The discovery of antibiotics revolutionized medicine in the 20th century. In 1900, infectious disease was a leading cause of death; in 2000, infectious diseases were responsible for only a small percentage of deaths in developed nations [4]. Antibiotic consumption continues to rise among the low- and middle-income countries in recent years [5]. Unfortunately, bacteria evolve rapidly, and resistance mechanisms have developed and spread for every antibiotic in clinical use soon after its introduction. Some fear a return to the pre-antibiotic era because of the increasing prevalence of multi-drug resistant superbugs, the decreasing number of novel antibiotics to enter the market, and the departure of several pharmaceutical companies from the field of antibiotic research due to scientific, economic, and regulatory challenges. There is an urgent need for solutions that can drive rapid discoveries of novel antibiotics, especially those against multidrug-resistant tuberculosis and Gram-negative bacteria and community-acquired infections such as Salmonella spp, Campylobacter spp, N gonorrhoeae, and H pylori, as prioritized by World Health Organization (WHO) [6]. WHO also reviews annually clinical and preclinical antibacterial pipelines [7]. Here, I review past and current methods for the discovery and development of antibiotics and overcoming bacterial resistance.

S.A. Waksman defined an antibiotic as “a chemical substance, produced by micro-organisms, which can inhibit the growth of and even to destroy bacteria and other micro-organisms” [8]. Today, the term antibiotic is used more broadly to include any anti-microbial compound, whether of natural or synthetic origin [4]. Antibiotics inhibit bacterial growth by targeting essential cellular processes such as the synthesis of the bacterial cell wall, DNA/RNA, and proteins (Figure 1). Natural product antibiotics are secondary metabolites produced by microbes at the end of the exponential phase of growth. Secondary metabolites are not essential for the growth of the organism [9]. Instead, they have diverse roles, such as in cellular differentiation, nutrient sequestration, metal transport, ecological interactions, and defense. They are usually strain-specific and are large molecules that have complex biosynthetic pathways [9]. Up until recently, it was assumed that microorganisms produced antibiotics as chemical warfare among species within an environmental niche. However, new studies have shown that sub-inhibitory concentrations of antibiotics can have broad transcriptional effects and suggest evolutionary roles other than mediating competitive interactions [10].

[enlarge]

Figure 1. Antibiotic targets and mechanisms of resistance. From [1].

In 1909, Paul Ehrlich discovered the very first antibiotics, synthetic arsenic-based drugs (Salvarsan), in a large screen of hundreds of organoarsenic compounds for use in the treatment of syphilis [4]. The next antibacterials to be clinically used were the synthetic sulfonamides or the sulfa drugs. Gerard Domagk discovered Prontosil, a red dye, while experimenting with azo dyes against bacterial infections in mice in 1935 [11]. The sulfonamide group was identified as the critical component in Prontosil, and this led to the synthesis of more than 5000 sulfa drugs between 1935 and 1945 [12, 13]. Sulfa drugs are broad-spectrum and were used against many bacterial infections such as pneumonia, gonorrhea and streptococcal infections. They act by inhibiting the dihydropteroate synthase (DHPS) enzyme in folate biosynthesis. Sulfa drugs dropped in their use because of bacterial resistance, which arises by mutations in the DHPS enzyme [14], and the discovery of penicillin and other natural product antibiotics. Nowadays they are given in combination with trimethoprim, which inhibits a later step in folate synthesis.

Alexander Fleming discovered penicillin, the first natural product antibiotic, in 1928. He observed that Penicillium molds produced a diffusible extract that had antibacterial activity against staphylococci [15]. Although Fleming performed several experiments in vitro, he did not test the extract against animal models, and penicillin was only used as a local antiseptic for many years [16]. Penicillin was also very difficult to isolate and purify, until the late 1930s. In 1939, Florey and Chain elucidated the structure of penicillin, and in 1940 they showed that penicillin was active against streptococcal infection in mice [17]. Following this discovery, penicillin started to be used systemically as an antibiotic, which then ushered in the golden age of antibiotics. In 1945, Fleming, Chain, and Florey were awarded the Nobel Prize in Physiology or Medicine for their discovery of penicillin.

Selman Waksman received the Nobel prize in 1952 for his “discovery of streptomycin, the first antibiotic effective against tuberculosis.” Waksman began systematically screening soil microorganisms for antibiotic production, and quickly discovered that soil actinomycetes are prodigious producers [18]. The screening method, often called the Waksman platform, involved preparing culture extracts from soil actinomycetes and overlaying filter paper discs infused with these extracts over a test organism on an agar plate, and then looking for zones of growth inhibition [19, 20]. There were efforts to systematically collect many soil samples and isolate as many novel actinomycetes as possible. The Waksman platform was adopted by several pharmaceutical companies and was very successful for two decades, resulting in the discovery of many of the classes of antibiotics currently in clinical use today. However, with the discovery of each novel antibiotic, the probability of finding the next novel compound became lower and lower. The platform collapsed because of the rediscovery of the same compounds [19, 20].

From the 1960s, many pharmaceutical companies started to employ rational screens for the discovery of cell wall inhibitors [19]. A spheroplasting method was used as a primary screen at Merck for many years and led to the discoveries of fosfomycin (a broad-spectrum phosphonate antibiotic), cephamycin C, thienamycin and other carbapenems. At Fujisawa, a Japanese company, an E. coli mutant that was hypersensitive to beta-lactams was used to screen for the first monobactam compound Nocardicin E. An L-form assay, which employed L-forms of bacteria that lack cell walls, were used at Lepetit to identify the glycopeptide antibiotics teicoplanin and ramoplanin [19].

The Center for Disease Control, USA, estimated in 2020 that antibiotic-resistant microbes cause 2.8 million infections and at least 35,000 deaths per year [21]. The Comprehensive Antibiotic Resistance Database / CARD (https://card.mcmaster.ca/) catalogs resistance genes, their products and associated phenotypes from 2685 publications as of June 6, 2020 [22]. Pathogenic strains may be multidrug resistant (resistant to three or more antibiotic categories), extensively-drug resistant (resistant to all except one or two antibiotic categories), and even pan-drug resistant (resistant to all known antibiotics) [23]. Antibiotic resistance mechanisms existed long before the use of antibiotics [24]. However, the widespread use of antibiotics has provided a selective pressure to maintain antibiotic resistance, and for the transfer of resistance genes either vertically from mother to daughter cell or horizontally through lateral gene transfer. Bacteria quickly gain advantageous mutations because their rapid growth allows even rare mutations to get selected. Thus, antibiotics are never effective long-term [25].

Resistance to antibiotics may be innate or acquired [25]. Innate resistance is provided by characteristics that are inherent in the physiology or structure of the bacteria [25, 26]. For instance, Gram-negative bacteria are intrinsically resistant to drugs like daptomycin because of differences in cell membrane composition, or to glycopeptides such as vancomycin that cannot cross the Gram-negative outer membrane. The other common mechanism of innate resistance, also prevalent among Gram-negative bacteria, is through efflux pumps that pump out the antibiotic [25, 27]. Bacteria can undergo DNA inversion–mediated phase variation to display novel phenotype [28].

Bacteria acquire resistance by 1) reducing intracellular concentration of the antibiotic, 2) modifying antibiotic targets, and 3) inactivating the antibiotic (Figure 1). The intracellular antibiotic concentration may be reduced by blocking entry of the antibiotic or pumping the antibiotic out by efflux pumps. Downregulation of porins or replacement of porins with more selective channels can reduce the permeability of the cell membrane to antibiotics [29, 30]. Overexpression of efflux pumps, often by modulation of transcription regulators, is a common resistance mechanism seen in clinical isolates [25]. Some efflux pumps are carried on mobilizable elements, allowing bacteria to acquire them by horizontal gene transfer [31].

Bacteria can modify the cellular target through mutations such that the antibiotic can no longer bind. Targets may be modified in alternate ways. 1) Natural transformation, where DNA is taken up from the environment, followed by homologous recombination can result in mosaic genes that are resistant [32]. 2) Sensitive strains could acquire, by horizontal gene transfer, a homologous target gene that is insensitive to the antibiotic. For example, the mec gene on a mobilizable cassette makes a penicillin-binding protein (PBP2a) that is insensitive to methicillin; acquisition of the mec cassette renders Staphylococcus aureus methicillin-resistant (MRSA) even though the native sensitive PBP is present [33]. 3) Targets may be altered enzymatically. Examples include resistance to macrolides and streptogramins by erythromycin ribosome methylase (erm) that methylates the 16S rRNA [34], aminoglycoside resistance through ribosome methylation [35], and quinolone resistance mediated by binding of Qnr proteins to DNA gyrase and topoisomerase IV [36]. The erm and qnr genes are also plasmid-borne and can be laterally transferred. 4) Through changes in expression of regulatory genes as seen in polymyxin resistance mediated by changes in the PmrAB two-component system that affects LPS production [37], and enterococcal resistance to daptomycin through mutations in the LiaFSR regulatory system that spatially reorganizes the lipid-rich domains in the membrane and blocks daptomycin binding [38].

Bacteria can inactivate antibiotics through hydrolysis or enzymatic modification [25]. Classic examples of hydrolytic enzymes are the beta-lactamases. Pathogens have evolved diverse beta-lactamases – extended-spectrum beta-lactamases, metallo-beta-lactamases, carbapenemases – that have kept up with the evolution of the semi-synthetic and synthetic beta-lactam antibiotics. Isolates have emerged in recent years that are resistant to all known beta-lactam antibiotics. The detection of different isolates is an active field of research [39]. Aminoglycoside resistance is commonly mediated through aminoglycoside-modifying enzymes (acetyltransferases, phosphotransferases, and nucleotidyltransferases) that block the binding of the antibiotic to its cellular target.

Apart from the antibiotic resistance mechanisms described above, bacteria can also be antibiotic tolerant through the formation of persister cells. Resistant strains elevate the minimum inhibitory concentration (MIC) of an antibiotic; while tolerant ones prolongs the treatment to kill the population without altering the MIC. Persisters are rare, less than 1% of a bacterial population, and they arise stochastically or through specific environmental cues. They are non-growing or dormant cells, and thus are resistant to most antibiotics that target active cellular processes. Persisters linger in the infection and extend treatment times, and therefore can lead to increasing populations of resistant bacteria [40]. For example, tissue-associated, antibiotic-resistant facultative intracellular entero-pathogen Salmonella enterica serovar Typhimurium could re-seed gut lumen and transfer the resistant plasmids to other strains of Enterobacteriaceae [41].

Semi-synthesis of antibiotics was the first response to the bacterial resistance problem. Semi-synthesis was a highly successful strategy where the natural antibiotic scaffold was modified chemically to produce new antibiotics with higher activity than the original (Table 1). The semi-synthetic antibiotics helped to prolong the life of the antibiotic by circumventing the resistance mechanisms. They were also designed to be more effective, have improved spectrum and potency, and better pharmacokinetics [42, 43].

Resistance to penicillin and other beta-lactams occur through hydrolysis of the beta-lactam ring. An important breakthrough came with the discovery of a fermentation route for the production of penicillin [16, 44]. The penicillin nucleus, 6-amino penicillanic acid (6-APA), was isolated from fermentation broths. 6-APA had no sidechains and also had no antibacterial activity [45]. This finding led to the idea that different sidechains could be attached to 6-APA to get molecules with different activities and properties such as resistance to beta-lactamases, improved activity against Gram-negative bacteria and better oral absorption. Thus, in the 1960s scientists created semi-synthetic penicillins such as methicillin, cloxacillin, ampicillin amoxicillin, carbenicillin and piperacillin [16, 46].

Abraham and Newton isolated cephalosporin C produced by the mold Cephalosporium acremonium in 1956 [47]. Its nucleus (7-ACA) was not naturally discovered, but because of insights from the penicillin nucleus, 7-ACA was synthesized chemically by removing the side-chain. New side-chains were added to create new cephalosporin derivatives [48]. Third- and fourth-generation cephalosporins were active against Gram-negative bacteria by their ability to penetrate the porins. The fifth-generation cephalosporin (ceftobiprole) was approved for use in 2013. The discovery of 6-APA and 7-ACA and their fermentations and semi-synthesis processes respectively have led to the commercialization of more than 50 antibiotics [42]. This illustrates the ability of medicinal chemists to tailor antibiotic scaffolds to produce newer and active molecules that can meet the clinical needs.

Eli Lilly discovered the glycopeptide vancomycin in the 1950s in a search for new antibiotics in soils [49]. Glycopeptides inhibit cell wall synthesis by binding to the D-Ala-D-Ala end of the lipid II stem and are active against Gram-positive bacteria. Vancomycin used to be a last resort treatment owing to its toxicity, but now it is becoming a frontline treatment, even in animal experiments [50]. Semi-synthetics have been made to combat the rise of vancomycin-resistant enterococci (VRE) and vancomycin-resistant S. aureus (VRSA): telavancin (approved in 2009) and oritavancin (approved in 2014) have reduced toxicity and higher potency [51]. Teicoplanin is another glycopeptide in clinical use that was isolated from Actinoplanes teichomyceticus in 1978, and dalbavancin is its semi-synthetic derivative [51]. Resistance to glycopeptides occurs by mutation of the D-Ala-D-Ala tails to D-Ala-D-lac, so chemists are synthesizing newer generations of glycopeptides that can recognize both chains [52]. Recent studies have reported vancomycin derivatives carrying lipid tails at the C-terminus that can penetrate the Gram-negative cell membrane [53].

Aminoglycosides are broad-spectrum antibiotics produced by actinomycetes. They block protein synthesis by binding to the ribosome and inhibiting translation [54]. Streptomycin, from Streptomyces griseus, was the first aminoglycoside to be discovered (1943). Other early aminoglycoside antibiotics were neomycin (1949), gentamicin (1963), tobramycin (1967), and sisomycin (1970). Resistance can arise through efflux pumps, mutations to the ribosome, and modifications through ribosomal methyltransferases, but most commonly through inactivation by aminoglycoside-modifying enzymes [55]. Scientists discovered that removing the 3’-OH group rendered the aminoglycoside resistant to the modifying enzymes. Dibekacin, the first semi-synthetic to be used clinically, was synthesized from kanamycin after removal of the 3’ and 4’-OH groups [43]. Other semi-synthetics were launched in the 1970s - arbekacin, amikacin, isepamicin, netilmicin - with improved pharmacological properties and less susceptibility to modifying enzymes [56]. The only new aminoglycoside to be recently approved (2018) was plazomicin (developed by Achaogen), a derivative of sisomycin; plazomicin is active against MRSA when combined with daptomycin [57, 58]. Spectinomycin analogs called spectinamides are promising narrow-spectrum antibiotic candidates currently in preclinical trials. They are effective against Mycobacterium tuberculosis by inhibiting ribosome translocation and evading the mycobacterial efflux pump [59].

Eli Lilly discovered the first macrolide antibiotic, erythromycin, in 1949 [42]. While erythromycin showed good activity against Gram-positive organisms, it had poor oral bioavailability and was unstable under acidic conditions. Chemists devised analogs exhibiting better stability and oral absorption (clarithromycin), and with extended activity spectrum (azithromycin in 1991). Azithromycin was the 7th most prescribed drug in the US in 2010. With the rise in resistance towards clarithromycin and azithromycin, Abbot labs developed the ketolide antibiotics semi-synthetically from erythromycin. The first ketolide, telithromycin, was approved in 2004, and a new ketolide solithromycin is now in phase III clinical trials [42].

Chlortetracycline, the first tetracycline to be discovered (1948), and oxytetracycline were the first broad-spectrum antibiotics that were active against Gram-negative bacteria [4, 42]. Tetracycline was first synthesized from chlortetracycline before being discovered as a natural product. Subsequently, more stable derivatives were synthesized, leading to the production of minocycline with improved activity spectrum [42]. Tigecycline was developed to combat the rise in tetracycline resistance. Tigecycline is often the last line of defense against multidrug-resistant bacteria [42].

Following the boom in natural product antibiotic discovery in the 1940s and 1950s, researchers once again turned toward the design of synthetic antibacterials. New classes of antibiotics and several approved drugs resulted from fully synthetic approaches. Many successful attempts were also made towards fully synthetic routes to derivatives of natural product antibiotics since natural products were often difficult to isolate. Here I discuss some of the important synthetic antibiotics.

In the 1960s, researchers at Wellcome Laboratories observed that purine and pyrimidine analogs could inhibit bacteria. They tested several diaminopyrimidines and discovered trimethoprim [60]. Trimethoprim inhibits folate synthesis and is given usually in combination with sulfonamides [4]. It is cheap to produce and is, therefore, a very widely used therapy in developing countries [42].

In the late 1950s, researchers at the University of Tokyo discovered a nitroimidazole that was active against the parasite Trichomonas vaginalis. Subsequent synthesis and testing of different nitroimidazoles led to the creation of metronidazole, the first drug that was effective against trichomoniasis. Later, it was accidentally discovered that metronidazole was effective against infections caused by anaerobic bacteria, and now it is used against Clostridium difficile infections [42].

Quinolone antibiotics were discovered in the 1960s when a by-product from the synthesis of the anti-malaria drug chloroquine was included in a screening program at Sterling-Winthrop Research Institute [4, 42]. The first clinically approved quinolone was nalidixic acid; derivatives were created for increased potency and broader spectrum with the fluoroquinolones being the most successful. Quinolones were initially used for the treatment of certain Gram-negative infections, but newer generations of fluoroquinolones are potent against Gram-positive bacteria as well. Examples of clinically used fluoroquinolones include norfloxacin (1977), ciprofloxacin (1987), levofloxacin (1996), and moxifloxacin (1999). The quinolones are a hugely successful class of antibiotics, with more than 10,000 quinolones synthesized and over 25 antibiotics clinically approved [4, 42]. In 2019, Luther A et al reported the development of chimeric peptidomimetic antibiotics against Gram-negative bacteria; its mechanism involves binding to the main component of the beta-barrel folding complex and lipopolysaccharide in Gram-negative bacteria [61]. Of the 30 antibiotics approved between 2000 and 2015, 11 were quinolones [62]. Resistance to quinolones was slower to develop than other antibiotics because they act on two targets - DNA gyrase and topoisomerase IV [4], but now resistance is increasingly common [63].

The next innovation came with the discovery of carbapenems, which are bicyclic beta-lactams. Thienamycin was the first carbapenem, which was isolated as a natural product by scientists at Merck in late 1970s [64]. Although it had broad-spectrum activity, it was chemically unstable and difficult to isolate and purify. This led chemists to devise a fully synthetic route to more stable and active carbapenems, with imipenem [65] being the first carbapenem to be used clinically (approved 1985), and meropenem (approved 1996) having better pharmacokinetics [66], and ertapenem (approved 2001) with less frequent dosing [67]. Monocyclic beta-lactams or monobactams were discovered as natural products in 1981. Chemists devised routes to synthesize monobactams fully, and aztreonam was approved in 1984. Aztreonam is useful against Gram-negative bacterial infections [42].

The oxazolidinones are an important class of fully synthetic antibiotics. In 1984, researchers at Dupont first observed the antibacterial properties of oxazolidinones while testing against plant pathogens [42]. Dupont synthesized many oxazolidinones that were active against human pathogens, but the compounds had toxicity issues. Researchers at Upjohn developed the first safe and potent oxazolidinone. Linezolid (2000) was the first antibiotic of a new structural class in 40 years (since the discovery of nalidixic acid in 1960). It is the last line of defense against MRSA and VRE. Research is ongoing for next-generation oxazolidinones [42].

Researchers have developed methods to fully synthesize tetracycline (in 2005) and macrolide antibiotics (in 2016), the advantage being that complete synthesis allows the building of diverse chemical structures that are not possible via semi-synthesis [68, 69]. Tetraphase Pharmaceuticals has synthesized more than 3000 tetracycline antibiotics [70] of which eravacycline was approved earlier this year (2018) for use in intra-abdominal infections. Scientists at Harvard University developed a platform to completely synthesize macrolide antibiotics instead of starting from erythromycin, resulting in far fewer synthesis steps, and have created more than 300 macrolide antibiotic candidates [68].

New derivatives of existing antibiotics continue to be synthesized today, and medicinal chemistry is likely to always play a role in antibiotic development.

From the 1970s to 2000, antibiotic research went through a dry spell with no new class of antibiotics discovered. Although the total numbers of antibiotics increased, they were analogs of known compounds that were susceptible to similar resistance mechanisms. With the availability of new tools such as high-throughput screening methods and analytical instrumentation with increased resolution, many pharmaceutical companies focused on developing synthetic antibiotics through target-based drug discovery campaigns. Glaxo Smith Kline ran 70 campaigns between 1995-2001 and Astra-Zeneca ran 65 between 2000-2010. While similar campaigns were very successful for anti-virals and other non-infectious diseases, not a single antibiotic was discovered through these pipelines [71-73].

Target-based drug discovery begins with identifying proteins that are essential to the pathogen, for example, the enzymes involved in RNA transcription, RNA translation, and DNA replication [71, 74]. After the proteins are characterized and validated as targets, they are screened against a panel of small molecules or chemical libraries to find those that inhibit the protein. The hits are then evaluated for their suitability as drugs – low toxicity to humans, good bacterial penetration and bioavailability – and chosen lead molecules are further studied, and attempts are made to improve their potency and pharmacokinetic properties.

There are many reasons why target-based campaigns failed in the quest for antibiotics. Many of the hits obtained during screening poorly penetrated bacterial cells. Bacteria have evolved to interact with a broad spectrum of small molecules for functions such as cell communication, differentiation, and defense, and so they have many methods to sense and block, inactivate or metabolize these molecules [75]. Bacterial cell membranes, particularly that of Gram-negative bacteria, act as an effective barrier blocking penetration of many substances. The Gram-negative cell membrane has an outer and an inner membrane. The porins in the outer membrane allow entry of only specific molecules and block others [76], and the efflux pumps of the inner membrane actively remove a wide range of molecules.

The screening campaigns adopted the Lipinski’s “rule of 5” to choose compounds with acceptable physicochemical properties for use in the screening library [77]. Drugs are likely to be poorly absorbed if: they have more than 5 H-bond donors, their molecular weight is > 500, their hydrophobicity measure is >5, and they have more than 10 H-bond acceptors. Lipinski based his rule on thousands of drugs but also observed that the rule of 5 did not apply to compounds such as antibiotics that were substrates for biological transporters [77]. Thus, the screening libraries themselves were biased against antibiotics. Since the failure of these campaigns, groups have attempted to define the “antibiotic chemical space” and identify properties that characterize antibiotic absorption and permeability. Such efforts revealed that the drugs that are active against Gram-positive bacteria have a different profile than those that are active against Gram-negative and these differ from other non-antibiotic drugs, especially in their molecular weight and hydrophobicity [78].

Another factor for the failure of the campaigns is that high concentrations of antibiotics are needed (as compared to drugs for non-infectious diseases) to prevent rapid bacterial growth and any resistance from developing during treatment [72, 79]. High drug concentrations lead to problems with drug toxicity and safety, which prevents many hits from being developed further. Also, pharmaceutical companies were preferentially searching only for broad-spectrum antibiotic activity, and it was challenging to find inhibitors that could effectively target both Gram-positive and Gram-negative bacteria [73].

The failure of drug discovery campaigns has caused many pharmaceutical companies to withdraw from antibiotic research. Apart from the scientific bottlenecks described above, the drug industry faces economic and regulatory barriers [71-73]. The FDA standards for clinical trials are particularly challenging for antibiotics. It is unethical to use placebos in antibiotic trials; expensive clinical trials are needed to demonstrate that the efficacy of the antibiotic is more significant than existing drugs. The return of investment on antibiotics as compared with other drugs is another challenge. Antibiotics are prescribed for acute infections and therefore only taken for a few days, whereas drugs for chronic diseases may be prescribed for years [79]. Antibiotics go off-patent quickly, and generic drugs become available at lower prices, which leads to public expectations for low prices on new antibiotics also. Since bacteria rapidly develop resistance towards new antibiotics, new drugs stay on the market for only a few years. In 2013, the US Congress passed the GAIN (Generating Antibiotics Incentives Now) initiative that lowered the regulations for industries. Several national and international initiatives incentivize research into antibiotics [80], such as the Innovative Medicines initiative, New Drugs for Bad Bugs (IM-ND4BB), a public-private partnership fund that supports clinical development of antibiotics and supports research into how drugs can be made to penetrate Gram-negative bacteria. It is hoped that such initiatives entice companies to continue much-needed research into antibiotic discovery.

Many research groups and drug companies are now revisiting natural products. Unlike synthetic molecules, natural microbial products have evolved to be able to penetrate the microbial cell membrane. The majority of clinically used antibiotics are derived from natural products. However, there are several reasons why drug pharma companies hesitate to pursue natural products. I discuss below two main challenges.

There are several thousand natural products known – 25-27,000 microbial antibiotics by a 2010 estimate [81] – but only a few scaffolds. From the 1970s, although the total numbers of natural products discovered increased, there were no new classes of antibiotics. Dereplication is the process of discarding known scaffolds. It is an expensive process because the product has to be purified and characterized. Dereplication is made more challenging because the natural products are made in small quantities, the producing organism may lose the ability to make the compound, or the product may be unstable [71].

Better analytical methods and instruments are addressing the problem of dereplication. Advances in NMR instrumentation and mass spectrometry methods allow analysis of sub-milligram quantities of material, generate data faster and have improved sensitivity and resolution. This enables chemists to rapidly identify if the compound is a known or novel scaffold; decisions on further purification can be then quickly made [71]. Next-generation sequencing and new bioinformatic tools such as antiSMASH [82] allow prediction of the numbers and structures of natural products synthesized by a genome, thus also aiding dereplication. Several computational tools have been created for dereplication. GNPS (Global Natural Product Social molecular networking) is a database that allows sharing and community curation of mass spectrometry data [83]. GNPS was instrumental in the recent analysis of 146 Salinispora and Streptomyces strains that identified 15 families of diverse natural products [84]. Tools such as DEREPLICATOR [85] and VarQuest [86] helps identify peptidic natural products based on mass spectra, and iSNAP helps identify non-ribosomal peptides [87].

Researchers at Merck created an S. aureus fitness test assay [88] with 245 S. aureus strains each with a xylose-inducible antisense RNA that targets an essential gene. The antisense RNA reduces levels of the target gene such that greater inhibition or sensitivity than wild-type strain can be observed with an antibiotic that acts on that particular gene product. When the strains are pooled and grown with test antibiotics, the abundance of each strain increases or decreases depending on the mode of action of the antibiotic. Such anti-sense strain abundance profiles were generated for 59 well-characterized antibiotics and are a useful dereplication tool for comparing new compounds. From a library of natural product extracts, coelomycin from a Coelomycete fungus produced a unique profile, proving the method’s validity for identifying novel antibiotic classes [89]. A similar method was developed in 2017 where unique transcriptome profiles were generated by RNA-seq for S aureus with different types of antibiotics [90]. S. aureus may be challenged with a crude natural product extract and its transcriptome profile compared to the database to determine if the product is novel or not. This method can facilitate high-throughput screening of extracts and identification of new leads without purification or characterization of the compounds.

Another dereplication tool, the Antibiotic Resistance Platform (ARP) developed in 2017 employs a library of E. coli strains, each expressing an individual resistance gene [91]. Over 40 genes conferring resistance to 15 classes of antibiotics are represented on the ARP. Natural product extracts are screened against the ARP, and identity to known antibiotics is evaluated based on the specific resistance gene-containing ARP strain that showed reduced susceptibility to the extract. Novel products do not have a corresponding resistance gene on the ARP.

Pharmaceutical companies focused their drug discovery campaigns on broad-spectrum antibiotics that can kill both Gram-positive and Gram-negative bacteria or at least a wide variety of pathogens. There are two reasons for this. An antibiotic that can be used in the treatment of multiple diseases would provide a better return on investment. Broad-spectrum antibiotics can be prescribed without knowledge of the specific pathogen, while a narrow-spectrum antibiotic can be prescribed only after diagnosis which causes a delay in treatment with current diagnostic methods [71]. Rapid and accurate diagnostics need to be developed before narrow-spectrum antibiotics can be prescribed routinely.

Most natural product antibiotics are narrow-spectrum or even species-selective, therefore screening campaigns that are designed towards pathogen-specific antibiotics are likely to be more successful [79]. One significant advantage of using a narrow-spectrum antibiotic is that it causes less harm to our gut flora. Broad-spectrum antibiotics play havoc with our normal gut flora and can cause resulting infections by opportunistic pathogens like Clostridium difficile. When gut microbes are affected by an antibiotic, it encourages resistance to develop and possibly transfer to the pathogen by lateral transfer. Narrow-spectrum antibiotics may thus slow the rise of resistance. Species-selective antibiotic screens can also eliminate nuisance compounds like detergents that are broadly toxic [79]. Useful antibiotics may be discovered from the extensive existing natural product database by re-examining them for species-selective activity.

Between 2000 and 2015, 30 new antibiotics have been launched, including five new classes of antibiotics [62]. Two of the five new antibiotics classes are synthetic: oxazolidinones (linezolid in 2000) and diarylquinolines (bedaquiline in 2012 for tuberculosis), which was found to accumulate in the lipid droplets of infected human macrophages [92]. Two are actinomycete natural products: the lipopeptide daptomycin in 2003 and fidaxomicin (of the tiacumicin family) in 2010. The fifth, retapamulin, is a fungal natural product derived from pleuromutilin and was approved in 2007 for topical use [93]. However, these five new antibiotic classes target only Gram-positives, and there is an urgent need for novel drugs for Gram-negative infections. Of the 30 antibiotics launched, 16 are natural products or their derivatives. Here I discuss the various approaches research groups are employing in their quest for new natural product antibiotics.

In the search for new antibiotics, scientists are tapping into the immense natural biodiversity available, developing strategies to select for novel antibiotic producers among soil actinomycetes, and examining previously uncultured bacteria by novel cultivation methods.

Soil actinomycetes, especially Streptomyces species, have been the source of most natural product antibiotics in use today. Decades of exploiting terrestrial streptomycetes have recovered the same classes of natural compounds rather than unique ones. Therefore, the search has shifted in recent years to rare actinomycetes (typically non-streptomycetes) and other taxa of bacteria such as Cyanobacteria [94, 95] and Proteobacteria. Several Gram-negative bacilli of genera Pseudomonas, Burkholderia, Janthinobacterium, to name a few, produce antibiotics [96, 97]. Novel actinomycetes species and their novel natural products have been discovered in deep ocean sediments [98], hyper-arid desert soils [99], and hot springs [100] to name a few unusual habitats. Endophytic bacteria present in plant tissues and plant rhizospheres [101], and symbiotic bacteria such as the actinobacteria living in mutualistic association with fungal growing attine ants [102], bacterial-nematode associations [103], or even human commensals [104] are being sourced for novel antibiotics. Examples of natural products obtained from unusual sources are listed in Table 2. Goodfellow and Fiedler [105] have outlined a bioprospecting strategy where species from extreme or unusual habitats are cultivated with selective isolation methods, novel taxa are recognized by dereplication and then screened for natural products.

It has been estimated that less than 10^-12 of the earth's soil has been screened for actinomycetes and that less than 3% of even the Streptomyces antibiotics have been discovered [120, 121]. The frequency of discovery of a particular antibiotic among actinomycetes has ranged from 1 in 10 to 1 in 10⁷. Streptomycin can be found if you screen 100 random soil actinomycetes, but Eli Lilly had to screen 5 million actinomycetes to find daptomycin [122]. Since the most abundant antibiotics have already been found, companies would need to screen tens of millions of strains to discover the remaining 97%.

Cubist Pharmaceuticals has employed a miniaturized high-throughput fermentation screen with calcium alginate microdroplet beads (~2 mm diameter); this method allows screening of millions of actinomycetes per year. To exclude known antibiotics, they screen using an engineered E. coli bearing 15 antibiotic resistance genes. This strategy excludes Gram-positive only antibiotics and most known broad-spectrum antibiotics [122].

Antibiotic selection can be used to weed out common and known antibiotic-producers while screening soil samples. Since antibiotic producers also encode genes for antibiotic resistance, Thaker et al, used vancomycin selection to enrich the number of glycopeptide producers [123]. The vancomycin-resistant strains were then examined for their glycopeptide biosynthetic genes, and a phylogenetic tree of the test strains and known glycopeptide producers was constructed. This method permitted an early dereplication to eliminate known glycopeptide producers and predict strains producing a novel scaffold. They successfully isolated a new glycopeptide antibiotic called pekiskomycin. Thaker’s method improved the discovery of new glycopeptides by four orders of magnitude [123].

Earlier this year, an ultra-high-throughput screening method for cultivation of slow-growing actinomycetes was published [124]. Environmental cells are encapsulated in picolitre droplets and allowed to grow for a month. Detection of secondary metabolites may be coupled to mass spectrometry, or by pico-injection of a fluorescently labeled reporter strain followed by fluorescence-based droplet sorting [124].

Less than 1% of the microbes present in the environment can be cultivated in the laboratory [125]. Traditional cultivation methods allow faster growing species to dominate, and rich medium supplies excess nutrients that may be toxic to oligophilic bacteria [126]. Many environmental bacteria only make microcolonies that require microscopic identification, and conventional methods of monitoring growth such as colony formation and turbidity have less utility. A major reason why many species are uncultivatable is that synthetic medium lacks essential nutrients or growth factors that are present in their natural environment [127].

There has been a move in recent years to simulate the natural environment in isolation efforts. A very low nutrient medium has been used successfully in high-throughput cultivation for new marine isolates [128]. Ferrari devised a microcolony cultivation method with a polycarbonate membrane as the growth support and soil extract as the medium. Microbes grew as microcolonies on the membrane and FISH (fluorescence in situ hybridization) was used to identify them [129]. Epstein introduced the diffusion chamber in 2002 for in situ cultivation; the chamber allowed growth in pure culture while permitting the free diffusion of nutrients from the native environment [130] (Figure 2). The ichip or isolation chip consists of hundreds of miniature diffusion chambers that allow parallel in situ cultivation of hundreds of isolates [3]. Each chamber is inoculated with a single environmental cell, and the chip is incubated in situ in the original environment. The method had not only high recovery rates (close to 50%) but also recovered novel species. There was minimal overlap in the species recovered with the traditional Petri dish method of cultivation. Cultivation by ichip also allows domestication of the isolates, whereby the cells can grow on synthetic medium in the laboratory after just one to four rounds of in situ cultivation [131]. While the ichip has been successful in aquatic environments and moist soils, it cannot be used in arid environments because the chip requires moisture. Another limitation of the ichip is that because it grows each cell in isolation, it cannot be used for microbes that only grow in co-culture [3]. The ichip platform has been licensed to Novobiotic Pharmaceuticals.

Examining uncultured bacteria for natural products has generated two significant antibiotic leads. Lassomycin, produced by Lentzea kentuckyensis, was identified from a screen of extracts from uncultured species against M. tuberculosis [132]. Lassomycin is a cyclopeptide that targets the Clp protease; It is selective for mycobacteria and is effective at killing both exponentially growing and dormant cells. The depsipeptide Teixobactin, produced by the new betaproteobacteria Eleftheria terrae, was discovered using the ichip in a screen of extracts from 10,000 isolates against S. aureus. Teixobactin blocks cell wall synthesis by binding to lipid II and lipid III and has excellent activity against Gram-positive pathogens [133]. Both Teixobactin and Lassomycin are being developed by Novobiotic pharmaceuticals. Efforts to cultivate new microbial taxa should continue to be a high priority, as it has been the most useful method for novel drug discoveries.

[enlarge]

Figure 2. In situ cultivation by A. diffusion chamber (adapted from [2] ). B. ichip (adapted from [3] ).

With advances in next-generation sequencing, several actinomycete genomes have been sequenced. Genomes of known antibiotic producers were found to have many other biosynthetic clusters of unknown functions. Many actinomycetes, particularly Streptomyces species, have 20-30 biosynthetic clusters [71, 134]. They are not routinely expressed under standard cultivation conditions, and there are several ongoing efforts to express these silent pathways and examine their products. Many of the following methods can also be used to increase antibiotic yields.

The secondary metabolome of many species can be perturbed by the simple alteration of culture conditions such as medium components, the addition of precursors, temperature, pH, and dissolved oxygen. The OSMAC (one strain, many compounds) approach was demonstrated with six strains (2 fungi, 4 Streptomyces) where modifying culture conditions resulted in 100 new compounds from 25 structural classes [135]. S. roseoporus produced daptomycin when decanoic acid was added to the culture [136]. Stresses such as ethanol and heat shock resulted in the production of jadomycin by S. venezuelae [137, 138]. Heavy metal stresses could stimulate antibiotic production in metal tolerant actinobacteria [139]. Antibiotics such as lincomycin at sub-inhibitory concentrations can stimulate silent pathways in Streptomyces species [140].

This method was discovered when scientists observed that actinorhodin production by the non-producing Streptomyces lividans was stimulated by mutation of the ribosomal S12 protein [141]. A simple way to introduce ribosomal mutations is by exposing strains to aminoglycoside antibiotics; strains that become resistant to these antibiotics harbor mutations in the ribosome. When more than 1000 actinomycetes from soil were tested, more than half of the non-producing strains could be made to produce antibiotics after they were made resistant to rifampicin or streptomycin. A new class of antibacterials named piperidamycin was discovered this way [142]. Resistance to other ribosome targeting antibiotics such as gentamicin or erythromycin could also stimulate antibiotic production [143, 144].

Knocking out genes for one secondary metabolite can direct resources for the production of other silent pathways. When rifamycin synthesis genes were deleted in the known rifamycin producer Amycolatopsis mediterranei, ten new amexanthomycins were produced [145]. Altering the expression of global regulators or introduction of a heterologous global regulator can change the secondary metabolome unpredictably. Introduction of the regulator AbsA2 into Streptomyces flavopersicus induced production of pulvomycin [146]. Many biosynthetic gene clusters include pathway-specific regulators. Such pathway-specific activators when overexpressed can increase the yield of the natural product or stimulate a previously silent cluster (e.g., production of stambomycin in Streptomyces ambofaciens [147] ). Similarly, deletion of pathway-specific repressor genes can stimulate expression of silent genes (e.g., production of angucyclinone in a Streptomyces sp [148]. Recent research efforts have made available a library of synthetic promoters for fine-tuned expression in actinomycetes [149].

Specific strains of Streptomyces coelicolor, S. lividans, S. albus, and S. avermitilis have been developed as host strains for heterologous expression of biosynthetic gene clusters. These strains have been deleted of their native biosynthetic gene clusters, which leaves resources free for the expression of the heterologous product and also allows easier detection of the product without interfering background from the native products [150]. Recently, large BAC libraries of Streptomyces rochei were heterologously expressed in S. albus in a high-throughput library transfer system. This led to the identification of known metabolites as well as novel, previously silent, natural products [151]. Another team cloned the minimal polyketide synthase gene clusters from environmental DNA and heterologously expressed them in S. albus. To express stubborn, silent gene clusters, they also overexpressed pathway specific activators and co-transformed S. albus with it. They screened the co-transformants for activity against pathogens and identified tetarimycin A with activity against MRSA [152].

Deletion or overexpression of histone methyltransferases or histone deacetylases can alter secondary metabolite production in fungi like Aspergillus by chromatin remodeling. The addition of inhibitors of histone deacetylases has been used to activate cryptic pathways in Aspergillus [153]. It was discovered that similarly, histone deacetylase inhibitors could activate cryptic pathways in Streptomyces, although the mechanism behind transcription induction is not known [154].

In nature, microbial species are very rarely present in isolation. There are several reports of growth in co-culture with another species resulting in induction of antibiotic production, often of previously silent gene clusters [153, 155]. Ueda tested pairings between 76 Streptomyces strains and found that 26 strains could stimulate antibiotic production in 10 strains [156]. At least some of these stimulations were because of sharing of the siderophore desferrioxamine, which may have provided iron for the co-cultured strain [157]. Some bacteria produce antibiotics which at sub-inhibitory concentrations can stimulate antibiotic production in other strains. Promomycin, produced by a Streptomyces strain, can stimulate antibiotic production in other Streptomyces strains [158].

Co-cultures can induce silent antibiotic genes because of competition. The antibiotic istamycin was produced by the marine Streptomyces tenjimariensis in competition with marine microbial species [159]. The fungi Fusarium tricinctum increased production of secondary metabolites and made new antibacterials only in co-culture with Bacillus subtilis [159].

There are also examples where physical contact between the two microbial species is needed for antibiotic production: pestalone production by a marine fungus in contact with a marine bacterium [160] ; orsellinic acid production in Aspergillus nidulans when it was co-cultured with Streptomyces hygroscopicus (only this particular strain out of 58 actinomycetes induced the cryptic pathway) [161] ; alchivemycin production by Streptomyces endus in co-culture with mycolic acid-containing bacteria [162].

Silent gene clusters may be induced by certain chemicals such as antibiotics at sub-inhibitory concentrations and histone deacetylase inhibitors as mentioned above. High-throughput elicitor screens typically use a chemical library to screen the organism of interest and look for altered secondary metabolite profiles [163, 164]. Craney et al used the Canadian compound collection of more than 30,000 chemicals to screen for modulation of secondary metabolomes in Streptomyces and found a class of effective elicitors that alter secondary metabolism by inhibiting fatty acid biosynthesis [165].

Genome mining can be used to identify new pathways and predict the synthesis of novel compounds. Genes synthesizing natural products are present in clusters and are easily identified [166, 167]. Many natural products were identified by mining for the biosynthetic enzymes [167] (a few examples are cilagicin for cil biosynthetic gene cluster [168], the polyketide stambomycin from Streptomyces ambofaciens [147], the non-ribosomal peptide coelichelin from S. coelicolor [169], the lantibiotic lichenicidin from Bacillus licheniformis [170] ). Biosynthetic gene clusters are very modular, and they follow an assembly line method, with the number of modules in the assembly line corresponding to the number of molecular building blocks incorporated in the product. Usually, there is a core cluster that makes the product and adjacent genes that modify the product such as glycosyltransferases and acyltransferases, genes that confer self-resistance, regulatory genes, and genes for efflux of the compound. Genome mining tools can find the accessory tailoring domains and use the information to predict structural features and physicochemical properties of the final product.

Two different approaches that used genome mining to look at cryptic pathways are the genomisotopic approach and the in vitro reconstitution approach. In the genomisotopic approach, labeled predicted precursors for the cryptic pathway of interest are fed to the growth culture, and NMR is used to identify the labeled metabolites (e.g., orfamides from P. fluorescens were discovered this way) [171]. In the second approach, the biosynthetic enzyme is purified and incubated with the predicted substrate, and the products are then examined [172]. Sugimoto Y et al mined human microbiome data, identified 13 complete biosynthetic gene clusters that potentially encode type II polyketides, examined two of the biosynthetic gene clusters through in vitro reconstitution and purified potential antibiotics metamycins and wexrubicin [173].

In recent years, several computational tools have been developed to improve genome mining [174]. The most important of these are antiSMASH (antibiotics and Secondary Metabolite Analysis SHell) [82] and PRISM (PRediction Informatics for Secondary Metabolomes) [175], both of which can predict the number of biosynthetic gene clusters in a given genome and the structures of the natural products. AntiSMASH also has a database of biosynthetic gene annotations for 6200 full and more than 18000 draft prokaryotic genomes [176]. Other useful tools that are limited to peptidic natural products are Pep2Path [177], RiPPQuest (for ribosomal peptides) [178], and NRPQuest (for non-ribosomal peptides) [179].

Here I highlight a few innovative screening methods that have been used recently. They include screens to identify inhibitors of new bacterial cellular targets and inhibitors of virulence factors. Researchers are increasingly addressing the problem of persistence in bacterial infections, and also seeing the value of screening in small organism models. With the dearth of new antibiotics, screens for combinatorial therapies that extend the life of existing antibiotics have great utility.

Target-based drug discovery screens, as mentioned above, search for inhibitors of target proteins that are essential to the pathogen. Such screens result in molecules with poor bacterial permeability or molecules that are effluxed. Empirical screening with whole cells results in molecules that can enter the cells; however empirical screening often selects for non-specific toxic compounds, cannot distinguish known from novel compounds, and does not identify the targets. Whole cell target-based discovery circumvents both problems by using whole cells in the screening method to look for inhibitors of specific pathways [180, 181]. To identify leads for antibiotics of new classes, different research groups are examining cellular pathways that are not targeted by currently available drugs. Stokes et al identified inhibitors of ribosome biogenesis that do not affect protein translocation. They screened a large chemical library for compounds that induce cold-sensitive phenotype in E. coli because mutations in ribosome biogenesis can result in cold-sensitivity. They identified the epilepsy drug lamotrigine as an inhibitor of ribosome biogenesis and proposed that this compound could be a lead for a new class of antibiotics [182].

Protein secretion is another attractive drug target. It is essential to bacterial growth and virulence, no current drugs target protein secretion, and it is distinct from eukaryotic secretion systems. Felise et al used a high-throughput phospholipase reporter assay to screen for inhibitors of the Salmonella typhimurium type 3 secretion system that is important for virulence. They also employed a secondary screen for removing non-specific inhibitors. Of the 92000 chemicals they screened, they found one that inhibited the assembly of the secretion system; the compound was a thiazolidinone [183]. Another group looked for inhibitors of the Sec pathway of protein secretion. They used a whole cell high-throughput assay with engineered E. coli where beta-galactosidase was exported outside the cell unless the Sec pathway was disrupted [184].

Fatty acid synthesis is another cellular process that is a promising drug target. Bacterial type II fatty acid synthesis is carried out by a discrete set of enzymes that is distinct from the mammalian fatty acid synthase, these genes are essential in bacteria, and homologous enzymes are easily identified across bacterial species [185]. The FabI, an essential enzyme, is a validated target for triclosan (antiseptic) and isoniazid (tuberculosis drug). Two groups used anti-sense RNA silencing screens with whole cells to identify inhibitors of FabH/FabF enzymes in S. aureus. Antisense screens reduce levels of the target genes such that greater inhibition or sensitivity than wild-type strain can be observed, thus making it easier to identify inhibitors. Natural product extracts from 83,000 strains (actinomycetes and fungi) grown in 3 different conditions were screened with the antisense screen, and they identified the fungal product phomallenic acids [186], platensimycin [187] and platencin [188] (products of Streptomyces platensis) as inhibitors of FabH and FabF. Platensimycin has strong Gram-positive activity.

Hosts such as macrophages or whole organism models such as Caenorhabditis elegans are increasingly being used to screen for antibiotics [180]. Such in situ screening has several advantages: it eliminates toxic molecules, directly demonstrates efficacy of the compound tested, does not select for compounds with poor penetration or those that get effluxed, allows discovery of different compounds such as those that require activation in vivo or those that target virulence but not growth, allows discovery of compounds that often are effective at low in vivo doses but require high doses in vitro (such compounds may have been missed in a conventional screen) [189].

C. elegans, a nematode worm, can be infected with several human pathogens and the infection can be rescued with antibiotics [189]. An automated high-throughput screening assay was used with C. elegans infected with Enterococcus fecalis to screen 37,000 synthetic and natural products and identified 28 new compounds that were not known to be antibiotics and that helped C. elegans survive the infection [190]. Six of these compounds were those that cured C. elegans but did not inhibit the growth of the microbe in vitro, suggesting that they were targeting virulence factors.

The zebrafish is another organism that is used for drug discovery and development because of its genetic tractability and ease of handling. The small size of zebrafish larvae (2 mm) allows growth in multi-well plates, and their optical transparency allows detection by fluorescence methods [191]. Zebrafish have a complex immune system similar to humans, and they have been used to examine the pathogenesis of Pseudomonas aeruginosa [192], Salmonella, Staphylococcus, Burkholderia, Shigella, Candida and M. tuberculosis [193]. A platform using zebrafish larvae infected with Mycobacterium marinum identified synergistic combinations of anti-tuberculosis drugs [191].

M. tuberculosis replicates inside human macrophages, but macrophages are also capable of restricting the pathogen growth under certain conditions. The host-pathogen interaction within the macrophage can affect the outcome of the infection. Macrophage-based screens with microscopic assays have been developed to understand the host-pathogen interactions better and to find potential lead drugs that can inhibit replication of M. tuberculosis inside macrophages [194, 195]. Macrophages were also used in a screen against inhibitors of the Bacillus anthracis toxin, and the identified compounds were also found to be active against the similar toxin from Clostridium difficile [196]. Another study identified small molecule inhibitors of C. difficile toxin B by screening in human fibroblasts coupled with a high-content imaging screen that measured changes in cell morphology induced by the toxin [197].

Silkworms are another example of a small organism model. Silkworms infected with S. aureus were used to screen natural product extracts from bacteria, and a new antibiotic lysocin E was identified that targets menaquinone in the bacterial membrane [198].

The potency and efficacy of an antibiotic may be dramatically improved by combination therapies with other chemicals such as antibiotic adjuvants, other antibiotics themselves, or other non-antibiotic drugs. Effective screens can help identify such combination therapies that help prolong the life of existing antibiotics [43, 180, 199].

Antibiotic adjuvants are non-antibiotic compounds that act synergistically with antibiotics and work by inhibiting or bypassing the resistance mechanisms. They may have little anti-bacterial activity on their own but can improve efficacy or extend the spectrum of the antibiotic. The best example is clavulanic acid which is given in combination with amoxicillin (Augmentin) and was also the first antibiotic/adjuvant pair to be approved clinically [200]. Clavulanic acid acts as a beta-lactamase inhibitor [201]. In its native host, the expression of clavulanic acid is coregulated with antibiotic cephamycin C [202]. Other Streptomyces producers of clavulanic acid also produce cephamycin C, suggesting that production of the two compounds evolved together as a way of fighting cephamycin resistance in its environment [203]. Another class of beta-lactam inhibitors, DABCO avibactams, is given in combination with cephalosporin ceftazidime (Avycaz) and was approved for clinical use in 2015 [200]. Murgocil was identified by Merck as an adjuvant with carbapenems against MRSA; murgocil bypasses resistance by inhibiting the MurG enzyme [204]. So far only beta-lactam adjuvants have been approved for clinical use.

Combining antibiotics with other antibiotics can expand the spectrum and act synergistically. In synergy, the efficacy of the combination is greater than the sum of the two individually. Well-known combinations include aminoglycoside + penicillin (e.g., gentamicin and ampicillin) and trimethoprim + sulfamethoxazole [200]. More recent examples are combinations of telavancin and colistin against Gram-negative infections [205], and doripenem with various antibiotics against Acinetobacter baumanii [206]. Many Streptomyces species produce multiple antibiotics, and examples suggest that these antibiotics evolved together to act synergistically [203]. The mechanisms of synergy can differ widely: the antibiotics could target linked biochemical pathways, or enhance each other’s uptake, or one antibiotic could suppress the efflux of the other. The number of essential genes in a bacterium is a relatively small number, but the number of synthetic lethal interactions (pairs of non-essential genes that when deleted become lethal) is much higher. This is also why the combination of antibiotics with other antibiotics or other drugs have such potential [200].

Antibiotics may be given in combinations with non-antibiotic drugs. Such drugs are already well characterized and so reduce the cost of drug discovery. The anti-diarrheal drug loperamide increased the intracellular concentration of tetracycline and was discovered in a screen for combinations with tetracycline and non-antibiotic drugs [207]. The antiplatelet drug ticlopidine potentiates the activity of cephalosporin against MRSA by inhibiting cell wall teichoic acid synthesis [208].

Since the number of possible combinations of antibiotics and possible small molecule adjuvants is staggeringly large, it is practical to screen large chemical or natural product libraries for synergistic effect with a chosen antibiotic against a chosen pathogen. Such screens have identified adjuvants that render antibiotics such as rifampicin and novobiocin active against Gram-negative bacteria (which are typically resistant) [209, 210]. Aspergillaminase A, a product of Aspergillus versicolor, was identified in a screen for adjuvants that restore resistance to meropenem. Aspergillaminase A functions as a potent inhibitor of the NDM-1 metallo-beta-lactamase that is responsible for the rise in carbapenem-resistant Enterobacteriaceae [211]. The Antibiotic Resistance Platform (ARP), described above in a previous section as a dereplication tool, also serves as a tool for adjuvant discovery [91].

Inhibitors of efflux pumps can also enhance antibiotic activity [200]. PAβN is a potent inhibitor of efflux pumps in Pseudomonas aeruginosa, and in combination with fluoroquinolones was an effective strategy to counter resistance [212]. Unfortunately, this compound and many others following it have had toxicity issues, and none have been used clinically. However, research into effective efflux pump inhibitors is ongoing [213].

Small molecules can also potentiate antibiotic action by stimulating the host immune system. A macrophage-based screen with natural product extracts was carried out to look for small molecules that enhance the killing of pathogens by macrophages, and streptazolin produced by Streptomyces sp was identified as a potent immune-enhancer. Streptazolin itself has no antibiotic action [214].

Persisters are dormant or non-growing cells that make up about 1% of a bacterial population. Since antibiotics act on functions that are needed in growing cells, they do not affect non-growing persister cells. Persister cells are thus antibiotic-tolerant, which is different from antibiotic resistance [215].

Traditional drug discovery methods have not considered persisters as a target, mainly because conventional screening methods cannot identify inhibitors of non-growing cells. Another challenge is that each bacterial species has several types of non-growing cells resulting in different sub-populations of persister cells with different antibiotic susceptibilities, and it is not clear which populations should be targeted [216]. Increasingly though, several groups are examining strategies to target persister populations.

One strategy is to target the membrane organization or membrane-associated respiratory enzymes, since disrupting membrane function can affect dormant bacteria as well as growing cells [217]. In anti-tuberculosis therapy, the drug pyrazinamide depletes membrane energy and targets non-growing cells. Bedaquiline, the new tuberculosis drug approved in 2012, also depletes microbial cellular energy by inhibiting the ATP synthase. Bedaquiline was discovered in a conventional screen against live cells but found to be also effective against dormant cells [217].

A second strategy that is also currently used in tuberculosis therapy is to use a combination of drugs that target both actively growing and non-multiplying cells [218]. Biofilm-dispersing compounds or compounds that inhibit persistence can be given in combination with antibiotics. Many membrane-disrupting agents also have activity against biofilms. Daptomycin can eradicate S. aureus biofilms, but it is not very effective against dormant or stationary phase cells – so not all membrane agents are effective against persisters [217]. A screen of protein kinase inhibitors identified palmitoyl-DL-carnitine as an inhibitor of E. coli and P. aeruginosa biofilms [219]. Another approach is to wake up the persister cells and shift them to a metabolically active state such that antibiotics can be effective against them. Quorum sensing inhibitors reduced persistence in E. coli [220] and P. aeruginosa [221] by waking up the persister cells and restoring antibiotic susceptibility. Kim et al identified a compound C10 that causes reversion of persisters to antibiotic-sensitive cells; C10 can be used as a single chemical supplement along with an antibiotic that targets growing cells [222].

There are a few recent examples of high-throughput screens to identify inhibitors of persistence in Mycobacterium tuberculosis. One study identified compounds that target ATP synthesis and maintenance under hypoxic conditions, which induce persistence in M. tuberculosis, and discovered three classes of compounds – imidazopyridines, benzimidazoles, and thiophenes – with activity against non-replicating and replicating mycobacteria [223]. Another study used carbon starvation to induce persistence and found compounds that inhibited non-replicating M. tuberculosis [224].

HT61 is the first example of an antibiotic that was developed intentionally against non-growing cells. Hu et al, screened quinolone like compounds and found two that were active against non-growing S. aureus cells. Three hundred analogs were synthesized and screened for activity, and HT61 was discovered to be more potent against non-replicating than growing cells [225]. HT61 is active against Gram-positive bacteria and works by depolarizing the cell membrane and nicking the cell wall. It is currently in phase III clinical trials (developed by Helperby).

Another proposed idea is to develop a sterilizing antibiotic, one that effectively kills all cells [215]. This would need to be a pro-drug. A pro-drug is active only after it enters the cell, perhaps by the action of a bacterial enzyme that converts it to an antibiotic. The drug shell may be exported by efflux pumps, but the active antibiotic would be free to bind to multiple targets inside the cell and kill it. There are four pro-drugs in use: three for tuberculosis therapy (isoniazid, pyrazinamide, ethionamide), and metronidazole (used for anaerobic bacteria) [215].

The aim is to search for drugs that would not kill the pathogen but would inhibit their toxins or production of other virulence factors. Such a strategy has several possible advantages: 1) if virulence factors rather than growth is inhibited, it is thought that resistance to such drugs would be slow to develop; 2) many virulence factors are secreted outside the cell, so drugs that block these factors do not need to enter the cell, and the problem of poor bacterial penetration does not arise; 3) virulence factors are absent in the host, so they provide a safe target ; 4) because they are pathogen-specific, such drugs would be expected not to affect the native gut flora; 5) because virulence factors are mostly species-specific, resistance cannot arise through horizontal gene transfer; 6) a large number of potential virulence factor targets [218, 226]. The primary challenge in targeting virulence factors are the difficulties in designing an effective screen since growth is not affected. Possible disadvantages of virulence factor therapy: 1) bacteria may persist and initiate virulence factor production after therapy has stopped; 2) may require rapid diagnostics of the specific pathogen or empiric therapy would require multiple anti-virulence drugs along with antibiotics to be effective [218, 226].

Anti-virulence campaigns need to be designed based on the specific pathogen and their specific virulence factors. The list of virulence factor targets is long: factors that promote bacterial adhesion and colonization, factors that promote biofilm formation and quorum sensing, toxins, secretion systems, and any regulators or biosynthetic enzymes corresponding to these factors [218, 226]. Two-component systems make attractive targets. Two-component systems are conserved across bacteria, they often control virulence factor production, and they are absent in mammals. The active sites of histidine kinases and response regulators are conserved, and this may allow one drug to target multiple two-component systems in one strain, which would also make it difficult for resistance to emerge. Some two-component systems also regulate antibiotic resistance, such as VanSR that regulates resistance to vancomycin. Targeting such systems helps attenuate the resistance mechanisms and make strains susceptible to the antibiotic [227].

FDA approved anti-virulence therapies - based on immunoglobulins and monoclonal antibodies - already exist for toxin-mediated diseases caused by Clostridium botulinum, Bacillus anthracis, and Clostridium difficile [226]. Monoclonal antibodies are also a source of promising therapies in preclinical stages or clinical trials for S. aureus (target - alpha-hemolysin) [228], P. aeruginosa (surface-associated virulence factors) [229], and M. tuberculosis (cell-surface antigens) [230] among others. Small molecule inhibitors against specific virulence factors have also been identified by high-throughput screens of chemical libraries or natural product extracts. Promising candidates are undergoing preclinical trials for infections of C. difficile (ebselen targets the C. difficile toxin) [231], S. aureus (triazolothiadiazole compounds against the sortase enzyme [232], savirin against the quorum sensing response regulator AgrA [233], and baulamycins against siderophore synthetase [234] ), P. aeruginosa (M64 blocks the PQS quorum sensing pathway [235], AHL mimics against the RhlR QS systems [236], and hydroxyquinoline compound inhibits the type 3 secretion system) [237], and Enterococcus fecalis (Fsr quorum sensing inhibitors [238] ).

New virulence factor targets are identified by identifying genes that are essential for in vivo infection using novel technologies such as dual RNA-seq or TraDIS. Inhibitors for the identified virulence factor can then be discovered using high-throughput screens. It is increasingly recognized that the genes essential for a bacterium are different under in vivo vs. in vitro conditions [180]. Guo et al used transposon mutant libraries and identified the enzyme PyrD in pyrimidine synthesis as a virulence factor target. They then used computer-aided screening to identify PyrD inhibitors and found one that was very effective in reducing virulence in mouse models [239].

Following the discovery of the first antibiotics, the field of antibiotic research went through a boom from 1930 to 1970 with the discovery of many of the clinically used antibiotics. This golden era was followed by a 30-year slump with a sharp drop in the number of new natural product antibiotics discovered. There was a renewed interest in synthetic antibiotics, and pharmaceutical companies dove into target-based antibiotic discovery campaigns. The unfortunate failure of these campaigns brought the attention back to natural products. It is clear that natural products continue to be a rich source of antibiotic candidates, and with new advances in microbial cultivation and screening techniques and increasing knowledge on the expression of cryptic pathways, there is a vast potential to be mined. Medicinal chemistry continues to be important in developing natural product derivatives that enable antibiotics to be one step ahead of bacterial resistance and in devising quicker and cheaper routes to synthesize antibiotics. Antibiotic discovery needs high tech platforms, and many promising platforms are already in development. The different non-traditional approaches described in this article show that there are many routes to new antibiotic discoveries and many routes to combinatorial therapies that can prolong the life of existing antibiotics. The future is likely also to see new antibiotics created by combinatorial biosynthesis where the biosynthetic gene clusters are manipulated to make "unnatural natural products" [240-243] or through artificial intelligence [244]. In addition, phage therapy is also having renewed interest in the fight against drug-resistant bacterial infections.