Sophisticated natural products as antibiotics

Summary:

In this review, we explore natural product antibiotics that do more than simply inhibit an active site of an essential enzyme. We review these compounds to provide inspiration for the design of much-needed new antibacterial agents. We examine the complex mechanisms that nature has evolved to effectively target bacteria including covalent binders, inhibitors of resistance, compounds that utilize self-promoted entry, those that evade resistance, prodrugs, target corrupters, inhibitors of “undruggable” targets, compounds forming supramolecular complexes, and selective membrane-acting agents. These are exemplified by β-lactams that bind covalently to inhibit transpeptidases and beta-lactamases; siderophore chimeras that hijack import mechanisms to smuggle antibiotics into the cell, compounds that are activated by bacterial enzymes to produce reactive molecules, and antibiotics such as aminoglycosides that corrupt, rather than merely inhibit their targets. Some of these mechanisms of antibiotic action would be very hard to imagine, such as the preformed β-strands of darobactins that target the “undruggable” β-barrel chaperone BamA, or teixobactin that binds to a precursor of peptidoglycan and then forms a supramolecular structure that damages the membrane impeding the emergence of resistance. Many of the compounds exhibit more than one notable feature, such as resistance evasion and target corruption. Understanding the surprising complexity of the best antimicrobials provides a roadmap for developing novel compounds to address the antimicrobial resistance crisis by both mining for new natural products and inspiring us to design comparably sophisticated antibiotics.

Introduction

Antibiotics are nature’s oldest toxins. Evolved as weapons of bacterial warfare, these small molecules have to perform several different, often opposing tasks. They need to hit homologous targets of different microorganisms, be hydrophilic in order to reach neighboring cells and penetrate the outer membrane of Gram-negative species, be hydrophobic to cross the cytoplasmic membrane, evade resistance mechanisms, and survive a harsh external environment. Borrowed for human use, compounds that become drugs have to possess two additional properties – good pharmacokinetics (PK) – an ability to move through tissues without being rapidly destroyed or sequestered, and be non-toxic to the host. No wonder natural products satisfying these criteria are rare, and making synthetic antibiotics is arguably the most challenging task of medicinal chemistry.

The low-hanging fruit, antibiotics commonly present in Actinomycetes, have been mostly discovered by the 1960s¹. Since then, the introduction of novel compounds has been sparse, resulting in the Antimicrobial Resistance Crisis (AMR) we are currently experiencing. This is a slow-moving global pandemic, according to the World Health Organization², with a burden of 5 million deaths world-wide associated with AMR³. The death rate from multidrug resistant Klebsiella pneumoniae, for example, is 40%, similar to pathogens of the pre-industrial era^4,5. Antibiotics are also an essential prophylactic in numerous medical procedures, from surgery to chemotherapy.

Currently, our most pressing need is for novel compounds acting against multidrug-resistant Gram-negative bacteria, such as K. pneumoniae, Escherichia coli, Acinetobacter baumannii and Pseudomonas aeruginosa³. The last class of synthetic antibiotics acting against Gram-negative bacteria that made it to the clinic, the fluoroquinolones, was discovered in the 1960s, while tetracyclines, the last class of natural product antibiotics acting against this group of bacteria, were discovered in 1949¹. Gram-negative bacteria have an additional outer membrane that restricts penetration of large and hydrophobic compounds, and anything that leaks through is extruded by multidrug efflux pumps⁶. It is relatively easier to find compounds acting against Gram-positive species, but even in this case, we have not found a novel compound that became a drug in over 40 years – daptomycin was discovered in 1978. There is no doubt a need for novel compounds acting against Gram-positive species such as Staphylococcus aureus MRSA (methicillin-resistant S. aureus) as well, it is a major killer in the hospital setting. Lack of economic incentives has led to the exodus of Pharmaceutical companies from the field, and very few companies now have programs devoted to antibiotic discovery, contributing to the paucity of novel compounds. However, it is clear from the above analysis that the main bottleneck for getting new drugs is in discovery, not in development⁷. Indeed, once a good lead compound is discovered and preclinical evaluation completed, development through clinical trials takes about 5 years.

In the realm of natural products, progress is being made by going outside of actinomycetes in search of novel compounds. The majority of bacteria are “uncultured”⁷ making up about 99% of the total diversity. Some of these bacteria depend on growth factors such as siderophores or quinones produced by neighboring species^8–10. Cultivation in their natural environment has led to the discovery of teixobactin¹¹ and clovibactin¹² from Eleftheria, β-proteobacteria, which will be discussed in this review. Interesting cultivable species, γ-proteobacteria Xenorhabdus and Photorhabdus, symbionts of the Steinernema and Heterorhabditis nematodes, make several target-specific compounds which we will discuss as well^13–17. Another vast untapped source of novel antibiotics is concealed in so-called “silent” biosynthetic gene clusters (BGCs)¹⁸. Metagenomics, and the direct cloning of environmental DNA into heterologous hosts, holds the promise of providing access to this hidden chemical diversity¹⁸. The current estimate of BGCs is about 1 million, based on genomes sequenced so far¹⁹.

Meanwhile, there has been encouraging progress in the discovery of synthetic antibiotics as well. Analysis of penetration of chemically unrelated compounds into E. coli produced the first “rules of permeation”^20–25, which has recently been extended to P. aeruginosa²⁵. One simple rule is that charged amino groups favor diffusion through porins. This rule was applied to synthesize Fabamycin, an aminated version of a Gram-positive synthetic inhibitor of FabI, a key enzyme in the synthesis of fatty acids which is now in development against Gram-negative bacteria²⁶. A simultaneous optimization of binding to target and penetration through porins led to the rational design of a novel broad-spectrum inhibitor of penicillin-binding proteins, ETX0462²⁷. It shares with penicillin a reactive carbonyl that enables covalent binding to the target. Two compounds based on a synthetic analog of a cyclic antimicrobial peptide inhibit essential targets in the outer membrane. Murepavadin²⁸ inhibits LptD, a translocase of lipopolysaccharide, and POL7306²⁹ is a peptide-polymyxin chimera that acts against BamA, the insertase of outer membrane proteins. Both compounds are active against Gram-negative bacteria, but have a common liability of polycations, nephrotoxicity. Screening of a library of synthetic cyclic peptides emulating natural products led to the development of Zosurabalpin that targets the transporter of lipopolysaccharide^30,31.

As we can see, of the five recently discovered novel synthetics acting against Gram-negative bacteria that made it to successful animal studies, four are loosely based on natural products. This trend is likely to continue. In choosing compounds to review, we focus on natural products that are already in the clinic or have shown systemic efficacy in animal models. Our emphasis is on more recently discovered compounds, or on recent findings that illuminate the fascinating properties of antibiotics that were discovered a long time ago. By looking at natural compounds, we are essentially reading a recipe on how to make antibiotics. While this recipe is not easy to follow, it points us in the right direction.

Covalent inhibitors

Penicillin was the first successful antibiotic. Discovered by Fleming in 1928³², penicillin is still in clinical practice, and served as the basis for developing several generations of analogs that expand its spectrum to cover Gram-negative bacteria, such as ampicillin, amoxicillin and carbenicillin (see Supporting information for structures). The targets of β-lactams are the essential “penicillin-binding proteins” (PBP) – the transpeptidases that link pentapeptides of Lipid II into a growing peptidoglycan polymer (Fig. 1).

Fig. 1. Beta lactams are molecular mimics of the D-Ala D-Ala terminal peptidoglycan motif.

a, Penicillin binding proteins normally recognize the terminal D-Ala D-Ala motif and form an acyl-enzyme intermediate by insertion into D-Ala D-Ala bond eliminating the terminal D-Ala. The acyl-enzyme intermediate subsequently reacts with a second strand of peptidoglycan to form a cross link completing the catalytic cycle. b, Penicillin binding proteins mistake penicillin for their natural substrate D-Ala D-Ala due to molecular mimicry reacting with the lactam ring and are then unable to eliminate the terminal D-Ala (thiazole) motif, irreversibly blocking subsequent deacylation and catalytic turnover.

The promiscuity of penicillin derives not only from the homology of its targets, but also from its reactive β-lactam ring that enables the compound to act in spite of differences among the catalytic centers of transpeptidases³³. Multitargeting of transpeptidases, covalent inhibition, and substrate mimicry are the keys to the success of the β-lactam class - dramatically limiting the rise of resistance by target mutation, and enabling selective killing of the target pathogen. Acting at the surface of Gram-positive bacteria, penicillin also avoids the other most common mechanism of resistance – efflux by MDR pumps. Target organisms then have to resort to spending resources to produce β-lactamases that destroy β-lactams. There are not that many targets in bacteria that are both essential and highly homologous. Penicillin is the first example of this multitargeting paradigm³⁴ that repeats itself with some other classes of successful antibiotics, as we will see below.

Anti-resistance and selective compounds

Closely linked to β-lactams both in nature and in clinical practice are anti-resistance compounds that inhibit β-lactamases. Clavulanic acid is produced by Streptomyces clavurigerus that also makes cephamycin C³⁵. A combination of amoxicillin with clavulanic acid became augmentin, the first combination antibiotic. The spread of additional classes of β-lactamases stimulated development of more effective inhibitors covering many types of these enzymes including diazabicyclooctane inhibitors such as avibactam that share a reactive carbonyl containing ring with penicillin and are also structural mimics ³⁶. The recently introduced boron-containing vaborbactam is the first synthetic β-lactamase inhibitor that is truly novel and does not contain a β-lactam mimic ring³⁷, but otherwise is remarkably similar in structure to penicillin making it difficult for the β-lactamases to avoid.

β-lactamase inhibitors remain the first, and only example of specialized anti-resistance compounds. Several natural inhibitors of MDR pumps have been described^38,39, but whether this is their intended function remains an open question. So far, a clinical candidate with a good safety profile has not emerged. Development of MDR pump inhibitors is probably the most realistic path forward for the next type of anti-resistance compounds, since they would increase potency, and spectrum, for many antibiotics⁴⁰.

Selective compounds

Historically, the quest for new antibiotics was synonymous with discovery of broad-spectrum compounds. Conveniently, β-lactams, fluoroquinolones and aminoglycosides enable physicians to treat infectious diseases without necessarily identifying the causative agent. This has led to overuse, prescribing antibiotics for viral infections, and contributes to the spread of resistance. We also have an increased appreciation for the role of the microbiome in human health, especially at an earlier age when it helps shape the immune system, and harm from broad-spectrum antibiotics can be considerable⁴¹. This harm can lead to life-threatening C. difficile infection. A minor member of the human microbiome, C. difficile is kept in check by the commensal bacteria in the gut⁴², but it can form spores that survive environmental threats. Systemic administration of broad-spectrum antibiotics has been shown to disrupt the microbiota ^42,43, which in turn enables germinating spores of C. difficile to propagate and produce toxins, leading to enterocolitis. The recommended treatment is with oral vancomycin or fidaxomicin, narrow-spectrum compounds that kill C. difficile. In the process, vancomycin and fidaxomicin also kill Gram-positive symbionts commensals, which leads to a relapsing, and often incurable infection. The current effective therapeutic for untreatable enterocolitis is fecal matter transplantation, and a simpler product from Seres, VOWST, pills containing spores of symbiotic Clostridia⁴⁴, was recently approved. A compound selective against C. difficile is highly desirable, but it is unclear how to do so without harming other Clostridia, a major group of gut symbionts⁴⁵.

The only selective antibiotics we currently have in the clinic are certain compounds acting against M. tuberculosis. All of them are synthetic and were identified by screening against this pathogen; selectivity was not a sought-after feature at the time, but an accident of discovery and the consequence of unique mycobacterial physiology. For example, ethambutol inhibits mycobacterial arabinogalactan synthesis⁴⁶, isoniazid is a prodrug activated by tuberculosis catalase-peroxidase KatG, and bedaquiline binds to the C-subunit of the H⁺-ATPsynthase that is sufficiently different from its non-mycobacterial homologs⁴⁷. Current treatment regimens for tuberculosis contain broader-spectrum antibiotics such as rifampicin, moxifloxacin, and linezolid, and the potential benefit of a completely tuberculosis selective regimen remains unrealized. Given the lengthy, 6-month treatment for tuberculosis, damage to the microbiome is especially pronounced⁴⁸.

Are species- or group-selective compounds present in nature? A screen against M. tuberculosis with a counterscreen against unrelated S. aureus provided an answer to this question. This differential screen of a small collection of actinomycetes produced several novel compounds, lassopeptides lassomycin and kitamycobactin targeting the ClpC1 chaperone of the unique mycobacterial ClpP1P2C1 protease; and amicobactin, an unusual macrocyclic polyketide lactone inhibiting SecY of the protein export machinery⁴⁹. Several conventional screens against M. tuberculosis uncovered additional inhibitors of ClpP1P2C1⁵⁰, suggesting that natural compounds acting selectively against mycobacteria are fairly common. A differential screen of P. noenieputensis against M. tuberculosis led to evybactin, which has an unusual mechanism of selectivity¹⁵. Evybactin targets the DNA gyrase which is well conserved. Analysis of resistance mutants showed that evybactin is smuggled into the cells of M. tuberculosis by BacA⁵¹, an unusual transporter of hydrophilic compounds with broad specificity. In a way, BacA resembles an inverted MDR pump for polar compounds. BacA apparently functions as a transporter of vitamin B12, and highly hydrophilic evybactin takes advantage of this entry port.

A somewhat similar mechanism of selectivity has been described for hygromycin A. This is an old abandoned antibiotic that is produced by S. hygroscopicus and was discovered in 1953. It was never developed because of poor activity against conventional pathogens. Hygromycin A was recently rediscovered in a selective screen of soil bacteria against Borreliella burgdorferi, the causal agent of Lyme disease⁵². Hygromycin A is highly potent against B. burgdorferi and other spirochetes. It binds to the conserved site of the 23S ribosomal RNA⁵³. This suggested that the selectivity must lie elsewhere, and further analysis indicated that the compound is taken up by an essential spirochetal transporter of nucleotides to which it bears some resemblance⁵².

From the cases we described, it appears that selectivity may be imparted in a variety of ways that probably cover all logical possibilities – acting against a unique target; acting against a unique site of a conserved target; or taking a selective route into the cell.

Resistance-evasive compounds

Here, we describe antibiotics that have a low propensity to select for resistance development in the bacteria they target. The evasion of resistance takes many forms, providing useful examples that we can emulate when designing synthetic compounds. Resistance to β-lactams was spreading in the 1950s. A novel antibiotic, vancomycin was discovered in 1958 and fast-tracked to treat penicillin-resistant staphylococci. Vancomycin targets Lipid II,⁵⁴ the central precursor of peptidoglycan biosynthesis localized in the cytoplasmic membrane. It ferries the GlcNAc-MurNAc-pentapeptide building block, attached to a pyrophosphate and a bactoprenol lipid tail across the membrane for incorporation into the growing peptidoglycan network⁵⁵. Vancomycin binds the acyl-D-Ala-D-Ala pentapeptide terminus of Lipid II (Fig. 2 a,b).

Fig. 2. Resistance-evasive compounds.

a, Lipid II/antibiotic interaction sites (pink). Vancomycin interacts with the C-terminus of the pentapeptide, teixobactin with the PPi-group and the MurNAc sugar, and Clovibactin with the PPi-group. b, Vancomycin binds the D-Ala-D-Ala C-terminus of the Lipid II-pentapeptide with five hydrogen bonds⁵⁸. c, Teixobactin – Lipid II complex interface. The backbone amino-protons of the depsi-cycle of teixobactin, the End10 sidechain and the N terminus of an adjacent teixobactin coordinate the lipid II PPi group. In addition, End10 interacts with the MurNAc sugar via hydrogen bonds. Blue spheres represent backbone nitrogens; numbers indicate the residue numbers. d, Clovibactin – Lipid II complex interface. The backbone amino-protons of the depsi-cycle of clovibactin tightly bind the PPi-group, while hydrophobic sidechains (Ala6, Leu7, Leu8) seamlessly wrap around PPi, but do not specifically interact with the sugars. e, Model of the mode of action of teixobactin. Teixobactin first forms small β-sheets upon binding of lipid II, then elongates into fibrils that eventually associate into lateral fibrillar sheets, obstructing biosynthesis of peptidoglycan and causing membrane defects. f, Model of the mode of action of clovibactin. At the membrane surface, clovibactin binds lipid II and forms small oligomers that serve as nuclei for the formation of fibrils. Fibril formation enables a stable binding of lipid II and other cell wall precursors, blocking cell wall biosynthesis Panels c, e reproduced from Shukla et al.⁵⁵

This finding explained the surprising ability of vancomycin to withstand resistance development. Lipid II is not directly coded by genes, as protein targets are, and mutations do not lead to changes in the pentapeptide. Thirty years after the introduction of vancomycin into the clinic, high-level resistance emerged in enterococci, traveling on a plasmid. The plasmid carries a van locus coding for a synthetic bypass that replaces the D-Ala residue of the pentapeptide by D-Lac or D-Ser⁵⁶. The van locus likely originated from Amycolatopsis orientalis, where it serves to protect the producer from vancomycin⁵⁷.

A screen of uncultured bacteria grown in situ led to the discovery of teixobactin, produced by a β-proteobacterium Eleftheria terrae⁵⁸. Teixobactin is an undecapeptide with a linear N-terminal 7-residue segment and a C-terminal 4-residue depsi-cycle, representing a novel class of antibiotics. Teixobactin contains several non-canonical residues including D-amino acids and the rare cationic residue enduracididine. Teixobactin is highly potent against Gram-positive pathogens, including S. aureus, S. pneumoniae, E. faecalis, C. difficile, B. anthracis, and M. tuberculosis. Teixobactin shows favourable pharmacokinetics, good safety, promising efficacy in mouse infection models and is currently in preclinical development.

Attempts to obtain mutants resistant to teixobactin failed. In an evolutionary experiment that maximizes the chances of developing resistance, teixobactin was first added at a sub-inhibitory level, and then its concentration was increased daily in small steps for 25 days. No resistance emerged. This suggested that it may be binding to Lipid II, which turned out to be the case⁵⁹. Teixobactin uses backbone amino-protons of the depsi-cycle to tightly coordinate the PPi moiety of lipid II while the unique enduracididine specifically interacts with the conserved MurNAc sugar (Fig. 2c)⁵⁹. The PP-sugar binding motif is also present in Lipid_WTA, the precursor of wall teichoic acid biosynthesis, and teixobactin targets it as well.

Remarkably, binding of teixobactin to Lipid II results in the formation of a large supramolecular structure (Fig. 2e). Teixobactin is anchored to the membrane by hydrophobic Ile and D-allo-Ile residues, which help the compound establish contact with its target. Intermittent localization of L- and D-amino acids favors the formation of an antiparallel β-sheet between adjacent teixobactin molecules. The N-end of teixobactin coordinates PPi of Lipid II, contributing to target capture. The β-sheet of teixobactins bound to Lipid II grows into a supramolecular fibrillar structure. Formation of this structure is likely irreversible, acting as a sink that further concentrates teixobactin at its site of action. Apart from efficiently and potently sequestering Lipid II, the supramolecular structure displaces phospholipids, thinning the membrane and compromising its integrity, leading to a drop in proton motor force and leakage.

A very large effort has been expended over decades to develop useful membrane-acting antibiotics, but obtaining leads with an acceptable therapeutic window proved elusive¹. As we see, the apparent solution to this problem is to start with a compound that is not membrane-acting. Teixobactin only damages membranes containing an accessible lipid-PP-sugar absent in eukaryotes, once it forms a supramolecular structure, elegantly resolving the toxicity problem. Teixobactin binds to a single moiety – PPi-sugar, but in doing so, attacks multiple targets and has at least 4 modes of action. These are inhibition of peptidoglycan synthesis, inhibition of WTA synthesis, release of autolysins and lysis; and cytoplasmic membrane disruption. This multi-prong attack on the entire cell envelope no doubt contributes to the effective killing and to the reduced propensity of teixobactin to select for resistance.

Another isolate of Eleftheria produces clovibactin⁶⁰, which bears some resemblance to teixobactin, but is considerably smaller, 903 vs. 1400 Da. The BGCs of the two compounds are related, suggesting a common ancestry. The compound has potent antibacterial activity against Gram-positive pathogens, and is efficacious in animal models of infection. Clovibactin has a C-terminal depsi-cycle warhead that directly binds to Lipid II, along with a linear N-terminal segment. Besides the unusual residue D-3-hydroxyasparagine that connects the N-terminal and C-terminal segments, the depsi-cycle of clovibactin contains only hydrophobic residues, lacking the enduracididine that teixobactin deploys to bind to MurNAc of Lipid II. Clovibactin binds to the pyrophosphate moiety of multiple essential cell wall precursors C₅₅PP, lipid II and lipid III_WTA, from different cell wall biosynthetic pathways. Backbone amino-protons of clovibactin’s depsi-cycle directly coordinate the PPi, while the hydrophobic depsi-cycle sidechains (Ala6, Leu7, Leu8) surround the PPi group like an adjustable glove, an unusual interaction which is presumably entropically favorable by replacing boundary water (Fig. 2e). Attacking a highly polar target (PPi) with a hydrophobic warhead is indeed a striking, counter-intuitive mode of action. PPi however seems as an unsuitable target for an antibiotic, since it will be released from dead cells together with PPi-containing nucleoside phosphates, and is commonly present in the environment. Adding to this puzzle is the lack of detectable binding of clovibactin to PPi in solution. Solid-state NMR and atomic force microscopy experiments show that potent and selective binding is apparently achieved not through the initial interaction with PPi, but by subsequent formation of an irreversible supramolecular structure. This fibrillar assembly is enabled by an antiparallel arrangement of clovibactin molecules, in which the short N-terminus acts as an oligomerization domain (Fig. 2F).

The fibrils of clovibactin do not damage the membrane, and appear to float just above it. A surprising feature of clovibactin is its superior ability to induce cell lysis. Lytic potency of cell-wall acting inhibitors ranks in the order: clovibactin>teixobactin>β-lactams>vancomycin. An intriguing possibility is that clovibactin fibrils have an additional function of displacing autolysins from wall teichoic acids, resulting in lysis.

The three antibiotics attacking Lipid II-type precursors present fascinating examples of compounds that appear to circumvent the typical mechanisms of resistance selection. Vancomycin binds to a target that is not directly mutable, but the pentapeptide moiety of Lipid II is modifiable. Teixobactin binds to a simpler PPi-sugar moiety that is shared with Lipid_wta, adding a target that induces lysis⁶¹. The formation of a supramolecular structure by teixobactin/Lipid II damages the membrane, adding yet another target that is not prone to direct mutation. While the MurNAc sugar of Lipid II is conserved, it is conceivably modifiable. Clovibactin appears to be the end of the road for the evolution of resistance evasion: PPi is present in all Lipid II-type compounds, and is a completely immutable target.

Apart from attacking immutable targets, teixobactin and clovibactin come with additional features that limit resistance development. Target overproduction is a simple mode of resistance, but formation of a supramolecular structure by teixobactin damages the membrane. This means that overproduction of Lipid II will only facilitate this damage. The Eleftheria producers are Gram-negative bacteria protecting themselves from these compounds by export across the outer membrane. The targets are exposed on the surface of the cytoplasmic membrane of Gram-positive bacteria. Borrowing an outer membrane from the producer does not seem like a realistic strategy for Gram-positive bacteria to acquire resistance. By contrast, vancomycin is produced by a Gram-positive bacterium and as we mentioned above, protects itself by substituting D-Ala residue of the pentapeptide by D-Lac or D-Ser⁵². The pathway coding for this modification can then be borrowed by other pathogens.

Antibiotics can be chemically modified or degraded, which is a major drawback for common compounds such as β-lactams hydrolyzed by β-lactamases. This however is not a problem for vancomycin - no deactivating enzyme has emerged in over 60 years since its introduction into the clinic. Teixobactin and clovibactin are even rarer than vancomycin. With Lipid II binders, nature provides us with a recipe for making antibiotics that are harder to evolve resistance to. These considerations do not mean that resistance to such compounds will not arise; rather, the probability is lower than that of antibiotics with single gene-encoded targets.

Target corrupting

Some of the best antibiotics corrupt their targets, transforming a useful molecular machine into a killing device. Effective killing is especially useful for treating chronic infections.

Aminoglycoside antibiotics (AGA) are produced by actinomycetes and were discovered by Selman Waksman and colleagues in 1944⁶², but only recently have we gained a proper understanding of their complex mechanism of killing. AGA are positively charged oligosaccharides that target bacterial ribosomes by binding to the rRNA close to the decoding center on the small ribosomal subunit. They induce the misreading of mRNA, resulting in the incorporation of incorrect amino acids into the growing peptide. In contrast to other groups of antibiotics targeting the ribosome such as tetracyclines, lincosamides and most macrolides, which usually stop translation and are bacteriostatic⁶³, AGAs only slow down translation, but are bactericidal⁶⁴. However, error-prone ribosomal mutations also cause mistranslation⁶⁵, but are not lethal. This puzzling contradiction was recently resolved – a mass spectrometry study revealed that AGAs induce strings of consecutive errors, with up to four incorrect amino acids incorporated along a stretch of seven amino acids in a protein⁶⁶. At low intracellular concentrations, AGAs target only a limited number of ribosomes, leading to few errors. However, once AGA is bound, it tends to stay attached through several translation cycles, inducing consecutive misreading events that result in accumulation of errors in the synthesized protein. The likelihood of these later errors is independent of the AGA concentration in the cell and is much higher (up to 10,000 times) than the first error. Proteins with error clusters are enriched in aggregates⁶⁶, indicating increased protein misfolding. While the quality control machinery that refolds defective proteins can buffer misfolding caused by single amino acid substitutions, accumulation of errors may reduce the protein stability beyond the level⁶⁷, that can be corrected by the quality control machinery leading to cell death.

Synthetic fluoroquinolones comprise another successful class of corrupting broad-spectrum antibiotics and were introduced in the 1960s. Fluoroquinolones inhibit breaks and cell death. Evybactin from Photorhabdus that selectively acts against gyrase in M. tuberculosis¹⁵ was described in the previous section on selectively acting compounds. Perhaps a natural counterpart to fluoroquinolones that is broad-spectrum or acts against Gram-negative bacteria is waiting to be discovered.

Double-strand breaks in the DNA can be achieved by an entirely different mechanism, according to a recent publication on griselimycin, an old antibiotic produced by streptomycetes. Griselimycin cyclic peptides act against DNA replication complex of mycobacteria by binding to the DNA-sliding clamp DnaN and blocking protein-protein interactions within the complex leading to double-strand breaks⁶⁸.

Anti-persister compounds and chronic infections

It is harder to treat chronic infections even though they are mainly caused by pathogens susceptible to antibiotics. This susceptibility is based on the standard in vitro MIC test that measures the minimal concentration of antibiotic required to stop the growth of cells. However, the MIC test misses the main culprit of chronic infections, dormant persister cells⁶⁹.

The burden of chronic infections is significant, including common child middle ear infections and female urinary tract infections, as well as complex infections of the cystic fibrosis lung to which the patients ultimately succumb. Many chronic infections are associated with biofilms that harbor dormant antibiotic-tolerant persister cells produced stochastically in bacterial populations. Once antibiotic concentration drops, persisters resuscitate and restore the population, causing a relapsing chronic infection. In E. coli, at least two specialized toxin mediated mechanisms of persister formation have been described. Upon DNA damage, an endogenous toxin TisB is expressed that decreases the proton motive force, ATP synthesis and causes antibiotic target shutdown, producing persisters^70–72. The HipA toxin is a protein synthesis inhibitor, and an active allele hipA7 present in clinical isolates from patients with chronic urinary tract infection causes a sharp increase in persister levels and drug tolerance⁷³. Apart from these specialized mechanisms, a stochastic variation in the expression of “persister genes” is the apparent general mechanism of persister formation in bacteria ⁷⁴ One main type of persister genes, and possibly the main one, results in perturbation of energy homeostasis. Rare E. coli cells with low levels of Krebs cycle enzymes have low ATP and are tolerant of killing by antibiotics⁷⁴. Similarly, a subpopulation of M. tuberculosis cells growing in a medium with acetate that have low levels of acetate catabolism enzymes have low ATP and high tolerance of antibiotics⁷⁵. Bactericidal antibiotics, as we have described in the preceding sections, require active targets in order to corrupt them. Shutting down energy-dependent targets appears as an ingenious mechanism to avoid killing by all antibiotics. A theoretical solution to this challenging problem is to identify compounds that do not require energy to kill. The metronidazole prodrug comes close to fulfilling this requirement (see section on Prodrugs below)⁷⁶.

There is a natural product antibiotic that perfectly fits the requirements for an anti-persister compound – acyldepsipeptide (ADEP). Produced by Streptomyces hawaiiensis^77,78, ADEP was initially overlooked since it only kills Gram-positive bacteria. In response to the rising epidemic of MRSA, Bayer undertook a medicinal chemistry program to optimize the compound, which resulted in ADEP4 with improved activity against a broad range of Gram-positive bacteria⁷⁹. The same team also determined that the target of ADEP is ClpP, an important protease conserved among bacteria. ClpP proteases digest defective misfolded proteins and also play an important role in regulating cellular functions by degrading transcriptional regulators. The ClpP proteases both structurally and functionally resemble the eukaryotic proteasome, even though these two molecular machines are not homologous.

ClpP is assembled from fourteen protomers to form a barrel, in which the 14 canonical catalytic triades of the serine proteases are shielded from the cytoplasm (Fig. 3a). Narrow entrance pores at the apical side of the barrel prevent access of proteins to the proteolytic compartment. Specialized ClpP-associated chaperones, ClpC and ClpX are ATPases that recognize defective proteins, unfold them and thread the peptide chains through the entrance pores of the protease into the proteolytic chamber⁸⁰. ClpP/chaperone therefore requires ATP for its functioning, and in this regard, is similar to other targets of bactericidal antibiotics. However, proteolysis per se is an energy-releasing reaction. ADEP exploits this intrinsic property by replacing the ATP-dependent chaperones, resulting in a non-specific protease. Co-crystal structures of ADEP bound to ClpP from S. aureus^81–83 show ADEP embedded in a hydrophobic pocket (H-pocket) at the apical face of the ClpP barrel (Fig. 3b). A ClpP tetradecamer consists of two heptameric rings stacked back-to-back and the H-pockets are located at the apical interface of the protomers, offering 14 ADEP binding sites in total. The same H-pockets serve as the primary interaction sites for the hexameric Clp-ATPases, which dock here via flexible loops terminating in the conserved tripeptide motif V/IGF/L^84–86. The side-chain and Phe-linker of ADEP mimic the IGF-loop perfectly, while the ADEP macrolactone core engages in additional interactions with ClpP (Fig. 3b). The affinity of ADEP surpasses that of the IGF loops, allowing ADEP to efficiently compete with the Clp-ATPases and prevent their association with ClpP. The result is an uncontrolled protease inside the cell (Fig. 3c). In growing bacteria, the central cell division protein FtsZ is particularly prone to degradation in the presence of ADEP^87,88. In non-growing, stationary cells of S. aureus, ADEP induces the degradation of >400 proteins⁸⁹. Stationary cultures of S. aureus are dominated by persister cells. Remarkably, ADEP eradicates both growing and stationary cells, including persisters⁸⁹. Conventional antibiotics such as rifampicin or ciprofloxacin have little effect on stationary populations of S. aureus. The ClpP protease is not essential, at least in vitro, and that presents a liability for ADEP. Combining ADEP with another antibiotic, rifampicin to prevent the rise of resistant mutants eradicates an untreatable S. aureus biofilm infection in a mouse thigh model⁸⁹.

Figure 3. Acyldepsipeptide (ADEP) as an anti-persister compound.

a, The S. aureus ClpP protease is a tetradecamer of stacked heptameric rings (closed conformation, PDB 6TTY). The individual protomers are rendered as protein surfaces and colored in shades of pink. b, The chemical structure of ADEP4 and its respective binding pose on S. aureus ClpP (open conformation, PDB 6PMD). ADEP4 is shown in blue, and ClpP is rendered as pink surfaces with cartoon ribbons in the binding site. c, The molecular mechanism of bacterial ClpP proteases. For regulated degradation in the presence of ATP, Clp chaperones bind specific protein substrates and bring them to ClpP for complex formation. Protein substrates unfold and translocate into the proteolytic chamber for controlled proteolysis. For dysregulated degradation, ADEPs bind to ClpP and compete out the chaperones for an unregulated, activated proteolytic state, leading to the proteolysis of essential proteins and causing cell death.

Pharmacologically advanced analogs of ADEP are being developed⁹⁰, and both ADEP and its urea derivatives show good efficacy in animal models with a number of pathogens^79,91,92. Whether pairing with another antibiotic is required is not obvious, since ClpP mutants lose virulence^93,94 and may not contribute to infection. The ADEP scaffold may very well lead to a useful anti-persister compound.

ADEP provides a general paradigm - dysregulation of hydrolysis for developing anti-persister compounds. Apart from proteases, there are other important and tightly regulated hydrolases in the cell – lipases, phosphatases, and nucleases. Many of them are potential targets for an anti-persister compound, yet to be explored.

“Undruggable” targets

The outer membrane of Gram-negative bacteria is a formidable barrier to penetration of antibiotics, and a particularly attractive approach is to attack the barrier itself, rather than try to bypass it. There are only two essential proteins in the outer membrane, the LPS transporter LptD, and the chaperone/insertase BamA. As mentioned in the introduction, synthetic polycation murepavadin targets LptD, but the development of the compound was paused due to nephrotoxicity. Other components of the Lpt transporter have been targeted as well. Thanatin7, derivative of the antimicrobial peptide thanatin, disrupts the LptA bridge connecting the inner membrane proteins LptBFG to LptD in E. coli and K. pneumoniae (Peptidomimetic antibiotics disrupt the lipopolysaccharide transport bridge of drug-resistant Enterobacteriaceae), and zosurabalpin binds to LptBF, trapping LPS, in A. baumannii^30,31. Both compounds are in development.

BamA folds and inserts β-barrel proteins such as porins and LptD into the outer membrane. It is the core unit of the BAM complex that contains additional chaperone proteins. BamA is itself a β-barrel protein, lacking the typical active site of an enzyme that can be engaged by a small-molecule inhibitor. Indeed, efforts to find synthetic compounds acting against BamA have not been particularly successful. The chimeric polymyxin-cyclic polycation OMPTA binds BamA only moderately, but anchoring of the polymyxin moiety to LPS results in potent action against bacteria²⁹.

Screening of a small library of Photorhabdus led to the discovery of darobactin A that binds to BamA and kills a variety of Gram-negative bacteria, such as E. coli and K. pneumoniae¹⁴. The compound is a 7-mer peptide made from a translationally-produced precursor and has an unusual structure. Two fused rings are formed through an unactivated C-C bond, and an ether linkage (Fig. 4a). Making these bonds is not trivial, either enzymatically or synthetically. The C-C linkage requires a free radical reaction, which is catalyzed by the radical SAM enzyme DarE, part of the BGC coding for darobactin. rSAM enzymes use an iron-sulfur cluster to produce a 5’-deoxyadenosyl radical, which in turn activates the substrate, resulting in formation of novel bonds. The Fe-S cluster that serves as an electron donor is sensitive to oxygen, and this is probably why rSAM enzymes acquire oxygen from water. Surprisingly, DarE incorporates oxygen from O₂ into the ether bond of darobactin⁹⁵. This unusual feature perhaps explains another mystery – Photorhabdus are the only terrestrial bacteria that are luminescent. The bacterial luciferase uses oxygen to oxidize luciferin, acting as an antioxidant.

Figure 4. Darobactin and Dynobactin.

a, Chemical structure of darobactin A. b, Close-up of the darobactin A binding site at the BamA lateral gate. c, Cryo-EM structure of BAM complex with bound darobactin A. d, Scheme of daro/dynobactin evolution. e, Chemical structure of dynobactin A. f, Comparison of darobactin A and dynobactin A binding to BamA. Panels b, c reproduced from Kaur et al.⁹³ Panels d, f reproduced from ref.¹⁷.

The two intricate fused rings of darobactin appear as an obvious warhead, but unexpectedly, the co-crystal structure with the target shows that the main warhead is actually the peptide backbone. The peptide forms a canonical antiparallel β-β contact with the lateral gate of BamA (Fig. 4b,c). The same site is used by incoming polypeptides to bind BamA, enter into the lumen of the barrel, and get folded. Remarkably, the fused rings of darobactin force its peptide backbone into a rigid, preformed β-strand conformation that linear peptides do not adopt in solution. The fused rings have the additional function to orient aromatic side chains into direct contacts with lipid headgroups at the lateral gate. As a result, darobactin binds to the lateral gate of BamA 1,000 times better than a typical substrate, jamming insertion and disrupting formation of the outer membrane⁹⁶.

Bioinformatic analysis identified a whole family of darobactin-related compounds, with a total of 12 different members thus far^14,16. The two tryptophan rings appear conserved in most analogs, while other residues on the 7-mer peptide are variable¹⁰. So far, darobactin A appears as the most active, but not all analogs have been examined. Apart from the natural variants, nucleotide replacement in the propeptide gene darA resulted in a compound with a simple C-terminal Phe-Trp substitution, which extends the activity of the compound to Pseudomonas aeruginosa⁹⁷. Other substitutions gave analogs with good activity against Acinetobacter baumannii⁹⁸, but these have an added positive charge, a problematic feature for nephrotoxicity.

Recently, the total chemical synthesis of darobactin has been reported by two groups, using Larock macrocyclization for indole synthesis as the key step to the non-canonical ring closures^99,100. Chemical synthesis, as well as the production of novel variants by nucleotide reshuffling⁹⁷ open attractive possibilities for producing analogs with improved potency and pharmacological properties to combat Gram-negative pathogens.

A bioinformatic approach aimed at uncovering very distant homologs of the dar operon led to the discovery of dynobactins (Fig. 4d,e). These compounds also bind to BamA (Fig. 4f). Dynobactins have so far 16 recognized members, with the 10-mer dynobactin A being the prototype for this class. Like darobactin, dynobactin A⁹⁶ has two rings, but these are located at different positions and include a histidine residue. When binding to the target, seven of the ten residues overlap with darobactin, and the remaining three are kinked into the BamA lumen where they displace the BamA C-terminus. Importantly, the two aromatic ring fusions are at different spatial positions compared to darobactin, providing information on the structure-activity relationship (SAR) aiding efforts for the rational design of synthetic compounds that target BamA. In a striking case of convergence, both nature and chemists arrived at a similar design for stabilizing linear peptides that are flexible by creating “stapled peptides” – inserting a staple that rigidifies and maintains the three dimensional structure that would be otherwise be rapidly lost in a short peptide ¹⁰¹. The darobactin- BamA interaction is based nearly entirely on peptide backbone contacts at the BamA lateral gate, making the acquisition of resistance by point mutations difficult⁹⁶. Indeed, replacing amino acids in the BamA binding site will not change its peptide backbone. Resistance mutations outside of the lateral gate of BamA have been detected, but these mutants have diminished fitness and are avirulent. The BamA lateral gate thus represents a true bacterial Achilles heel that nature has exploited by producing the bicyclic rings of darobactin and dynobactin.

Prodrugs

Prodrugs are compounds that have no activity on their own, but become active once they enter into the cell and are metabolized. The first natural product prodrug was azomycin, a simple 2-nitroimidazole produced by some species of Nocardia, Streptomyces and Pseudomonas^102–104. Its derivative, metronidazole was synthesized soon after and introduced into clinical practice¹⁰⁵. As a result, most of what we know about this series comes from studies of metronidazole. Azomycin and metronidazole apparently share the same mode of action¹⁰⁶. Metronidazole is reduced by bacteria-specific nitroreductases into a reactive nitroso radical that damages DNA and probably other targets as well. The activating reaction requires a low oxygen environment, and metronidazole is therefore active against anaerobic bacteria such as C. difficile and microaerophilic Helicobacter pylori. Metronidazole has several attractive features worth considering for designing prodrug antibiotics. Covalent attachment to the targets provides an irreversible sink by binding covalently to DNA that helps with penetration of the compound. Indeed, it is broad-spectrum, small, and penetrates well into Gram-negative bacteria. The reactive species is produced specifically inside the bacterial cell, minimizing damage to the host, as human reductases lack the potential to reduce these nitroimidazoles. There are typically more than one activating nitroreductases in pathogens, limiting resistance development. Finally, targeting DNA provides for excellent killing of both growing and non-growing cells such as persisters⁷⁶.

While covalent binding to a target improves potency, as once bound the antibiotics permanently inactivate the target and cannot be effluxed out of the cell, there is a class of prodrugs that evolved specifically to transport a warhead across bacterial membranes. In sideromycins, an antibiotic moiety is linked to a sideophore¹⁰⁷. In the well-studied albomycin produced by Actinomyces subtropicus, a ferrichrome-type siderophore is linked to a nucleotide analog that is an inhibitor of the Ser-tRNA synthetase. The ferrichrome moiety actually binds Fe³⁺, as would a proper siderophore, and iron is required for albomycin activity. In Gram-negative bacteria, the ferrichrome moiety binds to TonB-dependent transporters of the outer membrane, and the compound is translocated into the cell. The peptide bond linking the chimera is cleaved in the cytoplasm, releasing the active Ser-tRNA synthetase inhibitor that blocks protein synthesis. This concept was emulated in cefiderocol, a ceftazidime (a β-lactam antibiotic) and a siderophore chimera¹⁰⁸. Cefiderocol, however is not a prodrug, it is transported into the periplasm of Gram-negative bacteria by siderophore transporters enabling the ceftazidime moiety to target PBPs in the periplasm. Serendipitously, linking the ceftazidime β-lactam to a siderophore improved resistance to β-lactamases. Cefiderocol is the first “Trojan horse” antibiotic to be approved for clinical use.

As we can see from these examples, the general prodrug design enables a variety of strategies to target bacteria. One interesting case worth mentioning is a recent discovery of 3’-amino 3’-deoxyguanosine (ADG), a prodrug produced by Photorhabdus luminescens¹⁷. ADG is a close analog of guanosine, in which an amino group at the 3’ position of ribose replaces OH. This small change dramatically affects the properties of the molecule. The 3’ OH serves to connect nucleotides in the growing chain of nucleic acids. Once inside the cell, ADG is phosphorylated by E. coli, into ADG-Pi, and subsequently converted it into ADG-PPPi, an analog of GTP. When ADG-PPPi is miss-incorporated into a growing chain of RNA, it terminates transcription, since the next nucleotide triphosphate cannot link to NH₂. This is similar to the mechanism of action of AZT, interrupting the synthesis of HIV DNA¹⁰⁹, and the chain termination DNA sequencing method developed by Sanger ¹¹⁰. It appears however that this ingenious approach was used by bacteria long before Sanger and AZT.

Outlooks

Our needs for new antibiotics have evolved over time. Staying ahead by introducing new compounds to combat resistant pathogens is the main goal of discovery, but other needs became apparent over the past decades. Chronic infections in immunocompromised patients, or those associated with indwelling devices that produce a favorable environment for biofilms, require antibiotics capable of killing drug-tolerant persisters. In addition, the positive role of the microbiome in maintaining human health calls for selective compounds that specifically target pathogens. The nature of a “pathogen” has expanded as well. Multidrug-resistant A. baumannii is a major concern in the hospital setting, a fairly recent phenomenon. Bacteroides fragilis, an important symbiont, can acquire a plasmid coding for a protease that destroys tight junctions of epithelial cells, becoming a “toxigenic B. fragilis”, a risk factor for colon cancer¹¹¹. Similarly, Morganella morganii, a common commensal, produces DNA-damaging indolimines that can be causal to colon cancer as well¹¹². A bloom of pro-inflammatory Enterobacteriacea exacerbates IBD and is a general hallmark of dysbiosis¹¹³.

These varied needs call for an increasingly sophisticated arsenal of antimicrobial measures. Such measures include traditional vaccines and antibiotics, but also new approaches in development – including therapeutic antibodies, modulation of the immune system, phage therapy, CRISPR-based targeting of pathogen DNA¹¹⁴, and microbiota-based therapeutics that are designed to prevent/block pathogen colonization, as we described for curing C. difficile infection. Which of these new approaches will generally be successful is hard to tell, but for now, vaccines and small molecule antibiotics are our best defense against bacterial pathogens.

In this review, we considered antibiotics with sophisticated modes of action which would be difficult, and in most cases unrealistic, to anticipate and rationally design. It is encouraging that some of these compounds - teixobactin, darobactin and dynobactin were discovered very recently, and from small collections of producers. This suggests that useful natural product antibiotics will continue to be discovered, and advances in computational genomics and cheminformatics are likely to contribute to this effort by identifying attractive biosynthetic gene clusters^16,19.

Many of the compounds considered here have more than one property of particular interest (Box 1). Natural products show us that virtually everything we may wish to have in an antibiotic is possible, from a low propensity for resistance selection to the potential to eradicate dormant cells or selective activity against a particular pathogen. Notably, most natural products were introduced into clinical practice in their original form. While, this trend is likely to continue, increasing demands for safety and the urgency of the AMR crisis compel us to deploy the knowledge of natural products into a rational discovery platform for synthetic compounds, and for making advanced analogs of natural antibiotics. Development of ETX0462 is an example of what is probably the first successful rational design of an antibiotic²⁷. ETX0462 was designed to target penicillin binding proteins, to penetrate through porins of Gram-negative bacteria, and to avoid hydrolysis by β-lactamases. ETX0462 is based on avibactam, a β-lactamase inhibitor, which in turn was based on penicillin. There are other examples of surpassing nature, including the recent development of the anti-Gram-negative agent G0775, a bacterial type I signal peptidase inhibitor derived from the antibiotic arylomycin¹¹⁵.

Box 1. Manifold features of individual antibiotics.

Many of the antibiotics discussed in this article have more than one attractive feature linked to their mechanism of action. This does not necessarily result from chemical complexity – the simplest molecule we consider, the prodrug azomycin and its analog metronidazole display an impressive array of features^{102–104 105,106}. Activation of the prodrug by several nitroreductases and binding to DNA and other targets contribute to resistance evasion. Small size favors penetration, as does covalent binding to DNA, creating an irreversible sink for accumulation on the target in the cell. Action against DNA helps kill dormant cells, thus anti-persister activity⁷⁶. Through the targeting of DNA, these molecules have the capacity to kill dormant cells, and have demonstrated anti-persister activity⁷³. More complex aminoglycosides, aminated sugars, bind to rRNA, interfering with the proper hybridization between aminoacyl-tRNA anticodons and codons of the mRNA. Aminoglycosides engage their target for long enough to produce a string of errors – mistranslation that results in aberrant, toxic peptides leading to cell death⁶⁶. Acyldepsipeptide binds to and activates the ClpP protease, replacing native chaperones and their substrates with undesirable substrates for proteolysis^79–89. The result is a dysregulated protease that digests essential proteins, killing both regular cells and persisters. Darobactin, a 7-mer peptide is a rigid β-strand that forms when four of its amino acids are linked into two fused rings^14,96. This peptide backbone is the warhead that forms a β-β interaction with the lateral gate of BamA, thus preventing entry of incoming peptides that require folding on BamA, resulting in an aberrant outer membrane and death. Darobactin binds to the peptide backbone of its target which by definition cannot be changed through mutations, contributing to resistance evasion. Resistance mutations occur elsewhere in the protein, but at the price of diminished fitness. Teixobactin is complex, with hydrophobic side groups that anchor it to the membrane, interchanging L- and D-amino acids, a 13-member ring and a rare enduracididine at the C-terminal^58,59. Binding to the target – the PPi-sugar moiety of Lipid II, precursor of peptidoglycan – seems unremarkable, but once engaged, teixobactins form an anti-parallel β-β sheet supramolecular structure. The β-β interaction is accomplished by the strategically placed L- and D-amino acids. The supramolecular structure thins the membrane, aided by hydrophobic tails of its amino acids, producing leakage that contributes to cell death. The large structure that forms has high stability in the membrane, equivalent to covalent binding. Engaging non-protein targets that are not directly coded by genes results in resistance evasion. The related compound clovibactin also binds to Lipid II, but the target is perhaps the simplest possible, PPi¹². Surprisingly, the target is enclosed by a hydrophobic “glove” of the antibiotic, displacing water. This entropic interaction is weak, but becomes irreversible when the interacting clovibactins form a supramolecular structure. Binding to an immutable PPi is probably the ultimate feat in resistance evasion.

Of the targets considered here, some would not have made it onto the list of a typical drug discovery program. BamA is a β-barrel protein that lacks a catalytic center, but it is nonetheless an unexpected and powerful target of darobactin and dynobactin. Pyrophosphate seems particularly unsuitable as a target, being present in the environment as polyphosphate and in many cellular metabolites, but it is nonetheless a target of clovibactin. Nature seems oblivious to our notions of undruggable or unsuitable targets, broadening our appreciation for what is possible and pointing us in directions we would not have contemplated on our own.

Data availability –

Not applicable