The antibiotic resistance crisis and the development of new antibiotics

Abstract

The Global Burden of Disease report of 2019 estimated 14 million infection‐related deaths, making it the second leading cause of death after ischaemic heart disease. Bacterial pathogens accounted for 7.7 million deaths and deaths attributable to bacterial antibiotic resistance amounted to 1.3 million, describing a clear demand for novel antibiotics. Antibiotic development had its golden age in 1930–1960. Following failures in the screening of chemical libraries for novel antibiotics at the beginning of this century, the high cost of launching new antibiotics (estimated at US$ 1.4 billion per registered drug) and difficulties in achieving a return of investment for novel antibiotics, pharmaceutical industry has mostly left the field. The current Lilliput review analyses the question whether scientific or economic hurdles prevented the registration of new antibiotics. Scientifically, substantial progress has been achieved over recent years to define the chemical properties needed to overcome the permeation barrier in Gram‐negative pathogens; in extending the chemical space of antibiotic candidates by full modular synthesis of suitable molecules; by extending bioprospecting to previously ‘unculturable’ bacteria or unusual bacteria; by attacking bacterial targets on the outer bacterial membrane; and by looking for support from structural biology, genomics, molecular genetics, phylogenetic analyses and deep machine learning approaches. However, these research activities were mostly conducted by academic researchers and biotech companies with limited financial resources. It thus seems that the development of new antibiotics, frequently described as the drying of the pipeline, is less limited by lack of scientific insight than by lack of the mobilization of the monetary resources needed to bring these discoveries to the market despite recent financial push and pull efforts of the public sector.

The current Lilliput review analyses the question whether scientific or economic hurdles prevented the registration of new antibiotics. Scientifically, substantial progress has been achieved over recent years. The lack of profitability for costly novel antibiotic development necessitates push and pull efforts of the public sector to complement what the market forces alone cannot achieve.

Numbers are the backbone for many branches of human inquiry. Without exact numbers, it is not possible to assess the extent of a problem or to prioritize interventions, resources and research activities. This applies to human activities ranging from economics to politics, from science to medicine. One of the tasks of medical epidemiology is to define the numerical extent of diseases and provides thus a rational framework for allotting public health resources to cope with health issues commensurate with their societal impact. In that context, the Global Burden of Diseases (GBD) publications provide critical information at a global scale. GBD is a comprehensive research programme initiated in 1990 by the World Bank and now funded by the Gates Foundation.

GLOBAL BURDEN OF DISEASE REPORTS

The GBD metrics

The 2019 edition of the GBD was elaborated by more than 5000 scientists and covers data from 204 countries. GBD reports are coordinated by the Institute for Health Metrics and Evaluation at the University of Washington/USA. GBD lists disease burden by incidence, prevalence and associated mortality. The researchers have developed important public health parameters to characterize the impact of diseases, such as years of life lost (YLLs), years lived with disability (YLDs) and disability‐adjusted life‐years (DALYs) for 369 diseases and injuries (GBD 2019 Diseases and Injuries Collaborators, 2020). DALYs became a central public health parameter as it expresses the number of years lost due to ill health, disability or early death with respect to the life expectancy of an individual in a given country. In DALYs, mortality and morbidity data are combined into a widely used single, common metric. Over the past 30 years of GBD assessments, global health has significantly improved when measured by age‐standardized DALY rates.

Disease ranking and the role of infections

When all age groups are considered together, the 5 top‐ranking diseases as assessed by age‐standardized DALYs have not changed over the last 30 years. In both 1990 and 2019, these were, in decreasing order, neonatal disorders (they rank first because they affect the youngest which thus has a disproportional effect on the GBD metrics) followed by ischaemic heart disease, stroke, lower respiratory infections and diarrhoeal diseases (in 1990 respiratory and diarrhoeal diseases still ranked at positions 2 and 3). In comparison, AIDS, tuberculosis and malaria ranked at positions 11, 12 and 14, in 2019, respectively. Measles, meningitis and whooping cough which still came at positions 9, 16 and 25 in 1990 fell to positions 71, 40 and 55, respectively, in 2019 due not only to vaccination efforts, but also to the decrease of protein‐energy malnutrition, which fell in GBD from position 12 to 41 over this time period. In contrast, diabetes raised in ranks from position 20 to 8 over this 30‐year interval.

When different age groups were analysed, infectious diseases (ID) dominated the 15 top‐ranking DALYs in 0‐ to 9‐year‐old children, both in 1990 and 2019. In contrast, none of the top 7 DALYs in the 10–24 years age group were ID, but road injuries, self‐harm, depressive and anxiety disorders and interpersonal violence. In the 25‐ to 49‐year group, AIDS, followed directly by road injury and tuberculosis, was at position 10. In subjects older than 50 years, ischaemic heart disease, stroke and diabetes unsurprisingly took leading positions to which dementia was added in subjects older than 75 years. Lower respiratory infections, diarrhoeal diseases and tuberculosis were the major ID in older subjects.

The GBD 2019 report on infection mortality

In a follow‐up study, the GBD 2019 consortium analysed 343 million individual records for deaths in which infection played a role, specifically for deaths attributable to a given infectious syndrome or for deaths attributable to 33 bacterial pathogens across 204 countries (GBD 2019 Diseases and Injuries Collaborators, 2022). The researchers estimated 14 million infection‐related deaths globally, making it the second leading cause of death after ischaemic heart disease. The 33 selected bacterial pathogens accounted for 7.7 million ID deaths in 2019, yielding an all‐age mortality rate of 100 deaths per 100,000 population (representing 14% of all deaths). Five pathogens, namely Staphylococcus aureus, Escherichia coli, Streptococcus pneumoniae, Klebsiella pneumoniae and Pseudomonas aeruginosa, were responsible for 55% of deaths. S. aureus was associated with 1 million deaths, while the other four bacterial pathogens were each associated with about 500,000 deaths in 2019. With respect to infection syndromes, lower respiratory tract infections (LRTI) accounted for 4 million deaths, bloodstream infections for 3 million deaths and peritoneal and intra‐abdominal infections were responsible for 1.3 million deaths. When YLL was estimated, S. pneumoniae replaced S. aureus as leading bacterial pathogen, reflecting its prominence in young children. In different age groups, distinct bacterial pathogens represented the leading cause of infection deaths: K. pneumoniae was the pathogen associated with the most neonatal deaths (124,000, mostly sepsis); S. pneumoniae dominated deaths in post‐neonatal children up to age of 4 years (225,000 deaths, mostly lower respiratory tract infection); Salmonella enterica serovar Typhi was the most fatal infection in children aged 5–14 years (49,000 deaths); while S. aureus was the pathogen associated with the highest number of infection deaths in individuals older than 15 years (940,000 deaths, mostly bloodstream infections). E. coli was the leading cause of fatal peritoneal and intra‐abdominal infections. Geographical differences were detected with respect to death rates: the 33 investigated bacterial pathogens caused 230 deaths per 100,000 population in sub‐Saharan Africa compared to 52 deaths in high‐income regions. Geographical differences were also observed with respect to pathogens. E. coli was responsible for the highest age‐standardized mortality rate in Spain, Central and Eastern Europe, India and Indonesia; S. pneumoniae dominated deaths in the Sahel zone of Africa while S. aureus dominated in the rest of the world. Bloodstream infections were associated with S. aureus in 25% of cases in high‐income countries while only 5% bloodstream infections from sub‐Saharan Africa were associated with S. aureus; dominant isolates here were K. pneumoniae and Neisseria meningitidis. The authors of the study noted that research funding for infectious diseases does not reflect their global disease burden: HIV research received US$42 billion in funding compared with only $1.4 billion for research on Staphylococcus and $800 million for E. coli in the two decades preceding GBD 2019, despite the fact that the disease metrics for these bacterial pathogens ranked higher than that caused by HIV. The researchers also concluded that enormous GBD ameliorations can still be achieved with better hygiene, more and new vaccines and appropriate antibiotics.

THE ANTIBIOTIC RESISTANCE CRISIS IN NUMBERS

GBD data

In a next step, the GBD 2019 consortium estimated the number of deaths associated with or attributed to antibiotic resistance analysing 88 bacterial pathogen–drug combinations (Antimicrobial Resistance Collaborators, 2022). The researchers used two counterfactual scenarios (i.e. by considering in a thought experiment something that is contrary to what actually happened). With that approach, they evaluated what number of deaths would have been prevented if antibiotic‐resistant bacteria were all replaced by antibiotic‐sensitive bacteria (‘antibiotic‐attributed deaths’) or were replaced by no bacteria of the given pathogen species (‘antibiotic‐associated deaths’). For 2019, the researchers estimated 1.3 million antibiotic‐attributed deaths globally. Death attributable to bacterial antimicrobial resistance (AMR) would, by this calculation, be the twelfth leading cause of death, placing it ahead of both HIV and malaria. AMR death burden was highest in sub‐Saharan Africa with 27 deaths per 100,000 population and lowest in Australasia with 7 deaths per 100,000 people. The two infectious syndromes that dominated the AMR attributed deaths were LRTI and bloodstream infections, followed by death caused by AMR in tuberculosis, intra‐abdominal and urinary infections. With respect to pathogens involved, global deaths attributable to AMR were led by E. coli, K. pneumoniae and S. aureus, followed by Acinetobacter baumannii, S. pneumoniae and then Mycobacterium tuberculosis and Pseudomonas aeruginosa. In high‐income regions, half of the AMR attributed deaths were linked to S. aureus and E. coli while in sub‐Saharan Africa, S. pneumoniae and K. pneumoniae were the most frequent lethal AMR pathogens. Globally, methicillin‐resistant S. aureus (MRSA) accounted for more than 100,000 deaths, while multidrug‐resistant (MDR) tuberculosis, third‐generation (3G) cephalosporin‐resistant E. coli, carbapenem‐resistant A. baumannii, fluoroquinolone‐resistant E. coli, carbapenem‐resistant K. pneumoniae and 3G cephalosporin‐resistant K. pneumoniae each accounted for more than 50,000 deaths. Each of these antibiotic‐resistant bacterial pathogens showed a distinct geographical distribution. The study authors recommended five intervention strategies to cope with the problem of AMR attributed deaths. Proposed interventions targeted infection prevention by water, sanitation and hygiene quality control programmes, vaccination for S. pneumoniae, reduction of antibiotic use in farming, antibiotic stewardship in human therapy and investments in the development of new antibiotics.

The European perspective

The evaluation of the European Antimicrobial Resistance Surveillance Network (EARS‐Net) calculated 670,000 infections with antibiotic‐resistant bacteria in 2015. These infections resulted in 33,000 AMR attributable deaths; GBD 2019 estimated 23,000 deaths for this area. For the EU, these calculations translate into 131 AMR infections per 100,000 population (64% of which occurred in healthcare settings) and an attributable mortality of 6 per 100,000. With respect to both DALY and deaths, the leading AMR pathogens were third‐generation (3G) cephalosporin‐resistant E. coli, MRSA, carbapenem‐resistant P. aeruginosa and 3G cephalosporin‐resistant K. pneumoniae. The burden of disease expressed as DALYs was the highest in infants younger than 1 year and older citizens. The burden of AMR infections was much higher in southern and eastern than in central and northern Europe. The researchers noted a more than twofold increase in AMR attributable deaths between 2007 and 2015 (Cassini et al., 2019).

The US perspective

A 2019 report from the Centers for Disease Control and Prevention (CDC) estimated that 2.9 million persons in the US are infected each year with antibiotic‐resistant pathogens, of whom 36,000 die. Compared with a 2013 report from CDC, 18% fewer deaths from antibiotic resistance were observed in 2019; deaths due to antibiotic resistance in hospitals decreased by 28%. The CDC attributed this to a large panel of prevention measures: attention to medical device‐mediated infections; early AMR detection in the hospital and prevention of microbial spread between facilities; appropriate antibiotic use and infection control in long‐term facilities. The 2019 CDC report listed five urgent AMR threats, namely carbapenem‐resistant Acinetobacter (700 deaths in 2017), Candida auris (a yeast; 323 cases in 2018), Clostridioides difficile (12,800 deaths in 2017; provoked by antibiotic treatment), carbapenem‐resistant Enterobacteriaceae (1100 deaths in 2017) and drug‐resistant Neisseria gonorrhoeae (few deaths) (CDC 2013 report: ar‐threats‐2013‐508.pdf (cdc.gov)) CDC 2019 report: Antibiotic Resistance Threats in the United States, 2019 (cdc.gov). As can be seen from the mortality data, the urgent threats defined by CDC do not reflect the bulk of the AMR infections or mortality defined by GBD 2019, but their potential harm they might cause in the future.

The quantitative aspects of AMR infections in the US are given in a publication reporting on 42 million hospitalizations in 890 US hospitals between 2012 and 2017, representing 20% of all hospitalizations in the country. In 2017, the researchers detected 622,000 infections with AMR bacterial pathogens, only 17% of the infections had their onset in the hospital. MRSA and extended‐spectrum beta‐lactamase (ESBL) Enterobacteriaceae infections were the most common and accounted for 52% and 32%, respectively, of all cases in 2017. MRSA infections occurred in 94 cases per 10,000 hospitalizations (21% decrease from 2012) and ESBL‐producing Enterobacteriaceae accounted for 57 cases per 10,000 hospitalizations (53% increase from 2012). Vancomycin‐resistant Enterococcus (VRE) represented 15 cases, multidrug‐resistant P. aeruginosa accounted for 9 cases and carbapenem‐resistant Acinetobacter represented 3 cases per 10,000 hospitalizations (Jernigan et al., 2020).

Not enough with the sad numbers from GBD 2019, a frequently quoted projection for 2050 speaks of 10 million deaths per year attributed to antibiotic‐resistant bacterial infections (O'Neill, 2014). Should this prediction become a reality, AMR attributed deaths would even dwarf the death toll of a pandemic such as COVID‐19. While this projection for 2050 depends on several assumptions (de Kraker et al., 2016), the spectre of a such a pandemic‐sized death toll already calls for an assessment of the state of novel antibiotic development.

ANTIBIOTIC DEVELOPMENT: FROM GLORY TO GLOOM

Waksman platform

When you look at a table of commonly prescribed antibiotics in a standard medical textbook, the list of antibacterial agents seems long and impressive (Table 1). In fact, antibiotics represent triumphs of modern medicine, which have not only saved millions of lives from bacterial infections, but have also provided a protective shield to allow for surgical interventions not possible before antibiotic introduction into medical practice. The golden age of antibiotic development extended from the 1930s into the late 1950s where many antibiotics were discovered and introduced into medicine. In the Waksman platform (named after its pioneer), soil actinomycetes were screened for their ability to produce zones of growth inhibition on a Petri dish overlaid with a test pathogen. These natural antimicrobial compounds yielded the beta‐lactams, aminoglycosides, chloramphenicol, macrolides, tetracyclines and glycopeptides. Since the 1960s, it became clear that the Waksman platform increasingly failed to yield new natural antibiotics. Scientists concluded that actinomycetes as source material for novel antibiotics were overmined.

TABLE 1.

Common antibiotics in clinical use, their bacterial targets and resistance mechanisms.

Antibiotics (ab)	Target	Antibiotic resistance
Beta‐lactam antibiotics
Penicillins: aminobenzylpenicillins, acylaminopenicillins	Inhibition of transpeptidase reaction in peptidoglycan synthesis	Beta‐lactamase, mutated penicillin‐binding proteins (PBP)
Cephalosporins	Inhibition of transpeptidase reaction in peptidoglycan synthesis	Beta‐lactamase, mutated PBP
Carbapenems	Inhibition of peptidoglycan synthesis via interaction with PBP	Class A, D and metallo‐beta‐lactamases; loss of porin (imipenem), efflux pump (meropenem)
Glycopeptides: vancomycin, teicoplanin	Inhibition of transglycosylation reaction in peptidoglycan synthesis	Overproduction of peptidoglycans, replacement of D‐Ala‐D‐Ala by D‐Ala‐D‐Lac
Aminoglycosides: gentamycin, tobramycin, amikacin, streptomycin	Electrostatic binding to cell exterior leads to holes; ribosomal misreading	Reduced uptake, enzymatic inactivation, mutated ribosomal binding site, efflux pump
Tetracyclines: doxycycline, tigecycline	30S ribosome binding inhibiting tRNA association	Induction of efflux pumps, Ribosomal protection proteins detach doxycycline from ribosome, Mutation of 16S rRNA prevents ab binding
Lincosamine: clindamycin	Inhibition of ribosomal peptidyltransferase	Mutation of ribosomal binding site
Macrolides: erythromycin, azithromycin, clarithromycin	Inhibition of 50S ribosomal peptidyltransferase	Mutation of 23S rRNA prevents ab binding, Efflux pumps, Enzymatic inactivation (only for erythromycin)
Folic acid antagonists: trimethoprim sulfamethoxazole (TMS‐SMX), dapsone	TMP inhibits dihydrofolate reductase, SMX and dapson, inhibits dihydropteroate synthase, two successive steps in purine nucleotide synthesis	Mutation of targeted enzymes (?)
Quinolones: ciprofloxacin, moxifloxacin	Inhibition of topoisomerases (cipro); inhibition DNA gyrase, topoisomerase IV (moxi)	Decreased permeation, increased efflux (cipro) Mutation in targeted enzymes (moxi)
Antituberculous drugs
Isoniazid (only M. tuberculosis)	Inhibition of mycolic acid synthesis	Mutation in inhA, kasA involved in mycolic acid synthesis
Rifampicin	Inhibitor of RNA polymerase	Mutation in rpoB
Ethambutol (only mycobacteria)	Inhibits synthesis of arabinogalactan of mycobacteria	Mutation in arabinotransferase
Pyrazinamide (only M. tuberculosis)	Analog of nicotinamide	Mutation in pyrazinamidase
Diverse ab
Metronidazole	Inhibits nucleic acid synthesis creating a cytotoxic radical	Detoxification by a reductase, decreased uptake, decreased interaction with DNA
Fosfomycin	Inhibition of phosphoenol‐pyruvyltransferase in peptidoglycan synthesis	Decreased uptake, enzymatic inactivation, overproduction of pyruvyltransferase
Fusidic acid	Inhibits tRNA and elongation factor release from ribosome	Decreased uptake, reduced affinity for ribosome
Polymyxin and colistin	Cationic detergent for bacterial membrane	?
Mupirocin	Inactivation of isoleucyl‐tRNA synthase	Mutation of binding site or replacement by alternative synthetase
Oxazolidinone	Inhibits protein synthesis initiation by 70S ribosome	Mutation in 23S rRNA
Daptomycin	Ionophore	Reduced binding and depolarization
Chloramphenicol	Inhibits peptidyltransferase via binding to 23S rRNA	Decreased permeation, enzymatic inactivation, efflux pump

Semi‐synthetic chemistry

Therefore, pharmaceutical industry turned to synthetic chemistry approaches. Effective antituberculous drugs (isoniazid, pyrazinamide and ethambutol) were identified by this approach. The synthetic antimicrobials, metronidazole and oxazolidinone were added to the list of antibiotics. The DNA gyrase and topoisomerase inhibitor nalidixic acid led to a new class of synthetic antibiotics, the quinolones (Table 1). Synthetic chemistry also provided analogues that transformed narrow‐spectrum into broad‐spectrum antibiotics (penicillin → ampicillin; erythromycin → azithromycin) or yielded analogues active against antibiotic‐resistant pathogens (Lewis, 2020). However, concerns remained: The Waksman platform and synthetic antibiotics approaches have only identified a very limited number of bacterial targets for antibiotics, and resistance has evolved against most of these antibiotics in bacterial pathogens (Table 1).

Glaxo's experience

New efforts to develop novel antibiotics were stimulated by the whole‐genome sequencing (WGS) of bacterial pathogens starting in 1995. Antibiotic researchers from Glaxo Smith Kline summarized their experience with antibiotic development conducted between 1995 and 2001. These industrial researchers analysed the genomes of representative Gram‐negative (H. influenzae and Moraxella catarrhalis) and Gram‐positive (S. pneumoniae, S. aureus and Enterococcus faecalis) pathogens for conserved genes. Next, they determined by gene replacement mutagenesis that the genes were essential for bacterial growth and that growth varied with gene expression levels when using inducible promoters. Overall, 350 shared S. pneumoniae, S. aureus and H. influenzae candidate target genes were identified, 127 were essential for in vitro growth and 64 showed attenuated growth in an animal model. Finally, high‐throughput screening (HTS) campaigns were conducted with 67 bacterial target proteins against up to half a million chemical compounds. The 67 bacterial targets comprised proteins from fatty acid synthesis, DNA replication, protein modification, two‐component signal transduction, protein elongation, protein termination, RNA elongation, cell division, glycolytic pathway, amino acid synthesis, protein secretion, cell wall synthesis and tRNA synthetases. Whole‐cell antibacterial tests were also conducted against S. aureus and E. coli. A mere 16 screening projects gave rise to hits, but most hits turned out to be non‐specifically toxic to both mammalian and bacterial cells, usually as a result of indiscriminate cell membrane disruption. Five lead compounds were identified whose antibacterial spectra were, however, very limited. The level of success was thus unsustainably low in relation to the large time and cost invested in comparison with screen outcomes in other therapeutical areas. The costs per single HTS campaign were around US$1 million. Also, the outcome of whole‐cell antibacterial screening was disappointing: the E. coli screen did not yield a single exploitable hit, while S. aureus screens yielded thousands of hits which were finally attributed to non‐specific membrane interactions as shown by whole‐cell depolarization and erythrocyte lysis (Payne et al., 2007). In 34 other companies, 125 screens against 60 different bacterial target proteins were run, which yielded no antibiotic candidate for development (Chan et al., 2004). In 2002, GSK overhauled its antibacterial research strategy: they concentrated on novel chemical structures instead of novel targets and modified the pleuromutilin core structure leading to a clinical development. The GSK researchers also suggested to screen extremophiles which might elaborate novel forms of chemical defences not seen in actinobacteria. However, in view of the high attrition rates in industrial pipelines, they estimated that it will take 10–15 years for developing a novel mechanism antibiotic for the treatment of a relevant Gram‐negative hospital infection (Payne et al., 2007).

AstraZeneca's experience

In a separate publication, researchers from AstraZeneca summarized their antibiotic development efforts conducted between 2001 and 2010. The company conducted 65 HTS against their compound collection and identified 57 hits, defined by a reproducible dose response. Progression of these projects into lead optimization proved extremely difficult. Major barriers to whole cell activity were efflux pumps, defensive enzymes, polysaccharide cell capsules and the highly impenetrable outer membrane of Gram‐negative bacteria. The design of molecules with improved permeation of the outer membrane was considered crucial. Therefore, these researchers used transposon insertional mutations followed by sequencing to identify porins that transported the antibiotic carbapenem. Then they determined critical amino acids for molecular transport from crystal structures of these porins. As P. aeruginosa has a set of 30 porins controlling the entry of amino acids, sugars and phosphate as nutrients, the scientists proposed a Trojan horse strategy where antibiotics were conjugated to nutrients to get transported through the porins. They conclude that a basic understanding of the complex interplay of mechanisms driving compound uptake across bacterial membranes is an absolute prerequisite for the rational design of novel antibiotics (Tommasi et al., 2015).

The outer membrane barrier

The two membrane barriers in Gram‐negative bacteria make drug penetration a complex task. The outer membrane contains lipopolysaccharides inserted into the outer face of the membrane, which confer a high negative charge to the cell and prevents the penetration by hydrophobic molecules. The inner face of the outer membrane contains glycerophospholipids and confers a marked asymmetric character to the two faces of this bacterial membrane. The inner membrane of Gram‐negative bacteria in contrast consists of a conventional hydrophobic lipid bilayer that restricts the entry of hydrophilic molecules. These two membranes, with two clearly distinct physicochemical constraints, present a formidable challenge for medicinal chemists to design antibiotics that satisfy penetration requirements for both membranes, which explains the low success rate for developing novel antibiotics against Gram‐negative pathogens (Lewis, 2020).

Economic barriers

To these scientific challenges, come economic and strategical hurdles for antibiotic development. To cope with the problem of antibiotic resistance, WHO has released the ‘AWaRe’ classification database. ‘A’ stands for ‘access’ and describes antibiotics recommended as first‐ or second‐line therapies for specific infectious diseases. ‘Wa’ stands for ‘watch’, a stewardship programme for antibiotics sensitive to induce resistance development. ‘Re’ stands for ‘reserve’ and describes last‐line antibiotics only to be used in life‐threatening infections. While this approach has prolonged the longevity of some last‐line antibiotics, it also means that last‐line antibiotics are kept under key in hospital pharmacies, which seriously diminishes the economical return for the pharmaceutical industry developing the most valuable antibiotics. As resistance against some reserve antibiotics has also developed, the profit incentive for developing new antibiotics has critically decreased. As a consequence, only 43 phase 1–3 clinical trials were registered for antibiotics at the end of 2020 compared to 1300 trials for anticancer agents. To correct this situation in societies confronting increasing antibiotic resistance problems, push and pull mechanisms have been proposed to motivate industries to resume antibiotic development programmes. The push part comprises, for example, the GARDP (Global Antibiotic Research and Development Partnership) and the REPAIR (Replenishing and Enabling the Pipeline for Anti‐Infective Resistance) programmes that fund late‐stage antibiotic discovery and clinical trials with public money. The pull part comprises programmes such as one by the UK government where companies are paid for the development of antibiotics that meet specific clinical needs, providing companies a reasonable return for their efforts independent of later prescription sales. As clinical trials are major cost factors for drug development, it was proposed to replace cost‐prohibitive large randomized, double‐blind clinical trials by adaptive licensing strategies as for orphan diseases where a restricted approval is given for specific infections with MDR pathogens (Cook & Wright, 2022). These are difficult regulatory questions, but they can be addressed politically. The scientific obstacles to develop new antibiotics cannot be solved politically, but need new approaches and insights. In the following sections, some promising new approaches to develop new antibiotics are presented from the research literature of the last years.

EXPLORING THE ‘UNCULTURABLE’ BACTERIA

Teixobactin

Some researchers have argued that the reservoir of natural antimicrobial compounds is still largely untapped as only the about 1% of bacteria that can be cultured in laboratory media were used as source material. It was estimated that about 99% of all bacteria cannot be cultured or only with great difficulties. To circumvent this problem, a group of scientists from US and Germany used a multichannel device into which soil samples were diluted such that, statistically, only one bacterial cell is delivered to a given channel. The device is covered with two semi‐permeable membranes and placed back in the soil where it is supplied with its unknown nutrients and co‐factors. An encouraging 50% of channels started with growth. Extracts from 10,000 such isolates were screened for antimicrobial activity on plates overlaid with S. aureus. One extract from a new β‐proteobacterium produced an inhibitor, teixobactin (Figure S1A), which was characterized as a depsipeptide (a compound containing peptide and ester bonds) encoded by two large non‐ribosomal peptide synthetase genes. Teixobactin is bacteriocidal for numerous Gram‐positive pathogens and has a spectacular minimal inhibitory concentration (MIC) of 5 ng/mL against C. difficile. Teixobactin also inhibits M. tuberculosis. With respect to its mechanism of action (MoA), teixobactin inhibits the peptidoglycan synthesis, but also interacts with the bacterial membrane. The fact that no resistance developed against teixobactin suggested a non‐protein target. Lipid II, a precursor of peptidoglycan, was tentatively identified as its molecular target. Standard pharmacological tests, such as maintenance of its activity in serum, chemical stability, lack of toxicity, were favourable. In vivo efficacy of teixobactin was shown in two infection models against a lethal dose of MRSA demonstrating a bacterial titre reduction superior to that of vancomycin. Teixobactin was also active in a lung infection model with S. pneumoniae (Ling et al., 2015). Subsequent structural data revealed the mode of action of teixobactin: it binds to the conserved pyrophosphate sugar moiety of lipid II. Teixobactin sequesters lipid II by forming small β‐sheets upon binding of lipid II, which then elongate into long fibrils. This process not only sequesters lipid II and inhibits peptidoglycan synthesis but also damages the bacterial cell membrane. As lipid II is not found in eukaryotic cell membranes, the surprisingly low toxicity of this membrane disrupting compound for mammalian cells is easily understood. The bacterial membrane visibly gets thinner at the site of complex building which leads to the lysis of the bacterial cell (Shukla et al., 2022).

Clovibactin

Subsequently, these researchers argued that some of the ‘unculturable’ soil bacteria might only need an extended incubation time to induce growth in vitro. Indeed, they isolated colonies from soil after a 12‐week incubation period, which were then tested for antimicrobial activity against S. aureus. One colony showed strong inhibition of S. aureus, the colony was characterized as Eleftheria terrae, a β‐proteobacterium closely related to that which produced teixobactin. Genome sequencing of E. terrae identified 19 biosynthetic gene clusters (BGCs); 14 of them encoding non‐ribosomal peptide synthases. Isolation of the antibiotic was only achieved when the genes for a major, already known antibiotic was deleted. From that deletion mutant, the researchers isolated a minor compound, clovibactin (Figure S1B), related to, but distinct from, teixobactin. Clovibactin was active against Gram‐positive bacteria (including MRSA, vancomycin intermediate S. aureus (VISA) and VRE). In a mouse infection model, it was of comparable efficacy as vancomycin. Clovibactin showed low toxicity for mammalian cells and no resistant bacteria could be isolated. Using a mixture of cell biology, chemical and structural biology approaches, the researchers demonstrated that 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. Clovibactin binds the pyrophosphate moiety of lipid II. As they did not observe resistant cells, they calculated the frequency of resistance to <10⁻¹⁰ where 10⁻⁸ is considered enough to provide an extended lifetime for an antibiotic in the clinic. Chemically surprising, the hydrophobic residues of clovibactin embrace the charged pyrophosphate group like a glove (Shukla et al., 2023).

PHYSICOCHEMICAL APPROACHES TO THE ANTIBIOTIC PERMEATION PROBLEM

Outer membrane permeation rules for E. coli

Different approaches were taken to cope with the permeability barrier of the two membranes from Gram‐negative bacteria. Researchers from Illinois/USA started drug accumulation study in E. coli with 100 different chemical compounds, most of them natural products or their derivatives. Antibiotics active against Gram‐negative bacteria showed a molecular weight of <600 and were very polar, as measured by their distribution coefficient in two immiscible solvents (octan‐1‐ol/water). These two parameters were however not sufficient to convert Gram‐positive‐only antibiotics into broad‐spectrum antibiotics also active against Gram‐negative bacteria with two membranes. The researchers described three new criteria for outer membrane permeability. Positively charged compounds were the most likely to accumulate in E. coli; best were primary amines. The number of rotatable bonds should not exceed five and the globularity of the molecule should neither be flat nor bulky. The compound should be amphiphilic that is, containing hydrophilic and hydrophobic portions. Compounds with these chemical properties represent 0.1% of chemical collections and are thus exceedingly rare, explaining the low‐success rate of pharmaceutical screening programmes for new antibiotics. They confirmed the validity of their rules for a substance permeating E. coli by converting a Gram‐positive‐only antibiotic into an antibiotic active against several Gram‐negative ESKAPE organisms by modifying the compound according to these rules. A notable exception was lack of activity against P. aeruginosa (Richter et al., 2017).

Extending the rules to P. aeruginosa

These researchers in collaboration with Roche Pharma from Switzerland then explored the situation in P. aeruginosa (Pa). The biological difference with E. coli is substantial: Pa possess 40 channels for specific nutrient transport with a molecular size limit of 200 daltons and in addition efficient efflux pumps. In E. coli, the investigated drugs were transported by the general porin OmpF while drugs accumulated in Pa in a porin‐independent way as demonstrated by a Pa deletion mutant lacking all 40 porins. Drugs entered in Pa by a self‐promoted uptake. Not surprisingly, computational analysis revealed different rules than those observed in E. coli by identifying formal charge, positive polar surface area and hydrogen bond donor surface area as conditions for uptake in Pa. Based on these new rules, researchers tried to develop an anti‐Pa antibiotic modifying fusidic acid, a Gram‐positive only antibiotic. They created a prodrug with four ionizable nitrogens, which allowed the uptake of this drug (Figure S1C). The prodrug was split inside the cell to liberate the carboxyl group needed for engaging with its cellular target, elongation factor G, which leads to cell killing. These physicochemical discoveries will certainly help to extend the spectrum of antibiotics against ESKAPE organisms (Geddes et al., 2023).

Remodelling of inhibitors

In a collaboration of several pharmaceutical companies, medicinal chemists modulated the activity of diazabicyclooctane inhibitors of penicillin‐binding proteins (PBP) from Gram‐negative bacteria. Shifting the inhibitory activity from PBP2 to PBP1 and PBP3 led to a compound that inhibited several ESKAPE organisms. However, whole cell inhibitory activity against clinical isolates of Pa was low. This low activity could not be attributed to enzymatic digestion or the action of efflux pumps, but was a consequence of low permeation of the outer membrane of Pa. The inclusion of hydrogen bond donors and introduction of appropriate dipole moments into the inhibitor led to a compound ETX0462 showing potent in vivo inhibitory effects not only in two Pa mouse infection models, but also activity against several bioterrorist agents. The key message of this report was that substitutions that enhanced target potency of a compound concomitantly diminished permeation. It needed the optimization of permeation and a re‐engineering of target inhibition to obtain a promising new class of antibiotics that can be used as single agent against multidrug resistant Gram‐negative bacteria (Durand‐Reville et al., 2021).

EFFLUX PUMP INHIBITORS

Antibiotic adjuvants are an alternative approach to fight antibiotic resistant bacteria. The β‐lactamase inhibitors (e.g. clavulanic acid) are a prime example for a clinically successful inhibitor of a resistance mechanism. Antibiotic resistance is also frequently mediated by efflux pumps in the cytoplasmic bacterial membrane that export antibiotics and otherwise toxic chemical compounds out of the bacterial cell. An attractive approach to antibiotic resistance would be efflux pump inhibitors (EPI) (Douglas et al., 2023). Efflux pumps are energy dependent, hence agents that dissipate the electron transport and reduce the proton motive force across the membrane are potential EPIs. However, these compounds are cytotoxic to mammalian cells precluding their clinical use (Ding et al., 2023). Several plant alkaloids (reserpine), flavonoids, terpenoids and polyphenols are efficient EPI, but they are also toxic to mammalian cells such that none of these EPIs progressed beyond preclinical research (Douglas et al., 2023). To overcome the toxicity problem, Zimmermann et al. (2019) screened 1200 approved drugs as inhibitors of the NorA efflux pump from S. aureus. They identified nilotinib, a tyrosine kinase inhibitor, which showed synergistic activity with ciprofloxacin showing a 16‐fold reduction of ciprofloxacin MIC. A combination of nilotinib with ciprofloxacin reduced S. aureus biofilm formation (Zimmermann et al., 2019). Other researchers screened chemically synthesized indole derivatives for inhibitory activity of NorA efflux pump from S. aureus. They found that one indole compound in combination with ciprofloxacin achieved the eradication of S. aureus biofilm and showed in vivo efficacy against S. aureus in the neutropenic mouse thigh infection model (Chandal et al., 2023).

SYNTHETIC ANTIBIOTICS

From clindamycin to cresomycin based on target structure data

Lincomycin was isolated from a soil streptomycete and FDA approved in 1964 for the treatment of streptococcal and staphylococcal infections. By a one‐step chemical modification of this natural product (i.e. by semisynthesis, here: replacement of an OH group by Cl), clindamycin was developed, which had a broad‐activity spectrum and became a widely used antibiotic since its FDA approval in 1970. Clindamycin arrests bacterial protein synthesis by binding to the peptidyl transferase centre (PTC) of the bacterial ribosome. However, resistance to clindamycin is now worldwide. Three mechanisms preclude antibiotic binding to the ribosome: two methylases modify a nucleotide in the 23S rRNA close to the PTC and a nucleotidyltransferases modifies the antibiotic preventing its ribosome binding. Clindamycin (Figure S2A) can be imagined as composed of a “northern” (upper part in the chemical formula) amino‐octose moiety and a “southern” moiety containing a heterocyclic ring. Academic researchers from the US designed a chemical synthesis by replacing the “southern” part of the clindamycin structure by a two‐ring system. From this scaffold, they developed iboxamycin (Figure S2B), which displayed exceptional potency against ESKAPE pathogens including strains expressing ribosomal RNA methyltransferase enzymes that confer resistance to all clinically relevant antibiotics targeting the large ribosomal subunit. Iboxamycin is orally available, safe, and has shown efficacy against both Gram‐positive and Gram‐negative pathogens in mouse infection models (Mitcheltree et al., 2021). Subsequently, these researchers modified the “northern” part of iboxamycin by creating a macrobicyclic antibiotic. This compound, cresomycin (Figure S2C), showed superior in vitro activity against ESKAPE organisms. Cresomycin was safe in animals and protected mice in two mouse infection models. The basis for a successful chemical synthesis was structural data of antibiotic binding to the ribosome (Wu et al., 2024).

Arylomycin

Another interesting case is arylomycins, which inhibit the bacterial type I signal peptidase (SPase) that cleaves signal sequences from pre‐proteins, following their translocation across the cytoplasmic membrane. Arylomycins (Figure S2D) consist of a macrocyclic tripeptide core with an N‐terminal lipopeptide tail, a C‐terminal carboxylic acid and two phenols on the aromatic rings of the macrocycle. The crystal structure of the Gram‐negative SPase LepB in complex with arylomycin guided chemical modifications of the compound by chemists from the biotechnological company Genentech. They modified the lipopeptide, replaced the carboxylic acid with an acetonitrile chemical ‘warhead’ to covalently inactivate an active site amino acid from LepB and they replaced the two phenols by primary amines to increase the outer membrane permeation (Figure S2E). These modifications resulted in a 500‐fold potency increase of the modified over the original compound. It showed good activity against ESKAPE organisms and a representative collection of MDR clinical isolates from CDC. Resistance against the modified arylomycin was obtained with low frequency, it mapped around the active site of LepB. Access to the inner membrane was porin‐independent; it was by ‘self‐promoted uptake’. Outer membrane permeability promotion by the added amines was measurable, but moderate. In mouse infection models, the modified arylomycin significantly decreased titres of ESKAPE organisms in a dose‐dependent way (albeit at higher MIC for P. aeruginosa and A. baumannii) and prevented death in a peritoneal mouse infection model at a dose of 5 mg/kg. The report is significant as most antibacterial discovery efforts focus on reinvigorating existing classes of clinically approved antibiotics by means of chemical modification. Owing to arylomycin's novel mechanism of action, it was not susceptible to existing resistance mechanisms (Smith et al., 2018).

Fully synthetic platforms using modular building blocks

Semisynthesis or the chemical modification of natural products derived from fermentation, is challenging. Macrolide antibiotics (macrocyclic lactones with one or more pendant glycosidic residues), such as the clinical antibiotic candidate solithromycin, is produced by a 16‐step linear sequence from erythromycin (Figure S2F,G), a natural product. An international consortium of academic and industrial scientists developed a new fully synthetic strategy to new macrolide antibiotics. They started with eight simple building blocks, two of which are commercially available, and the other six can be produced at reasonable 30–150 g amounts. Three building blocks lead in a linear chemical sequence to an intermediate, four others to another intermediate; the intermediates are fused and the last building block is introduced to finalize the synthesis of a 14‐membered ring structure azaketolide, a hybrid of a ketolide and an azalide, which combines advantageous pharmacological properties from two different drugs. The synthesis scheme is adaptable: a 15‐membered ring structure azaketolide can be synthesized by modifying three of the eight building blocks. A chemical library of 300 fully synthetic macrolide antibiotic candidates were produced by varying the building blocks (e.g. having an azido group, an amino group, a β‐keto lactone function and an allyl group). Notably, 83% of the library showed a low MIC against S. pneumoniae containing resistance factors and S. aureus (Seiple et al., 2016). This report is an extension of an earlier one where the academic US laboratory of this consortium had demonstrated the value of this modular synthetic approach when modifying tetracycline antibiotics. Tetracyclines contain four linearly fused rings, labelled A through D. The C‐ring construction was used to couple structurally varied D‐ring precursors with an AB ring precursor, where AB assures binding to the bacterial ribosome. A compound incorporating a two‐ring structure at the place of the D ring (hence a pentacycline) showed good activity against strains with resistance to tetracycline, methicillin, and vancomycin (Charest et al., 2005).

CIRCUMVENTING THE PERMEATION PROBLEM BY TARGETING THE BACTERIAL OUTSIDE

As it is difficult to design antibiotics that penetrate the two membranes of Gram‐negative bacteria, one might hypothesize that antibiotics targeting components inserted in the outer membrane and accessible from the outside might offer opportunities for novel classes of antibiotics. Progress along this line has in fact been achieved.

Targeting the LPS translocon

Researchers from the University of Zurich/Switzerland had synthesized libraries of β‐hairpin–shaped peptidomimetics based on the membranolytic host‐defence peptide protegrin I. Screening for antimicrobial activity resulted in a compound that had MICs in the nanomolar range against P. aeruginosa (Srinivas et al., 2010). Resistance mutants were obtained and mutations were located in lptD encoding an LPS translocon, located in the outer membrane. This protein mediates the last step in the transport of LPS into the outer membrane after its synthesis in the inner membrane (Okuda et al., 2012). The peptidomimetic antibiotic induced abnormal membrane convolutes in the bacterial cell and impaired the permeability barrier of the outer membrane, leading to a slow bacterial lysis. The peptidomimetic antibiotic displayed protection in a sepsis mouse infection model at concentrations of a 0.5 mg/kg dose. However, the activity of this novel antibiotic was limited to P. aeruginosa. Murepavadin (Figure S3A), the lead compound for LptD inhibitors, is now in clinical development.

Targeting the BamA insertase

Researchers from Zurich and Basel/Switzerland screened cyclic peptides related to murepavadin and found compounds displaying MIC <0.25 μg mL⁻¹ against A. baumannii, P. aeruginosa, E. coli and K. pneumoniae, but showed no activity against Gram‐positive bacteria. Effective compounds consisted of a disulfide stabilized, murepavadin‐type peptide hairpin linked to a polymyxin‐like cyclic peptide. The compounds showed a rapid bactericidal activity with a 3‐log pathogen reduction, low lytic activity against mammalian membranes, and potent in vivo efficacy in mouse infection models against several ESKAPE bacteria causing minimal nephrotoxicity. Fluorescence‐marked drugs located at the outer membrane and caused permeability changes and membrane convolutes. Photolabelling showed an association of the drug with the outer membrane proteins LptE, LamB, and, most consistently, with BamA. Resistant mutants to these peptidomimetic antibiotics showed an amino acid change in an external loop of the BamA β‐barrel. Chemical studies showed that the drug stabilized the closed form of BamA channel. BamA insertase facilitates the folding and insertion of outer membrane proteins as part of the β‐barrel assembly machinery from Gram‐negative bacteria. Binding of this antibiotic inhibits the foldase activity of the BAM complex. The resulting incorrectly folded outer membrane proteins (OMPs), when mislocated to the inner membrane, lead to cell permeabilization and death. If confirmed in clinical trials, peptidomimetic antibiotics might open a new tool against several ESKAPE organisms (Luther et al., 2019).

Darobactin targeting BamA from a nematode symbiont

The above‐mentioned antimicrobial compounds are products of chemical synthesis. An international consortium of researchers screened natural compounds for novel antibiotics. As actinobacteria which yielded aminoglycosides, tetracyclines and β‐lactams have failed to provide new antibiotics, the researchers looked for alternative natural sources. They argued that microbial symbionts of nematodes might contain novel antibiotics. When entomophagous nematodes attack insect larvae they release their bacterial symbionts that paralyse their prey and produce antimicrobials to defend the victim's body against digestion by competing environmental microbes. Initial screens with the nematode symbiont Photorhabdus spotted on agar plates and overlaid with target pathogens were frustrating. When Photorhabdus supernatants were concentrated, a weak antimicrobial activity from a ‘silent’ operon was detected. A bioassay‐guided isolation revealed a novel antibiotic, darobactin (Figure S3B), a heptapeptide with two unusual macrocycle cross‐links. The yield of the ribosome synthesized darobactin was low and needed 2 weeks of fermentation, but it could be expressed in E. coli. It showed in vitro activity against several ESKAPE organisms but was inactive against Gram‐positive bacteria. Darobactin induced membrane blebbing and lysis of the pathogens but not of gut commensals. Mutants resistant against darobactin occurred with a frequency of 10⁻⁸ and mapped to BamA. Importantly the mutated, drug‐resistant pathogens had lost their virulence in mice infections. Binding assays suggested that darobactin bound to a lateral gate structure of BamA, closing the barrel structure. Darobactin protected mice against resistant E. coli, K. pneumoniae and P. aeruginosa in two infection models, making it a promising lead compound for clinical development. The researchers concluded that the location of the target on the bacterium's surface resolves the problem of antibiotic penetration across the permeability barrier of Gram‐negative bacteria (Imai et al., 2019). Researchers isolated BamA from E. coli, inserted it into micelles and incubated it with darobactin and solved the structure by cryo‐EM analysis. Substrate‐free BamA exists in two interchanging conformations with open and closed gates. Darobactin binds to the open form and prevents the first step in the insertion process of outer membrane proteins by BamA, namely the interaction with the signal sequence of the transported protein with the lateral gate of BamA (Kaur et al., 2021).

Macrocyclic peptides blocking LPS transport

A collaboration led by Roche Pharma in Basel/Switzerland discovered further antibiotics targeting other members of the LPS transport system in Gram‐negative bacteria (Zampaloni et al., 2024). In a whole‐cell phenotypic screening of 45,000 tethered macrocyclic peptides (MCPs) (Figure S3C), they detected a compound consisting of a tripeptide subunit and a diphenyl‐sulfide tether to close the ring. The compound showed a MIC of <0.5 μg mL⁻¹ against MDR A. baumannii but was inactive against other Gram‐negative bacteria. In two mouse models the compound protected against lethal infection with A. baumannii. However, in rats the compound led to lipid dysregulation and disease which could be alleviated when synthesizing a zwitterionic derivative. Whole‐genome sequencing of spontaneous bacterial isolates with elevated MIC against this MCP derivative identified mutations in LptG and LptF, proteins of the periplasmic LPS transport system. The lead compound, zosurabalpin (Figure S3D), which is now in clinical trials, inhibits the growth of 90% of 130 mostly MDR A. baumannii clinical isolates at a MIC of 1 μg mL⁻¹. Overall, it lacked off‐target activity, displayed a >5 log bacterial load reduction of pan‐drug resistant A. baumannii in several mouse infection models, and was insensitive to known resistance mechanisms. Cryo‐electron microscopy structural studies revealed that MCP binds to a transmembrane helix of the LPS transporter LptF in the inner membrane. This interaction blocks the release of LPS to LptA, located in the periplasmic space from where LPS is further transported via LptD to the external face of the outer membrane. However, it is not the lack of LPS arrival in the outer membrane, but the accumulation of LPS in the periplasmic space which kills the cell. The seven component LPS transporter is similarly organized in Gram‐negative bacteria but the homologous proteins in different bacterial species share only low aa sequence identity, explaining the species‐specific inhibitory activity of both peptidomimetics against LptD and MCP against LptF (Pahil et al., 2024).

GENOMICS AND PHYLOGENETIC TREES LEAD TO NEW COMPOUNDS

Macolacin

Colistin (Figure S4A) is the last line of defence against a number of Gram‐negative pathogens. Colistin belongs to the polymyxin family of antibiotics, which are cationic cyclic lipo‐decapeptides that arise from non‐ribosomal peptide synthetases organized in biosynthetic gene clusters (BGC). Colistin binds to the lipid A moiety of LPS, disrupting bacterial membrane integrity leading to cell death. The plasmid‐borne mobilized colistin‐resistance (mcr‐1) gene encodes a transferase that modifies lipid A and prevents the interaction with colistin. US researchers argued that while colistin resistance development was accelerated by the veterinary and medical use of this antibiotic, colistin resistance certainly evolved as part of the microbial competition fought between microbes with antibiotics and resistance to them. Consequently, they postulated that polymyxin‐type antibiotics must exist that evolved insensitivity against the resistance genes repurposing modified antibiotics for microbial warfare. They screened more than 10′000 bacterial genomes for polymyxin‐like BGC that showed a characteristic operon structure. They found a cluster which they called mac. From the predicted binding pocket of the synthetic enzymes, they could predict the amino acid sequence of the decapeptide which differs in three positions from colistin. They synthesized this decapeptide, called it macolacin and described potent activity against some ESKAPE organisms, including those containing the mcr‐1 plasmid. The lipid attached to the cyclic peptide of the antibiotic makes non‐specific hydrophobic interactions with the long acyl substituents of lipid A, enhancing its inhibitor action. A biphenyl‐macolacin derivative turned out to be the best antibiotic, inhibiting even pan‐resistant A. baumannii with low MIC. Animal infection demonstrated a 1000‐fold higher inhibitory activity of biphenyl‐macolacin against A. baumannii than colistin (Wang et al., 2022a). In this way, a combined genomics, bioinformatic and peptide synthetic approach represents an alternative pathway against some antibiotic‐resistant pathogens.

Cilagicin

The same authors illustrated the value of this approach with another example. Using the condensation domain from non‐ribosomal peptide synthetase (NRPS)‐encoded lipopeptides, they created a phylogenetic tree for BGC. Many branches of this BGC tree were known to produce antimicrobial lipopeptides. They concentrated on an uncharacterized branch of BGC from Paenibacillus mucilaginosus (class: Bacilli). Bioinformatically, they predicted an 11‐aa peptide which they called cilagicin and produced a synthetic‐bioinformatic natural product (‘syn‐BNP’). Cilagicin was in vitro active against a wide range of Gram‐positive pathogens, but lacked activity against Gram‐negative pathogens with the exception of A. baumannii. Cilagicin's mode of action is its ability to sequester two distinct, indispensable undecaprenyl phosphates used in cell wall biosynthesis. In contrast to antibiotics such as bacitracin that sequester only one form of undecaprenyl phosphates, no resistance development was observed against cilagicin. No reduction of S. aureus was observed with cilagicin in a mouse infection model. When replacing the first guess lipid tail myristic acid of cilagicin by a biphenyl (Figure S4B), they observed a 4‐log reduction of S. aureus in the animal model, demonstrating the utility of this bioinformatic genomics approach for novel antibiotic development (Wang et al., 2022b).

Corbomycin

Canadian researchers had taken a comparable approach: they created a phylogenetic tree for non‐ribosomal peptide synthase condensation domains from 71 BGC of the glycopeptide antibiotic family. Major branches belonged to known glycopeptide groups, but two branches were previously not described. From two Streptomyces representing these new branches, they isolated two type V family members of glycopeptides, complestatin and corbomycin (Figure S4C). Both compounds showed bacteriostatic activity against Gram‐positive pathogens, including MRSA and VRE. The mode of action of both compounds were additive and distinct from that of previously reported antibiotics. Treated cells showed septal defects and both compounds bound to the cell wall. It was difficult to obtain mutants resistant to these compounds; those showing weak resistance carried mutations in diverse autolysins. Corbomycin and complestatin bind to peptidoglycan and block its access to autolysins interfering with cell wall modification during cell growth and division. In a skin infection with MRSA in neutropenic mice, corbomycin reduced the bacterial load by 100‐fold over a vector control and showed similar efficacy as fusidic acid. To avoid cross‐resistance with existing antibiotics, the development of compounds with novel targets is an important goal. The combined genomics and phylogenetic approach shows a way how this can be achieved even in a purportedly overmined groups as the actinomycetes (Culp et al., 2020).

Hypomorphs resolve a dilemma

Antibiotic development confronts a dilemma. Primary chemical screening approaches that use biochemical, target‐based assays frequently yield compounds that lack whole‐cell activity. However, when conducting whole‐cell assays using wild‐type bacteria, no mechanism of action (MoA) was determined, which is crucial for further development of an antibiotic. To address this dilemma researchers designed the PROSPECT platform. They generated hundreds of mutant strains that were depleted in essential targets by fusing to their C‐terminus a tag, which targets the protein for degradation. These ‘hypomorphs’ mutants were barcoded allowing the simultaneous testing of 100 strains against chemical compounds in 384‐well plates. Inhibition results were evaluated by sequencing and barcode counting, yielding large chemical–genetic interaction profiles. The researchers conducted the initial screens with 470 hypomorphs of the about 600 essential genes from M. tuberculosis (Mtb). In their screen, they identified MoA for 45 new molecules. When validating their platform with known targets, they found new chemical scaffolds against validated targets (DNA gyrase, mycolic acid biosynthesis, folate and tryptophan biosynthesis), which is a valuable strategy to overcome antimicrobial resistance with antibiotics targeting distinct proteins. Subsequently, they extended the screen to a larger, unbiased library of 50,000 compounds and tested them against a pool of 150 hypomorphs. They rapidly identified new chemotypes against established targets using reference data and turned to the discovery of inhibitors with completely new MoA. One of the hits inhibited the efflux pump EfpA. With two chemical modifications the hit compound killed specifically non‐replicating, drug‐tolerant Mtb. The researchers isolated mutants resistant to this compound, they all mapped to a single aa in this efflux pump proving the absence of off‐targets for this inhibitor. With this target identified and an efflux test, the researchers identified six further compounds representing three chemical classes that inhibited EfpA (Johnson et al., 2019).

REVITALIZATION OF OLDER ANTIBIOTICS

Natural product antibiotics have co‐evolved with resistance mechanisms. The spread of co‐evolved resistance mechanisms limits the clinical use of many antibiotics. This is the case for group A streptogramin antibiotics: they bind to the peptidyl transferase centre of the bacterial ribosome. Together with group B streptogramin, which blocks the nascent peptide exit tunnel, they inhibit bacterial protein translation. Multiple resistance mechanisms have evolved against group A streptogramins (Figure S4D): they block the binding of the antibiotic to the ribosome by methylating the 23S rRNA target site, a protein dislodges the antibiotic from the binding site, or the antibiotic is deactivated by acetyltransferases. US researchers started with structural analyses of the group A streptogramin virginiamycin in complex with either the bacterial ribosome or in complex with the inactivating acetyltransferase. Then they continued with a structure‐guided rational drug design and synthesized 62 analogues by convergent assembly of diversifiable chemical building blocks that maintain ribosome binding but block the inactivating acetyltransferase. The analogues were tested in vitro against streptogramin‐resistant S. aureus and E. faecalis strains. The best candidate (Figure S4E) was tested in a mouse infection model where the compound showed a 100‐fold reduction in bacterial load compared with control infections. The authors are optimistic that this approach may permit chemical adaptations to extend the clinical longevity not only of the streptogramin class (Li et al., 2020).

REGISTERED NON‐ANTIBIOTIC DRUGS WITH ANTIMICROBIAL ACTIVITIES

Researchers from Germany screened more than 1000 registered non‐antibiotic drugs against 40 bacterial strains and found that 24% of these drugs inhibited the growth of at least one strain in vitro. This approach has the obvious advantage that these drugs have already passed clinical trials and registration procedures, are therefore to be considered as safe for human use. The question was only whether they had—unexpected—antimicrobial activities. The focus of the study was, however, the impact of drugs on the human gut microbiome and, therefore, most of the test strains were gut commensals. The growth of E. coli was for example inhibited by seven human‐targeted drugs (Maier et al., 2018). US researchers extended these observations by asking for the mechanistic and genetic underpinning of this antimicrobial activity of non‐antibiotic drugs. They screened 1050 non‐antibiotic drugs for growth inhibition of E. coli: 176 non‐antibiotics were inhibitory at 10 μM drug concentration. They noted that the antimicrobial activity of these drugs was not due to chemical similarity with antibiotics. To get an indication about the MoA for these non‐antibiotic drugs, they screened these drugs against 6709 barcoded knockout E. coli strains covering 3500 non‐essential genes. By sequencing the barcodes, they calculated the frequency of each knockout for each drug. Depleted knockouts reflected gene loss for functions that increases drug sensitivity, whereas enriched knockouts reflect loss of function that increases resistance. Overall, non‐antibiotic drugs showed MoA distinct (‘orthogonal’) from that of standard classes of antibiotics. Network analysis revealed that non‐antibiotics with antimicrobial activity clustered into highly intra‐connected modules enriched in specific cellular pathways: biosynthetic processes (vitamin, pyrimidine), LPS biosynthesis, DNA damage, siderophore synthesis and oxidative stress largely distinct from the network of antibiotics. In contrast, transport systems (outer membrane proteins, efflux pumps) affected similarly both antibiotics and non‐antibiotics. The observations underlined three representative strategies for adaptive resistance to antimicrobial activity: decreased import, mutation of target and increased export. However, the data also revealed that non‐antibiotics target bacterial cellular components not currently exploited by standard antibiotics, for example, the translation initiation factor IF2 (Noto Guillen et al., 2024).

ANTIBIOTIC COMBINATION VERSUS DUAL‐MECHANISM ANTIBIOTIC

Researcher from Princeton University / USA screened a chemical library of 33,000 molecules for activity against an E. coli test strain that had a compromised outer membrane, making it more permeable to antibiotics. The most potent hit, SCH‐79797 (Figure S5A), was active against both Gram‐positive and Gram‐negative pathogens in vitro and in vivo against A. baumannii in a wax worm model. Resistant bacteria could not be found, raising the problem to characterize the MoA. The researchers used thermal proteome profiling combined with mass spectrometry to answer this question. By increasing the temperature, proteins were precipitated and collected by centrifugation. Antibiotic target protein interaction increases the thermal stability of the protein that remains in solution, and its presence was diagnosed by mass spectrometry. Dihydrofolate reductase was identified and confirmed as an antibiotic target. However, a bacterial cytological profiling showed a different pattern for SCH‐79797 than for trimethoprim, a known folate antagonist antibiotic, suggesting a distinct MoA. Flow cytometry with fluorescent dyes demonstrated that SCH‐79797 disrupted both the membrane potential and permeability barrier. Chemical modification of SCH‐79797 showed that two different parts of the chemical's structure were responsible for the two distinct MoAs and allowed the synthesis of a derivative with increased potency and lesser toxicity. In a mouse infection model with N. gonorrhoeae, the derivative was safe, had favourable pharmacokinetic properties, and reduced the bacterial load. The researchers concluded that compounds combining dual MoAs on a single chemical scaffold can realize potencies not achievable by combining two different antibiotics, each providing one MoA. In fact, a combination of trimethoprim with nisin or polymyxin B led to antagonistic interaction (Martin et al., 2020). This is not an isolated case: researchers from Israel quantified S. aureus survival during prolonged exposure to pairwise bacteriocidal drug combinations. When testing pairs of 14 antibiotics, they often observed a reciprocal suppression whereby the efficacy of the drug mixture is weaker than any of the individual drugs alone (Lázár et al., 2022).

DEEP MACHINE LEARNING APPROACHES

A group of academic engineers, computer scientists and pharmacists from different institutes at Boston/ USA explored the efficacy of deep learning approaches to antibiotic discovery (Stokes et al., 2020). They first established a training set of 1760 FDA‐approved drugs of diverse structure and function, complemented with 800 natural products isolated from plant, animal and microbial sources. Experimentally, they screened them for growth inhibition against E. coli where 120 compounds showed inhibitory activity. The compounds were binarized as hit or non‐hit and a deep neural network model translated the graph representation of the chemical structure into a vector, with finally a single continuous vector representing the entire molecule. This vector model was then applied to the Drug Repurposing Hub, a collection of 6111 compounds at various stages of investigation against human diseases. Of them, 99 were predicted to display antibacterial activities; and 51 indeed inhibited the growth of E. coli. The researchers then selected these 51 compounds for absence of toxicity, being in preclinical or clinical studies and for showing low structural chemical similarity to the training set molecules. This identified a single compound, a small nitrothiazole in preclinical test against diabetes, which they called halicin (Figure S5B). Halicin was bactericidal even for non‐growing E. coli strains containing plasmids expressing five clinically relevant resistance mechanisms were killed; it rapidly killed M. tuberculosis, carbapenem‐resistant Enterobacteriaceae, A. baumannii, but not P. aeruginosa. Halicin has a new mechanism of antibiotic action: it dissipates the ∆pH component of the membrane proton motive force. No resistant cells could be isolated after halicin treatment. In a skin infection model with a pan‐resistant A. baumannii in neutropenic mice, halicin showed a potent therapeutic effect. Halicin also cleared C. difficile in an intraperitoneal mouse infection model. In controls for specificity, none of 10′000 compounds from an anti‐tuberculosis library scored positive in their approach and, from 300 compound tested experimentally, none inhibited E. coli growth. Encouraged by these results, the researchers extended their test to a virtual database of 1.5 billion (!) compounds. After screening for physicochemical properties that are observed in antibiotic‐like compounds, 100 million compounds remained. After 4 days of computing time, their programme identified 1000 compounds with high scores for antibiotic activity. When further selecting for low‐structural similarity to known antibiotics (to obtain new classes of antibiotics), they were left with 23 compounds. Eight chemicals mediated growth inhibition of E. coli and two overcame an array of resistance mechanisms. The researchers were convinced that comparable approaches can refill the antibiotic development pipeline. They warn that machine learning is imperfect and need to be combined with appropriate experimental designs for training. Investigators can also encounter limitations in acquiring the predicted compounds by chemical synthesis needed for experimental verification. However, compared with prior antibiotic screening programmes, deep learning offers enormous cuts in both cost and time.

The same team of researchers extended their approach of deep learning to the Gram‐positive pathogen S. aureus. They first tested 39,000 compounds ranging in molecular weight from 40 to 4200 Da and identified 512 strongly growth inhibiting compounds. Screening for cytotoxicity against three human cell lines in parallel led to the exclusion of about 10% of the compounds. With these experimental structural data as training set, the researchers predicted the cytotoxicity and antibiotic activity of a collection of more than 10 million compounds that can commercially be purchased, thus avoiding the later chemical synthesis of promising candidates by the team. With these two filters, the programme sorted the 39,000 chemicals into 3000 possible hits and 3000 clearly non‐hits. Subsequently, the scientists searched for a structural chemical rationale of the programme identifying antibiotic candidates, which are understandable by humans (‘white vs. black box’). This process eliminated known antibiotic structures and the researchers filtered for novel structures which had previously not been associated with antibiotics. Five such chemical scaffolds were filtered; four of them were growth inhibiting for S. aureus, two also for MRSA. These two compounds were less bactericidal than vancomycin, but showed only a twofold increase of MIC against MRSA (Figure S5C). Resistant mutants did not evolve in standard experiments and physiological experiments revealed—as for halicin—dissipation of ΔpH as a primary MoA. The two compounds were insensitive to various established resistance mechanisms and included activity against VRE and Bacillus subtilis persister cells. In standard mouse infection models, one compound showed a 1.2‐log reduced bacterial load of MRSA compared to controls. This reduction was lower than that achieved with vancomycin or fusidic acid, but, according to these researchers, hold promise for further optimization by chemical modification (Wong et al., 2024).

OUTLOOK

It is still unclear if the projection of 10 million deaths per year worldwide due to antibiotic‐resistant bacterial infections will materialize in 2050. However, the writing is on the wall, and we should take this warning seriously, particularly since the GBD 2019 report calculated 1.3 million deaths per year due to antibiotic‐resistant infections, which is not a small number. Should this number further increase, and the spread of antibiotic resistance mechanisms suggests that this problem will do so, we get quickly close to the death toll of a pandemic. Complacency is thus not in place. The crisis feeling is further accentuated by the fact that major pharmaceutical companies have shifted their commercial focus away from antibiotic development. Indeed, venture investments for cancer drugs has steadily increased over the last decade to US$ 7 billion in 2020 while investments in antibiotics over the same time period remained flat and did not even reach a quarter billion US$ per year (McKenna, 2024). Glaxo and AstraZeneca reported objective technical hurdles to identify promising lead compounds. The overview of the scientific literature over the last decade indicates that some of these scientific hurdles have not only been defined in detail, but also new promising antibiotic candidates were identified by various approaches. While this could give rise to some optimism, there is a caveat: most of this work has been conducted by academic institutions and most of these efforts stopped after demonstrating in vivo efficacy of the antibiotic candidate in experimental animals, mostly mice. The reason for this situation is that academic institutions do not have the financial resources to develop an antibiotic through clinical trials to FDA registration. An estimate from 2016 gives a price tag of 1.4 billion US$ for the process from lead compound identification, optimization to drug registration (McKenna, 2024). Small wonder that small biotech companies that specialized on antibiotic development went bankrupt before reaching a profitability zone. Even for big pharmaceutical companies the peculiar characteristics of the antibiotic market (e.g. the AWARE strategy) make it uncertain whether the costly process of developing a new antibiotic will assure a return of investment. Due to this profitability dilemma, the market forces are unlikely to provide a solution for the development of novel antibiotics. Alternative solutions are needed. One is CARB‐X (Combating Antibiotic‐Resistant Bacteria Biopharmaceutical Accelerator), which has allotted nearly half a billion US$ from government and philanthropic organizations. Additionally, the UK government in its National Institute for Health and Care Excellence (NICE) programme will support two companies by guaranteeing with £10 million per year the purchase of antibiotics they are developing against hospital infections, a typical pull initiative.

Push approaches had already some effects in the past: the US NIH has supported a successful phase II clinical trial of zoliflodacin, a spiro pyridmidine trion (Figure S5D), against uncomplicated gonorrhoeae (17% of N. gonorrhoeae isolates were antibiotic‐resistant) (Taylor et al., 2018). The compound was identified by AstraZeneca, which passed it to a spin‐off, Entasis Therapeutics, when AstraZeneca withdrew from antibiotic development. This trial can be quoted as an example for the push strategy. However, one should note that this clinical trial enrolled just 179 patients, and much bigger amounts of money are needed to support phase III clinical trials. It is currently not obvious how the public sector could politically support antibiotic development with the billions of US$ needed to bring novel antibiotics to the market. The public sector has no factories to produce novel antibiotics. Even for conventional antibiotics, the pharmaceutical industry in many Western countries outsourced their production to China, leading to a shortage of conventional antibiotics in the last year.

The Spanish flu pandemic claimed more victims than World War I on all its battlefields. In 2019, fatal infections with antibiotic resistant bacteria claimed a number of lives in the US that amounted to two thirds of US soldiers killed in the Vietnam War. Based on these comparisons, one might wonder whether prevention measures against the current, not to speak of a future full‐blown antibiotic crisis, should not become part of homeland security or the defence departments where billions of dollars of investment are politically easier swallowed than investments into public health as seen by the repeated refusals of the PASTEUR Act by the US Congress (McKenna, 2024).

FUNDING INFORMATION

No funding was received.

AUTHOR CONTRIBUTIONS

Harald Brüssow: Conceptualization; writing – original draft.

CONFLICT OF INTEREST STATEMENT

The author has no conflict of interest.

Supporting information

Figure S1.

Brüssow, H. (2024) The antibiotic resistance crisis and the development of new antibiotics. Microbial Biotechnology, 17, e14510. Available from: 10.1111/1751-7915.14510

DATA AVAILABILITY STATEMENT

Data sharing is not applicable to this article as no new data were created or analyzed in this study.