Platform to Discover Protease-Activated Antibiotics and Application to Siderophore–Antibiotic Conjugates

Abstract

Here we present a platform for discovery of protease-activated prodrugs and apply it to antibiotics that target Gram-negative bacteria. Because cleavable linkers for prodrugs had not been developed for bacterial proteases, we used substrate phage to discover substrates for proteases found in the bacterial periplasm. Rather than focusing on a single protease, we used a periplasmic extract of E. coli to find sequences with the greatest susceptibility to the endogenous mixture of periplasmic proteases. Using a fluorescence assay, candidate sequences were evaluated to identify substrates that release native amine-containing payloads. We next designed conjugates consisting of (1) an N-terminal siderophore to facilitate uptake, (2) a protease-cleavable linker, and (3) an amine-containing antibiotic. Using this strategy, we converted daptomycin—which by itself is active only against Gram-positive bacteria—into an antibiotic capable of targeting Gram-negative Acinetobacter species. We similarly demonstrated siderophore-facilitated delivery of oxazolidinone and macrolide antibiotics into a number of Gram-negative species. These results illustrate this platform’s utility for development of protease-activated prodrugs, including Trojan horse antibiotics.

Graphical Abstract

INTRODUCTION

The well-recognized term, “ESKAPEE” (previously ESKAPE),¹ encompasses the names of seven species of clinically relevant pathogens (Enterococcus faecium, Staphylococcus aureus, Klebsiella pneumoniae, Acinetobacter baumannii, Pseudomonas aeruginosa, Enterobacter spp., and Escherichia coli) that are associated with resistance to commonly prescribed antibiotics and are largely responsible for the world’s nosocomial infections.² Five of these pathogens are Gram-negative species, whose outer membrane and associated resistance–nodulation–cell division (RND) efflux pumps render them resistant to many classes of antibiotics.³ Indeed, the outer membrane shields the bacteria from molecules that are unable to pass through porins,⁴ providing an effective barrier to many molecules that would otherwise be effective antibiotics against these pathogens.⁵ Gram-negative therapies can be delivered by siderophore-mediated antibiotic delivery using Nature’s Trojan-horse approach.^6–8 Siderophores are small-molecule chelators that are produced by bacteria to sequester Fe(III),⁹ which is an essential nutrient required for bacterial growth and virulence.¹⁰ In the case of Gram-negative pathogens, outer membrane proteins (e.g., TonB-dependent transporters)¹¹ bind ferric siderophores and provide opportunities for their active transport into the periplasm,¹² where they may undergo further translocation into the cytoplasm by alternate transport mechanisms depending on the siderophore and the species of bacteria.¹³ Owing to the promiscuity of their transport systems, bacteria also use siderophores in warfare against other microbes.^12,14 For example, streptomycetes produce albomycins, which are natural siderophore–antibiotic conjugates (SACs) and highly effective antibiotics against Gram-negative Enterobacteriaceae.¹⁵ Albomycins are recognized by siderophore uptake machinery, transported into the cytoplasm, and activated by peptidase N, which cleaves the N-terminal serine-amide bond and releases the serine-bound t-RNA synthetase inhibitor to bind to its target.^16–19 Inspired by Nature’s strategy, we develop an unbiased platform for the discovery of linkers that are cleaved by periplasmic proteases,²⁰ which demonstrates that this platform can produce SACs with both broad and narrow spectra of activity. We target proteases in the periplasm, a compartment that contains >20 known proteases, because all SACs pass through the periplasm.

There are two categories of SACs, depending on the type of linker they possess: non-cleavable and cleavable. There has been significant progress in the development of non-cleavable SACs,^21,22b,23 with the first siderophore–β-lactam conjugate Fetroja (cefiderocol) recently approved by the FDA.²⁴ However, their use is often limited to periplasmic-targeting antibiotics (e.g., daptomycin, vancomycin, and β-lactams). The few examples of cytoplasmic-targeting, non-cleavable SACs may be less effective than the parent antibiotic for two reasons:^{25,26b–d,27–34} (1) the conjugate may not pass through the inner membrane to reach the cytoplasm, or (2) the bulky siderophore component may interfere with binding to the target.^12,25b,35 Therefore, cleavable linkers are traditionally thought to be required for SAC compatibility with cytoplasmic-targeting antibiotics.²¹ The majority of Gram-positive antibiotics are cytoplasmic-targeting and may require a cleavable linker to be converted into SACs for Gram-negative pathogens.^12,21,36 We show that protease-cleavable linkers improve the activity of SACs containing periplasmic- and cytoplasmic-targeting antibiotics.

Over the past 30 years, only a few cleavable linker strategies have been developed for SACs and a number of challenges remain.^{22,25–27,35,21,37–39} Despite optimization for hydrolytic stability,³⁵ ester linkers for SACs (e.g., A, Figure 1A) are susceptible to premature cleavage prior to bacterial-cell entry.^{25,26c,d,27,35,39} SACs with disulfide (e.g., B) and trimethyl-lock linkers based on reduction- (e.g., C), phosphatase (e.g., D), and esterase-triggered cleavage mechanisms were less active than the parent antibiotic.^26,38 Recent work by Nolan demonstrated that cytoplasmic siderophore degradation by the siderophore-hydrolase IroD confers high activity to a conjugate with a non-cleavable linker.³⁷ Miller and co-workers developed a dual drug conjugate with the cleavable β-lactam linker E.^22a However, more alternatives are needed as the physiological instability of β-lactams can lead to hyper-sensitivity and allergic reactions in patients.^40a Protease-cleavable linkers have the potential to overcome the limitations of cleavable SACs reported to date.

Figure 1.

(A) Selected cleavable linkers that have previously been used for SACs. (B) Concept for SACs that contain a linker that can be cleaved by periplasmic proteases. (C) Workflow for the development of protease-cleavable SACs.

With few exceptions,^22a,27,39 cleavable SACs have incorporated a DNA-gyrase-inhibiting fluoroquinolone antibiotic, ciprofloxacin or norfloxacin,^40b,41 which are already active against Gram-negative pathogens. These SACs provide a method to study siderophore-mediated antibiotic delivery because release of the antibiotic from the conjugate is often required to observe significant growth inhibition.^{21,25,26,35,37,38} In line with this trend, we investigate a macrolide that is already active in Gram-negative bacteria as one strategy to interrogate SAC cleavage.

Protease-cleavable SACs are classified as prodrugs.^42,43 Protease-activated prodrugs that are cleaved by mammalian proteases include antibody–drug conjugates,^44–50 antibody–antibiotic conjugates,^51,52 peptide–drug conjugates,⁵³ macromolecular prodrugs,^42,54 and protease-activatable photosensitizers,^55,56 with the cathepsin B-sensitive valyl-citrulline (Val-Cit) linker being the most successful and widely known.^47,49 Although most protease-activated prodrugs target cancer, antibody–antibiotic conjugates are undergoing clinical trials for intracellular bacterial infections associated with difficult-to-treat persisters.^51,52 However, antibody–antibiotic conjugates have been limited to the treatment of intracellular S. aureus, using a linker that was optimized for cleavage by mammalian proteases. Protease-activated prodrugs that are activated by bacterial proteases would clearly be of value for treatment of a variety of Gram-negative bacterial infections.

Several technologies have been developed to screen libraries of protease substrates,⁵⁷ including indexed arrays of fluorogenic substrates,^58,59 positional scanning of synthetic combinatorial libraries,^60,61 substrate phage,^62,63 multiplex substrate profiling by mass spectrometry,^57,64–66 and others.^58,67–71 Given the large size of achievable libraries, substrate phage display⁶² provides an unbiased selection tool to discover cleavable linkers for SACs. Conventional phage display has been used in vitro and in vivo to design targeted-peptide conjugates.⁷² For example, Wilfred van der Donk used phage display to select for ribosomally synthesized and post-translationally modified peptides (RiPPs) that bind to lipid II.^72g However, “substrate phage display” has not been applied to prodrug development,⁶³ and its use for profiling complex biological mixtures is limited.⁷³ Here, we extend substrate phage display to identify cleavable peptide linkers for SACs that are activated by periplasmic proteases.

In designing protease-activated prodrugs, there are advantages to targeting multiple proteases over an individual protease.^74–76 For example, in vivo deletion mutants of cathepsin B retain the ability to release prodrugs from the combined activity of several proteases.^77,78 Indeed, targeting multiple proteases might minimize resistance when designing antibiotic prodrugs. Thus, we screened broadly for peptides that are cleaved by the proteases present in an unfractionated periplasmic extract.

RESULTS

Substrate Phage Display Leads to WSPKYM-RFG and WSWC-KWASG as Substrates for Periplasmic Cleavage.

To discover efficient peptide substrates using the method of substrate phage,⁶² we built a random hexapeptide library genetically fused to the pIII gene of M13 bacteriophage. A phagemid vector allows monovalent display of the corresponding protein on the tip of the phage. A GGS spacer was incorporated at each end of the randomized peptide to enhance flexibility. An AviTag sequence was also incorporated at the N-terminus for biotinylation of the displayed peptides. The biotin is used to immobilize the phage library on a streptavidin-coated surface, and a protease can then cleave at favorable peptide sequences. Proteolysis releases the phage, which are then amplified and sequenced to determine the favorable substrates for the protease of interest.

The process of “biopanning” entails the following steps: (1) enzymatic biotinylation of the AviTag sequence,^79,80 (2) immobilization of the biotinylated library on streptavidin 96-well plates, (3) cleavage of the immobilized library by incubation with the periplasmic extract of E. coli K12 MG1655 at 37 °C, (4) amplification of the eluted phage using E. coli TG-1 cells, and (5) isolation and purification of phage for the next round of selection. The periplasmic extract used in panning was obtained by osmotic shock of E. coli K12 MG1655.⁸¹ We carried out four rounds of selection (see Supporting Information, section VB, page S79), with the stringency being increased with each succeeding round by reducing extract concentration and decreasing the incubation time. The phagemids from the input library and final round of biopanning were isolated, barcoded, and submitted for next-generation sequencing.⁸² Sequences for further characterization were ranked based on the extent of enrichment relative to the original library (Table 1).

Table 1.

Highly Enriched Sequences Found through Substrate Phage Display

sequence	reads	initial reads	enrichment factor
KNQSLG	10652	0.5	21304
GSDSSV	9239	0.5	18478
NHADVH	8138	0.5	16276
KSEMLS	7742	0.5	15484
WCKWAS	15307	1	15307
PKYMRF	13192	1	13192

Six highly enriched sequences (SKNQSLGG, SGSDSSVG, SNHADVHG, SKSEMLSG, SWCKWASG, and SPKYMRFG) were synthesized with a Ser at the N-terminus and a Gly at the C-terminus to mimic the GGS spacers in the phage library. A tryptophan was added to the N-terminus to facilitate detection by HPLC. Each peptide was found to be cleaved to varying extents following treatment with periplasmic extract for 18 h at 37 °C. The cleavage sites and extent of proteolysis were evaluated by LC/MS (Table S5), which revealed that the sequences WSWC-KWASG (dash indicates site of cleavage) and WSPKYM-RFG may be optimal linkers for cleavable SACs.⁸² In addition to their promising cleavage profiles described in Table S5, the presence of these sequences in the original library contributed to their selection as potential linker candidates. In protease substrate nomenclature, the Cys and Met residues at the N-terminal side of the scissile bond for WSWC-KWASG and WSPKYM-RFG are designated as the P1 positions, while the Lys and Arg residues at the C-terminal side are designated as the P1′ positions. Although the residues on the P′ side are sometimes important for efficient cleavage,⁸³ this is not always the case.⁸⁴

WSPKYM Conjugates Are Efficiently Cleaved without a P′ Peptide.

With the candidate sequences WSWC-KWASG and WSPKYM-RFG in-hand, we asked whether the residues on the C-terminal P′ sequence were required for proteolysis (Figure 2A). To probe this question, we used a solid-phase method to synthesize fluorescent substrates. A 7-amino-4-carbamoylmethylcoumarin (acc)^85,86 was coupled directly to the P1 Cys and Met residues as an antibiotic surrogate (see Supporting Information, section IIIF, pp S67–S72). We were indeed pleased to find that the treatment of peptide 1 (25 μM) with periplasmic extract led to rapid release of the fluorescent amino coumarin. HPLC analysis also showed the substrate was fully consumed at the end of the reaction (Figure S9). As expected, the rate of cleavage was dependent on the substrate concentration, showing partial saturation at concentrations over 12.5 μM (Figures S4 and S5). However, we did not attempt to fit a value of K_M, given the heterogeneous nature of the proteolytic composition of the extract. On the other hand, the amount of acc released from peptide 2 (WSWC-coumarin) was comparably insignificant under these conditions (Figure 2B). Thus, WSPKYM was determined to be more suitable than WSWC for the development of cleavable SACs.

Figure 2.

(A) Peptides with turn-on fluorophore, acc, as an antibiotic surrogate. (B) Evaluation of acc cleavage from peptides 1 and 2 in a periplasmic extract of E. coli K12 MG1655 (400 μg/mL) at 37 °C.

To determine the type of protease responsible for acc release in a periplasmic extract of E. coli, we evaluated a variety of protease inhibitors (Figure S6), which revealed that the enzymes of interest include a metalloprotease or a calcium-dependent protease. We also used alanine positional scanning to evaluate the sequence dependence for efficient cleavage. The starting peptide Ac-WSPKYM-acc (1) was the most efficient substrate (Figure S7). Individual substitutions of Ala at Tyr (P2), Trp (P6), and Lys (P3) resulted in large decreases in rate, while substitutions at other positions resulted in an approximately 2-fold decrease in rate. These findings indicate that substrate phage was effective in discovering an optimized substrate for cleavage.

We also examined the ability of human and mouse serum to cleave 1. We were pleased to find that under conditions where 1 is rapidly cleaved by periplasmic extract, human serum released acc with a half-life of approximately 3–4 h (Figure S8). However, the compound is cleaved more rapidly by mouse serum, indicating that some optimization would be necessary for applications using mouse models.

Design and Synthesis of SACs That Incorporate Solithromycin, Daptomycin, and Eperezolid-NH₂.

To explore the versatility of SACs, we selected three structurally and mechanistically diverse antibiotics that act on targets in either the periplasm or the cytoplasm. Each antibiotic has an amine, which can be unmasked upon proteolysis of the WSPKYM linker. The lipopeptide daptomycin (4) interacts with the cytoplasmic membrane in Gram-positive bacteria, leading to increased membrane permeability and membrane depolarization.^87a–c However, it is ineffective against Gram-negative bacteria and challenging to functionalize without loss of potency.^87d Nevertheless, the Miller group has shown that daptomycin can gain activity in Gram-negative species if conjugated to a siderophore with a non-cleavable linker.^22b,23 Here, we examine the use of a protease-cleavable linker for this system.

We also chose two ribosomal protein synthesis inhibitors, amino-oxazolidinone 5 and solithromycin 6, as examples of antibiotics that must gain access to the cytoplasm to be active.^22a Since both oxazolidinones and macrolides bind deep within the large ribosomal subunit in fairly occluded binding sites, siderophore conjugates without cleavable linkers have been met with limited success, potentially due to interference of the linkers with binding.^{12,21,36,88,89} Our strategy would avoid this complication by enabling release of the parent antibiotics.

For attachment to the N-terminal side of the linker, we sought a siderophore that was synthetically accessible and compatible with a variety of bacterial siderophore uptake systems. The bis-catecholate, azotochelin-like⁹⁰ siderophore (Heinisch–Möllmann–Miller (HMM) siderophore, Table 2)^22,91a,b was selected due to its ease of synthesis and its ability to carry large cargo (e.g., daptomycin) into A. baumannii, E. coli, and P. aeruginosa.^22,23 Furthermore, β-lactam conjugates with non-cleavable linkers that used tetra-acetate derivatives of siderophore 10a were active against MDR strains capable of efflux.^91c We used a modified version of Miller’s protocol to access siderophore 10, which has acid-labile ketal protecting groups that can be removed concomitantly with tert-butyl and tert- butoxycarbonyl (Boc) protecting groups on the amino acid side chains.⁸²

Table 2.

Modular Synthetic Platform for SAC Synthesis

^aFor detailed synthetic methods, see Supporting Information.

We developed a modular synthetic route that enables the facile incorporation of a variety of linkers, antibiotics, and siderophores (Table 2). Gram-scale linker assembly and subsequent siderophore attachment were accomplished on solid-phase to provide the partially protected intermediate 3 in 50% overall yield, and the antibiotic was then coupled to the C-terminus in solution. Following acidolytic deprotection, the final SACs (7–9) were obtained in 12–53% yield over two to four steps.

Several aspects of our route merit further discussion. The majority of the synthesis proceeds on solid phase, simplifying purification and facilitating parallel synthesis of analogues. Antibiotics are directly attached in the penultimate step, enabling rapid access to the final antibiotic conjugates from intermediate 3.⁸² The synthesis requires only one HPLC purification, and the final products are purified by trituration. Daptomycin and solithromycin are commercially available and the oxazolidinones were synthesized following the protocols of Miller^22a and Rafai Far.^91d

We also synthesized several controls to probe the mechanism of action of 7, 8, and 9 using modifications of our existing protocol (11–18, Table 2). These included conjugates with d-amino acid linkers (e.g., 13, 14, and 16), which are not readily cleaved by proteases, enabling us to determine if proteolytic linker cleavage is responsible for the observed activity (see Supporting Information, section VIII, page S139).⁹² To confirm on-target activity of the antibiotic, we evaluated conjugates that lack an antibiotic or contain an inactive enantiomer of the antibiotic (e.g., 11, 12, and 15). To compare the effectiveness of conjugates containing a peptide linker that did not release acc in periplasmic extract (Figure 2B), we synthesized WSWC conjugate 17. We also synthesized a siderophore-free conjugate (18) to determine the dependence of activity on the siderophore.

Determination of the Antibacterial Activity of SACs 7–9 and Iron-Dependent Activity.

The minimum inhibitory concentrations (MICs) of conjugates 7–9 were evaluated according to the standard CLSI antimicrobial susceptibility testing guidelines in Meuller–Hinton-II (MH-II) broth with dipyridyl to sequester iron from the media and promote siderophore-mediated transport (Tables 3 and 4, Table S1A,B).^82,93 Controls that lacked a siderophore did not show activity dependence on dipyridyl concentration, while the siderophore conjugate became increasingly active at higher concentrations of dipyridyl (Table S3A–C). This phenomenon can be explained by the enhanced expression of outer-membrane transport proteins for siderophore uptake in iron-deficient media.^91e The absence of dipyridyl from the growth medium dramatically attenuated siderophore–conjugate activities without influencing the MIC of the free antibiotic.⁸² These results correlate well with expected growth-inhibitory activity of SACs.

Table 3.

Antibacterial Activity (MIC in μM) and In Vitro Cleavage of Eperezolid-NH₂ Conjugate 8 and Derivatives Thereof^a

^aFor strain descriptions and extract cleavage procedure, see Supporting Information, Table S1B. ND = not determined.

Table 4.

Antibacterial Activity (MIC in μM) of Daptomycin SAC 7 and Derivatives Thereof^a

^aFor strain descriptions, see Supporting Information. ND = not determined. Daptomycin is a calcium-dependent lipopeptide. MIC assays of daptomycin conjugates were conducted in the presence of 100 μg/mL CaCl₂ in MH-II media.¹⁰³

We included 15 bacterial strains in our assay (14 Gram-negative and one Gram-positive)^94–98 and have highlighted selected activities below (for full-activity tables and strain details, see Table S1B). Two genetically modified strains of E. coli were included: a ΔsurA strain that is deficient in outer-membrane proteins and has increased permeability,^95a and a ΔbamBΔtolC mutant, which has a deficient BamACDE outer-membrane-assembly complex and lacks the TolC-transport protein.^95a,b This strain is widely used because it is defective in small-molecule efflux.

Solithromycin is active in many Gram-negative and Gram-positive species, and the strategy of using a Gram-negative antibiotic is frequently used to evaluate the efficiency of linker cleavage for cleavable SACs (see Introduction).^{21,25,26,35,37,38} The l-linker solithromycin conjugate 9 was comparably active to solithromycin in several pathogenic Gram-negative strains, and the d-linker conjugate 16 was inactive (Table S1A,B). In E. coli, however, conjugates 9 and 16 had similar activity, suggesting that the d-linker conjugate should not be used to evaluate linker cleavage in this species. We found that this may be due to the ability of the entire conjugate to inhibit the 70S E. coli ribosome (Figure S1A). However, the differences in activity between the d- and the l-solithromycin conjugates in pathogenic strains suggested that the linker may enable the release of Gram-positive antibiotics (Table S1A and Figure S1A). With this promising result in-hand, we then proceeded to investigate Gram-positive-only antibiotics (vide infra).

Oxazolidinone Conjugate 8 Was Active against E. coli ΔbamBΔtolC (Table 3).

The oxazolidinone class of antibiotics are active against Gram-positive bacteria, but members of this class lack activity against Gram-negative bacteria, due to the presence of endogenous efflux pumps (Figure S1B). Nevertheless, mutants of E. coli such as ΔbamBΔtolC are susceptible to oxazolidinones because these bacterial strains have disruptions in their efflux systems. This strain is susceptible to eperezolid (Table S1B), but the corresponding amine variant, eperezolid-NH₂ (5), was inactive (MIC > 171 μM). This is likely a result of the inability of 5 to diffuse through the outer and inner membranes as indicated by the data in Table 3 (vide infra). We therefore asked whether conjugate 8 could deliver 5 into a ΔbamBΔtolC mutant of E. coli. We were pleased to discover an MIC of 1 μM for 8 in this mutant; the corresponding derivative 14 with an all-d linker showed strongly decreased activity with an MIC of 19 μM, which is consistent with low activities of previously reported oxazolidinone conjugates with non-cleavable linkers.³⁹ In contrast to eperezolid-NH₂ (5), eperezolid conjugate 8 displayed only 10% inhibition in a cell-free translation assay at a concentration of 38 μM (Figure S1A), indicating that the intact conjugate does not inhibit the ribosome. These findings suggest that potent inhibition of bacterial growth requires enzymatic cleavage of the linker. Supporting this suggestion, 34% cleavage of 8 to the parent antibiotic eperezolid-NH₂ occurred after 11 h of incubation with bacterial periplasmic extract (Table 3). Finally, as expected, conjugate 8 was not active in wild-type strains with functional endogenous efflux pumps (Table S1B).

Eperezolid-NH₂ (5) has only minimal activity (MIC = 43 μM) against both E. coli ΔsurA and S. aureus Newman, which lack outer membrane proteins, indicating that the cytoplasmic membrane of both Gram-positive and Gram-negative bacteria provide a barrier for the diffusion of 5 into the cytoplasm. There are two possible explanations for the potent activity of 8 given the lack of activity of 5 in E. coli ΔbamBΔtolC: (1) conjugate 8 may be actively transported to the cytoplasm and activated by a cytoplasmic protease⁹⁹ or (2) cleavage in the periplasm may lead to large differences in the concentrations of molecules in the periplasm versus the cytoplasm, enhancing the effective concentration of 5 in the cytoplasm and hence potency.

We also synthesized a number of additional control molecules to probe the antibacterial mechanism of 8, including conjugate 17 with the WSWC linker (Table 3); this linker did not release an acc fluorophore from peptide 2 (Figure 2). Not surprisingly, this analogue had only weak (MIC = 37 μM) activity, as did compounds 11, 12, and 15 that lacked an active antibiotic payload. Similarly, the conjugate 18 without a siderophore was inactive. However, methyl ester 15 retained a modicum of activity (9 μM), possibly by a mechanism similar to many non-helical proline-containing cationic antimicrobial peptides.¹⁰⁰

Miller and co-workers published a highly similar eperezolid conjugate using the same siderophore and eperezolid-NH₂ derivative with a cleavable β-lactam linker.^22a This conjugate exhibited activity against a number of strains of Gram-negative bacteria, including some with high-level β-lactam resistance. Against strains with high β-lactamase content for which the β-lactam linker did not measurably contribute to activity (up to 50 μM), the MIC of the conjugates was 6 μM. We did not observe activity of conjugate 8 in the same A. baumannii strain (ATCC BAA-1797), which could be due to inherent differences in the two linkers.

Daptomycin Conjugate 7 Exhibits High Activity against Acinetobacter Species (Table 4).

Daptomycin is used to treat Gram-positive infections, but it lacks activity against Gram-negative species. Therefore, we were gratified to find that the cleavable l-linker daptomycin conjugate 7 showed species-specific activity against A. nosocomialis, A. baumannii, and E. coli, with MIC values in the 1 to 10 μM range, while daptomycin itself was inactive against these species (MIC > 39 μM). The highest activities were observed in Acinetobacter species (MIC = 1–5 μM). Similarly, Miller and co-workers revealed that non-cleavable daptomycin conjugates displayed selective activity against A. baumannii.^22b,23 These findings indicate that this approach has the potential to produce precision antibiotics with relatively narrow-spectrum activity. Moreover, as expected, 7 was inactive against S. aureus as it was not proteolytically activated due to the absence of a periplasm and periplasmic proteases. The activities of ornithine-functionalized daptomycin analogs in S. aureus are highly dependent on the side chain, and the WSPKYM linker reduces the activity of daptomycin in this species.^87d The activity of 7 cannot be compared to previously reported non-cleavable linkers because the specific linker contributes to the activity of non-cleavable daptomycin conjugates (Figure S3).^22b,23 Therefore, we limit our comparisons of the cleavable conjugate 7 to its closely related non-cleavable diastereomer 13.

The d-linker conjugate 13 was 2- to 11-fold less active than 7 against E. coli, A. baumannii, and A. nosocomialis and did not entirely lose activity as would be expected because daptomycin can still engage its target in the periplasm with a noncleavable linker.^22,23 Nevertheless, the l-linker conjugate 7 improved the activity up to 11-fold, suggesting that cleavable linkers may benefit SACs containing periplasmic-targeting antibiotics. Finally, derivatives of 7 lacking the daptomycin payload (11 and 12) were essentially inactive. Taken together, the data in Table 4 indicates that we have successfully repurposed daptomycin for Gram-negative bacteria, and the enhanced activity of 7 relative to 13 is consistent with our guiding hypothesis of stereospecific proteolytic activation. It is also clear that proteolytic activation likely occurred in the periplasm, rather than by an extracellular protease in the medium given that daptomycin lacks activity against Gram-negative species.

To understand the lower activity observed for E. coli (MIC = 11 μM) relative to Acinetobacter species (MIC = 1–5 μM), we note that the release of daptomycin from conjugate 7 in periplasmic extract of E. coli was not observed by HPLC analysis (see Supporting Information, section VIII, pp S120–S124), which may be due to unfavorable steric interactions between the approaching protease(s)¹⁰¹ and the large daptomycin molecule. Following incubation of 7 in periplasmic extracts of A. baumannii and A. nosocomialis, however, the HPLC trace revealed cleavage products with retention times that overlapped with daptomycin (see Supporting Information, section VIII, pp S115–S119).

These results are also of interest with respect to the mechanism of action of daptomycin. It has been previously reported that the target for daptomycin (4) may be absent in Gram-negative species due to the differing membrane compositions between Gram-positive and Gram-negative bacteria.¹⁰² Given that 7 is active against E. coli, it would appear that daptomycin is able to act on the cytoplasmic membranes of these Gram-negative species once they gain access. Also, our finding that daptomycin itself is equipotent against S. aureus Newman and in the outer membrane-compromised E. coli ΔsurA (MIC = 0.6 μM) is consistent with this conclusion.

CONCLUSION

The strategy developed here should be broadly applicable for discovery of protease-activated peptide prodrugs for a variety of applications. Here, we focused on delivering antibiotics by designing protease-cleavable siderophore conjugates. By targeting E. coli periplasmic proteases, we were able to design conjugates that act against a broad (or narrow) spectrum of Gram-negative bacteria, illustrating the potential of this approach. Our results provide strong support for the overall mechanism of proteolytic release of the antibiotic from conjugates 7–9. Although we have not yet identified the proteases responsible for activity against our substrates, we purposefully avoided targeting a single protease to decrease the chances of resistance arising from mutants of a single protein. Moreover, the use of chemically stable amide linkers provides an advantage to targeting proteases over esterases and β-lactamases by avoiding the need for esters and β-lactams, which are chemically more labile. Importantly, the modular design and facile synthetic route provides opportunity for rapid synthesis of SACs with different siderophores, linkers, and antibiotics. This has led to the discovery of cleavable conjugates with activity against several clinically relevant Gram-negative pathogens.

Throughout the course of this work, we made a number of unexpected discoveries with impacts that extend beyond the scope of protease-cleavable prodrugs. The l-linker daptomycin conjugate 7 completely lacks the Gram-positive activity of daptomycin and has gained Gram-negative activity, effectively “flipping” the spectrum of activity of this potent antibiotic. We found that conjugates with d-linkers, which are unlikely to be cleaved proteolytically, have moderate activity against several strains of Gram-negative bacteria. Perhaps the most unexpected results are the activities of the solithromycin conjugates 9 and 16 in a cell-free translation assay, which indicate that these large (MW > 2000) conjugates may directly inhibit the ribosome in E. coli (see Supporting Information, Figure S1A). These results are extremely surprising in the context of solithromycin–ribosome structural data¹⁰⁴ and may provide the basis for new macrolide–peptide hybrid antibiotics.

Additional studies will be required to optimize the current linkers for use as practical drugs. The current linker has reasonable stability in human serum, which could doubtlessly be enhanced by limited structure–activity relationships (Figure S8). The peptide WSPKYM also contains more than one cleavage site (Table S5), which may complicate the drug activation mechanism. Our conjugates do not show significant activity against P. aeruginosa (Table S1B and Figure S2), a pathogen prone to resistance, which could be due to the following: (1) insufficient linker cleavage may not lead to growth inhibition because the linker was optimized for E. coli, (2) the target for daptomycin may be present in minimal amounts, depending on the strain or species of Gram-negative bacteria,¹⁰² or (3) solithromycin and eperezolid might be efflux substrates in this organism.

In summary, this work provides a robust methodology for selection and screening of Trojan-horse prodrugs applied to the persistent and growing problem of antibacterial resistance. The resulting conjugates from this platform improve upon existing cleavable linkers for SACs. Using phage display, one can rapidly screen vast peptide libraries, and by varying the selection strategy one can screen for linkers with desired characteristics. For example, by using periplasmic extracts from different species of bacteria in succeeding selections, one can ensure broad activity over the desired range of bacteria. Alternatively, negative selection could be incorporated to select against cleavage of serum proteases or beneficial members of the microbiome. Thus, the potential for fine-tuning the protocol for future practical applications is substantial.