Overuse of antibiotic drugs is leading to the appearance of antibiotic-resistant bacteria; this is, bacteria with mutations that allow them to survive treatment with specific antibiotics. This has made some bacterial infections difficult or impossible to treat. Learning more about how bacteria evolve resistance to antibiotics could help scientists find ways to prevent it and develop more effective treatments.

Changing antibiotics frequently may be one way to prevent bacteria from evolving resistance. That way if a bacterium acquires mutations that allow it to escape one antibiotic, another antibiotic will kill it, stopping it from dividing and preventing the appearance of descendants with resistance to several antibiotics. In order to use this approach, testing is needed to find the best sequences of antibiotics to apply and the optimal timings of treatment.

To find out more, Batra, Roemhild et al. grew bacteria in the laboratory and exposed them to different sequences of antibiotics, switching antibiotics at different time intervals. This showed that sequential treatments with different antibiotics can limit bacterial evolution, especially when antibiotics are switched quickly. Unexpectedly, one of the most effective sequences used very similar antibiotics. This was surprising because using similar antibiotics should lead to the evolution of cross-resistance, which is when a drug causes changes that make the bacterium less sensitive to other treatments. However, in the tested case, cross-resistance did not evolve when antibiotics were switched quickly, thereby ensuring efficiency of treatment.

Batra et al. show that alternating sequences of antibiotics may be an effective strategy to prevent drug resistance. Because the experiments were done in a laboratory setting it will be important to verify the results in studies in animals and humans before the approach can be used in medical or veterinary settings. If the results are confirmed, it could reduce the need to develop new antibiotics, which is expensive and time consuming.

The efficacy of antibiotics for the treatment of infections is diminishing rapidly as bacteria evolve new mechanisms to resist antibiotics (Laxminarayan et al., 2013). Resistance evolution is frequently observed during antibiotic therapy and can happen within days (Bloemberg et al., 2015; Hjort et al., 2020; Tueffers et al., 2019). A failure to account for such rapid bacterial adaptation is likely a common reason for treatment failure (Woods and Read, 2015; Zhou et al., 2020). For this reason, the field of evolutionary medicine specifically accounts for bacterial evolvability and seeks treatment solutions that are hard to overcome by genetic adaptation (Andersson et al., 2020; Merker et al., 2020). While an evolution-proof antibiotic remains to be found, the mechanisms that restrict evolutionary escape are starting to be revealed (Bell and MacLean, 2018). Such evolutionary insight may guide the design of effective and sustainable antibiotic therapy.

An effective way of reducing the amount of evolutionary solutions is to administer several antibiotics either simultaneously (i.e., combination therapy) or sequentially (i.e., sequential therapy). Tailored combination treatments make use of physiological and evolutionary constraints (Baym et al., 2016). The emergence of resistance is delayed by combinations, when evolutionary escape requires multiple mutations and when drug interactions eliminate the intermediate genetic steps of single-drug resistance (Chait et al., 2007), antibiotic tolerance (Levin-Reisman et al., 2017), and heteroresistance (Band et al., 2019). However, when genetic resistance to the combination is easily accessible, for example, through gene amplification of efflux pumps, then combination therapy can accelerate resistance emergence (Pena-Miller et al., 2013). This undesired selective effect is potentially avoided by sequential drug application. Evolutionary escape from sequential treatments is constrained by negative hysteresis responses induced by specific antibiotics (Roemhild et al., 2018) and/or the emergence of genetic collateral sensitivity trade-offs (Barbosa et al., 2019; Yoshida et al., 2017). Negative hysteresis occurs when exposure to an antibiotic induces changes to bacterial physiology that transiently increase the killing efficacy of other antibiotics (Roemhild et al., 2018). Collateral sensitivity is a genetic side effect of evolved resistance that too increases the efficacy of other antibiotics (Szybalski and Bryson, 1952). Collateral sensitivity is prevalent among pathogens and occurs especially between antibiotics with distinct mechanism of action (i.e., heterogeneous sets of antibiotics), while cross-resistance often emerges towards antibiotics with similar mode of action (i.e., homogeneous sets of antibiotics) (Barbosa et al., 2017; Imamovic and Sommer, 2013; Lázár et al., 2013; Maltas and Wood, 2019). Thus, conventionally, multidrug treatments would avoid antibiotics from similar classes, with the rationale of limiting the overlap in the respective sets of resistance mutations, and thus the ensuing cross-resistance.

The particular efficacy of sequential therapy has been confirmed with the help of evolution experiments under controlled laboratory conditions. Different types of sequential treatments have been tested. Some regimens involved a single switch between antibiotics, while others included multiple switches at short time intervals. One of the main findings was that the efficacy of sequential treatments depended both on the included antibiotics and the particular treatment sequence (Fuentes-Hernandez et al., 2015; Maltas and Wood, 2019; Roemhild et al., 2015). While fast sequential treatments did not exclude the eventual emergence of multidrug resistance, many significantly delayed bacterial adaptation compared to monotherapy (Kim et al., 2014; Roemhild et al., 2015; Yoshida et al., 2017). A single antibiotic switch can also delay adaptation, dependent on the drug order, and it can additionally reverse previous resistance and resensitize bacterial populations to specific antibiotics (Barbosa et al., 2019; Hernando-Amado et al., 2020; Imamovic and Sommer, 2013; Yen and Papin, 2017). Moreover, our group previously demonstrated that fast sequential treatments with a heterogeneous set of three antibiotics – the fluoroquinolone ciprofloxacin (CIP), the β-lactam carbenicillin (CAR), and the aminoglycoside gentamicin (GEN) – delayed the emergence of multidrug resistance in the pathogen Pseudomonas aeruginosa (Roemhild et al., 2018). The observed inhibition of evolutionary escape was manifested by the occurrence of population extinction, although antibiotic concentrations were below the minimal inhibitory concentration (MIC). We further found that negative hysteresis at antibiotic switches reduced adaptation rates because it selected for distinct genetic changes. Several populations adapted to fast sequential treatment by independent mutations in the histidine kinase cpxS that only mildly increased resistance, thereby explaining the low rate of adaptation to the used antibiotics. Instead, the cpxS mutations suppressed negative hysteresis, demonstrating that adaptation was specific to the selective constraint imposed by the drug switches. Based on these findings, we assumed that the acting selective dynamics were ultimately a consequence of antibiotic heterogeneity. However, is this so? Do selective dynamics differ for a homogenous set of drugs?

The primary aim of our current study was to assess the efficacy of sequential treatments with either heterogeneous or homogeneous sets of three antibiotics. We focused on P. aeruginosa strain PA14 as a tractable pathogen model system, for which comprehensive experimental reference data is available on resistance evolution (e.g., Barbosa et al., 2019; Barbosa et al., 2018; Hernando-Amado et al., 2020; Roemhild et al., 2018; Sanz-García et al., 2018; Yen and Papin, 2017). We performed similar evolution experiments as before, with three new sets of bactericidal antibiotics, two of which included only β-lactams, and one the three previously considered modes of action (Figure 1—figure supplement 1A). The new heterogeneous drug set CIP, streptomycin (STR), and doripenem (DOR) involved drug synergy and was expected to contribute to collateral sensitivity (Barbosa et al., 2018; Barbosa et al., 2017). The drug sets comprising three β-lactams, however, had all properties that would typically be avoided for the design of multidrug treatments. The three β-lactams CAR, cefsulodin (CEF), and DOR have the same core structure and individually inhibit the DD-transpeptidase activity in cell-wall synthesis (Walsh, 2003). The collateral effects landscape between CAR-CEF-DOR was expected to be dominated by cross-resistance (Barbosa et al., 2017) and the three antibiotics showed neither synergy nor antagonism (Barbosa et al., 2018). Resistance to these antibiotics may potentially be achieved through single mutations. The situation is replicated by the set of ticarcillin (TIC), azlocillin (AZL), and ceftazidime (CEZ). In contrast to expectations, the triple β-lactam sequences showed high treatment potency. Therefore, the secondary aim of our study was to assess which characteristics constrained the ability of the bacteria to adapt to the β-lactam sequential treatments. We focused on one triple β-lactam set (CAR-CEF-DOR) and specifically tested the influence of antibiotic switching rate, switching regularity, negative hysteresis, the potential for spontaneous resistance evolution, and resulting cross-resistances on treatment efficacy.

Treatment with multiple β-lactam antibiotics is generally avoided due to the perceived fear of therapy failure from cross-resistance. Our work now challenges this widespread belief. We characterized the ability of replicate P. aeruginosa populations to evolve de novo resistance to sequential treatments with different drug sets. To our surprise, we found that sets of three β-lactams constrained bacterial adaptation by reducing bacterial survival. We demonstrate that treatment potency was determined by variation in the spontaneous rate of resistance to the β-lactams and the resulting collateral effects across sequential treatment protocols.

Our initial screen of sequential protocols with different antibiotic triplets revealed that the triple β-lactam sequences are at least as effective at causing extinction as sequences of antibiotics with distinct modes of actions. This finding is at first sight counterintuitive, but at second sight not completely unexpected. The joint application of two β-lactam drugs was in fact tested and found effective in a few previous studies (Rahme et al., 2014). For example, the β-lactam aztreonam was shown to interact synergistically with four other β-lactam drugs against multiple resistant isolates of Enterobacteriaceae and P. aeruginosa in vitro (Buesing and Jorgensen, 1984). A combination of ticarcillin with ceftazidime produced high efficacy in a rat peritonitis model (Shyu et al., 1987). In a treatment of bacterial soft tissue infections, the combination of cefotaxime and mecillinam led to higher clinical response rates than the tested monotherapy (File and Tan, 1983). Further, the dual β-lactam combination of ceftazidime plus piperacillin was as effective as the combination of ceftazidime and tobramycin in granulocytopenic cancer patients (Joshi et al., 1993). More recent studies demonstrated that a triple combination of meropenem, piperacillin, and tazobactam successfully constrained resistance evolution in Methicillin-resistant Staphylococcus aureus (MRSA), both in vitro and in a mouse model (Gonzales et al., 2015). In addition, the combination of cefotaxime and mecillinam was effective against Salmonella enterica harboring a mutant β-lactamase in a mouse model (Rosenkilde et al., 2019). Our findings add to the high potency of treatments with multiple β-lactams. We conclude that the use of multiple β-lactams, either as a combination or sequentially, is a commonly underappreciated form of therapy and its use opens new avenues to better utilize our existing antibiotic armamentarium.

Spontaneous rate of antibiotic resistance was found to play a critical role in the success of the CAR-CEF-DOR sequential treatment. The probability of spontaneous resistance on all three β-lactams was significantly different, with the rate of DOR resistance being the lowest. These rates determined the overall probability of acquiring direct resistance in treatment, which significantly correlated with the frequency of population extinction (Figure 5B). Resistance rates were previously shown to vary towards different antibiotics, for example, in Escherichia coli (El Meouche and Dunlop, 2018) and P. aeruginosa (Oliver et al., 2004). This variation can arise from genetic factors such as mutational target space and physiological factors like activation of the bacterial SOS response (Martinez and Baquero, 2000). Such information on resistance rates has so far been used for predicting the occurrence of resistance against single drugs, and antibiotics that target multiple pathways in a cell are considered advantageous in this context (Ross-Gillespie and Kümmerli, 2014). One example of the latter are compounds against S. aureus that inhibit both DNA gyrase and topoisomerase IV (Nyerges et al., 2020). The rate of resistance emergence may also be reduced by using adjuvants that target the SOS response (Bell and MacLean, 2018), as previously shown for compounds interfering with LexA activity leading to reduced resistance rates to ciprofloxacin and rifampicin in E. coli (Cirz et al., 2005). Our study extends the role of resistance rates of antibiotics beyond this convention. We show that inclusion of an antibiotic with relatively low spontaneous resistance emergence can enhance the potency of a sequential treatment design.

What could be the underlying reasons for the particular importance of DOR compared with the other β-lactams? DOR belongs to the carbapenem subclass of the β-lactam antibiotics. Carbapenems possess broad activity against Gram-positive and Gram-negative bacteria (Papp-Wallace et al., 2011) and are active against many β-lactamase-producing microbes since their thiazolidinic ring makes them relatively resistant to β-lactamase-mediated hydrolysis (Schafer et al., 2009). In contrast, the penicillin CAR is active mostly (albeit not exclusively) against Gram-negative bacteria (Castle, 2007) while the activity of the cephalosporin CEF is restricted to P. aeruginosa (Wright, 1986). Within P. aeruginosa, all three antibiotics show high potency against a large variety of clinical isolates (Castanheira et al., 2009; Neu and Scully, 1984; Traub and Raymond, 1970). Resistance rates for β-lactam antibiotics were assessed with different approaches across P. aeruginosa strains and clinical isolates, consistently showing that DOR has a particularly low propensity to select for resistance mutations, even when compared to other carbapenems (Barbosa et al., 2017; Barbosa et al., 2021; Sakyo et al., 2006; Tanimoto et al., 2008; Mushtaq et al., 2004; Fujimura et al., 2009). Therefore, the phenotype of reduced spontaneous resistance to DOR appears to be robustly expressed across different P. aeruginosa genotypes and does not extend to other carbapenems or β-lactams. One possible reason for this pattern may be variation in the range of β-lactam target proteins, in this case the penicillin binding proteins (PBPs), and where DOR is known to bind more of these PBPs than do CAR or CEF (Davies et al., 2008; Fontana et al., 2000; Rodriguez‐Tebár et al., 1982; Rodríguez-Tebar et al., 1982; Zapun et al., 2008; Zimmermann, 1980). Thus, target resistance to DOR would likely require a larger number of mutations than that to other β-lactams. Interestingly, another carbapenem, meropenem, targets the same PBPs as DOR (Davies et al., 2008) but has a higher resistance rate, suggesting that the underlying reasons for resistance rate variation are multifactorial. Taken together, effective resistance mutations against DOR seem to be less commonly available in P. aeruginosa in comparison to that against other drugs, including the here used CEF and CAR.

A key determinant of treatment potency was the reduced level of spontaneous cross-resistance to the sequentially applied drugs (Figure 5B). This effect was maximized by the switching rate (Figure 5—figure supplement 1). Our findings are consistent with the previously and repeatedly proposed importance of collateral sensitivity for the efficacy of sequential treatment protocols (Barbosa et al., 2019; Hernando-Amado et al., 2020; Imamovic and Sommer, 2013; Kim et al., 2014; Maltas and Wood, 2019; Yen and Papin, 2017). Even though we did not measure collateral sensitivity directly, the lack of cross-resistance is related as it indicates that the mutant cells, which have become resistant to one drug, maintain at least ancestral levels of susceptibility against the second drug. Moreover, our study focused on spontaneous emergence of cross-resistance (or lack thereof). By contrast, many previous studies established collateral effects after bacteria evolved resistance to the first drug over many generations, often followed by only a single antibiotic switch to assess the impact of collateral sensitivity on therapy success (Barbosa et al., 2019; Hernando-Amado et al., 2020; Imamovic and Sommer, 2013; Yen and Papin, 2017). Surprisingly, our study revealed potentially beneficial collateral effects between antibiotics of the same class. In fact, we chose these three β-lactams because our previous work demonstrated cross-resistance between most of them, although inferred upon multigenerational adaptation to the first drug (Barbosa et al., 2017). Our current finding of a lack of cross-resistance among some of these drugs now suggests that spontaneous mutants may have different collateral profiles than the lines, which adapted over many generations. Our results further suggest that the collateral effects of spontaneous mutants are likely to be more pertinent for the design of sequential treatments with fast switches among antibiotics. This suggestion is supported by two previous studies, in which the efficacy of fast sequential treatments was optimized by considering collateral effects for either single-step mutants of S. aureus, obtained after 20 hr exposure to three distinct antibiotics for 20 hr (Kim et al., 2014), or from Enterococcus faecalis populations adapted over 2 days to four distinct antibiotics (Maltas and Wood, 2019). As a side note, it is particularly interesting that our detailed resistance analysis consistently revealed almost all treatments to cause the evolution of collateral sensitivity towards the aminoglycoside gentamicin, but not the fluoroquinolone ciprofloxacin (Figure 2B, C), possibly indicating yet another treatment option – in cases where the applied triple β-lactam sequential protocols fail.

Temporal irregularity was additionally found to constrain bacterial adaptation. When bacteria experienced the antibiotics in an irregular pattern, this caused significantly increased extinction and to some degree reduced multidrug resistance. With CAR-DOR-CEF, the lowest multidrug resistance was observed in random sequential treatments (Figure 2—figure supplement 5), as also previously observed with CIP-GEN-CAR (Roemhild et al., 2018). Environmental change anticipation has been documented in several microorganisms (Mitchell et al., 2009; Mitchell and Pilpel, 2011), indicating their capability to specifically adapt to regular environmental change. Stochastic changes can make it harder to evolve anticipation (Roemhild and Schulenburg, 2019). Stochastic changes in environmental parameters were indeed found to constrain fitness in evolving bacteria (Hughes et al., 2007) and viruses (Alto et al., 2013). We show that irregular antibiotic sequences have potential to inhibit bacterial resistance evolution.

Unexpectedly, we further identified negative hysteresis for multiple combinations of the three β-lactams. However, cumulative hysteresis levels per treatment did not significantly associate with any of our measured evolutionary responses. In our previous study (Roemhild et al., 2018), within the CAR-CIP-GEN combination, negative hysteresis was expressed for the switches from CAR to GEN and CIP to GEN. Yet, only the CAR-GEN hysteresis was significantly associated to the evolutionary responses. Thus, hysteresis interactions can exist between antibiotics from the same or different classes, but they need not impact the evolutionary outcome of a sequential treatment protocol each time. In the current study, it appears that spontaneous resistance effects and the resulting cross-resistance effects are dominant over the β-lactam hysteresis. One potential explanation could be that insensitivity to β-lactam hysteresis evolves quickly. Nevertheless, it clearly warrants further research to assess whether negative hysteresis between the β-lactam drugs is robustly shown across strains of P. aeruginosa or other bacterial species and can somehow be exploited in sequential therapy, in analogy to the previous results with antibiotics from different classes (Roemhild et al., 2018).

Taken together, our study highlights that the available antibiotics offer unexplored, highly potent treatment options that can be harnessed to counter the spread of drug resistance. It further underscores the importance of evolutionary trade-offs such as reduced cross-resistance in treatment design and introduces spontaneous resistance rates of component antibiotics as a guiding principle for sequential treatments. It is ironic that the differential cross-resistance landscape of the β-lactams was a key factor contributing to treatment potency, even though the risk of cross-resistance is usually used to reject β-lactam-exclusive treatments. The underlying reasons for differential spontaneous and long-term cross-resistance between these drugs (including the underlying molecular mechanisms) are as yet unknown and clearly deserve further attention in the future. We conclude that a detailed understanding of both spontaneous resistance rates and resulting cross-resistances against different antibiotics should be of particular value to further improve the potency of sequential protocols.

Sequencing data have been deposited at NCBI under the BioProject number: PRJNA704789. All other data is provided in the supplementary source data files.

1. 1. Batra A

   (2021) NCBI BioProject

   ID PRJNA704789. Sequential beta-lactam treatment genomics.

   https://www.ncbi.nlm.nih.gov/bioproject/PRJNA704789

1. 1. Alto BW
   2. Wasik BR
   3. Morales NM
   4. Turner PE

   (2013) Stochastic temperatures impede rna virus adaptation

   Evolution 67:969–979.

   https://doi.org/10.1111/evo.12034

   • Google Scholar
2. 1. Andersson DI
   2. Balaban NQ
   3. Baquero F
   4. Courvalin P
   5. Glaser P
   6. Gophna U
   7. Kishony R
   8. Molin S
   9. Tønjum T

   (2020) Antibiotic resistance: turning evolutionary principles into clinical reality

   FEMS Microbiology Reviews 44:171–188.

   https://doi.org/10.1093/femsre/fuaa001

   • Google Scholar
3. Software

   1. Andrews S

   (2010) FastQC: A Quality Control Tool for High Throughput Sequence Data, version 0.11.9

   Babraham Bioinformatics.

   https://www.bioinformatics.babraham.ac.uk/projects/fastqc/
4. 1. Band VI
   2. Hufnagel DA
   3. Jaggavarapu S
   4. Sherman EX
   5. Wozniak JE
   6. Satola SW
   7. Farley MM
   8. Jacob JT
   9. Burd EM
   10. Weiss DS

   (2019) Antibiotic combinations that exploit heteroresistance to multiple drugs effectively control infection

   Nature Microbiology 4:1627–1635.

   https://doi.org/10.1038/s41564-019-0480-z

   • PubMed
   • Google Scholar
5. 1. Barbosa C
   2. Trebosc V
   3. Kemmer C
   4. Rosenstiel P
   5. Beardmore R
   6. Schulenburg H
   7. Jansen G

   (2017) Alternative evolutionary paths to bacterial antibiotic resistance cause distinct collateral effects

   Molecular Biology and Evolution 34:2229–2244.

   https://doi.org/10.1093/molbev/msx158

   • PubMed
   • Google Scholar
6. 1. Barbosa C
   2. Beardmore R
   3. Schulenburg H
   4. Jansen G

   (2018) Antibiotic combination efficacy (ACE) networks for a Pseudomonas aeruginosa model

   PLOS Biology 16:e2004356.

   https://doi.org/10.1371/journal.pbio.2004356

   • Google Scholar
7. 1. Barbosa C
   2. Römhild R
   3. Rosenstiel P
   4. Schulenburg H

   (2019) Evolutionary stability of collateral sensitivity to antibiotics in the model pathogen Pseudomonas aeruginosa

   eLife 8:e51481.

   https://doi.org/10.7554/eLife.51481

   • Google Scholar
8. 1. Barbosa C
   2. Mahrt N
   3. Bunk J
   4. Graßer M
   5. Rosenstiel P
   6. Jansen G
   7. Schulenburg H

   (2021) The genomic basis of rapid adaptation to antibiotic combination therapy in Pseudomonas aeruginosa

   Molecular Biology and Evolution 38:449–464.

   https://doi.org/10.1093/molbev/msaa233

   • PubMed
   • Google Scholar
9. 1. Baym M
   2. Stone LK
   3. Kishony R

   (2016) Multidrug evolutionary strategies to reverse antibiotic resistance

   Science 351:aad3292.

   https://doi.org/10.1126/science.aad3292

   • Google Scholar
10. 1. Bell G
    2. MacLean C

    (2018) The search for 'Evolution-Proof' Antibiotics

    Trends in Microbiology 26:471–483.

    https://doi.org/10.1016/j.tim.2017.11.005

    • PubMed
    • Google Scholar
11. 1. Bloemberg GV
    2. Keller PM
    3. Stucki D
    4. Trauner A
    5. Borrell S
    6. Latshang T
    7. Coscolla M
    8. Rothe T
    9. Hömke R
    10. Ritter C
    11. Feldmann J
    12. Schulthess B
    13. Gagneux S
    14. Böttger EC

    (2015) Acquired resistance to bedaquiline and delamanid in therapy for tuberculosis

    New England Journal of Medicine 373:1986–1988.

    https://doi.org/10.1056/NEJMc1505196

    • Google Scholar
12. 1. Bolger AM
    2. Lohse M
    3. Usadel B

    (2014) Trimmomatic: a flexible trimmer for Illumina sequence data

    Bioinformatics 30:2114–2120.

    https://doi.org/10.1093/bioinformatics/btu170

    • Google Scholar
13. 1. Buesing MA
    2. Jorgensen JH

    (1984) In vitro activity of aztreonam in combination with newer beta-lactams and amikacin against multiply resistant gram-negative bacilli

    Antimicrobial Agents and Chemotherapy 25:283–285.

    https://doi.org/10.1128/AAC.25.2.283

    • Google Scholar
14. 1. Cabot G
    2. Florit-Mendoza L
    3. Sánchez-Diener I
    4. Zamorano L
    5. Oliver A

    (2018) Deciphering β-lactamase-independent β-lactam resistance evolution trajectories in Pseudomonas aeruginosa

    The Journal of Antimicrobial Chemotherapy 73:3322–3331.

    https://doi.org/10.1093/jac/dky364

    • PubMed
    • Google Scholar
15. 1. Castanheira M
    2. Jones RN
    3. Livermore DM

    (2009) Antimicrobial activities of doripenem and other carbapenems against Pseudomonas aeruginosa, other nonfermentative bacilli, and Aeromonas spp

    Diagnostic Microbiology and Infectious Disease 63:426–433.

    https://doi.org/10.1016/j.diagmicrobio.2009.01.026

    • Google Scholar
16. Book

    1. Castle SS

    (2007) Carbenicillin

    In: Enna S. J, Bylund D. B, editors. XPharm: The Comprehensive Pharmacology Reference. New York: Elsevier. pp. 1–5.

    https://doi.org/10.1016/B978-008055232-3.61381-9

    • Google Scholar
17. 1. Chait R
    2. Craney A
    3. Kishony R

    (2007) Antibiotic interactions that select against resistance

    Nature 446:668–671.

    https://doi.org/10.1038/nature05685

    • PubMed
    • Google Scholar
18. 1. Cingolani P
    2. Platts A
    3. Wang leL
    4. Coon M
    5. Nguyen T
    6. Wang L
    7. Land SJ
    8. Lu X
    9. Ruden DM

    (2012) A program for annotating and predicting the effects of single Nucleotide Polymorphisms, SnpEff: snps in the genome of Drosophila Melanogaster strain w1118; iso-2; iso-3

    Fly 6:80–92.

    https://doi.org/10.4161/fly.19695

    • PubMed
    • Google Scholar
19. 1. Cirz RT
    2. Chin JK
    3. Andes DR
    4. de Crécy-Lagard V
    5. Craig WA
    6. Romesberg FE

    (2005) Inhibition of mutation and combating the evolution of antibiotic resistance

    PLOS Biology 3:e176.

    https://doi.org/10.1371/journal.pbio.0030176

    • PubMed
    • Google Scholar
20. 1. Davies TA
    2. Shang W
    3. Bush K
    4. Flamm RK

    (2008) Affinity of Doripenem and Comparators to Penicillin-Binding Proteins in Escherichia coli and Pseudomonas aeruginosa

    Antimicrobial Agents and Chemotherapy 52:1510–1512.

    https://doi.org/10.1128/AAC.01529-07

    • Google Scholar
21. 1. El Meouche I
    2. Dunlop MJ

    (2018) Heterogeneity in efflux pump expression predisposes antibiotic-resistant cells to mutation

    Science 362:686–690.

    https://doi.org/10.1126/science.aar7981

    • PubMed
    • Google Scholar
22. 1. File TM
    2. Tan JS

    (1983) Amdinocillin plus cefoxitin versus cefoxitin alone in therapy of mixed soft tissue infections (including diabetic foot infections)

    The American Journal of Medicine 75:100–105.

    https://doi.org/10.1016/0002-9343(83)90103-1

    • Google Scholar
23. 1. Fontana R
    2. Cornaglia G
    3. Ligozzi M
    4. Mazzariol A

    (2000) The final goal: penicillin-binding proteins and the target of cephalosporins

    Clinical Microbiology and Infection 6:34–40.

    https://doi.org/10.1111/j.1469-0691.2000.tb02038.x

    • Google Scholar
24. 1. Fuentes-Hernandez A
    2. Plucain J
    3. Gori F
    4. Pena-Miller R
    5. Reding C
    6. Jansen G
    7. Schulenburg H
    8. Gudelj I
    9. Beardmore R

    (2015) Using a sequential regimen to eliminate Bacteria at sublethal antibiotic dosages

    PLOS Biology 13:e1002104.

    https://doi.org/10.1371/journal.pbio.1002104

    • Google Scholar
25. 1. Fujimura T
    2. Anan N
    3. Sugimori G
    4. Watanabe T
    5. Jinushi Y
    6. Yoshida I
    7. Yamano Y

    (2009) Susceptibility of Pseudomonas aeruginosa clinical isolates in Japan to doripenem and other antipseudomonal agents

    International Journal of Antimicrobial Agents 34:523–528.

    https://doi.org/10.1016/j.ijantimicag.2009.07.008

    • Google Scholar
26. 1. Gonzales PR
    2. Pesesky MW
    3. Bouley R
    4. Ballard A
    5. Biddy BA
    6. Suckow MA
    7. Wolter WR
    8. Schroeder VA
    9. Burnham C-AD
    10. Mobashery S
    11. Chang M
    12. Dantas G

    (2015) Synergistic, collaterally sensitive β-lactam combinations suppress resistance in MRSA

    Nature Chemical Biology 11:855–861.

    https://doi.org/10.1038/nchembio.1911

    • Google Scholar
27. 1. Hauser AR
    2. Kang PJ
    3. Engel JN

    (1998) PepA, a secreted protein of Pseudomonas aeruginosa, is necessary for cytotoxicity and virulence

    Molecular Microbiology 27:807–818.

    https://doi.org/10.1046/j.1365-2958.1998.00727.x

    • Google Scholar
28. 1. Hernando-Amado S
    2. Sanz-García F
    3. Martínez JL

    (2020) Rapid and robust evolution of collateral sensitivity in Pseudomonas aeruginosa antibiotic-resistant mutants

    Science Advances 6:eaba5493.

    https://doi.org/10.1126/sciadv.aba5493

    • Google Scholar
29. 1. Hjort K
    2. Jurén P
    3. Toro JC
    4. Hoffner S
    5. Andersson DI
    6. Sandegren L

    (2020) Dynamics of Extensive Drug Resistance Evolution of Mycobacterium tuberculosis in a Single Patient During 9 Years of Disease and Treatment

    The Journal of Infectious Diseases 5:jiaa625.

    https://doi.org/10.1093/infdis/jiaa625

    • Google Scholar
30. 1. Hughes BS
    2. Cullum AJ
    3. Bennett AF

    (2007) An experimental evolutionary study on adaptation to temporally fluctuating pH in Escherichia coli

    Physiological and Biochemical Zoology 80:406–421.

    https://doi.org/10.1086/518353

    • PubMed
    • Google Scholar
31. 1. Imamovic L
    2. Sommer MOA

    (2013) Use of collateral sensitivity networks to design drug cycling protocols that avoid resistance development

    Science Translational Medicine 5:204ra132.

    https://doi.org/10.1126/scitranslmed.3006609

    • Google Scholar
32. 1. Joshi JH
    2. Newman KA
    3. Brown BW
    4. Finley RS
    5. Ruxer RL
    6. Moody MA
    7. Schimpff SC

    (1993) Double ?-lactam regimen compared to an aminoglycoside/?-lactam regimen as empiric antibiotic therapy for febrile granulocytopenic cancer patients

    Supportive Care in Cancer 1:186–194.

    https://doi.org/10.1007/BF00366445

    • Google Scholar
33. 1. Kim S
    2. Lieberman TD
    3. Kishony R

    (2014) Alternating antibiotic treatments constrain evolutionary paths to multidrug resistance

    PNAS 111:14494–14499.

    https://doi.org/10.1073/pnas.1409800111

    • Google Scholar
34. 1. Langmead B
    2. Salzberg SL

    (2012) Fast gapped-read alignment with Bowtie 2

    Nature Methods 9:357–359.

    https://doi.org/10.1038/nmeth.1923

    • Google Scholar
35. 1. Laxminarayan R
    2. Duse A
    3. Wattal C
    4. Zaidi AKM
    5. Wertheim HFL
    6. Sumpradit N
    7. Vlieghe E
    8. Hara GL
    9. Gould IM
    10. Goossens H
    11. Greko C
    12. So AD
    13. Bigdeli M
    14. Tomson G
    15. Woodhouse W
    16. Ombaka E
    17. Peralta AQ
    18. Qamar FN
    19. Mir F
    20. Kariuki S
    21. Bhutta ZA
    22. Coates A
    23. Bergstrom R
    24. Wright GD
    25. Brown ED
    26. Cars O

    (2013) Antibiotic resistance—the need for global solutions

    The Lancet Infectious Diseases 13:1057–1098.

    https://doi.org/10.1016/S1473-3099(13)70318-9

    • Google Scholar
36. 1. Lázár V
    2. Pal Singh G
    3. Spohn R
    4. Nagy I
    5. Horváth B
    6. Hrtyan M
    7. Busa‐Fekete R
    8. Bogos B
    9. Méhi O
    10. Csörgő B
    11. Pósfai G
    12. Fekete G
    13. Szappanos B
    14. Kégl B
    15. Papp B
    16. Pál C

    (2013) Bacterial evolution of antibiotic hypersensitivity

    Molecular Systems Biology 9:700.

    https://doi.org/10.1038/msb.2013.57

    • Google Scholar
37. 1. Levin-Reisman I
    2. Ronin I
    3. Gefen O
    4. Braniss I
    5. Shoresh N
    6. Balaban NQ

    (2017) Antibiotic tolerance facilitates the evolution of resistance

    Science 355:826–830.

    https://doi.org/10.1126/science.aaj2191

    • Google Scholar
38. 1. Li H
    2. Handsaker B
    3. Wysoker A
    4. Fennell T
    5. Ruan J
    6. Homer N
    7. Marth G
    8. Abecasis G
    9. Durbin R
    10. 1000 Genome Project Data Processing Subgroup

    (2009) The sequence alignment/Map format and SAMtools

    Bioinformatics 25:2078–2079.

    https://doi.org/10.1093/bioinformatics/btp352

    • PubMed
    • Google Scholar
39. 1. Liao X
    2. Hancock RE

    (1995) Cloning and characterization of the Pseudomonas aeruginosa pbpB gene encoding penicillin-binding protein 3

    Antimicrobial Agents and Chemotherapy 39:1871–1874.

    https://doi.org/10.1128/AAC.39.8.1871

    • Google Scholar
40. 1. Livermore DM

    (1995) beta-Lactamases in laboratory and clinical resistance

    Clinical Microbiology Reviews 8:557–584.

    https://doi.org/10.1128/CMR.8.4.557

    • Google Scholar
41. 1. Luria SE
    2. Delbrück M

    (1943) Mutations of Bacteria from virus sensitivity to virus resistance

    Genetics 28:491–511.

    https://doi.org/10.1093/genetics/28.6.491

    • PubMed
    • Google Scholar
42. 1. Maltas J
    2. Wood KB

    (2019) Pervasive and diverse collateral sensitivity profiles inform optimal strategies to limit antibiotic resistance

    PLOS Biology 17:e3000515.

    https://doi.org/10.1371/journal.pbio.3000515

    • Google Scholar
43. Software

    1. Mangiafico SS

    (2016) Summar and analysis of extension program evaluation in R, version 1.18.8

    Rutgers.

    https://scholar.google.com/citations?user=W9CxPFIAAAAJ&hl=en
44. 1. Martinez JL
    2. Baquero F

    (2000) Mutation frequencies and antibiotic resistance

    Antimicrobial Agents and Chemotherapy 44:1771–1777.

    https://doi.org/10.1128/AAC.44.7.1771-1777.2000

    • Google Scholar
45. 1. Merker M
    2. Tueffers L
    3. Vallier M
    4. Groth EE
    5. Sonnenkalb L
    6. Unterweger D
    7. Baines JF
    8. Niemann S
    9. Schulenburg H

    (2020) Evolutionary approaches to combat antibiotic resistance: opportunities and challenges for precision medicine

    Frontiers in Immunology 11:1938.

    https://doi.org/10.3389/fimmu.2020.01938

    • PubMed
    • Google Scholar
46. 1. Mitchell A
    2. Romano GH
    3. Groisman B
    4. Yona A
    5. Dekel E
    6. Kupiec M
    7. Dahan O
    8. Pilpel Y

    (2009) Adaptive prediction of environmental changes by microorganisms

    Nature 460:220–224.

    https://doi.org/10.1038/nature08112

    • Google Scholar
47. 1. Mitchell A
    2. Pilpel Y

    (2011) A mathematical model for adaptive prediction of environmental changes by microorganisms

    PNAS 108:7271–7276.

    https://doi.org/10.1073/pnas.1019754108

    • Google Scholar
48. 1. Mushtaq S
    2. Ge Y
    3. Livermore DM

    (2004) Doripenem versus Pseudomonas aeruginosa In Vitro: Activity against Characterized Isolates, Mutants, and Transconjugants and Resistance Selection Potential

    Antimicrobial Agents and Chemotherapy 48:3086–3092.

    https://doi.org/10.1128/AAC.48.8.3086-3092.2004

    • Google Scholar
49. 1. Neu HC
    2. Scully BE

    (1984) Activity of cefsulodin and other agents against Pseudomonas aeruginosa

    Clinical Infectious Diseases 6:S667–S677.

    https://doi.org/10.1093/clinids/6.Supplement_3.S667

    • Google Scholar
50. 1. Nichol D
    2. Rutter J
    3. Bryant C
    4. Hujer AM
    5. Lek S
    6. Adams MD
    7. Jeavons P
    8. Anderson ARA
    9. Bonomo RA
    10. Scott JG

    (2019) Antibiotic collateral sensitivity is contingent on the repeatability of evolution

    Nature Communications 10:334.

    https://doi.org/10.1038/s41467-018-08098-6

    • PubMed
    • Google Scholar
51. 1. Nyerges A
    2. Tomašič T
    3. Durcik M
    4. Revesz T
    5. Szili P
    6. Draskovits G
    7. Bogar F
    8. Skok Žiga
    9. Zidar N
    10. Ilaš J
    11. Zega A
    12. Kikelj D
    13. Daruka L
    14. Kintses B
    15. Vasarhelyi B
    16. Foldesi I
    17. Kata D
    18. Welin M
    19. Kimbung R
    20. Focht D
    21. Mašič LP
    22. Pal C

    (2020) Rational design of balanced dual-targeting antibiotics with limited resistance

    PLOS Biology 18:e3000819.

    https://doi.org/10.1371/journal.pbio.3000819

    • Google Scholar
52. 1. Oliver A
    2. Levin BR
    3. Juan C
    4. Baquero F
    5. Blázquez J
    6. Blázquez J

    (2004) Hypermutation and the Preexistence of Antibiotic-Resistant Pseudomonas aeruginosa Mutants: Implications for Susceptibility Testing and Treatment of Chronic Infections

    Antimicrobial Agents and Chemotherapy 48:4226–4233.

    https://doi.org/10.1128/AAC.48.11.4226-4233.2004

    • Google Scholar
53. 1. Papp-Wallace KM
    2. Endimiani A
    3. Taracila MA
    4. Bonomo RA

    (2011) Carbapenems: past, present, and future

    Antimicrobial Agents and Chemotherapy 55:4943–4960.

    https://doi.org/10.1128/AAC.00296-11

    • Google Scholar
54. 1. Pena-Miller R
    2. Laehnemann D
    3. Jansen G
    4. Fuentes-Hernandez A
    5. Rosenstiel P
    6. Schulenburg H
    7. Beardmore R

    (2013) When the most potent combination of antibiotics selects for the greatest bacterial load: the Smile-Frown transition

    PLOS Biology 11:e1001540.

    https://doi.org/10.1371/journal.pbio.1001540

    • Google Scholar
55. Preprint

    1. Poplin R
    2. Ruano-Rubio V
    3. DePristo MA
    4. Fennell TJ
    5. Carneiro MO
    6. der A
    7. Kling DE
    8. Gauthier LD
    9. Levy-Moonshine A
    10. Roazen D
    11. Shakir K
    12. Thibault J
    13. Chandran S
    14. Whelan C
    15. Lek M
    16. Gabriel S
    17. Daly MJ
    18. Neale B
    19. MacArthur DG
    20. Banks E

    (2018) Scaling accurate genetic variant discovery to tens of thousands of samples

    bioRxiv.

    https://doi.org/10.1101/201178

    • Google Scholar
56. Software

    1. R Development Core Team

    (2020) R: A language and environment for statistical computing, version 2.6.2

    R Foundation for Statistical Computing, Vienna, Austria.

    https://www.R-project.org/
57. 1. Rahme L
    2. Stevens E
    3. Wolfort S
    4. Shao J
    5. Tompkins R
    6. Ausubel F

    (1995) Common virulence factors for bacterial pathogenicity in plants and animals

    Science 268:1899–1902.

    https://doi.org/10.1126/science.7604262

    • Google Scholar
58. 1. Rahme C
    2. Butterfield JM
    3. Nicasio AM
    4. Lodise TP

    (2014) Dual beta-lactam therapy for serious Gram-negative infections: is it time to revisit?

    Diagnostic Microbiology and Infectious Disease 80:239–259.

    https://doi.org/10.1016/j.diagmicrobio.2014.07.007

    • PubMed
    • Google Scholar
59. 1. Ritz C
    2. Baty F
    3. Streibig JC
    4. Gerhard D

    (2015) Dose-Response analysis using R

    PLOS ONE 10:e0146021.

    https://doi.org/10.1371/journal.pone.0146021

    • PubMed
    • Google Scholar
60. 1. Rodríguez-Tebar A
    2. Rojo F
    3. Dámaso D
    4. Vázquez D

    (1982) Carbenicillin resistance of Pseudomonas aeruginosa

    Antimicrobial Agents and Chemotherapy 22:255–261.

    https://doi.org/10.1128/AAC.22.2.255

    • Google Scholar
61. 1. Rodriguez‐Tebár A
    2. Rojo F
    3. Montilla JC
    4. Vázquez D

    (1982) Interaction of β-lactam antibiotics with penicillin-binding proteins from Pseudomonas aeruginosa

    FEMS Microbiology Letters 14:295–298.

    https://doi.org/10.1111/j.1574-6968.1982.tb00016.x

    • Google Scholar
62. 1. Roemhild R
    2. Barbosa C
    3. Beardmore RE
    4. Jansen G
    5. Schulenburg H

    (2015) Temporal variation in antibiotic environments slows down resistance evolution in pathogenic Pseudomonas aeruginosa

    Evolutionary Applications 8:945–955.

    https://doi.org/10.1111/eva.12330

    • Google Scholar
63. 1. Roemhild R
    2. Gokhale CS
    3. Dirksen P
    4. Blake C
    5. Rosenstiel P
    6. Traulsen A
    7. Andersson DI
    8. Schulenburg H

    (2018) Cellular hysteresis as a principle to maximize the efficacy of antibiotic therapy

    PNAS 115:9767–9772.

    https://doi.org/10.1073/pnas.1810004115

    • Google Scholar
64. 1. Roemhild R
    2. Schulenburg H

    (2019) Evolutionary ecology meets the antibiotic crisis: can we control pathogen adaptation through sequential therapy?

    Evolution, Medicine, and Public Health 2019:37–45.

    https://doi.org/10.1093/emph/eoz008

    • PubMed
    • Google Scholar
65. 1. Rosenkilde CEH
    2. Munck C
    3. Porse A
    4. Linkevicius M
    5. Andersson DI
    6. Sommer MOA

    (2019) Collateral sensitivity constrains resistance evolution of the CTX-M-15 β-lactamase

    Nature Communications 10:618.

    https://doi.org/10.1038/s41467-019-08529-y

    • Google Scholar
66. 1. Ross-Gillespie A
    2. Kümmerli R

    (2014) 'Evolution-proofing' antibacterials

    Evolution, Medicine, and Public Health 2014:134–135.

    https://doi.org/10.1093/emph/eou020

    • PubMed
    • Google Scholar
67. 1. Sakyo S
    2. Tomita H
    3. Tanimoto K
    4. Fujimoto S
    5. Ike Y

    (2006) Potency of carbapenems for the prevention of Carbapenem-Resistant mutants of Pseudomonas aeruginosa: the high potency of a new carbapenem doripenem

    J Antibiot 59:220–228.

    https://doi.org/10.1038/ja.2006.31

    • Google Scholar
68. 1. Sanz-García F
    2. Hernando-Amado S
    3. Martínez JL

    (2018) Mutation-Driven evolution of Pseudomonas aeruginosa in the presence of either ceftazidime or Ceftazidime-Avibactam

    Antimicrobial Agents and Chemotherapy 62:e01379-18.

    https://doi.org/10.1128/AAC.01379-18

    • Google Scholar
69. 1. Schafer J
    2. Goff D
    3. Mangino J

    (2009) Doripenem: a new addition to the carbapenem class of antimicrobials

    Recent Patents Anti-Infect Drug Disc 4:18–28.

    https://doi.org/10.2174/157489109787236283

    • Google Scholar
70. 1. Shyu WC
    2. Nightingale CH
    3. Quintiliani R

    (1987) Pseudomonas peritonitis in neutropenic rats treated with amikacin, ceftazidime and ticarcillin, alone and in combination

    Journal of Antimicrobial Chemotherapy 19:807–814.

    https://doi.org/10.1093/jac/19.6.807

    • Google Scholar
71. 1. Sobel ML
    2. Hocquet D
    3. Cao L
    4. Plesiat P
    5. Poole K

    (2005) Mutations in PA3574 ( nalD ) Lead to Increased MexAB-OprM Expression and Multidrug Resistance in Laboratory and Clinical Isolates of Pseudomonas aeruginosa

    Antimicrobial Agents and Chemotherapy 49:1782–1786.

    https://doi.org/10.1128/AAC.49.5.1782-1786.2005

    • Google Scholar
72. 1. Szybalski W
    2. Bryson V

    (1952) Genetic studies on microbial cross resistance to toxic agents I

    Journal of Bacteriology 64:489–499.

    https://doi.org/10.1128/jb.64.4.489-499.1952

    • Google Scholar
73. 1. Tanimoto K
    2. Tomita H
    3. Fujimoto S
    4. Okuzumi K
    5. Ike Y

    (2008) Fluoroquinolone Enhances the Mutation Frequency for Meropenem-Selected Carbapenem Resistance in Pseudomonas aeruginosa , but Use of the High-Potency Drug Doripenem Inhibits Mutant Formation

    Antimicrobial Agents and Chemotherapy 52:3795–3800.

    https://doi.org/10.1128/AAC.00464-08

    • Google Scholar
74. 1. Traub WH
    2. Raymond EA

    (1970) Susceptibility of Pseudomonas aeruginosa to carbenicillin

    Applied Microbiology 20:630–632.

    https://doi.org/10.1128/am.20.4.630-632.1970

    • PubMed
    • Google Scholar
75. 1. Tueffers L
    2. Barbosa C
    3. Bobis I
    4. Schubert S
    5. Höppner M
    6. Rühlemann M
    7. Franke A
    8. Rosenstiel P
    9. Friedrichs A
    10. Krenz-Weinreich A
    11. Fickenscher H
    12. Bewig B
    13. Schreiber S
    14. Schulenburg H

    (2019) Pseudomonas aeruginosa populations in the cystic fibrosis lung lose susceptibility to newly applied β-lactams within 3 days

    Journal of Antimicrobial Chemotherapy 74:2916–2925.

    https://doi.org/10.1093/jac/dkz297

    • Google Scholar
76. 1. von der Schulenburg JH
    2. Hancock JM
    3. Pagnamenta A
    4. Sloggett JJ
    5. Majerus ME
    6. Hurst GD

    (2001) Extreme length and length variation in the first ribosomal internal transcribed spacer of ladybird beetles (Coleoptera: coccinellidae)

    Molecular Biology and Evolution 18:648–660.

    https://doi.org/10.1093/oxfordjournals.molbev.a003845

    • PubMed
    • Google Scholar
77. Book

    1. Walsh C

    (2003)

    Antibiotics: Actions, Origins, Resistance

    Washington: ASM Media & PR.

    • Google Scholar
78. 1. Woods RJ
    2. Read AF

    (2015) Clinical management of resistance evolution in a bacterial infection

    Evolution, Medicine, and Public Health 2015:281–288.

    https://doi.org/10.1093/emph/eov025

    • Google Scholar
79. 1. Wright DB

    (1986) Cefsulodin

    Drug Intelligence & Clinical Pharmacy 20:845–849.

    https://doi.org/10.1177/106002808602001104

    • Google Scholar
80. 1. Yen P
    2. Papin JA

    (2017) History of antibiotic adaptation influences microbial evolutionary dynamics during subsequent treatment

    PLOS Biology 15:e2001586.

    https://doi.org/10.1371/journal.pbio.2001586

    • Google Scholar
81. 1. Yoshida M
    2. Reyes SG
    3. Tsuda S
    4. Horinouchi T
    5. Furusawa C
    6. Cronin L

    (2017) Time-programmable drug dosing allows the manipulation, suppression and reversal of antibiotic drug resistance in vitro

    Nature Communications 8:15589.

    https://doi.org/10.1038/ncomms15589

    • Google Scholar
82. 1. Zapun A
    2. Contreras-Martel C
    3. Vernet T

    (2008) Penicillin-binding proteins and β-lactam resistance

    FEMS Microbiology Reviews 32:361–385.

    https://doi.org/10.1111/j.1574-6976.2007.00095.x

    • Google Scholar
83. 1. Zheng Q

    (2017) rSalvador: an R package for the fluctuation experiment

    G3: Genes, Genomes, Genetics 7:3849–3856.

    https://doi.org/10.1534/g3.117.300120

    • Google Scholar
84. 1. Zhou S
    2. Barbosa C
    3. Woods RJ

    (2020) Why is preventing antibiotic resistance so hard? Analysis of failed resistance management

    Evolution, Medicine, and Public Health 2020:102–108.

    https://doi.org/10.1093/emph/eoaa020

    • Google Scholar
85. 1. Zimmermann W

    (1980) Penetration of beta-lactam antibiotics into their target enzymes in Pseudomonas aeruginosa: comparison of a highly sensitive mutant with its parent strain

    Antimicrobial Agents and Chemotherapy 18:94–100.

    https://doi.org/10.1128/AAC.18.1.94

    • Google Scholar

Acceptance summary:

This study compared effectiveness of three combinations of antibiotic drug therapy regimes against the bacterial pathogen Pseudomonasaeruginosa and found that an unlikely drug combination promoted bacterial extinction. In contrast to prior studies using combinations of drugs with similar chemistries (eg. β-lactams that inhibit cell wall synthesis), which lead to cross-resistance and treatment failure, the sequential use of three β-lactams CAR-CEF-DOR (carbenicillin, cefsulodin, doripenem) was highly effective because of constraints caused by DOR. These results revitalize the potential use of similar β-lactams for treatment against bacterial pathogens and promote utility of predicting evolutionary trajectories to improve antimicrobial drug therapies.

Decision letter after peer review:

Thank you for submitting your article "High potency of sequential therapy with only beta-lactam antibiotics" for consideration by eLife. Your article has been reviewed by 3 peer reviewers, including Vaughn S Cooper as the Reviewing Editor and Reviewer #1, and the evaluation has been overseen by George Perry as the Senior Editor.

The reviewers have discussed their reviews with one another, and the Reviewing Editor has drafted this to help you prepare a revised submission.

Essential Revisions:

1) The sole strain used in this work is P. aeruginosa PA14. This is a good strain choice but the paper seems to presume that this isolate represents the species well, or even more broadly. What is known in the literature about variation within PA or related species in responses to DOR, and/or other drugs in the triple combinations? Is the hysteresis effect limited to this strain?

2) Given that much of the paper focuses on the doripenem effect, what are the relevant attributes of this drug relative to others? At face value, it's a long-lived carbapenem, so is it the fact that it's a carbapenem, its durability, or both the cause of this physiological stress that diminishes subsequent survival? Might the hysteresis effect simply be the killing of susceptible cells causing subsequent failure?

3) A primary conclusion of this paper (L 359-61) is that the spontaneous rate of resistance was a key variable dictating treatment potency. However, this conclusion does not appear to account for the diminished population size of DOR-treated populations and the high potency of this drug relative to others, which would increase the likelihood of error in DOR dosage. While Figure 4 shows that mutation rates to DOR are low, and the methods are appropriate, the results could reflect stronger selection imposed by DOR concentrations and smaller genomic targets to resistance. The authors use this to make their case, but do not adequately discuss the differences between these three β-lactam drugs. In other words, I wonder if PA14 is on a knife-edge in the presence of DOR at the concentrations used, and this reduces population sizes, strengthens selection beyond other drugs, and leads to more variable extinction probabilities. Discussion of the population-genetic consequences of evolution at 0.75x the MIC, and the role of error around this 0.75x threshold, are needed.

4) Figure 1A: I found the different dashed lines to be difficult to distinguish. Maybe place each label near each closed contour so it's easy to see? Also, consider emphasizing those from this study (perhaps with thicker contours?). The positioning of the drug classes makes it unclear what STR is. Is it DNA gyrase or Ribosome since it is in the middle of those terms? Perhaps use color coding. Also, please define what FQ, AG and BL are in Figure 1A.

5) Lines 47-50: It is unclear how competitive release occurs in this scenario. It is my understanding that competitive release occurs when one lineage acquires a highly beneficial mutation early that leads to competitive exclusion of the others. And the paper cited suggests that this phenomenon occurs due to selective pressure not strong enough to prevent this mutation. However, how it is written suggests that easily accessible genetic resistance is the source. But if it were indeed easily accessible, then there would be higher chances that other lineages would also acquire it? If my understanding is incorrect, perhaps provide a bit more of an explanation for the term competitive release.

6) Line 80-84: It is unclear how negative hysteresis was suppressed by "strong selection dynamics" with a mild increase in resistance. Can you provide a specific genetic example from the paper? This will also help conceptualize types of negative hysteresis that can occur during strong antibiotic pressures.

7) Line 91: Replace "As before" with "The new" and add "expected to contribute to collateral sensitivity…". As written, it is confusing what was previously tested and what the new design and expectations are in this study.

8) Line 97: Similar to above. Add "expected to be dominated by cross-resistance…" Helps to clarify past from present.

9) The term "season" is confusing and often used interchangeably with "transfer". I prefer the term transfer and suggest keeping this terminology throughout.

10) Figure Supplemental 1, Table1: I like the extra information about the stock concentration and storage. It would also be helpful if the drugs were ordered by class and maybe even identified using the BL, FQ, AG abbreviations.

11) Figure 2: Please describe that the dotted lines represent the three growth phases in A. For B, it is unclear what the arrows signify. It looks like they are aligned with transfers 12 and 48. Either make them more continuous or remove them altogether. Otherwise, it looks as if the populations are what is being highlighted. Also, what is the significance of the +1 to -0.5 range vs the 16x to 0.33x? The MIC column is most explanatory to me.

12) Line 192-193: To my eye, the resistance profiles between the early- and mid- phases look pretty similar. The sum of resistances is mentioned. Perhaps add the sum to each of the columns in Figure 2 to make it clearer.

13) Line 210-213: DOR monotherapy is stated to not show a noticeable amount of resistance at the early stage. However, it looks like population 1 shows 4x-8X MIC and looks similar to the mid phase. This is contradictory to what is stated. Can you please rephrase.

14) Line 210-213: It is interesting that monotherapy population one at transfer 12 showed increased resistance but resulted in no variants found. Is it possible that this was a result of heteroresistance or persistence? If you sequenced the whole population instead of isolating three clones, you might be able to distinguish variants at low frequency that would contribute to heteroresistance.

15) Figure 3C. The Blue and black are hard to distinguish as a gradient.

16) Line 361: Is it treatment potency due to cross resistance or the lack of cross-resistance. I thought that it was due to the latter because neither of the other two drugs promoted cross-resistance to DOR.

17) Line 414: Is "disadvantageous cross-resistance" the same as collateral sensitivity? If so, the double negative is confusing.

Recommendations for follow up work

1. Whole genome sequencing was limited to three clones per population. While it is understandable that the number of treatment regimens might make sequencing costly. Population dynamics and allele frequencies may help explain why the population during the DOR monotherapy struggled to acquire mutations before transfer 12.

2. One of the most interesting aspects of the study is that the principles identified for designing sequential therapies are not dependent on a unique molecular mechanism, and one could therefore envision multiple drug combinations that achieve similar evolution-slowing effects but do so in very different ways at the molecular level (where resistance has often been studied). Do the authors believe there is some inherent feature of β-lactams-for example, the fact that they compromise the cell wall and promote lysis-that may underlie the success of homogeneous sequential therapies? Or might one expect similar success with, say, triplets of fluoroquinolones or other drugs, as long as the triplet is chosen with similar qualitative features (e.g. low levels of cross-resistance, low spontaneous resistance to at least one drug). Relatedly, are there reasons to believe that certain classes of drugs will be more likely to show those qualitative features? It would be remarkable if, for example, these important features could be predicted based on, say, the known diversity of resistant mutations for the component drugs. In any event, I think one of the striking conclusions from this work is that considerable optimization may be possible without detailed knowledge of molecular mechanism.

3. The results suggest that the chosen β lactams exhibit negative hysteresis, yet the negative hysteresis does not play a dominant role in slowing resistance (unlike in the authors' previous study with heterogeneous drug combinations). Does this discrepancy arise simply because other effects (e.g. collateral effects) are dominant, so the impacts of negative hysteresis are (by comparison) negligible? Or is there some feature of the negative hysteresis-and perhaps its correlation with the collateral effects-that renders it non-beneficial? I'd venture a guess that in general, there is some underlying relationship between the hysteresis profiles, the collateral profiles, and the timescale of switching that could conceivably be tuned to optimize these therapies. It seems to be an exciting topic for future work.

Reviewer #1 (Recommendations for the authors):

The authors conducted an ambitious set of hundreds of evolution experiments with P. aeruginosa exposed to different sets of 3 antibiotics amounting to ~500 generations each, under monotherapy, fast switching, slow switching, and random switching. A primary outcome was extinction, ie failure to persist, while secondary outcomes included the extent and breadth of evolved resistance and a sampling of evolved resistance mutations. Treatments including the antibiotic doripenem led to more extinctions, likely because growth in doripenem reduced the ability to grow in alternative antibiotics, which they term negative hysteresis. This is an interesting and potentially clinically relevant discovery. The paper is well written and engaging.

I see three major limitations that the authors could reasonably address.

1. The sole strain used in this work is P. aeruginosa PA14. This is a good strain choice but the paper seems to presume that this isolate represents the species well, or even more broadly. What is known in the literature about variation within PA or related species in responses to DOR, and/or other drugs in the triple combinations? Is the hysteresis effect limited to this strain?

2. Given that much of the paper focuses on the doripenem effect, what are the relevant attributes of this drug relative to others? At face value, it's a long-lived carbapenem, so is it the fact that it's a carbapenem, its durability, or both the cause of this physiological stress that diminishes subsequent survival? Might the hysteresis effect simply be the killing of susceptible cells causing subsequent failure?

3. A primary conclusion of this paper (L 359-61) is that the spontaneous rate of resistance was a key variable dictating treatment potency. However, this conclusion does not appear to account for the diminished population size of DOR-treated populations and the high potency of this drug relative to others, which would increase the likelihood of error in DOR dosage. While Figure 4 shows that mutation rates to DOR are low, and the methods are appropriate, the results could reflect stronger selection imposed by DOR concentrations and smaller genomic targets to resistance. The authors use this to make their case, but do not adequately discuss the differences between these three β-lactam drugs. In other words, I wonder if PA14 is on a knife-edge in the presence of DOR at the concentrations used, and this reduces population sizes, strengthens selection beyond other drugs, and leads to more variable extinction probabilities. Discussion of the population-genetic consequences of evolution at 0.75x the MIC, and the role of error around this 0.75x threshold, are needed.

Reviewer #2 (Recommendations for the authors):

Batra, Aditi et al., compared the effectiveness of three in vitro antibiotic drug therapy regimes against the bacterial pathogen Pseudomonasaeruginosa and found that an unlikely drug combination promoted bacterial extinction. Previous studies have shown that using similar drug classes with similar chemistries (eg. β-lactams that inhibit cell wall synthesis) lead to increased cross resistance and therefore, treatment failure. However, the authors found that the fast and sequential use of three β-lactams CAR-CEF-DOR (carbenicillin, cefsulodin, doripenem) was the most effective treatment against P. aeruginosa. The authors further investigated the potential factors contributing to a lack of evolutionary rescue such as negative hysteresis, collateral sensitivity, and rate of spontaneous direct/indirect resistance. By conducting additional experiments to test multiple hypotheses, the authors concluded that the fast-switching three-drug series was constrained by DOR and that the cause of the constraint was due to a low rate of spontaneous resistance and cross-resistance to DOR. These results revitalize the potential use of similar β-lactams for treatment against bacterial pathogens and the utility of predicting evolutionary trajectories to improve antimicrobial drug therapies.

Strengths:

This paper introduces both physiological and evolutionary concepts as possible hypotheses for explaining antibiotic drug constraints. When discussing consequences to drug therapies, cross resistance and collateral sensitivity are often described whereas physiological constraints such as negative hysteresis is overlooked. The experiments conducted in this paper demonstrate how to test for hysteresis using short exposures to a drug and then switching to a different drug and comparing growth killing effects. Drug exposure is carried out quickly to maintain a low mutational load and reduce the probability that a beneficial mutation would arise. Mutations would interfere with a test for non-genetic physiological conditioning.

Measurements of spontaneous resistance rates is a useful phenotype to compare mechanisms of mutations for each drug. The experiments and results that compare the rate of spontaneous direct resistance and indirect (ie. Cross resistance) as well as collateral sensitivity are clear and well replicated.

Weakness:

Whole genome sequencing was limited to three clones per population. While it is understandable that the number of treatment regimens might make sequencing costly. Population dynamics and allele frequencies may help explain why the population during the DOR monotherapy struggled to acquire mutations before transfer 12.Reviewer #3 (Recommendations for the authors):

The manuscript by Batra and colleagues investigates the impact of sequential antibiotic treatments on evolving laboratory populations of P. aeruginosa, a clinically important microbial pathogen. A growing body of work suggests that cycling between different antibiotics can slow resistance in the lab, though the conventional view is that successful treatments would require drugs from different classes, limiting the propensity for cross resistance and potentially harnessing sensitizing collateral effects or negative hysteresis. This study provides a large-scale investigation of the efficacy of sequential treatments comprised of different 3-drug treatments. The drugs are chosen to span different classes, with triplicates including drugs targeting the ribosome, DNA gyrase, or the cell wall (β lactams) in some combination. Previous work from the same group showed that fast switching between a heterogeneous set (3 different classes) of drugs promoted population extinction at sub-MIC concentrations, with selection favoring genetic suppression of negative hysteresis. In this work, they investigate the response to different 3-drug sequences (both homogeneous and heterogeneous) in a total of 756 (!!) replicate populations. Treatments included monotherapy, fast or slow switching, and random switching. Interestingly, they found that switching among two different sets of β lactams produced high levels of extinction-comparable to that previously observed with a heterogeneous set-while switching between a new heterogeneous set did not produce extinction. Extinction tended to occur early in the treatment and was significantly reduced when the period of treatment switching was extended. The authors focus on one drug set (CAR-CEF-DOR) for an in-depth characterization. Using general linear models (GLM) to evaluate growth adaptation and total β lactam resistance in the three distinct growth phases, they showed that early phase adaptation is decelerated in the fast-switching group relative to monotherapy and slow-switching group; interestingly, the fast and random treatments showed relatively similar rates of extinction, suggesting that precise ordering is not the primary driver of the effects. They also observed generally increased β lactam resistance in the second phase, but the different treatment types led to largely distinct resistance profiles. Population diversity increased between early (transfer 12) and later (transfer 48) days of adaptation, and resistance in the early phase was often constrained by susceptibility to a particular drug (DOR) (and whole-genome sequencing found DOR resistance mutations only at later times). The authors go on to show that negative hysteresis is common among the three β-lactams (Figure 3), while DOR exhibits the lowest rates of direct and cross resistance (as measured by classic fluctuation tests and a patching assay). Finally, they used another GLM to quantify the impacts of different features on slowing resistance; they found that switching rate, temporal irregularity, spontaneous resistance probability, and (most prominently) the collateral effects dominate, with low levels of spontaneous resistance to DOR a primary constraint.

This study is timely, important, extremely thorough, and perhaps most importantly, it helps to identify a number of new principles to guide the rational design of sequential antibiotic therapies. It represents a monumental effort to thoroughly characterize adaption to 3-drug cycles in several representative examples, an achievement in itself. But as impressively, the authors draw on GLM's to extract meaningful patterns from the web of complex phenotypic data, and in doing so they uncover features of these treatments (e.g. surprising cross-resistance landscapes between drugs from the same class; potential importance of having one drug with low rates of spontaneous resistance) that may guide general design of new therapies. As with all good studies, it raises a number of questions for future work, and I list a few that come to mind below (that authors may wish to respond, perhaps in the discussion, at their discretion). But in general, I have no major concerns with the study at a technical or conceptual level (a rarity!). I believe this is an excellent contribution to the literature that will interest a broad range of readers in evolutionary biology and microbiology.

1) To me, one of the most interesting aspects of the study is that the principles identified for designing sequential therapies are not dependent on a unique molecular mechanism, and one could therefore envision multiple drug combinations that achieve similar evolution-slowing effects but do so in very different ways at the molecular level (where resistance has often been studied). Do the authors believe there is some inherent feature of β-lactams-for example, the fact that they compromise the cell wall and promote lysis-that may underlie the success of homogeneous sequential therapies? Or might one expect similar success with, say, triplets of fluoroquinolones or other drugs, as long as the triplet is chosen with similar qualitative features (e.g. low levels of cross-resistance, low spontaneous resistance to at least one drug). Relatedly, are there reasons to believe that certain classes of drugs will be more likely to show those qualitative features? It would be remarkable if, for example, these important features could be predicted based on, say, the known diversity of resistant mutations for the component drugs. In any event, I think one of the striking conclusions from this work is that considerable optimization may be possible without detailed knowledge of molecular mechanism.

2) The results suggest that the chosen β lactams exhibit negative hysteresis, yet the negative hysteresis does not play a dominant role in slowing resistance (unlike in the authors' previous study with heterogeneous drug combinations). Does this discrepancy arise simply because other effects (e.g. collateral effects) are dominant, so the impacts of negative hysteresis are (by comparison) negligible? Or is there some feature of the negative hysteresis-and perhaps its correlation with the collateral effects-that renders it non-beneficial? I'd venture a guess that in general, there is some underlying relationship between the hysteresis profiles, the collateral profiles, and the timescale of switching that could conceivably be tuned to optimize these therapies. It seems to be an exciting topic for future work.

Essential Revisions:

1) The sole strain used in this work is P. aeruginosa PA14. This is a good strain choice but the paper seems to presume that this isolate represents the species well, or even more broadly. What is known in the literature about variation within PA or related species in responses to DOR, and/or other drugs in the triple combinations? Is the hysteresis effect limited to this strain?

Many thanks for this comment. We agree that the strain PA14 does not necessarily capture the phenotypes present within the species Pseudomonas aeruginosa and it is also possible that other bacterial species may show different evolutionary responses to the treatments tested. We here use PA14 as an established, tractable pathogen model system. In response to the reviewer’s comment, we now emphasize this point at the end of the introduction. Moreover, some information is indeed available on the response of other P. aeruginosa strains to the three antibiotics used. We now summarized this information in a new paragraph in the discussion. In contrast, hysteresis tests with the considered three antibiotics have thus far only been done by us, and the results are presented in the manuscript. It would indeed be of particular value to assess such hysteresis also in other strains, as we now point out in the final part of the discussion. In fact, this is something which we will address in a new project in my lab.

2) Given that much of the paper focuses on the doripenem effect, what are the relevant attributes of this drug relative to others? At face value, it's a long-lived carbapenem, so is it the fact that it's a carbapenem, its durability, or both the cause of this physiological stress that diminishes subsequent survival? Might the hysteresis effect simply be the killing of susceptible cells causing subsequent failure?

This is indeed a very interesting point. We now added more information on the characteristics of Doripenem in comparison with the other drugs in the first part of the discussion. Moreover, as indicated above, we now also provide a focused summary of the available information on resistance rate evolution of P. aeruginosa to Doripenem and the other two drugs. As to hysteresis: The used assay is set up to minimize any killing or differential replication of cells during the pretreatment, in order to ensure that any responses seen in the main treatment are based on physiological changes and not selection. This aspect is now highlighted in the Results section in the chapter on hysteresis.

3) A primary conclusion of this paper (L 359-61) is that the spontaneous rate of resistance was a key variable dictating treatment potency. However, this conclusion does not appear to account for the diminished population size of DOR-treated populations and the high potency of this drug relative to others, which would increase the likelihood of error in DOR dosage. While Figure 4 shows that mutation rates to DOR are low, and the methods are appropriate, the results could reflect stronger selection imposed by DOR concentrations and smaller genomic targets to resistance. The authors use this to make their case, but do not adequately discuss the differences between these three β-lactam drugs. In other words, I wonder if PA14 is on a knife-edge in the presence of DOR at the concentrations used, and this reduces population sizes, strengthens selection beyond other drugs, and leads to more variable extinction probabilities. Discussion of the population-genetic consequences of evolution at 0.75x the MIC, and the role of error around this 0.75x threshold, are needed.

Many thanks for this interesting point. However, we politely disagree that a stronger reduction in population size through DOR in comparison to the other drugs could have influenced the results.

Firstly, the dose response curves in the first supplementary figure (Figure 1—figure supplement 1) indicate very little variation in yield upon DOR treatment, especially at the IC75 concentration used to initiate the evolution experiments. Please note the 2nd panel in the top row and the right panel in the middle row; error bars are standard deviation of the mean (using 6 biological replicates) and hardly visible for the IC75 point because of the lack of variation (see red data points). Much more variation was for example found for CAR.

Secondly, the main evolution experiments were initiated using specifically standardized IC75 drug concentrations. Therefore, we do not expect initial variation in population size across treatments. Such variation should only arise thereafter, as a consequence of hysteresis effects or spontaneously evolved resistances and collateral effects.

Thirdly, the same argument holds for the fluctuation assays, which were performed with specifically standardized antibiotic concentrations, leading to identical selection being imposed to capture the spontaneous mutants.

Therefore, we consider it most likely that the lower adaptation to DOR in the monotherapies and the DOR transfer periods within the sequential treatments and also the reduced resistance rate within the fluctuation assays are due to a lower rate of DOR-specific resistance mutations (i.e., smaller genomic targets to DOR resistance) and also a lower rate of cross-resistance to DOR. These arguments are now presented in the revised Results section – in the chapter on the dynamics of resistance for the triple β-lactam treatment, and the chapter on spontaneous resistance rates.

4) Figure 1A: I found the different dashed lines to be difficult to distinguish. Maybe place each label near each closed contour so it's easy to see? Also, consider emphasizing those from this study (perhaps with thicker contours?). The positioning of the drug classes makes it unclear what STR is. Is it DNA gyrase or Ribosome since it is in the middle of those terms? Perhaps use color coding. Also, please define what FQ, AG and BL are in Figure 1A.

Many thanks for your suggestions. We hope that the revised figure is more accessible.

5) Lines 47-50: It is unclear how competitive release occurs in this scenario. It is my understanding that competitive release occurs when one lineage acquires a highly beneficial mutation early that leads to competitive exclusion of the others. And the paper cited suggests that this phenomenon occurs due to selective pressure not strong enough to prevent this mutation. However, how it is written suggests that easily accessible genetic resistance is the source. But if it were indeed easily accessible, then there would be higher chances that other lineages would also acquire it? If my understanding is incorrect, perhaps provide a bit more of an explanation for the term competitive release.

We agree that the sentence was not entirely clear. We have now removed the concept of competitive release from the sentence, as it is not important for the particular argument given and also not for the focus of our paper.

6) Line 80-84: It is unclear how negative hysteresis was suppressed by "strong selection dynamics" with a mild increase in resistance. Can you provide a specific genetic example from the paper? This will also help conceptualize types of negative hysteresis that can occur during strong antibiotic pressures.

Many thanks for helping us to clarify this section. We now refer to a genetic example from our previous publication in the revised text.

7) Line 91: Replace "As before" with "The new" and add "expected to contribute to collateral sensitivity…". As written, it is confusing what was previously tested and what the new design and expectations are in this study.

Many thanks. This is now corrected.

8) Line 97: Similar to above. Add "expected to be dominated by cross-resistance…" Helps to clarify past from present.

Corrected.

9) The term "season" is confusing and often used interchangeably with "transfer". I prefer the term transfer and suggest keeping this terminology throughout.

Many thanks for this suggestion, which we now followed throughout the revised manuscript.

10) Figure Supplemental 1, Table1: I like the extra information about the stock concentration and storage. It would also be helpful if the drugs were ordered by class and maybe even identified using the BL, FQ, AG abbreviations.

We followed the advice of the reviewer and changed the table accordingly.

11) Figure 2: Please describe that the dotted lines represent the three growth phases in A. For B, it is unclear what the arrows signify. It looks like they are aligned with transfers 12 and 48. Either make them more continuous or remove them altogether. Otherwise, it looks as if the populations are what is being highlighted. Also, what is the significance of the +1 to -0.5 range vs the 16x to 0.33x? The MIC column is most explanatory to me.

Many thanks for your suggestions.

First, we now describe that the dotted lines indicate the three growth phases in the text (line 165-166) and in the figure legend.

Second, we have replaced the misleading arrows by curled brackets to more clearly indicate that all adjacent columns characterize the same transfer (see revised Figure 2).

Third, we completely agree with the reviewer that the MIC column of the resistance scale is more intuitive. However, the visualized resistance scale that ranges from -0.5 to +1 is a better representation of the observed biological variation as it captures not only the x-axis intercept, but the entire shape of the dose-response curve. Thus, the used parameter will differentiate sub-populations that have equivalent MIC, but differ in dose-sensitivity or metabolic fitness cost. Still, it is important to know the approximate range of effect sizes, for which we indicated approximate change of MIC corresponding to resistance value. In the revised manuscript, we now provide more information about the double scaling of the resistance quantification in the new Figure 2—figure supplement 3.

12) Line 192-193: To my eye, the resistance profiles between the early- and mid- phases look pretty similar. The sum of resistances is mentioned. Perhaps add the sum to each of the columns in Figure 2 to make it clearer.

Many thanks for this comment. The reviewer is correct that there is no general increase in resistance across the two considered time points. Instead, resistance changes vary and depend on the exact treatment. In response to the reviewer’s comments, we now adjusted our description of the results, where we now point out that “Resistance to the used β-lactams increased across the two time points only in some treatments, but not all (Figure 2B and C, Figure 2—figure supplement 4, Figure 2—figure supplement 5, Supplementary File 1F), suggesting treatment-dependent evolutionary responses to the antibiotics.”. We also adjusted our conclusions of this part of the result, where we now state that “the population analysis of resistance profiles indicates that resistance evolution depends on the exact treatment protocol and that the dynamics of resistance emergence to DOR may be key for the observed deceleration of beta-lactam adaptation in the fast-regular treatments.”. To enhance clarity, we further added two supplementary figures.

Figure 2—figure supplement 4 shows the early- and mid-phase resistance values side-by-side for each antibiotic in a box plot design. Please see also the related new statistical analysis in Supplementary File 1F that shows which population resistance profiles changed over time.

The sum of resistances is now presented in Figure 2—figure supplement 5, separately for the two considered transfer time points in panels A and B and also separately for three β-lactams (left columns) and the two collateral antibiotics CIP and GEN (right columns). We now refer to this figure to further support our observation of the comparatively high efficacy of irregular sequential protocols in the discussion.

13) Line 210-213: DOR monotherapy is stated to not show a noticeable amount of resistance at the early stage. However, it looks like population 1 shows 4x-8X MIC and looks similar to the mid phase. This is contradictory to what is stated. Can you please rephrase.

Many thanks for this comment. It is based on a misunderstanding of the design of panels B and C of Figure 2, which were apparently not sufficiently clear. In detail, Mono 1 is monotherapy with CAR, mono 2 is monotherapy with DOR, and mono 3 is monotherapy with CEF. For the early time point, the considered isolates from DOR treatment (=2nd row) did not produce any resistances against DOR (=2nd column) and almost no resistance against any of the other drugs. In the revised manuscript, we now describe the design of the figure in more detail in order to prevent any misunderstandings.

14) Line 210-213: It is interesting that monotherapy population one at transfer 12 showed increased resistance but resulted in no variants found. Is it possible that this was a result of heteroresistance or persistence? If you sequenced the whole population instead of isolating three clones, you might be able to distinguish variants at low frequency that would contribute to heteroresistance.

Many thanks for this comment. It shows that our figure was not sufficiently clear, as the comment is based on the same misunderstanding as mentioned above for comment 13. In detail, the results for the DOR monotherapy are shown in the second row of panels B and C of Figure 3 (not the first row). Thus, the isolates from the DOR treatment at the early time point do not show increases in resistance. This is then entirely consistent with our finding that the sequenced genomes for this treatment at the early time point do not bear any resistance mutations. This is something, which we now emphasize in the revised manuscript. These consistent results indicate that evolution of resistance to DOR is constrained, possibly because of a smaller range of genomic targets for DOR resistance. We apologize that the design of Figure 2 panels B and C led to a misunderstanding. As outlined above, we adjusted the design and description of the two panels, in order to prevent such misunderstandings.

We further agree with the reviewer that a more detailed population genomic analysis could have provided further insights into adaptation to DOR monotherapy. The conducted sequencing does not have the resolution to detect low-frequency variants or potential gene amplifications that could confer heteroresistance. However, we are also convinced that such additional genomics data is not essential for our conclusions. Firstly, the phenotypic and genomic results are consistent, as outlined above. Secondly, the indicated lower mutation rate towards DOR resistance was independently validated by fluctuation assays. And this test did indeed confirm the initial finding. This is a point, which we now emphasize more clearly in the Results section. Therefore, we are convinced that our conclusion of a constrained rate of DOR resistance evolution is fully valid and justified and would not receive further support by population sequencing.

Nevertheless, we agree that an understanding of the very early steps of antibiotic adaptation is very important. We are therefore considering to perform a detailed population genomic sequence analysis in the future. Unfortunately, we are currently constrained by consequences of the Corona pandemic, because our sequencing center cannot promise results in the close future. Hence, we are grateful that the eLife guidelines suggest to perform such interesting further analyses separate from the current manuscript with the possibility of linking them in the future. Most importantly, as outlined above, the suggested population genome analyses would be a useful addition, but they are not essential for the final conclusions drawn.

15) Figure 3C. The Blue and black are hard to distinguish as a gradient.

The color scheme has been adjusted to improve clarity.

16) Line 361: Is it treatment potency due to cross resistance or the lack of cross-resistance. I thought that it was due to the latter because neither of the other two drugs promoted cross-resistance to DOR.

Many thanks for this comment. We agree that our previous statement was not sufficiently clear. We now point out that treatment potency was determined by variation in the spontaneous rate of resistance to the β-lactams and by variation in the resulting collateral effects across sequential treatment protocols.

17) Line 414: Is "disadvantageous cross-resistance" the same as collateral sensitivity? If so, the double negative is confusing.

This expression is indeed not very clear. We deleted the word “disadvantageous”, because it is not really needed here.

Recommendations for follow up work

1. Whole genome sequencing was limited to three clones per population. While it is understandable that the number of treatment regimens might make sequencing costly. Population dynamics and allele frequencies may help explain why the population during the DOR monotherapy struggled to acquire mutations before transfer 12.

Many thanks for this recommendation. We agree that a more detailed population genomic analysis of the evolving bacteria could yield important insights into the exact dynamics of resistance evolution. We are indeed planning to perform such analyses in the future – as soon as this is reliably possible again, and we would then be happy to link the results to the current manuscript. We would still like to point out that these population genomic analyses will be insightful, yet they are not essential for our conclusion on the constrained rate of DOR resistance evolution, as explained in detail in our above reply.

2. One of the most interesting aspects of the study is that the principles identified for designing sequential therapies are not dependent on a unique molecular mechanism, and one could therefore envision multiple drug combinations that achieve similar evolution-slowing effects but do so in very different ways at the molecular level (where resistance has often been studied). Do the authors believe there is some inherent feature of β-lactams-for example, the fact that they compromise the cell wall and promote lysis-that may underlie the success of homogeneous sequential therapies? Or might one expect similar success with, say, triplets of fluoroquinolones or other drugs, as long as the triplet is chosen with similar qualitative features (e.g. low levels of cross-resistance, low spontaneous resistance to at least one drug). Relatedly, are there reasons to believe that certain classes of drugs will be more likely to show those qualitative features? It would be remarkable if, for example, these important features could be predicted based on, say, the known diversity of resistant mutations for the component drugs. In any event, I think one of the striking conclusions from this work is that considerable optimization may be possible without detailed knowledge of molecular mechanism.

Many thanks for these comments. We agree that the raised questions would be extremely fascinating to study in more detail. They cannot really be answered based on the current data. In fact, we are in the process of setting up new projects that address for example the last point and also the molecular basis of interactions between different β-lactam drugs. We will hopefully be able to present interesting results on at least these points in the future.

3. The results suggest that the chosen β lactams exhibit negative hysteresis, yet the negative hysteresis does not play a dominant role in slowing resistance (unlike in the authors' previous study with heterogeneous drug combinations). Does this discrepancy arise simply because other effects (e.g. collateral effects) are dominant, so the impacts of negative hysteresis are (by comparison) negligible? Or is there some feature of the negative hysteresis-and perhaps its correlation with the collateral effects-that renders it non-beneficial? I'd venture a guess that in general, there is some underlying relationship between the hysteresis profiles, the collateral profiles, and the timescale of switching that could conceivably be tuned to optimize these therapies. It seems to be an exciting topic for future work.

Again, very interesting points. And again, we cannot answer the raised questions in consideration of the current data. There is clearly a need for an even more systematic analysis on the relationship between hysteresis, collateral effects, and the rate of antibiotic changes. As part of the above mentioned new projects, we do indeed plan to study the molecular basis of hysteresis so that this information could be linked to known mechanisms of collateral sensitivity. So again, we hope to be able to answer at least some of the raised questions with our new projects in the future.

Reviewer #1 (Recommendations for the authors):

The authors conducted an ambitious set of hundreds of evolution experiments with P. aeruginosa exposed to different sets of 3 antibiotics amounting to ~500 generations each, under monotherapy, fast switching, slow switching, and random switching. A primary outcome was extinction, ie failure to persist, while secondary outcomes included the extent and breadth of evolved resistance and a sampling of evolved resistance mutations. Treatments including the antibiotic doripenem led to more extinctions, likely because growth in doripenem reduced the ability to grow in alternative antibiotics, which they term negative hysteresis. This is an interesting and potentially clinically relevant discovery. The paper is well written and engaging.

I see three major limitations that the authors could reasonably address.

1. The sole strain used in this work is P. aeruginosa PA14. This is a good strain choice but the paper seems to presume that this isolate represents the species well, or even more broadly. What is known in the literature about variation within PA or related species in responses to DOR, and/or other drugs in the triple combinations? Is the hysteresis effect limited to this strain?

Many thanks for this comment. We agree that the strain PA14 does not necessarily capture the phenotypes present within the species Pseudomonas aeruginosa and it is also possible that other bacterial species may show different evolutionary responses to the treatments tested. We here use PA14 as a tractable model system. In response to the reviewer’s comment, we now emphasize this point at the end of the introduction. Moreover, some information is indeed available on the response of other P. aeruginosa strains to the three antibiotics used. We now summarized this information in a new paragraph in the discussion. In contrast, hysteresis tests with the considered three antibiotics have thus far only been done by us, and the results are presented in the manuscript. It would indeed be of particular value to assess such hysteresis also in other strains, as we now point out in the final part of the discussion.

2. Given that much of the paper focuses on the doripenem effect, what are the relevant attributes of this drug relative to others? At face value, it's a long-lived carbapenem, so is it the fact that it's a carbapenem, its durability, or both the cause of this physiological stress that diminishes subsequent survival? Might the hysteresis effect simply be the killing of susceptible cells causing subsequent failure?

This is indeed a very interesting point. We now added more information on the characteristics of Doripenem in comparison with the other drugs in the first part of the discussion. Moreover, as indicated above, we now also provide a focused summary of the available information on resistance rate evolution of P. aeruginosa to Doripenem and the other two drugs. As to hysteresis: The used assay is set up to minimize any killing or differential replication of cells during the pretreatment, in order to ensure that any responses seen in the main treatment are based on physiological changes and not selection. This aspect is now highlighted in the Results section in the chapter on hysteresis.

3. A primary conclusion of this paper (L 359-61) is that the spontaneous rate of resistance was a key variable dictating treatment potency. However, this conclusion does not appear to account for the diminished population size of DOR-treated populations and the high potency of this drug relative to others, which would increase the likelihood of error in DOR dosage. While Figure 4 shows that mutation rates to DOR are low, and the methods are appropriate, the results could reflect stronger selection imposed by DOR concentrations and smaller genomic targets to resistance. The authors use this to make their case, but do not adequately discuss the differences between these three β-lactam drugs. In other words, I wonder if PA14 is on a knife-edge in the presence of DOR at the concentrations used, and this reduces population sizes, strengthens selection beyond other drugs, and leads to more variable extinction probabilities. Discussion of the population-genetic consequences of evolution at 0.75x the MIC, and the role of error around this 0.75x threshold, are needed.

Many thanks for this interesting point. However, we politely disagree that a stronger reduction in population size through DOR in comparison to the other drugs could have influenced the results.

Firstly, the dose response curves in the first supplementary figure (Figure 1—figure supplement 1) indicate very little variation in yield upon DOR treatment, especially at the IC75 concentration used to initiate the evolution experiments. Please note the 2nd panel in the top row and the right panel in the middle row; error bars are standard deviation of the mean (using 6 biological replicates) and hardly visible for the IC75 point because of the lack of variation (see red data points). Much more variation was for example found for CAR.

Secondly, the main evolution experiments were initiated using specifically standardized IC75 drug concentrations. Therefore, we do not expect initial variation in population size across treatments. Such variation should only arise thereafter, as a consequence of hysteresis effects or spontaneously evolved resistances and collateral effects.

Thirdly, the same argument holds for the fluctuation assays, which were performed with specifically standardized antibiotic concentrations, leading to identical selection being imposed to capture the spontaneous mutants.

Therefore, we consider it most likely that the lower adaptation to DOR in the monotherapies and the DOR transfer periods within the sequential treatments and also the reduced resistance rate within the fluctuation assays are due to a lower rate of DOR-specific resistance mutations (i.e., smaller genomic targets to DOR resistance) and also a lower rate of cross-resistance to DOR. These arguments are now presented in the revised Results section – in the chapter on the dynamics of resistance for the triple β-lactam treatment, and the chapter on spontaneous resistance rates.

Reviewer #2 (Recommendations for the authors):

Batra, Aditi et al., compared the effectiveness of three in vitro antibiotic drug therapy regimes against the bacterial pathogen Pseudomonas aeruginosa and found that an unlikely drug combination promoted bacterial extinction. Previous studies have shown that using similar drug classes with similar chemistries (eg. β-lactams that inhibit cell wall synthesis) lead to increased cross resistance and therefore, treatment failure. However, the authors found that the fast and sequential use of three β-lactams CAR-CEF-DOR (carbenicillin, cefsulodin, doripenem) was the most effective treatment against P. aeruginosa. The authors further investigated the potential factors contributing to a lack of evolutionary rescue such as negative hysteresis, collateral sensitivity, and rate of spontaneous direct/indirect resistance. By conducting additional experiments to test multiple hypotheses, the authors concluded that the fast-switching three-drug series was constrained by DOR and that the cause of the constraint was due to a low rate of spontaneous resistance and cross-resistance to DOR. These results revitalize the potential use of similar β-lactams for treatment against bacterial pathogens and the utility of predicting evolutionary trajectories to improve antimicrobial drug therapies.

Strengths:

This paper introduces both physiological and evolutionary concepts as possible hypotheses for explaining antibiotic drug constraints. When discussing consequences to drug therapies, cross resistance and collateral sensitivity are often described whereas physiological constraints such as negative hysteresis is overlooked. The experiments conducted in this paper demonstrate how to test for hysteresis using short exposures to a drug and then switching to a different drug and comparing growth killing effects. Drug exposure is carried out quickly to maintain a low mutational load and reduce the probability that a beneficial mutation would arise. Mutations would interfere with a test for non-genetic physiological conditioning.

Measurements of spontaneous resistance rates is a useful phenotype to compare mechanisms of mutations for each drug. The experiments and results that compare the rate of spontaneous direct resistance and indirect (ie. Cross resistance) as well as collateral sensitivity are clear and well replicated.

Weakness:

Whole genome sequencing was limited to three clones per population. While it is understandable that the number of treatment regimens might make sequencing costly. Population dynamics and allele frequencies may help explain why the population during the DOR monotherapy struggled to acquire mutations before transfer 12.

Many thanks for this comment. We agree with the reviewer that a more detailed population genomic analysis could have provided further insights into the absence of resistance changes in the bacteria from the DOR monotherapy treatment. However, such additional genomics data is not essential for our main conclusions. Firstly, the phenotypic and genomic results are consistent, as both demonstrate a lack of resistance to the DOR treatment. Secondly, the phenotypic and current genomic data only provide an indication of a lower mutation rate towards DOR resistance. Additional population genomic data could confirm it, but not validate it. Such a validation may best be achieved by an independent experimental approach, which directly assesses the rate of DOR resistance mutations. This is the reason why we performed such an independent experimental test, using fluctuation assays. This test indeed confirmed the initial finding. Therefore, we consider our conclusion of a constrained rate of DOR resistance evolution to be valid and justified.

Reviewer #3 (Recommendations for the authors):

[…]

This study is timely, important, extremely thorough, and perhaps most importantly, it helps to identify a number of new principles to guide the rational design of sequential antibiotic therapies. It represents a monumental effort to thoroughly characterize adaption to 3-drug cycles in several representative examples, an achievement in itself. But as impressively, the authors draw on GLM's to extract meaningful patterns from the web of complex phenotypic data, and in doing so they uncover features of these treatments (e.g. surprising cross-resistance landscapes between drugs from the same class; potential importance of having one drug with low rates of spontaneous resistance) that may guide general design of new therapies. As with all good studies, it raises a number of questions for future work, and I list a few that come to mind below (that authors may wish to respond, perhaps in the discussion, at their discretion). But in general, I have no major concerns with the study at a technical or conceptual level (a rarity!). I believe this is an excellent contribution to the literature that will interest a broad range of readers in evolutionary biology and microbiology.

1) To me, one of the most interesting aspects of the study is that the principles identified for designing sequential therapies are not dependent on a unique molecular mechanism, and one could therefore envision multiple drug combinations that achieve similar evolution-slowing effects but do so in very different ways at the molecular level (where resistance has often been studied). Do the authors believe there is some inherent feature of β-lactams-for example, the fact that they compromise the cell wall and promote lysis-that may underlie the success of homogeneous sequential therapies? Or might one expect similar success with, say, triplets of fluoroquinolones or other drugs, as long as the triplet is chosen with similar qualitative features (e.g. low levels of cross-resistance, low spontaneous resistance to at least one drug). Relatedly, are there reasons to believe that certain classes of drugs will be more likely to show those qualitative features? It would be remarkable if, for example, these important features could be predicted based on, say, the known diversity of resistant mutations for the component drugs. In any event, I think one of the striking conclusions from this work is that considerable optimization may be possible without detailed knowledge of molecular mechanism.

We share the enthusiasm of the reviewer that the available drugs offer many currently unsuspected options to optimize antibiotic therapy. We agree that the raised questions would be extremely fascinating to study in more detail. They cannot really be answered based on the current data. They do require future additional studies that systematically assess the ability of bacteria to adapt to a variety of multi-drug sequential therapy as well as the underlying mechanisms. Further work is also required to understand the exact molecular basis of interactions between different β-lactam drugs or the general influence of variation in spontaneous resistance to different antibiotics on the efficacy of treatments with these drugs.

2) The results suggest that the chosen β lactams exhibit negative hysteresis, yet the negative hysteresis does not play a dominant role in slowing resistance (unlike in the authors' previous study with heterogeneous drug combinations). Does this discrepancy arise simply because other effects (e.g. collateral effects) are dominant, so the impacts of negative hysteresis are (by comparison) negligible? Or is there some feature of the negative hysteresis-and perhaps its correlation with the collateral effects-that renders it non-beneficial? I'd venture a guess that in general, there is some underlying relationship between the hysteresis profiles, the collateral profiles, and the timescale of switching that could conceivably be tuned to optimize these therapies. It seems to be an exciting topic for future work.

Many thanks for these interesting comments. Again, the raised questions cannot really be answered in consideration of the current data. There is clearly a need for a more systematic analysis of the relationship between hysteresis, collateral effects, and additionally the rate of antibiotic switches. Our earlier work found a two-component regulator cpxS to contribute to negative hysteresis, but only mildly to resistance, possibly suggesting that hysteresis relies on molecular processes that are distinct to those mediating resistance or collateral sensitivity. However, this is only a single case and more examples need to be characterized in detail before general conclusions can be drawn

Author details

1. Aditi Batra

   1. Department of Evolutionary Ecology and Genetics, University of Kiel, Kiel, Germany
   2. Max Planck Institute for Evolutionary Biology, Ploen, Germany

   Contribution

   Conceptualization, Data curation, Formal analysis, Validation, Investigation, Visualization, Methodology, Writing - original draft, Writing - review and editing

   Contributed equally with

   Roderich Roemhild

   Competing interests

   No competing interests declared

2. Roderich Roemhild

   1. Department of Evolutionary Ecology and Genetics, University of Kiel, Kiel, Germany
   2. Max Planck Institute for Evolutionary Biology, Ploen, Germany
   3. Institute of Science and Technology, Klosterneuburg, Austria

   Contribution

   Conceptualization, Data curation, Formal analysis, Validation, Investigation, Visualization, Methodology, Writing - original draft, Writing - review and editing

   Contributed equally with

   Aditi Batra

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0001-9480-5261

3. Emilie Rousseau

   Borstel Research Centre, National Reference Center for Mycobacteria, Borstel, Germany

   Contribution

   Investigation, Methodology, Writing - review and editing

   Competing interests

   No competing interests declared

4. Sören Franzenburg

   Competence Centre for Genomic Analysis Kiel, University of Kiel, Kiel, Germany

   Contribution

   Investigation, Project administration, Whole genome sequencing

   Competing interests

   No competing interests declared

5. Stefan Niemann

   Borstel Research Centre, National Reference Center for Mycobacteria, Borstel, Germany

   Contribution

   Supervision, Project administration, Writing - review and editing

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0002-6604-0684

6. Hinrich Schulenburg

   1. Department of Evolutionary Ecology and Genetics, University of Kiel, Kiel, Germany
   2. Max Planck Institute for Evolutionary Biology, Ploen, Germany

   Contribution

   Conceptualization, Formal analysis, Supervision, Funding acquisition, Writing - original draft, Project administration, Writing - review and editing

   For correspondence

   hschulenburg@zoologie.uni-kiel.de

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0002-1413-913X

• 3,029

  Page views

• 336

  Downloads

• 14

  Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.