An engineered prodrug selectively suppresses β-lactam resistant bacteria in a mixed microbial setting

Abstract

The rise of β-lactam resistance necessitates new strategies to combat bacterial infections. We purposefully engineered the β-lactam prodrug AcephPT to exploit β-lactamase activity to selectively suppress resistant bacteria producing extended-spectrum-β-lactamases (ESBLs). Selective targeting of resistant bacteria requires avoiding interaction with penicillin-binding proteins, the conventional targets of β-lactam antibiotics, while maintaining recognition by ESBLs to activate AcephPT only in resistant cells. We show AcephPT selectively suppresses gram-negative ESBL-producing bacteria in clonal populations and in mixed microbial cultures, with effective selectivity for both lab strains and clinical isolates expressing ESBLs. Time-course NMR experiments confirm hydrolytic activation of AcephPT exclusively by ESBL-producing bacteria. In mixed microbial cultures, AcephPT suppresses proliferation of an ESBL-producing strain while sustaining growth of β-lactamase-non-producing bacteria, highlighting its potential to combat β-lactam resistance while promoting antimicrobial stewardship.

Graphical Abstract

Drugs containing a β-lactam core are the most prescribed antibiotics in the United States.¹ Resistance to β-lactam antibiotics, however, is a serious and growing threat to human health, posing challenges worldwide for the prevention and treatment of infections.^2,3 A significant resistance mechanism in gram-negative bacteria is the acquisition and production of β-lactamases (BLs), enzymes that degrade β-lactam antibiotics.^4,5 The rise of extended-spectrum-β-lactamases (ESBLs) that degrade all classes of β-lactams makes the development of new strategies to overcome BL-mediated resistance an urgent medical need.

Counterproductively, the wide use of β-lactams has fueled the spread of resistance. Like other broad-spectrum antibiotics, β-lactams act indiscriminately on both pathogenic and commensal bacteria, thus negatively perturbing the diversity and composition of the host microbiome.^6–8 Microbiomes are complex communities of organisms, genes, and metabolites that interact in myriad ways with each other and the host to maintain health. Depleting resident microbiota can diminish the homeostatic mechanisms within a microbial community that otherwise restrict pathogen growth. Altered access to nutrients or loss of production of bioactive metabolites can provide growth opportunities for selection and enrichment of opportunistic pathogens, including those that are drug resistant.^9–12 Administering broad spectrum antibiotics to resistant infections incurs risk of gastrointestinal diseases by commensal fungal and C. difficile infection, as well as overpopulation of drug-resistant strains of Enterococcus, Salmonella, and others.^6,10,13 The long-term and cumulative collateral damage of antibiotic-induced dysbiosis on individual patients is still being elucidated,^14–18 with the growing list of negative health effects including various allergic, metabolic, immunological and inflammatory conditions with broad-reaching impact on intestinal, neurological, cardiovascular, and respiratory systems.^6,7,14,19

Given the negative consequences of broad-spectrum antibiotics, treatment strategies are needed that eliminate infection while preserving the diversity and function of healthy polymicrobial communities.¹⁴ Desirable agents would maintain the benefits of broad-spectrum coverage to increase the probability of clearing an infection of unknown composition, while simultaneously minimizing collateral damage and the spread of resistance (Fig. 1a). Toward this goal, we introduce AcephPT, a reactivity-based prodrug intentionally designed to turn BL reactivity into a vulnerability to achieve selective suppression of β-lactam resistant bacteria among polymicrobial mixtures (Fig. 1b).

Figure 1. Designing antibacterial prodrugs to overcome the limitations of current antibiotics.

a, Model of bacteria communal growth from different treatment strategies. Classic antibiotics (left) are ineffective at suppressing the growth of ESBL-producing bacteria, while there is collateral damage of ESBL-non-producing bacteria. A prodrug that is selectively activated by ESBL-producing bacteria (right) suppresses the growth of the resistant population while sparing the non-producing bacteria. b, Schematic of the prodrug AcephPT, which remains intact and ineffective at killing ESBL-non-producing bacteria (left) but is activated by ESBL-producing bacteria (right) leading to localized release of pyrithione, a bactericidal agent that also has metallo-β-lactamase inhibiting properties.

AcephPT’s design is rooted in principles of bacterial population dynamics and of molecular recognition. In a mixed bacterial community, strains with a growth advantage will overtake the population. Conversely, strains that experience a burden will have a growth disadvantage in a communal setting, allowing other strains to thrive. Broad spectrum antibiotics impose a burden on commensal strains, but not on resistant strains, thereby favoring population takeover by the resistant strain. Not only do conventional antibiotics fail to impose a burden on resistant strains, but the resistance mechanisms carried by these strains also may not impose a significant burden. Whereas chromosomally encoded resistance determinants typically incur a fitness cost to bacteria, plasmid-encoded resistance does not necessarily impose a fitness burden. For example, it has been found that plasmids carrying multiple resistance genes, including ESBLs, can be cost neutral to bacteria that carry them, a phenomenon that can allow resistance to persist in bacterial populations even in the absence of antibiotic pressure.^20,21

We envisioned that reactivity-based prodrugs with preferential activation by ESBLs could provide a strategy to selectively burden and suppress resistant bacteria while reducing collateral damage. Selectively suppressing the proliferation of resistant, foreign strains (i.e. pathogens) will allow other non-producers in the community (i.e. commensals) to outcompete the pathogen, even if the agent does not eliminate the pathogen. Here we describe a cephalosporin prodrug that successfully biases reactivity toward ESBLs to release a cytotoxin, while remaining unreactive and therefore nontoxic to bacteria that lack BL. Importantly, the selectivity observed against clonal populations is maintained in a mixture, showing for the first time that a BL-activated prodrug can selectively suppress a drug-resistant pathogen even in a mixture containing drug-sensitive bacteria.

Results

AcephPT selectively suppresses ESBL-producing bacteria in clonal populations

To achieve selective targeting of ESBL-expressing pathogens, we designed prodrugs to co-opt the BL enzymatic reactivity that generates resistance to conventional β-lactam antibiotics to release agents that preferentially act on resistant cells without collateral damage to bacteria that do not produce BL. The design focuses on the specific challenge of achieving reactivity bias for ESBL producers vs nonproducers, and does not explicitly address design challenges surrounding uptake or efflux, for example by strains with porin modifications or efflux pump overexpression. The premise for BL-mediated release of a cytotoxin is known from previous efforts to develop dual-action β-lactams,^22–31 but these molecules do not achieve and were not designed for selective killing amongst polymicrobial communities. Indeed, the molecular features of dual-action β-lactams were optimized for broad-spectrum activity against both sensitive and resistant organisms.³⁰ Through principles of molecular recognition, we reverse-engineered cephalosporin scaffolds to be poor substrates of penicillin-binding proteins (PBPs), the biological targets of conventional β-lactam antibiotics, while retaining substrate recognition with ESBLs. Our strategy differs from conventional approaches that optimize chemical structures to prevent degradation by BLs. Traditional broad-spectrum antibiotics generally possess bulky substituents on their β-lactam rings to increase engagement with PBPs, whereas serine-β-lactamase (SBL) inhibitors do not. Since the inhibitors target SBLs and avoid interaction with PBPs, we prepared AcephPT and AcetamidocephPT with small functional groups at this position.

AcephPT and AcetamidocephPT are in the family of cephalosporin prodrugs that hijack a resistance mechanism to release an active agent upon hydrolytic cleavage of the β-lactam ring by BL enzymes.^22–29 Like our previous generation prodrug PcephPT,²⁸ AcephPT is designed to release pyrithione, a metal-binding antimicrobial linked to the cephalosporin core via one of its metal-binding atoms. As we previously showed for PcephPT, the conjugate has minimal metal-binding ability or cytotoxicity when pyrithione is linked in the prodrug state. However, in BL-producing Escherichia coli this linkage is broken to release antimicrobial pyrithione (Fig. 1b).²⁸ Our mechanistic investigations revealed two modes of action for PcephPT. Against strains expressing SBLs, the in-situ released pyrithione resulted in bacterial cell death. In carbapenem-resistant strains expressing the metallo-β-lactamase NDM-1, PcephPT also acted as an auto-inhibitor, with released pyrithione being retained at the Zn-active site of NDM-1, allowing NDM-1-producing bacteria to become sensitive to β-lactam antibiotics.^27,32 While PcephPT was active against BL-producing E. coli, it also retained conventional β-lactam activity and was cytotoxic against bacteria not producing BLs.²⁸ This off-target activity is undesirable, as it would leave commensals present in a healthy microbiome susceptible to suppression.

To evaluate the relative activity of AcephPT and AcetamidocephPT compared to PcephPT, we used a phenotypic screen against BL producers vs nonproducers. The compounds were tested against a BL-non-producing strain of E. coli MG1655 and MG1655 engineered to express a panel of BL enzymes consisting of serine-BLs (SBLs): OXA-1, TEM-1, CMY-2 and CTX-M-1, as well as zinc-containing MBLs: NDM-1, VIM-2, and IMP-1. Among this series, CTX-M-1, NDM-1, VIM-2, and IMP-1 (Fig. 2a-c, greens) are classified as ESBLs and carbapenemases, presenting promiscuous β-lactam recognition that confers resistance to a wide range of antibacterials, including last-resort carbapenems.^33–39 OXA-1, TEM-1, CMY-2 are also clinically significant BLs with narrower scope of β-lactam recognition compared to ESBLs and carbapenemases (Fig. 2a-c, grays).^40–42 As a positive control, all our engineered bacteria were challenged by free pyrithione, for which EC₅₀ values ranged from 15–20 μM (Fig. S24). The narrow range in pyrithione EC₅₀ against each of these strains indicates BL expression does not impact pyrithione’s activity. The average of these bacterial growth dose-response curves is depicted as the blue curve for nonconjugated PT in Fig. 2a-c and represents the toxicity threshold for the active agent. In BL-producing strains, an EC₅₀ value for pyrithione-conjugated prodrugs near the pyrithione toxicity threshold of 17 μM indicates drug activation by BLs.

Figure 2. AcephPT selectively suppresses ESBL-producing lab variety bacteria and clinical isolates.

Dose-response curves of a, PcephPT b, AcetamidocephPT, and c, AcephPT treatment of E. coli K-12 MG1655 engineered to express the indicated BLs, with a non-producer shown in translucent black and those classified as ESBLs in green. Treatment of pyrithione as a non-conjugated small molecule shown in blue. d, Dose-response curves of AcephPT against a panel of clinical isolates producing BLs (see Table S1 for species and BLs produced). Points and error bars are mean ± SE from 3 biological replicates of 4 technical replicates (N = 12 each group). For Nonconjugated PT, points and errors bars are mean ± SE from 3 biological replicates from each bacterial strain tested (N = 24 each group). Fitted lines are logistic functions determined from non-linear regression.

PcephPT inhibited growth of all strains tested, with EC₅₀ values ranging from 6–40 μM (Fig. 2a). The BL-non-producing strain was the most sensitive while the NDM-1-producing strain was the least sensitive, highlighting PcephPT’s lack of desired selectivity for BL-producing bacteria.

Removing the phenyl moiety from the R¹ functionalization to produce AcetamidocephPT resulted in ten-fold decreased activity against BL-non-producing bacteria compared to PcephPT (Fig. 2b). Although AcetamidocephPT was least toxic to the strain producing the penicillinase OXA-1, for all other BL-producing strains tested AcetamidocephPT retained a similar level of activity to PcephPT, with EC₅₀ values ranging from 15–36 μM. Removal of the phenyl moiety enhanced the selectivity of this pryrithione conjugated prodrug by retaining activity with BL-producing bacteria and by better sparing non-resistant bacteria.

Following the trend of decreasing chemical space at the R¹ position of the prodrug core, removal of the acyl group on AcetamidocephPT leaves the amine functionality of AcephPT. When tested against our BL-non-producing bacteria, AcephPT was remarkably non-toxic (Fig. 2c), as no growth suppression was observed for the concentrations tested. The strains producing the BLs OXA-1, TEM-1, and CMY-2 were also less sensitive to AcephPT, with EC₅₀ thresholds not being reached by the concentrations tested (EC₅₀ > 256 μM). The strains producing the ESBLs CTX-M-1, NDM-1, VIM-2 and IMP-1, however, were sensitive to AcephPT, with EC₅₀ values ranging from 14–43 μM. The non-ESBL BL-producing variants do not recognize AcephPT, while the more promiscuous ESBL-producing bacteria recognize and are sensitive to AcephPT. Notably, AcephPT is active against BL types that produce the greatest amount of resistance to current antibiotics.

Given AcephPT’s differential activity against our panel of engineered BL-non-producing and ESBL-producing bacteria, we tested for toxicity against mammalian cells and across a panel of bacterial clinical isolates acquired from patients treated in the United States. Using an LDH activity assay, no cytotoxicity was observed for AcephPT or PT treatment against Caco-2 cells up to 1 mM. Against HepG2 cells, cytotoxicity was below 40 % for AcephPT and PT at the concentrations tested (Fig. S23). The panel of clinical isolates was selected to represent different BL types and bacterial species (Fig. 2d and Table S1). For all strains except the ARLG 3667 Pseuodomonas aeruginosa, the positive control of free pyrithione was effective at suppressing the growth of the BL-producing clinical isolates, with EC₅₀ ranging from 11–37 μM (Fig. S25).

AcephPT’s EC₅₀ values determined against E. coli, Klebsiella pneumoniae, and Enterobacter cloacae clinical isolates (Fig. 2d) were similar to those found for our BL-producing engineered E. coli (Fig. 2d). The exceptions to this trend were ARLG 3667, the P. aeruginosa isolate that was also not susceptible to free pyrithione, and DICON isolates 007 and 029. The observation that AcephPT was ineffective against DICON 007, an E. coli isolate that produces TEM-1, is consistent with the poor activity found against our engineered strain producing TEM-1, which is not considered an ESBL (Fig. 2c). DICON 029 has been annotated to coproduce CTX-M-15, an ESBL, and OXA-1, not an ESBL. The weak activity of AcephPT against this strain may suggest additional mechanisms of resistance, such as reduced membrane permeability or active efflux. Overall, AcephPT suppressed the growth of clinically relevant bacteria that are otherwise resistant to broad-spectrum antibiotics.

AcephPT is hydrolytically activated exclusively by ESBL-producing E. coli

Results from the dose-response curves are consistent with a mechanism in which the β-lactam structure of AcephPT is stable to non-specific hydrolysis, yet susceptible to ESBL-triggered release of pyrithione as the active antimicrobial agent. To verify the molecular details of drug activation in cells, we adapted⁴³ and designed a time-course ¹H NMR spectroscopy experiment to profile the activation of AcephPT in our engineered ESBL-non-producing and ESBL-producing E. coli strains. Proton resonances of the prodrug were observed and monitored for consumption by the expressed ESBL. Changes in resonance frequencies indicate a new species is being produced, such as the release of pyrithione from the conjugate. As shown in Fig. 3a, AcephPT remained intact with no hydrolysis or other decomposition being observed upon incubation with the BL-non-producing strain across the incubation time. In contrast, notable changes in the resonance peaks of AcephPT are observed upon incubation with the strain that expresses the SBL CTX-M-1 (Fig. 3a, middle). As the starting prodrug’s resonances decrease in intensity, resonances corresponding to free pyrithione increase in the aromatic region (8.4–6.9 ppm) with the identical intensity change. Additionally, the appearance of alkenyl resonances in the 5.9–5.6 ppm region is indicative of alkene formation within the hydrolyzed prodrug fragment (see Fig. 1b), which is mechanistically anticipated to form during the release of pyrithione.^23,24,27 Similar resonance changes were observed when AcephPT was incubated with our engineered NDM-1-producing (Fig. 3a, bottom) and IMP-1-producing strains (Fig. S26).

Figure 3. AcephPT is selectively activated by ESBL-producing E. coli.

a, Time-course NMR spectra profile the chemical changes of the prodrug AcephPT incurred upon incubation at 37 °C with whole cell E. coli for (top) BL-non-producing, (middle) CTX-M-1-producing, and (bottom) NDM-1-producing E. coli. The unchanged resonances in the top panel show stability of AcephPT in the non-producer strain. b, c, d, Speciation curves for the conversion of AcephPT to pyrithione show that (b) CTX-M-1-producing E. coli hydrolyze all available AcephPT while (c) NDM-1-producing E. coli hydrolyze an initial amount of AcephPT rapidly before the reaction stalls. (d) IMP-1-producing E. coli hydrolyze nearly all AcephPT, but addition of supplemental pyrithione (PT) inhibits AcephPT hydrolysis. Each point is an average of the integrations for resonances in the aromatic region (6.9–8.5 ppm) for the corresponding species scaled to a relative amount of AcephPT (t₀).

When AcephPT was incubated with our CTX-M-1-producing strain, the intensity of the resonances observed for the prodrug’s consumption matched the intensity of the peaks corresponding to free pyrithione (Fig. 3b), indicating stoichiometric and complete conversion from prodrug to active drug. When incubated with our NDM-1-producing strain, AcephPT hydrolysis was initially more rapid than with the CTX-M-1-producing strain, as more than 0.2 molar equivalents of AcephPT were converted to free pyrithione by the acquisition of the first spectrum (Fig. 3c). This activity does not persist, as the hydrolysis of AcephPT by the NDM-1-producing strain significantly slows down, leaving 0.7 molar equivalents of the prodrug still intact after two hours. This stalling of prodrug activation is reminiscent of our previous work with PcephPT, in which NDM-1 undergoes product inhibition by pyrithione released upon prodrug activation.²⁷ In contrast, prodrug activation occurred fastest and most completely against the IMP-1 strain, as AcephPT was nearly completely hydrolyzed even before the acquisition of the first spectrum (Fig. 3c). IMP-1, like NDM-1, is an MBL capable of being inhibited by molecules containing metal-binding moieties.^44,45 To determine if IMP-1 activity can be inhibited by pyrithione, the whole cell mixtures were supplemented with free pyrithione before the addition of AcephPT. Indeed, supplemental pyrithione showed a dose-dependent inhibition of AcephPT hydrolysis in the IMP-1 strain.

Holistically, these in-cell ¹H NMR experiments demonstrate the excellent stability of AcephPT in aqueous buffered and cellular environments at physiological temperature in the absence of BL-producing bacteria. Furthermore, they validate the release of pyrithione upon hydrolysis of AcephPT by ESBL-positive bacteria, with different extents of product inhibition for MBL-producing strains.

AcephPT selectively suppresses a drug-resistant clinical isolate in a pairwise mixed microbial environment.

To determine the selectivity of the prodrug in a communal setting, we created mixed cultures of 1:999, 1:99, and 1:9 proportions of ESBL-producing to non-producing strains (Fig. 4a). The ESBL-producing strain was the pathogenic DiRTE 16096 E. coli clinical isolate that expresses NDM-1 (AcephPT EC₅₀ = 29 μM). To model a BL-non-producing “commensal” strain, we used an engineered laboratory Top10 E. coli that expresses mCherry and lacks β-galactosidase (AcephPT EC₅₀ > 256 μM, Fig. S27). The mixed cultures were prepared from individual overnight cultures equalized to an OD₆₀₀ of 1.0. The cultures were diluted 100-fold and were untreated or treated with cefotaxime (a third-generation β-lactam) or AcephPT in 96-well plates. While the mixtures were incubating over 24 hours, OD₆₀₀ readings were taken to indicate total culture density of the mixture (Fig. 4b) and fluorescence intensity measurements (Ex₅₈₆/Em₆₂₀) were taken to monitor mCherry expression (Fig. 4c), a distinct reporter of growth of the Top10 BL-non-producing strain.

Figure 4. In communal mixtures, AcephPT suppresses the growth of ESBL-producing E. coli while recovering the growth of non-producers.

a, Schematic of communal experimental design. Different producer:non-producer ratios were prepared from individual overnight starter cultures before being treated. Conditions were monitored optically for b, culture growth by OD₆₀₀ and c, for expression of mCherry by fluorescence (Ex₅₈₆/Em₆₂₀) for 24 h. b, Growth curves of cultures for each producer:non-producer ratio under each treatment condition. The producer strain (top left) is sensitive to AcephPT treatment but not cefotaxime. The non-producer strain (top right) is sensitive to cefotaxime treatment but not AcephPT treatment. The producer strain grows faster than the non-producer strain. All communal mixtures show no change in growth upon cefotaxime treatment. Treatment with AcephPT shows growth suppression for each mixture, with the greater starting ratios of producers being more sensitive to AcephPT. Each curve is a biological replicate, showing 3 for each group. c, Fluorescence measurements taken in tandem with culture growth curves to monitor expression of mCherry. No fluorescence is detected for producers (top left), as they do not express mCherry. For the non-producers (top right), fluorescence increases over time in the untreated and AcephPT conditions, but not the cefotaxime-treated conditions, as the non-producers are sensitive to this antibiotic (b, top left). All communal mixtures show no change in fluorescence in the untreated and cefotaxime-treated conditions, indicating culture take over by the producers. Fluorescence for the AcephPT treated conditions are present in the 1:999 and 1:99 mixtures, indicating growth recovery of the non-producer. Each curve is a biological replicate, showing 3 for each group. d, CFU/mL enumerations of viable producers (greys) and non-producers (greens) for different treatment conditions determined from blue/white (producers/non-producers) screening. Cefotaxime treatment results in no change culture ratio compared to untreated, indicating it cannot combat the ESBL-producing population. AcephPT suppresses growth of the ESBL-producing population at all starting culture ratios and recovers growth of the non-producing population. AcephPT’s activity is dose-dependent in communal environments, as changes in culture composition are more extreme from 100 to 250 μM. Bars are average ± SE from 3 biological replicates of 2 technical replicates (N = 6 each group). Black circles are individual data points; points under the axis indicate no viable growth for that replicate. Data areas with NG labels had no viable cell growth for that condition. e, Fraction of ESBL-producing E. coli determined from viable cell enumerations (d) by dividing raw counts of producers by the total for each ratio/treatment condition. AcephPT suppresses the growth of ESBL-producing E. coli. Bars are average ± SE from 3 biological replicates of 2 technical replicates (N = 6 each group). Black circles are individual data points; points under the axis have an undefined value as no viable producers or non-producers were found for the given replicate.

The ESBL-producing strain is sensitive to AcephPT treatment, but not cefotaxime (Fig. 4b, top left). In contrast, the BL-non-producing strain (Fig. 4b, top right) is sensitive to cefotaxime but not AcephPT. The ESBL-producer grows faster than the non-producer, as the culture density is greater in the untreated condition for the ESBL-producer. For the communal mixtures (Fig. 4b, middle and bottom) treated with cefotaxime, there is no change in culture density from their respective untreated conditions. Treatment with AcephPT, however, results in suppressed growth for each mixture compared to untreated or cefotaxime-treated conditions (Fig. 4b, middle and bottom). Decreases in overall culture density from AcephPT treatment correlates with the starting proportion of producers – as the starting culture ratio of producers increases, the AcephPT treated communal culture grows less.

The fluorescence measured in tandem with culture growth curves were used to monitor expression of mCherry and therefore growth of the non-producer. No fluorescence was detected for producers (Fig. 4c, top left), as they do not express mCherry. For the non-producers (Fig. 4c, top right), fluorescence increases over time in the untreated and AcephPT conditions, correlating well with increases in the cell density curves (Fig. 4b, top right). The cefotaxime-treated non-producer condition presented no increase in fluorescence, validating that non-producers are sensitive to this antibiotic (Fig. 4b, top right). For the non-producers, the lack of fluorescence from cefotaxime treatment but increase in fluorescence from AcephPT treatment means the non-producers are sensitive to cefotaxime but not sensitive to AcephPT. None of the communal mixtures showed any increase in fluorescence in the untreated and cefotaxime-treated conditions, indicating culture takeover by the producers. This producer takeover highlights the importance of bacterial growth dynamics in complex environments with and without exposure to classic antibiotics. In contrast, fluorescence for the AcephPT treated conditions increase in the 1:999 and 1:99 mixtures, indicating growth recovery of the non-producer (Fig. 4c, middle). Together, these data indicate AcephPT is targeting the resistant bacteria and allowing for recovery of the ESBL-non-producing bacteria.

While the fluorescence readout of mCherry expression allows for the rapid screening of ESBL-producing and ESBL-non-producing pairs, it does not quantify the proportion of viable bacteria from each strain. To enumerate colony-forming units (CFUs) of viable bacteria, treatment and mixture combinations were grown for 24 hours, diluted, and spotted onto agar plates containing X-Gal and IPTG (Fig. 4d). Since the E. coli Top10 engineered non-producer strain does not have β-galactosidase, these colonies appear white, allowing for differentiation from the blue colonies of the clinical isolate producer strain during blue-white screening to quantify viable cells of each type.⁴⁶ In the 1:99 and 1:9 mixtures, no growth of non-producer bacteria was observed in the conditions treated with vehicle or cefotaxime. This result recapitulates those depicted in Fig. 4b-c: the growth rate of the producer allows for its takeover of the communal environment. Occurring in starting ratios as low as 1:99, this takeover again highlights the importance of bacterial growth dynamics in mixed cultures. In contrast, AcephPT significantly suppresses the producer population in all mixtures, while encouraging the growth of the non-producers (Fig. 4d). AcephPT selectively suppresses the ESBL-producing strain in a manner depending on both the starting culture ratio and the dose. The proportion of pathogenic cells in the mixed population treated with AcephPT was reduced up to 100-fold compared to untreated and cefotaxime-treated mixtures.

To highlight the suppression of the producer, the fraction of producers remaining after treatment was determined for each treatment and mixture combination (Fig. 4e). Cefotaxime presented no suppression of the producer population, being no different from the untreated conditions. AcephPT, in contrast, suppressed the growth of ESBL-producing E. coli. These key data show prodrug AcephPT preferentially suppresses the producer population in each mixed microbial environment, allowing reversal of the fraction of pathogenic vs commensal bacteria, in favor the ESBL-non-producing bacteria.

These results reinforce that treatment with a classic antibiotic such as cefotaxime is not only ineffective for combating a drug-resistant mixture but selects for growth of drug-resistant bacteria. AcephPT overcomes this limitation by exploiting the resistance mechanism to suppress the growth of the ESBL-producer and promote the growth of the non-producer.

Discussion

Antibiotic use has counter-productively created selection conditions for spreading extended spectrum β-lactamases (ESBLs), which have promiscuous recognition and enhanced kinetic activity to hydrolyze and inactivate β-lactam antibiotics. Prodrug AcephPT was designed to take advantage of these properties by being purposefully hydrolyzed, and therefore activated, by ESBLs to become cytotoxic. Remarkably, AcephPT does not affect BL-non-producing bacteria, consistent with the hypothesis of diminished recognition by penicillin-binding protein (PBP) that was envisioned by its molecular design. While the phenotypic screen used here is naïve to potential off-target interactions that may be occurring with nonessential PBPs or other proteins, the outcome provides evidence that a small molecule can be engineered to be selective for resistant bacteria. AcephPT demonstrates a first-in-class ability to suppress ESBL-producing, pathogenic bacteria while simultaneously allowing ESBL-non-producing bacteria to recover. By turning ESBL expression into a weakness, reactivity-based prodrugs like AcephPT may improve antimicrobial stewardship by selectively suppressing resistant populations.

Key to the design strategy that led to AcephPT was de-engineering the susceptibility of non-ESBL-producing strains to what is otherwise a classic cephalosporin antibiotic scaffold. The unabated growth of non-ESBL-producing E. coli in the presence of increasing concentrations of AcephPT substantiated this idea, which was validated by whole cell NMR experiments confirming the molecular stability of AcephPT against unintended hydrolysis reactions or degradation in a cellular context of non-ESBL-producing E. coli. Furthermore, the NMR experiments were critical for characterizing the chemical products of prodrug activation by ESBL producers. This characterization is important for establishing the mechanism of action, as hydrolysis of other reactivity-based β-lactam compounds has been shown to occur without release of the leaving group.²³

Quantifying the release of pyrithione by bacteria expressing different ESBLs further revealed that rates of release and level of autoinhibition vary based on the expressed ESBL. To be amenable to the sensitivity of the NMR experiment, the cell densities and concentrations of compound used for these experiments were significantly greater than those used in microculture cell assays. Under these conditions, the in-situ NMR experiments revealed variable levels of auto-inhibition by strains expressing MBLs. Comparatively, IMP-1 was inhibited less than NDM-1 under these conditions as IMP-1 turned over more AcephPT without reaching autoinhibition, although it was found to be inhibited upon adding free pyrithione. These observations raise an interesting conundrum, as autoinhibition by reactivity-based prodrugs could undercut the mechanistic advantage providing their selectivity. When the active agent can serve as both cytotoxic agent and inhibitor, as it does for AcephPT, the interplay of the rate of release, the EC₅₀ of the cytotoxic agent, and the enzyme inhibition constant will influence the overall capability of generating sufficient biocide without prematurely inhibiting the enzyme that releases it. A further consideration is the possibility of collateral damage by the released agent, which if overproduced could become available to inflict nonspecific damage to cells within its diffusion reach, or upon lysis of BL-producers. Under these scenarios, the dynamics of autoinhibition of MBLs may be favorable for self-tempering such an outcome. The observation that MBL-expressing strains were the most sensitive of the strains tested to growth inhibition by AcephPT (Fig. 3c) suggests the conditions in the microdilution experiments were such that sufficient pyrithione was released to reach efficacious levels to inhibit growth of the lab strain monocultures without reaching auto-inhibition levels. In the mixed microbial cultures, the additional factor of population dynamics comes into play.

As our relatively simple two-strain pairwise mixtures show, even a small percentage of a faster-growing pathogen readily outcompetes to take over the whole culture (Fig. 4d). While pyrithione was found not to be bactericidal to the clinical isolate producer strain used in the mixed microbial experiment, it was bactericidal to the non-producer strain at 250 μM (Fig. S27). This difference in cidality between the two strains explains why the producer was not fully eliminated in our mixed microbial model and implies pyrithione was not over-produced to reach collateral bactericidal levels available to the non-producer. Although the producing strain was not eliminated, the reduced growth of the strain is noteworthy, and the recovery of the slower-growing non-producer, most evident at the lowest ratios of producer, is remarkable.

Our work illustrates AcephPT as a potential targeted therapeutic agent against drug-resistant bacteria, opening the door for further development of antibacterials that are selectively recognized by bacteria producing ESBLs. While this work focused on ESBL-mediated resistance for design inspiration, there are, of course, other resistance and drug tolerance mechanisms that are distinct from BL expression. The current work raises new questions regarding the scope of ESBL-targeted prodrugs against microbial communities that harbor multiple resistance determinants and susceptibilities. We demonstrate the pharmacophore pyrithione can selectively suppress ESBL-positive pathogenic bacteria by its conjugation to a cephalosporin, with modification to its β-lactam core that maintains activity with ESBLs but appears to disfavor PBP recognition. The enzymatic activity generating drug resistance (i.e. cleavage of β-lactam drugs by BLs) can be hijacked to release antimicrobial compounds that selectively suppress the pathogens harboring these resistance enzymes, while sparing others.

Materials and Methods

General Chemical Information

Synthetic procedures and characterization details for all compounds can be found in the Supporting Information. All reagents and solvents were purchased from commercial sources (Millipore Sigma, Oakwood Chemical, Fischer Scientific, AA Blocks, Combi Blocks) and used as received. Silica gel (230–400 mesh) was used as a stationary phase for column chromatography. Eluent details are listed with the associated compound in the Supplementary Information. TLC was performed on silica gel 60 F₂₅₄. No unexpected or unusually high safety hazards were encountered.

NMR Spectroscopy

¹H, ¹³C{¹H}, DEPT-90, and DEPT-135 NMR data were acquired on a Bruker Ascend 500 MHz instrument at ambient temperature. Chemical shift values are reported in ppm with coupling constants in Hz. Abbreviations for multiplicity are: br = broad singlet, s = singlet, d = doublet, t = triplet, q = quartet, dd = doublet of doublets, dt = doublet of triplets, ABq = H_AH_B quartet, m = multiplet. The solvent signal for each spectrum was used for axis calibration. NMR data were processed using a Bruker TopSpin academic license. Resonance assignments and spectra for ¹H and ¹³C{¹H} spectra are available in the Supplementary Information.

Mass Spectrometry

Liquid chromatography/mass spectrometry (LC/MS) analyses were acquired on an Agilent 1260/6460 instrument. Positive-ion mass spectra were acquired in full-scan mode over the range of 100–2500 m/z using the following source parameters: gas temperature 300 °C, gas flow 5 L/min, nebulizer pressure 45 psig, VCap 3500 V, and fragmentor voltage 135 V. RP-HPLC separations were achieved on a Phenomenex Luna C18(2) column (100 mm x 2 mm, 3 μm, 100 Å) using a linear gradient of mobile phase B in A, a flow rate of 0.5 mL/min, and a column temperature of 40 °C. Mobile phase A was prepared by combining 400 mL ultrapure water with 12 mL methanol and 1.2 mL formic acid. Mobile phase B was prepared by combining 400 mL acetonitrile with 12 mL ultrapure water and 1.2 mL formic acid. The gradient program included an initial hold at 0 % B for 2 min, linear increase to 60 % B from 2–9 min, linear increase to 100 % B from 9–10 min, hold at 100 % B from 10–12 min, linear decrease to 0 % B from 12–13 min, final hold at 0 % B from 13–15 min for a total run time of 15 min. HILIC HPLC separations were achieved on a SeQuant ZIC-HILIC column (100 mm x 2.1 mm, 3.5 μm, 100 Å) using a linear gradient of mobile phase B in A, a flow rate of 0.5 mL/min, and a column temperature of 40 °C. Mobile phase A was 5 mM ammonium formate at pH 2.5, prepared by dissolving ammonium formate in ultrapure water and adjusting pH with formic acid. Mobile phase B was prepared by combining 400 mL acetonitrile with 12 mL ultrapure water and 1.2 mL formic acid. The gradient program included an initial hold at 95 % B for 2 min, linear decrease to 40 % B from 2–9 min, linear decrease to 5 % B from 9–10 min, hold at 5 % B from 10–12 min, linear increase to 95 % B from 12–13 min, final hold at 95 % B from 13–15 min for a total run time of 15 min. In addition to MS detection, the DAD was used to acquire a UV chromatogram at 220 nm, 254 nm, and 280 nm. Samples were analyzed using a 2 μL injection volume. Processed raw data files were exported from Agilent MassHunter and plotted in Microsoft Excel. UV chromatograms and mass spectra of isolated compounds are available in the Supplementary Information.

High-resolution mass spectra (HRMS) were acquired on an Agilent 1200/6224 instrument. The mass spectrometer was equipped with a Dual ESI source, and accurate mass data were obtained by internal calibration (reference ions 121.050873 and 922.009798 m/z) using a secondary nebulizer to continuously deliver the reference solution. Positive-ion mass spectral data were acquired in full-scan mode over the range of 100–3200 m/z using the following source parameters: gas temperature 325 °C, gas flow 11 L/min, nebulizer pressure 33 psig, VCap 3500 V, and fragmentor voltage 220 V. Samples were analyzed using a 1–5 μL injection volume. HRMS results of isolated compounds are available in the Supplementary Information.

Communal Growth Model

For simulations shown in Figure 1, ordinary differential equations were formulated to model the qualitative dynamics of a bacterial system with ESBL-producing and BL-non-producing subpopulations responding to an antibiotic or a prodrug. The dynamics were formulated as the interactions between six main components: ESBL-producing population density (n_r), BL-non-producing population density (n_s), nutrient level (s), antibiotic concentration (a), extracellular ESBL concentration (b), prodrug concentration (p), and released drug concentration (e). In this model, ESBL production extracts a fitness cost (α) and grants resistance to the antibiotic (β) relative to the BL-non-producing population. Nutrient is consumed by growth and released incompletely (ξ) with lysis. Antibiotic degradation and prodrug cleavage are mediated either by extracellular ESBLs (κ_b, κ_p) which are released on lysis, or living ESBL producers (φ, σ). All simulations were conducted in MATLAB R2023b. Details of the differential equations and the model can be found in Supplementary Information.

Whole Plasmid Sequencing

Transformed E. coli were grown overnight at 37 °C on Luria Bertani Broth (Lennox) (LB) agar plates containing the appropriate antibiotics (50 μg/mL kanamycin and/or 100 μg/mL ampicillin). A single colony was used to inoculate 5 mL of liquid LB with the appropriate antibiotics and grown at 37 °C and 200 rpm for 20 h. Cells were harvested by centrifugation (4000 rcf, 5 min, rt), resuspended in 0.5 mL of ultrapure water, and DNA was isolated using Zymo Research Zyppy Plasmid Miniprep Kit. Purified DNA was sent to plasmidsaurus for whole-plasmid Oxford Nanopore sequencing.

Monoclonal Microplate Microdilution Assays

Bacteria from frozen glycerol stocks streaked onto LB agar plates containing the appropriate antibiotics (50 μg/mL kanamycin and/or 100 μg/mL ampicillin) were grown at 37 °C for 20 h. With a single colony from the agar plate, a 2 mL culture of bacteria in MHB 2 media with the appropriate antibiotics was grown at 37 °C and 200 rpm for 20 h. Prodrug stock solutions (100 mM in DMSO) were diluted to 4x of the highest tested concentration with MHB 2 media. A final solution volume of 200 μL was used for all wells in a 96-well microplate. In a 96-well microplate, 100 μL 2x serial microdilutions were performed across the long axis of the plate. Bacteria from the 2 mL culture were diluted 1:500 in 10 mL of MHB 2 media and 100 μL was added to the appropriate wells in the prepared 96-well microplate (1:1000 bacteria dilution total). The 96 well plate was incubated at 37 °C and 200 rpm for 20 h. Each microplate had a media dam around the outside of the plate, an entire row of no treatment growth control, and a positive control 2x serial microdilution of pyrithione (20 mM in DMSO), prepared as described for the prodrug microdilutions. Using a Perkin Elmer Victor³ V 1420 plate reader, OD₆₀₀ was read at the initial time of plating in the microplate and subtracted from the OD₆₀₀ readings after 20 h incubation. These values were normalized by division to the no treatment growth control for their given column.

Bactericidal Assay

Bacteria from frozen glycerol stocks streaked onto LB agar plates containing the appropriate antibiotics (50 μg/mL kanamycin and/or 100 μg/mL ampicillin) were grown at 37 °C for 20 h. With a single colony from the agar plate, a 2 mL culture of bacteria in LB media with the appropriate antibiotics was grown at 37 °C and 200 rpm for 18 h. Overnight cultures were equalized to OD₆₀₀ 1.0 and diluted 1000-fold in a 96-well plate filled with liquid media containing a final concentration of either 100 μM (12.7 μg/mL) pyrithione, 250 μM (31.8 μg/mL) pyrithione, 100 μM (37.5 μg/mL) AcephPT, 250 μM (94.0 μg/mL) AcephPT, or pure liquid media. The 96-well plate was incubated at 37 °C and 200 rpm for 20 h. After incubation, the contents of each of the wells were 10x serially diluted and drip-plated on LB agar plates. After 16 h of incubation at 37 °C, colony counts were enumerated to determine viable bacteria in CFU/mL.

Profiling Hydrolysis of Prodrugs in Whole Cell Bacteria by NMR Spectroscopy

BL-non-producing or BL-producing engineered E. coli K-12 MG1655 from frozen glycerol stocks streaked onto LB agar plates containing the appropriate antibiotics (50 μg/mL kanamycin and/or 100 μg/mL ampicillin) were grown overnight at 37 °C for 20 h. With a single colony from the agar plate, a 5 mL culture of the bacteria in LB with the appropriate antibiotics was grown at 37 °C and 200 rpm for 20 h. The 5 mL culture was poured into 125 mL of LB media with the appropriate antibiotics and grown at 37 °C and 200 rpm to OD₆₀₀ of 0.1–0.2. The cells were harvested by centrifugation (2,500 rcf, 20 min, 4 °C) and washed three times with 50 mM phosphate buffer at pH 7.5. The cells were taken up with 3 mL of 50 mM phosphate buffer in 9:1 H₂O:D₂O at pH/pD 7.5. Cell density was determined by diluting a sample of the cells to OD₆₀₀ of 0.2–0.6. The cell suspension was normalized to 2.5 OD₆₀₀ and loaded into an NMR tube. NMR experiments were taken on a Bruker Avance III 700 MHz spectrometer or a Bruker Avance III 600 MHz spectrometer. Sample temperature was set to 310 K and the sample was allowed to equilibrate before performing an experiment. A standard ¹H NMR experiment was performed on 50 mM phosphate buffer in 9:1 H₂O:D₂O at pH 7.5 to determine the location of the water peak in the spectrum. A presaturation ¹H NMR experiment was performed on the buffer sample, presaturating each scan at the determined resonance of the water peak. The cell suspension was loaded into the spectrometer and an identical presaturation ¹H NMR experiment was performed after lock, tune, shim, and gain parameters were adjusted to collect a baseline spectrum of the cell suspension. Quickly, the cell suspension was ejected from the instrument, an aliquot of 100 mg/mL AcephPT was added to the sample by mixing with a 9” glass Pasteur pipette (AcephPT concentration = 1 mg/mL; 2.66 mM), and the sample was added back to the instrument. Sequential presaturation ¹H NMR experiments (number of scans = 16, relaxation delay = 3, total experiment time = 83 s) were performed for the indicated time. The first scan or two were excluded from the data sets as the sample equilibrated back to 310 K. NMR spectra were processed using a Bruker TopSpin academic license. Speciation data were computed using serial integration on TopSpin. The sample containing AcephPT in buffer (t₀) was used as an external calibrant to globally scale indicated resonances of sequentially acquired spectra of AcephPT incubated with E. coli. Resonance integration values were exported and plotted in Microsoft Excel.

Communal Growth Competition Assay

ESBL-producing E. coli (16096 NDM-1-producing strain) and BL-non-producing E. coli (Top10 carrying mCherry-expression plasmid) were incubated in LB with the appropriate antibiotics (50 μg/mL kanamycin and/or 100 μg/mL carbenicillin) at 37 °C and 200 rpm for 16 h. Three different overnight cultures were used as biological replicates. Overnight cultures were equalized to OD₆₀₀ 1.0 and mixed in different ratios to 1000 μL. The mixtures were diluted 100-fold in a 96-well plate filled with liquid media containing a final concentration of either 2.2 μM (1 μg/mL) cefotaxime, 100 μM (37.5 μg/mL) AcephPT, 250 μM (94.0 μg/mL) AcephPT, or pure liquid media. Monoculture wells were also prepared as controls for each treatment. Liquid media for OD₆₀₀ and Ex₅₈₆/Em₆₂₀ curves was LB, while for colony enumeration it was MHB 2. The 96-well plate was incubated at 37 °C for 24 h in a Tecan Infinite® 200Pro plate reader. Five seconds of orbital shaking followed by OD₆₀₀ and Ex₅₈₆/Em₆₂₀ measurements were acquired in 10 min intervals. After incubation, the contents of each of the wells were 10x serially diluted, and drip-plated on LB agar plates containing 40 μg/mL X-gal and 1 mM IPTG. After 20 h, blue-white colony counts were enumerated to differentiate ESBL-producing E. coli (blue colonies) from BL-non-producing E. coli (white colonies). CFU/mL was determined, and the ESBL-producer fraction was calculated by dividing the blue CFU/mL by the total CFU/mL.

Statistical Analyses

IBM SPSS Statistics was used for all statistical analyses. For pairwise comparisons, complete datasets were assessed for significance (p<0.05) by global analysis of variance (ANOVA) before subdivision of data. If needed, additional ANOVA tests were performed before further subdivision of data and pair-wise comparison by Tukey’s group mean comparison post hoc test. For nonlinear regression, a Levenberg-Marquardt estimation method was used with a sum-of-squares convergence of 1×10⁻⁸ and parameter convergence of 1×10^-8.

Supplementary Material

Supporting Information and Data Availability

The data that support the findings in this work are included in the article and Supporting Information, which is available free of charge at https://pubs.acs.org/XXXXX: Synthetic procedures and characterization, including NMR and mass spectral data, together with details of the communal growth model, clinical isolate characterization, and supplemental figures of dose-response curves, hydrolysis profiling, and bactericidal assays. Plasmid sequencing results for engineered strains are available for free of charge download from the Duke University Research Data Repository at https://doi.org/10.7924/r4vm4hd42.

Abbreviations

BL

β-lactamase

ESBL

extended-spectrum-β-lactamase

SBL

serine-β-lactamase

MBL

metallo-β-lactamase

PBP

penicillin-binding protein

LDH

lactate dehydrogenase

EC₅₀

half maximal effective concentration

PT

pyrithione

ARLG

Antibacterial Resistance Leadership Group

DICON

Duke Infection Control Outreach Group

DiRTE

Disinfection, Resistance, and Transmission Epidemiology Lab

NMR

nuclear magnetic resonance

DEPT

distortionless enhancement by polarization transfer

HPLC/MS

high performance liquid chromatography/mass spectrometry

RP-HPLC

reversed-phase high performance liquid chromatography

HILIC

hydrophilic interaction chromatography

HRMS

high resolution mass spectrometry

ESI

electrospray ionization

OD₆₀₀

optical density at 600 nm

Ex₅₈₆/Em₆₂₀

excitation at 585 nm, emission at 620 nm

CFU

colony forming unit

X-Gal

5-bromo-4-chloro-3-indolyl-beta-D-galacto-pyranoside

IPTG

isopropyl β-D-1-thiogalactopyranoside

TLC

thin layer chromatography

LB

Luria Bertani Broth (Lennox)

MHB 2

Mueller Hinton Broth 2

DMSO

dimethyl sulfoxide

DCM

dichloromethane

TFA

trifluoracetic acid

TIPS

triisopropylsilane