J. Math. Biol.

(1997) 35: 523—544

Competitive exclusion in a vector-host model for the

dengue fever
Zhilan Feng1, Jorge X. Velasco-Hernández2,w
1 Biometrics Unit, Cornell University, Ithaca, NY 14853—7801, USA
2 Departamento de Matemáticas, UAM-Iztapalapa, México

Received 4 December 1995; received in revised form 5 March 1996

Abstract. We study a system of differential equations that models the popula-

tion dynamics of an SIR vector transmitted disease with two pathogen strains.
This model arose from our study of the population dynamics of dengue fever.
The dengue virus presents four serotypes each induces host immunity but only
certain degree of cross-immunity to heterologous serotypes. Our model has
been constructed to study both the epidemiological trends of the disease and
conditions that permit coexistence in competing strains. Dengue is in the
Americas an epidemic disease and our model reproduces this kind of dynam-
ics. We consider two viral strains and temporary cross-immunity. Our analy-
sis shows the existence of an unstable endemic state (‘saddle’ point) that
produces a long transient behavior where both dengue serotypes cocirculate.
Conditions for asymptotic stability of equilibria are discussed supported by
numerical simulations. We argue that the existence of competitive exclusion in
this system is product of the interplay between the host superinfection process
and frequency-dependent (vector to host) contact rates.

Key words: Mathematical epidemiology — Dengue — Differential equations —

Vectorborne diseases — Population dynamics

1 Introduction

Dengue fever is a viral disease endemic in many areas of the world that is
invading and recolonizing regions where either it was absent or it had been
eradicated. The dengue virus has 4 different serotypes. We construct and
analyze a mathematical model for its transmission dynamics. The model is

—————
w
Present address: Biometrics Unit, Cornell University, Ithaca, NY
14853—7801, USA
524 Z. Feng, J. X. Velasco-Hernández

a system of differential equations that incorporates variable population size in

both host and mosquito populations, two co-circulating strains and frequency
dependent biting rates. The model constitutes a framework to discuss condi-
tions for coexistence or competitive exclusion of closely related pathogen
strains.
In the next section we give a basic background on the disease summarizing
its epidemiological importance as well as the main features incorporated into
our model. Then we proceed with the model formulation and discuss previous
work on the principle of competitive exclusion. This section is followed by
model analysis in absence of disease induced mortality (negligible virulence).
Then we discuss the computational results for the case of nonnegligible
virulence and, finally, in the last section, we compare our findings with other
models on superinfection, variable population size and frequency dependent
infection rates.

1.1 Dengue fever

We start with a brief summary of the epidemiology of dengue. We follow

references [23, 22] as well as others that are indicated where appropriate.
In developing countries population growth is an important factor that
contributes to the increase in the incidence of communicable diseases which
affect mainly the urban poor, with infants and children among the groups
particularly at risk [23, 22]. Urbanization and population growth increase the
demand on basic essential services such as housing, water supply, etc., and at
the same time induce conditions that increase the transmission potential of
some vector-borne diseases [23]. Inadequacies in water supplies require
large-scale water storage which are ideal breeding habitats for Aedes spp
mosquitoes, the vectors of dengue fever, dengue hemorrhagic fever and dengue
shock syndrome as well as yellow fever. Changes in food habits lead to
increase use of tinned food and more use of disposable containers that provide
breeding sites to vectors of this type. In summary population growth, urban-
ization and poverty enhance presence and transmission of infectious diseases.
Unfortunately not only dengue has increased its incidence in urban centers
of the developing world but also yellow fever, malaria and Chagas disease
have been benefited. The destruction of city water supplies, temporary hous-
ing for refugees, high fertility and rural to urban migration and the steady
deterioration of urban environments have led to sustained growth in density
and area occupied by Aedes aegypti and Aedes albopictus, two of the main
vectors of dengue virus.
Other important problems of the dengue virus in the Americas and
elsewhere are the public health consequences of global warming [25]. Of
concern is the potential spread of dengue through the vector Aedes albopictus,
recently introduced to the American continent [24].
Dengue causes a spectrum of illnesses in humans ranging from clinically
inapparent to severe and fatal hemorrhagic disease [8]. Classical dengue fever
Competitive exclusion in dengue 525

is generally observed in older children and adults and is characterized by

sudden onset of fever, frontal headache, nausea, vomiting and other symp-
toms. The acute illness last for 3 to 7 days is usually benign. The hemorrhagic
form of dengue and its associated dengue shock syndrome (DHF-DSS) is most
commonly observed in children under the age of 15 years but it can also occur
in adults [8]. It is characterized by acute onset of fever and a variety of
symptoms that last 2 to 7 days. This form of dengue can terminate in death of
the patient.
Dengue is produced by viruses of the genus ¹ogaviridae, subgenus
Flavivirus. Four distinct dengue viruses have been distinguished, denoted by
types 1, 2, 3 and 4. Dengue viruses can infect only a restricted number of
vertebrates but it is an essentially human disease [10]. Infection by any
dengue virus strain produces long lasting immunity but only temporary
cross-immunity to other serotypes. Three of the vectors are Aedes aegypti
Linnaeus, Aedes albopictus Skuse, and Aedes scutellaris Walk. Aedes aegypti
mosquitoes acquire infection from infected individuals 6 to 18 h before onset
of fever and then for the duration of the fever. A minimum extrinsic incubation
period of 8 to 14 days is required after an infective blood meal before the
mosquito becomes infectious. The infection in the vector is for life. Dengue
virus is transmitted in two cycles: urban and sylvan although, as mentioned
before, it is predominantly a human virus [10].
There are no effective programs for vector control and, as a consequence,
the absolute numbers of dengue infection and dengue infection rates have
increased during the last 40 years [11]. Unfortunately, countries where posit-
ive results exist for vector eradication have been suffering from epidemic
outbreaks: the disease is coming back. Dengue viruses were introduced in the
Americas around 1960 and since, dengue has been reported in countries where
it was absent before as Cuba, México [19], the United States, most Central
America, Ecuador, Perú, Paraguay, Bolivia, Argentina and Brazil [11].
Dengue transmission occurs throughout the year in endemic tropical areas
but there exists, however, a distinct cyclical pattern associated with the rainy
season [8]. In particular, in Thailand where the vector life cycle is highly
domiciliary, temperature and humidity conditions during the rainy season
favor survival of infected mosquitoes. In the Americas the situation is different
since in these areas larvae develop in the outdoors. Here, peak transmission
takes place in the days of highest rainfall and warmer temperatures season [8].
In regions where mosquito and humans exists, an introduction of dengue
virus may produce an epidemic depending upon a) the strain of the virus
(influencing magnitude and duration of viremia), b) the susceptibility of the
human population, c) the density, behavior, and competence of the mosquito
vector population, and d) the introduction of the virus into an area where it
has contact with the local mosquito population [8]. Severity of dengue fever
has been associated with secondary dengue infections although its causes are
far from being explained. Epidemiological studies in Thailand suggest that an
important risk factor for DHF-DSS is the presence of preexisting dengue
antibody at subneutralizing levels. Also, endemic DHF-DSS is found in areas
526 Z. Feng, J. X. Velasco-Hernández

Table 1. Absolute numbers of dengue hemorrhagic fever cases and deaths reported to the
World Health Organization regional offices. NR: not reported. (From Halstead,1992)

where Aedes aegypti densities are high and dengue virus of multiple types are
endemic. Moreover, DHF-DSS is associated with individuals older than
1 year with a secondary type antibody response and with primary infections in
newborn babies whose mother were immune to dengue [8, 10]. These facts led
to the formulation of the secondary infection or immune enhancement hy-
pothesis to explain it [8]. This hypothesis states that only those persons
experiencing a second infection with heterologous dengue serotype present
DHF-DSS. In particular it has been found that only secondary dengue-2
disease is immunologically enhanced and that infection with this virus sero-
type cause the majority of DSS cases [11]. Other factors are also associ-
ated with DHF-DSS. These are sex (more frequent infections in females),
nutritional status (higher incidence in well-nourished babies of middle
and upper socioeconomic class), and the interval between first and second
infections.
To conclude this brief review of the epidemiology of dengue, we show in
Table 1 a summary of dengue epidemic in several countries of the world.

1.2 Superinfection and coexistence

In dengue certain sequences of infection appear to be more damaging to the

host than others. We model this process using a susceptibility coefficient (cf.
[12, 4]). This coefficient allows us to explore varying degrees of susceptibility
to secondary infections and their effect on the asymptotic dynamics of the
disease. Through model analysis we explore the consequences of coupling two
populations that differ in the infection pattern (SIR with superinfection in the
Competitive exclusion in dengue 527

host, SI in the vector), and the effect of frequency-dependent infection rates on

the coexistence of closely related strains.
Most diseases are produced by an spectrum of closely related pathogens
rather than by single strains and dengue is clearly an example of this assertion.
In dengue an analogous phenomenon to superinfection (Nowak and May
[21] May and Nowak [16]) occurs. One strain invades the host population,
produces a brief period of temporary immunity to other strains but when the
immunity is lost the host becomes susceptible to reinfection with another
strain. In dengue, before reinfection can occur, there is a period where the host
is resistant, in varying degrees, to all strains [8]. In dengue fever we are thus
confronted with a vector-transmitted disease, co-circulating strains, certain
degree of cross-immunity or even increased susceptibility to infection, and
a variable host population size. Under these conditions one important theo-
retical problem that we address here is that of the coexistence of all strains or
the eventual extinction of some of them. A similar problem has been theoret-
ically explored by several authors [12, 2, 13, 5].
There are numerous published results discussing the problem of coexis-
tence in pathogen-host interactions. Levin and Pimentel [12] constructed
a mathematical SI model where the population in the absence of disease
grows exponentially. Two strains with different virulences compete with each
other. The most virulent strain can ‘takeover’ hosts already infected with the
less virulent strain. With these assumptions a globally stable equilibrium is
possible where both strains may coexist [3]. The stability of the positive
equilibrium is only guaranteed for certain range of values of superinfection.
Outside this range one of the boundary equilibria is asymptotically stable.
Bremermann and Thieme [2] postulate a competitive exclusion principle
in an SIR epidemic in a population with variable size. Several strains compete
for a single host population. The pathogens differ on their virulence. In this
model virulence is a strictly convex function of the transmission rate implying
that the evolution of virulence leads to a transmission rate that maximizes the
basic reproductive number of the pathogen [2].
Castillo-Chavez et al. [13] find, for a SIS two-sex model with variable
population size, that competitive exclusion is the norm: the strain with the
highest reproductive number persists in both host types. Mena-Lorca,
Velasco-Hernandez and Castillo-Chavez [17] studied the effect of variable
population, virulence and density-dependent population regulation. In this
model too, coexistence is feasible only in certain window of parameter space.

2 A mathematical model for dengue with two strains

Previous models for dengue fever are reported in [26, 20]. The former is
a cost-effectiveness model for the management of dengue. It addresses socio-
logical-epidemiological issues that we do not consider here. The second model
follows the same basic methodology that we adopt. This model [20] incorpo-
rates an incubation or latent period for both mosquitoes and humans. Both
528 Z. Feng, J. X. Velasco-Hernández

total populations of hosts and vectors are considered constant. The model
predicts an asymptotically stable endemic state if the basic reproductive
number is greater than one.
In the model that we analyze here, we consider variable population sizes of
both hosts and vector populations, we do not incorporate the exposed
compartment, but include instead the existence of a second co-circulating
strain that can produce secondary infections in those individuals either
susceptible or already recovered from a primary infection with a different
strain.

2.1 Model equations

Consider a human population settled in a region where a mosquito popula-

tion of the genus Aedes is present and carrier of the dengue virus.
Model equations then stand as follows (@"d/dt):
S@(t)"h!(B #B ) S!uS ,
1 2
I@ (t)"B S!p B I !uI ,
1 1 2 2 1 1
I@ (t)"B S!p B I !uI ,
2 2 1 1 2 2 (1)
½@ (t)"p B I !(e #u#r) ½ ,
1 1 1 2 1 1
½@ (t)"p B I !(e #u#r) ½ ,
2 2 2 1 2 2
R@(t)"r(½ #½ )!uR ,
1 2
and
M@(t)"q!(A #A ) M!dM ,
1 2
»@ (t)"A M!d» , (2)
1 1 1
»@ (t)"A M!d» .
2 2 2
N"S#I #I #½ #½ #R and ¹"M#» #» are the host and
1 2 1 2 1 2
vector total population sizes respectively (see Table 2 for other parameter
definitions and values). Primary infections in human hosts are produced by
either of the two strains at rates
B "b » /(c#u N)
i i i h
for i"1, 2 (in vector to host transmission). Primary infections in vectors are
produced at rates
A "a (I #½ )/(c#u N) .
i i i i v
These function forms describe frequency-dependent disease transmission.
Both are special cases of the Holling type II functional response [6] and are
also generalizations the contact rates of the Ross-Macdonald model for
Malaria [1] and for Chagas disease [27].
Competitive exclusion in dengue 529

Table 2. Parameter definitions and values used in the simulations illuestrated in the figures

We assume that once a mosquito is infected it never recovers and it cannot

be reinfected with a different strain of virus. Secondary infections, therefore,
may take place only in the host. Two cases can occur: either previously
I individuals are infected by strain 2, through contact with infected mos-
1
quitoes » , becoming ½ hosts, or previously I individuals are infected with
2 2 2
strain 1, through contact with » mosquitoes, to become ½ infected hosts, at
1 1
rates p B I and p B I , respectively. Here, p is a positive real number that
1 1 2 2 2 1 i
may mimic either cross-immunity (p (1) or increased susceptibility (p '1)
i i
by immune enhancement. This type of dynamics is analogous to superinfec-
tion (cf. [21]). In dengue, the immunity developed after infection is a factor
that does not appear in superinfection models. In dengue, either of the
primary infected populations can be reinfected with the other strain. General
results on the effects of cross-immunity in SIS and SI models respectively
[21, 17], indicate that for certain values of p coexistence of competing strains
is possible. As we will show, the existence of cross-immunity together with the
induction of specific permanent immunity, and frequency dependent contact
rates, prevent coexistence. The generic outcome of our model is competitive
exclusion although, in some cases, in a very long time scale.
To summarize, if p (1, primary infections confer partial immunity to
i
strain i; if p "1 secondary infections with strain i take place as if they were
i
primary infections, and if p '1 primary infections increase susceptibility to
i
strain i. Once an individual has suffered from both infections it gets immunity
to both strains at a rate r independent of the sequence of infections.
Since the equation for the total vector population is
¹@"q!d¹ ,
we have that as tPR, ¹(t)Pq/d. This allows us to substitute M"
q/d!» !» making the equation for M in (2) redundant.
1 2
530 Z. Feng, J. X. Velasco-Hernández

2.2 Basic reproductive number

Let
X"M(S, I , I , ½ , ½ , R, » , » ) :
1 2 1 2 1 2
S#I #I #½ #½ #R6h/u,
1 2 1 2
» #» 6q/dN
1 2
be the set bounded by the total host and vector population in the absence of
disease. We can immediately identify three equilibrium solutions to (1—2), the
disease-free equilibrium E*"(S*, 0, 0, 0, 0, 0, 0, 0) and the two (boundary)
0
equilibria

E*"(S*, I*, 0, 0, 0, 0, »*, 0) ,

1 1 1 1
where only strain 1 survives, and

E*"(S*, 0, I*, 0, 0, 0, 0, »*)

2 2 2 2
where only strain 2 survives.
The basic reproduction number is defined as the number of secondary
infections that a single infectious individual produces in a population where
all hosts are susceptible. It provides an invasion criterion for the initial spread
of the virus in a susceptible population
To find the basic reproductive number for our model we equate (1—2) to
zero and rewrite it as

S"h/(u#B #B ) ,
1 2
I "B S/(u#p B ), iOj, i, j"1, 2 ,
i i j j
½ "p B I /(u#e #r), iOj, i, j"1, 2 ,
i i i j j
R"r(½ #½ ) /u ,
1 2
» "qA /(d#A #A ) ,
1 1 1 2
» "qA /(d#A #A ) .
2 2 1 2
Let K"(B , B , A , A )@ (@ denotes transpose). Substituting the above expres-
1 2 1 2
sions into the definition of B and A we obtain a system of four non-linear
i i
algebraic equations in terms of B and A . We denote this system by U(K) . The
i i
solutions of K"U(K) give, by construction, all the equilibrium points of
(1—2). This equation also allows an easy computation of the next generation
operator [7] and the associated basic reproduction number. By construction,
Competitive exclusion in dengue 531

the next-generation operator is simply the Jacobian of U evaluated at the

disease-free equilibrium (given by B "B "A "A "0):
1 2 1 2

A B
b h
0 0 1 0
u2(c#hu /u) b h
h 2
0 0 0
DU(0)" a q u2(c#hu /u) .
1 h
0 0 0
d2(c#hu /u) a q
v 2
0 0 0
d2(c#hu /u)
v
The basic reproduction number is therefore

R "maxMJR , JR N
0 1 2
with
a b hq/du
R" i i .
i ud(c#hu /u) (c#hu /u)
h v
This formula is a generalization of the Ross-Macdonald basic reproductive
number to the case of multiple strains, frequency-dependent contact rates and
variable population size in both host and vector.
It follows then that if R '1, then the disease is able to invade the host
0
population. Otherwise, if R 61 the virus eventually disappears from the host
0
population (local result).

3 Equilibrium points

We are interested in the conditions that guarantee the permanence of dengue

as an endemic disease. There are, in our model, boundary equilibria (where
only one strain is present), and the coexistence equilibrium. In the following
section we analyze the former. Its existence is determined by the relative
magnitude of the basic reproductive number of each strain. After that, we
present the numerical results that characterize the stability properties of the
coexistence equilibrium.

3.1 Boundary equilibria

Whenever R (1 then B , A both vanish. Therefore, R '1, R (1,

j j j i j
i9j, i, j"1, 2, implies the existence of E* which is also a unique endemic
i
equilibrium (where only one strain is present). Under these conditions the
asymptotic dynamics of system (1—2) is completely contained in an invariant
manifold X LX defined as
i
X "M(S , I , » ): 0(S #I 6h/u, 0(» 6q/dN .
i i i i i i i
532 Z. Feng, J. X. Velasco-Hernández

For » "I "½ "R"0 (within X ) our model is described by the

j j j i
equations
S@(t)"h!B S!uS ,
i
I@ (t)"B S!uI , (3)
i i i
aI
»@ (t)" i i (q/d!» )!d» ,
i c#u N i i
v
with the total host population size (in X ) satisfying the equation
i
N@(t)"h!uN .
In (3) the unique (non-trivial) equilibrium point is the projection of E* onto
i
the subspace X . Moreover, virulence (extra mortality induced by the virus)
i
plays no role in this asymptotic system.
Since N(t)Ph/u as tPR, (3) can be reduced further to the equations
b»
I@ (t) " i i (h/u!I )!uI ,
i c#u h/u i i
h
aI
»@ (t) " i i (q/d!» )!d» ,
i c#u h/u i i
v
that are equivalent to the Ross Macdonald malaria model (Aron and May
[1]). Therefore, whenever R '1, this limiting system is globally asymp-
i
totically stable in X [18].
i
3.1.1 Boundary equilibria when virulence is negligible
To obtain precise results on the existence and stability properties of equilib-
rium points we assume that dengue does not produce significant mortality
(e "0). The assumption is not justifiable in all regions where dengue is
j
endemic but it is a useful first approximation.
In the absence of virulence, the host population is asymptotically constant
and therefore we can reduce the dimension of our model equations by one. We
drop the equation for S. Let u "u "u, N] "h/u and ¹] "q/d and take
h v
e "0.
i
To facilitate the analysis and interpretation of our results we compute the
Jacobian matrix of (1—2) by reordering it as indicated in (4) below.
Consider the boundary equilibrium
E*"(»*, I*, »*, I*, ½*, ½*, R*) (4)
1 1 1 2 2 1 2
where
»*"I*"½*"½*"R*"0
2 2 1 2
and
ud(R !1) ud(R !1)
»*" 1 , I*" 1 ,
1 b (d#a Nª ) 1 a (b ¹ª #u )
1 1 1 1
with
a b
a" i , b" i .
i c#uNK i c#uNK
Competitive exclusion in dengue 533

With this new order the Jacobian has the form

A B
G G
J(E*)" 1 2 ,
1 0 G
4
where

A B
!d!a I* a (¹] !»*)
G " 1 1 1 1 ,
1 b (N] !I*) !u!b »*
1 1 1 1

A B
!d a (¹] !»*) 0 a (¹] !»*) 0
2 1 2 1
b (N] !I*) !u!p b »* 0 0 0
2 1 1 1 1
G " 0 p b »* !(u#r) 0 0 .
4 1 1 1
p b I* 0 0 !(u#r) 0
2 2 1
0 0 r r !u

The eigenvalues of J(E*) are given by the eigenvalues of G and G . When

1 1 4
R '1, G has two eigenvalues with negative real parts. Eigenvalues of G are
1 1 4
given by !u, !(u#r) and by the roots of the polynomial
j3#Aj2#Bj#C"0
where
A"2u#p b »*#r#d ,
1 1 1
R d»*
B"(u#r) (u#p b »*)#d(u#p b »*)!ud 2!p a b 1,
1 1 1 1 1 1 R 2 2 2 a
1 1
R d»*
C"d(u#p b *) (u#r)!(u#r) ud 2!(u#p b »*) p a b 1.
1 11 R 1 1 1 2 2 2 a
1 1
In the above we have used the equivalencies
N] ¹] u d#a N] d»*
(N] !I*) (¹] !»*)" , " 1 , I* (¹] !»*)" 1 .
1 1 R b »* d(R !1) 1 1 a
1 1 1 1 1
Our threshold parameters are given in terms of bounds for the superinfec-
tion coefficients p and p . Thus, they set bounds for the level of induced
1 2
resistance or increased susceptibility that each strain produces in the host.
Let

G A B H
R d#a N]
p*"max 0, 2!1 1 ,
1 R d(R !1)
1 1

A B
d(u#r) u R
p*" f (p )" 1! 2 . (5)
2 1 a b I* (¹K !» *) u#p b » * R
2 2 1 1 1 1 1 1
Then the following hold
Lemma 1.
1. p 'p* implies p*'0 ;
1 1 2
2. p 'p* and p (p* implies A'0, B'0, C'0 and AB'C ;
1 1 2 2
3. p (p* or p 'p* imply C(0.
1 1 2 2
534 Z. Feng, J. X. Velasco-Hernández

The proofs of 1 and 3 are straightforward. To show 2, let p 'p* and

1 1
p (p*. It is easy to see that p (p* implies C'0. Note that
2 2 2 2
R
d(u#p b »*)'ud 2 .
1 1 1 R
1
Hence

B(u#p b » *)
1 1 1
7(u#d) (u#p b »*) d!(u#p b »*) p!2a b I* (¹] !»*)
1 1 1 1 1 1 2 2 1 1
7C ,

but C'0, therefore B'0. To show that AB'C we note that p ( f (p ),

2 1
thus obtaining

u#r R
p a b I* (¹] !»*)6d(u#r)!ud 2.
2 2 2 1 1 u#p b » * R
1 1 1 1
It follows then that

1 1 1 A
AB7(2u#p b »*#r#d) · (u#r) (u#p b » *#d)#d(u#p b »*)
1 1 1 1 1 1

B
R u#r R
!ud 2!d(u#r)#ud 2
R u#p b »* R
1 1 1 1 1

A B
R
'(u#r) d(u#p b »*)!ud 2 .
1 1 1 R
1
Since
R
C(d(u#p b »*) (u#r)!(u#r) ud 2 ,
1 1 1 R
1
it follows that
AB'C .

Using the Routh-Hurwitz criteria we have the following result:

Corollary ¹he boundary equilibrium E* is locally asymptotically stable if

1
p 'p* , and p (p* . Otherwise it is unstable.
1 1 2 2
Lemma 1 and the corollary say that whenever the superinfection coeffic-
ient of the first strain is above threshold there are values of the second
superinfection coefficient that give asymptotic stability of the boundary equi-
librium E*. Since it is only required that p*'0 and p*'0, the asymptotic
1 1 2
stability of E* is guaranteed in principle either when the first strain induces
1
resistance or when it increases susceptibility to the second strain.
Using the definition of f given in (5) we have
Competitive exclusion in dengue 535

Fig. 1. Graph on the parameter space (p , p ) for case 1 of Lemma 1. In this case R (R ,
1 2 1 2
f (0)"0. The shaded area corresponds to parameters values that render the boundary
equilibrium for strain 1 locally asymptotically stable

Fig. 2. Graph on the parameter space (p , p ) for case 2 of Lemma 1. In this case R (R ,
1 2 1 2
f (0)"0. The shaded area corresponds to parameters values that render the boundary
equilibrium for strain 1 locally asymptotically stable.

Lemma 2.
1. If R (R , then p*"0, f (p*)"f (0) '0, f @(p )'0 .
2 1 1 1 1
2. If R 'R , then p*'0, f (p*)"0, f (0)(0 and f @(p )'0 .
2 1 1 1 1
Define D "limp f (p ). Then the region of stability of E* given by
1 1?= 1 1
Lemma 2 is shown in Figures 1 and 2 for both cases ( f (0) '0 and f (0) (0).
Using the symmetry between the two dengue strains, we can perform
a similar analysis for the other boundary equilibrium E* (where the second
2
strain wins). In this case we have »*'0 and I*'0, and we can define
2 2

G A B H
R d#a N]
p f "max 0, 1!1 2 ,
2 R d(R !1)
2 2

A B
d(u#r) u R
p f "g(p )" 1! 1 . (6)
1 2 a b I* (¹K !»*) u#p b »* R
1 1 2 2 2 2 2 2
536 Z. Feng, J. X. Velasco-Hernández

Fig. 3. Region of parameter space (p , p ) where both boundary equilibria are locally
1 2
asymptotically stable. Fixed parameter values are r"0.71/day, k"0.000039/day,
d"0.71/day, h"0.9775, a "0.002, a "0.015, b "0.001, b "0.001, c"10,
1 2 1 2
¹] "50 000, N] "25 000. The corresponding basic reproduction numbers are R "2.4 and
1
R "2.08
2

Without loss of generality assume R 'R '1. Then we can draw a bifurca-
1 2
tion diagram in parameter space (p , p ). See Fig. 3.
1 2
Putting together the three lemmas and the corollary above, we summarize
our results in the following lemma about the local stability properties of both
boundary equilibria (where we use the definition of g given in (6)):

Lemma 3.
1. E* is locally asymptotically stable if p (f (p ) for every p '0, and
1 2 1 1
unstable if p 'f (p ).
2 1
2. E* is locally asymptotically stable if p 'g~1(p ) for every p '0, and
2 2 1 1
unstable if p (g~1(p ).
2 1
3. E* and E* are locally asymptotically stable if g~1 (p ) (p (f (p ).
1 2 1 2 1

Note that it is possible to have threshold values of p* and p* such that

1 2
both E* and E* are locally asymptotically stable. This conclusion indicates, at
1 2
the very least, that there are situations where the asymptotic dynamics of our
model depends on the initial conditions.

3.1.2 Boundary equilibria and virulence

When e '0, we can no longer claim that the total host population is
j
asymptotically constant and one has to work with the full system (1—2). In this
section we provide a rough sketch of the stability properties of the boundary
equilibria for this case.
Competitive exclusion in dengue 537

The local stability analysis of E* (equilibrium with only strain i present)

i
can be determined by computing DU(E*). The eigenvalues of this matrix
i
around this equilibrium are

j "R~1, and k "R t (p , p , R ), i9j ,

i i i j i j i
where t is a multiplicative perturbation of R that depends on the basic
j
reproductive number of strain i as well as on the superinfection indices p and
1
p . This perturbation has the general form
2
t"t (p , R )#t (p , R ), i9j ,
i i i j j i
where
c c p (R !1)
t" 1 , t" 3 j i ,
i u#c p (R !1) j c (e #r#u)
2 i i 4 j
with c , c , c and c positive constants. In particular t satisfies Lt /Lp (0,
1 2 3 4 i i
and Lt /Lp '0 and Lt /Le (0.
j j j j
Note that the stability of the equilibrium point E* is favored (k tends to
i i
reduce its magnitude) when the virulence (extra host mortality induced by the
virus) of strain j is large.
Moreover, in the limiting case were p "p "0 (no superinfection occur-
i j
ring), t"R~1. Therefore, the condition R (1(R makes U a local contrac-
1 j i
tion around E*. Thus, the local asymptotic stability of E* in X follows.
i 1
However, if 0(p (p , E * may cease to be an attractor under U(with the
i j i
condition R (1(R holding). In this case E* is still a global attractor in
j i i
X and (locally) a saddle point in X.
i
It is important to mention that whenever R '1 and R '1 simulta-
i j
neously, both equilibria E* and E* exists. Their stability properties depend on
i j
the magnitude of k and k respectively (j and j are always less than one
i j i j
whenever R '1 and R '1). For e '0 numerical simulations give essen-
i j i
tially the same result as for the case e "0.
i

3.2 Characterization of the interior endemic equilibrium

In this section we present results from numerical simulations that provide

strong evidence for the existence of an interior endemic equilibrium, that is, an
equilibrium point with positive densities of both infected host types. In these
simulations we explored the interdependence of three key parameters: the
basic reproductive number, the superinfection coefficients and the disease-
induced death rate for each strain.
Variable population size can have a very dramatic effect on the result of
a competitive interaction [17]. In particular it can ‘reduce’ the area of
parameter space on which coexistence is possible (compared with the equiva-
lent model with constant total population size). The following results were
found through the numerical simulation of model (1—2) when e '0:
i
538 Z. Feng, J. X. Velasco-Hernández

Fig. 4. Phase plot in the space (I , I ) for values of the superinfection indices outside the
1 2
shaded area shown in Fig. 3. The graph was computed with the same parameter values
shown in Fig. 3 but with p "5, p "0.05, and positive disease-induced death rates
1 2
e "0.0001/day, e "0.0005/day. These parameter values give p (g~1(p )"0.1. In this
1 2 2 1
case strain 1 competitively excludes strain 2. The final outcome of the disease (which strains
wins) is independent of initial conditions. The black square indicates the boundary equilib-
rium point. The unit of measurement of I and I is number of cases
1 2

1. Whenever R '1 for i"1, 2 there exists an equilibrium point in the

i
interior of X. This point has a local unstable and a local stable manifold
of positive dimension (see Fig. 5).
2. If R '1'R then the boundary equilibrium E* and the interior en-
i j j
demic equilibrium do not exist and the boundary equilibrium E* is
i
globally asymptotically stable (Fig. 4).
3. When R 'R '1, the superinfection coefficients p and p may change
i j 1 2
the asymptotic behavior of the system, rendering strain j as the winner
over strain i (which would be the winner if p "p "1, see Fig. 6).
1 2
4. When R 'R '1 and both boundary equilibria are locally
i j
asymptotically stable, there exists a separatrix that cuts X into
two disjoint basins of attraction (one for each boundary equilibrium, see
Fig. 5).
Model simulations show that in the host population there is no long-term
persistence of both strains. However, the unusual nature of the endemic
equilibrium (a ‘saddle’ point) produces a relatively prolonged (years of dura-
tion) quasi-steady state when both R are greater than one. Given the inherent
i
Competitive exclusion in dengue 539

time-scale of the disease (months), this quasi-steady state would look as an

stable endemic equilibrium (see Figs. 5 and 6). Under these conditions
there are two possibilities depending on how many of the boundary equi-
libria are locally stable. If only one of them is locally asymptotically stable
our computer simulations indicate that this equilibrium is also globally
asymptotically stable. Thus the competitive exclusion of one of the strains
occurs.
Our computer simulations also indicate that even though the initial
outbreak of primary infection is driven by the strain with the highest repro-
ductive number, it is precisely this strain the one that can be competitively
excluded. This occurs if the primary infection enhances (increases susceptibil-
ity to) secondary infections (see Fig. 6). Therefore, the strain with the smallest
reproductive number may end up persisting in the host.
The second possibility occurs when both boundary equilibria exist
and both are locally asymptotically stable. In this case the outcome of the
interaction — competitive exclusion of one strain —, depends on the initial
conditions (Fig. 5).

Fig. 5. Phase plot in the space (I , I ) for values of the superinfection indices outside
1 2
the shaded area shown in Fig. 3. The graph was computed with the same parameter values
shown in Fig. 3 but with p "1, p "4.2, and positive disease-induced death
1 2
rates e "0.0001/day, e "0.0005/day. The presence of a saddle point in the interior of
1 2
the region and the existence of a separatrix may be conjectured. Note that the final
outcome of the disease (which strains wins) depends on initial conditions. The circle
indicates the endemic equilibrium point. The unit of measurement of I and I is number of
1 2
cases
540 Z. Feng, J. X. Velasco-Hernández

Fig. 6. Time plot of model (1—2) for a period of 5 years. The graph shows the total numbers
of infected individuals for each strain I #½ and I #½ . Parameter values are the same
1 1 2 2
as for Fig. 3 except for the following: a "0.005, a "0.005, b "0.005, b "0.007, p "4,
1 2 1 2 1
p "1.2. There are two curves, one for each strain. For about 3 years both strains seem to
2
increase and coexist. Only in the fourth year strain 1 clearly wins over strain 2. Note that
strain 2 increases faster at the beginning of the epidemic but it is this strain the one that goes
extinct

4 Discussion

The incorporation of full vector-host dynamics in a multiple strain epi-

demiological system has been partially analyzed in this work. Conditions for
existence and stability properties of the interior endemic equilibrium point are
somewhat unusual. Although existence of the endemic equilibrium is still
a function of the basic reproduction numbers of each strain (both basic
reproduction numbers must be greater than one), the endemic equilibrium is
always unstable with stable and unstable manifolds of non-zero dimension.
To illustrate this point numerical simulations were carried out based on the
numbers reported in [15] for the basic reproductive number of the 1990—91
epidemic in Brazil. R had an average value of 2.03. Previous work in Mexico
0
reported an average reproduction number of 1.33 with a maximum of 2.41
(cited in [15]). The parameter values that we have chosen give basic reproduc-
tion numbers for both strains of about 2.
Competitive exclusion in dengue 541

The existence of an interior endemic equilibrium point with this character-

ization produces a potentially unpredictable long-term behavior (see Fig. 5).
New infection waves after the primary epidemic burst will settle to a transient,
apparently stable, low level of endemicity where strains cocirculate. However,
as time passes, the prevalence of one of the strains will eventually and steadily
increase while the other strain disappears. Which of the strains will be the
winner depends on the initial conditions preexisting when the new strain
arrived and on the level of susceptibility induced by the primary infection
(Figs. 4 and 5). Of course, if during the transient phase, a new wave of infection
appears, it will not be possible to predict which strain will become endemic. In
general, as Fig. 3 shows, the region where both boundary equilibria are locally
asymptotically stable is large and, according to our model, most dengue
epidemics fall into this situation.
As shown in Fig. 3, the existence of two locally asymptotically stable
boundary equilibria, and therefore the existence of an (unstable) interior
endemic equilibrium is guaranteed for a biologically feasible range of values of
the superinfection indices p and p . This range covers cases when both
1 2
primary infections induce resistance to secondary infections or when suscepti-
bility is enhanced.
It appears that for feasible values of p (e.g. p 3(0.1, 2) ) and whenever
i i
R 'R , the interior equilibrium exists unless p is either very small or very
i j j
large. In Fig. 3 we illustrate the case when R 'R (see Fig. 3 legend for
1 2
parameter values used). We conclude that the presence of superinfection forces
the existence of the interior equilibrium. However, it appears that superinfec-
tion cannot induce stability in this state. In the case of Fig. 3, the endemic
interior equilibrium would cease to exist for a value of p +0.001 but then
2
only the boundary equilibrium E* would be asymptotically stable.
1
Gupta, Swinton and Anderson [9] show in a model for malaria that
coexistence is a likely outcome when cross-immunity is taken into account.
Although malaria is a parasitic, not a viral disease, the mathematical structure
of the model allows some comparisons with ours since both deal with a vector
transmitted disease. Gupta et al. generalize directly the Ross-Macdonald
model for malaria studied by Aron and May [1] introducing cross-immunity
and two infected host subtypes: those that are infected and infectious, and
those that are infected but uninfectious. Thus, essentially there is a reduction
in the net number of infected individuals that can transmit the disease.
However all infected individuals can hold the parasite. In particular, the rate
at which parasites become ineffective to transmission, i.e., the hosts become
infected but not infectious, is exponential, guaranteeing the presence of posi-
tive densities (however small) of each type of infected hosts for all time. Thus,
the Gupta et al. model effectively creates a refuge for each parasite strain.
Moreover, the total host population is considered constant. The assumption
of constant host population size is achieved by defining the recruitment rate in
such a way as to balance the output from all system compartments. This factor
alone when associated with cross-immunity is enough to enhance coexistence
in models for directly transmitted diseases [21, 16]. In the case of our general
542 Z. Feng, J. X. Velasco-Hernández

model all infected individuals are infectious; thus there are no refuges. Also, by
definition, we take virulence as extra mortality induced by the disease. This
prevents the existence of a constant population size for the host. We do not
define the recruitment rate for the total population so as to balance disease
mortality losses (and therefore achieve a constant size in the host population).
This would be equivalent, in our case, to require that the extra-mortality rate
is compensated exactly by the cure rate of the disease and, therefore, popula-
tion variability would be independent of disease dynamics. However, even in
this case (no virulence) our model predicts competitive exclusion of one of the
strains.
The main reason that explains why in our model coexistence is an improb-
able outcome resides, we believe, in the coupling of two populations, each with
a different pattern of disease progression.
The structure of the equations that describe the transmission dynamics in
the host population is that of an SIR model with superinfection and variable
population size. In a directly transmitted disease with this structure and no
virulence one would expect analogous results to those of Nowak and May
[21]: coexistence of both strains as a rule. Our model also incorporates an SI
model without superinfection in the vector population. In a directly transmit-
ted disease this structure would predict competitive exclusion of the strain
with lower basic reproductive number [2].
When we couple both of these types of epidemics into one, our host-vector
model (1—2), the outcome is competitive exclusion of one of the strains if at
least one of the basic reproductive numbers is greater than one. In a sense, the
vector dynamics dominates the dynamics of the coupled system. The reason
for this is that the vector-host relationship is asymmetric. The vector chooses
the host. In this case we have modeled the contact rates according to
a generalization of the Ross-Macdonald model: the contact rate is frequency
dependent [6] (depends on the ratios of vector numbers to host numbers for
both types of strains). Thus, what our results show is that coexistence pro-
moted by superinfection in the host population is ‘broken’ by frequency dependent
dynamics in the biting (contact) rates, thus resulting in the competitive exclusion
of one strain even when an interior steady-state exists.
Other models that incorporate cross-immunity and multiple strains have
been studied [12, 14, 17]. We compare our results with the original one that
introduced this idea of competition of multiple strains in epidemic models,
namely the Levin and Pimentel paper [12]. In summary, the conclusions of
[12] are that in a variable host population system coexistence is possible in
a bounded region of parameter space. Outside this region, depending on the
relative magnitudes of parameter values one of the two strains wins and
competitively excludes the other. This model was originally designed for the
theoretical study of myxomatosis as a control factor of an exponentially
growing population. The fact that virulence is the growth regulatory factor
in this model determines the existence of a coexistence region in parameter
space. In the dengue model that we analyze here, the disease is not the
unique factor that regulates growth. Permanent immunity is also explicitly
Competitive exclusion in dengue 543

introduced into the model. Even in the case when virulence is negligible
competitive exclusion is the rule. The existence of frequency-dependent con-
tact rates closes the window of coexistence.
The model analyzed here does not incorporate the effects of age structure.
Dengue in tropical Asia affects particularly children with ages between 5 and
15 years old, with a modal age of 5 years [26]. The same reference indicates
that in 1987 more than 600 000 cases of dengue were reported in Southeast
Asia with 24 000 deaths: 90% of both cases and deaths were children. The risk
of infection is obviously an age dependent factor. Moreover, the influence of
physiological structure into the dynamics of dengue may have an influence in
the likelihood of coexistence of both strains. The need for a model that
incorporates age structure into the dengue population dynamics is thus
justified.

Acknowledgements. We thank Carlos Castillo-Chavez for his continuous support and

encouragement. This research was initiated while JXVH was a visiting fellow of the Royal
Society of London/Academia de la Investigación Cientı́fica, at the Zoology Department of
the University of Oxford. JXVH acknowledges the support of CONACYT grant 400200-5-
3551E. This research has been partially supported by grant DEB-9253570 (Presidential
Faculty Award) to Carlos Castillo-Chavez. JXVH dedicates this work to the memory of
Diego Bricio Hernández Castan8 os.

References

[1] Aron, J. and May, R. M., The population dynamics of Malaria, pages 139—179.
Population Dynamics of Infectious Diseases. Chapman and Hall, London, 1982
[2] Bremermann, H. J. and Thieme, H. R., A competitive exclusion principle for pathogen
virulence. Journal of Mathematical Biology, 27: 179—190, 1989
[3] Capasso, V. Mathematical Structures of Epidemic Systems. Springer Verlag, New
York, 1993. Lecture Notes in Biomathematics 97
[4] Castillo-Chavez, C., Hethcote, V. A. H. W., Levin, S. A. and Liu, W.-M., Epidemiologi-
cal models with age structure, proportionate mixing, and cross-immunity. Journal of
Mathematical Biology, 27: 233—258, 1989
[5] Castillo-Chavez, C., Huang, W. and Li. J., Competitive exclusion in gonorreha models
and other sexually-transmitted diseases. Technical Report 93-36, Mathematical
Science Institute, Cornell University, 1993
[6] Castillo-Chavez, C., Velasco-Hernandez, J. X. and Fridman, S., Modelling contact
structures in biology. Lecture Notes in Biomathematics, 100: 454—492, 1995
[7] Diekmann, O., Heesterbeek, J. A. P. and Metz, J. A. J., On the definition and the
computation of the basic reproduction ratio r in models for infectious diseases in
0
heterogeneous populations. Journal of Mathematical Biology, 28: 365—382, 1990
[8] Gubler, D. J., Dengue, volume 2 of The Arboviruses: Epidemiology and Ecology,
pages 223—261. CRC Press, Inc., 1986
[9] Gupta, S., Swinton, J., and Anderson, R. M., Theoretical studies of the effects of
heterogeneity in the parasite population on the transmission dynamics of malaria.
Proceedings of the Royal Society of London B, 256: 231—238, 1994
[10] Halstead, S. B., Dengue, Tropical and Geographical Medicine. Mac Graw Hill, New
York, pp. 652—659, 1984
[11] Healstead, S. B., The XXth century dengue pandemic: need for surveillance and
research. World Health Statistics Quarterly, 45: 292—298, 1992
[12] Levin, S. and Pimentel, D., Selection of intermediate rates of increase in parasite-host
systems. American Naturalist, 117(3): 308—315, 1981
544 Z. Feng, J. X. Velasco-Hernández

[13] Li, J., Castillo-Chavez, C. and Huang, W., Dynamics of multiple pathogen strains in
heterosexual epidemiological models pages 289—298. Differential Equations and Ap-
plications to Biology and to Industry. Singapore, New Jersey, London, Hong Kong,
1996
[14] Lipsitch, M. and Nowak, M. A., The evolution of virulence in sexually transmitted
HIV/AIDS. Journal of Theoretical Biology, 174: 427—440, 1995
[15] Marques, C. A., Foratini, O. P. and Masad, E., The basic reproduction number for
dengue fever n sao paulo state, brazil 1990—1991 epidemic. Transactions of the Royal
Society of Tropical Medicine and Hygene, 88: 58—59, 1994
[16] May, R. M. and Nowak, M. N., Superinfection, metapopulation dynamics, and the
evolution of virulence. Journal of Theoretical Biology, 170: 95—114, 1994
[17] Mena-Lorca, J., Velasco-Hernandez, J. X. and Castillo-Chavez, C., Superinfection,
virulence and density dependent mortaility in an epidemic model. Technical Report
BU 1299-M, Biometrics Unit, Cornell University, 1995
[18] Moreira, H. N., Lienars-type equations and the epidemiology of malaria. Ecological
Modelling, 60: 139—150, 1992
[19] CNN World News. Deadly fever sweeping Central, South America. Lucia Newman,
Correspondant September 28, CNN Food and Health Page, http://www.cn.com, 1995
[20] Newton, E. A. C. and Reiter, P., A model of the transmission of Dengue fever with an
evaluation of the impact of ultra-low volume (ULV) insecticide applications on
Dengue epidemics. American Journal of Tropical Medicine and Hygene, 47: 709—720,
1992
[21] Nowak, M. N., and May, R. M., Superinfection and the evolution of parasite viru-
lence. Proceedings of the Royal Society of London B, 255: 81—89, 1994
[22] World Helath Organization. Selected health problems. World Health Statistics Quar-
terly, 44: 208—233, 1991
[23] World Health Organization. The urban crisis. World Health Statistics Quarterly, 44:
189—197, 1991
[24] Rai, K. S., Aedes albopictus in the Americas. Annual Review of Entomology, 36:
459—484, 1991
[25] Reeves, W. C., Hardy, J. L., Reisen, W. K. and Milby, M. M., Potential effect of global
warming on mosquito-borne arboviruses. Journal of Medical Entomology, 31:
323—332, 1994
[26] Shepard, D. S., and Halstead, S. B., Dengue (with notes on Yellow fever and Japanese
encephalitis), pages 300—320. Disease Control Priorities in Developing Countires.
Oxford University Press, 1993
[27] Velasco-Hernandez, J. X., A model for Chagas disease involving transmission by
vectors and blood transfusion. Theoretical Population Biology, 46(1): 1—31, 1994