Quantifying Subpopulation Synergy for Antibiotic Combinations via

Mechanism-Based Modeling and a Sequential Dosing Design

Cornelia B. Landersdorfer,a,b Neang S. Ly,b Hongmei Xu,b Brian T. Tsuji,b Jürgen B. Bulittaa,b
Centre for Medicine Use and Safety, Monash University (Parkville campus), Parkville, Victoria, Australiaa; School of Pharmacy and Pharmaceutical Sciences, University at
Buffalo, State University of New York, Buffalo, New York, USAb

Quantitative modeling of combination therapy can describe the effects of each antibiotic against multiple bacterial populations.
Our aim was to develop an efficient experimental and modeling strategy that evaluates different synergy mechanisms using a
rapidly killing peptide antibiotic (nisin) combined with amikacin or linezolid as probe drugs. Serial viable counts over 48 h were
obtained in time-kill experiments with all three antibiotics in monotherapy against a methicillin-resistant Staphylococcus aureus
USA300 strain (inoculum, 108 CFU/ml). A sequential design (initial dosing of 8 or 32 mg/liter nisin, switched to amikacin or lin-
ezolid at 1.5 h) assessed the rate of killing by amikacin and linezolid against nisin-intermediate and nisin-resistant populations.

Downloaded from http://aac.asm.org/ on July 8, 2018 by guest

Simultaneous combinations were additionally studied and all viable count profiles comodeled in S-ADAPT and NONMEM. A
mechanism-based model with six populations (three for nisin times two for amikacin) yielded unbiased and precise (r ⴝ 0.99,
slope ⴝ 1.00; S-ADAPT) individual fits. The second-order killing rate constants for nisin against the three populations were 5.67,
0.0664, and 0.00691 liter/(mg · h). For amikacin, the maximum killing rate constants were 10.1 hⴚ1 against its susceptible and
0.771 hⴚ1 against its less-susceptible populations, with 14.7 mg/liter amikacin causing half-maximal killing. After incorporating
the effects of nisin and amikacin against each population, no additional synergy function was needed. Linezolid inhibited suc-
cessful bacterial replication but did not efficiently kill populations less susceptible to nisin. Nisin plus amikacin achieved sub-
population synergy. The proposed sequential and simultaneous dosing design offers an efficient approach to quantitatively char-
acterize antibiotic synergy over time and prospectively evaluate antibiotic combination dosing strategies.

T he emerging global health care crisis caused by multidrug-

resistant bacteria and a lack of effective antibiotics presents a
major challenge and is one of the three greatest threats to human
health (1–5). While resistance is increasing rapidly, the number of
new approved antibiotics has declined dramatically since the
1980s (6). Synergistic combinations of available antibiotics offer a
promising and timely option to combat multidrug-resistant bac-
teria. However, quantitative approaches for antibiotic combina-
tions that characterize the time course of synergistic killing or of
resistance prevention are scarce (7–9). Existing methods to de-
scribe synergy are limited to outcomes at a single time point (often
24 h) and do not implement the mechanism(s) of synergy (10, 11).
In vitro experiments and mathematical modeling offer advan-
tages, since they can assess the time course and proposed mecha-
nisms of synergy and can therefore more rationally optimize an-
tibiotic combination therapy.
Bacterial populations that are less susceptible or even resistant
to antibiotics may cause failure of antibiotic therapy, in particular
in severe infections with a high bacterial burden or infections in
patients with an impaired immune system (9, 12, 13). High-inoc-
ulum infections likely harbor preexisting populations that are less
susceptible to one or both antibiotics. To optimally treat such
infections, antibiotic combination strategies may utilize subpop-
ulation synergy with one antibiotic killing the resistant popula-
tion(s) of the second antibiotic and vice versa (Fig. 1) (7–9). In the
absence of a population that is resistant to both antibiotics in mono-
therapy, subpopulation synergy can yield eradication of a bacterial
inoculum without resistance. Such synergy seems particularly prom-
ising for combinations of antibiotics that do not share an important
resistance mechanism which affects both antibiotics.
For the simplest case of subpopulation synergy, bacterial kill-
ing by antibiotic A does not affect killing by antibiotic B and vice
versa (i.e., independent effects). Such combinations may be de-
signed to maximize killing during the first hours of therapy to
prevent bacteria from becoming tolerant or resistant to both an-
tibiotics. Importantly, emergence of resistance is a time-depen-
dent process involving both phenotypic tolerance and genotypic
resistance mechanisms (14). To ensure survival of the predomi-
nant population during the first hours of therapy, bacteria com-
monly utilize phenotypic tolerance mechanisms that do not re-
quire a mutation.
To minimize both upregulation of tolerance and de novo for-
mation of resistant mutants, we used a rapidly killing first antibi-
otic (nisin) and tested the efficacy of the second antibiotic against
the population(s) surviving the first antibiotic in real time (i.e.,
without isolation of resistant colonies after growth for ⱖ24 h on
antibiotic-containing agar plates, for example). Some studies as-
sessed the susceptibility (MICs) of colonies resistant to the first
antibiotic that were generated via long-term (ⱖ24 h) passaging
toward a panel of second antibiotics (15). We are not aware of
systematic methods to evaluate subpopulation synergy that ac-
Received 12 January 2013 Returned for modification 20 February 2013
Accepted 3 March 2013
Published ahead of print 11 March 2013
Address correspondence to Cornelia B. Landersdorfer,
cornelia.landersdorfer@monash.edu, or Jürgen Bulitta,
jurgen.bulitta@monash.edu.
Supplemental material for this article may be found at http://dx.doi.org/10.1128
/AAC.00092-13.
Copyright © 2013, American Society for Microbiology. All Rights Reserved.
doi:10.1128/AAC.00092-13

May 2013 Volume 57 Number 5 Antimicrobial Agents and Chemotherapy p. 2343–2351 aac.asm.org 2343
Landersdorfer et al.
FIG 1 Subpopulation synergy concept. Bacterial cells inside the solid line are
susceptible to drug A; bacterial cells inside the long dashed lines are susceptible
to drug B. Bacteria within the dotted lines are resistant to either drug A or drug
B. The combination of drug A and drug B will eradiate these bacterial popu-
lations due to the lack of a population resistant to both drugs.
counted for the time-dependent upregulation and minimized the
impact of phenotypic tolerance mechanisms during the first hours
of therapy. Therefore, methods that can rapidly isolate less-suscepti-
ble populations may be a valuable alternative. This may be important
to better understand the reasons for success of simultaneously dosed
combinations, as bacteria have limited time to become tolerant or
resistant after simultaneous dosing of two antibiotics.
Nisin, a lantibiotic, is a peptide antibiotic that induces pore
formation in bacterial membranes and inhibits peptidoglycan
synthesis. Due to its rapid killing of methicillin-resistant Staphy-
lococcus aureus (MRSA) (16, 17), nisin served as a probe drug.
Our primary objective was to assess an experimental strategy
that identifies and characterizes potential types of synergy and to
develop quantitative, mechanism-based models for antibiotic
combinations. In addition to monotherapy and simultaneously
dosed antibiotic combinations, we utilized a new sequential de-
sign that exposed bacteria to nisin for 1.5 h to rapidly kill nisin-
susceptible bacteria. Subsequently, nisin was removed and the ef-
ficacy of amikacin or linezolid was assessed against the bacteria
surviving 1.5 h exposure to nisin.
(Part of this work has been presented as a short oral presenta-
tion and a poster at the National Institute of General Medical
Sciences Quantitative and Systems Pharmacology Workshop II,
Bethesda, MD, 9 to 10 September 2010 and as a poster at the
American Conference on Pharmacometrics, San Diego, CA, 3 to 6
April 2011.)
MATERIALS AND METHODS
Bacterial strains, media, and susceptibility testing. An MRSA USA300
strain obtained from the Network on Antimicrobial Resistance was uti-
lized for all experiments (18). Static time-kill studies and MIC tests were
performed in Luria-Bertani broth (Difco Laboratories, Detroit, MI) sup-
plemented with 12.5 mg/liter Mg2⫹ and 25 mg/liter Ca2⫹. The MICs were
additionally determined in duplicate in cation-adjusted Mueller-Hinton
broth. Nisin A and amikacin were obtained from Sigma-Aldrich, St.
Louis, MO, and linezolid was provided as a kind gift by Pfizer. Nisin A was
dissolved at pH 2 in diluted hydrochloric acid, and the denatured milk
solids were removed by centrifugation. Viable counts were determined on
Luria-Bertani agar (Difco Laboratories, Detroit, MI).
Time-kill experiments. Time-kill experiments were performed as de-
scribed previously (19–21). In brief, fresh bacterial colonies were grown
on Luria-Bertani agar for approximately 20 h prior to each experiment. A
bacterial suspension of ⬃109 CFU/ml in saline was prepared spectropho-
tometrically and diluted into 20 ml of fresh, prewarmed, sterile broth to
obtain an initial inoculum of ⬃107.7 CFU/ml. Bacteria were then grown
for 60 min to ⬃108.0 CFU/ml prior to addition of the respective antibiot-
2344 aac.asm.org
FIG 2 Mechanistic synergy with drug B enhancing the rate of killing by drug
A for the population susceptible to drug A, the population resistant to drug A,
or both populations.
ic(s). Viable counts representing the initial inoculum were obtained
within less than 10 min prior to dosing.
Monotherapy. Static time-kill studies assessed nisin (4, 8, 32, and 128
mg/liter), amikacin (1, 4, 16, and 64 mg/liter), and linezolid (2, 8, and 32
Downloaded from http://aac.asm.org/ on July 8, 2018 by guest
mg/liter) in monotherapy.
Combinations. Sequential combinations with nisin switched to ami-
kacin or linezolid quantified the rate of growth and bacterial killing by
amikacin or linezolid against the nisin-intermediate and nisin-resistant
populations. Both the nisin-intermediate and nisin-resistant populations
were selected by exposing the initial bacterial inoculum to 8 mg/liter nisin
over 1.5 h. The nisin-resistant population only was selected separately by
exposure to 32 mg/liter nisin over 1.5 h. These nisin concentrations and
the 1.5 h duration of exposure were optimized in pilot time-kill studies
(results not shown). Nisin was then removed by centrifugation of the
conical tubes and careful removal of the supernatant. The bacterial pellets
were immediately resuspended in prewarmed broth containing no anti-
biotic (control), 8 or 16 mg/liter amikacin, or 8 or 32 mg/liter linezolid at
approximately 1.75 h. In addition to the sequential combinations with
pretreatment by nisin, simultaneous combinations also were studied,
where the bacteria were simultaneously exposed to nisin plus amikacin or
nisin plus linezolid throughout the whole experiment. The simultane-
ously dosed combinations were studied for selected informative combi-
nations of nisin (8, 16, and 32 mg/liter) with amikacin (4, 8, and 16 mg/
liter) or linezolid (2, 8, and 32 mg/liter).
Viable counting. For all arms, serial viable counts were quantified
over 48 h as described previously (19–21). For sequential combinations,
additional viable counts were assessed before centrifugation and ⬃5 min
after resuspension of the bacterial pellets.
Pharmacodynamic modeling and synergy concept. (i) Subpopula-
tion synergy. We define two types of synergy: subpopulation synergy and
mechanistic synergy. Subpopulation synergy (Fig. 1) occurs if antibiotic A
kills the bacteria that are less susceptible to antibiotic B and vice versa. If
there are no bacteria resistant to both antibiotics, the combination will
eradicate all bacteria. In the simplest case of independent effects, antibi-
otic A does not affect the rate of killing by antibiotic B and vice versa.
(ii) Mechanistic synergy. The second type of synergy defined here is
mechanistic synergy (Fig. 2), which, in the present analysis, refers to an
enhanced rate of killing of a bacterial population due to the simultaneous
presence of two antibiotics. Mechanistic synergy was identified via mod-
eling if a bacterial population was killed more rapidly by the combination
of antibiotics A and B than by the independent effects of both antibiotics
in monotherapy. For example, antibiotic B may be a ␤-lactamase inhibi-
tor or an efflux pump inhibitor that achieves no (or limited) killing in
monotherapy. In combination with antibiotic A, antibiotic B may in-
crease the target site concentration and therefore the killing by antibiotic
A. In our modeling analysis, mechanistic synergy was considered for ei-
ther specific population(s) or all populations.
Life cycle growth model. For each population, a life cycle growth
model (22, 23) was applied that accounts for the underlying biology of
bacterial growth (24) and contains bacteria that are preparing for replica-
tion (state 1) and bacteria immediately before the replication step (state
2). The transition from state 1 to state 2 occurred via a first-order growth
Antimicrobial Agents and Chemotherapy

FIG 3 Life cycle growth model for the population susceptible to both nisin
and amikacin. This two-state life cycle growth model was applied to each of the
six populations in Fig. 4. k12, first-order growth rate constant; k21, first-order
replication rate constant (representing doubling); InhRep, inhibition of the
probability of successful replication; Inhk12, inhibition of the first-order
growth rate; other symbols are explained in Table 1 and Materials and
Methods.
rate constant, k12 (Fig. 3), and replication (k21, or doubling) was assumed
to be fast. This two-state life cycle growth model offers the advantage to
describe a lower growth rate due to inhibition of protein synthesis by
linezolid, for example. Alternatively, a less-susceptible population may
have a decreased biofitness and a longer mean generation time (1/k12)
compared to susceptible bacteria (Fig. 3). The potentially decreased
biofitness of less-susceptible populations was estimated as a growth rate
factor (fk12) that was multiplied with the growth rate constant (k12) of the
susceptible population. For example, the growth rate constant (k12_RR) of
a population resistant to both antibiotics was expressed as k12_RR ⫽
fk12_RR · k12.
Population model. Models with one, two, or three preexisting popu-
lations with different susceptibilities to the respective antibiotic were con-
sidered similarly to previously described models (8, 14, 19, 22, 25–30).
The susceptibility of each population was estimated via a specific rate of
bacterial killing by the respective antibiotic (Fig. 4).
Nisin-plus-amikacin combination. The mathematical model for ni-
sin plus amikacin contained six populations, including all combina-
tions of a nisin-susceptible (Niss), -intermediate (Nisi), and -resistant
(Nisr) population and of an amikacin-susceptible (Amis) and -resis-
tant (Amir) population. With the two states (i ⫽ 1 and i ⫽ 2 in equa-
tion 1) for each population from the life cycle growth model, the total
population (CFUALL) contained 12 compartments (i.e., two compart-
ments for each of the six populations).
2 killing rate constants (i.e., k2 was either k2S, k2I, or k2R) as FrSurvive ⫽
CFUALL ⫽ 冱 CFUSS,i ⫹ CFUIS,i ⫹ CFURS,i ⫹ CFUSR,i
i⫽1
⫹ CFUIR,i ⫹ CFURR,i (1)
The differential equation for the concentration of bacteria from the
Niss/Amis population in state 1 (CFUSS,1) comprised the second-order
killing by nisin and a saturable killing by amikacin (initial condition de-
scribed below):
dCFUSS,1
dt 冉
⫽ REP · k21 · CFUSS,2 ⫺ k12 ⫹ k2S · CNis
冊
KmaxS · CAmiHill
⫹ · CFUSS,1
CAmiHill ⫹ KC50Hill
where CNis is the nisin and CAmi the amikacin concentration in broth
medium. CFUSS,2 is the concentration of bacteria in state 2 of the Niss/
Amis population. The rate constants k21, k12, and k2S and the parameters
for the Hill-type killing function for amikacin (KmaxS, KC50, and Hill) are
defined in Table 1. The replication factor REP ensures that the total con-
centration of bacteria in all populations (CFUALL) does not exceed the
maximum population size (CFUmax):
冉
REP ⫽ 2 · 1 ⫺
CFUALL
CFUmax ⫹ CFUALL 冊 (3)
At low viable counts, REP approaches 2, representing a 100% proba-
bility of successful replication (i.e., doubling). As CFUALL approaches
May 2013 Volume 57 Number 5
Subpopulation Synergy of Antibiotic Combinations
FIG 4 Subpopulation synergy model for nisin and amikacin comprising six
populations with different susceptibility to each antibiotic. Niss, nisin suscep-
tible; Nisi, nisin intermediate; Nisr, nisin resistant; Amis, amikacin susceptible;
Amir, amikacin resistant. Thick arrows represent fast growth or fast killing.
Solid downward arrows represent killing by nisin, and dashed arrows represent
Downloaded from http://aac.asm.org/ on July 8, 2018 by guest
killing by amikacin. See Table 1 for parameter explanations.
CFUmax, REP approaches 1, representing a 50% probability of successful
replication where bacteria continue to transition between states 1 and 2
but the total viable count is constant (22). The differential equation for
CFUSS,2 also included killing by nisin and amikacin:
dCFUSS,2
dt 冉
⫽ k12 · CFUSS,1 ⫺ k21 ⫹ k2S · CNis
冊
KmaxS · CAmiHill
⫹ · CFUSS,2 (4)
CAmiHill ⫹ KC50Hill
The differential equations for the five populations that were less sus-
ceptible to nisin, amikacin, or both are shown in the supplemental mate-
rial. These differential equations accounted for the decreased susceptibil-
ity to nisin (using k2I or k2R instead of k2S; Table 1) or to amikacin (using
KmaxR instead of KmaxS). Some less-susceptible populations were esti-
mated to have a decreased biofitness described by a longer mean genera-
tion time (denoted by the rate constants k12_RS_SR, k12_IR, and k12_RR
[Table 1]).
Survival fraction. The survival fraction after pretreatment with nisin
for 1.5 h was calculated for each population based on the second-order
exp(⫺k2 · CNis · 1.5 h).
Initial conditions. The total inoculum and the mutation frequency of
each less-susceptible population were estimated model parameters. The
product of the total inoculum and the mutation frequency defined the size
of each of the less-susceptible populations at 0 h (i.e., initial condition).
The initial condition of the population susceptible to both antibiotics was
calculated by subtracting the size of all less-susceptible populations from
that of the total initial inoculum. All bacteria of a population were initial-
ized into state 1 of the respective population (i.e., for i ⫽ 1 in equation 1).
Initial conditions for the compartments representing state 2 (i ⫽ 2 in
equation 1) were zero, since k21 was rapid.
(2)
Nisin-plus-linezolid combination. The model for nisin plus linezolid
contained three populations, including nisin-susceptible and linezolid-
susceptible (Niss/Lins), nisin-intermediate and linezolid-susceptible
(Nisi/Lins), and nisin-resistant and linezolid-intermediate (Nisr/Lini)
populations. The differential equation for the Niss/Lins population in
state 1 (CFUSS,1) contained second-order killing by nisin and two inhib-
itory effects of linezolid (Fig. 3; initial condition calculated as for the
nisin-plus-amikacin model):
dCFUSS,1
dt
⫽ REP · (1 ⫺ InhRep) · k21 · CFUSS,2 ⫺ (k12 · Inhk12
⫹ k2S · CNis) · CFUSS,1 (5)
where REP was defined as above and CFUALL was the sum of both states
aac.asm.org 2345
Landersdorfer et al.

TABLE 1 Parameter estimates for nisin combinations with amikacin or linezolid

Estimate (relative SE [%])

Nisin ⫹ amikacin Nisin ⫹ linezolid

Parameter Symbol (unit) S-ADAPT NONMEM S-ADAPT NONMEM

Growth parameters
Log10 (initial inoculum) 7.89 (2.0)c 7.84 (2.2) 7.88 (1.4)c 7.83
Mean generation time
Niss/Amis and Nisi/Amis k12⫺1 (min) 57.3 (7.0)b 67.5 (9.5)
Niss/Lins and Nisi/Lins k12⫺1 (min) 83.9 (7.5)b 56.6
Growth rate factors
Nisr/Amis and Niss/Amir fk12_RS_SR Very lowa Very lowa
Nisi/Amir fk12_IR 0.532d 0.519 (14)
Nisr/Amir fk12_RR 0.413d 0.386 (12)
Nisr/Lini fk12_R 0.144 (15) 0.102
Replication rate constant k21 (h⫺1) 50 (fixed) 50 (fixed) 50 (fixed) 50 (fixed)
Log10 (maximum population size) CFUmax 9.23 (2.5)c 9.23 (2.0) 9.38 (1.7)c 9.25

Log10 (mutation frequencies)

Downloaded from http://aac.asm.org/ on July 8, 2018 by guest

Nisi/Amis ⫺4.25 (15)c ⫺3.63 (8.2)
Nisr/Amis ⫺3.25 (7.1) ⫺3.89 (8.0)
Niss/Amir ⫺2.57 (23) ⫺2.58 (18)
Nisi/Amir ⫺4.37 (9.3) ⫺4.48 (8.2)
Nisr/Amir ⫺7.42 (7.7) ⫺7.47 (4.5)
Nisi/Lins ⫺2.88 (6.2)c ⫺3.02
Nisr/Lini ⫺4.15 (7.3) ⫺3.59

Second-order killing rate constants for nisin

Susceptible population k2S [liters/(mg · h)] 5.67 (14)b 3.58 (17) 4.49 (12)b 3.08
Intermediate population k2I [liters/(mg · h)] 0.0664 (6.9) 0.0605 (19) 0.0209 (12) 0.0605
Resistant population k2R [liters/(mg · h)] 0.00691 (13) 0.00484 (21) 0.00318 (13) 0.00383

Killing by amikacin
Maximum killing rate constants
Susceptible population KmaxS (h⫺1) 10.1 (19) 9.10 (21)
Resistant population KmaxR (h⫺1) 0.771 (18) 0.635 (10)
Amikacin concn causing 50% of Kmax KC50 (mg/liter) 14.7 (7.7) 16.5 (7.2)
Hill coefficient Hill 2.45 (12)b 2.83 (14)

Inhibition of replication and growth rate by linezolid

Inhibition of successful replication by linezolid
(InhRep) for Niss/Lins and Nisi/Lins
Maximum inhibition of replication ImaxRep 1.0 (fixed) 1.0 (fixed)
Linezolid concn causing half-maximum IC50,Prot (mg/liter) 3.92 (19) 7.35
inhibition of protein synthesis
Inhibition of rate of growth (all populations, Inhk12)
Maximum inhibition of rate of growth Imax,k12 0.858 (12) 0.918
Linezolid concn causing 50% of effect IC50,k12 (mg/liter) 4.25 (31) 19.4
Hill coefficient Hillk12 10 (fixed) 10 (fixed)
Turnover rate constant for protein pool kProt (h⫺1) 0.72 (7.1) 0.458
Standard deviation of additive residual error on SDCFU 0.395 (9.5) 0.775 (15) 0.307 (7.6) 0.549
log10 scale
a
Replication for these two populations was estimated to be very slow in the present static time-kill experiment. Eventually, fk12_RS_SR was fixed to 0.
b
For the population PD modeling analysis in S-ADAPT, the biological variability between the viable count profiles (between-curve variability) was set to a coefficient of variation of
15% for all parameters except those estimated on a log10 scale. The between-curve variability was set to 10% for the Hill coefficient.
c
The between-curve variability for parameters estimated on a log10 scale in S-ADAPT was set to a variance of 0.25. For the maximum population size and the experimentally
standardized initial inocula, the variances were estimated to be small and therefore eventually fixed to 0.01.
d
As these fractions were estimated via a logistic transformation in S-ADAPT, their mean estimate was close to zero on a transformed scale (equivalent to 50% on a linear scale).
This resulted in relative standard errors of 222% for fk12_IR and of 65.4% for fk12_RR, since the mean on transformed scale was almost zero. These artificially inflated relative
standard errors did not impact the performance of the model or the curve fits, as also suggested by the small standard error in the NONMEM analysis which did not use this
transformation.

for the three populations of the nisin-plus-linezolid model. Parameters in the linezolid concentration (CLin) required for half-maximal inhibition of
equations 5 to 9 are explained in Table 1. Linezolid inhibited protein protein synthesis. In the absence of linezolid, the protein pool is at its
synthesis, and the protein constituent pool (P), with an initial condition of hypothetical baseline of 100%. Depletion of the protein pool by linezolid
1.0, was defined as: increases the probability of unsuccessful replication (InhRep):
dP
dt
⫽ kprot · 冋冉 1⫺
CLin
CLin ⫹ IC50,Prot 冊 册
⫺P (6)
InhRep ⫽ ImaxRep · (1 ⫺ P)
As described previously (22), an InhRep (equations 5 and 7) of 0.50
(7)

kprot is the turnover rate constant for the protein pool, and IC50,Prot is results in net stasis of the respective bacterial population and an InhRep of

2346 aac.asm.org Antimicrobial Agents and Chemotherapy


⬎0.5 results in bacterial killing, since bacteria that replicate unsuccessfully
are eliminated (i.e., lost from the system) (Fig. 3). Linezolid additionally
inhibited the growth rate (Inhk12):
Imax,k12 · CLinHillk12
Inhk12 ⫽ 1 ⫺ (8)
CLinHillk12 ⫹ IC50,k12Hillk12
The differential equation for bacteria in state 2 (CFUSS2) was:
dCFUSS,2
⫽ k12 · Inhk12 · CFUSS,1 ⫺ (k21 ⫹ k2S · CNis) · CFUSS,2
dt
(9)
The differential equations for state 1 (CFUIS,1) and state 2 (CFUIS,2) of
the Nisi/Lins population were:
dCFUIS,1
⫽ REP · (1 ⫺ InhRep) · k21 · CFUIS,2 ⫺ (k12 · Inhk12
dt
⫹ k2S · CNis) · CFUIS,1 (10)
dCFUIS,2
⫽ k12 · Inhk12 · CFUIS,1 ⫺ (k21 ⫹ k2S · CNis) · CFUIS,2
dt
(11)
The same growth rate constant (k12) was used for the two Lins popu-
lations (Niss/Lins and Nisi/Lins), and a lower growth rate constant (k12_R)
was used for the Nisr/Lini population to reflect a decreased biofitness. The
differential equations for state 1 (CFURI,1) and state 2 (CFURI,2) of the
Nisr/Lini population contained the inhibitory effect of linezolid on
the growth rate (Inhk12) but lacked the effect on the probability of suc-
cessful replication (InhRep):
dCFURI,1
⫽ REP · k21 · CFURI,2 ⫺ (k12_R · Inhk12 ⫹ k2S · CNis) · CFURI,1
dt
(12)
dCFURI,2
⫽ k12_R · Inhk12 · CFURI,1 ⫺ (k21 ⫹ k2S · CNis) · CFURI,2
dt
(13)
Observation model. All log10 viable counts for nisin plus amikacin or
nisin plus linezolid were simultaneously fitted using an additive residual
error on a log10 scale. For observations below 100 CFU/ml, the number of
colonies per plate was directly fitted using a previously developed residual
error model (19).
Computation and model selection. To provide a robust mathemati-
cal analysis, model development was performed independently in two
software packages (S-ADAPT and NONMEM) which utilize different es-
timation algorithms. The importance sampling Monte Carlo parametric
expectation-maximization method (pmethod ⫽ 4) in S-ADAPT (version
1.57 [31]) using SADAPT-TRAN (32, 33) and the first-order conditional
estimation method with the interaction option (FOCE⫹I) in NONMEM
VI (level 1.2; using the ADVAN9 subroutine; NONMEM Project Group,
Icon Development Solutions, Ellicott City, MD [34]) were applied. Mod-
els were evaluated based on the S-ADAPT objective function value (⫺1⫻
log likelihood), NONMEM objective function value (⫺2⫻ log likeli-
hood), and a series of standard diagnostic plots as previously described
(35–37).
RESULTS
Nisin and amikacin monotherapies. The MIC of the studied
MRSA strain was 16 mg/liter for nisin and 8 mg/liter for amikacin.
Both nisin and amikacin displayed (relatively) rapid killing and
prevented regrowth over 48 h at the highest tested concentration
in monotherapy (Fig. 5A and B). Nisin at 4 and 8 mg/liter yielded
2.3 to 3.5 log10 killing during the first 2 h followed by near-com-
plete regrowth at 24 h.
Survival fractions during pretreatment. Pretreatment with 8
or 32 mg/liter nisin for 1.5 h resulted in 2 to 3 log10 killing (Fig. 5C
May 2013 Volume 57 Number 5
Subpopulation Synergy of Antibiotic Combinations
and Fig. 6C). The modeled survival fraction (FrSurvive; based on
estimated k2S in Table 1) of the Niss populations after 1.5 h pre-
treatment was ⬍10⫺18, indicating a complete killing of Niss bac-
teria (calculated from k2S in Table 1). For the Nisi populations,
FrSurvive was 0.056 (equivalent to 1.25 log10 killing) for 32 mg/liter
nisin and 0.49 (equivalent to 0.3 log10 killing) for 8 mg/liter nisin
pretreatment. Thus, 32 mg/liter nisin pretreatment killed the Nisi
populations to levels below those of the Nisr populations. For the
Nisr populations, the FrSurvive values were 0.79 for 32 mg/liter and
0.94 for 8 mg/liter nisin pretreatment, suggesting that these pop-
ulations were killed by ⱕ0.1 log10 by either pretreatment. These
FrSurvive results were in good agreement with results from the pilot
experiments (results not shown) that optimized the nisin concen-
trations and duration of pretreatment.
Nisin switched to amikacin. Control arms that switched from
nisin pretreatment to no antibiotic showed growth and thus con-
Downloaded from http://aac.asm.org/ on July 8, 2018 by guest
firmed the viability of the surviving bacteria (results not shown).
Nisin control arms that contained a 2-fold-lower nisin concentra-
tion compared to that in pretreatment showed stasis followed by
growth and thus confirmed the decreased susceptibility of bacteria
surviving nisin pretreatment (results not shown).
Against bacteria surviving 32 mg/liter nisin pretreatment, se-
quential therapy with 16 mg/liter amikacin achieved viable counts
below the limit of counting (1.3 log10 CFU/ml equivalent to 1
colony per plate) at 9.5 and 32 h (Fig. 5C). Thus, the surviving
bacteria were susceptible to amikacin monotherapy, since amika-
cin remained active against Nisi and Nisr bacteria selected via pre-
treatment.
Simultaneous combination of nisin and amikacin. Consis-
tent with the assumption that amikacin killed the Nisi and Nisr
populations, the simultaneous combination of 32 mg/liter nisin
plus 16 mg/liter amikacin achieved near-eradication from 4 h, and
16 mg/liter nisin plus 16 mg/liter amikacin reduced viable counts
to below the limit of counting from 8 h (Fig. 5E).
Modeling nisin and amikacin. The rate of bacterial killing by
nisin over the studied concentration range was satisfactorily de-
scribed when modeled to be proportional to the nisin concentra-
tion, and a more complex function was not necessary to describe
the profiles. This second-order killing function for nisin was in
agreement with its ability to form pores in bacterial membranes
(16, 38). The killing rate constant was 85.5-fold lower for the Nisi
and 821-fold lower for the Nisr populations compared to the Niss
populations (Table 1). Bacterial killing by amikacin was relatively
rapid and adequately described by a Hill function that takes into
account the fact that amikacin reaches a maximum effect on bac-
terial killing at high concentrations and the sigmoidicity of the
concentration effect relationship. This choice of a Hill function
was in agreement with previous papers on aminoglycosides (39,
40). The estimated maximal killing rate constant for amikacin
against the Amis populations was 13-fold higher than that of the
Amir populations (Table 1). Only a very small fraction (10⫺7.42 for
Nisr/Amir [Table 1]) of the initial inoculum was estimated to be
resistant to both nisin and amikacin.
A model with independent (i.e., additive) effects of nisin and
amikacin against each of the six populations fitted all viable count
profiles well and was in agreement with the observation that ami-
kacin killed the Nisi and Nisr populations. Thus, a model with
subpopulation synergy was suitable to describe the viable counts
(Fig. 5C, D, and E).
All 20 viable count profiles with nisin and amikacin mono-
aac.asm.org 2347
Landersdorfer et al.

Downloaded from http://aac.asm.org/ on July 8, 2018 by guest

FIG 5 Observed and individual fitted viable count profiles from S-ADAPT for nisin (A) and amikacin (B) in monotherapy and sequential (C) and simultaneous
(D and E) combinations. Time is the time after start of the experiment. For the sequential combinations, the time axis shows the time after switching treatment
to amikacin at 1.75 h after the start of the experiment. Plates with zero colonies are plotted as 0 log10 CFU/ml.

therapy, sequential and simultaneous combinations (Fig. 5) were
successfully comodeled with unbiased and precise fits (r, 0.99;
slope, 0.996; intercept, 0.05 log10 CFU/ml for observed versus in-
dividually fitted log10 viable counts). For the 1- and 4-mg/liter
amikacin curves, the observation at 4 h was overpredicted
(Fig. 5B); however, this did not have a large influence, since the
overall extent of killing by 1 and 4 mg/liter amikacin was very
modest and the 1-h and 8-h samples were reasonably fitted. The
observed versus population fitted log10 viable counts (i.e., in the
absence of random between curve variability) were also unbiased
and reasonably precise (r ⫽ 0.94). All parameters for the nisin and
amikacin model were estimated with good precision (relative
standard errors, ⱕ23% except for fk12_IR and fk12_RR [Table 1]).
Inclusion of an enhanced rate of killing for the simultaneous com-
bination improved neither the objective function nor the curve
fits, suggesting a lack of mechanistic synergy. This result on the
types of synergy was based on thorough modeling analyses in
S-ADAPT and NONMEM which yielded comparable parameter
estimates (Table 1) and led to the same conclusion regarding a lack
of mechanistic synergy.
Linezolid and nisin. The MIC was 2 mg/liter for linezolid. At
the high inoculum, 32 mg/liter linezolid achieved slow killing by 1
log10 from 8 to 56 h (Fig. 6B), whereas 2 mg/liter linezolid essen-
tially paralleled the growth control. After nisin pretreatment, lin-
ezolid yielded mostly static profiles, suggesting that linezolid dis-
played some activity against the Nisi and Nisr populations
(Fig. 6C). The simultaneous combinations of linezolid with 16
or 32 mg/liter nisin achieved 3 to 6 log10 killing at 48 h (Fig. 6D
and E).
Modeling nisin and linezolid. For nisin and linezolid, all 20
profiles were adequately fit simultaneously (Fig. 6) with an overall
r of 0.99, a slope of 1.001, and an intercept of 0.002 log10 CFU/ml
for the observed versus individual fitted log10 viable counts and an
r of 0.96 for the population fits. Only the 8- and 32-mg/liter curves
for the nisin monotherapy and the 2-mg/liter linezolid mono-
therapy showed slight misfits (Fig. 6A and B) which did not affect
the characterization of the type of synergy. In contrast to nisin and
amikacin, linezolid did not exhibit fast bacterial killing (Fig. 6B).
In agreement with its effect on inhibiting protein synthesis, lin-
ezolid prolonged the mean generation time (Inhk12 in Fig. 3) of all
populations. For the Niss/Lins and Nisi/Lins populations, linezolid
additionally inhibited the probability of successful replication

2348 aac.asm.org Antimicrobial Agents and Chemotherapy


FIG 6 Observed and individual fitted viable count profiles from S-ADAPT for nisin (A) and linezolid (B) in monotherapy and sequential (C) and simultaneous
(D and E) combinations.
(InhRep), most likely by affecting the synthesis of proteins that are
necessary for replication. Therefore, linezolid (slowly) killed these
populations at linezolid concentrations above the IC50,Prot of 3.92
mg/liter (Table 1) at the high inoculum studied. Linezolid
achieved a static response only against the Nisr/Lini population
which was modeled by linezolid prolonging the mean generation
time (Inhk12). Nisin killed this population slowly.
The Nisr/Lini population was estimated to have a lower growth
rate than the Lins populations. The estimated log10 mutation fre-
quencies suggested that the initial inoculum comprised a rela-
tively high fraction (10⫺4.15) of the Nisr/Lini population (Table 1).
This result indicated that high nisin and linezolid concentrations
in combination would be required to achieve more than ⬃4.15
log10 killing against this high inoculum.
DISCUSSION
The experimental and modeling strategy developed in the present
study was designed to efficiently identify and characterize synergy
mechanisms for antibiotic combinations. We proposed to distin-
guish between two types of synergy. Subpopulation synergy can be
greatly beneficial, if one antibiotic kills the resistant population of
a second antibiotic and vice versa (Fig. 1). Mechanistic synergy
May 2013 Volume 57 Number 5
Subpopulation Synergy of Antibiotic Combinations
Downloaded from http://aac.asm.org/ on July 8, 2018 by guest
was defined here as a higher rate of killing of a bacterial population
under the simultaneous presence of two antibiotics (Fig. 2) com-
pared to the rate of killing of this bacterial population predicted by
the independent effects of both antibiotics in monotherapy.
The proposed approach accounted for the limited time avail-
able to bacteria to elaborate their tolerance and resistance mech-
anisms to survive standard, simultaneously dosed antibiotic com-
binations. Short-term (1.5-h) pretreatment with a rapidly killing
first antibiotic (nisin) was utilized to select less-susceptible popu-
lations. This approach allows one to test second antibiotics against
the entire surviving bacterial population and does not require
isolating presumably more stably resistant colonies after
growth over 24 h or even longer on agar plates containing the
first antibiotics.
Nisin very rapidly killed MRSA at a high inoculum which al-
lowed an efficient selection of the Nisi and Nisr populations. Ami-
kacin achieved substantial killing against the Nisi and Nisr popu-
lations (Fig. 5B), whereas linezolid yielded only stasis or slow
killing (Fig. 6B) against these populations. The Nisr/Amir popula-
tion had a log10 mutation frequency of ⫺7.42 (Table 1), suggest-
ing that nisin in combination with amikacin can achieve more
than 7 log10 killing at high drug concentrations. This was success-
aac.asm.org 2349
Landersdorfer et al.
fully confirmed for both sequential and simultaneously dosed
combinations (Fig. 5).
The nisin-plus-amikacin combination leverages subpopula-
tion synergy via nisin killing the Amir populations and amikacin
killing the Nisi and Nisr populations (except for the small Nisr/
Amir population that was only [slowly] killed by 16 mg/liter ami-
kacin [Fig. 5E]). This sequential dosing strategy was critical, since
it provided experimental insights into why eradication was
achieved by the simultaneously dosed combination. In immuno-
competent patients the small Nisr/Amir population might be erad-
icated by the immune system (12, 13), as the vast majority of
bacteria are rapidly killed by the antibiotic combination.
In contrast, the Nisr/Lini population had a high mutation fre-
quency of ⫺4.15 log10 (Table 1). This suggested limited promise
for the nisin-plus-linezolid combination against this MRSA strain
at a high inoculum. However, future studies are required to assess
whether the inhibition of protein synthesis and inhibition of bac-
terial growth by linezolid may postpone the rate of emergence of
resistance.
It was critical for the proposed approach to characterize sub-
population synergy against a high total inoculum exceeding the
inverse of the mutation frequency of the less-susceptible popula-
tions. Standard MIC testing and checkerboard synergy studies in
96-well plates with a low initial inoculum usually utilize ⱕ5 ⫻ 105
CFU/ml (equivalent to ⱕ1 ⫻ 105 CFU for a 200-␮l volume per
well). Such low inocula have an extremely low probability (here,
0.38%) to carry a single bacterial cell of the Nisr/Amir population
that had a log10 mutation frequency of ⫺7.42. Therefore, subpop-
ulation synergy likely cannot be detected in checkerboard synergy
studies at low inocula. In contrast, the high inoculum used in this
study of ⬃2 · 109 CFU per arm (⫽ 20 ml ⫻ 108 CFU/ml) had a
⬎99.9999% probability of containing the Nisr/Amir population.
Mathematical modeling and pilot experiments consistently in-
dicated that pretreatment with 8 mg/liter nisin completely killed
the Niss populations but (essentially) did not kill the Nisi and Nisr
populations. Pretreatment with 32 mg/liter nisin completely
killed the Niss populations and killed the Nisi populations to via-
ble counts below the Nisr populations but had very limited effect
against the Nisr populations. This model fitted the viable count
profiles well (Fig. 5) based on the assumption of subpopulation
synergy and independent killing effects of nisin and amikacin.
Inclusion of a mechanistic synergy term with an enhanced rate
of killing of one or multiple bacterial populations for the simulta-
neously dosed combination was not needed for nisin and amika-
cin. In general, an enhanced rate of killing due to the simultaneous
presence of two antibiotics (i.e., mechanistic synergy) may be dif-
ficult to estimate for antibiotic combinations with subpopulation
synergy, since eradication is already achieved by the first antibiotic
killing the resistant population of the second antibiotic and vice
versa. Mechanistic synergy is easier to identify and estimate, if one
antibiotic has no (or limited) effect in monotherapy, such as for a
␤-lactam/␤-lactamase inhibitor combination or for combina-
tions with an efflux pump inhibitor.
The present study also showed that the development of life
cycle growth models with multiple (here, up to six) bacterial pop-
ulations is robust and the model parameters were well estimated
in two software packages utilizing different estimation algorithms.
The proposed model for nisin employed a second-order killing
function similar to previously developed models for colistin (19),
a peptide antibiotic against Gram-negative pathogens. Published
2350 aac.asm.org
models for aminoglycosides against Pseudomonas aeruginosa and
Staphylococcus aureus utilized Hill-type killing functions in agree-
ment with the present combination model (39, 40). Consistent
with previously developed models for linezolid against Enterococ-
cus faecalis (41) and MRSA (18), the present model included an
effect of linezolid prolonging the mean generation time and inhib-
iting the probability of successful replication due to linezolid in-
hibiting protein synthesis. The population analysis in S-ADAPT,
in addition to the analysis in NONMEM, demonstrated that the
model structure is sufficiently complex to capture all viable count
profiles well, as shown by the r ⱖ 0.99 for the individual fitted
versus observed log10 CFU/ml, and it suggested there is a small
between-curve variability that is reflected by variability in the pa-
rameter estimates.
In summary, nisin achieved subpopulation synergy with ami-
kacin but not with linezolid against a high inoculum of MRSA.
Downloaded from http://aac.asm.org/ on July 8, 2018 by guest
Additional mechanistic synergy was not needed to simultaneously
model the time course of all viable counts for nisin and amikacin.
The proposed experimental and modeling strategy is valuable to
predict the potential benefits of combination therapies, as shown
here with nisin as a probe drug. The proposed strategy can effi-
ciently identify and quantify subpopulation synergy and explain
why simultaneously dosed combinations yield eradication. Short-
term (1.5 h) pretreatment with a rapidly killing first antibiotic
allows this approach to minimize the impact of upregulation of
tolerance and resistance mechanisms. This may be important, as
bacteria have limited time to elaborate their tolerance and resis-
tance mechanisms to survive standard, simultaneously dosed an-
tibiotic combinations in patients. Therefore, this sequential dos-
ing strategy may be beneficial for future studies on combinations
with established and new antibiotics.
ACKNOWLEDGMENTS
This study was partly supported by a grant from the Center for Protein
Therapeutics at SUNY, Buffalo, NY. N.S.L. is supported by a predoctoral
fellowship from the American Foundation for Pharmaceutical Education.
H.X. was supported by a postdoctoral fellowship from Pfizer. J.B.B. is an
Australian Research Council DECRA fellow (DE120103084).
We thank Silvia V. Brown for excellent technical assistance during the
performance of the experiments.
We report no conflict of interest.
REFERENCES
1. Boucher HW, Talbot GH, Bradley JS, Edwards JE, Gilbert D, Rice LB,
Scheld M, Spellberg B, Bartlett J. 2009. Bad bugs, no drugs: no ESKAPE!
An update from the Infectious Diseases Society of America. Clin. Infect.
Dis. 48:1–12.
2. Drusano GL. 2007. Pharmacokinetics and pharmacodynamics of antimi-
crobials. Clin. Infect. Dis. 45(Suppl 1):S89 –S95.
3. Payne DJ, Gwynn MN, Holmes DJ, Pompliano DL. 2007. Drugs for bad
bugs: confronting the challenges of antibacterial discovery. Nat. Rev. Drug
Discov. 6:29 – 40.
4. Walker B, Barrett S, Polasky S, Galaz V, Folke C, Engstrom G, Acker-
man F, Arrow K, Carpenter S, Chopra K, Daily G, Ehrlich P, Hughes T,
Kautsky N, Levin S, Maler KG, Shogren J, Vincent J, Xepapadeas T, de
Zeeuw A. 2009. Looming global-scale failures and missing institutions.
Science 325:1345–1346.
5. Bradley JS, Guidos R, Baragona S, Bartlett JG, Rubinstein E, Zhanel
GG, Tino MD, Pompliano DL, Tally F, Tipirneni P, Tillotson GS,
Powers JH. 2007. Anti-infective research and development—problems,
challenges, and solutions. Lancet Infect. Dis. 7:68 –78.
6. Spellberg B, Blaser M, Guidos RJ, Boucher HW, Bradley JS, Eisenstein
BI, Gerding D, Lynfield R, Reller LB, Rex J, Schwartz D, Septimus E,
Tenover FC, Gilbert DN. 2011. Combating antimicrobial resistance:
Antimicrobial Agents and Chemotherapy

policy recommendations to save lives. Clin. Infect. Dis. 52(Suppl 5):S397–
S4428.
7. Bulitta JB, Li J, Poudyal A, Yu HH, Owen RJ, Tsuji BT, Nation RL,
Forrest A. 2009. Quantifying synergy of colistin combinations against
MDR gram negatives by mechanism-based models, abstr A1-573. Intersci.
Conf. Antimicrob. Agents Chemother., San Francisco, CA, 12 to 15 Sep-
tember 2009.
8. Bulitta JB, Li J, Poudyal A, Yu HH, Owen RJ, Tsuji BT, Nation RL,
Forrest A. 2010. Mechanism-based modelling of the synergy of colistin
combinations against multidrug-resistant gram negative bacteria, abstr
1918. Abstr. Annu. Meet. Population Approach Group in Europe (PAGE).
www.page-meeting.org/?abstract⫽1918.
9. Bulitta JB, Landersdorfer CB, Forrest A, Brown SV, Neely MN, Tsuji
BT, Louie A. 2011. Relevance of pharmacokinetic and pharmacodynamic
modeling to clinical care of critically ill patients. Curr. Pharm. Biotechnol.
12:2044 –2061.
10. Greco WR, Bravo G, Parsons JC. 1995. The search for synergy: a critical
review from a response surface perspective. Pharmacol. Rev. 47:331–385.
11. White DB, Slocum HK, Brun Y, Wrzosek C, Greco WR. 2003. A new
nonlinear mixture response surface paradigm for the study of synergism: a
three drug example. Curr. Drug Metab. 4:399 – 409.
12. Drusano GL, Fregeau C, Liu W, Brown DL, Louie A. 2010. Impact of
burden on granulocyte clearance of bacteria in a mouse thigh infection
model. Antimicrob. Agents Chemother. 54:4368 – 4372.
13. Drusano GL, Vanscoy B, Liu W, Fikes S, Brown D, Louie A. 2011.
Saturability of granulocyte kill of Pseudomonas aeruginosa in a murine
model of pneumonia. Antimicrob. Agents Chemother. 55:2693–2695.
14. Jumbe N, Louie A, Leary R, Liu W, Deziel MR, Tam VH, Bachhawat R,
Freeman C, Kahn JB, Bush K, Dudley MN, Miller MH, Drusano GL.
2003. Application of a mathematical model to prevent in vivo amplifica-
tion of antibiotic-resistant bacterial populations during therapy. J. Clin.
Invest. 112:275–285.
15. Li J, Nation RL, Owen RJ, Wong S, Spelman D, Franklin C. 2007.
Antibiograms of multidrug-resistant clinical Acinetobacter baumannii:
promising therapeutic options for treatment of infection with colistin-
resistant strains. Clin. Infect. Dis. 45:594 –598.
16. Lubelski J, Rink R, Khusainov R, Moll GN, Kuipers OP. 2008. Biosyn-
thesis, immunity, regulation, mode of action and engineering of the model
lantibiotic nisin. Cell. Mol. Life Sci. 65:455– 476.
17. McAuliffe O, Ross RP, Hill C. 2001. Lantibiotics: structure, biosynthesis
and mode of action. FEMS Microbiol. Rev. 25:285–308.
18. Tsuji BT, Brown T, Parasrampuria R, Brazeau DA, Forrest A, Kelchlin
PA, Holden PN, Peloquin CA, Hanna D, Bulitta JB. 2012. Front-loaded
linezolid regimens result in increased killing and suppression of the acces-
sory gene regulator system of Staphylococcus aureus. Antimicrob. Agents
Chemother. 56:3712–3719.
19. Bulitta JB, Yang JC, Yohonn L, Ly NS, Brown SV, D’Hondt RE, Jusko
WJ, Forrest A, Tsuji BT. 2010. Attenuation of colistin bactericidal activity
by high inoculum of Pseudomonas aeruginosa characterized by a new
mechanism-based population pharmacodynamic model. Antimicrob.
Agents Chemother. 54:2051–2062.
20. Tsuji BT, von Eiff C, Kelchlin PA, Forrest A, Smith PF. 2008. Attenu-
ated vancomycin bactericidal activity against Staphylococcus aureus
hemB mutants expressing the small-colony-variant phenotype. Antimi-
crob. Agents Chemother. 52:1533–1537.
21. Harigaya Y, Bulitta JB, Forrest A, Sakoulas G, Lesse AJ, Mylotte JM,
Tsuji BT. 2009. Pharmacodynamics of vancomycin at simulated epithelial
lining fluid concentrations against methicillin-resistant Staphylococcus
aureus (MRSA): implications for dosing in MRSA pneumonia. Antimi-
crob. Agents Chemother. 53:3894 –3901.
22. Bulitta JB, Ly NS, Yang JC, Forrest A, Jusko WJ, Tsuji BT. 2009.
Development and qualification of a pharmacodynamic model for the pro-
nounced inoculum effect of ceftazidime against Pseudomonas aeruginosa.
Antimicrob. Agents Chemother. 53:46 –56.
23. Lin HY, Landersdorfer CB, London D, Meng R, Lim CU, Lin C, Lin S,
Tang HY, Brown D, Van Scoy B, Kulawy R, Queimado L, Drusano GL,
Louie A, Davis FB, Mousa SA, Davis PJ. 2011. Pharmacodynamic
May 2013 Volume 57 Number 5
Subpopulation Synergy of Antibiotic Combinations
modeling of anti-cancer activity of tetraiodothyroacetic acid in a perfused
cell culture system. PLoS Comput. Biol. 7:e1001073. doi:10.1371/journal
.pcbi.1001073.
24. Maidhof H, Johannsen L, Labischinski H, Giesbrecht P. 1989. Onset of
penicillin-induced bacteriolysis in staphylococci is cell cycle dependent. J.
Bacteriol. 171:2252–2257.
25. Meagher AK, Forrest A, Dalhoff A, Stass H, Schentag JJ. 2004. Novel
pharmacokinetic-pharmacodynamic model for prediction of outcomes
with an extended-release formulation of ciprofloxacin. Antimicrob.
Agents Chemother. 48:2061–2068.
26. Campion JJ, Chung P, McNamara PJ, Titlow WB, Evans ME. 2005.
Pharmacodynamic modeling of the evolution of levofloxacin resistance in
Staphylococcus aureus. Antimicrob. Agents Chemother. 49:2189 –2199.
27. Mouton JW, Vinks AA, Punt NC. 1997. Pharmacokinetic-
pharmacodynamic modeling of activity of ceftazidime during continuous
and intermittent infusion. Antimicrob. Agents Chemother. 41:733–738.
28. Czock D, Keller F. 2007. Mechanism-based pharmacokinetic-
pharmacodynamic modeling of antimicrobial drug effects. J. Pharmaco-
kinet. Pharmacodyn. 34:727–751.
29. Nielsen EI, Viberg A, Lowdin E, Cars O, Karlsson MO, Sandstrom M.
Downloaded from http://aac.asm.org/ on July 8, 2018 by guest
2007. Semimechanistic pharmacokinetic/pharmacodynamic model for
assessment of activity of antibacterial agents from time-kill curve experi-
ments. Antimicrob. Agents Chemother. 51:128 –136.
30. Gumbo T, Louie A, Deziel MR, Parsons LM, Salfinger M, Drusano GL.
2004. Selection of a moxifloxacin dose that suppresses drug resistance in
Mycobacterium tuberculosis, by use of an in vitro pharmacodynamic in-
fection model and mathematical modeling. J. Infect. Dis. 190:1642–1651.
31. Bauer RJ. 2010. S-ADAPT/MCPEM user’s guide (version 1.57). Software
for pharmacokinetic, pharmacodynamic and population data analysis,
Berkeley, CA.
32. Bulitta JB, Bingolbali A, Shin BS, Landersdorfer CB. 2011. Development
of a new pre- and post-processing tool (SADAPT-TRAN) for nonlinear
mixed-effects modeling in S-ADAPT. AAPS J. 13:201–211.
33. Bulitta JB, Landersdorfer CB. 2011. Performance and robustness of the
Monte Carlo importance sampling algorithm using parallelized S-ADAPT
for basic and complex mechanistic models. AAPS J. 13:212–216.
34. Beal SL, Sheiner LB, Boeckmann AJ. 2006. NONMEM users guides
(1989 –2006). Icon Development Solutions, Ellicott City, MD.
35. Bulitta JB, Duffull SB, Kinzig-Schippers M, Holzgrabe U, Stephan U,
Drusano GL, Sorgel F. 2007. Systematic comparison of the population
pharmacokinetics and pharmacodynamics of piperacillin in cystic fibrosis
patients and healthy volunteers. Antimicrob. Agents Chemother. 51:
2497–2507.
36. Bulitta JB, Kinzig M, Jakob V, Holzgrabe U, Sörgel F, Holford NHG.
2010. Nonlinear pharmacokinetics of piperacillin in healthy volunteers—
implications for optimal dosage regimens. Br. J. Clin. Pharmacol. 70:682–
693.
37. Landersdorfer CB, Kirkpatrick CM, Kinzig M, Bulitta JB, Holzgrabe U,
Drusano GL, Sorgel F. 2009. Competitive inhibition of renal tubular
secretion of gemifloxacin by probenecid. Antimicrob. Agents Chemother.
53:3902–3907.
38. Vollmerhaus PJ, Breukink E, Heck AJ. 2003. Getting closer to the real
bacterial cell wall target: biomolecular interactions of water-soluble lipid
II with glycopeptide antibiotics. Chemistry 9:1556 –1565.
39. Mouton JW, Vinks AA. 2005. Pharmacokinetic/pharmacodynamic mod-
elling of antibacterials in vitro and in vivo using bacterial growth and kill
kinetics: the minimum inhibitory concentration versus stationary con-
centration. Clin. Pharmacokinet. 44:201–210.
40. Tam VH, Kabbara S, Vo G, Schilling AN, Coyle EA. 2006. Comparative
pharmacodynamics of gentamicin against Staphylococcus aureus and
Pseudomonas aeruginosa. Antimicrob. Agents Chemother. 50:2626 –
2631.
41. Tsuji BT, Bulitta JB, Brown T, Forrest A, Kelchlin PA, Holden PN,
Peloquin CA, Skerlos L, Hanna D. 2012. Pharmacodynamics of early,
high-dose linezolid against vancomycin-resistant enterococci with ele-
vated MICs and pre-existing genetic mutations. J. Antimicrob. Che-
mother. 67:2182–2190.
aac.asm.org 2351