + a2,1 1 + a1 + a1
3,1 3,3
1 1 V (0)
a2,1 a3,2 1
 1
a3,3 t (a2,2
1 −a 1 )t
+ 1 e −e 3,2
(a2,2 − a3,21 − a 1 + a 1 + a 1 )(a 1 − a 1 + a 1 )
1,1 2,1 3,1 3,3 2,2 3,2
1 a 1 V (0)
a2,1 3,2 1

(a2,2
1 − a 1 − a 1 + a 1 + a 1 )(−a 1 + a 1 + a 1 + a 1 )
3,2 1,1 2,1 3,1 1,1 2,1 3,1 3,3
 1
a3,3 t (a1,1
1 −a 1 −a 1 )t (a 1 −a 1 )t
× e −e 2,1 3,1 + V2 (0)e 2,2 3,2

123
826 Y. Tao et al.

1 V (0)
a3,2 2
 1
ea3,3 t − e(a2,2 −a3,2 )t
1 1
+ 1 − (a 1 − a 1 )
a3,3 2,2 3,2
1
+V3 (0)ea3,3 t for all t > 0. (3.14)

Proof We first note that by (3.5), (3.6), and the definitions of V1 (t), V2 (t), and V3 (t),
we have

V1 (t) ≥ 0, V2 (t) ≥ 0 and V3 (t) ≥ 0 for all t > 0. (3.15)

Since V1 (0) ≥ 0, V2 (0) ≥ 0, V3 (0) ≥ 0, and

4π R03
V1 (0) + V2 (0) + V3 (0) = V (0) = > 0,
3

we can distinguish the following three cases.


Case 1: V1 (0) > 0. Solving the first ODE in (3.9) yields

V1 (t) = V1 (0)e(a1,1 −a2,1 −a3,1 )t > 0


1 1 1
for all t > 0. (3.16)

1 > 0, a 1 > 0, the third equation in (3.9), and (3.15), this entails that
Along with a3,2 3,3

d V3 (t)
≥ a3,1
1
V1 (t) > 0 for all t > 0,
dt

and thus

V3 (t) > V3 (0) ≥ 0 for all t > 0.

In particular,

V3 (1) > 0. (3.17)

1 > 0, a 1 > 0, the third equation in (3.9), and (3.15), we find


However, using a3,1 3,2

d V3 (t)
≥ a3,3
1
V3 (t) for all t > 0.
dt

In conjunction with (3.17), this yields

V3 (t) ≥ V3 (1)ea3,3 (t−1) → +∞


1
as t → +∞.

Therefore,

V (t) = V1 (t) + V2 (t) + V3 (t) ≥ V3 (t) → +∞ as t → +∞.

123
A partial differential equation model and its reduction 827

Case 2: V2 (0) > 0. Using a2,1


1 > 0, the second equation in (3.9), and (3.15), we obtain

V2 (t) ≥ V2 (0)e(a2,2 −a3,2 )t > 0


1 1
for all t > 0.

1 > 0, a 1 > 0, the third equation in (3.9), and (3.15), this yields
Along with a3,1 3,3

d V3 (t)
≥ a3,2
1
V2 (t) > 0 for all t > 0.
dt

Thus,

V3 (1) > V3 (0) ≥ 0.

Using this, and proceeding as for Case 1, we obtain

V (t) ≥ V3 (t) ≥ V3 (1)ea3,3 (t−1) → +∞


1
as t → +∞.

Case 3: V3 (0) > 0. From a3,1


1 > 0, a 1 > 0, the third equation in (3.9), and (3.15),
3,2
we infer that

d V3 (t)
≥ a3,3
1
V3 (t) for all t > 0,
dt

and thus
1
V3 (t) ≥ V3 (0)ea3,3 t → +∞ as t → +∞.

This completes the proof of (3.13).


We now turn to proving the formula (3.14). We first note that

V1 (t) = V1 (0)e(a1,1 −a2,1 −a3,1 )t


1 1 1
for all t > 0. (3.18)

Substituting this expression into the second equation in (3.9), we have

d V2 (t) 1 (a1,1
1 −a 1 −a 1 )t
− (a2,2
1
− a3,2
1
)V2 (t) = V1 (0)a2,1 e 2,1 3,1 .
dt

Solving this equation, we obtain

V2 (t) = V2 (0)e(a2,2 −a3,2 )t


1 1

1 V (0)
a2,1 1
 1
(a2,2 −a3,2
1 )t (a1,1
1 −a 1 −a 1 )t
+ 1 e − e 2,1 3,1
a2,2 − a3,2
1 − a1 + a1 + a1
1,1 2,1 3,1
(3.19)

123
828 Y. Tao et al.

for all t > 0. Substituting (3.18) and (3.19) into the third equation in (3.9) and solving
the resultant equation, we have

1 V (0)
a3,1 1
 1
ea3,3 t − e(a1,1 −a2,1 −a3,1 )t
1 1 1
V3 (t) =
−a1,1
1 + a1 + a1 + a1
2,1 3,1 3,3
a3,2 V2 (0)
1 
1 t
a3,3 (a2,2
1 −a 1 )t
+ 1 e − e 3,2
a3,3 − (a2,2
1 − a1 )
3,2
1 a 1 V (0)
a2,1  1
3,2 1
ea3,3 t − e(a2,2−a3,2 )t
1 1
+
(a2,2
1 − a 1 −a 1 +a 1 + a 1 )(a 1 − a 1 + a 1 )
3,2 1,1 2,1 3,1 3,3 2,2 3,2
1 a 1 V (0)
a2,1 3,2 1

(a2,2
1 − a 1 − a 1 + a 1 + a 1 )(−a 1 + a 1 + a 1 + a 1 )
3,2 1,1 2,1 3,1 1,1 2,1 3,1 3,3
 1
a3,3 t (a1,1
1 −a 1 −a 1 )t a 1 t
e −e 2,1 3,1 + V3 (0)e 3,3 for all t > 0. (3.20)

Adding (3.18)–(3.20) and using (3.8), we obtain (3.14).




First of all, it is well known that AD cells are sensitive to androgen deprivation; in
other words, hormone therapy takes suppressive effect on AD cells. Therefore, CAS
therapy can be used to bring about initial regression of a prostatic tumor, which has
been confirmed by experimental studies (see Bruchovsky et al. 1990; Akakura et al.
1993; Jackson 2004b; Abrahamsson 2010). Since AD cells decrease in an androgen-
deprived circumstance and a1,1 1 represents the own growth rate of AD cells under CAS
1 is negative; namely, (3.10)
therapy, it is physiologically reasonable to assume that a1,1
holds.
At the very early stage of CAS therapy, we suppose that the adaption or muta-
tion changes of AD cells do not take place and the AI cells have not appeared yet.
Accordingly, at the initial condition of CAS therapy, we may assume that

V2 (0) = V3 (0) = 0,

and thus V (t) ≈ V1 (t) = V1 (0)e(a1,1 −a2,1 −a3,1 )t decreases for small time t because
1 1 1

1
a1,1 < 0 and the changing rates a2,1 1 and a 1 are positive. However, the tumor
3,1
will eventually progress to an androgen-independent state due to the aforemen-
tioned adaption and mutation changes of AD and AI cells under CAS therapy.
Moreover, the physiological meaning of hormone therapy mainly lies in its middle-
term and long-term efficacy. To better understand the mechanisms of tumor relapse
and help to design possible optimal protocols of treatment, in our model (2.2) we
assume that AI cells appear and proliferate, and the adaption and mutation mech-
anisms play important roles in the progression of a prostatic tumor under CAS
therapy (see Rennie et al. 2005; Feldman and Feldman 2001). Thus, we made the
assumption (3.11).

123
A partial differential equation model and its reduction 829

Since AI cells are malignant cells believed to be androgen-independent, it is natural


to assume that

1
a1,1 − a2,1
1
− a3,1
1
(the net growth rate of AD cells)
< a2,2
1
− a3,2
1
(the net growth rate of reversible AI cells),

under hormone therapy. Furthermore, the presence of irreversible AI cells is due to


somatic mutations (see Feldman and Feldman 2001), and these are likely to proliferate
mainly even in an androgen-deprived circumstance (see Jackson 2004a) as a probable
cause of relapse. Accordingly, we may assume that

1
a2,2 − a3,2
1
(the net growth rate of reversible AI cells)
< a3,3
1
(the net growth rate of irreversible AI cells),

under hormone therapy. The first inequality in (3.12) says that the net growth rate of
AD cells is not equal to that of reversible AI cells, whereas the second inequality in
(3.12) means that the net growth rate of reversible AI cells is also not equal to that of
irreversible AI cells. Therefore, the assumption (3.12) is physiologically reasonable.
We also note that adding the three equations in (3.9) yields

d V (t)
= a1,1
1
V1 (t) + a2,2
1
V2 (t) + a3,3
1
V3 (t) for all t > 0.
dt

Using this in conjunction with the smoothness and continuity of the solution (see
Lemma 3.1), we find

d V (t)
= a1,1
1
V1 (0) + a2,2
1
V2 (0) + a3,3
1
V3 (0).
dt t=0

From this we infer that if V2 (0) and V3 (0) are sufficiently small, then

d V (t)
< 0, (3.21)
dt t=0

1 < 0 and V (0) = 4π R 3


3 − V2 (0) − V3 (0) > 0. Biologically, (3.21)
0
owing to a1,1 1
reflects the fact that androgen-deprivation sensitivity immediately follows hormonal
therapy; however, (3.13) indicates that the eventual relapse of a tumor is unavoidable
due to the progression of malignant AI cells.
We further numerically study the tumor growth under the CAS therapy. The typical
parameter values for our numerical work are as follows:

123
830 Y. Tao et al.

80

70

60
Tumor Volumn
50

40

30

20

10

0
0 50 100 150 200 250 300 350 400 450 500
Time (days)

Fig. 2 The dynamics of tumor growth under the CAS therapy

1
a1,1 = −0.0608 (see Ideta et al. 2008; Jackson 2004a),
1
a2,1 = 0.0013 (see Hirata et al. 2010a),
1
a2,2 = 0.0064 (a perturbation of a3,3
1 ),
1
a3,1 = 5.8852 × 10−7 (see Hirata et al. 2010a),
1
a3,2 = 9.0160 × 10−7 (see Hirata et al. 2010a),
1
a3,3 = 0.0074 (see Ideta et al. 2008; Jackson 2004a).

In our simulation, we take R0 = 2 mm (see Jackson 2004a), x10 = 0.93892, x20 =


0.04095, and x30 = 0.02013 (see Hirata et al. 2010a).
Figure 2 shows that the CAS therapy leads to a tumor relapse, even though initially
effective in controlling the growth of a tumor.

4 Tumor growth under off-treatment

In this section, we study the growth dynamics of a tumor under off-treatment and thus
we consider the system (2.1, 2.3–2.6). As in Section 3, adding the three equations in
(2.3) and using (2.1) yield an equation for the cellular velocity as follows:

1 ∂  2 
r u(r, t) = a1,1
0
x1 (r, t) + a2,2
0
x2 (r, t) + a3,3
0
x3 (r, t). (4.1)
r ∂r
2

As before, we also have

d R(t)
= u(R(t), t). (4.2)
dt
Then, using (4.1), (4.2), and the transformation (3.3), we can rewrite the system (2.1,
2.3–2.6) in the following form in {0 < ρ < 1, t > 0}:

123
A partial differential equation model and its reduction 831



⎪ x1 (ρ, t) + x2 (ρ, t) + x3 (ρ, t) = 1,
⎪ ∂ x1
⎪ ∂ x1
∂t (ρ, t) + [u(ρ, t) − ρu(1, t)] ∂ρ (ρ, t) = a1,1 x 1 (ρ, t) + a1,2 x 2 (ρ, t) − g1 (x 1 , x 2 , x 3 ),
0 0



⎪ ∂ ∂


x 2 x
(ρ, t) + [u(ρ, t) − ρu(1, t)] ∂ρ (ρ, t) = (a2,2 − a1,2 )x2 (ρ, t) − g2 (x1 , x2 , x3 ),
2 0 0

⎪ ∂∂tx3
⎨ (ρ, t) + [u(ρ, t) − ρu(1, t)] ∂∂ρ x3
(ρ, t) = a3,3
0 x (ρ, t) − g (x , x , x ),
∂t 3 3 1 2 3
(4.3)

⎪ x 1 (ρ, 0) = x 10 (ρ), x 2 (ρ, 0) = x 20 (ρ), x 3 (ρ, 0) = x 30 (ρ) := 1 − x10 (ρ) − x20 (ρ),

⎪ ∂ x1 ∂ x2 ∂ x3

⎪ ∂ρ (0, t) = ∂ρ (0, t) = ∂ρ (0, t) = 0,


0

⎪ u(ρ, t) = ρ12 0ρ a1,1 x1 (s, t) + a2,2
0 x (s, t) + a 0 x (s, t)]s 2 ds,

⎪ 2 3,3 3
⎩ d R(t)
dt = R(t)u(1, t), R(0) = R 0 ,

where gi (x1 , x2 , x3 ) := [a1,1


0 x (ρ, t) + a 0 x (ρ, t) + a 0 x (ρ, t)]x (ρ, t) for i =
1 2,2 2 3,3 3 i
1, 2, 3.
The global existence issue of the problem (4.3) can be addressed as in Lemma 3.1,
so we focus on the dynamical behavior of (4.3). To this end, we introduce three
volume variables V1 (t), V2 (t), and V3 (t) as defined in (3.7). Then, the dynamics of
(V1 (t), V2 (t), V3 (t)) is described by a linear ODE system as below.

Lemma 4.1 The triple (V1 (t), V2 (t), V3 (t)) satisfies the following linear ODE sys-
tem:
⎧ d V1 (t)

⎪ = a1,1
0 V (t) + a 0 V (t),
1 1,2 2 t > 0,
⎪ dt

⎨ d V2 (t) = (a 0 − a 0 )V (t), t > 0,
dt 2,2 1,2 2
(4.4)
⎪ 3 = a 0 V (t),
⎪ d V (t)
t > 0,

⎪ dt 3,3 3

V1 (0), V2 (0) and V3 (0) are prescribed.

Proof The proof is similar to that of Lemma 3.2, so we refrain from repeating the
derivations here. 

From (4.4) we can derive the following theorem.

Theorem 4.2 Assume that


0
a1,1 > 0, a1,2
0
> 0, a2,2
0
> 0, a3,3
0
> 0 and a1,1
0

= a2,2
0
− a1,2
0
. (4.5)

Then, the tumor will continually grow to infinite size as time goes to infinity; i.e.,

d V (t)
> 0 for all t > 0 and V (t) → +∞ as t → +∞. (4.6)
dt
Moreover, the tumor volume V (t) at any time t can be predicted with the following
explicit formula :

V (t) = V1 (0)ea1,1 t + V2 (0)e(a2,2 −a1,2 )t


0 0 0

0 V (0)
a1,2 2
 0
(a2,2 −a1,2
0 )t 0 t
a1,1 0
+ 0 e − e + V3 (0)ea3,3 t (4.7)
a2,2 − a1,2 − a1,1
0 0

for all t > 0.

123
832 Y. Tao et al.

Proof Adding the three equations in (4.4) yields

d V (t) d V1 (t) d V2 (t) d V3 (t)


= + +
dt dt dt dt
= a1,1
0
V1 (t) + a2,2
0
V2 (t) + a3,3
0
V3 (t)
≥ min{a1,1
0
, a2,2
0
, a3,3
0
}V (t) for all t > 0.

Along with V (t) = 43 π R 3 (t) > 0, owing to R(t) ≥ R0 e−βt > 0 (see Lemma 3.1),
this entails
d V (t)
> 0 for all t > 0 and V (t) ≥ V (0)emin{a1,1 , a2,2 , a3,3 }t → +∞ as t → +∞.
0 0 0

dt
The former indicates that the tumor grows continually, and the latter indicates that the
tumor grows without limit. Moreover, straightforward derivations yield

V3 (t) = V3 (0)ea3,3 t , V2 (t) = V2 (0)e(a2,2 −a1,2 )t ,


0 0 0
(4.8)

and
0 V (0)
a1,2 2
 0
e(a2,2 −a1,2 )t − ea1,1 t
0 0 0
V1 (t) = V1 (0)ea1,1 t + 0 − a0 − a0
for all t > 0.
a2,2 1,2 1,1
(4.9)

Adding the three equations in (4.8) and (4.9) and using V (t) = V1 (t) + V2 (t) + V3 (t)
prove (4.7).

Under the off-treatment condition, the net growth rate of AD cells is generally larger
than that of AI cells (see Jackson 2004a,b); i.e.,
0
a1,1 > a2,2
0
− a1,2
0
, 0
a1,1 > a3,3
0
.

Hence, the assumption a1,1 0


= a 0 − a 0 in (4.5) is biologically acceptable.
2,2 1,2
Theorem 4.2 clearly shows that a tumor will continually grow to reach unmanage-
able size if the off-treatment time is sufficiently large.
Remark 4.3 The system (4.4) can be used to study the pre-treatment growth of an early
tumor. Indeed, an untreated early tumor usually consists of only AD cells. Accordingly,
we may assume that

V2 (0) = V3 (0) = 0.

Using this and (4.7) yields


0
V (t) = V1 (0)ea1,1 t for all t ∈ [0, T ],

where T > 0 is some constant. Thus, an untreated early tumor will grow exponentially.

123
A partial differential equation model and its reduction 833

Although the ODE systems (3.9) and (4.4) are both linear, the combination of these
provides a piecewise linear hybrid system for modeling the growth dynamics of a
prostate tumor under IAS therapy, which is the main objective of the next section.

5 Dynamics of a hybrid system for modeling IAS therapy

We are now concerned with the question of whether intermittent androgen suppression
(IAS) (see Akakura et al. 1993) is a possible strategy to delay the relapse of a tumor
in the clinical treatment of prostate cancer. To address this issue, we use a piecewise
linear hybrid ODE system that is a combination of system (3.9) and system (4.4).
First, we choose a time sequence {ti } (i = 0, 1, 2, . . .) with 0 = t0 < t1 < t2 < · · · <
tk < tk+1 < · · · ,where (t2i , t2i+1 ] (i = 0, 1, 2, . . .) denote the on-treatment periods
and (t2i+1 , t2(i+1) ] (i = 0, 1, 2, . . .) denote the off-treatment periods. Then, for the
on-treatment periods we solve the sub-system
⎧ 1,i

⎪ d V1 (t) 1 − a 1 − a 1 )V 1,i (t),
= (a1,1 t ∈ (t2i , t2i+1 ],




dt 2,1 3,1 1

⎨ d V2 (t)
1,i
1 V 1,i (t) + (a 1 − a 1 )V 1,i (t),
dt = a2,1 1 2,2 3,2 2 t ∈ (t2i , t2i+1 ],
(5.1)

⎪ 1,i
d V3 (t)

⎪ 1 V 1,i (t) + a 1 V 1,i (t) + a 1 V 1,i (t), t ∈ (t , t
= a3,1 2i 2i+1 ],

⎪ dt 1 3,2 2 3,3 3

⎩ 1,i 1,i 1,i
V1 (t2i ), V2 (t2i ) and V3 (t2i ) are prescribed,

and for the off-treatment periods we solve the sub-system


⎧ 0,i

⎪ d V1 (t) = a1,1
0 V 0,i (t) + a 0 V 0,i (t), t ∈ (t2i+1 , t2(i+1) ],




dt 1 1,2 2

⎨ d V2 (t)
0,i
0 − a 0 )V 0,i (t),
dt = (a2,2 1,2 2 t ∈ (t2i+1 , t2(i+1) ],
(5.2)
⎪ d V (t)
⎪ 0,i

⎪ 3 0 0,i
= a3,3 V3 (t), t ∈ (t2i+1 , t2(i+1) ],

⎪ dt

⎩ 0,i 0,i 0,i
V1 (t2i+1 ), V2 (t2i+1 ) and V3 (t2i+1 ) are prescribed.

We prescribe V11,0 , V21,0 , and V31,0 as initial conditions, and then we solve the
systems (5.1) and (5.2) alternately. In other words, we proceed as follows: first, we
solve the sub-system (5.1) on a time interval (t2i , t2i+1 ]; then, we take

V10,i (t2i+1 ) = V11,i (t2i+1 ), V20,i (t2i+1 ) = V21,i (t2i+1 ), V30,i (t2i+1 ) = V31,i (t2i+1 ),

and solve the sub-system (5.2) on a time interval (t2i+1 , t2(i+1) ]; next, we take

V11,i+1 (t2(i+1) ) = V10,i (t2(i+1) ), V21,i+1 (t2(i+1) ) = V20,i (t2(i+1) ),


V31,i+1 (t2(i+1) ) = V30,i (t2(i+1) )

123
834 Y. Tao et al.

80

70

60
Tumor Volumn

50

40
CAS

30

20 IAS

10

0
0 100 200 300 400 500 600 700 800 900 1000
Time (days)

Fig. 3 The dynamics of tumor growth under CAS and IAS therapies, where the on-treatment period and
the off-treatment period are each 40 days for the IAS therapy. The solid blue and broken red lines show
on-treatment and off-treatment periods, respectively

and solve the sub-system (5.1) once again on a time interval (t2(i+1) , t2(i+1)+1 ]. Contin-
uing the above iterative procedure, we can obtain the large-time behavior of a prostate
tumor under IAS therapy.
The typical parameter values for our numerical work are as follows:

1
a1,1 = −0.0608 (see Ideta et al. 2008; Jackson 2004a),
1
a2,1 = 0.0013 (see Hirata et al. 2010a),
1
a2,2 = 0.0064 (a perturbation of a3,3
1 ),
1
a3,1 = 5.8852 × 10−7 (see Hirata et al. 2010a),
1
a3,2 = 9.0160 × 10−7 (see Hirata et al. 2010a),
1
a3,3 = 0.0074 (see Ideta et al. 2008; Jackson 2004a),
0
a1,1 = 0.010653 (see Ideta et al. 2008; Jackson 2004a),
0
a1,2 = 0.1000 (see Hirata et al. 2010a),
0
a2,2 = 0.002 (a perturbation of a3,3
0 ),
0
a3,3 = 0.001 (see Ideta et al. 2008; Jackson 2004a).

In our simulations, we take

V11,0 = 31.4635, V21,0 = 1.3723, and V31,0 = 0.6745,

4π R 3
since V j1,0 = 3 0 × x j0 ( j = 1, 2, 3), where R0 = 2 mm (see Jackson 2004a),
x10 = 0.93892, x20 = 0.04095, and x30 = 0.02013 (see Hirata et al. 2010a). Since
0 > 0, a 0 > 0, a 0 > 0, a 0 > 0, and a 0 − a 0 < 0, the parameter values for
a1,1 1,2 2,2 3,3 2,2 1,2
the numerical work satisfy the assumption of Theorem 4.2.

123
A partial differential equation model and its reduction 835

Fig. 4 The dynamics of tumor growth under IAS therapies with different on-treatment and off-treatment
periods. (a) The on-treatment period is 10 days, and the off-treatment period is 70 days. (b) The on-treatment
period is 20 days, and the off-treatment period is 60 days. (c) The on-treatment period is 30 days, and the
off-treatment period is 50 days. (d) The on-treatment period is 50 days, and the off-treatment period is 30
days

Figure 3 indicates that the IAS therapy remarkably postpones the relapse time when
compared with the CAS therapy. However, it seems impossible to avoid the eventual
relapse under the IAS therapy with our typical parameter values. While androgen
deprivation is temporarily effective for inhibiting the increase of AD cells in a prostate
tumor, IAS offers the possibility of limiting the emergence of AI cells, which has been
observed in experiments (e.g., see Bruchovsky et al. 1990, 2000, 2006, 2007). Figure 3
predicts that IAS would be more appropriate than CAS.
Figure 4 partially addresses how to optimize treatment schedules for the IAS ther-
apy. Figure 4(a) shows that the IAS therapy with short on-treatment intervals leads to
cyclic growth and regression of a tumor, with an eventual relapse. Figure 4(d) implies
that the IAS therapy with long on-treatment intervals may also be ineffective in pre-
venting the relapse of a tumor. Figure 4 thus suggests that there exists an optimal
schedule for IAS therapy, which should be carefully planned (see Hirata et al. 2010b;
Suzuki et al. 2010 for details of optimal scheduling).
0 of the reversible change from state 2 to
Figure 5 suggests that increasing the rate a1,2
state 1 may improve the effect of the IAS therapy. The higher rate will increase the AD

123
836 Y. Tao et al.

0 of the reversible change from state 2 to state 1


Fig. 5 The dependence of tumor growth on the rate a1,2

cell sub-population and decrease the reversible AI cell sub-population during the off-
treatment periods. The effect during the on-treatment periods may thus be improved,
since androgen deprivation is a fundamental therapy for inhibiting the increase of AD
cells in a prostate tumor, as mentioned above.

6 Discussion

We have proposed a PDE model for prostate tumor growth under IAS. Compared with
previous models for prostate tumor growth under hormone therapy (see Jackson 2004a;
Guo et al. 2008; Tanaka et al. 2010a; Tao et al. 2010b), this model has the following two
novel features: on the one hand, it includes the coexistence of the AI cells generated
by reversible changes due to adaptation and the AI cells generated by irreversible
changes due to genetic mutations; on the other hand, this PDE model can be reduced
to a piecewise linear ODE model by introducing new volume variables. The model can
reflect the temporary regression of a tumor following androgen deprivation, forecast
the emergence of androgen resistance, indicate that IAS may be more appropriate than
CAS, and imply optimal treatment schedules. For simplicity, the PSA level during IAS
can be related to the tumor volume, which is equivalent to the number of cancer cells.
This is the first model that succeeded to derive a piecewise linear ODE model of
prostate cancer under IAS on the basis of partial differential equations. This is the most
important result of this paper because if a piecewise linear ODE model is obtained, our
previous results (see Hirata et al. 2010a,b; Suzuki et al. 2010; Tanaka et al. 2010a; Tao
et al. 2010b) are available to fit the model to real clinical PSA data and provide optimal
IAS scheduling in a personalized way. Since these processes need complicated and
statistical analysis of real PSA data beyond the scope of the present paper, we make
them our future study.

123
A partial differential equation model and its reduction 837

Our simulation shows that the effect of the IAS therapy is sensitive to certain para-
0 (see Figure 5). Therefore, estimating parameters and overcoming
meters, such as a1,2
estimation error are two important issues. One strategy is using past data on patients
as a priori knowledge about the parameters. However, the validity of these parameter
values requires assessment in future clinical trials.
Finally, we note that the IAS therapy may help not only to prolong the survival
times for patients with advanced prostate cancer but also to improve the quality of life
for those patients.

Acknowledgments The authors are very grateful to Professor Mats Gyllenberg and the two anonymous
referees for their value comments and suggestions, which helped the authors to improve the original man-
uscript. Tao is supported by the National Natural Science Foundation of China (No. 11171061) and by the
Innovation Program of the Shanghai Municipal Education Commission (No. 13ZZ046). Guo is supported
by the E-Institutes of the Shanghai Municipal Education Commission (No. E03004) and by the National
Natural Science Foundation of China (No. 10901106). Aihara is supported by the Aihara Innovative Math-
ematical Modelling Project of the Japan Society for the Promotion of Science (JSPS) through the Funding
Program for World-Leading Innovative R&D on Science and Technology (FIRST Program) initiated by the
Council for Science and Technology Policy (CSTP).

References

Abrahamsson PA (2010) Potential benefits of intermittent androgen suppression therapy in the treatment
of prostate cancer: a systematic review of the literature. Eur Urol 57:49–59
Adam JA (1996) A simplified mathematical model of tumor growth. Math Biosci 81:224–229
Akakura K, Bruchovsky N, Goldenbeg SL, Rennie PS, Buckley AR, Sullivan LD (1993) Effects of inter-
mittent androgen suppression on androgen-dependent tumors: apoptosis and serum prostate-specific
antigen. Cancer 71:2782–2790
Bellomo N, Li NK, Maini PK (2008) On the foundations of cancer modelling: selected topics, speculations
and perspective. Math Mod Meth Appl Sci 18:593–646
Bladou F, Vessella RL, Buhler KR, Ells WJ, True LD, Lange PH (1996) Cell proliferation and apoptosis
during prostatic tumor xenograft involution and regrowth after castration. Int. J. Cancer 7:785–790
Bruchovsky N, Rennie PS, Coldman AJ, Goldenberg SL, To M, Lawson D (1990) Effects of androgen
withdrawal on the stem cell composition of the Shionogi carcinoma. Cancer Res 50:2275–2282
Bruchovsky N, Klotz LH, Sadar M, Crook JM, Hoffart D, Godwin L, Warkentin M, Gleave ME, Goldenberg
SL (2000) Intermittent androgen suppression for prostate cancer: Canadian prospective trial and related
observations. Mol Urol 4:191–199
Bruchovsky N, Klotz L, Crook J, Malone S, Ludgte C, Morris WJ, Gleave ME, Goldenberg SL (2006)
Final results of the Canadian prospective phase II trial of intermittent androgen suppression for men
in biochemical recurrence after radiotherapy for locally advanced prostate cancer: clinical parameters.
Cancer 107:389–395
Bruchovsky N, Klotz L, Crook J, Larry S, Goldenberg SL (2007) Locally advanced prostate cancer –
biochemical results from a prospective phase II study of intermittent androgen suppression for men with
evidence of prostate-specific antigen recurrence after radiotherapy. Cancer 109:858–867
Byrne HM, Alarcon T, Owen MR, Webb SD, Maini PK (2006) Modelling aspects of cancer dynamics: a
review. Phil Trans R Soc Lond Ser A Math Phys Eng Sci 36:1563–1578
Feldman BJ, Feldman D (2001) The development of androgen-independent prostate cancer. Nat Rev Cancer
1:34–45
Friedman A (2007) Mathematical analysis and challenges arising from models of tumor growth. Math Mod
Meth Appl Sci 17:1751–1772
Friedman A, Hu B (2006) Bifurcation from stability to instability for a free boundary problem arising in a
tumor model. Arch Ration Mech Anal 180:293–330
Friedman A, Tao Y (2003) Analysis of a model of a virus that replicates selectively in tumor cells. J Math
Biol 47:391–423
Greenspan H (1972) Models for the growth of a solid tumor by diffusion. Stud Appl Math 52:317–340

123
838 Y. Tao et al.

Guo Q, Tao Y, Aihara K (2008) Mathematical modeling of prostate tumor growth under intermittent andro-
gen suppression with partial differential equations. Int J Bifurcat Chaos 18:3789–3797
Hirata Y, Bruchovsky N, Aihara K (2010a) Development of a mathematical model that predicts the outcome
of hormone therapy for prostate cancer. J Theor Biol 264:517–527
Hirata Y, di Bernardo M, Bruchovsky N, Aihara K (2010b) Hybrid optimal scheduling for intermittent
androgen suppression of prostate cancer. Chaos 20:045125
Ideta AM, Tanaka G, Takeuchi T, Aihara K (2008) A mathematical model of intermittent androgen sup-
pression for prostate cancer. J Nonlinear Sci 18:593–614
Jackson TL (2004a) A mathematical model of prostate tumor growth and androgen-independent relapse.
Discrete Contin Dyn Syst B 4:187–201
Jackson TL (2004b) A mathematical investigation of multiple pathways to recurrent prostate cancer: com-
parsion with experiment data. Neoplasia 6:697–704
Rennie PS, Bruchovsky N, Coldman AJ (1990) Loss of androgen dependence is associated with an increase
in tumorigenic stem cells and resistance to cell-death genes. J Steroid Biochem Mol Biol 37:843–847
Rennie P, Read J, Murphy L (2005) Hormones and Cancer. In: Tannock IF, Hill RP, Bristow RG, Harrington
L (eds) The basic science of oncology. McGraw-Hill, New York, pp 400–430
Suzuki T, Bruchovsky N, Aihara K (2010) Piecewise affine systems modelling for optimizing hormone
therapy of prostate cancer. Phil Trans R Soc A 368:5045–5059
Tanaka G, Hirata Y, Goldenberg SL, Bruchovsky N, Aihara K (2010a) Mathematical modelling of prostate
cancer growth and its application to hormone therapy. Phil Trans R Soc 368:5029–5044
Tao Y, Chen M (2006) An elliptic-hyperbolic free boundary problem modelling cancer therapy. Nonlinearity
19:419–440
Tao Y, Guo Q, Aihara K (2009) A model at the macroscopic scale of prostate tumor growth under intermittent
androgen suppression. Math Mod Meth Appl Sci 19:2177–2201
Tao Y, Guo Q, Aihara K (2010b) A mathematical model of prostate tumor growth under hormone therapy
with mutation inhibitor. J Nonlinear Sci 20:219–240
Tello JI (2013) On a mathematical model of tumor growth based on cancer stem cells. Math Biosci Eng
10:263–278
Ward JP, King JR (2003) Mathematical modelling of drug transport in tumour multicell spheroids and
monolayer cultures. Math Biosci 181:177–207

123

You might also like