Nonlinear Analysis: Real World Applications 38 (2017) 49–67

Contents lists available at ScienceDirect

Nonlinear Analysis: Real World Applications

www.elsevier.com/locate/nonrwa

Modeling the spatiotemporal variations in brucellosis transmission

Chayu Yang a , Paride O. Lolika b , Steady Mushayabasa b , Jin Wang a,∗
a
Department of Mathematics, University of Tennessee at Chattanooga, Chattanooga, TN 37403, USA
b
Department of Mathematics, University of Zimbabwe, P.O. Box MP 167, Harare, Zimbabwe

article info abstract

Article history: In this paper, we propose a nonlinear modeling framework to investigate the trans-
Received 31 December 2016 mission dynamics of brucellosis, incorporating both the spatial and seasonal varia-
Received in revised form 10 April tions. The spatial modeling is based on a patch structure, and the seasonal impact
2017
is represented by utilizing time-periodic model parameters. We demonstrate this
Accepted 17 April 2017
framework through a two-patch model and conduct detailed analysis, for the cases
with and without seasonal oscillations, respectively. In particular, we establish the
Keywords: threshold dynamics results using the reproduction numbers defined under different
Brucellosis scenarios. Our findings underscore the importance of including spatial and seasonal
Spatial variation heterogeneities in the design of control strategies for brucellosis.
Seasonal oscillation © 2017 Elsevier Ltd. All rights reserved.
Threshold dynamics

1. Introduction

Brucellosis, an infectious bacterial disease, is one of the world’s major zoonoses. Caused by various species
of the bacteria Brucella [1], the disease can be transmitted to animals and humans with exposure to infected
animals or ingestion of contaminated water, food, and dust, etc. [2]. In animals, especially among sheep and
goats, brucellosis mainly affects the reproduction process and can lead to fertility problems and abortion,
and reduce the survival of newborns [3]. In humans, mortality is negligible, but the illness can last for
several years [4], characterized by such symptoms as intermittent fever, headache, fatigue, joint and bone
pain, psychosis, and disturbance [5].
Currently there are more than 500,000 new cases of brucellosis reported annually and the disease remains
endemic in many areas of the world, including Spain, Latin America, the Middle East, and Africa [6,2].
Among these, the majority of brucellosis cases are found in sub-Sahara Africa, where Ethiopia, Chad,
Tanzania, Nigeria, Uganda, Kenya, Zimbabwe and Somalia have been reporting persistence of brucellosis
in humans attributed to the infection of domestic cattle, camels, goats and sheep [7]. With large pastoral

∗ Corresponding author.
E-mail address: Jin-Wang02@utc.edu (J. Wang).

http://dx.doi.org/10.1016/j.nonrwa.2017.04.006
1468-1218/© 2017 Elsevier Ltd. All rights reserved.
50 C. Yang et al. / Nonlinear Analysis: Real World Applications 38 (2017) 49–67

communities, and the demand for meat and livestock products to double by 2050, brucellosis poses a major
threat to the public health and economic growth of the region and demands serious control efforts.
Mathematical modeling, analysis and simulation offer a useful means to understand the transmission
and spread of brucellosis so that effective disease control measures could be designed. A few mathematical
models have been published in recent years to investigate brucellosis dynamics. For example, Ainseba et al. [8]
constructed and analyzed a mathematical model for ovine brucellosis that incorporated direct, indirect and
vertical transmission routes. Hou and co-workers [9] employed a system of ordinary differential equations
(ODEs) to model the transmission of brucellosis and the effects of vaccination on brucellosis prevention
and intervention. Lolika et al. [10] proposed a brucellosis model and conducted an optimal control study
on the use of animal vaccination and environmental decontamination as disease control measures against
brucellosis infection. Li et al. [11] proposed a model to investigate the transmission of brucellosis among
sheep and from sheep to humans, and their findings indicated that a combination of intervention methods
(such as prohibiting mixed feeding, vaccination, and detection and elimination) is useful in controlling human
brucellosis.
Despite these efforts, however, several challenges remain in the mathematical modeling of brucellosis.
First, different places likely have different geographic, ecological and environmental structures, and animals
living in various locations likely exhibit different contact and communication patterns. In particular, animals
make regular migration from one place to another, which directly contributes to the disease spread. So far
these differences of the transmission dynamics have not been taken into account, leading to inadequate
understanding of the influence of the spatial factors in the transmission and spread of brucellosis. Another
limitation in brucellosis modeling is that the impact of seasonal variation is insufficiently addressed. In fact,
like many other infectious diseases [12–14], brucellosis is significantly influenced by seasonal changes, and
prior field studies have already demonstrated a strong correlation between brucellosis outbreaks and seasonal
oscillations [15–17]. For example, a recent analysis of brucellosis datasets in a few countries [17] reveals that
there is a marked seasonal variation in the incidence of acute brucellosis, with most cases occurring in the
spring and summer. Factors such as periodic changes in temperature, seasonal precipitation which directly
affects the availability of forage, environmental fluctuations in humidity and exposure to UV light which
impact the survival of Brucella, and seasonal rituals in Africa which are associated with animal migration
and slaughtering, all contribute to seasonal fluctuations in the transmission and spread of brucellosis.
Such spatial and temporal heterogeneities have strong impacts on the dynamics of brucellosis that are not
captured by homogeneous, autonomous differential equation models. In the present study, we will make a
first step toward integrating the spatial and seasonal variations into a single framework for a comprehensive
modeling of brucellosis dynamics. To that end, we make use of the patch modeling approach (see, e.g.,
[18,19]) — we propose a two-patch deterministic model, where each patch has distinct populations and
infection characteristics, to study the transmission of brucellosis among animals. In each patch, the
animal population is subdivided into the susceptible and the infected compartments; meanwhile, another
compartment is introduced to represent the concentration of the pathogen (i.e., Brucella) in the environment.
Both the indirect (i.e., environment-to-host) and direct (i.e., host-to-host) transmission routes are considered
in our model, representing the multiple pathways in the force of infection for brucellosis. Animals may move
from one patch to the other, representing their migration in space. Additionally, we will incorporate the
effects of seasonal oscillation by employing time-periodic model parameters, which leads to a non-autonomous
patchy ODE system.
We organize the remainder of this paper as follows. In Section 2, we first introduce our two-patch model
in the autonomous form, where each model parameter is fixed as a constant, and then conduct a thorough
equilibrium analysis for this model. In Section 3, we extend the autonomous model to a periodic two-patch
model and analyze the threshold dynamics. In Section 4, we use numerical simulation results to validate our
analytical predictions. Finally, we conclude the paper with some discussion in Section 5.
 C. Yang et al. / Nonlinear Analysis: Real World Applications 38 (2017) 49–67 51

2. A two-patch autonomous model

We consider the spatial spread of brucellosis in an environment of two patches, where brucellosis can
spread from one patch to the other due to animal movement, particularly, through migration. We further
assume a unidirectional form of migration; that is, animals move from patch 1 to patch 2.
Let Sj and Ij (j = 1, 2) denote the number of susceptible and infectious animals, respectively, in each
patch. Meanwhile, we introduce another compartment Bj that represents the population of the free-living
pathogen (i.e., Brucella) in the environment. The Brucella population can be replenished by infectious
hosts that excrete the pathogen to the environment. In contrast, the natural decay of the pathogen and
the decontamination practices reduce the environmental persistence of the pathogen. Susceptible animals
become infected (and infectious) either by adequate contacts with infectious animals or the contaminated
environment. The following system of ordinary differential equations (ODEs) describe the brucellosis
transmission dynamics:

S˙1 (t)


 = A1 − (α1 I1 + β1 B1 )S1 − (θS + µ1 )S1 ,
I˙1 (t) = (α1 I1 + β1 B1 )S1 − (θI + c1 + µ1 )I1 ,





Ḃ (t)
1 = φ1 I 1 − d 1 B 1 ,
˙2 (t) (1)

S = A2 − (α2 I2 + β2 B2 )S2 − µ2 S2 + θS S1 ,
I˙2 (t)




 = (α2 I2 + β2 B2 )S2 − (µ2 + c2 )I2 + θI I1 ,
Ḃ2 (t) = φ2 I 2 − d 2 B 2 ,


where all model parameters are non-negative. The parameter Aj (j = 1, 2) is the constant recruitment
rate for animals in each patch, µj is the natural animal death rate, αj and βj denote the host-to-host and
environment-to-host disease transmission rates, respectively. The mean infectious period for animals in each
patch is denoted by c−1
j . For the pathogen population, φj denotes the pathogen shedding rate, and dj is the
pathogen removal rate that includes the effects of both the natural decay and the decontamination practices.
In addition, we assume that the susceptible and infectious animals migrate from patch 1 to patch 2 at rates
θS and θI , respectively.

2.1. Reproduction numbers

By direct calculation, we find that (1) always has a disease-free equilibrium (DFE) given by
 
0 A1 θS A1 + (θS + µ1 )A2
S10 , I10 , B10 , S20 , I20 , B20
 
E = = , 0, 0, , 0, 0 . (2)
θS + µ1 µ2 (θS + µ1 )

Let N (t) = S1 (t) + I1 (t) + S2 (t) + I2 (t), then it can be easily verified that the domain of biological interest
 
φj (A1 + A2 ) (A1 + A2 )
Γ = (S1 , I1 , B1 , S2 , I2 , B2 ) ∈ R6+ : Sj ≤ Sj0 , Bj ≤ , j = 1, 2; N (t) ≤
dj µj min(µ1 , µ2 )

is positively invariant and attracting all orbits with respect to the model (1).
The basic reproduction number, denoted R0 , measures the average number of secondary infections
generated by a single infectious case in a fully susceptible population during its average infectious period [20].
The reproduction number is commonly regarded as a threshold quantity for the disease dynamics, essential in
determining the transmission and spread of the disease. Using the next-generation matrix notations in [20],
the non-negative matrix F that denotes the generation of new infection and the non-singular matrix V that
52 C. Yang et al. / Nonlinear Analysis: Real World Applications 38 (2017) 49–67

denotes the disease transfer among compartments, are respectively given by

   
α1 S10 β1 S10 0 0 (c1 + θI + µ1 ) 0 0 0
 0 0 0 0   −φ1 d1 0 0
F =  , and V =  (3)
   
 0 0 α2 S20 β2 S20  −θI 0 (c2 + µ2 ) 0 


0 0 0 0 0 0 −φ2 d2
Let us use Ri to denote the reproduction number associated with patch i (i = 1, 2), where
(α1 d1 + β1 φ1 )A1 (α2 d2 + β2 φ2 )(θS A1 + (θS + µ1 )A2 )
R1 = , R2 = . (4)
d1 (θS + µ1 )(c1 + θI + µ1 ) d2 µ2 (θS + µ1 )(µ2 + c2 )
Biologically, R1 and R2 represent the disease risks for patches 1 and 2, respectively. We observe that, based
on our assumption of the unidirectional animal movement, R1 does not depend on the properties of patch
2, whereas R2 depends on some characteristics of patch 1. In particular, the disease risk for patch 2 has
been increased due to the animal migration from patch 1.
From (3), we know that the basic reproductive number R0 for the entire system is determined by the
spectral radius of the next-generation matrix F V −1 . It then follows that
R0 = max(R1 , R2 ), (5)
showing that the disease risk for the entire system depends on that associated with each of the two patches.

2.2. Disease-free equilibrium

From the work in [20], we know that the DFE is locally asymptotically stable when R0 < 1, and unstable
when R0 > 1. Indeed, we can establish below a stronger result regarding the global dynamics of the DFE.
We will utilize the approach of Lyapunov functions [21–24] in the analysis of the global asymptotic stability.

Theorem 2.1. If R0 ≤ 1, the DFE is globally asymptotically stable in Γ . If R0 > 1, the system is uniformly
persistent.

Proof. Let Y(t) = (I1 , B1 , I2 , B2 ). Since



 I˙1 (t) = (α1 I1 + β1 B1 )S1 − (θI + c1 + µ1 )I1 ,

Ḃ (t) = φ I − d B ,
1 1 1 1 1
˙2 (t) = (α2 I2 + β2 B2 )S2 − (µ2 + c2 )I1 + θI I1 , (6)

 I

Ḃ (t) = φ I − d B ,
2 2 2 2 2

it follows that
Ẏ(t) ≤ (F − V )Y,
where F and V are defined in (3). Motivated by [24], we define a Lyapunov function as follows
L = wT V −1 Y.
Differentiating L along solutions of (1), we have
L̇(t) = wT V −1 Ẏ
≤ wT V −1 (F − V )Y
= (R0 − 1)wT Y ≤ 0, if R0 ≤ 1.
It can be easily verified that the largest invariant subset of Γ where L̇ = 0 is the singleton {E 0 }. Therefore,
by LaSalle’s invariance principle [25], E 0 is globally asymptotically stable in Γ when R0 ≤ 1.
 C. Yang et al. / Nonlinear Analysis: Real World Applications 38 (2017) 49–67 53

If R0 > 1, then by continuity, L̇ > 0 in a neighborhood of E 0 in Γ̊ . Solutions in Γ̊ sufficiently close to E 0

move away from the DFE, implying that the DFE is unstable. Using a uniform persistence result from [26]
and an argument as in the proof of Proposition 3.3 of [27], it can be shown that when R0 > 1, the instability
of the DFE implies the uniform persistence of the model (1). 
The result in Theorem 2.1 shows that R0 = 1 is a sharp threshold for disease dynamics: the disease will
die out when R0 ≤ 1, whereas the disease will persist when R0 > 1 (we refer to [28,29,14] for more details
on the persistence theory). Next, we turn to the analysis of the nontrivial equilibria of the system and their
dynamical properties.

2.3. Nontrivial equilibria

Any nontrivial equilibrium (S1 , I1 , B1 , S2 , I2 , B2 ) for system (1) satisfies the following algebraic
equations:

A1 = (α1 I1 + β1 B1 + θS + µ1 )S1 ,
(α1 I1 + β1 B1 )S1 = (θI + c1 + µ1 )I1 ,
d1 B1 = φ1 I1 ,
(7)
A2 = (α2 I2 + β2 B2 + µ2 )S2 − θS S1 ,
(α2 I2 + β2 B2 )S2 = (µ2 + c2 )I2 − θI I1 ,
d2 B2 = φ2 I2 .
φ1
We have B1 = d1 I1 from the third equation in (7), and plug it into the second equation to obtain
 
β1 φ 1
α1 + I1 S1 = (θI + c1 + µ1 )I1 . (8)
d1
A1 1 = S 0 . Combining Eqs. (7) and (8), we obtain
If I1 = 0, then B1 = 0, S1 = θS +µ1 := S
1 , where S
1

α2 d2 + β2 φ2
I2 S2 = (µ2 + c2 )I2 . (9)
d2
Notice that I2 must be positive for a nontrivial equilibrium (since I1 = 0 already). It then yields
d2 (µ2 + c2 )
S2 = := S
2 . (10)
α2 d2 + β2 φ2
By substituting Eq. (10) into (7), we obtain
d2 µ2 (R2 − 1) φ2 µ2 (R2 − 1)
I2 = := I2 , B2 = := B
2 .
α2 d2 + β2 φ2 α2 d2 + β2 φ2
Therefore, we conclude that there exists a nontrivial boundary equilibrium

E1 = (S
1 , 0, 0, S
2 , I2 , B
2 ) (11)

for system (1) if and only if R2 > 1.

Next, we consider the case I1 = I1∗ > 0. We can calculate S1 = S1∗ directly from (8), which yields
d1 (θI + c1 + µ1 )
S1∗ = . (12)
α1 d1 + β1 φ1
Then substitution of Eq. (12) into the second equation in (7) yields
d1 (θI + c1 + µ1 )
I1∗ = (R1 − 1). (13)
α1 d1 + β1 φ1
54 C. Yang et al. / Nonlinear Analysis: Real World Applications 38 (2017) 49–67

Clearly, I1∗ > 0 if and only if R1 > 1. Subsequently, B1 = B1∗ = φd11 I1∗ is uniquely determined by I1∗ .
Meanwhile, Eqs. (7) and (8) yield
 
α2 d2 + β2 φ2
A2 = I2 + µ2 S2 − θS S1∗ ,
d2 (14)
α2 d2 + β2 φ2
I2 S2 = (µ2 + c2 )I2 − θI I1∗ .
d2
Combining the two equations in (14) and canceling out S2 , we have
 
α2 d2 + β2 φ2 α2 d2 + β2 φ2
(µ2 + c2 )I22 + µ2 (µ2 + c2 ) − (A2 + θS S1∗ + θI I1∗ ) I2 − µ2 θI I1∗ = 0,
d2 d2
α2 d2 +β2 φ2
which indicates that I2 has a unique positive solution I2∗ > 0 since d2 (µ2 + c2 ) > 0 and µ2 θI I1∗ > 0.
A2 +θS S1∗ φ2 ∗
Consequently, S2∗ = r2 I2∗ +µ2 and B2∗ = d2 I2 are uniquely determined by I2∗ . Therefore, we have a positive
endemic equilibrium
E2 = (S1∗ , I1∗ , B1∗ , S2∗ , I2∗ , B2∗ ) (15)
for system (1) if and only if R1 > 1.
Now we may summarize the above analysis by the following theorem.

Theorem 2.2. If R0 = max{R1 , R2 } ≤ 1, then system (1) only has the trivial, disease-free equilibrium E 0 .
If R0 = max{R1 , R2 } > 1, then in addition to the DFE E 0 , nontrivial equilibria exist:
(1) If R1 ≤ 1, R2 > 1, there is only a nontrivial boundary equilibrium E1 in Γ .
(2) If R1 > 1, R2 ≤ 1, there is only a positive endemic equilibrium E2 in Γ .
(3) If R1 > 1, R2 > 1, both the boundary equilibrium E1 and the endemic equilibrium E2 exist in Γ .

2.4. Local and global dynamics

We proceed to investigate the dynamical behavior of the nontrivial equilibria. The following result
characterizes the local dynamics of the boundary equilibrium E1 .

Theorem 2.3. (ii) If R1 < 1 and R2 > 1, then E1 is locally asymptotically stable.
(i) If R1 > 1 and R2 > 1, then E1 is unstable.

Proof.
 Linearizing the system (1) at the boundary equilibrium E1 , we obtain the Jacobian matrix J =
J1 0
J∗ J2
, where
 
−(θS + µ1 ) −α1 S
1 −β1 S1
J
1 = 
 0 1 − (θI + c1 + µ1 ) β1 S
α1 S 1 ,
0 φ1 −d1
 
−(α2 I2 + β2 B
2 + µ2 ) −α2 S
2 −β2 S2
J
2 = 
 α2 I2 + β2 B
2 2 − (µ2 + c2 ) β2 S
α1 S 2 .
0 φ2 −d2

It is easy to verify that the characteristic polynomial of J is det(λI − J) = det(λI − J

1 ) det(λI − J
2 ), and
1 ) = λ3 + x1 λ2 + y1 λ + z1 ,
det(λI − J
2 ) = λ3 + x2 λ2 + y2 λ + z2 ,
det(λI − J
 C. Yang et al. / Nonlinear Analysis: Real World Applications 38 (2017) 49–67 55

where
 
α1 d1
x1 = d1 + Y1 + X1 1 − R1 ,
Z1
  
α1 d1
y1 = Y1 d1 + X1 1 − R1 + d1 X1 (1 − R1 ),
Z1
z1 = d1 X1 Y1 (1 − R1 ),
β2 φ 2
x2 = d2 + µ2 R2 + X2 ,
Z2
β2 φ 2
y2 = µ2 X2 (R2 − 1) + µ2 X2 + d2 µ2 R2 ,
Z2
z2 = µ2 d2 X2 (R2 − 1),
with
X1 = θI + c1 + µ1 ,
Y1 = θS + µ1 ,
Z1 = α1 d1 + β1 φ1 ,
X2 = µ2 + c2 ,
Z2 = α2 d2 + β2 φ2 .
Clearly, if R1 > 1 and R2 > 1, then z1 < 0, and there exists an eigenvalue of J that has a positive real
part. Hence E1 is unstable in this case. On the other hand, x1 , y1 , z1 , x2 , y2 , z2 are all positive when R1 < 1
and R2 > 1. Furthermore, we have x1 y1 > z1 and x2 y2 > z2 , since x1 > Y1 , y1 > d1 X1 (1 − R1 ) and
x2 > d2 , y2 > µ2 X2 (R2 − 1). It follows from the Routh–Hurwitz criterion that E1 is locally asymptotically
stable if R1 < 1 and R2 > 1. 
We already know that the disease-free equilibrium E0 is unstable when there exist nontrivial equilibria;
i.e., when R0 > 1. Theorem 2.3 shows that when E1 is the only nontrivial equilibrium, it must be (locally)
stable; and when both E1 and E2 exist, E1 becomes unstable. This, consequently, implies that the endemic
equilibrium E2 is stable whenever it exists. The local stability of E2 can be similarly analyzed by examining
its characteristic polynomial and using the Routh–Hurwitz criterion, though the algebraic manipulation
becomes extremely tedious. Instead of engaging the (somehow unnecessary) algebraic complications, we
proceed to establish the stronger results regarding the global asymptotic stabilities of both E1 and E2 . To
that end, we introduce two more assumptions:
θI +c1
(C1) supt≥0 (S1 ) ≤ 2α1 .
c2
(C2) supt≥0 (S2 ) ≤ 2α2 .
These conditions provide additional regulations on the upper bounds of the susceptible populations in
both patches to ensure global stability. In particular, if S10 ≤ θI2α
+c1
1
c2
and S20 ≤ 2α 2
, then (C1) and (C2) will
be automatically satisfied.
We will follow the geometric approach originally proposed by Li and Muldowney [30] to investigate the
global asymptotic stabilities of the nontrivial equilibria. For completeness, we first present the following
result from [30].

Lemma 1. Consider a dynamical system dX n 1

dt = f (X), where f : D → R is a C function and D ⊂ R is
n

a simply connected domain. Assume that there exists a compact absorbing set K ⊂ D and the system has a
unique equilibrium point X ∗ in D . Then X ∗ is globally asymptotically stable in D if q¯2 < 0, where
1 t 


q¯2 = lim sup sup m Q(X(s, X0 )) ds. (16)
t→∞ X0 ∈K t 0
56 C. Yang et al. / Nonlinear Analysis: Real World Applications 38 (2017) 49–67

In Eq. (16), Q is a matrix-valued function defined as

Q = Pf P −1 + P J [2] P −1 ,

where P (X) is a n2 × n2 matrix-valued C 1 function in D, Pf is the derivative of P (entry-wise) along the

   

direction of f , and J [2] is the second additive compound matrix of the Jacobian J(X) = Df (X) . Meanwhile,
m(Q) is the Lozinskiǐ measure of Q with respect to a matrix norm; i.e.,
|I + hQ| − 1
m(Q) = lim+ ,
h→0 h
where I represents the identity matrix.
Now we are ready to prove the following global stability result.

Theorem 2.4. (i) If R1 > 1, then the endemic equilibrium E2 exists and is global asymptotically stable,
provided that the assumptions (C1) and (C2) hold.
(ii) If R1 < 1 and R2 > 1, then the boundary equilibrium E1 is global asymptotically stable.

Proof. We apply the geometric approach, summarized in Lemma 1, to analyze the global stabilities.

(i). Note that patch 1 does not depend on patch 2 and that the endemic equilibrium of patch 1 is
(1)
(uniquely) represented by the first three components of E2 ; i.e., E2 = (S1∗ , I1∗ , B1∗ ). This indicates that
(1)
the global stability of E2 in patch 1 can be analyzed independently, based on the first three equations in
(1)
system (1). Thus we will first prove that E2 is globally asymptotically stable in patch 1, using the geometric
approach.

By direct calculation, we find that the Jacobian matrix of the linearized subsystem in patch 1 is
 
−(α1 I1 + β1 B1 + Y1 ) −α1 S1 −β1 S1
J1 =  α1 I1 + β1 B1 α1 S1 − X1 β1 S1  ,
 
0 φ1 −d1
and the associated second additive compound matrix is
 
α1 S1 − (α1 I1 + β1 B1 + Y1 + X1 ) β1 S 1 β1 S1
[2]
J1 =  φ1 −(α1 I1 + β1 B1 + Y1 + d1 ) −α1 S1 .
 
0 α1 I1 + β1 B1 α1 S1 − X1 − d1
 
Define P1 = diag 1, BI11 , BI11 and let F1 denote the vector field of patch 1, then
 ˙
Ḃ1 I˙1

I1 Ḃ1
P1F1 P1−1 = diag 0, − , − ,
I1 B1 I1 B1
and
 B1 B1 
α S − (α1 I1 + β1 B1 + Y1 + X1 ) β1 S1 β1 S1
 1 1 I1 I1 
[2]
P1 J1 P1−1 = I1 .
 
φ 1 −(α I
1 1 + β 1 1 + Y1 + d1 )
B −α1 S1
 B 1

0 α1 I1 + β1 B1 α1 S1 − X1 − d1
[2]
The matrix Q(1) := P1F1 P1−1 + P1 J1 P1−1 can be written in the block form as follows
 
(1) (1)
Q Q
Q(1) = 11
(1)
12
(1) ,
Q21 Q22
 C. Yang et al. / Nonlinear Analysis: Real World Applications 38 (2017) 49–67 57

where
(1)
Q11 = α1 S1 − (α1 I1 + β1 B1 + Y1 + X1 ),
 
(1) B1 B1
Q12 = β1 S1 , β1 S1 ,
I1 I1
 T
(1) I1
Q21 = φ1 , 0 ,
B1
I˙1
 
Ḃ1
−(α1 I1 + β1 B1 + Y1 + d1 ) + − −α1 S1
(1) 
Q22 =  I 1 B1 
.
 ˙
I1 Ḃ1 
α1 I1 + β1 B1 α1 S1 − X1 − d1 + −
I1 B1
We now define the vector norm for any (x1 , x2 , x3 ) ∈ R3 as

|(x1 , x2 , x3 )| = max(|x1 |, |x2 | + |x3 |).

Let m denote the Lozinskiǐ measure with respect to this norm. By direct calculation, we find
   
(1) (1)
m Q(1) = sup g1 , g2
           
(1) (1)  (1)  (1)  (1)  (1)  (1)   (1) 
with g1 = m1 Q11 + Q12  , g2 = Q21  + m1 Q22 , where Q12  and Q21  are the matrix norms
induced by the L1 norm, and m1 denotes the Lozinskiǐ measure with respect to the L1 norm. Specifically,
(1) B1
g1 = α1 S1 − (α1 I1 + β1 B1 + Y1 + X1 ) + β1 S1 ,
I1
(1) I˙1
g2 = −Y1 + + sup(0, 2α1 S1 − X1 + Y1 ).
I1
I˙1
Observing that I1 = (α1 S1 + β1 S1 BI11 ) − X1 and using assumption (C1), we have

(1) I˙1 I˙1

g1 = − (α1 I1 + β1 B1 + Y1 ) ≤ − µ1 ,
I1 I1
(1) I˙1
g2 ≤ − µ1 .
I1
I˙1 A1 +A2
Hence m(Q(1) ) ≤ I1 − µ1 . In view of 0 ≤ I1 (t) ≤ N (t) ≤ min(µ1 ,µ2 ) , if t is large enough, then

ln(I1 (t)) − ln(I1 (0)) µ1

≤ .
t 2
Consequently,
t  t ˙ 
ln(I1 (t)) − ln(I1 (0))

1 1 I1 µ1
m(Q(1) )ds ≤ − µ1 ds = − µ1 ≤ −
t 0 t 0 I1 t 2
for t sufficiently large. Therefore, we obtain
 t
(1) 1 µ1
q¯2 := lim sup m(Q(1) )ds ≤ − < 0,
t→∞ t 0 2
(1)
which shows that E2 is globally asymptotically stable in patch 1.
Thus, to establish the global asymptotic stability of E2 in the entire domain, it is sufficient to show that
(2)
the endemic equilibrium of patch 2; i.e., E2 = (S2∗ , I2∗ , B2∗ ), is globally asymptotically stable in patch 2
under the condition (S1 , I1 , B1 ) = (S1∗ , I1∗ , B1∗ ). This can be proved in a similar way and details are provided
below.
58 C. Yang et al. / Nonlinear Analysis: Real World Applications 38 (2017) 49–67

For patch 2, the Jacobian matrix of the linearized subsystem is

 
−(α2 I2 + β2 B2 + µ2 ) −α2 S2 −β2 S2
J2 =  α2 I2 + β2 B2 α2 S2 − X2 β2 S2  ,
 
0 φ2 −d2
and the associated second additive compound matrix is
 
α2 S2 − (α2 I2 + β2 B2 + µ2 + X2 ) β2 S2 β2 S2
[2]
J2 =  φ2 −(α2 I2 + β2 B2 + µ2 + d2 ) −α2 S2 .
 
0 α2 I2 + β2 B2 α2 S2 − X2 − d2
 
Also define P2 = diag 1, BI22 , BI22 and let F2 denote the vector field of patch 2, then
 ˙
Ḃ2 I˙2

I2 Ḃ2
P2F2 P2−1 = diag 0, − , − ,
I2 B2 I2 B2
and
 B2 B2 
α S − (α2 I2 + β2 B2 + µ2 + X2 ) β2 S2 β2 S2
 2 2 I2 I2 
[2]
P2 J2 P2−1 = I2 .
 
φ2 −(α2 I2 + β2 B2 + µ2 + d2 ) −α2 S2
 B 2

0 α2 I2 + β2 B2 α2 S2 − X2 − d2
[2]
The matrix Q(2) := P2F2 P2−1 + P2 J2 P2−1 can be written in the block form as follows
 
(2) (2)
(2) Q11 Q12
Q = (2) (2) ,
Q21 Q22
where
(2)
Q11 = α2 S2 − (α2 I2 + β2 B2 + µ2 + X2 ),
 
(2) B2 B2
Q12 = β2 S2 , β2 S2 ,
I2 I2
 T
(2) I2
Q21 = φ2 , 0 ,
B2
I˙2
 
Ḃ2
−(α2 I2 + β2 B2 + µ2 + d2 ) + − −α2 S2
(2) 
Q22 =  I 2 B2 
.
 ˙
I2 Ḃ2 
α2 I2 + β2 B2 α2 S2 − X2 − d2 + −
I2 B2
 
  (2) (2)
Then we have m Q(2) = sup g1 , g2 , where

(2) B2
g1 = α2 S2 − (α2 I2 + β2 B2 + µ2 + X2 ) + β2 S2 ,
I2
(2) I˙2
g2 = −µ2 + + sup(0, 2α2 S2 − c2 ).
I2
I˙2 I∗
Since I2 = (α2 S2 + β2 S2 BI22 ) − X2 + θI I12 and the assumption (C2) holds, we have

(2) I˙2 − θI I1∗ I˙2

g1 = − (α2 I2 + β2 B2 + µ2 ) ≤ − µ2 ,
I2 I2
(2) I˙2
g2 ≤ − µ2 .
I2
 C. Yang et al. / Nonlinear Analysis: Real World Applications 38 (2017) 49–67 59

I˙2 A1 +A2
Hence m(Q(2) ) ≤ I2 − µ2 . Also notice that 0 ≤ I2 (t) ≤ N (t) ≤ min(µ1 ,µ2 ) . Then

ln(I2 (t)) − ln(I2 (0)) µ2

≤
t 2
if t is large enough. Therefore,

1 t 1 t I˙2
  
ln(I2 (t)) − ln(I2 (0))

µ2
m(Q(2) )ds ≤ − µ2 ds = − µ2 ≤ −
t 0 t 0 I2 t 2
t
for t sufficiently large. This implies q¯2 (2) := lim supt→∞ 1
t 0
m(Q(2) )ds ≤ − µ22 < 0 which establishes the
(2)
global stability of E2 in patch 2. Consequently, the proof of the global asymptotic stability of E2 in Γ is
complete.
(1) (2)
(ii). If R1 < 1 and R2 > 1, there is only one nontrivial equilibrium E1 and we may write E1 = (E1 , E1 ),
(1) 1 , 0, 0) is the disease-free equilibrium of patch 1, and E (2) = (S
where E1 = (S 2
2 , I2 , B
2 ) is the unique positive
equilibrium of patch 2 under the condition (S1 , I1 , B1 ) = (S1 , 0, 0). Similar to the proof of Theorems 2.1

(1) (2)
and 2.4 (i), it can be easily verified that E1 is globally asymptotically stable in patch 1, and E1 is globally
asymptotically stable in patch 2 provided that (S1 , I1 , B1 ) = (S 1 , 0, 0). Hence we conclude that E1 is globally
asymptotically stable. 

3. A two-patch periodic model

Having thoroughly analyzed the dynamics of the autonomous model, we now incorporate the seasonal
variation into our modeling framework. As mentioned before, brucellosis exhibits a strong seasonal pattern,
with concentrated mortality and morbidity burden in a few months each year. These seasonal fluctuations
could be represented by periodic changes in the various contact and transmission rates in our model.
For illustration, let us consider the temporal oscillation of the direct (i.e., host-to-host) and indirect
(i.e., environment-to-host) transmission rates. We define
   
πt
βj (t) = βj0 1 + a1 cos , j = 1, 2


  6 
πt
αj (t) = αj0 1 + a2 cos , j = 1, 2


6

where αj0 and βj0 denote the respective time-averaged (or, basic) transmission rates. Meanwhile, we
represent the pathogen shedding rate by
  
πt
φj (t) = φj0 1 + a3 cos , j = 1, 2
6

where φj0 denotes the basic shedding rate in the absence of seasonal forcing. Note that 0 ≤ ak ≤ 1 (k = 1, 2, 3)
denote the magnitude of seasonal fluctuations for these three parameters. For simplicity, we assume that
all other model parameters are the same as defined in system (1). Let again Sj , Ij and Bj (j = 1, 2)
denote, respectively, the susceptible animals, the infectious animals, and the environmental pathogens in the
jth patch. Within each patch, susceptible animals are infected through direct transmission from infectious
animals or indirect transmission from the contaminated environment, both in a time-periodic manner.
Meanwhile, the time-dependent pathogen shedding rate leads to seasonal growth of the Brucella population
in the environment. Thus, our new two-patch system incorporating both spatial and seasonal variations is
60 C. Yang et al. / Nonlinear Analysis: Real World Applications 38 (2017) 49–67

given by

S˙1 (t)


 = A1 − [α1 (t)I1 + β1 (t)B1 ]S1 − [θS + µ1 ]S1 ,
I˙1 (t) = [α1 (t)I1 + β1 (t)B1 ]S1 − [θI + c1 + µ1 ]I1 ,





Ḃ (t)
1 = φ1 (t)I1 − d1 B1 ,
˙2 (t) (17)

 S = A2 − [α2 (t)I2 + β2 (t)B2 ]S2 − µ2 S2 + θS S1 ,
I˙2 (t)




 = [α2 (t)I2 + β2 B2 (t)]S2 − (µ2 + c2 )I2 + θI I1 ,
Ḃ2 (t) = φ2 (t)I2 − d2 B2 .


It again can be easily verified that system (17) has a unique and bounded solution with any initial value
(Sj (0), Ij (0), Bj (0)) ∈ Γ , and that the compact set Γ is positively invariant with respect to system (17).
The model, however, is now a non-autonomous patchy system which exhibits more complex dynamics than
those of the previous autonomous model.

3.1. The basic reproduction number

 
It is straightforward to see that E 0 = S10 , 0, 0, S20 , 0, 0 remains the disease-free equilibrium for the model
(17). Thus, we can similarly introduce the next-generation matrices F (t) and V (t) (evaluated at the disease-
free equilibrium) as
 
α1 (t)S10 β1 (t)S10 0 0
 0 0 0 0 
F (t) =  ,
 
 0 0 α2 (t)S20 β2 (t)S20 
0 0 0 0
 
(c1 + θI + µ1 ) 0 0 0
 −φ1 (t) d1 0 0
and V (t) =  .
 
 −θI 0 (c2 + µ2 ) 0
0 0 −φ2 (t) d2

In order to define the basic reproduction number of this non-autonomous model, we follow the work of Wang
and Zhao [13]. They introduced the next-infection operator L for a model in periodic environments by
 ∞
(Lφ)(t) = Y (t, t − s)F (t − s)φ(t − s)ds, (18)
0

where Y (t, s), t ≥ s, is the evolution operator of the linear ω-periodic system dy/dt = −V (t)y and φ(t), the
initial distribution of infectious individuals, is ω-periodic and nonnegative. The basic reproduction number
is then defined as the spectral radius of the next-infection operator,

R0 = ρ(L). (19)

For our model (17), the evolution operator can be determined by solving the system of differential equations
dy/dt = −V (t)y with the initial condition Y (s, s) = I4×4 ; thus, we obtain
 
y11 (t, s) 0 0 0
y (t, s) y (t, s) 0 0 
 21 22
Y (t, s) =  (20)

y31 (t, s) 0 y33 (t, s) 0 

y41 (t, s) 0 y43 (t, s) y44 (t, s)
 C. Yang et al. / Nonlinear Analysis: Real World Applications 38 (2017) 49–67 61

where
y11 (t, s) = e−(c1 +θI +µ1 )(t−s) ,
 t   πx 
y21 (t, s) = φ10 e−d1 t ed1 x 1 + a3 cos y11 (x, s)dx,
s 6
 t
y31 (t, s) = θI e−(c2 +µ2 )t e(c2 +µ2 )x y11 (x, s)dx,
s
 t   πx 
y41 (t, s) = φ20 e−d2 t ed2 x 1 + a3 cos y31 (x, s)dx,
s 6
y22 (t, s) = e−d1 (t−s) ,
y33 (t, s) = e−(c2 +µ2 )(t−s) ,
 t   πx 
−d2 t
y43 (t, s) = φ20 e ed2 x 1 + a3 cos y33 (x, s)dx,
s 6
y44 (t, s) = e−d2 (t−s) .
The basic reproduction number defined in Eq. (19) can be numerically evaluated by using, for example, the
method described in [31].

3.2. Threshold dynamics

Using the basic reproduction number R0 , we aim to establish the threshold type result, stated in the
theorem below, for the periodic model (17). To that end, we first note that R2+ is positively invariant for the
following cooperative system [32]:

S˙1 (t) = A1 − (θS + µ1 )S1 ,

(21)
S˙2 (t) = A2 − µ2 S2 + θS S1 ,
and that (S10 , S20 ) is the unique equilibrium solution which is globally attractive in R2+ .

Theorem 3.1. (i) If R0 < 1, then the disease-free equilibrium E0 of system (17) is globally asymptotically
stable;
(ii) If R0 > 1, then system (17) admits at least one positive ω-periodic solution, and solutions of system
(17) are uniformly persistent.

Proof. If (S1 (t), I1 (t), B1 (t), S2 (t), I2 (t), B2 (t)) is a nonnegative solution of (17), then we have

S˙1 (t) ≤ A1 − (θS + µ1 )S1 ,
(22)
S˙2 (t) ≤ A2 − µ2 S2 + θS S1 .

Note that any nonnegative solution (S1 (t), S2 (t)) of system (21) approaches (S10 , S20 ) as t → ∞. It then
follows from the standard comparison theorem (see, e.g., [33, Theorem A.4]) that for any ϵ > 0, there is a
T > 0 such that
Si (t) < Si0 + ϵ, i = 1, 2, for t > T. (23)
Thus, for t > T , we have

I˙1 (t) ≤ [α1 (t)I1 + β1 (t)B1 ](S10 + ϵ) − [θI + c1 + µ1 ]I1 ,

Ḃ1 (t) ≤ φ1 (t)I1 − d1 B1 ,
(24)
I˙2 (t) ≤ [α2 (t)I2 + β2 B2 (t)](S20 + ϵ) − [µ2 + c2 ]I2 + θI I1 ,
Ḃ2 (t) ≤ φ2 (t)I2 − d2 B2 .
62 C. Yang et al. / Nonlinear Analysis: Real World Applications 38 (2017) 49–67

Define
 
α1 (t)(S10 + ϵ) β1 (t)(S10 + ϵ) 0 0
 0 0 0 0 
Fϵ (t) =  .
 
 0 0 α2 (t)(S20 + ϵ) β2 (t)(S20 + ϵ) 
0 0 0 0
By [13, Theorem 2.2], we have R0 < 1 ⇐⇒ ρ(φF −V (ω)) < 1, where ρ(φF −V (ω)) is the spectral radius of
φF −V (ω), and φF −V (ω) is the monodromy matrix of the linear ω-periodic system dy/dt = (F − V )y. Then
we can set ϵ sufficiently small such that ρ(φFϵ −V (ω)) < 1. As a consequence, the trivial solution (0, 0, 0, 0)
of the following linear ω-periodic system

I˙1 (t) = [α1 (t)I1 + β1 (t)B1 ](S10 + ϵ) − [θI + c1 + µ1 ]I1 ,

Ḃ1 (t) = φ1 (t)I1 − d1 B1 ,
(25)
I˙2 (t) = [α2 (t)I2 + β2 B2 (t)](S20 + ϵ) − [µ2 + c2 ]I2 + θI I1 ,
Ḃ2 (t) = φ2 (t)I2 − d2 B2 ,
is globally asymptotically stable. Again by the comparison theorem, we know that Ii (t) → 0, Bi (t) → 0 as
t → ∞ for i = 1, 2. Finally, the first and fourth equations of system (17) imply that Si (t) → Si0 as t → ∞
for i = 1, 2. This proves the result in part (i).
Now we consider the case R0 > 1. We define X = R6+ , X0 = R2+ × Int(R4+ ), ∂X0 = X\X0 , and denote
S(t) = (S1 (t), S2 (t)), I(t) = (I1 (t), I2 (t)), B(t) = (B1 (t), B2 (t)). It is easy to see that both X and X0
are positively invariant. Let P : R6+ → R6+ be the Poincaré map associated with system (17); that is,
P (x0 ) = u(ω, x0 ) for all x0 ∈ R6+ , where u(t, x0 ) is the unique solution of (17) with u(0, x0 ) = x0 . Set
M∂ = {(S(0), I(0), B(0)) ∈ ∂X0 : P m (S(0), I(0), B(0)) ∈ ∂X0 , ∀m ≥ 0},
M = {(S, I, B) : S ≥ 0, I = (0, I2 ), B = (0, B2 ), I2 ≥ 0, B2 ≥ 0}.
We first show that
M∂ = M. (26)
Clearly, M ⊆ M∂ . For any (S(0), I(0), B(0)) ∈ ∂X0 \M , if I1 (0) > 0, I2 (0) = 0, B(0) = 0, then
Ḃ1 (0) = φ1 (0)I1 (0) > 0, I˙2 (0) = θI (0)I1 (0) > 0. It follows that I(t) > 0, B1 (t) > 0 for 0 < t ≪ 1,
hence Ḃ2 (t) > 0 for 0 < t ≪ 1, which implies B(t) > 0 for 0 < t ≪ 1. If B1 (0) > 0, I1 (0) = 0,
then I˙1 (0) = α1 (0)B1 (0)S1 (0) > 0, we can still obtain I(t) > 0, B(t) > 0 for 0 < t ≪ 1. Thus,
we have (S(t), I(t), B(t)) ̸∈ ∂X0 for 0 < t ≪ 1. By the positive invariance of X0 , we know that
P m (S(0), I(0), B(0)) ̸∈ ∂X0 for m ≥ 1, hence (S(0), I(0), B(0)) ̸∈ M∂ , and thus (26) holds.
Now consider the fixed point M0 = (S 0 , 0, 0) of the Poincaré map P , where S 0 = (S10 , S20 ). Define
W (M0 ) = {x0 : P m (x0 ) → M0 , m → ∞}. We show that
S

W S (M0 ) ∩ X0 = ∅. (27)
Based on the continuity of solutions with respect to the initial conditions, for any ϵ > 0, there exists δ > 0
small enough such that for all (S(0), I(0), B(0)) ∈ X0 with ∥(S(0), I(0), B(0)) − M0 ∥ ≤ δ, we have
∥u(t, (S(0), I(0), B(0))) − u(t, M0 )∥ < ϵ, ∀t ∈ [0, ω]. (28)
To obtain (27), we claim that
lim sup ∥P m (S(0), I(0), B(0)) − M0 ∥ ≥ δ, ∀(S(0), I(0), B(0)) ∈ X0 . (29)
m→∞

We prove this claim by contradiction; that is, we suppose lim supm→∞ ∥P m (S(0), I(0), B(0)) − M0 ∥ < δ for
some (S(0), I(0), B(0)) ∈ X0 . Without loss of generality, we assume that ∥P m (S(0), I(0), B(0)) − M0 ∥ <
 C. Yang et al. / Nonlinear Analysis: Real World Applications 38 (2017) 49–67 63

Table 1
Parameters and their values.
Symbol Definition Value Unit Source
−1
cj Elimination rate due to brucellosis, (j = 1, 2) 0.15 year [9]
αj0 Averaged direct transmission rate, (j = 1, 2) 1.48 × 10−8 animal−1 year −1 [9]
βj0 Averaged indirect transmission rate, (j = 1, 2) 1.7 × 10−10 pathogen−1 year −1 [9]
φj0 Averaged Brucella shedding rate, (j = 1, 2) 15 pathogen animal−1 [9]
year −1
θS Averaged susceptible animals migration rate Varied year −1 –
θI Averaged infectious animals migration rate Varied year −1 –
a1 Amplitude of oscillation in βj (t), (j = 1, 2) 0.8 – Assumed
a2 Amplitude of oscillation in αj (t), (j = 1, 2) 0.8 – Assumed
a3 Amplitude of oscillation in φj (t), (j = 1, 2) 0.8 – Assumed
a4 Amplitude of oscillation in θS (t) 0.8 – Assumed
a5 Amplitude of oscillation in θI (t) 0.8 – Assumed
µj Natural elimination rate, (j = 1, 2) 0.22 year −1 [9]
Aj Animal recruitment rate, (j = 1, 2) 11629200 animals year −1 [9]
dj Pathogen decay rate, (j = 1, 2) 3.6 year −1 [9]
Sj (0) Initial susceptible animals, (j = 1, 2) 5.185 × 107 animals [9]
Ij (0) Initial infected animals, (j = 1, 2) 1.33 × 106 animals [9]
Bj (0) Initial load of Brucella, (j = 1, 2) 6 × 106 pathogens [9]

δ, ∀m ≥ 0. Thus,
∥u(t, P m (S(0), I(0), B(0))) − u(t, M0 )∥ < ϵ, ∀t ∈ [0, ω] and m ≥ 0. (30)
Moreover, for any t ≥ 0, we write t = t0 + kω with t0 ∈ [0, ω) and k = [t/ω], the greatest integer less than
or equal to t/ω. Then we obtain
∥u(t, (S(0), I(0), B(0))) − u(t, M0 )∥ = ∥u(t0 , P m (S(0), I(0), B(0))) − u(t0 , M0 )∥ < ϵ (31)
for any t ≥ 0. Let (S(t), I(t), B(t)) = u(t, (S(0), I(0), B(0))). It follows that −ϵ < S(t) − S 0 < ϵ, 0 < I(t) < ϵ
and 0 < B(t) < ϵ. Again based on [13, Theorem 2.2], R0 > 1 if and only if ρ(ΦF −V (ω)) > 1. Thus, for ϵ
small enough, we have ρ(ΦFϵ −V (ω)) > 1 which immediately yields the contradiction as
lim Ii (t) = ∞, lim Bi (t) = ∞, i = 1, 2.
t→∞ t→∞

Let P1 : R2+ −→ R2+ be the Poincareé map associated with (21). Then S 0 is globally attractive in R2+ \{0}
for P1 . It follows that M0 is isolated invariant set in X, and notice that W S (M0 )∩X0 = ∅. Hence, every orbit
in M∂ converges to M0 and M0 is acyclic in M∂ . By [34, Theorem 1.3.1], for a stronger repelling property
of ∂X0 , we conclude that P is uniformly persistent with respect to (X0 , ∂X0 ), which implies the uniform
persistence of the solutions of system (17) with respect to (X0 , ∂X0 ) [34, Theorem 3.1.1]. Consequently,
based on [34, Theorem 1.3.6], the Poincaré map P has a fixed point (S̄(0), I(0), ¯ B̄(0)) ∈ X0 , and it can be
¯ 6 ¯ ¯
easily seen that S̄(0) ̸= 0. Thus, (S̄(0), I(0), B̄(0)) ∈ Int(R+ ) and (S̄(t), I(t), B̄(t)) = u(t, (S̄(0), I(0), B̄(0)))
is a positive ω-periodic solution of the system. 

4. Numerical results

In this section, we conduct some numerical simulation in order to verify our analytical findings. We list
the model parameters and their numerical values in Table 1.
Using the parameter values in Table 1, we first perform a numerical study on the autonomous model
(1). Our main analytical result, Theorem 2.4, states the global asymptotic stability of the two nontrivial
equilibria, E1 and E2 . Numerically, we vary the animal migration rates θS and θI so as to generate
different values of R1 and R2 . For each of these scenarios, we pick a number of initial conditions and
conduct separate simulations to system (1). Fig. 1 illustrates a case where we set θS = θI = 0.2
. .
which results in R1 = 0.75 < 1, R2 = 3.27 > 1, and all solution curves converge to the equilibrium
64 C. Yang et al. / Nonlinear Analysis: Real World Applications 38 (2017) 49–67

Fig. 1. Phase portrait illustrating the global stability of E1 for system (1) in the S2 –I2 plane with R1 < 1, R2 > 1. Each
curve in the plot corresponds to a different initial condition, and all these curves converge to the equilibrium E1 (where
. .
S2 = 2.39 × 107 , I2 = 3.22 × 107 ) over time.

.
E1 = (2.77 × 107 , 0, 0, 2.39 × 107 , 3.22 × 107 , 1.34 × 108 ) over time. The same convergence pattern is
observed throughout the regime R1 < 1, R2 > 1, though E1 changes as θS and θI vary. In contrast, Fig. 2
.
illustrates that when we set θS = θI = 0.1 which results in R1 = 1.2 > 1, all solution curves converge to
.
the equilibrium E2 = (3.03 × 107 , 4.11 × 106 , 1.71 × 107 , 2.29 × 107 , 2.71 × 107 , 1.13 × 108 ) . As θS and θI
vary, E2 also changes but the same type of convergence to E2 is numerically observed for all R1 > 1. These
results demonstrate the analytical predictions in Theorem 2.4.
Next, we turn to the periodic model (17) and numerically demonstrate the threshold dynamics result in
Theorem 3.1. Fig. 3 plots the time evolution of two infection curves, I1 for patch 1 and I2 for patch 2, when
R0 < 1. We clearly observe that both curves approach the disease-free equilibrium E0 (where I1 = I2 = 0)
over time. Though not shown here, the same pattern is observed when we vary the initial conditions for
I1 and I2 , an evidence for the global asymptotic stability of E0 when R0 < 1. Fig. 4, on the other hand,
illustrates the case with R0 > 1. We observe that each infection curve converges to a periodic solution with
a period ω = 12 months, highlighting the persistence of the infection when R0 > 1. Particularly, we note
that the infection curve I2 stays at a much higher level and exhibits a much stronger oscillation than those
with the curve I1 , showing a higher disease prevalence and risk associated with patch 2 due to the animal
migration. This result indicates that in a (simple) population system of two patches where animals migrate
from patch 1 to patch 2, with otherwise identical characteristics between the two patches, more prevention
and intervention efforts should be devoted to patch 2 in order to control brucellosis outbreaks. Furthermore,
such disease control strategies should take into account the seasonal fluctuations of brucellosis so as to make
best use of available resources.

5. Conclusion and discussion

We have proposed a new mathematical modeling framework for the dynamics of brucellosis, incorporating
multiple transmission pathways and both spatial and temporal heterogeneities. As a demonstration of this
framework, we have focused on a two-patch model throughout this paper. We started our analysis on the
two-patch model with fixed coefficients (an autonomous system) where detailed results were obtained,
showing the rich dynamics of brucellosis transmission due to the spatial variation. In particular, we have
 C. Yang et al. / Nonlinear Analysis: Real World Applications 38 (2017) 49–67 65

Fig. 2. Phase portrait illustrating the global stability of E2 for system (1) in the S2 –I2 plane with R1 > 1. Each curve in the plot
. .
corresponds to a different initial condition, and all these curves converge to the equilibrium E2 (where S2 = 2.29 × 107 , I2 =
2.71 × 107 ) over time.

Fig. 3. The infection curves for the two patches associated with system (17) when R0 < 1. Both curves converge to the disease-free
equilibrium E0 over time.

thoroughly characterized the multiple equilibria of the system and their stabilities, using the reproduction
numbers associated with the model. In the second part of this study, we extended our model to a time-periodic
environment that represents seasonal oscillation. We then calculated the basic reproduction number, R0 , for
this periodic two-patch model and established the threshold result: when R0 < 1, the disease-free equilibrium
is globally stable; when R0 > 1, the disease is uniformly persistent.
For our autonomous model, we were able to completely determine its local and global dynamics.
Particularly, we applied the geometric approach to prove the global asymptotic stability of the nontrivial
equilibria. For the periodic two-patch model, however, the dynamics are more complex. Although we were
able to establish the uniform persistence result, we have not resolved the stabilities of the nontrivial periodic
orbits when R0 > 1 and that remains an interesting topic for our future research. Nevertheless, our current
study demonstrates that the incorporation of spatial and temporal variations leads to rich and complex
66 C. Yang et al. / Nonlinear Analysis: Real World Applications 38 (2017) 49–67

Fig. 4. The infection curves for the two patches associated with system (17) when R0 > 1. Each curve converges to a periodic
oscillation with a period ω = 12 months.

dynamics that are distinct from those observed from prior models based on homogeneous environments.
Our results also indicate that the prevention and intervention strategies need to take into account the
spatial and temporal heterogeneities in order to effectively control brucellosis while optimize the use of
available resources.
A potential limitation of the present paper is that we have employed the bilinear incidence for both
the direct and indirect transmission routes. Such bilinear incidence forms, with the advantage of making
model calculations more tractable, have been extensively used in previously published brucellosis modeling
work (see, e.g., [8,9,11]). On the other hand, for realistic applications, the indirect (i.e., environment-to-
host) transmission might be better represented by saturated type functional responses, and such type of
incidence forms have been used in modeling some other environmentally transmitted diseases (such as cholera
[35,36]). It would be interesting to employ saturated type incidence in our future work on brucellosis
modeling, for both the autonomous and time-periodic cases.
Our current study on the spatial modeling of brucellosis is based on the patch structure, and we plan
to extend this work to a more general setting with an arbitrary number of patches. We expect that many
results presented in this paper can be similarly established for the general spatial setting. On the other hand,
a different approach to model spatial heterogeneity is to utilize partial differential equations (PDEs), for
example, by adding diffusion terms to an ODE model so as to represent the movement and dispersal of the
animals and the pathogen, and by adding convection terms to represent the migration of animals. Seasonal
variation can be similarly incorporated by considering periodic model parameters, resulting in a periodic
PDE system. Such a periodic PDE model can be possibly analyzed using techniques recently proposed by
Zhao and co-workers [37–39]. In particular, the basic reproduction number can still be defined which can
be used to investigate the threshold properties of the model. It will be very interesting to compare the
threshold dynamics from these two modeling approaches: one based on multi-patch periodic ODEs and the
other based on periodic PDEs.

Acknowledgments

This work was partially supported by the National Science Foundation under grant numbers 1412826 and
1557739. The authors are grateful to the three anonymous referees for their helpful comments that have
improved the original manuscript.
 C. Yang et al. / Nonlinear Analysis: Real World Applications 38 (2017) 49–67 67

References

[1] M.L. Boschiroli, V. Foulogone, D. O’Callaghan, Brucellosis: A worldwide zoonosis, Curr. Opin. Microbiol. 4 (2001) 58–64.
[2] G. Pappas, N. Akritidis, M. Bosikovski, E. Tasianos, Brucellosis, N. Engl. J. Med. 352 (2005) 2325–2336.
[3] M.M.H. Sewel, D.W. Brocklesby (Eds.), Handbook on Animal Diseases in the Tropics, fourth ed., Bailliere Tindall, London,
1990.
[4] M.M. Madkour, Madkour’s Brucellosis, Springer Verlag, Heidelberg, Berlin, 2001.
[5] E.J. Richey, C. Dix Harrel, Brucella Abortus (Brucellosis) in Beef Cattle, University of Florida, IFAS Extension, VM 100
(1997), pp. 1–6.
[6] M.J. Corbel, Brucellosis: An overview, Emerg. Infect. Diseases 3 (1997) 213–221.
[7] M.J. Mangen, J. Otte, D. Pfeiffer, P. Chilonda, Bovine brucellosis in Sub-Sahara Africa: Estimation of sero-prevalence and
impact on meat and milk offtake potential, Livestock Policy Discussion Paper N. 8, Food and Agriculture Organization of
the United Nations (FAO) Livestock Information and Policy Branch, Rome, 2002.
[8] B. Ainseba, C. Benosman, P. Magal, A model for ovine brucellosis incorporating direct and indirect transmission, J. Biol.
Dyn. 4 (2010) 2–11.
[9] Q. Hou, X. Sun, J. Zhang, Y. Liu, Y. Wang, Z. Jin, Modeling the transmission dynamics of sheep brucellosis in Inner
Mongolia autonomous region, China, Math. Biosci. 242 (2013) 51–58.
[10] P.O. Lolika, S. Mushayabasa, C.P. Bhunu, C. Modnak, J. Wang, Modeling and analyzing the effects of seasonality on
brucellosis infection, 2016 (submitted for publication).
[11] M. Li, G. Sun, J. Zhang, Z. Jin, X. Sun, Y. Wang, B. Huang, Y. Zheng, Transmission dynamics and control for brucellosis
model in Hingaan League Inner Mongolia, China, Math. Biosci. Eng. 11 (2014) 1115–1137.
[12] D. Posny, J. Wang, Modelling cholera in periodic environments, J. Biol. Dyn. 8 (2014) 1–19.
[13] W. Wang, X.-Q. Zhao, Threshold dynamics for compartment epidemic models in periodic environments, J. Dynam.
Differential Equations 20 (2008) 699–717.
[14] X.-Q. Zhao, Uniform persistence and periodic coexistence states in infinite-dimensional periodic semiflows with applications,
Can. Appl. Math. Q. 3 (1995) 473–495.
[15] K. Aune, J.C. Rhyan, R. Russell, T.J. Roffe, B. Corso, Environmental persistence of Brucella abortus in the Greater
Yellowstone area, J. Wildl. Manage. 76 (2012) 253–261.
[16] W. Beauvais, I. Musallam, J. Guitian, Vaccination control programs for multiple livestock host species: An age-stratified,
seasonal transmission model for brucellosis control in endemic settings, Parasites Vectors 9 (2016) 55.
[17] World Health Organization Brucellosis in Humans and Animals, WHO Press, 2006.
[18] W.D. Wang, P. Fergola, C. Tenneriello, Innovation diffusion model in patch environment, Appl. Math. Comput. 134 (2003)
51–67.
[19] W.D. Wang, X.Q. Zhao, An epidemic model in a patchy environment, Math. Biosci. 190 (2004) 97–112.
[20] P. van den Driessche, J. Watmough, Reproduction number and subthreshold endemic equilibria for compartment models
of disease transmission, Math. Biosci. 180 (2002) 29–48.
[21] A. Korobeinikov, Lyapunov functions and global properties for SEIR and SEIS epidemic models, Math. Med. Biol. 21
(2004) 75–83.
[22] C.C. McCluskey, Lyapunov functions for tuberculosis models with fast and slow progression, Math. Biosci. Eng. 3 (2006)
603–614.
[23] Z. Mukandavire, C. Chiyaka, G. Magombedze, G. Musuka, N.J. Malunguza, Assessing the effects of homosexuals and
bisexuals on the intrinsic dynamics of HIV/AIDS in heterosexual settings, Math. Comput. Modelling 49 (2009) 1869–1882.
[24] Z. Shuai, J.A.P. Heesterbeek, P. van den Driessche, Extending the type reproduction number to infectious disease control
targeting contact between types, J. Math. Biol 67 (2013) 1067–1082.
[25] J.P. LaSalle, The Stability of Dynamical Systems, in: CBMS-NSF Regional Conference Series in Applied Mathematics,
vol. 25, SIAM, Philadelphia, 1976.
[26] H.I. Freedman, S. Ruan, M. Tang, Uniform persistence and flows near a closed positively invariant set, J. Dynam.
Differential Equations 6 (1994) 583–600.
[27] M.Y. Li, J.R. Graef, L. Wang, J. Karsai, Global dynamics of a SEIR model with varying total population size, Math.
Biosci. 160 (1999) 191–213.
[28] H.R. Thieme, Convergence results and a Poincare’–Bendixon trichotomy for asymptotically autonomous differential
equations, J. Math. Biol. 30 (1992) 755–763.
[29] H.R. Thieme, Persistence under relaxed point-dissipativity (with application to an endemic model), SIAM J. Math. Biosci.
166 (1993) 407–435.
[30] M.Y. Li, J.S. Muldowney, A geometric approach to global-stability problems, SIAM J. Math. Anal. 27 (1996) 1070–1083.
[31] D. Posny, J. Wang, Computing basic reproductive numbers for epidemiological models in nonhomogeneous environments,
Appl. Math. Comput. 242 (2014) 473–490.
[32] X.-Q. Zhao, Z.J. Jing, Global asymptotic behavior in some cooperative systems of functional differential equations, Can.
Appl. Math. Q. 4 (1996) 421–444.
[33] H.L. Smith, P. Waltman, The Theory of the Chemostat, Cambridge University Press, Cambridge, 1995.
[34] X.-Q. Zhao, Dynamical Systems in Population Biology, Springer, New York, 2003.
[35] C.T. Codeço, Endemic and epidemic dynamics of cholera: the role of the aquatic reservoir, BMC Infect. Dis. 1 (2001) 1.
[36] Z. Mukandavire, S. Liao, J. Wang, H. Gaff, D.L. Smith, J.G. Morris, Estimating the reproductive numbers for the 2008–2009
cholera outbreaks in Zimbabwe, Proc. Natl. Acad. Sci. 108 (2011) 8767–8772.
[37] R. Peng, X.-Q. Zhao, A nonlocal and periodic reaction–diffusion-advection model of a single phytoplankton species,
J. Math. Biol. 72 (2016) 755–791.
[38] R. Peng, X.-Q. Zhao, A reaction–diffusion SIS epidemic model in a time-periodic environment, Nonlinearity 25 (2012)
1451–1471.
[39] X. Yu, X.-Q. Zhao, A nonlocal spatial model for Lyme disease, J. Differential Equations 261 (2016) 340–372.