Mining for novel antibiotics

Keeping pace with the development and spread of drug resistant bacteria will require a strong and diverse antibiotic pipeline. Currently, clinical phase development remains quite limited and narrow, consisting largely of derivatives from known antibiotics [1,2]. The modification of existing antibiotics increases the lifetime and useful breadth of known antibiotics, but its ability to keep up with antibiotic resistance is finite. Thus, discovery and clinical development of novel classes of antibiotics with new targets is essential.

Finding and advancing new drugs to treat disease is shared by all therapeutic sectors. However, relatively poor financial returns on antibiotics make them a less attractive investment [3,4]. Thus improving the efficiency with which we identify, select, and validate new antibiotic leads for clinical investigation will be important for reviving and maintaining a robust antibiotic pipeline.

Natural reservoirs

Historically, the discovery of novel antibiotics has come from mining microorganisms (predominantly from soil) for natural products with antimicrobial activity [5–7]. During the mid 20^th century, this approach identified the majority of antibiotic classes we use today. After an initial burst of discovery, this approach quickly had diminished returns, as rediscovery of previously identified antibiotics became the norm. However, this early work investigated only a fraction of potential natural habitats and microbial sources for antibiotics. New microbe hunters are probing underexplored niches and microbes with encouraging results.

Recent work has highlighted the diversity of compounds that can be found in native microbiomes. Investigation of bacteria from the nematode microbiome led to the discovery of darobactin, a compound with a new mechanism of action that targets the essential Gram-negative outer membrane protein BamA [8*]. Staphylococcus lugdunensis from the human nose was shown to produce lugdunin, which inhibits the growth of Staphylococcus aureus [9]. Our own microbiome has also been found to produce a wide spectrum of natural products, many with antibacterial activity [10]. In addition to exploration of microbiomes and terrestrial niches [11,12], the marine environment is also proving a rich source of antibiotic leads [13,14], though in all cases issues of cultivability remain.

The vast majority of bacteria are difficult or impossible to cultivate under standard laboratory conditions historically used for antibiotic identification. Furthermore, the gene clusters responsible for antibiotic production frequently remain silent until they receive proper stimuli. Both of these hurdles prevent facile investigation of potentially new antibiotics. Thus, while we push to explore new environments we also need to advance methodologies used to grow and trigger bacteria to produce any hidden antibiotics they may contain. New techniques are being implemented to grow and test uncultivable samples in their natural environment both alone and in association with other microbes. The development of methods to grow bacteria in situ, such as the iCHIP diffusion chamber, has facilitated the cultivation of previously uncultivable bacteria by allowing them access to the natural compounds required to meet their nutritional needs [15,16]. This general approach aided in the discovery of lassomycin, which targets the ATP-dependent protease ClpC1P1P2 to kill Mycobacterium tuberculosis [12], and teixobactin, which has potent activity against Gram-positive bacteria through inhibition of cell wall synthesis [11]. Microbes may only produce new compounds when stimulated by other prokaryotic or eukaryotic organisms. Co-culturing approaches can provide the necessary signaling molecules to help coax expression of latent gene clusters and their associated natural products [17–19]. The advancement of approaches that closer approximate native microbial environments is likely to continue to help identify novel antibiotics.

Integration of omics data in antibiotic discovery

As we expand our exploration of the natural world, approaches that can focus our efforts will help increase the throughput and efficiency of antibiotic discovery. Advances in DNA sequencing technology have been instrumental in identifying new bacteria and gene clusters from uncultured communities likely to produce new antibiotics. The use of genome-mining approaches can help prioritize communities for screening, and provide routes for synthetic investigation. A recent method that incorporated phylogenic analysis of biosynthetic genes along with a lack of known co-associated resistance determinates was used to predict glycopeptide antibiotics that were likely to have unique biological activities [20]. This process led to the identification of complestatin and corbomycin, which were shown to have a novel mechanism of action that targets autolysins in Gram-positive bacteria.

Once identified, heterologous expression of latent biosynthetic gene clusters can be used to produce new compounds in sufficient quantities for study and development, rather then relying on production from the parent organism [21,22]. Alternatively, as computational prediction of gene cluster products improves, putative antibiotics may be synthesized de novo based solely on genomic information. Humimycins were identified by genome mining of human-associated bacteria [10*]. Based on a bioinformatics prediction, select humimycins were chemically synthesized and shown to be active against several bacteria including Streptococcus and Staphylococcus species.

All of these advances offer fresh hope of identifying antibiotic leads from natural sources. However, it remains unclear how deep this well of clinically viable natural products goes. Increasing screening efficiency will be important as some estimates suggest thousands if not millions of bacteria will need to be investigated to find each new viable lead [5,6]. Once identified, production at scale of new natural product scaffolds necessary for optimization and eventually distribution can also be challenging. Clearly, natural sources will play an important role in identifying novel leads going forward. However, development of additional strategies will both bolster natural product discovery, as well as offer further paths to antibiotic development that are needed to restore and maintain a robust clinical pipeline.

Computation provides new paths to discovery

Past attempts to discover new antibiotics using high-throughput screening of small molecule chemical libraries against defined targets proved largely unsuccessful [23,24]. However, the data from screens like these, and the ever-growing list of active and inactive compounds generated over the past several decades, may provide information needed to advance new routes to identify novel antibiotics. This importantly includes datasets describing factors influencing the ability to bypass the bacterial membrane barrier in addition to antimicrobial activity [25–27]. New computational approaches, such as machine learning, now allow us to churn through the massive antimicrobial datasets at our disposal to look for hidden patterns associated with antibiotic activity. Unlike previous computational exploration of drug design, these new approaches do not require human theories of antibiotic action and can find patterns in nature itself. Importantly, these methods can factor in aspects of toxicity and large-scale synthesis prior to discovery, which is not possible with traditional antimicrobial screening alone.

A recent study developed a neural network that identified antibacterial compounds from a set of more than ten million molecules [28**]. Importantly, some of the identified compounds were structurally distinct from known antibiotics. This work demonstrates the power of combining experimental and in silico approaches to probe new areas of antibiotic chemical space. Investigation of antibacterial peptides has also benefited greatly from new computer-aided approaches. Ribosomally synthesized antimicrobial peptides have been advocated for years as potent antibacterials, but clinical success has been largely elusive. However, databases of thousands of natural and synthetic antibacterial peptides that are now available can facilitate computationally-aided de novo generation and optimization of peptide antibiotics [29]. Recent studies have shown that design principles can be applied to identify stable, active, and non-toxic peptides to treat infection [30]. Furthermore, computational approaches are also being developed to perform in silico evolution and further increase the efficiency with which we can modify leads for improved antibacterial activity [31]. The distinction between small molecule and peptide antibiotics is fluid as many old and newly discovered antibiotics are based on a macrocyclic peptide core structure. However, our understanding of rules supporting macrocyclic peptide antibacterial activity is relatively small compared to their linear counterparts. Expansion of peptide databases to fill this knowledge gap may enable the development of more potent macrocyclic peptide based antibiotics. New approaches to generate and test diverse macrocyclic sequences will help meet this need [32–34].

Beyond small molecules

Treatments in other therapeutic areas, such as oncology, have expanded from small molecules, to biologics, to cell-based treatments in recognition that a one size fits all model is not appropriate to target all factors of a disease. While antibiotic development has largely been confined to small molecules, it also appears to be expanding to include new modalities. Monoclonal antibodies have been wildly successful in treating a range of diseases. They have several unique properties compared to small molecules including generally low toxicity, high specificity, an ability to engage the immune system, and extended half-life [35]. While efforts to identify antibodies with direct antibacterial activity have been challenging [36,37], other antibody-based approaches have shown promise. A bispecific antibody from MedImmune targeting the virulence factor PcrV and exopolysaccharide Psl of Pseudomonas aeruginosa has shown efficacy in mouse models, and is preceding through clinical development [38,39]. Still other approaches combine the benefits of antibody and small molecule approaches. Genentech developed a novel antibody-antibiotic conjugate to combat S. aureus. The antibody binds the bacterium, but the antibiotic only becomes active once inside host cells that have taken up the bacterium [40*]. This antibody-antibiotic conjugate is also proceeding through clinical development [41]. The natural repertoire of infected and convalescent patients may also prove a fertile ground for identifying protective antibodies [42] and help reveal new bacterial epitopes that can be targeted to control infection. A major hurdle in pursuing innovative new approaches is the lack of standardized proof-of-concept assays [2]. In this regard it is encouraging to observe two respected companies pursuing this new direction and it will be exciting to see how they validate and position monoclonal treatments in the market.

The unceasing emergence and spread of antibiotic resistant bacteria is beginning to leave the clinical antibiotic cupboard bare. Previously stalled developmental strategies using natural sources and synthetic screening have been reinvigorated by new technologies and computational approaches providing renewed hope (Figure 1). Still, continued advancement and investment in the field will be required to restore the clinical antibiotic pipeline back to its needed levels.

Figure.1.

Exploration of new antimicrobial sources and advances in computational approaches can synergize with advanced methods for high-throughput screening and production to help generate novel antibiotics and expand a limited clinical pipeline

Highlights.

• Exploration of new microbial habitats provides opportunity for antibiotic discovery

• Computational approaches advance detection of hidden antibiotic sources

• Large datasets provide opportunity for in silico antibiotic development

• Antibody based approaches expand options to fight bacterial infection