Drug efflux pump deficiency and drug target

resistance masking in growing bacteria

David Fangea, Karin Nilssona, Tanel Tensonb, and Mns Ehrenberga,1
aDepartment of Cell and Molecular Biology, Uppsala University, 75124 Uppsala, Sweden; and bInstitute of Technology, University of Tartu,
50411 Tartu, Estonia

Edited by Peter G. Wolynes, University of California at San Diego, La Jolla, CA, and approved March 24, 2009 (received for review November 12, 2008)

Recent experiments have shown that drug efflux pump deficiency exposed bacteria in a drug efflux pump proficient background
not only increases the susceptibility of pathogens to antibiotics, but virtually no advantage in a drug efflux pump deficient
but also seems to mask the effects of mutations, that decrease background. That is, the effect of the drug resistance mutation
the affinities of drugs to their intracellular targets, on the growth was masked by drug efflux pump deficiency, despite the fact
rates of drug-exposed bacteria. That is, in the presence of drugs, that the mutation per se seemed not to stimulate drug efflux in
the growth rates of drug-exposed WT and target mutated strains the pump proficient situation. Similar masking results have
are the same in a drug efflux pump deficient background, but the been reported for L22 and L4 mutations in Campylobacter jejuni
mutants grow faster than WT in a drug efflux pump proficient and Campylobacter coli (14) and for a 23S rRNA mutation in C.
background. Here, we explain the mechanism of target resistance coli (15).
masking and show that it occurs in response to drug efflux pump Here, we develop a theory that accounts for the interplay
inhibition among pathogens with high-affinity drug binding tar- between drug efflux pump efficiency and intracellular drug
gets, low cell-membrane drug-permeability and insignificant in- degradation, on the one hand, and, on the other, mutations that
tracellular drug degradation. We demonstrate that target resis- reduce the affinity of drugs to their targets. We show how the
tance masking is fundamentally linked to growth-bistability, i.e., reduced drug susceptibility conferred by such target mutations
the existence of 2 different steady state growth rates for one and in a drug efflux pump proficient background may be completely
the same drug concentration in the growth medium. We speculate lost (masked) in a drug efflux deficient background and define
that target resistance masking provides a hitherto unknown mech- the conditions under which such target resistance masking
anism for slowing down the evolution of target resistance among occurs. The theory does not invoke drug efflux pump stimulation
pathogens. of the target mutations to explain the synergism between pump
efflux efficiency and resistance. Instead, it shows resistance
antibiotic resistance ! efflux pump inhibition ! macrolides masking to emerge when intracellular drug dilution by growth
dominates over drug dilution by efflux or degradation. The

I n the years after the Second World War, the wide spread masking effect is strong for intracellular targets of high drug
introduction of antibiotics in the treatment of bacterial infec- affinity and high concentration in conjunction with drug efflux
tions revolutionized medicine and dramatically improved the pump deficiency, insignificant drug degradation in the cyto-
health condition on a global scale. Now, 60 years later, the ever plasm, and low cell wall permeability. Accordingly the theory is
evolving antibiotic resistance among pathogens has heavily mandatory for both the design and the interpretation of exper-

COMPUTATIONAL BIOLOGY
depleted the arsenal of effective antibiotic drugs. We seem to be iments addressing to the interplay between drug efflux pumps

BIOPHYSICS AND
running out of options, and a return to the pitiful health and target resistance.
conditions preceding the Second World War has become an We also demonstrate that target resistance masking is closely
ominous scenario. linked to growth-bistability (16), i.e., that a pathogen may grow
Resistance against antibiotic drugs among bacterial pathogens with either one of 2 different steady state rates in the presence
may depend on alterations in intracellular drug targets, reducing of one and the same concentration of an antibiotic drug in the
their drug affinity or rendering drug binding harmless to their medium. This finding may have baring on the observation that
function (1). Bacteria may also evolve enzyme systems degrading drug efflux pump deficiency greatly slows down the evolution of
the antibiotic drugs (1), or highly efficient drug efflux pumps for target resistance mutations among pathogens (5, 6, 8, 17). Our
rapid export of antibiotics from the bacterial cytoplasm to the analysis suggests, in other words, target resistance masking in
growth medium (2, 3). combination with growth-bistability to be a previously unrecog-
Increased attention is now being paid to the interplay between nized cause of delayed drug resistance development among
drug efflux pump efficiency and inhibition of target function in pathogens under drug efflux pump deficient conditions.
the cytoplasm. Experiments show that the drug susceptibility of Results
pathogens increases dramatically upon reduction of drug efflux
Synopsis. First, we set up differential equations for drug transport
pump efficiency by either genetic manipulation (4, 5) or addition
through the cell envelope and drug binding to intracellular
of efflux pump inhibitors (69). This feature of drug efflux pump
targets of growing bacteria. Second, we derive the conditions
deficiency is intuitively obvious, but recent findings for the
under which steady state growth-bistability can arise for bacterial
ribosome targeting macrolide antibiotic erythromycin suggest
populations exposed to antibiotic drugs. Thirdly, we describe in
the existence of more subtle aspects of the interplay between
drug efflux pump efficiency and drug susceptibility. An Esche-
richia coli mutant with an amino acid deletion in protein L22 of Author contributions: M.E. designed research; D.F., K.N., T.T., and M.E. performed research;
the large ribosomal subunit (10, 11), causing reduced affinity of D.F. contributed new reagents/analytic tools; D.F., K.N., T.T., and M.E. analyzed data; and
erythromycin to the ribosome (12), displayed reduced suscepti- D.F. and M.E. wrote the paper.
bility to erythromycin in relation to a ribosome WT strain in a The authors declare no conflict of interest.
drug efflux pump proficient but the same susceptibility in a drug This article is a PNAS Direct Submission.
efflux pump deficient background (12, 13). These experiments 1To whom correspondence should be addressed. E-mail: [email protected].
demonstrated how a mutation reducing drug affinity to a vital This article contains supporting information online at www.pnas.org/cgi/content/full/
intracellular target may give a distinct growth advantage to drug 0811514106/DCSupplemental.

www.pnas.org"cgi"doi"10.1073"pnas.0811514106 PNAS ! May 19, 2009 ! vol. 106 ! no. 20 ! 8215 8220
 ae Table 1. Summary of model parameters with corresponding
pe kpassive normalized parameters
ae envelo
kactive cell kpassive
Normalized
lux Parameter Description parameter
eff mp af af
pu ka r0 Total intracellular target concentration rf % rf/r0
af rf rb rf Free intracellular target concentration
kd #0 Growth rate* in drug absence # % #/#0
af kdeg # Growth rate in drug presence
ae Concentration of drug in growth jin % kinae/#0r0
= antibiotic = antibiotic medium
target kin Rate constant for drug influx
kout Sum of rate constant for drug kout % kout/#0
Fig. 1. Diagram illustrating drug flows over the cell envelope, drug inacti- efflux/degradation
vation and target binding. Drugs enter the cell by passive transport (rate kd Rate constant for drug dissociation kd % kd/#0
constant kpassive) and exit by passive transport (rate constant kpassive), active
ka Rate constant for drug association ka % kar0/#0
pumping (rate constant kactive), drug degradation in the cytoplasm (rate
constant kdeg) and dilution by growth (rate #, equal to #0 in drug free *Growth rate % log2/(generation time).
medium). ae is the external, af the free and a0 the total drug concentration in k % k
in passive.
the cell. The total, free and drug-bound target concentrations are r0, rf and rb, k
out % kpassive $ kactive $ kdeg.
respectively.

Drug Induced Growth-Bistability. Steady state growth-bistability for

detail the steady state aspects of drug resistance masking and its
drug-exposed bacterial populations was discussed in the special
connection with growth-bistability. Finally, we discuss the dy-
case of equilibration between free and target-bound drug in the
namics of growth inhibition in response to drug addition to or
bacterial cytoplasm (16), and here we generalize the treatment
removal from the medium. Detailed derivations and extensions
by taking also nonequilibrated scenarios into account. When the
of the theory are given in SI Appendix.
normalized growth rate, #, increases monotonically with the
fraction, rf % rf/r0, of drug free targets, there is an inverse
Cell Growth of Drug Exposed Bacteria with Intracellular Drug Targets.
function, rf % rf(#), relating jin to # in Eq. 2:
The starting point of our theory, based on the model in Fig. 1,
is 2 differential equations that account for drug influx, drug
outflux and drug target binding:

drf
j in ! !1 " r f! # "" # !k d $ # "!k out $ # "
k a r f! # "
$
$ # . [3]

! !r 0 " r f"k d " a f r f k a " # r f $ # r 0 When # is approximated by the drug-free fraction of target
dt
[1] molecules, rf, Eq. 3 simplifies to
da f
dt
! a ek in " a f k out $ !r 0 " r f"k d " a f r f k a " # a f .

Here, r0 and rf are the total and free intracellular target

j in ! !1 " # " # !k d $ # " !k out $ # "
k a
!
#
$ # . $ [4]

concentrations, respectively, whereas ae is the external, and af the

In Eq. 4, when kd && #, free and target-bound drugs are
free intracellular drug concentration. The target-bound drug (or
occupied target) concentration, rb, is r0 # rf, so that the total drug equilibrated and the target part of the growth inhibition (left
concentration, a0, equals af $ rb. The rate constants for drug factor in first term of the big parenthesis) equals KD/r0, where
dissociation from and drug association to the target are kd and KD % kd/ka (16). When kd '' #, this factor is independent of
ka, respectively. # is the growth rate of the bacterial population, kd and equals #/(kar0). For one and the same value of ae or,
equal to #0 when ae % 0, whereas kin is the first order rate equivalently, jin, there may be 2 stable steady state values of #
constant for drug influx from the medium determined by passive in Eqs. 3 and 4, as illustrated in Fig. 2A for the # % rf case. The
transport: kin % kpassive, where kpassive is determined by the cell possibility of growth-bistability depends on the normalized
membrane permeability, % (cm#1), membrane area, A (cm2), and rate constants for association to (ka) and dissociation from (kd)
cell volume V (cm3). The rate constant kout is determined by the target and on the rate constant kout (Eq. S11, Fig. S3, and
kpassive, active transport by drug efflux pumping, kactive, and Fig. 2B). The analytical results below are derived for the # %
intracellular first order drug degradation, kdeg, so that kout % rf case, but the inhibitory action of a drug may deviate from
kpassive $ kactive $ kdeg (see SI Appendix and Fig. S1) for how kout this linear dependence of # on rf. For instance, the inhibitory
is determined for one (Gram-positive bacteria) and two (Gram- action of erythromycin on ribosomes and growth rate is quite
negative bacteria) cell membranes). The inflow per cell volume, complex (12) and intracellular control systems may respond in
jin, is equal to kinae and the outflow, jout, to koutaf $ #a0, where different ways to target inhibition. However, we have found
the last term is due to drug dilution by growth. To obtain the that the existence of growth-bistability is robust to variation in
steady state solution of Eq. 1, we set the time derivatives on the the exact functional relation between # and rf, as exemplified
left hand side to zero and define dimensionless (normalized) by a weaker (# % rf1/2 in Eq. 3) and a stronger (# % rf2 in Eq.
variables # % #/#0, jin % jin/(#0r0) % kinae/(#0r0), kout % kout/#0, 3) dependence of the growth rate on the fraction of free
ka % kar0/#0 and kd % kd/#0 (Table 1). The normalized growth target, compared with the case in Eq. 4 (see SI Appendix and
rate # is then related to the normalized influx jin through (see SI below).
Appendix and Fig. S2) The plots in Fig. 2B reveal growth-bistability to be favored by
inefficient drug transport out from the cell and small rate of drug

j in ! !1 " r f" # !k d $ # "!k out $ # "

k a r f
$# . $ [2]
degradation (small kout), large drug affinity to the target and
large target concentration (large ka and small kd). All these
characteristics reflect the ultimate cause of growth-bistability,

8216 ! www.pnas.org"cgi"doi"10.1073"pnas.0811514106 Fange et al.

 kdk out
# ! . [8]
j ink a

This means that when the system leaves the fast and enters the
slow growth regime, # drops vertically from 12 to

4kdk out 4K Dk out

!
k a r0

and is thereafter proportional to the dissociation constant, KD,

for drug binding to the target.
Fig. 2. Growth bistability and the conditions under which it occurs. (A) The The slow growth regime is stable for decreasing values of
growth rate, # (y axis), is plotted as a function of the drug influx, j in (x axis), crit
jin until the slow growth critical point (Fig. 2 A): jin, s %
with ka % 106, kout % 10, and kd % 1. Upper red curve: when jin increases from
2 kdkout/ka, where # % kdkout/ka. When jin decreases further,
) )
crit
zero to jin % jin, f % 1/4, then #, approximated by !1 $ %
1 " 4jin"/2, decreases
the real-valued solution to Eq. 7 disappears and # returns to the
from 1.0 to 12, at which point the fast growth regime disappears. Lower red
fast growth regime. It follows that the growth-bistable regime is
curve: when jin decreases from high values to jin % j in,s crit
% 2 %kdkout/ka, then
#, approximated by jin!1 " %1 " 4kdkout/! j in
2
ka""/2, increases to %kdkout/ka, at
boxed in by the following inequalities
which point the slow growth regime disappears (see Table 1 for parameter
definitions). In gray: steady-state growth rate according to Eq. 4, where the % crit
2 kdk out/k a ! j in crit
s & j in & j inf ! 1/4. [9]
solid curves (partially hidden by approximation in red) represent stable
steady-states and the dashed curve represents unstable steady-states. (B) Note that according to Eq. 6 the fast growth regime behavior of
The association, ka (y axis), and dissociation, kd (x axis), rate constants for the cell population does not depend on the association and
drug-target interaction at which growth-bistability can arise for different dissociation rate constants for drug-target interactions. One
values of the drug efflux rate constant, kout. Growth-bistability is confined consequence of this is that as long as the inequality in Eq. 5 holds,
to areas northwest of the solid curves, illustrated by the shaded area in
the insert. For kd values much &1, all curves are straight lines, correspond-
target resistance alterations decreasing the association rate
ing to equilibration between free and target-bound drug (16). Dashed lines constant, ka, or increasing the dissociation rate constant, kd, do
indicate the result of an equilibrium assumption, in regions where the not affect the minimal inhibitory drug concentration (MIC
equilibrium condition is not valid. The inequality defining the northwest value, see next section). This means, in other words, that the
region is given by Eq. S11 in SI Appendix. ultimate cause of growth-bistability is also the ultimate cause of
masking of the growth effects of target resistance mutations
in the fast growth regime as described in the next section.
namely that in the fast growth regime, where the growth rate, #,
decreases from 1 to (1/2 (see Fig. 2 A and below), the major Drug Efflux Deficiency Masking of Target Resistance Mutations. In
pathway for reduction of the total intracellular drug concentra- this section, we inspect how target resistance mutations that
tion is dilution-by-growth, rather than drug efflux and/or drug decrease ka values and increase kd values affect pathogen growth
degradation. The dilution-by-growth condition is defined by in the presence of antibiotic drugs under conditions of drug
the inequality: efflux proficiency and deficiency. For this, we use Eq. 4 to

COMPUTATIONAL BIOLOGY
compute the growth rate at varying external drug concentration

BIOPHYSICS AND
!kd $ # "!k out $ # " for bacterial WT and resistance mutated populations. In one
'' # . [5] test-case, there is a 100-fold or a 10,000-fold larger kd value for
k a#
the target-mutated than for the WT pathogen. The parameters
The right hand side is the normalized rate of dilution by growth, are chosen so that the target-WT pathogen displays growth-
operating on the total intracellular drug concentration, whereas bistability in the drug-efflux deficient but not the drug-efflux
the left hand side describes the out flow and degradation (kout) proficient background, and the result is shown in Fig. 3A.
operating on the fraction of intracellular drug concentration that As expected, drug-efflux deficiency promotes greatly in-
is free, as mainly determined by kd and ka. When the inequality creased drug susceptibility of all 3 strains (Fig. 3A). We can also
in Eq. 5 is satisfied, it follows from Eq. 4 that # is approximated observe that the MIC50% values, here defined as the normalized
minimal inhibitory drug concentration, ae/r0, conferring at least
by (SI Appendix, Eq. S16)
50% reduction of growth rate, increase with increasing drug

# !
1
2
# %
1 $ 1 " 4 j in . $ [6]
efflux (see Fig. 3B). Note that the normalized drug concentra-
tion, ae/r0, is proportional to the drug inflow (Table 1) and that
in Fig. 3B we have assumed that kin % kin/#0 % 1, so that the
normalized drug inflow, jin, becomes jin % kinae/r0 % ae/r0. More
Accordingly, # decreases from 1.0 down to 1/2, as jin increases
unexpected, the normalized MIC50% values for WT and mutants
from zero in the absence of drug in the medium to its critical
crit are identical (MIC50% ( 0.25) under the drug-efflux deficient
value j in, f % 1/4 (Fig. 2 A). When jin increases above 1/4, the
condition but distinct under the drug-efflux proficient condition
real-valued solution to Eq. 6 disappears as the system exits the
(Fig. 3A, intersections with upper horizontal dotted line where
fast growth regime. As a result, # decreases abruptly to a value # % 0.5). In fact, the MIC50% values remain identical as the
much '1.0 and the intracellular drug concentration increases drug-efflux rate constant (kout) increases from 10, defining
dramatically. It follows from Eq. 4 that in the slow growth drug-efflux deficiency in Fig. 3A, up to 100 (Fig. 3B), after which
regime, # is approximated by (SI Appendix, Eq. S18) the MIC50% value of the lowest affinity mutant increases (green

# % $
curve in Fig. 3B), whereas the MIC50% values of the WT and the
jin 4k dk out middle affinity mutant remain at 0.25. The latter 2 MIC50%
# ! 1" 1" . [7]
2 2
j in k a values do not visibly diverge until the kout value supersedes
10,000 (blue and black curves in Fig. 3B). The underlying reason
For large values of jin, Eq. 7 can be Taylor expanded to for this behavior can be traced to the inequality in Eq. 5 as

Fange et al. PNAS ! May 19, 2009 ! vol. 106 ! no. 20 ! 8217
 Dynamics of Growth Inhibition. The time for bacterial growth to
A 10 0
% = 0.5 B 10 2
reach steady state after drug addition to or drug removal from
resistance mut. k%d = 10 4 k%dwt

Normalized MIC50%, ae / r0
Normalized growth rate, %

-1
10 resistance mut. k%d = 102 k%dwt the growth medium is in general much longer for drug efflux
-2 1 wildtype k% = k% wt pump deficient than for drug efflux pump proficient cells, as
10 % = 0.004 10 d d

described in detail in SI Appendix. Strikingly long adjustment

-3
10 times are observed when drug efflux pump deficient strains,
-4
10
0
10
displaying target resistance masking and growth-bistability, are
-5
exposed to concentrations close to the critical points at which
10 the cell population moves from the fast to the slow growth
-6
10
-1
10 regime or moves from the slow to the fast growth regime (Fig.
-2 0 2 4 0 2 4 6
10 10 10 10 10 10 10 10 S5 A and C). Accordingly, such dynamic measurements can be
Normalized drug inflow, %jin Norm. efflux rate constant, k%out used to reject or corroborate the existence of target resistance
masking and growth-bistability according to the mechanism
Fig. 3. Drug target resistance masking and its dependence on low drug-
proposed in the present work. To illustrate, we compare the
efflux pump efficiency. (A) Normalized steady state growth rate (#, y axis) as
function of drug normalized influx (jin, x axis) into the bacterial cells for
approach to the steady state of the drug-efflux deficient strains
drug-efflux proficient (thick lines) and deficient (thin lines) cells for target WT in Fig. 3, when they first grow in the absence of antibiotics and
(black) and target resistance mutated (blue, 100 * WT dissociation rate then are rapidly exposed to the drug at a normalized drug
constant) and (green, 10,000 * WT dissociation rate constant) strains. For the concentration (ae/r0) of 0.3, just above the critical concentration
target WT strain we use ka % 106, kd % 0.01 and we use kout % 10 in the of 0.25 at which the fast growth regime disappears (Fig. S5A).
drug-efflux deficient and kout % 106 in the drug-efflux proficient background The growth rates of all 3 strains decrease similarly and very
(see Table 1 for parameter definitions). Solid and dashed lines are stable and slowly in the fast growth regime, until they decrease sharply at
unstable steady-states, respectively. The upper and lower dotted horizontal the same point in time to their different values in the slow growth
lines indicate normalized growth rates of 0.5 and 0.004, respectively. (B)
regime. Similarly, very slow transition times are observed for cell
Normalized minimal inhibitory concentration at 50% reduction of growth
rate (MIC50% % ae/r0, y axis) as function of the drug efflux rate constant (kout,
populations in the slow growth regime that are exposed to a drug
x axis) for the WT and resistance mutated strains as in A with the same color concentration just below the critical point at which the slow-to-
code. Here, it is assumed that kin/#0 % 1, so that jin % ae/r0. fast transition occurs (Fig. S6E and Fig. S7).

Discussion
illustrated in Fig. 3B: As long the inequality is fulfilled for both Mutations that reduce the affinity of antibiotic drugs to intra-
WT and resistance mutants, they all display growth-bistability cellular targets are common among pathogens and they are
with identical drug inhibition patterns in their fast growth known to reduce the drug susceptibility and thus increase the
regime. Divergent inhibition patterns and distinct MIC50% values MIC value of bacterial mutants. Here, we have shown that
appear when the inequality in Eq. 5 is violated by large changes affinity reducing target mutations, which lead to reduced anti-
in ka or kd values or by greatly increased drug-efflux pump biotic susceptibility in the presence of efficient drug efflux
efficiency (Fig. 3B). Accordingly, masking of target resistance pumps and/or drug inactivation in the cytoplasm, may have no
mutations and growth-bistability are but two sides of the same effect on the drug susceptibility when the efflux pumps are
underlying phenomenon. Furthermore, under the drug-efflux inhibited or absent and intracellular drug inactivation is negli-
deficient condition in Fig. 3A, also MIC values defined as the gible. The growth advantage for drug exposed pathogens con-
normalized external drug concentration (ae/r0) at which the ferred by reduced target affinity may, in other words, be
fractional growth rate, #, has been reduced to '0.4% (MIC0.4%) conditional on the presence of rapid drug efflux from the
are identical for target WT and target mutants and the same as cytoplasm or efficient drug inactivation. The disappearance of
the MIC50% values (Fig. 3A, between upper and lower horizontal this growth advantage of a drug exposed mutant in relation to
dotted line). This reflects the vertical drop in growth rate as WT, due to drug efflux pump and drug inactivation deficiency,
the strains leave the fast growth regime (Eq. 6) and transit we have named target resistance masking.
through the no mans land without stationary solution down Questions we have addressed are: How can the same drug
to its slow growth regime (Eq. 8 and Fig. 3A). efflux reduction in WT and mutant eliminate the relative growth
In the slow growth regime, in contrast, the inequality in Eq. 5 advantage of the drug-exposed mutant? Under what condition
is always violated because of small # values and here the target does target resistance masking occur? Is target resistance mask-
affinity differences between mutant and WT strains are fully ing medically relevant?
expressed as shown in Eq. 8 and illustrated by the lower curves
in Fig. 3A. This means, in summary, that WT and mutant strains What Is the Working Principle of Target Resistance Masking? To get
are equally inhibited by antibiotic drugs in the fast growth an intuitive understanding of target resistance masking, we focus
regime, but in the slow growth regime they are inhibited in on scenarios where the degree of growth inhibition by the drug
proportion to the affinity of their target to the antibiotic drug. is determined by the fraction of drug-bound targets (see Fig. 1).
When the relation # % rf is replaced by the power law # % rf1/n, When the target has high affinity to the drug and/or is present
where n values &1 reflect a weak and n values '1 reflect a strong at high concentration in the cytoplasm, virtually all intracellular
dependence of the growth rate on the fraction of free targets, the drug molecules are target bound as long as their total concen-
main results regarding growth-bistability and target resistance tration is smaller than that of the target. The inflow of drug
masking remain unaltered. For instance, both growth-bistability molecules from the external medium to the cytoplasm is deter-
(SI Appendix and Fig. S4) and target resistance masking are mined by passive transport over the cell membrane(s). It depends
similar for n % 2, n % 1 and n % 1/2 (Fig. 3A, Fig. S4), but the on the external drug concentration and the drug permeability of
parameter regions allowing for growth-bistability (Fig. 2B) are the cell membrane(s) and is the same for WT and target
somewhat different in the 3 cases. In the slow growth regime, the resistance mutants. The out flow is determined by passive
1/n
growth rate decreases in proportion 1/ jin (See SI Appendix, Eq. transport of drugs by diffusion over the cell membrane(s), by
S20) and thus more rapidly for small than for large n values. One pump-mediated active transport through the cell membrane(s),
consequence of this is that the growth rate drops vertically to by enzymatic drug degradation and, finally, by dilution-by-
crit
a much lower value when jin increases beyond jin,f for small than growth. The flows in the first 3 pathways for drug reduction in
for large values of n (see Fig. 3A and Fig. S4). the cell are proportional to the free concentration of the drug in

8218 ! www.pnas.org"cgi"doi"10.1073"pnas.0811514106 Fange et al.

the cytoplasm, whereas the last flow is proportional to the total MIC values for various macrolides and ketolides were re-
drug concentration and the growth rate # (Fig. 1). For one and corded for nonisogenic WT and 23S rRNA mutated (A2057G)
the same total drug concentration in the cytoplasm, the free strains of C. coli (C. coli) with an intact and a cmeB deleted
intracellular drug concentration will be higher for the target CmeABC efflux pump (15). In the case of the macrolide
mutant than for the WT. In drug efflux pump proficient cells, erythromycin, the A2057G mutation conferred an (250-fold
where drugs are mainly removed by efflux over the cell mem- higher MIC value than WT in the efflux pump proficient
brane(s) with a rate proportional to the free drug concentration, background but only an (8-fold higher MIC value in the drug
reduced target affinity to the drug will lead to larger efflux and efflux pump deficient background. In the case of the ketolide
thus to decreased total drug concentration in the cytoplasm. This telitromycin, the A2057G mutation conferred an (6-fold higher
means that mutants with reduced target affinity to drugs will MIC value in the drug efflux proficient and virtually no MIC
have smaller drug susceptibility and higher MIC values than WT. value increase in the drug efflux deficient background. The first
What about drug efflux pump deficient strains with small cell observation is in line with partial masking of target resistance,
membrane permeability? Here, the drug outflow through the whereas the second is in line with complete masking of target
cell membrane(s) may be so small that drug efflux is dominated resistance by the drug efflux pump deficiency (compare with
by drug dilution by growth, rather than by transport over the cell Fig. S8).
membrane(s). Then, for the same growth rate, the outflow will Spontaneous macrolide-resistance mutations in ribosomal
be virtually the same for WT and target mutant. Because also the proteins L4 and L22 were obtained by plating C. jejuni and C. coli
drug inflows are equal for WT and mutant, they will have in the presence of erythromycin and tylosin (14). MIC values
virtually the same total drug concentration in the cytoplasm, the were recorded for the 2 macrolides in isogenic drug efflux
same fraction of free target and thus the same drug susceptibility proficient and deficient (cmeB-deleted) strains containing either
and growth rate. Accordingly, the target affinity difference is one of the L4 and L22 mutations. In the C. coli strains, the target
functionally masked, as long as the growth rate remains WT had an erythromycin MIC value (#g/mL) of 4 in the drug
sufficiently high to maintain dilution-by-growth as the dominant efflux proficient and 0.25 in the drug efflux deficient back-
pathway for drug reduction in both target WT and mutant. ground, whereas the L22 mutant had a MIC value of 256, i.e., 64
Target resistance masking requires low cell membrane perme- times larger than WT, in the drug efflux proficient and 0.25, i.e.,
ability, high target affinity to drug and high intracellular target the same as WT, in the drug efflux deficient background. In the
concentration as defined by the inequality in Eq. 5. From this tylosin case, the L22 and L4 mutations conferred greatly en-
follows that target resistance masking may occur for relatively hanced and varying MIC values for both the C. coli and C. jejuni
high growth rates, but what happens when the external drug strains in the drug efflux pump efficient but invariably the same
concentration increases further so that growth inhibition be- MIC value of 0.5 for WT and target mutants in the drug efflux
comes more severe? To answer this question, we need to take deficient background. Also these striking observations are com-
into account another finding of the present work, namely that patible with complete target resistance masking (Fig. S8), but the
target resistance masking is fundamentally linked to growth- experiments are semiquantitative and lack biochemical data on
bistability. target affinity, as in the erythromycin case in the previous
That is, when the dominant pathway for drug reduction in the paragraph.
cytoplasm is dilution by growth, the cell population also displays Very recently, it was demonstrated by more quantitative
growth-bistability and hysteresis with 2 possible growth rates for experiments that an amino acid deletion in protein L22 of an E.
one and the same external drug concentration (Fig. 2 A). When coli strain reduced its susceptibility to erythromycin in a drug
the external drug concentration exceeds a critical value at which efflux proficient, but not in a drug efflux deficient (+tolC)

COMPUTATIONAL BIOLOGY
the growth rate is 50% inhibited (Eq. 6), the fast growth regime background (12, 13). Interestingly, when the experiment was

BIOPHYSICS AND
disappears and the growth rate drops to a very small value (Eq. carried out in +acrB strains, retaining residual drug efflux pump
8), as the cell population enters the slow growth regime (Fig. activity, the L22 mutation conferred a susceptibility reduction
2 A). Here, the growth rate is so small that drug diffusion over that was significant but much smaller than the reduction seen in
the membrane(s) is the dominant pathway for drug reduction in the drug efflux pump proficient background (12, 13). Biochem-
the cell and, thus, the target affinity difference between WT and ical experiments showed that the L22 alteration reduced (50-
mutant is fully expressed as a growth rate difference (Eq. 8 and fold the rate constant for erythromycin binding to and (10-fold
Fig. 3A). the rate constant for erythromycin dissociation from the ribo-
One may ask whether the condition for target resistance some, corresponding to an (5-fold overall affinity loss (12). It
masking and growth-bistability is too stringent to be relevant for was, furthermore, shown by modeling, taking the complex mode
real bacterial pathogens. To inspect this, and to further of action of erythromycin into account (12), that the erythro-
illustrate the fundamental connection between growth- mycin affinity loss due to the L22 alteration could account for
bistability and target resistance masking, we discuss a simple the erythromycin induced growth inhibition observed in the drug
example (see SI Appendix and Fig. S8). In this, we use what we efflux proficient, the +acrB and + tolC backgrounds (12). At the
deem as realistic values for growth rate, drug permeability, drug same time, the L22 mutation did not stimulate drug efflux by
affinities to target and target concentration. The experimental up-regulation of the expression of existing pump protein genes
conditions mimic those of recently published observations (12, in the drug efflux proficient background (13), which led Moore
13), that will be discussed in the next section. and Sauer to the ad hoc proposition that the L22 mutation
favored ribosomal synthesis of unfolded proteins, thereby caus-
Experimental Evidence for Resistance Masking in Bacteria. The bac- ing a change in the cell wall permeability.
terial ribosome is a common target for clinically relevant anti-
biotic drugs (18, 19), and its intracellular concentration is high, Target Resistance Masking and the Evolution of Drug Resistance.
i.e., (20 #M in E. coli (20). Macrolides and ketolides bind with There are several reports in the literature that drug efflux pump
high affinity to the 23S rRNA of the 50S ribosomal subunit (21). deficiency greatly slows down the evolution of drug target
Quantitative estimates of cell wall permeability for these drugs resistance among pathogens exposed to antibiotics (e.g., refs. 4,
are missing, but it is expected to be small, because they diffuse 6, 7, and 17). The rapid evolution of drug resistance among
slowly through the phospholipid bilayer, because of their hydro- bacterial pathogens is today a major clinical problem (23, 24),
philic character, and slowly through the porins, because of their and the reports on slow drug resistance evolution are important
large size (22). in that they suggest a combination of efflux pump inhibitory

Fange et al. PNAS ! May 19, 2009 ! vol. 106 ! no. 20 ! 8219
drugs and growth inhibitory drugs as a future strategy for growth at any concentration of the growth inhibitory drug and
effective treatment of bacterial disease with minimal risk of thus no possibility for the mutant to take over the population. In
target resistance development (6). One reason for the slowing the slow growth regime, in contrast, target resistance mutations
down of target resistance emergence by drug efflux pump are fully expressed as fitness gains in relation to WT (Fig. 3A),
deficiency may simply be the greatly enhanced drug susceptibil- but here the growth rates of both WT and mutants may be too
ity that arises by drug efflux inhibition (6). Another, that the slow to allow survival of either strain when attacked by the
growth rate recovery time after drug removal is much longer immune system. There may, in other words, be clinically impor-
under conditions of drug efflux deficiency than proficiency (Fig. tant scenarios in which target resistance masking and growth-
S5 B and D). In the present work, we have shown that when drug bistability contribute to the containment of drug resistance
efflux pump deficiency is associated with target resistance evolution. These may be further analyzed with population ge-
masking and growth-bistability, there are 2 regimes with greatly netic approaches in conjunction with the present findings.
separated growth rates (Fig. 2 A). In the fast growth regime,
ACKNOWLEDGMENTS. We thank Drs. Johan Elf and Martin Lovmar for valu-
there is no difference between WT and all those mutants for able comments on the manuscript. This work was supported by the Swedish
which growth-bistability is retained (Fig. 3A). Here, there is no Medical Research Council (M.E.) and the European Regional Development
fitness advantage for the mutant in relation to the WT during Fund through the Center of Excellence in Chemical Biology (T.T.).

1. Walsh C (2000) Molecular mechanisms that confer antibacterial drug resistance. Nature 12. Lovmar M, et al. (2009) Kinetic coupling between target binding and efflux pump
406:775781. efficiencyThe case of erythromycin resistance. EMBO J 28:736 744.
2. Lomovskaya O, Zgurskaya HI, Totrov M, Watkins WJ (2007) Waltzing transporters and 13. Moore SD, Sauer RT (2008) Revisiting the mechanism of macrolide-antibiotic resistance
the dance macabre between humans and bacteria. Nat Rev Drug Discovery 6:56 65. mediated by ribosomal protein L22. Proc Natl Acad Sci USA 105:1826118266.
3. Piddock LJ (2006) Multidrug-resistance efflux pumps - not just for resistance. Nat Rev 14. Cagliero C, Mouline C, Cloeckaert A, Payot S (2006) Synergy between efflux pump
Microbiol 4:629 636. CmeABC and modifications in ribosomal proteins L4 and L22 in conferring macrolide
4. Lomovskaya O, et al. (1999) Use of a genetic approach to evaluate the consequences resistance in Campylobacter jejuni and Campylobacter coli. Antimicrob Agents Che-
of inhibition of efflux pumps in Pseudomonas aeruginosa. Antimicrob Agents Che- mother 50:38933896.
mother 43:1340 1346. 15. Cagliero C, Mouline C, Payot S, Cloeckaert A (2005) Involvement of the CmeABC efflux
5. Oethinger M, Kern WV, Jellen-Ritter AS, McMurry LM, Levy SB (2000) Ineffectiveness of pump in the macrolide resistance of Campylobacter coli. J Antimicrob Chemother
topoisomerase mutations in mediating clinically significant fluoroquinolone resis- 56:948 950.
tance in Escherichia coli in the absence of the AcrAB efflux pump. Antimicrob Agents 16. Elf J, Nilsson K, Tenson T, Ehrenberg M (2006) Bistable bacterial growth rate in response
Chemother 44:10 13. to antibiotics with low membrane permeability. Phys Rev Lett 97:258104.
6. Lomovskaya O, et al. (2001) Identification and characterization of inhibitors of multi- 17. Ricci V, Tzakas P, Buckley A, Piddock LJ (2006) Ciprofloxacin-resistant Salmonella
drug resistance efflux pumps in Pseudomonas aeruginosa: Novel agents for combina- enterica serovar Typhimurium strains are difficult to select in the absence of AcrB and
tion therapy. Antimicrob Agents Chemother 45:105116. TolC. Antimicrob Agents Chemother 50:38 42.
7. Markham PN (1999) Inhibition of the emergence of ciprofloxacin resistance in Strep- 18. Poehlsgaard J, Douthwaite S (2005) The bacterial ribosome as a target for antibiotics.
tococcus pneumoniae by the multidrug efflux inhibitor reserpine. Antimicrob Agents Nat Rev Microbiol 3:870 881.
Chemother 43:988 989. 19. Tenson T, Mankin A (2006) Antibiotics and the ribosome. Mol Microbiol 59:1664 1677.
8. Markham PN, Neyfakh AA (1996) Inhibition of the multidrug transporter NorA pre- 20. Bremer, H. & Dennis, P (1996) in Escherichia coli and Salmonella: Cellular and Molecular
vents emergence of norfloxacin resistance in Staphylococcus aureus. Antimicrob Biology, ed Neidhardt FC (ASM, Washington, DC).
Agents Chemother 40:26732674. 21. Douthwaite S, Hansen LH, Mauvais P (2000) Macrolide-ketolide inhibition of MLS-
9. Markham PN, Westhaus E, Klyachko K, Johnson ME, Neyfakh AA (1999) Multiple novel resistant ribosomes is improved by alternative drug interaction with domain II of 23S
inhibitors of the NorA multidrug transporter of Staphylococcus aureus. Antimicrob rRNA. Mol Microbiol 36:183193.
Agents Chemother 43:2404 2408. 22. Nikaido, H (2003) Molecular basis of bacterial outer membrane permeability revisited.
10. Apirion D (1967) Three genes that affect Escherichia coli ribosomes. J Mol Biol Microbiol Mol Biol Rev 67:593 656.
30:255275. 23. Furuya EY, Lowy FD (2006) Antimicrobial-resistant bacteria in the community setting.
11. Wittmann HG, et al. (1973) Biochemical and genetic studies on two different types of Nat Rev Microbiol 4:36 45.
erythromycin resistant mutants of Escherichia coli with altered ribosomal proteins. Mol 24. Martinez JL, Baquero F, Andersson DI (2007) Predicting antibiotic resistance. Nat Rev
Gen Genet 127:175189. Microbiol 5:958 965.

8220 ! www.pnas.org"cgi"doi"10.1073"pnas.0811514106 Fange et al.