Report

Effect of tolerance on the evolution of antibiotic resistance under drug combinations

See allHide authors and affiliations

Science  10 Jan 2020:
Vol. 367, Issue 6474, pp. 200-204
DOI: 10.1126/science.aay3041

Challenges of drug combinations

Combinations of antibiotics are used to treat intractable infections such as methicillin-resistant Staphylococcus aureus. Clinically, however, drugs tend to be used empirically, and results can be contradictory. Liu et al. translated observations made in vitro to patient samples to understand the role of antibiotic tolerance in promoting or suppressing resistance when drug combinations are used (see the Perspective by Berti and Hirsch). Although bacterial populations exposed to multiple antibiotics can develop tolerance to multiple drugs, one drug in a combination may be able counter resistance to a partner drug and provide effective therapy. However, if tolerance has already emerged to one drug, the combination may end up promoting the transmission of resistance to a partner drug.

Science, this issue p. 200; see also p. 141

Abstract

Drug combinations are widely used in clinical practice to prevent the evolution of resistance. However, little is known about the effect of tolerance, a different mode of survival, on the efficacy of drug combinations for preventing the evolution of resistance. In this work, we monitored Staphylococcus aureus strains evolving in patients under treatment. We detected the rapid emergence of tolerance mutations, followed by the emergence of resistance, despite the combination treatment. Evolution experiments on the clinical strains in vitro revealed a new way by which tolerance promotes the evolution of resistance under combination treatments. Further experiments under different antibiotic classes reveal the generality of the effect. We conclude that tolerance is an important factor to consider in designing combination treatments that prevent the evolution of resistance.

Evolution experiments have shown that tolerance evolves quickly under cyclic antibiotic treatments (13) and subsequently promotes the evolution of antibiotic resistance (4). In contrast to resistance mutations that decrease the effectiveness of the antibiotic and elevate the minimum inhibitory concentration (MIC) (5), tolerance mutations increase the minimum time to kill the population without changing the MIC (610). To understand whether the evolutionary trajectory of evolving tolerance—and thereafter resistance—occurs in patients, we followed sequential isolates of life-threatening methicillin-resistant Staphylococcus aureus (MRSA) blood infections in which the bacterial infection persisted for at least 2 weeks despite antibiotic treatment (fig. S1).

Between May 2017 and May 2018, 2 of 48 adult patients (>18 years old) admitted to Shaare Zedek Hospital with MRSA bacteremia fitted our inclusion criteria (see materials and methods). Frozen stocks of streaks obtained from the blood cultures were prepared. The single-cell distribution of growth phenotype of the bacterial population was measured by plating serial dilutions of the frozen stock on agar plates and following the appearance of each colony with the Scanlag setup (see materials and methods).

In patients 1 and 2, we observed that colonies, arising from the bacterial population isolated 1 week after the start of treatment, occurred much later than those from the population isolated on day 1 (Fig. 1A and fig. S2). In some samples, early- and late-appearing phenotypes coexisted (Fig. 1A; see day 5), but eventually only the late-appearing phenotype remained in both patients. From each subpopulation, single colonies were selected randomly for further analyses. MIC assays, tolerance detection tests (TDtests), and killing assays performed on these clones revealed that the late-appearing phenotype—observed on day 5 and associated with tolerance to vancomycin (VAN) (Fig. 1, B and C) (1115)—was characterized by a reduced killing rate but showed no change to the MIC (Fig. 1D and fig. S3). The VAN-tolerant phenotype was always associated with impaired bacterial growth (Fig. 1, E and F, and fig. S4) (1618). Whole-genome reconstruction of the ancestral strains isolated before antibiotic treatment (day 1), and comparison with whole-genome sequencing (WGS) of isolates from later days, identified a few single-point mutations—single-nucleotide polymorphisms (SNPs)—attributed to tolerance in several genes [e.g., RNA polymerase subunit (rpoC), transcriptional repressor of purine biosynthesis (purR) (in patient 1), and Clp protease subunit (clpX) (in patient 2) (table S1 and table S2)]. Identical SNPs were detected in isolates from different days, indicating clonal evolution, which enabled phylogenetic reconstruction (Fig. 2A).

Fig. 1 Evolution of tolerance in patient 1.

(A) Distribution of the time of appearance of colonies on solid medium from subsequent samples was measured by ScanLag. The y axis represents the normalized proportion of colony-forming units (CFUs) detected at each sampling time. Sample sizes N = 244, 310, 660, 456, 932, 411, 498, 244, 240, and 759, respectively. (B) Tolerance detection using the TDtest (11). Step I: Single colonies isolated on day 1, 5, and 7 are exposed to a VAN disk (10 μg), which results in a zone of inhibition with similar radius in all strains (i.e., no resistance increase). Step II: Addition of a nutrient disk after the antibiotic concentration has decreased below MIC allows detection of increased survival of the tolerant strains. (Bottom) Enlargement of a region within the inhibition zone shows numerous colonies that regrew at step II for the strains isolated on day 5 and 7. Strain names are based on patient (P#), day of isolation (D#), and clone isolated (C#). (C) Killing assay in liquid medium of clones isolated on different days from patient 1 under VAN (30 μg/ml). Data are presented as the means ± SD from at least three biological replicates. (D) MIC measurement of VAN with Etest (epsilometer test) (micrograms per milliliters). The white dashed line denotes the value of the MIC on day 1. (E and F) Growth impairment detected in the clinical samples in the absence of antibiotics. (E) Representative time-lapse phase-contrast microscopy images of clones isolated on day 1 (upper row) and day 7 (lower row). Scale bars, 5 μm. (F) Decreased growth rate in the tolerant strain (orange). Data extracted from the experiments shown in (E) display bacteria numbers over time, normalized by initial number at the beginning of observation.

Fig. 2 Within-patient evolution of antibiotic tolerance is followed by resistance.

(A) Nonsynonymous mutations identified in the subsequent strains isolated from the blood of patient 1. Phylogenetic analysis suggests that they all originated from clonal evolution of the ancestral strain. tol, tolerant; res, resistant. (B) Relative MDK99 [minimum duration for killing 99% (6)] for VAN and DAP of strains isolated from patient 1. (C) MIC for VAN (cyan), RIF (purple), and DAP (red). Strains used in (B) and (C) are shown underlined in (A). Colored bars below (B) and (C) indicate the antibiotic treatment regime of patient 1 during hospitalization. Data are presented as the means ± SD from at least three biological replicates.

In vitro experiments tend to use a single drug, but—as is frequently the case for patients with severe infections, such as patient 1—treatment in these cases involved several drugs. Thus, our bacterial isolates were exposed to three antibiotics: VAN, rifampicin (RIF), and daptomycin (DAP) (Fig. 2). First, only VAN was administered, which selected for strains tolerant to VAN within 5 days (Fig. 2B). RIF was combined with the VAN treatment on day 4. Because bacteremia persisted, VAN was replaced by DAP, so that from day 8 to 14, patient 1 was under treatment with RIF and DAP. Despite the combination treatment, mutations in the polymerase gene rpoB, which are known to generate RIF resistance (19), emerged after the tolerance mutations (Fig. 2C and figs. S5 to S7). The evolution of antibiotic resistance in patient 1 is notably similar to the rapid evolution of antibiotic resistance observed in vitro after a tolerant phenotype has been established in experimental evolution (4).

To understand whether the rapid evolution of RIF resistance in the patient was facilitated by the tolerance phenotype, we quantified in vitro survival to each drug—separately and in combination—at drug concentrations close to those occurring in vivo (see supplementary materials). All isolates tolerant to VAN were also highly tolerant to DAP (hereafter called VAN/DAP-tolerant) (Fig. 1C and fig. S5) (20). A representative VAN/DAP-tolerant strain showed a survival advantage of three orders of magnitude in the DAP killing assay after 1 hour (Fig. 3A; blue bars). We discovered that the RIF treatment was as effective at killing the VAN/DAP-tolerant strains as it was at killing the ancestral day 1 strain (Fig. 3A; red bars). Similarly, the combination of DAP and RIF (Fig. 3A; purple bars) did not kill the ancestral strain significantly more effectively than the VAN/DAP-tolerant strains.

Fig. 3 Tolerance promotes the evolution of resistance under suppressive antibiotic combination treatment [DAP (12.5 μg/ml) and RIF (1 μg/ml)].

(A) Survival after 1 hour of the wt strain (P1D1C1), tolerant strain (P1D7C1), and their RIF-resistant derivative (rpoB H481Y) (patterned fill) under treatments. The P value for analysis of variance (ANOVA) F test of the interaction effect of RIF resistance in wt or tolerant background under DAP and RIF in combination is 4.9 × 10−6. P values for the pairwise comparison were estimated with Student’s t test. Data are presented as the means ± SD from at least three biological replicates. (B) In vitro evolution experiments in wt or its DAP tolerant mutant under intermittent combination treatment of DAP and RIF. RIF resistance evolved repeatedly in tolerant (four of five experiments) but not in wt background (zero of five experiments), P = 0.048 with Fisher’s exact test. (C) Competition experiments for RIF sensitive (empty bars) and resistant mutant (patterned fill). Around 103 RIF-resistant derivatives (rpoB H481Y) were mixed with 106 of their parental strains (wt, P1D1C1; and tolerant, P1D7C1). The mixed population were killed with combination treatment DAP (12.5 μg/ml) and RIF (1 μg/ml) for 1 hour then regrown overnight. The RIF-resistant population (patterned fill) went extinct in the wt background but survived in the tolerance background. Asterisk indicates below detection limit (<10 CFU/ml). (D) Rate of evolution of treatment failure (RIF resistance or DAP tolerance fixation) under different treatments and for different background (wt, P1D1C1; or tolerant, P1D7C1). This rate is estimated by the number of treatment and regrowth cycles (K) required for the mutation to fix (see supplementary materials).

To understand whether the evolution of RIF resistance in the VAN/DAP-tolerant strains can be reproduced in vitro, we performed evolution experiments with the ancestral isolate and its in-host–evolved, VAN/DAP-tolerant mutant, using the DAP and RIF combination treatment under which RIF resistance emerged. Within a few treatment cycles, in four out of five experiments, we observed selection for RIF resistance mutations in the VAN/DAP-tolerant background, reproducing the evolutionary trajectory observed in the patient. None of the five experiments led to selection of RIF resistance in the ancestral isolate (Fig. 3B; P = 0.048). This result led to two questions: (i) If the combination treatment leads to RIF resistance so quickly, why is DAP and RIF used in combination in the clinic? (ii) Why did RIF resistance mutations become established only in the VAN/DAP--tolerant background, despite the similar survival capacities of ancestral and VAN/DAP-tolerant strains under the combination treatment?

Paradoxically, we observed a significant decrease in survival of RIF-resistant bacteria compared with the ancestral wild-type strain (wt) under the DAP and RIF combination (Fig. 3A). DAP alone kills the wt very effectively (decrease in survival by approximately five orders of magnitude after 1 hour), whereas the combination of DAP and RIF is less effective by a factor of 100. Such a drug combination, which is less effective than a single drug, is called suppressive (fig. S8) (21). A well-known example of such an interaction is the suppressive effect of chloramphenicol—which slows growth—on beta-lactam killing, which requires growth to be effective (22). Correspondingly, RIF suppresses killing by DAP, such that a mutant that is fully RIF resistant—i.e., is unaffected by the presence of RIF—will experience full unsuppressed killing by DAP. It will therefore be killed more efficiently under DAP and RIF in combination than a RIF-susceptible strain. Thus, RIF resistance will not be established in the wt background under the DAP and RIF combination. This phenomenon may answer our first question: explaining the empirical use of DAP and RIF combination in the clinic (23), in alignment with the strategy of using suppressive drug combinations for the prevention of resistance in vitro (24).

We now turn to the effect of VAN/DAP tolerance on survival under the DAP and RIF combination treatment. In contrast to the reduction in survival of RIF resistance mutation in the wt strain when exposed to combination treatment, we observed that the RIF resistance mutation increased survival in the VAN/DAP-tolerant strain (Fig. 3A). RIF-resistant, VAN/DAP-tolerant mutants survive DAP exposure because of the protective effect of VAN/DAP tolerance. Thus, VAN/DAP tolerance allows RIF resistance to evolve under the DAP and RIF combination (Fig. 3B).

To verify that the rescue of resistance mutations by VAN/DAP tolerance occurs under combination with DAP and RIF, we performed competition experiments in which a small number (~1000 bacteria) of RIF-resistant mutants were mixed with their wt ancestral strain (~106 bacteria) and exposed to DAP and RIF. Strong suppression of resistance to RIF by DAP and RIF combination treatment caused the extinction of the RIF-resistant mutants in the wt background (Fig. 3C; wt). In contrast, RIF-resistant mutants survived in the VAN/DAP-tolerant background during DAP and RIF combination treatment (Fig. 3C; tolerant, P = 0.0003).

The survival measurements (Fig. 3A) can be used to quantitatively determine how quickly treatment would fail because of the rapid fixation of a mutation (see “Calculation of the number of treatment cycles for the fixation of a mutation” in the supplementary materials). Because of the high probability of RIF resistance mutations (19), fixation to resistance occurred rapidly (after approximately one cycle) under RIF monotherapy (shown schematically in Fig. 3D). By contrast, evolution of RIF resistance in the wt background was delayed by more than 20 cycles of treatment with the DAP and RIF combination (Fig. 3D). However, for a VAN/DAP-tolerant strain, RIF resistance occurs much faster than for the wt (7 cycles versus >20 cycles) (Fig. 3D). This analysis answers our second question, explaining why VAN/DAP-tolerant strains have evolved RIF resistance rapidly in the host and in vitro. Thus, the DAP and RIF combination treatment is effective at delaying both the evolution of DAP tolerance and RIF resistance, and the treatment might have been effective in patient 1 if administered before DAP tolerance had been established.

To understand whether prior evolution of tolerance is required for the evolution of resistance under other suppressive combination treatments, we measured bacterial survival under drug combinations from four different antibiotic classes. We evaluated all six possible combinations in Escherichia coli strain KLY and its ampicillin (AMP)– and norfloxacin (NOR)–tolerant mutant KLY metGT (Fig. 4, A to D) (4). We found that four of the six combinations were suppressive in the wt strain (Fig. 4, A and C), but only two were suppressive in the tolerant strains (Fig. 4, B and D). This result indicates that some of the combinations may act similarly to the DAP and RIF combination on the evolution of RIF resistance. Thus, resistance was suppressed in the wild type but protected in a tolerant background.

Fig. 4 Tolerance promotes resistance in other suppressive combinations.

(A to D) All six possible combination treatments for antibiotics from four different classes: AMP (50 μg/ml), KAN (kanamycin, 30 μg/ml), NOR (1 μg/ml), and RIF (200 μg/ml) for [(A) and (C)] E. coli KLY (wt) and [(B) and (D)] its tolerant derivative KLY metGT. Survival was measured after 4 hours. The numbers in (C) and (D) represent the suppression factor (see fig. S8). P values in (A) and (B) were estimated with Student’s t test. (E) Survival after 4 hours of the wt strain, tolerant strain, and their NOR-resistant derivative (gyrA S83L) (patterned fill) under combination treatment of AMP and NOR. Survival data are presented as the means ± SD from at least three biological replicates. The P value for ANOVA F test of the interaction effect of NOR resistance in wt or tolerant background under AMP and NOR in combination is 2.8 × 10−6. P values for the pairwise comparison data were estimated with Student’s t test. (F) Competition experiments for sensitive (empty bars) and NOR-resistant mutant (patterned fill). Around 103 NOR-resistant mutants (gyrA S83L) were mixed with their parental strains (~107) (wt, KLY; and tolerant, KLY-metGT). The mixed populations were killed with combination treatment of AMP and NOR then regrown overnight. The NOR-resistant population went extinct in the wt background but survived in the tolerance background. Asterisk indicates below detection limit (<10 CFU/ml). Data are presented as the means ± SD from at least three biological replicates.

We tested these predictions for the AMP and NOR combination by constructing a known NOR resistance mutation (gyrA-S83L) (25), which increased MIC to NOR by more than 10-fold. As predicted, the survival of the NOR-resistant E. coli under combination treatment was suppressed in the wt background but not in the tolerant strain (Fig. 4E), similarly to the DAP and RIF combination for the S. aureus clinical strains (Fig. 3A). We repeated the assay for the rescue of resistance mutations by tolerance under the combination NOR and AMP and found that tolerance could rescue NOR resistance mutations that would otherwise go extinct in the wt background (Fig. 4F; P = 0.0003). We conclude that rescue of resistance mutations by tolerance is a general phenomenon that may have crucial implications for the evolution of resistance in patients treated with combinations of antimicrobials.

Our analysis of strains isolated from blood infections revealed a notable similarity to in vitro experiments. In both, the rapid evolution of tolerance was identified as a major survival factor, followed by the emergence of resistance. However, in contrast to the in vitro evolution results (where a single drug was used), our study shows how tolerance promotes the evolution of resistance under combination treatments that are expected to prevent resistance.

Despite the drawbacks of using suppressive combination treatment, this strategy can be effective at preventing the evolution of resistance (21, 24) and may explain the motivation for the empirical use of such treatments in patients (23, 26). But it does need to be deployed before tolerance to those drugs has established. Diagnostic tools for detecting the emergence of tolerance under treatment would provide crucial information for guiding combination treatments (11, 27). Moreover, combination treatments such as DAP and RIF may be effective at preventing the evolution of tolerance, which has been shown to underlie difficult-to-treat infections in immunocompromised patients (12). Designing combination therapies that account for tolerance or persistence (28) may be especially relevant for the treatment of tuberculosis, where tolerance has been suggested to be a major factor for survival of the pathogen (29) and where the prevention of the de novo evolution of resistance in patients is crucial (30). The generality of the mechanism by which tolerance promotes resistance suggests that it is relevant not only to de novo evolution of resistance by mutations (31, 32) but also to other mechanisms for acquiring resistance, such as horizontal gene transfer. Our study could also apply to the recognition of the effect of drug tolerance on the outcome of anticancer combination treatments (3335).

Supplementary Materials

science.sciencemag.org/content/367/6474/200/suppl/DC1

Materials and Methods

Supplementary Text

Figs. S1 to S8

Tables S1 to S3

References (3648)

References and Notes

Acknowledgments: We thank A. O’Neill for providing the strains 8325-4 and derivatives R23 (rpoB S486L) and R35 (rpoB S464P) and J. Strahilevitch, M. V. Assous, I. Levin-Reisman, A. Brauner, S. P. Mizrahi, and O. Fridman for discussions and suggestions. Ethics approval of this work was granted by the Shaare Zedek Medical Center ethics committee (#0277-17-SZMC). Funding: The work was supported by the European Research Council (Consolidator grant no. 681819), the Israel Science Foundation (grant no. 492/15), and the Minerva Foundation. J.L. acknowledges support from the HUJI-CSC scholarship program. Author contributions: Conceptualization, J.L., O.G., and N.Q.B.; Methodology, J.L., O.G., and N.Q.B.; Investigation, J.L. and M.B.-M.; Writing – original draft, J.L. and N.Q.B.; Writing – review & editing, J.L., O.G., and N.Q.B.; Funding acquisition, N.Q.B.; Resources, M.B.-M. and I.R.; Supervision, N.Q.B. Competing interests: N.Q.B. and O.G. submitted U.S. patent application 62/244,809, which covers the fabrication of the TDtest technique utilized in this paper (11). I.R., J.L., and M.B.-M. declare no competing interests. Data and materials availability: The accession number for the sequencing data is National Center for Biotechnology Information BioProject PRJNA503808. Code for ScanLag is available at http://bio-site.phys.huji.ac.il/Materials. The clinical strains are available under a material transfer agreement from MBM under a material agreement with Shaare Zedek Hospital.

Stay Connected to Science

Navigate This Article