Evaluation of a Library of FDA-Approved Drugs for Their Ability To Potentiate Antibiotics against Multidrug-Resistant Gram-Negative Pathogens

The Prestwick library was screened for antibacterial activity or “antibiotic resistance breaker” (ARB) potential against four species of Gram-negative pathogens. Discounting known antibacterials, the screen identified very few ARB hits, which were strain/drug specific.

For ARBs the corrected percent inhibition was defined by growth inhibition in the presence of compound plus antibiotic minus the growth inhibition in the presence of compound plus DMSO control. The maximum inhibition was then defined as the highest corrected percentage inhibition observed between the two duplicates of the HTS. Finally, ARB hits were defined as maximum corrected percent inhibitions that were greater than the mean + 3SD of the maximum inhibitions ( Figure 2 and supplementary figures 2-6). False positive ARBs, particularly from wells where the use of DMSO alone gave an aberrantly high growth inhibition which masked any ARB activity, were included in the initial analysis, but haven't been discussed further in this manuscript.
There are some striking differences with the numbers of ARB hits between bacterial species (Table  S2). The highest number of ARB hits was found in combination with CST, which is most likely due to the ability of CST to permeabilise the outer membrane of select Gram-negative species, thereby effectively becoming the ARB for the test compounds and allowing them into the cell to interact with an intracellular target 4 . The majority of hits for K. pneumoniae were known antimicrobials or antiseptics, whereas for other species, up to half of the hits were compounds with no previously described antimicrobial activity. Interestingly, for combinations of compounds with GEN, the highest number of ARB hits was found in E. coli (20 ARBs), which had ten times the amount of hits identified for K. pneumoniae (2 ARBs). Very few hits were observed in P. aeruginosa with any of the compounds directly (16 hits) nor in combination with antibiotics (40 ARBs across all antibiotic combinations). Aside from the high number of ARB hits identified with CST, hits were distributed by antibiotic with the following frequency, irrespective of species (GEN (107) > MEM (81) > CIP (75)).
There were interesting observations around potentiation of different antibiotic classes by a number of antiseptics and interactions between existing antimicrobials (mostly weak antibacterials which had greater potency when combined with CST, such as rifampicin, furazolidone, clioquinol and nitrofurantoin). The strongest combination of known antimicrobials, CST and rifampicin, has known synergistic interactions in vitro, which has so far failed to translate in vivo 5 . As these were not the focus of the current study, these have not been discussed further. The focus of this study was to identify new compounds lacking known or demonstrable antimicrobial effects which might work synergistically with one or more antibiotic class. The intention would be to look at these as candidates for direct repositioning as adjunct therapies. Amongst the hits identified, it was possible to identify three groups of compounds that showed potentiation in more than one member of the class (Table 1). However, most are cytotoxic and future studies will need to carefully consider whether there is a therapeutic window which would support development of such molecules as adjunct therapies.

Anthracyclines and derivatives.
The anthracyclines are 4-ring heterocyclic compounds originally identified from natural products derived from Streptomyces and related species. They are some of the most widely used drugs for chemotherapy for a variety of different tumour types. Within this class, the clearest potentiation in this screen was with daunorubicin/epirubicin and CST, against both K. pneumoniae and A. baumannii, with >93% reduction in growth observed in both species at 20 M. ARB activity with mitoxantrone was limited to potentiation of TGC against P. aeruginosa in this study, but the literature has reported synergy with imipenem in S. aureus and E. coli and with CST in the laboratory strain PA01 3, 6 .
The mechanism of action of anthracyclines is multifactorial, but key to their efficacy as chemotherapy agents is their ability to intercalate between bases in DNA or RNA and poison topoisomerase II, leading to toxic double stranded DNA breaks and causing cell death. The observed activity in bacteria is likely to be via a similar mechanism. Whilst this likely conservation of mechanism may preclude clinical development as an adjunct antimicrobial therapy, encapsulation of anthracyclines to minimise off-target effects and targeting to increase specificity might offer possible routes forward. Such approaches have been widely explored for improving the target specificity and reducing the general toxicity for chemotherapy drugs 7,8 . Interestingly the related anthracycline compound, doxorubicin, was also included in the screen but showed no activity, either as a direct hit or an ARB. This perhaps suggests that the mechanism of potentiation is not simply related to toxicity of the compounds, but is mediated by a specific cellular mechanism. Anthracycline analogues with an improved therapeutic index as antibacterials have also been identified, suggesting that there may be merit in exploring this class of molecule further 9 . Strategies to explore other DNA binding molecules, currently being used as chemotherapy agents, have also been explored with pyrollobenzodiazepines being the subject of recent papers 10, 11 (Picconi, Hind et al. manuscript in preparation).

Antimetabolites.
Zidovudine (ZDV), also known as azidothymidine (AZT), is a thymidine analogue that inhibits the reverse transcriptase of HIV 12 . The compound showed direct antibacterial activity against K. pneumoniae at 20 M and E. coli at 7 M. ARB effects are also seen with MEM, GEN and CST in K. pneumoniae and in A. baumannii with CST only. Direct activity of ZDV is well characterised in E. coli 13 , where ZDV appears to have dual targets, thymidine phosphorylase 14 and thymidine kinase 15 . ZDV requires activation to its triphosphate version by cellular enzymes and, whilst this may improve its safety profile, the rate of this activation mechanism in different species may contribute to differential activity between species. ARB activity has also been reported previously, most notably with CST in Enterobacteriaceae 16,17 . There are also reports of ZDV working synergistically with trimethoprim in Enterobacteriaceae with the effects being enhanced by inclusion of floxouridine, perhaps by modulating intracellular nucleotide/nucleoside pools 18 .
Floxuridine (5-fluorodeoxyuridine), a pyrimidine analog used in cancer therapy, showed direct activity against E. coli at 7 M and K. pneumoniae at 20 M and significant potentiation of MEM, CIP and CST in K. pneumoniae and CST in A. baumannii at 7 M. Activity of floxuridine has been reported previously, also against E. coli and K. pneumoniae, but was shown to be bacteriostatic (unlike ZDV which was rapidly bactericidal) 19 . The study highlighted the potentially poor prospects of developing the drug as an antimicrobial, given its high toxicity, low recommended dose of 0.3 mg/kg daily 20 and very short plasma half-life of around 15 minutes 21 . This is perhaps exacerbated by common side effects, seen in around 30% of patients, which include low blood counts (platelets and white cells) and diarrhoea 22 .
The third antimetabolite compound, didanosine, a nucleoside analogue of adenosine used as an antiviral, showed ARB activity with MEM at 7 M and CST at 20 M against K. pneumoniae and CST against A. baumannii at 20 M . Previous studies have reported direct antibacterial activity of didanosine against E. coli and Salmonella, but this was relatively modest (31 -62 and 2 -125 mg/L respectively) which was reflected here in K. pneumoniae with direct activity of just 30 % growth inhibition at 20 M. Previous ARB effects have not been reported to our knowledge. Opportunities to further develop this compound as an adjunct therapy may be limited by the low Cmax (~2 mg/L) described at normal doses while didanosine might require significant dose escalation to achieve relevant therapeutic effects 23 .
A fourth antimetabolite drug, gemcitabine, showed limited but intriguing potentiation with CIP in E. coli at 20 μM. Gemcitabine is another nucleoside analogue used to treat patients with a variety of different cancers. Recently there has been considerable interest in the relationship between the microbiome and gemcitabine, notably the suggestion that intratumour bacteria are capable of inactivating the drug via the action of cytidine deaminase enzymes, generating the inactive form 2',2'-difluorodeoxyuridine 24 . The activity is a function of the long form of the bacterial cytidine deaminase enzyme, which is expressed in E. coli along with many other Proteobacteria. The use of prodrug versions of floxuridine and gemcitabine similar to those generated for ZDV 25 , might be a possible route forward, as might the generation of amino-acid conjugated prodrugs with improved safety profiles 26 , and targeting bacterial uptake.
Other nucleoside analogues, abacavir, emtricitabine, lamivudine, stavudine and zalcitabine were included in the library but showed no direct or ARB activity.

Psychoactive drugs.
Gabapentin, an analogue of the neurotransmitter gamma-aminobutyric acid (GABA), used to treat neuropathic pain, was an ARB at 20 M in E. coli with MEM and CIP. Gabapentin is already used to treat side effects of diarrhoea infections caused by E. coli and the ARB activity may be a useful component of this therapy. Gabapentin has a Cmax of 8.7 mg/L which is compared to a concentration of 3.4 mg/L at 20 M. Previous studies have suggested direct antibacterial activity against S. aureus, P. aeruginosa and E. coli 27 , but that was not observed in other studies 28 . The latter study identified a number of other psychoactive compounds with broad ranging antibacterial activity including sertraline (not a hit or ARB in this study) and fluoxetine (ARB with CST in A. baumannii). Thioridazine, an antipsychotic phenothiazine drug, also showed ARB activity with CST in A. baumannii at 20 M. Interestingly, thioridazine has been shown to have ARB properties in studies with oxacillin and dicloxacillin in MRSA 29,30 as well as direct antimicrobial effects against TB 31 . We have also previously shown that fluoxetine and thioridazine can impact on biofilm formation in Proteus mirabilis, with a mechanism of action that relates to inhibition of a specific MFS-family efflux pump 32 .
Fluspirilene, is a diphenylbutylpiperidine drug used in the treatment of schizophrenia and anxiety disorders, working via blockade of the dopamine D2 receptors. The compound showed potentiation of CST with both A. baumannii (7 M) and with K. pneumoniae (20 M) as well as direct antibacterial activity against A. baumannii at 20 M. This data largely confirms the results from a recent publication, which also described ARB activity in A. baumannii with both CST and azithromycin 33 . Interestingly, fluspiriline has also been explored to identify new drug targets via target-independent hit expansion and target identification, suggesting this scaffold may have general properties that are useful for repurposing 34 .
Oxethazaine is used as a local anaesthetic and a component of treatments for ulcers, and works by altering the sensitivity of the membranes of gastric or sensory cells to sodium 35 . It acted as an ARB with CST at 7 M in K. pneumoniae and 20 M in A. baumannii. There are no previous descriptions of activity of oxethazaine as either a direct antibacterial or as an ARB.
A more in-depth exploration of the ARB activities of fluspirilene and oxethazaine in combination with colistin was conducted in a larger panel of colistin-resistant strains. When MICs of fluspirilene and oxethazaine were assessed as growth curves, a number of additional strains showed significant growth defects at sub-MIC concentrations. The highest concentration that did not reduce growth by more than 20% was selected for ARB studies. This is reflected in the lower concentrations of both compounds used for potentiation studies with 2/3 P. aeruginosa strains (5 μM) and with A. baumannii strains (5 μM oxethazaine; Table S4). The compounds were used at 10 or 20μM for all other strains tested. The reason for the strains dependence and/or heteroresistance to these compounds was not investigated further.
The CST resistance associated mutations were the only mutations identified by whole genome sequence analysis in these particular strains (Table S3). NCTC 13439 CST 2A is the only K. pneumoniae strain tested with a mutation in pmrA and as such, this could be directly linked to the colistin-potentiation by fluspirilene. With respect to the two additional strains which show potentiation with fluspirilene, there are other strains with mutations in the same genes (pmrB and mgrB) which do not show CST potentiation. As other CST-resistant mutations in the same strain backgrounds (MGH 78578 and m109) did not show CST potentiation, the data suggests that the response in all three strains is also not linked to specific strain backgrounds. This suggests that the activity of fluspiriline as an ARB for CST may be related to specific alleles in each of the genes associated with CST resistance.
Intriguingly, one strain of K. pneumoniae (141 CHX) showed a significant loss of sensitivity to CST in the presence of either oxethazaine or fluspirilene. This strain has an MIC of 2 mg/L for CST in the absence of potentiator and had the lowest MIC of any strain tested. The strain has a mutation in wbaP (encoding a single amino acid change WbaP D445E), one of the glycosyltransferases involved in O antigen lipopolysaccharide and polysaccharide capsule synthesis in other Gram-negative species. Whilst this mutation appears to have only a minor effect on CST resistance in this strain, it is linked to modestly elevated chlorhexidine resistance (from 16 mg/L to 32 mg/L; unpublished). Why this relatively conservative amino acid change should generate such a strong antagonistic effect to CST activity is not clear.

Other miscellaneous ARB hits
Thonzonium bromide was identified as a potent ARB with CST in both A. baumannii and K. pneumoniae and has previously been identified as a CST ARB against P. aeruginosa 3 . Thonzonium bromide is a cationic surface-active compound, used to promote dispersion and penetration of cellular debris as part of various formulations, promoting tissue contact of the administered medication. Examination of the structure suggested the presence of a quaternary amine group, which might explain its function as an ARB. Interestingly it did not show direct antibacterial activity, which is distinct from other cationic antiseptics (chlorhexidine, alexidine) which showed both direct activity and broad-spectrum ARB function in this screen. Its efficacy as an antibacterial is unclear, but it has been tested as a nanoparticle formulation for Streptococcus mutans biofilms 36 and described as a potential new anthelmintic following a drug repurposing screen with nematodes 37 .
Pyrvinium pamoate, a quinolone-derived cyanine dye with a quaternary ammonium centre, used for treatment of pinworms, was an ARB for various drug and species combinations. It was most effective in E. coli where it was an ARB for GEN and MEM at 7 M, also being an effective ARB with CST in A. baumannii (7 M) and one of the few compounds to show effects in P. aeruginosa where it weakly potentiated GEN at 20 M. Previous studies have described direct activity against MRSA 38 and C. albicans 2 and synergy with azoles against yeasts 39 .
Auranofin, an antirheumatic agent, displayed ARB activity with MEM and CIP in E. coli and K. pneumoniae, and GEN and CST in K. pneumoniae. It was also directly antimicrobial against A. baumannii, K. pneumoniae and P. aeruginosa. Auranofin, a gold containing compound with antiinflammatory properties, has been widely described as either a direct-acting antimicrobial against C. albicans biofilms 6 , as an antiparasitic in phase I clinical trials against MRSA 7 and K. pneumoniae 8 and against both planktonic and biofilm cultures of P. aeruginosa (PAO1) in combination with CST 3 .

Conclusions.
Aspects of the design of the current study, particularly the screen for ARB activity with a range of MDR Gram-negative species, was novel and produced some possible starting points for future drug discovery. The study reiterated the potential value of nucleoside/nucleotide analogues, anthracycline derivatives and psychoactive drugs as potential series which might merit additional investigation. There have already been a number of studies looking at either individual members of these classes as synergising agents (e.g. ziduvidine 19 ) or more systematically 33,40,41 .
The study also highlighted the challenges of extrapolating from single species/strains used in HTS to general utility in the species of interest. While fluspirilene and oxethazaine showed clear ARB activity in the original screen, the CST-resistant K. pneumoniae isolate was not representative of the majority of other strains tested in extended studies, with only three isolates showing ARB effects and only with fluspirilene. This certainly asks interesting questions about the allele-specific nature of potentiation and its possible mechanisms, but it also highlights the pitfalls of such potentiation studies which are frequently reported in the literature with only a single or small number of closely related strains. The ARB effects seen with fluspirilene in the other CST-resistant Gram-negative species are encouraging, particularly with E. coli, where the MIC is reduced to below the breakpoint, but are again based on a very small sample size. Table S1: MICs (mg/L) and relevant resistance genes for each organism/antibiotic combination in the high throughput combination screen. S = MIC below EUCAST breakpoints and therefore not relevant in this screen. (MEM; meropenem, CIP; ciprofloxacin, GEN; gentamicin, TGC; tigecycline, CST; colistin) EUCAST breakpoints were accessed by the following link (http://www.eucast.org/clinical_breakpoints/; accessed 3 May 2019).   (3) 13 (8) 12 (6) 14 (7) 6 (4) 9 (7) 6 (4) 11 (     For key, see Figure S1.