Applied Mathematical Modelling 76 (2019) 238–251

Contents lists available at ScienceDirect

Applied Mathematical Modelling

journal homepage: www.elsevier.com/locate/apm

Stability and periodic solutions for a model of bacterial

resistance to antibiotics caused by mutations and plasmids
Eduardo Ibargüen-Mondragón a,∗, Jhoana P. Romero-Leiton c, Lourdes Esteva b,
Miller Cerón Gómez a, Sandra P. Hidalgo-Bonilla d
a
Grupo de Investigación en Biología Matemática y Matemática Aplicada (GIBIMMA), Facultad de Ciencias Exactas y Naturales,
Departamento de Matemáticas y Estadística, Universidad de Nariño, C. U. Torobajo, Clle 18 – Cra 50, Pasto, PBX 27311449, Colombia
b
Facultad de Ciencias, Departamento de Matemáticas, Universidad Nacional Autónoma de México, 04510, CDMX, México
c
School of Mathematics and Information Technology, Yachay Tech University, Yachay City of Knowledge, Urcuquí 100650, Ecuador
d
School of Chemical Sciences and Engineering, Yachay Tech University, Yachay City of Knowledge, Urcuquí 100650, Ecuador

a r t i c l e i n f o a b s t r a c t

Article history: Bacterial resistance is one of the most prominent public health problems affecting the en-
Received 15 June 2018 tire world population. Although some infectious diseases are no longer a problem as they
Revised 29 May 2019
were in the past, the acquisition of bacterial resistance continues to increase. In particular,
Accepted 11 June 2019
antibiotics have been losing their effectiveness after decades of misuse and overuse, which
Available online 18 June 2019
has generated an emergency situation. In this work, we formulate and analyse a determin-
Keywords: istic model for the population dynamics of susceptible and resistant bacteria to antibiotics,
Bacteria assuming that drug resistance is acquired through mutations and plasmid transmission.
Drug resistance Qualitative analysis reveals the existence of a bacteria-free equilibrium, a resistant bacteria
Plasmids equilibrium, an a coexistence equilibrium and a limit cycle arising from Hopf bifurcation.
Ordinary differential equations The stability of the equilibria are given in terms of the growth rate of bacteria, the ac-
Stability quisition of resistance, as well as the elimination of bacteria due to the immune system
and the action of antibiotics. Numerical simulations corroborate our analytical results, and
illustrate the temporal dynamics of the susceptible and resistant bacteria.
© 2019 Elsevier Inc. All rights reserved.

1. Introduction

The diversity of antibiotics used in the treatment of infections and the acquisition of bacterial resistance to these antibi-
otics are currently increasing. One possible explanation for this phenomenon may be the fact that bacteria contain complex
groups of genes in their structures, such as plasmids, integrons or transposons, that are linked to resistance to a specific type
of antibiotic. The interaction between these groups of genes also causes the emergence of resistance to other antibiotics [1].
In most cases, the transformations that take place inside the bacterium during the process of acquiring resistance to
antibiotics arise due to the action of plasmids, which are circular or linear extrachromosomal DNA molecules that are
replicated and transcribed independently of chromosomal DNA. They are normally present in bacteria and are sometimes
founded in eukaryotic organisms such as yeast. Their size varies from 1 to 250 KB (one kilobase is 10 0 0 pairs of DNA or
RNA bases). The number of plasmids can vary, depending on their type, from a single copy to a few hundred copies per
cell [2]. Bacteria that acquire plasmids are capable of simultaneously becoming resistant to multiple antibiotics [2], and this

∗
Corresponding author.
E-mail address: edbargun@udenar.edu.co (E. Ibargüen-Mondragón).

https://doi.org/10.1016/j.apm.2019.06.017
0307-904X/© 2019 Elsevier Inc. All rights reserved.
 E. Ibargüen-Mondragón, J.P. Romero-Leiton and L. Esteva et al. / Applied Mathematical Modelling 76 (2019) 238–251 239

type of resistance acquisition is called horizontal transmission. Another mechanism by which bacteria acquire resistance to
antibiotics is through mutations in a chromosomal gene which can be natural or acquired [3]; this process is called vertical
transmission from one generation to another. According to these ideas, bacteria acquire resistance through two different
mechanisms: introduction of a plasmid R for resistance and mutation of a chromosomal gene.
The scientific community is tackling the problem described above from different perspectives, and mathematics are par-
ticularly focusing on models that can describe the phenomena that occur in the dynamics between populations of bacteria
and the acquisition of resistance to antibiotics. Analysis of these models allows researchers to describe the interaction dy-
namics of susceptible and resistant bacteria to antibiotics when an antibiotic treatment has been delivered to an infected
patient.
To date, mathematical models have been widely used to simulate the spread of antibiotic resistant-bacteria [4–6], to
identify the factors responsible for the prevalence of bacterial resistance [4,7,8], to examine the behaviour of bacteria be-
fore the use of different antibiotic treatments [9–11], to optimize the use of antibiotics [12], and to control the propagation
of resistance [13–19]. In 2011, Romero et al. [20] formulated and analysed a mathematical model to simulate the propaga-
tion dynamics of the bacterial resistance of Staphylococcus aureus to a single bactericidal antibiotic. In 2014, Ibarguen et al.
[3] generalized the previous model by considering simultaneous resistance to multiple antibiotics with bactericidal action.
In 2014, the previously constructed model was examined once more to consider the dynamics of propagation of bacterial re-
sistance to antibiotics with simultaneous bactericidal and bacteriostatic action [21], and in 2016, they analysed the incidence
of plasmids as a parameter in the transmission dynamics of bacterial resistance [22].
In this work, a mathematical model for population dynamics of susceptible bacteria, resistant bacteria, plasmids and
antibiotics is proposed and analysed. The results of the qualitative analysis show the existence of a trivial equilibrium P0
without bacteria and an equilibrium P1 with only susceptible bacteria. When P1 loses stability, an equilibrium solution of
coexistence P2 appears. In addition, the existence of a Hopf bifurcation is shown for certain values of parameters, which
implies the existence of periodic solutions around P2 , suggesting that control strategies should be implemented to eliminate
the total population of bacteria or at least control the spread of resistance. Numerical simulations are performed using
infection data for Staphylococcus aureus found in [20].

2. Mathematical model

In this section, we formulate a model of bacterial resistance that describes the interaction of susceptible bacteria, re-
sistant bacteria and plasmid populations. Let us denote by S(t) and R(t) the population sizes of susceptible and resistant
bacteria to antibiotics at time t, respectively, P(t) the number of plasmids at time t and C(t) the concentration of dissolved
antibiotics in the medium at time t. As in [3], we assume that bacteria follow logistical growth with carrying capacity K. Let
β s and β r represent the reproduction rates of susceptible and resistant bacteria, respectively. Specific mutations that confer
resistance to chemical control often have an inherent fitness cost that may be manifested through reduced reproductive
capacity or competitive ability [21]. In this work, we quantify the fitness cost as a reduction in the reproduction rate of
the resistant strain, which is β r ≤ β s . During administration of the antibiotic, several resistant bacteria can emerge due to
mutations of bacteria exposed to antibiotics, and we model this situation with the term q̄CS where q̄ is the mutation rate of
susceptible bacteria due to exposure to antibiotic treatment. Susceptible and resistant bacteria have per capita natural death
rates given by μs and μr , respectively. Susceptible bacteria also die due to the action of the antibiotics, and we assume that
the rate at which they are eliminated by the antibiotic is equal to ᾱS SC. Furthermore, this approach assumes that during
treatment with antibiotics, the process of bacterial conjugation for the transfer of resistant plasmids is carried out. In this
process susceptible bacteria are receivers and resistant bacteria are donors of genetic material which is represented by the
term δ̄ SR, where δ̄ is the rate of transfer of resistant plasmids among bacteria. Susceptible and resistant bacteria are elimi-
nated by the host immune system at the per capita rate γ . Plasmids reproduce proportionally to the population of resistant
bacteria at a constant rate σ p and die at a constant rate μp . Finally, to model the concentration of bactericidal antibiotics,
we assume the diffusion layers model with the saturation solubility rate  and a degradation rate of μc C [23].
Under the assumptions mentioned above, we obtain the following system of ordinary differential equations:

dS
 S+R

= βs S 1 − − (q̄ + ᾱS )CS − δ̄ P S − γ S − μs S
dt K
dR
 S+R

= βr R 1 − + q̄CS + δ̄ P S − γ R − μr R
dt K
dP
= σpR − μpP
dt
dC
=  − μcC. (1)
dt
Fig. 1 shows the flow diagram of the system (1). With the following change of variables

S R P C
s= , r= , p= ,c= , (2)
K K Kσp /μc
240 E. Ibargüen-Mondragón, J.P. Romero-Leiton and L. Esteva et al. / Applied Mathematical Modelling 76 (2019) 238–251

Fig. 1. The flow diagram of the system (1).

the system (1) is written as

ds
= βs s[1 − (s + r )] − (q + αs )cs − δ ps − (γ + μs )s
dt
dr
= βr r[1 − (s + r )] + qcs + δ ps − (γ + μr )r
dt
dp
= r − μp p
dt
dc
= μc − μc c, (3)
dt
 
where q = q̄ , α = ᾱS and δ = δ̄σ p K. The region of biological interest of system (3) is given by
μc s μc
 
1
 = (s, r, p, c ) ∈ R4+ : 0 ≤ s, r, c ≤ 1, 0 ≤ s + r ≤ 1, 0 ≤ p ≤ . (4)
μp
The following lemma assures that system (3) is well posed in the sense that solutions with initial conditions in  remain
there for all t ≥ 0.

Lemma 2.1. The set  defined in (4) is positively invariant with respect system (3).

See [22] for proof of above lemma.

3. Equilibrium solutions

The equilibria of the system (3) are given by the solutions of the system of algebraic equations
βs s[1 − (s + r )] − (q + αs )cs − δ ps − (γ + μs )s = 0
βr r[1 − (s + r )] + qcs + δ ps − (γ + μr )r = 0
r − μp p = 0
μc ( 1 − c ) = 0 . (5)
From the last two equations of system (5) we have c = 1 and
r
p= . (6)
μp
By replacing c and p in the first two equations of (5), we obtain
δ
βs s[1 − (s + r )] − (q + αs )s − rs − (γ + μs )s = 0
μp
δ
βr r[1 − (s + r )] + qs + rs − (γ + μr )r = 0. (7)
μp
For s = 0, system (7) is reduced to
[βr (1 − r ) − (γ + μr )]r = 0, (8)
 E. Ibargüen-Mondragón, J.P. Romero-Leiton and L. Esteva et al. / Applied Mathematical Modelling 76 (2019) 238–251 241

Rr −1
the solutions of (8) are r = 0 and r = Rr where

βr
Rr = . (9)
γ + μr
By substituting the values of r in (6) we obtain the following equilibrium solutions

P0 = (0, 0, 0, 1 )
 
r1
P1 = 0, r1 , ,1 , (10)
μp
where
Rr − 1
r1 = . (11)
Rr
From (11), it follows that a necessary and sufficient condition for the biological sense of P1 is Rr > 1. Now, for s = 0 the first
equation of (7) is reduced to
δ
βs [1 − (s + r )] − (q + αs ) − r − ( γ + μs ) = 0 ,
μp
or equivalently
 
Rs − 1 δ
= s+ 1+ r, (12)
Rs βs μ p
where
βs
Rs = . (13)
q + αs + γ + μs
From (12), it is concluded that a necessary condition for the existence of susceptible and resistant bacteria is Rs > 1. By
solving for s in (12) we obtain
 
Rs − 1 δ
s= − 1+ r. (14)
Rs βs μ p
Further, s defined in (14) is positive if and only if r < rmax , where
Rs − 1
rmax =  . (15)
Rs 1 + βsδμ p

From the second equation of (7) we obtain

  
2 δ q
−r + r1 r + + − 1 r s = 0. (16)
βr μ p βr
By substituting s defined in (14) in Eq. (16), we obtain the following quadratic equation p(r ) = 0 where

p(r ) = −h2 r 2 + h1 r + h0 , (17)

with h2 , h1 and h0 constants defined by
  
δ δ
h2 = 1 + −1 1+
μ p βr μ p βs
 
δ δ
= β − βr +
μ p βs βr s μp
   
δ Rs − 1 q δ
h1 = r 1 + −1 − 1+
μ p βr Rs βr μ p βs
q Rs − 1
h0 = . (18)
βr Rs
Since h2 and h0 are positive, Eq. (17) has a unique positive solution given by

h1 + h21 + 4h2 h0
r2 = . (19)
2 h2
242 E. Ibargüen-Mondragón, J.P. Romero-Leiton and L. Esteva et al. / Applied Mathematical Modelling 76 (2019) 238–251

Now, Rr < 1/(1 − rmax ) is equivalent to r1 < rmax which implies that p(rmax ) < 0 (see Appendix A). Since p(0 ) = h0 > 0 and
p(rmax ) < 0 then r2 ∈ (0, rmax ). Finally, the fact that

Rs − 1 δ
s+r = − r
Rs μ p βs
1 δ
= 1− − r
Rs μ p βs
≤1

implies that s and r satisfy the conditions of . The above results are summarized in the following proposition.

Proposition 3.1. System (3) always has a trivial equilibrium P0 = (0, 0, 0, 1 ). If Rr > 1, in addition to P0 there exists equilibrium
r
P1 = (0, r1 , μ1p , 1 ).
1
If Rs > 1 and Rr < , in addition to P0 there exists an equilibrium in which susceptible and resistant bacteria co-exist
1 − rmax
P2 = (s2 , r2 , p2 , 1 ).

3.1. Biological interpretation of the parameters

Following the classical definition of the basic reproductive number, the quantity

βr
Nr =
μr
represents the product of the reproduction rate of resistant bacteria β r and the average life span of resistant bacteria 1/μr .
The above parameter is interpreted as the number of bacteria produced by a resistant bacterium during its average life.
Similarly,

βs
Ns =
μs
is interpreted as the number of bacteria produced by a susceptible bacterium during its average life. On the other hand, Rs
defined in (13) is rewritten as
μs
Rs = Ns , (20)
q + αs + γ + μs
where αs = αS /μc is the rate at which the antibiotic eliminates bacteria at their equilibrium level. Since
μs
(21)
q + αs + γ + μs
defines the fraction of susceptible bacteria that have not mutated spontaneously and escape to the action of the antibiotic
and the immune response, then Rs represents the number of bacteria produced by the fraction of susceptible bacteria that
do not present spontaneous mutations and that are not eliminated by the antibiotic or the immune system. Analogously, Rr
defined in (9) is rewritten as
μr
Rr = N,
γ + μr r
which defines the number of bacteria produced by the fraction of resistant bacteria that evade the immune response.
The results of Proposition 3.1 establish that: (a) if the average amount of bacteria produced by the fraction of resistant
bacteria that evade the immune response is greater than one (Rr > 1), then the population of resistant bacteria will persist,
(b) if the average amount of bacteria produced by the fraction of susceptible bacteria that do not have spontaneous muta-
tions and that evade both the effect of the antibiotic and the immune response is greater than one (Rs > 1) and the average
amount of bacteria produced by the fraction of resistant bacteria that evade the immune response is bounded by a thresh-
old (Rr < 1−r1max ), then both susceptible and resistant bacteria will persist. An important result is that if the population of
bacteria that evade all the mechanisms of elimination are not able to reproduce, then the population of bacteria could be
controlled or eliminated.

4. Stability of equilibria P0 and P1

In this section, we determine the local asymptotic stability of the equilibrium solutions of the system (3). To this end,
let us start with the trivial equilibrium P0 = (0, 0, 0, 1 ). Linearization of system (3) around P0 is given by x = J (P )x, where
x = (s, r, p, c )T and the matrix J evaluated at P is
 E. Ibargüen-Mondragón, J.P. Romero-Leiton and L. Esteva et al. / Applied Mathematical Modelling 76 (2019) 238–251 243

⎡ ⎤
j11 (P ) −βs s −δ s −(q + αs )s
⎢−βr r + qc + δ p j22 (P ) δs qs ⎥
J (P ) = ⎣ ⎦, (22)
0 1 −μ p 0
0 0 0 −μc
with
j11 (P ) = βs [1 − (2s + r )] − (q + αs )c − δ p − (γ + μs )
j22 (P ) = βr [1 − (2r + s )] − (γ + μr ). (23)
By evaluating the Jacobian J in P0 we obtain
⎡ ⎤
j11 (P0 ) 0 0 0
⎢ q j22 (P0 ) 0 0 ⎥
J (P0 ) = ⎣
0 ⎦
.
0 1 −μ p
0 0 0 −μc
The eigenvalues of J(P0 ) are given by
ξ1 = j11 (P0 ) = (q + αs + γ + μs )(Rs − 1 )
ξ2 = j22 (P0 ) = (γ + μr )(Rr − 1 )
ξ3 = −μ p
ξ4 = −μc .
Since ξ 1 and ξ 2 are negative for Rs < 1 and Rr < 1 respectively, then P0 is locally and asymptotically stable. This result is
summarized in the following proposition.

Proposition 4.1. If Rs < 1 and Rr < 1, then the trivial equilibrium P0 is locally and asymptotically stable in . If Rs > 1 or Rr > 1,
then P0 is unstable.

Now, we will determine the conditions for which the equilibrium P1 is locally and asymptotically stable. To this end, let
us observe that the Jacobian given in (22) evaluated in P1 is given by
⎡ ⎤
j11 (P1 )  0 0 0
⎢ Rr − 1 δ ⎥
⎢ − βr + q j22 (P1 ) 0 0 ⎥
J (P1 ) = ⎢ Rr μp ⎥.
⎣ 0 1 −μ p 0
⎦
0 0 0 −μ p
By following a procedure similar to the previous case, the eigenvalues of J(P1 ) are k1 = −μc , k2 = −μ p and
  
δ 1
k3 = βs 1 + − (1 − rmax )
μ p βs Rr
k4 = (γ + μr )(1 − Rr ).
We see that k4 < 0 if and only if Rr > 1 and that k3 < 0 if and only if Rr > 1/(1 − rmax ). From above we have the following
proposition.
1 1
Proposition 4.2. If Rr > , then the equilibrium P1 is locally and asymptotically stable in . If Rr < , then P1 is
1 − rmax 1 − rmax
unstable.

5. Occurrence of a Hopf bifurcation

In this section, we will prove existence of a Hopf bifurcation for a suitable parameter values of system (3). In this case,
the equilibrium P2 changes its stability, and a limit cycle appears with an amplitude and frequency that depends on the
value of the parameter. To this end we will use the following theorem

Theorem 5.1. Let the n-dimensional autonomous system of the differential equation be given by
x˙ = F (x, μ ), (24)
which depends on the real parameter μ and where F(x, u) is twice differentiable in both variables. We suppose that

1. System (24) possesses an analytic family x(μ) of equilibrium points; that is, F (x(μ ), μ ) = 0.
2. For a certain value of μ, say μ0 , the Jacobian matrix J (x(μ0 ), μ0 ) = Fx (x(μ0 ), μ0 ) has two purely imaginary eigenvalues
λ± (μ0 ) = ±iβ , and no other eigenvalue of J(x(μ0 ), μ0 ) is an integral multiple of iβ .
3. If λ(μ ) = α (μ ) + iβ (μ ) is the continuation of eigenvalue iβ , then α  (μ0 ) = d (Re(λ )(μ0 ) )/dμ = 0.
244 E. Ibargüen-Mondragón, J.P. Romero-Leiton and L. Esteva et al. / Applied Mathematical Modelling 76 (2019) 238–251

Under the above conditions, there exist differentiable functions μ( ) and T( ) that depend on a parameter  with μ0 and
T (0 ) = 2π β −1 such that there are no constant periodic solutions x(t,  ) of (24) with period T( ) which collapse into x(μ) as
 → 0.
See [24] for a proof of Theorem 5.1. On the other hand, we observe that system (3) is rewritten as system (24), where
x = (s, r, p, c )T and μ is one of the following parameters: β s , β r , q, α s , δ , γ , μs , μr , μp or μc ; additionally, F(x, μ) is the
vector field defined by the right side of (3). In Proposition 3.1, we proved the existence of the coexistence equilibrium P2 ,
which depends on μ and is denoted by x(μ), such that the first literal of Theorem 5.1 is satisfied. Now, from the two first
equations of (5) we obtain
δr
j11 (x ) = βs [1 − (s + r )] − (q + αs )c − − (γ + μs ) − βs s
μp
= −βs s
j22 (x ) = βr [1 − (s + r )] − (γ + μr ) − βr r
s
= − (qc + δ p) − βr r. (25)
r
Substituting (25) in (22) the Jacobian is rewritten as
⎡ ⎤
−βs s −βs s −δ s −(q + αs )s
⎢−β r + qc + δ p j22 (x ) δs ⎥
⎢ r qs ⎥
J (x ) = ⎢ ⎥. (26)
⎣ 0 1 −μ p 0 ⎦
0 0 0 −μc
As a result of evaluating the Jacobian defined in (26) in x(μ) we obtain
⎡ ⎤
−βs s2 −βs s2 −δ s2 −(q + αs )s2
⎢−β r + q + δ p j22 (x(μ )) δ s2 ⎥
⎢ r2 2 qs2 ⎥
J (x(μ )) = ⎢ ⎥. (27)
⎣ 0 1 −μ p 0 ⎦
0 0 0 −μc
The eigenvalues of J(x(μ)) defined in (27) are given by the roots of the following algebraic equation
 
λ3 + a1 λ2 + a2 λ + a3 (λ + μc ) = 0, (28)
where

a1 = βs s2 + μ p − j22 (x(μ )),

a2 = −μ p j22 (x(μ )) − δ s2 + βs s2 μ p − βs s2 j22 (x(μ )) + βs s2 (−βr r2 + q + δ p2 ),
a3 = −βs s2 [δ s2 + j22 (x(μ ))μ p ] + (δ s2 + βs s2 μ p )(−βr r2 + q + δ p2 ). (29)
Let
s2
A = βs s2 + βr r2 + q . (30)
r2
In terms of parameter A, the constants a1 , a2 and a3 are rewritten as
δ
a1 = A + μ p + s
μp 2
s 
a2 = μ p A + βs s2 (q + δ p2 ) 2 + 1
r2
s   s  
a3 = βs s2 μ p (q + δ p2 ) + 1 − δ s2 A + δ s2 q + 1 + δ p2 .
2 2
r2 r2
Now, we will prove that x(μ) is a non-hyperbolic equilibrium that has two negative real eigenvalues and a pair of eigen-
values on the imaginary axis. To this end, we observe that the parameter A defined in (30) is the only positive root of the
equation a1 a2 = a3 (see Appendix B). By substituting the previous equation in (28), we obtain

(λ3 + a1 λ2 + a2 λ + a1 a2 )(λ + μc ) = 0
(λ + a1 )(λ2 + a2 )(λ + μc ) = 0. (31)
√
From (31), we establish that the other eigenvalues of J(P2 ) are −μc , −a1 and λ± (μ )
= ± −a2 . Since a1 > 0 and a2 > 0,
√
then −a1 is a negative real number and λ± (μ ) = ±i a2 are imaginary numbers, which implies that the second part of
Theorem 5.1 is satisfied.
 E. Ibargüen-Mondragón, J.P. Romero-Leiton and L. Esteva et al. / Applied Mathematical Modelling 76 (2019) 238–251 245

In order to determine the bifurcation parameter, we solve s2 and p2 in Eqs. (6) and (14), respectively, in terms of r2 ,
obtaining the following expressions
 
Rs − 1 δ
s2 = − 1+ r
Rs βs μ p 2
r2
p2 = . (32)
μp
On the other hand, after some algebraic manipulations the constants a1 , a2 and a3 are rewritten as
s2
a1 = βs s2 + μ p + (q + δ p2 ) + βr r2
r2
s ( s )2
a2 = μ p 2 q + μ p βr r2 + βs s2 μ p + βs 2 (q + δ p2 ) + βs s2 (q + δ p2 )
r2 r2
    
s2 qμ p Rs − 1 δ βr
a3 = βs δ + 1+ − ( r2 )2 , (33)
r2 δ Rs μ p βs βs
By substituting (32) into (33), we obtain a1 , a2 and a3 in terms of r2 as follows
      
δ Rs − 1 1 δ Rs − 1 δ
a1 = βr − βs 1 + 2
r2 + q + βs 1 + + μp − q 1 +
βs μ p Rs r2 βs μ p Rs βs μ p
 
δ δ δ
a2 = βs 1+ r2
μ p μ p βs βs μ p 2
       
δ δ δ δ Rs − 1
+ μ p βr − βs 1 + 2
− βs q 1 + + r2
βs μ p βs μ p μ p μ p βs Rs
      
Rs − 1 δ δ δ δ Rs − 1
+μ p βs −q 1+ + βs 1+ −q 1+
Rs βs μ p βs μ p μ p βs μ p Rs
 
Rs − 1 Rs − 1 1
+ μp + q
Rs Rs r2
    
δ δ βr 2 δ βr Rs − 1
a3 = − 1 + 1+ − r + 1+ − r2
βs μ p βs μ p βs 2 βs μ p βs Rs
    1
δ qμ R − 1 p s qμ R − 1 p s 2
− 1+ + . (34)
βs μ p δ Rs δ Rs r2
From (34) and (19), we notice that the calculation of the crossing speed is cumbersome regardless of which bifurcation
parameter is chosen. For this reason, we consider the infection rate of susceptible bacteria, β s , as a bifurcation parameter,
that is, μ = βs . Let μ0 = βs0 and (z, μ ) = a2 (z, μ )a1 (z, μ ) − a3 (z, μ ) where z = (βr , q, αs , δ, γ , μs , μr , μ p , μc ), then from
Appendix B, we verify that (z0 , μ0 ) = 0, where z0 is the parameter vector corresponding to the parameter μ0 . In addition,
from (34) we verified that ∂ (z0 , μ0 )/∂μ = 0. As a consequence, from the implicit function theorem, there is an open ball
U ∈ R9 containing z0 and an interval V ⊂ R containing μ0 such that there is a unique function μ = g(z ) defined for z ∈ U and
μ ∈ V which satisfies F (x(μ0 ), μ0 ) = 0, where F is the right side of system (3) [25]. On the other hand, since
the eigenvalues
of the Jacobian matrix defined in (27) are given by −μc , −a1 (μ ) and λ± (μ ) = ±iβ (μ ) where β (μ ) = a2 (μ ), then using
the canonical form theory of Jordan [26] we obtain
⎛ ⎞
0 −β0 0 0
⎜β0 0 0 0 ⎟
J (x(μ0 ), μ0 ) = DF (x(μ0 ), μ0 ) ∼ ⎝
0 ⎠
,
0 0 −a01
0 0 0 −μc

with β0 = β (μ0 ) and a01 = a1 (μ0 ). Suppose that for μ ≈ μ0 , we have

λ+ = α (μ ) + iβ (μ )
λ− = α (μ ) − iβ (μ ),
where α (μ0 ) = 0 and β (μ0 ) = β0 . We wish to calculate
d
α  (μ ) = [Re(λ(μ ))]|μ=μ0 .
dμ
which is the crossing speed of the eigenvalues λ ± (μ) in the imaginary axis. For μ ≈ μ0 the characteristic polynomial asso-
ciated with the Jacobian matrix
246 E. Ibargüen-Mondragón, J.P. Romero-Leiton and L. Esteva et al. / Applied Mathematical Modelling 76 (2019) 238–251

⎛ ⎞
α (μ ) −
β (μ ) 0 0
⎜β (μ ) α (μ ) 0 0 ⎟
Jμ = DF (x(μ ), μ ) = ⎝
0 ⎠
, (35)
0 0 −a1 (μ )
0 0 0 −μc
is
Pμ = det (λI − Jμ )
 
= λ3 + L1 (μ )λ2 + L2 (μ )λ + L3 (μ ) (λ + μc ),
where
L1 ( μ ) = a1 ( μ ) − 2α ( μ )
L2 (μ ) = −2α (μ )a1 (μ ) + α (μ )2 + β (μ )2
L 3 ( μ ) = a 1 ( μ )[ α ( μ ) 2 + β ( μ ) 2 ] . (36)
The coefficients L1 , L2 and L3 must satisfy the same conditions as coefficients a1 , a2 and a3 of the characteristic Eq. (28), that
is L3 (μ ) − L1 (μ )L2 (μ ) = 0 (see Appendix B). By substituting (36) in the above equation we obtain the following equation
2α (μ )[(a1 (μ ) + α (μ ))2 + β (μ )2 ] = 0. (37)
For α (μ) = 0, (37) is rewritten as
(a1 (μ ) + α (μ ))2 + β (μ )2 = 0. (38)
By deriving (38) with respect to μ, we obtain

β (μ ) dβ d a1
α  (μ ) = − +
a1 ( μ ) + α ( μ ) d μ dμ

1   α (μ )
=− β ( μ )2 + a1 ( μ )2 + ( a1 ( μ )2 ) . (39)
2(a1 (μ ) + α (μ )) a1 ( μ )
By evaluating (39) at μ = μ0 , we obtain
1  
α  ( μ0 ) = − 0 β ( μ ) 2 + a 1 ( μ ) 2 ( μ0 )
2a
 1 
β0  
=−
a 2 ( μ0 ) + a 1 ( μ0 )
2a01 a2 (μ0 )

1  
=− a ( μ0 ) + a 1 ( μ0 ) .
0 2
(40)
2a1
Now, since μ = βs , from the expressions of a1 and a2 defined in (33), we obtain
s22
a1 (μ0 ) = s2 , a2 (μ0 ) = μ p s2 + + s2 ( q + δ p2 ). (41)
r2
By substituting (41) in (40), we obtain
  
1 s22
α  ( μ0 ) = − 0
μ s
p 2 + + s 2 ( q + δ p 2 ) + s 2 = 0. (42)
2a1 r2
Eq. (42) implies the existence of a Hopf bifurcation to the system (3) at the endemic equilibrium x(μ ) = P2 = (s2 , r2 , p2 , 1 ).

6. Numerical simulations

This section presents numerical simulations and graphics that illustrate the population growth of the bacteria Staphylo-
coccus aureus susceptible and resistant to different types of antibiotics. These bacteria are mainly found in the skin and nasal
mucosa but can also be found in hair, nails and other areas. Many healthy people are carriers of this microorganism, and the
spread of this bacteria from one person to another can achieved through by different mechanisms, such as direct contact or
contaminated objects [27]. Among the diseases produced by this bacterium are pneumonia, sialoadenitis (inflammation of
one of the salivary glands), and sepsis (an infection characterized by generalized injury of the vascular endothelium, which
lines the inside of blood vessels).
Since 1940, the treatment of infections by Staphylococcus aureus was accomplished with penicillin, but strains resistant to
this antibiotic have been discovered; these strains are capable of producing an enzyme called betalactamase that degrades
penicillin and makes it inactive. Currently, penicillin-resistant Staphylococcus aureus predominate in almost everyone, and
 E. Ibargüen-Mondragón, J.P. Romero-Leiton and L. Esteva et al. / Applied Mathematical Modelling 76 (2019) 238–251 247

8
s(t)
r(t)
7
p(t)
c(t)
6
Population density

−1
0 20 40 60 80 100 120
Time (minutes)
Fig. 2. Temporal course of s(t), r(t), p(t) and c(t) using the parameter values given in Table 1. In this case Rr < 1 and Rs < 1 which implies that solutions of
the system (1) approach P0 .

Table 1
Values of parameters for the model (1). rep = replica of plasmid, bact = bacterium.

Parameter Definition Value Reference

βS Growth rate of S 0.4 min−1 [28]

βR Growth rate of R 0.1 min−1 [28]
ᾱS Elimination rate of S 0.9617 (mg min)−1 [29]
K Carrying capacity of S and R 1012 bact. [30]
q̄ Mutation rate of S 0.00063785 (mg min)−1 [29]
δ̄ Transfer rate of resistant plasmids 0.2 (rep min)−1 [13]
μS Natural death rate of S 0.16 min−1 [28]
μR Natural death rate of R 0.16 min−1 [28]
μC Degradation rate of C 0.0083 min−1 [31]
 Saturation rate of C 0.8328 mg/min [31]
γ Elimination rate of S and R by IS 0.04 min−1 [3]
σp Replication rate of P 1.2 rep/(bact min) [32]
μp Death rate of P 0.012 min−1 [32]

therefore, this drug is almost no longer used to treat infections caused by this bacterium [27]. For this reason other antibi-
otics similar to penicillin were introduced that able to resist the action of betalactamases, and therefore, they are effective
for the treatment of infections caused by Staphylococcus aureus resistant to penicillin. One of these antibiotics is methicillin.
Unfortunately, strains resistant to methicillin have also appeared; these strains are often resistant to many other antibiotics
in addition to penicillin and methicillin. Some effective antibiotics are still available for the treatment of these infections,
but they usually have to be administered intravenously and are not usually as safe as penicillin [27].
The values of the parameters used in the simulations are constant and were determined based on data from Table 1. The
parameter ᾱS represents the elimination rate of bacteria by LINEZOLID [33], which is an antibiotic with ninety-nine percent
effectiveness against Staphylococcus aureus.
The trivial equilibrium solution P0 can only be stable when the number of bacteria produced by a resistant bacterium
is less than one (Rr < 1) and the number of bacteria produced by a susceptible bacterium is less than one (Rs < 1), as seen
in Fig. 2. Otherwise, when the average
 amount  of bacteria produced by the fraction of resistant bacteria that evade the
immune response is large enough Rr > 1−r1max the non-trivial equilibrium solution with no susceptible bacteria is locally
and asymptotically stable, as seen in Fig. 3.
When the saturation rate is reduced to ( = 0.0 0 08328), the populations of susceptible and resistant bacteria are peri-
odic and always coexist (see Fig. 4).
248 E. Ibargüen-Mondragón, J.P. Romero-Leiton and L. Esteva et al. / Applied Mathematical Modelling 76 (2019) 238–251

3.5

2.5 s(t)
Population density

r(t)
2 c(t)
p(t)

1.5

0.5

−0.5
0 10 20 30 40 50 60
Time (minutes)
Fig. 3. Temporal course of s(t), r(t), p(t) and c(t) using the parameter values given in Table 1. In this case Rr > 1 which implies that solutions of the
system (1) approach P1 .

3
s(t)
r(t)
2.5 p(t)
c(t)

2
Population density

1.5

0.5

−0.5
0 500 1000 1500 2000 2500 3000
Time (minutes)
Fig. 4. Temporal course of s(t), r(t), p(t) and c(t) using the parameter values given in Table 1. In this case, s(t), i(t) and r(t) are periodic solutions.
 E. Ibargüen-Mondragón, J.P. Romero-Leiton and L. Esteva et al. / Applied Mathematical Modelling 76 (2019) 238–251 249

7. Discussion and conclusion

Bacterial resistance to antimicrobials, especially to many first-line medicines, is a global human health threat. Actually,
the monitoring conducted in 2016 by the UN on drug-resistant infections and the amounts of antimicrobials used in hu-
mans have shown the need for new alternatives such as new antibiotics. In response to this finding, the WHO developed
a priority list of antibiotic-resistant bacteria to underpin the renewed efforts for research and development of new antibi-
otics [34]. In this sense, it is vital to know which are the main mechanisms of bacterial resistance acquisition and how
they influence the results of the infection. For this reason, in this work, we presented a mathematical model to explore the
dynamic relations among susceptible and resistant bacteria, plasmids and antibiotics with the assumption that drug resis-
tance is acquired through mutations and plasmid transmission. Qualitative analysis reveals the existence of a bacteria-free
equilibrium, P0 , a resistant bacteria equilibrium, P1 , and an endemic equilibrium P2 , in which both bacteria co-exist. We
found two parameters determining the existence and stability of these equilibria, namely, Rs and Rr . The parameter Rs is the
number of bacteria generated by the fraction of susceptible bacteria that survive the effects due to mutations, antibiotics
and the immune system response. Similarly, Rr represents the number of bacteria produced by the fraction of resistant bac-
teria that evade the immune response. When Rs < 1 and Rr < 1, the solutions approach the bacteria-free equilibrium which
means that susceptible and resistant bacteria are eliminated by the drugs and the immune system. This result tells us that
when the bacteria surviving both the antibiotic treatment and the immune system response are not able to reproduce, then
the infection will be controlled or eliminated. When bacteria have the ability to reproduce, there are several scenarios. Thus,
when Rr > 1−r1max , where rmax is given in (15), the solutions approach the equilibrium P1 , which implies that the infection is
caused only by resistant bacteria. This finding means that the treatment is not the most appropriate and must be changed or
reformulated. When Rs > 1 and Rr > 1−r1max , the endemic equilibrium P2 emerges, in which susceptible and resistant bacteria
co-exist.
In addition, using the Hopf Theorem we proved the existence of a limit cycle around P2 . In this case, susceptible and
resistant bacteria have an oscillatory behavior with initial period T0 = 2π β −1 . This result suggests that treatment of the
host should be re-evaluated.
To carry out the numerical simulations, we chose biological parameters of the bacteria Staphylococcus aureus which is
well known for its ability to become resistant to antibiotics, especially via plasmids and mutations. The simulations pre-
sented in Figs. 2–4 were made by using the data from Table 1, and as we can see, they corroborated the three possible
scenarios. Although the model proposed in this work does not consider all the multiple factors in bacterial resistance, it
establishes the basic relations among the different kind bacteria taking into account the most important mechanisms that
lead to acquisition of resistance
Bacterial resistance may be due to the spontaneous mutation of as a result of exposure to antibiotics as well as through
the acquisition of new resistance genes from other bacteria. A particular way to acquire resistance genes is through plasmids
which are fragments of DNA that contain genes that have the ability to replicate independently of the genetic system of the
bacterium. Model (1) is a generalization of the model formulated in [3]. Here, in addition to consider the acquisition of
resistance due to spontaneous mutations generated by contact with the antibiotic, we also consider resistance by plasmids.
The model analysis reveals the existence of the same kind of equilibria obtained in [3], i.e., the trivial equilibrium with no
bacteria; the one with resistant bacteria; and the equilibrium in which susceptible and resistant bacteria co-exist. In this
model, a Hopf bifurcation arises, which leads to periodic behaviours that can be observed in recurrent infections.
Fig. 4 shows that both susceptible and resistant bacteria present growth peaks, which decrease rapidly. Subsequently,
both bacteria regulate their growth during the period T0 = 2π β −1 maintaining a population density close to zero. This
process of self regulation is well known in the dynamics of bacterial growth, in fact it is a fundamental characteristic of
cells in response to changing environmental conditions [35]. In our case, we observe in the same figure that plasmids also
present growth peaks that decrease slowly during the period T0 = 2π β −1 , the above suggests that bacteria regulate their
growth to control the transfer of plasmids; that is, they try to transfer the minimum amount of plamids.
From the results of the model presented in this paper, it is inferred that a strategy to eliminate the bacterial load consists
of applying an adequate combination of antibiotics that help to quickly eliminate the bacterial load, generating little or no
resistance, and all this combined with an adequate stimulation of the immune system.

Acknowledgments

We want to thank anonymous referees for their valuable suggestions and comments that helped us to improve the paper.
L. Esteva acknowledges support from project IN113716 from PAPIIT-UNAM. E. Ibarguen acknowledges support from project
No 106-25/09/2018 (VIPRI-UDENAR).

Appendix A. Inequality

Substituting rmax defined in (15) in the inequality −rmax + r1 < 0 we obtain

Rs − 1
−   + r1 < 0.
Rs 1 + βsδμ p
250 E. Ibargüen-Mondragón, J.P. Romero-Leiton and L. Esteva et al. / Applied Mathematical Modelling 76 (2019) 238–251

Adding the terms

   
Rs − 1 δ q δ
−1 − 1+ ,
Rs μ p βr βr μ p βs
in both sides of the above inequality we obtain
   
δ δ Rs − 1
− 1+ −1 1+   + r1
μ p βr βs μ p Rs 1 + βsδμ p
     
Rs − 1 δ q δ q δ
+ −1 − 1+ <− 1+ ,
Rs μ p βr βr μ p βs βr μ p βs
or equivalently
 
q δ
−h2 rmax + h1 < − 1+ . (A.1)
βr μ p βs
Oh the other hand, we observe that
  q Rs −1
q δ βr R s h0
1+ = = .
βr μ p βs Rs −1 1
Rs 1+ δ
rmax
μ β p s

In consequence, substituting above equation in the inequality (A.1) we obtain

h0
−h2 rmax + h1 < − .
rmax
Therefore p(rmax ) = −h2 rmax
2 + h1 rmax + h0 < 0.

Appendix B. Equation

For a1 , a2 and a3 defined in (29) we define

3 = a 1 a 2 − a 3 . (B.1)
Observe that 3 defined in (B.1) is rewritten as
3 = α2 A2 + α1 A + α0 ,
where A is defined in (30), and
α2 = μ p
s  s δ s 
α1 = βs s2 (q + δ p2 ) 2 + 1 + 2 μ2p + δ s2 − s2 ( q + δ p2 )
2
+1
p2 r2 μp r2
        
δ s2 s2 s2
α0 = − s2 q + βr r2 (q + δ p2 ) + 1 + δ s2 q + 1 + δ p2 .
μp r2 r2 r2
Since α 2 > 0 and α 0 < 0, then A defined in (30) is the only positive root of the equation 3 = 0.

References

[1] J.M. Blair, M.A. Webber, A.J. Baylay, D.O. Ogbolu, L.J. Piddock, Molecular mechanisms of antibiotic resistance, Nat. Rev. Microbiol. 13 (2015) 42–51,
doi:10.1038/nrmicro3380.
[2] H. Lodish, A. Berk, P. Matsudaira, C.A. Kaiser, M. Krieger, M.P. Scott, S.L. Zipursky, J. Darnell, Biología Celular y Molecular, 2016.
[3] E. Ibargüen-Mondragón, L. Esteva, J.P. Romero-Leiton, S.P. Hidalgo-Bonilla, E.M. Burbano-Rosero, M. Cerón, S. Mosquera, Mathematical modeling on
bacterial resistance to multiple antibiotics caused by spontaneus mutations, Biosystems 117 (2014) 60–67, doi:10.1016/j.biosystems.2014.01.005.
[4] M.J. Bonten, D.J. Austin, M. Lipsitch, Understanding the spread of antibiotic resistant pathogens in hospitals: mathematical models as tools for control,
Clin. Infect. Dis. 33 (2007) 1739–1746, doi:10.1086/323761.
[5] B. Daşbaşi, I. Öztürk, Mathematical modelling of bacterial resistance to multiple antibiotics and immune system response, Springerplus 5 (2016) 1–17,
doi:10.1186/s40064- 016- 2017- 8.
[6] M. Merdan, Z. Bekiryazici, T. Kesemen, T. Khaniyev, Comparison of stochastic and random models for bacterial resistance, Adv. Differ. Equ. 133 (2017)
1–19, doi:10.1186/s13662- 017- 1191- 5.
[7] D. Austin, R. Anderson, Studies of antibiotic resistance within the patient, Philos. Trans. R. Soc. Lond. B Biol. Sci. 354 (1999) 721–738, doi:10.1098/rstb.
1999.0425.
[8] T.T. Nguyen, J. Guedj, E. Chachaty, J. Gunzburg, A. Andremont, F. Mentre, Mathematical modeling of bacterial kinetics to predict the impact of antibiotic
colonic exposure and treatment duration on the amount of resistant enterobacteria excreted, PLoS Comput. Biol. 10 (2014) 1–10, doi:10.1371/journal.
pcbi.1003840.
[9] E. D’Agata, P. Magal, D. Olivier, S. Ruan, G.F. Webbe, Modeling antibiotic resistance in hospitals: the impact of minimizing treatment duration, J. Theor.
Biol. 249 (2008) 487–499, doi:10.1016/j.jtbi.2007.08.011.
 E. Ibargüen-Mondragón, J.P. Romero-Leiton and L. Esteva et al. / Applied Mathematical Modelling 76 (2019) 238–251 251

[10] E. Ibargüen-Mondragón, L. Esteva, L. Chávez-Galán, A mathematical model for cellular immunology of tuberculosis, Math. Biosci. Eng. 8 (2011) 973–
986, doi:10.3934/mbe.2011.8.973.
[11] E. Ibargüen-Mondragón, L. Esteva, E.M. Burbano-Rosero, Mathematical model for the growth of mycobacterium tuberculosis in the granuloma, Math.
Biosci. Eng. 15 (2018) 407–428, doi:10.3934/mbe.2018018.
[12] E. Massad, M. Nascimento, F. Bezerra, An optimization model for antibiotic use, Appl. Math. Comput. 201 (2008) 161–167, doi:10.1016/j.amc.2007.12.
007.
[13] E. Ibargüen-Mondragón, L. Esteva, On the interactions of sensitive and resistant mycobacterium tuberculosis to antibiotics, Math. Biosci. 246 (2013)
84–93, doi:10.1016/j.mbs.2013.08.005.
[14] S. Ahmadin, F. Fatmawati, Mathematical modeling of drug resistance in tuberculosis transmission and optimal control treatment, Appl. Math. Sci.
(Ruse) 8 (2014) 4547–4559, doi:10.12988/ams.2014.46492.
[15] F. Augusto, A. Adekunle, Optimal control of a two strain tuberculosis and HIV/AIDS co-infection model, Biosystems 119 (2014) 20–44, doi:10.1016/j.
biosystems.2014.03.006.
[16] R.J. Hem, Optimal control of an HIV immunology model, Optim. Control Appl. Methods 23 (2002) 199–213, doi:10.1002/oca.710.
[17] J. Lowden, R. Miller, M. Yahdi, Optimal control of vancomycin-resistant enterococci using preventive care and treatment of infections, Math. Biosci.
249 (2014) 8–17, doi:10.1016/j.mbs.2014.01.004.
[18] J. Tan, X. Zou, Optimal control strategy for abnormal innate inmune response, Comput. Math. Methods Med. 2015 (2015) 1–16, doi:10.1155/2015/
386235.
[19] B.A.D. van Bunnik, M.E.J. Woolhouse, Modelling the impact of curtailing antibiotic usage in food animals on antibiotic resistance in humans, R. Soc.
Open Sci. 4 (2017) 1–8, doi:10.1098/rsos.161067.
[20] J. Romero-Leiton, E. Ibargüen-Mondragón, L. Esteva, Un modelo matemático sobre bacterias sensibles y resistentes a antibióticos, Rev. ERM 19 (2011)
67–85.
[21] J. Romero-Leiton, E. Ibargüen-Mondragón, Sobre la resistencia bacteriana a antibióticos de acción bactericida y bacteriostática, Integración 32 (2014)
101–116.
[22] E. Ibargüen-Mondragón, J.P. Romero-Leiton, L. Esteva, E.M. Burbano-Rosero, Mathematical modeling of bacterial resistance to antibiotics by mutations
and plasmids, J. Biol. Syst. 24 (2016) 129–146, doi:10.1142/s021833901650 0 078.
[23] P. Macheras, A. Iliadis, Modeling in Biopharmaceutics, Pharmacokinetics, and Pharmacodynamics: Homogeneous and Heterogeneous Approaches,
Springer, New York, 2006.
[24] D.S. Schmidt, Hopf’s bifurcation theorem and the center theorem of Liapunov with resonance cases, J. Math. Anal. Appl. 63 (1978) 354–370, doi:10.
1016/0 022-247X(78)90 081-1.
[25] T.M. Apostol, Mathematical Analysisis, Addison-Wesley, Massachusetts, 1960.
[26] M.W. Hisrch, S. Smale, Differential equations, in: Dynamical Systems, and Linear Algebra, Academic Press, New York, 1974.
[27] L. Fainboim, J. Geffner, Introducción a la Inmunología Humana, Editorial Médica Panamericana, Buenos Aires, sixth ed., 2011.
[28] H. Tzagolof, R. Novick, Geometry of cell division in staphylococcus Aureus, J. Bacteriol. 129 (1977) 343–350.
[29] S. Ager, K. Gould, Clinical update on Linezolid in the treatment of gram-positive bacterial infections, Infect. Drug. Resist. 5 (2012) 87–102, doi:10.2147/
IDR.S25890.
[30] T.L. Bannam, W.L. Teng, D. Bulach, D. Lyras, J.I. Rood, Functional identification of conjugation and replication regions of the tetracycline resistance
plasmid pCW3 from clostridium perfringens, J. Bacteriol. 188 (2006) 4942–4951, doi:10.1128/JB.00298-06.
[31] R. Taylor, B. Sunderland, G. Luna, P. Czarniak, Evaluation of the stability of linezolid in aqueous solution and commonly used intravenous fluids, Drug
Des. Dev. Ther. 11 (2017) 2087–2097, doi:10.2147/DDDT.S136335.
[32] A.S. Mirt, M.P. Blanco, E. Caleiras, O. Rangel, Histopatología y ultraestructura de la cromomicosis causada por cladosporium carrionii, Invest. Clin. 36
(1995) 173–182.
[33] S. Tsiodras, H.S. Gold, G. Sakoulas, G.M. Eliopoulos, C. Wennersten, L. Venkataraman, M.J. Ferraro, et al., Linezolid resistance in a clinical isolate of
staphylococcus aureus, Lancet 358 (2001) 207–208, doi:10.1016/S0140-6736(01)05410-1.
[34] T. Stalder, L.M. Rogers, C. Renfrow, H. Yano, Z. Smith, E.M. Top, Emerging patterns of plasmid-host coevolution that stabilize antibiotic resistance, Sci.
Rep. 7 (2017) 1–10, doi:10.1038/s41598- 017- 04662- 0.
[35] C.S. Hayes, D.A. Low, Signals of growth regulation in bacteria, Curr. Opin. Microbiol. 12 (2009) 667–673, doi:10.1016/j.mib.2009.09.006.