Re-Evaluating the Approach to Drug Target Discovery in Multidrug-Resistant Gram-Negative Bacilli

KEYWORDS:

• Acinetobacter baumannii
• antibiotic target
• bacterial drug target
• essential genes
• Gram-negative bacilli
• multidrug resistant

The incidence of infections due to multi-, extreme- and pan-drug resistant (MDR, XDR and PDR) bacteria is increasing. The term ESKAPE pathogens was coined by the Infectious Disease Society of America to designate bacterial pathogens that cause life-threatening infections and exhibit a high incidence of drug resistant strains [1]. ESKAPE represents Enterococcus faecium, Staphylococcus aureus, Klebsiella pneumoniae, Acinetobacter baumannii, Pseudomonas aeruginosa and Enterobacter spp. Escherichia coli has joined the line-up with its ability to acquire and express extended-spectrum beta-lactamases and carbapenemases. A. baumannii is one of the most problematic species and has been termed a poster child for XDR and PDR Gram-negative bacilli (GNB) [1]. Treatment of drug-resistant infections is challenging, with resultant increased morbidity, mortality and healthcare costs. The mortality rate is up to 50% for infections caused by XDR K. pneumoniae or XDR A. baumannii and clinicians often must resort to therapeutics possessing significant toxicity or side effects, such as colistin. Strains of A. baumannii resistant to colistin have recently been reported, with concomitant development of cross-resistance to host antimicrobial peptides [2]. The fear of a postantibiotic era was initially predicted to be due to methicillin-resistant S. aureus (MRSA) or vancomycin-resistant Enterococcus (VRE) but never realized. It is now on the cusp of being fulfilled by GNB such as A. baumannii. Over 2 million people acquire serious drug-resistant bacterial infections annually in the USA, resulting in more than 23,000 deaths. Many more deaths are attributed to medical conditions complicated by antibiotic-resistant infections [3]. The annual cost to the US economy of treating drug-resistant infections is estimated to be $20–$35 billion.

New classes of antibiotics against all types of bacterial pathogens are required, but the situation is particularly urgent with respect to GNB. Unfortunately, a 2013 review revealed that there were virtually no new antimicrobial agents active against GNB in the pharmaceutical pipeline [4]. In fact, avibactam, a non-β-lactam β-lactamase inhibitor, delivered in combination with ceftazadime was the only novel anti-GNB in clinical trials. Development of new antimicrobials is a long, arduous and expensive process. As a result, many major pharmaceutical companies have exited antimicrobial R&D. In order to address the dearth of antibiotic development, in 2012 the Generating Antibiotic Incentives Now Act was enacted in the USA to provide incentives for the development of new antibiotics, including priority review by the US FDA and a 5-year extension of exclusivity. Even with new incentives, there has been little advancement in the treatment of GNB infections.

A variety of approaches are being used to fill this need. Examples include analogs of currently approved antibiotics, repurposing of approved or candidate drugs, development of efflux pump and β-lactamase inhibitors and discovery of new chemical entities. Screening campaigns to identify new small molecule compounds follow two general strategies: whole-cell phenotype or target based. Phenotype screens offer the advantage of directly identifying compounds possessing bactericidal or bacteriostatic activity. However, this strategy often does not connect activity to a specific target(s), adding significant difficulty during the hit-to-lead optimization process. Target-based screens offer the promise of a theoretically elegant approach, but have yet to result in a bountiful yield of new antibiotics. The limiting factor of this approach is that compounds active against purified targets may not display activity against live bacteria, most often due to permeability and/or efflux issues. Target-based methods gained favor in the mid 1990s when whole-genome sequencing data became available. However, the increased focus on target-based screening occurred while FDA approvals of new antibiotics were decreasing, particularly those representing novel classes. Therefore, this inverse relationship has given target-based antibiotic discovery a poor reputation. The trend of decreased antibiotic discovery began well before target-based methods came into vogue (only a single new class, mupirocin, was launched between the early 1970s and 1999), due to a variety of economic, regulatory and scientific factors [4]. In natural product screens, the low-hanging fruit were picked (and repicked) over the past decades and now methods are required to avoid rediscovering these known natural products while identifying new natural products with antibacterial activity [5]. Many target-based screens suffered from nonoptimal selections of chemical libraries [6]. For pragmatic reasons, large chemical libraries are often biased toward chemical space favorable for identifying drug-like molecules against eukaryotic therapeutic targets (e.g., oncology and cardiovascular, among others). Chemoinformatic studies have demonstrated that drugs acting upon eukaryotic versus prokaryotic targets possess significantly different chemical properties [7]. The lack of positive results may be partially due to looking in the wrong place rather than due to the target-based strategy itself. Lessons learned from past antibiotic discovery efforts will inform the newly re-emerging and urgently needed efforts.

Target-based antibiotic discovery strategies place great weight on gene essentiality. However, selection and prioritization of putative antibacterial targets have largely relied on inferred essentiality annotations due to the relatively small number of genome-wide essentiality screens, with the majority conducted on rich lab media [8,9]. Unfortunately, inferred essentiality often neglects the significant disparity in annotation for a given orthologous family across experimental essentiality screens. Pairwise comparison of genome-wide prokaryotic essential gene sets typically have 50–70% overlap, but this rapidly decreases as the number of species being compared increases [10]. This disparity reflects real genetic and phenotypic differences between species and even strains, but also reflects experimental approaches and data analysis. Confidence levels vary greatly for gene essentiality predictions and in the past (and present) have resulted in the selection of nonoptimal targets to pursue for antibiotic drug discovery or the exclusion of targets perhaps worthy of consideration.

Further, gene essentiality has been traditionally defined as the minimal gene set required for growth of a specific organism under optimal growth conditions. However, gene essentiality is increasingly being viewed as contextual [11]. Altered nutrient levels, changes in carbon sources and environmental stressors have been demonstrated to change the set of genes required for growth or survival [12,13]. Microorganisms infrequently encounter ideal growth conditions, except for in the lab and so they have evolved to grow and survive in multiple and changing environments. For example, bacterial pathogens have coevolved with their host(s) and will typically encounter a very different environment during infection of a host (nutrient poor, host defenses) versus during growth on lab media (nutrient rich). Inclusion of genes required to grow and survive under nonoptimal conditions (contextual essential genes) may expand the set of potential antibacterial targets. Conversely, genes experimentally determined as essential under optimal in vitro growth conditions have been later discovered to be dispensable during infection of a host, resulting in reconsideration of drugs under development [14,15].

Our group has recently published a unique methodology that employed screening of A. baumannii mutants in ascites, a clinically relevant human body fluid and subsequent in vivo validation to specifically and efficiently select for and identify what we termed in vivo essential genes [16–18]. These in vivo essential genes are not essential for growth and survival under standard rich lab media growth conditions, but are essential for growth and survival during infection of a host. Comparison of these A. baumannii in vivo essential genes against the Database of Essential Genes (DEG) demonstrated that 89% of these genes could not be readily predicted as being essential in vivo. This figure is even more profound considering that essentiality data for a related species, Acinetobacter baylyi is present within DEG. Likely, this observation was due to most genome-wide screens cataloged in DEG being performed in vitro using rich laboratory media and none used a clinically relevant medium. The A. baylyi essentiality screen used a defined minimal medium, but that failed to fully represent in vivo conditions. These results emphasize that when prioritizing antibacterial targets, caution is required when trying to predict essential genes either for a single species under different growth environments or across species. Neglecting these important caveats may result in unnecessarily excluding potential targets or assigning falsely high rankings to targets. In other words, there is no replacement for experimental data conducted under conditions that best represent those of an actual infection. Our method is efficient and can be extended for use with any pathogens that grow in a relevant human body fluid. This philosophy has been implemented at Pfizer in screen mimicking carbon source conditions encountered during infection that identified antimicrobial lead compounds acting upon the two glyoxylate shunt enzymes in P. aeruginosa [19].

There is no silver bullet for antimicrobial drug development and especially for agents effective against drug-resistant GNB. At the target identification stage, it would seem prudent to design a screening strategy that identifies essential genes under clinically relevant conditions. Likewise, if a previously identified putative target is chosen for further study, it is critical to confirm essentially under conditions similar to those the pathogen will experience during human infection. In this manner, the chances of discovering that a target is dispensable during human infection after significant resource investment will be minimized [14,20]. We are faced with an extremely difficult challenge in an era of decreasing resources for antimicrobial development. It is imperative that logical drug design is based on logical target identification.

Financial & competing interests disclosure

This work was supported in part by a Telemedicine and Advance Technical Research Center (TATRC) cooperative agreement W81XWH-11-2-0218 (TA Russo, LW Schultz and TC Umland) and a VA Merit Review from the Department of Veterans Affairs (TA Russo). The authors have no other relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials d­iscussed in the manuscript apart from those disclosed.

No writing assistance was utilized in the production of this manuscript.

Additional information

Funding

This work was supported in part by a Telemedicine and Advance Technical Research Center (TATRC) cooperative agreement W81XWH-11-2-0218 (TA Russo, LW Schultz and TC Umland) and a VA Merit Review from the Department of Veterans Affairs (TA Russo). The authors have no other relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials d­iscussed in the manuscript apart from those disclosed. No writing assistance was utilized in the production of this manuscript.