Combinatorial libraries as a tool for the discovery of novel, broad-spectrum antibacterial agents targeting the ESKAPE pathogens

Abstract

Mixture based synthetic combinatorial libraries offer a tremendous enhancement for the rate of drug discovery, allowing the activity of millions of compounds to be assessed through the testing of exponentially fewer samples. In this study we used a scaffold-ranking library to screen 37 different libraries for antibacterial activity against the ESKAPE pathogens. Each library contained between 10,000 and 750,000 structural analogs for a total of >6 million compounds. From this we identified a bis-cyclic guanidine library that displayed strong antibacterial activity. A positional scanning library for these compounds was developed and used to identify the most effective functional groups at each variant position. Individual compounds were synthesized that were broadly active against all ESKAPE organisms at concentrations <2μM. In addition, these compounds were bactericidal, had anti-biofilm effects, showed limited potential for the development of resistance, and displayed almost no toxicity when tested against human lung cells and erythrocytes. Using a murine model of peritonitis, we also demonstrate that these agents are highly efficacious in vivo.

Graphical Abstract

Introduction

Nosocomial infections are a significant cause of human morbidity and mortality. In the United States alone there are 2 million such infections every year caused by bacterial pathogens, leading to approximately 100,000 deaths.¹ These infections are a significant public health concern as they are typically caused by broadly multidrug resistant organisms, which have become virtually unmanageable with existing antibacterial chemotherapeutics.² It is thus no surprise that the World Health Organization has identified antimicrobial resistance as one of the three greatest threats to mankind in the 21^st century.³ In light of this, the Infectious Disease Society of America (IDSA) coined the term ESKAPE pathogens almost a decade ago, referring to the six bacterial species that collectively cause around two-thirds of all US nosocomial infections, and have effectively escaped the ability to be treated by existing drugs.⁴ These bacteria are: Enterococcus faecium, Staphylococcus aureus, Klebsiella pneumoniae, Acinetobacter baumannii, Pseudomonas aeruginosa, and Enterobacter species.

In spite of the rapid and continued emergence of drug resistant ESKAPE pathogen isolates, there has been an alarming decline in drug discovery efforts in the pharmaceutical industry; resulting in a 75% reduction in FDA approval of antibacterial agents from 1983–2007.⁴ For three of the Gram-negative ESKAPE organisms (K. pneumoniae, A. baumannii and P. aeruginosa), a post antibiotic era has effectively been realized, with pan-resistant isolates identified on numerous occasions over the last decade.⁵ As a result, the IDSA recently issued a call to action, indicating the urgent necessity of developing sustainable antibacterial research and development that responds to current resistance trends, and anticipates the development of resistance in the future.⁴

The use of positional scanning libraries (PSL) provides a fundamental shift in the drug discovery processes for diseases. These libraries allow the evaluation of thousands to millions of synthetic compounds through the use of exponentially fewer test samples, compared to traditional approaches of screening individual agents against a given target. PSLs contain diverse chemical structures, and large numbers of compounds in each library, which increases the rate of identifying compounds with useful chemical characteristics.⁶ Such approaches expedite the screening process, allowing for rapid selection of specific PSLs that can be deconvolved, generating strong and detailed SAR data due to high structural density of the libraries.⁷ Furthermore, it has been demonstrated through chemoinformatic approaches that PSLs have the ability to expand currently known medicinal chemistry space.⁸ Taking all these considerations together, the use of such libraries can rapidly enhance the drug discovery process, which is of significant benefit in trying to keep pace with increasing rates of antibacterial resistance.

In this study, we began with a scaffold ranking library containing 37 different combinatorial libraries composed of 10,000 to 750,000 compounds each, leading to a total of >6 million compounds tested. From this initial screening, a PSL based around a core bis-cyclic guanidine scaffold was selected for further evaluation. The library contained 45,864 different bis-cyclic guanidines systematically formatted into 110 mixture samples. By screening this library, we identified a series of individual bis-cyclic guanidine compounds that have strong antibacterial activity against both Gram-positive and Gram-negative organisms.

Guanidine based compounds have been found to possess extensive functional bioactivities.⁹ In the late 1960s, guanidine hydrochloride was used to treat Clostridium botulinum infections as it was found to block presynaptic potassium channels, which stimulated neurotransmitter release, and alleviated toxin mediated paralysis.¹⁰ Guanidine related compounds like bisbiguanides (such as chlorhexidine) and bisamidines (such as hexamidine) have shown broad spectrum activity against both Gram-negative and Gram-positive pathogens and are commonly used as antiseptics. Recently, the attachment of guanidinium groups to both neomycin B and kanamycin A has been shown to restore and extend their activity towards gentamicin-resistant Pseudomonas aeruginosa and methicillin-resistant Staphylococcus aureus.¹¹ Finally, a highly novel antimicrobial agent (teixobactin) containing a cyclic guanidine substituent was recently discovered, demonstrating excellent activity against Gram-positive organisms, and displaying essentially no propensity for the development of resistance.¹²

More relevant to this study, in work by Rideout et al.,¹³ pyrrolidine bis-cyclic guanidines were identified, with antibacterial activity towards Gram-positive and Gram-negative species. These agents were found to interfere with DNA replication, and induce envelope stress in target organisms. Interestingly, the bis-cyclic guanidine library tested herein was also screened in these holiday junction accumulation assays, and found to bind these structures as well, although at a much lower frequency than the their pyrrolidine-based counterparts. Further to this, the pyrrolidine bis-cyclic guanidines were actually part of the combinatorial library screened in the present study (library 1955); however the simpler bis-cyclic guanidine scaffold identified herein possessed a broader spectrum of activity at more promising concentrations.

As such, in this study we have identified a novel series of bis-cyclic guanidine compounds that have broad activity against all of the ESKAPE pathogens, limited toxicity to human cells, a strong ability to eradicate bacterial biofilms, and show promising efficacy in mammalian models of infection. We contend that employing positional scanning approaches, and the accompanying strategies described herein, create a fundamental shift away from traditional antibacterial testing methodologies, by introducing a rapid approach to discover novel compounds that possess broad spectrum activity.

Results and Discussion

Scaffold Ranking Library.

In order to rapidly assess the available chemical scaffolds in our combinatorial collection for their potential broad-spectrum antibacterial activity, a scaffold ranking library approach was utilized. We have previously described in detail the construction,^{14 6} advantages and limitations of the scaffold ranking library,⁶ as well as its successful implementation for the discovery of several classes of novel ligands for a range of targets and indications;^{15 16 17} including antimicrobials that inhibit tyrosine recombinases and Holliday junction-resolving enzymes.^{13a 18} In the current project we utilized a scaffold ranking library containing 37 mixture samples, each of which was comprised of approximately equal molar concentrations of individual compounds containing the same common core scaffold (Supplemental Table S1). The 37 mixtures were screened for antimicrobial activity against all six ESKAPE pathogens using a microbroth dilution assay. From the initial scaffold ranking data (Figure 1) we determined that the most potent broad spectrum library was 2157. This sample (Figure 1) effectively inhibited E. faecium, S. aureus, A. baumannii, P. aeruginosa, and E. cloacae at 100 μM. More importantly the sample retained broad spectrum activity at 5 μM, where it inhibited E. faecium, S. aureus, A. baumannii, and E. cloacae. Two other samples, 2161 and 1952 (both polyamines) were active against all six ESKAPE pathogens at 100 μM, but did not retain broad spectrum activity at lower concentrations, which led to a less significant stacked scale score. In general scaffolds containing cyclic guanidines, piperazines, and polyamines were amongst the most active scaffolds (see Supplementary Table 1 for list of core scaffolds). However the broad antimicrobial activity, even at low concentrations, led us to further investigate the 2157 positional scanning library.

Figure 1. Screening the scaffold ranking library for antibacterial activity against the ESKAPE pathogens.

Compound mixtures were assayed against the ESKAPE pathogens using a micro broth dilution assay. Data is presented as a stacked scaled score, which is determined by dividing 100 μM (the maximum concentration tested) by the individual doses tested. Each library is given a scaled score for each pathogen, and these are then stacked to determine the library with the broadest activity, at the lowest concentration.

Deconvolution of the 2157 library.

Library 2157 is a positional scanning library containing 45,864 individual bis-cyclic guanidines (Scheme 1 and Figure 2) systematically synthesized into 110 mixture samples (Supplemental Table S2). These separate 110 mixtures were next screened against the ESKAPE pathogens to deconvolve specific antibacterial activity, and begin to generate a structure activity relationship. The first 42 of these 110 samples contain the 45,864 bis-cyclic guanidines arranged by fixing the R₁ position (Figure 2A, Supplemental Table 2); the next 26 samples are arranged by R₂ position (Figure 2B); and the last 42 samples are arranged by R₃ (Figure 2C). By way of example the first sample in Figure 2A contains an equal molar amount of the 1,092 individual compounds in the library that have phenethyl fixed at the R₁ position; likewise the last sample in Figure 2C contains an equal molar amount of the 1,092 individual compounds in the library that have adamantan-1-yl-ethyl fixed at the R₃ position.

Scheme 1. Synthetic scheme of bis-cyclic guanidines.

a) 5% DIEA/DCM; b) Fmoc-Lys(Boc)-OH, DIC, HOBt, DMF; c) 20% Piperidine/DMF; d) R1COOH, DIC, HOBt, DMF; e) 55% TFA/DCM; f) Boc-AA(R2), DIC, HOBt, DMF; g) R3COOH, DIC, HOBt, DMF; h) BH3-THF, 65oC, 96 hours; i) Piperidine, 65°C, 24 hours; j) CNBr, DCM; k) HF, anisole, 0°C

Figure 2. Deconvolving the antibacterial activity of the bis-cyclic guanidine library.

The bis-cyclic guanidines were systematically synthesized into a positional scanning library containing 110 samples (shown in supplemental table S2). These were fixed at: A = the R1 (42 samples); B = R2 (26 samples); or C = R3 (42 samples) position. For example, the first sample in A, is an approximate equal molar mixture of 1,092 compounds. The 1,092 compounds contain hydrogen fixed in the R1 position and all 1,092 combinations of the 26 R2 and 42 R3 functionalities. Similarly, the first sample in B is 1,764 compounds generated from fixing R2 with S-methyl and utilizing all 1,764 combinations of the 42 R1 and 42 R3 functionalities. The height for each color of individual bars is determined by dividing 100 μM (the maximum concentration tested) by the individual MIC for each agent. Libraries are then given a scaled score for each pathogen, and these are stacked to determine the library with the broadest activity, at the lowest concentration.

The 110 samples from Library 2157 were screened for antimicrobial activity against all six ESKAPE pathogens in a similar manner to the Scaffold Ranking Library, generating MIC data for each sample (Figure 2, Supplemental Table 3). From this we determined a clear differentiation in the potency of mixtures. For example, those fixed with large aromatic or aliphatic substitutions, such as 2-(3-trifluoromethyl-phenyl)-ethyl and adamantan-1-yl-ethyl, respectively, at R₁, were more potent than any of the mixtures fixed with small aliphatic groups, such as butyl and isobutyl. However, we noted that samples fixed at the R₂ position with different butyl functionalities are actually amongst the most potent, although as the butyl group is shortened to a propyl and then a methyl, there appears to be step-wise reduction in potency. Additionally there is no apparent preference for absolute configuration at this position. For the R3 position a number of samples with aliphatic (cyclic and acyclic) and aromatic functionalities fixed at the R3 position show activity, however there were a few trends that seem to affect activity at this position such as the size of the aliphatic group (larger favored) as well as the preference for aromatic electron withdrawing groups over electron donating groups. For example changing from a heptyl, six carbon chain functionality, to a butyl, four carbon chain in R3 effectively eliminates activity of the sample; and switching from a weak meta-electron donating group such as 2-(3-fluoro-phenyl)-ethyl or 2-(3-bromo-phenyl)-ethyl to a strong meta-donating group such as 2-(3-methoxy-phenyl)-ethyl at the R3 position reduced the overall activity of the sample.

From this data we could have chosen a number of different functionalities (active samples) to fix at each of the positions; however in order to reduce the number of compounds produced, we selected 27 individual compounds for synthesis. These compounds were selected by combining the functionalities of the most potent mixtures from each of the R positions, while biasing to include as much structural diversity as possible (Supplemental Table S4 Samples 1–27).

Screening of Individual Compounds.

The 27 individual compounds were synthesized (Scheme 1, Figure 3) and screened for antimicrobial activity against all six ESKAPE pathogens, again using MIC assays, and are reported using stacked scaled scores (Figure 3, 1–27). There are several SAR trends of note. Using a S-cyclohexylmethyl to a S-butyl (ie 7 vs 4, 9 vs 6, and 17 vs 14) corresponds with a decrease in total activity score and the decrease in total activity score for each pair is most significantly driven by a decrease in activity against Pseudomonas aeruginosa. Additionally when R2 was fixed with either S-butyl or S-cyclohexylmethyl there was a stepwise decrease in total activity when R1 was fixed with 2-(3-trifluormethyl-phenyl)-ethyl versus cyclohexyl-butyl versus adamantan-1-yl-ethyl (ie 4 vs. 5 vs. 6). When R2 was fixed with R-2-napthylmethyl the trend was slightly different, there was a stepwise decrease in total activity when R1 was fixed with cyclohexyl-butyl versus 2-(3-trifluormethyl-phenyl)-ethyl versus adamantan-1-yl-ethyl (ie 2 vs. 1 vs. 3). Finally, in general substituting the R3 2-biphenyl-4-yl-ethyl group with either a cyclohexyl-buty (1–9 vs 10–18) or a heptyl (1–9 vs 19–27) resulted in a reduction in total activity and this reduction in activity was not necessarily bacteria specific but rather attributed to a slight loss in activity against several bacteria.

Figure 3. Assessing the antibacterial activity of individual bis-cyclic guanidines synthesized based on library SAR data.

Fifty-four individual compounds were synthesized for testing against the ESKAPE pathogens. 1–27 were generated based on SAR data from ESKAPE testing with the combinatorial libraries; 28–54 were included as they were predicted to be significantly less active based on PSL data. Data is presented as stacked, scaled scores, with the height for each color of individual bars determined by dividing 100 μM (the maximum concentration tested) by the individual MIC for each agent. Compounds are then given a scaled score for each pathogen, and these are then stacked to determine which have the broadest activity, at the lowest concentration. Note data is generated using “crude” compounds (see Materials and Methods Section for details).

Additionally, a separate set of 27 structural analogs predicted not to be potent based on the SAR of the positional scanning library data were tested to verify as much, and that the SAR generated holds value for predicting potent inhibitors (Supplemental Table S4a and Figure 3, 28–54). Although these additional 27 compounds (28–54) are close structural analogs to the 27 compounds selected for synthesis in the ESKAPE project (1–27); based on the screening of library 2157 the additional compounds were predicted to be significantly less active towards the ESKAPE pathogens. We included these compounds to validate that the activity observed is being driven by the correct combination of functionalities around the core bis-cyclic guanidine scaffold, and not just generally by any compound from this library. The 27 novel compounds synthesized for the ESKAPE project displayed an increase in broad spectrum antibacterial activity at low concentrations. At a concentration of 45 μM, 25 of the 27 compounds inhibited growth of all six ESKAPE organisms, with 14 of these retaining activity against all organisms when the concentration decreased to 10 μM. Even more promising, 5 of the individual compounds tested (1, 2, 7, 16 and 19; Figure 4) had antibacterial activity against all 6 species at concentrations < 2 μM (Table 1). Conversely, and as expected, the 27 additional compounds (28–54) displayed almost no activity towards the ESKAPE pathogens (Figure 3, 28–54), further validating our structure-guided design of individual compounds.

Figure 4.

Lead bis-cyclic guanidine compounds.

Table 1.

Antimicrobial activity of front-runner bis-cyclic guanidines.

CMPD	MIC (μM)						MBC90 (μM)						MBEC50 (μM)						IC50 (μM)	AI
	E	S	K	A	P	E	E	S	K	A	P	E	E	S	K	A	P	E	A549	-
1	1.6	1.6	1.6	1.6	1.6	1.6	1.9	3.1	2.4	3.1	4.0	3.1	39	14	7.0	15	20	13	164	100
2	1.5	1.5	1.5	1.5	1.5	1.5	1.7	2.0	4.0	2.8	2.3	4.1	29	2.2	4.7	16	31	13	125	82
7	1.5	1.5	1.5	1.5	1.5	1.5	3.3	3.3	2.2	3.2	14	6.4	26	4.4	4.8	13	46	15	66	43
16	1.6	1.6	1.6	1.6	1.6	1.6	2.9	3.9	4.5	2.8	4.0	9.6	28	7.6	6.3	14	34	13	> 225	> 139
19	1.7	1.7	1.7	1.7	1.7	1.7	3.6	2.9	2.6	4.7	2.8	2.7	13	6.4	8.6	24	5.0	15	146	88

The in vitro antibacterial and cytotoxic properties of the lead bis-cyclic guanidines were assessed. Shown are the antibacterial activity (MIC), the bactericidal capacity (MBC₉₀), anti-biofilm properties (MBEC₅₀), and toxicity towards human A549 cells (IC₅₀). Selectivity windows were also determined in the form of an Activity Index (IC₅₀ / MIC): note that only 1 value is given because all compounds have the same MIC against each of the ESKAPE pathogens. All data in table generated with purified compounds.

Preliminary SAR based on rapid deconvolution and potential activity cliffs for the individual compounds.

As previously described, the 27 compounds (1–27) were selected based on SAR information inherently contained in positional scanning library 2157 as the individual compounds most likely to be active. While compounds 28–54, having the same scaffold as all the compounds in 2157, are close structural analogs to compounds 1–27, they clearly possess little-to-no activity when compared to compounds 1–27. An exploration of compound structures, in order to identify parameters that distinguished these two sets of analogs, is therefore warranted. As a first step, for each of the 54 compounds Canvas¹⁹ was used to generate six physicochemical properties commonly used to characterize and compare compound data sets in drug discovery^{20 19}: molecular weight (MW), polar surface area (PSA), AlogP (logP as calculated by Canvas), number of rotatable bonds (RB), number of hydrogen bond acceptors and donors (HBA and HBD respectively). A list of all computed values for each compound can be found in Supplemental Table S4; the average and standard deviation for each of the six properties for the two sets, as well as the subset of 5 leads (1, 2, 7, 16, 19), is shown in Table 2. From these data it is evident that the average MW, AlogP, and RB for the two sets is markedly different, with the active group (1–27) having a higher average MW, AlogP and number of RB than the inactive set. However, it is important to note the five lead compounds (1, 2, 7, 16, and 19) had, on average, slightly lower MW and AlogP values, making the most potent compounds more similar (based on properties) to the inactive group than the other members of the active group were. Thus it is clear that size, lipophilicity, and flexibility do not fully capture the SAR of these data sets, even though the groupings are grossly categorized by these properties (Table 2), there are numerous examples of compounds with similar physicochemical properties having large activity differences, e.g. activity cliffs with respect to property similarity.²¹ In order to investigate the potential presence of activity cliffs in a systematic manner, we used Structure-Activity Similarity (SAS) Maps,^{22 7 20} which were one of the first methods developed to characterize SARs by using the concept of activity landscape modeling. The impact of activity landscape modeling and activity cliffs in medicinal chemistry has been extensively discussed as well as the principles of the computational approaches. The interested reader is referred to recent reviews of this topic.^{23 24 25} SAS maps systematically compare a given representation of molecular similarity with activity similarities for all possible combinations of compounds in a data set.²⁶ Because the SAR observed with any dataset is highly dependent on the molecular representation used,²⁷ different structural similarity methods can have drastically different behavior with regards to activity cliffs. Using SAS maps, each of the 1,431 non-redundant pairs of the 54 compounds in the series was evaluated for potency, similarity and relative molecular similarity (Figure 5). The left panel of Figure 5 shows a SAS map with molecular similarity computed using the six physicochemical properties (following a method we have previously described,²⁷ and summarized in the Materials and Methods) on the x-axis and potency similarity on the y-axis. The data points in the lower right hand quadrant of this plot indicate pairs of compounds with high similarity in the six physicochemical properties used, but low activity similarity (i.e. large differences in potency). Such points thus represent activity cliffs. Unsurprisingly, it is clear from the large number of similar points in the plot that property differences alone are not sufficient to explain the activity differences in these 54 compounds. Two example pairs are highlighted in figure 5: pair 19 (an active compound) and 35 (a relatively inactive compound) are shown by open blue circles, whilst pair 2 (an active compound) and 32 (a relatively inactive compound) are shown by open black circles. The right panel shows a SAS map where the molecular similarity was computed using a different molecular representation: radial fingerprints. Radial fingerprints entail growing a set of fragments radially from each heavy atom over a series of iterations,²⁸ and are equivalent to extended connectivity fingerprints (ECFPs).²⁹ In sharp contrast to the SAS map obtained with physicochemical properties, the SAS map generated with radial fingerprints does not show activity cliffs. For example, the two pairs of compounds 19–35 and 2–32 are now appropriately located in the lower left quadrant of the SAS map (right panel Figure 5). Notably, we recently reported the superior performance of radial fingerprints over other fingerprint-based methods for activity landscape studies.⁶ We would like to emphasize that, despite the fact that a comprehensive description of the activity landscape study can be performed with the 54 compounds (using SAS maps and/or other approach), the discussion of the two SAS maps presented in Figure 5 is focused on showing the relevance of changing the representation to explore the SAR, from continuous physicochemical properties (SAS map on the left), the atom connectivity (SAS map on the right). By no means, the distribution of the data points in the SAS maps are meant to suggest specific substitutions to improve potency. However, the fact that it successfully explains the structure-activity relationship extant in this data set means this fingerprint method will be a useful tool in the ongoing more intensive exploration of the SAR associated with the lead compounds.

Table 2.

Physicochemical properties of individual bis-cyclic guanidines.

Set	MW	AlogP	RB	HBA	HBD	PSA
1–27	618.24 ± 50.99	8.72 ± 0.66	16.33 ± 1.27	6 ± 0	3± 0	69.45 ± 0
28–54	421.32 ± 55.46	4.78 ± 1.35	11.67 ± 2.59	6 ± 0	3.3 ± 0.5	72.38 ± 4.22
Lead 5	566.03 ± 23.33	8.00 ± 0.62	17.40 ± 0.55	6 ± 0	3 ± 0	69.45 ± 0

Shown are data for the active set (1–27), inactive set (28–54), and front-runner compounds (lead 5; 1, 2, 7, 16, and 19). Molecular Weight (MW), number of rotatable bonds (RB), number of hydrogen bond acceptors (HBA) and donors (HBD), polar surface area (PSA).

Figure 5. Computational Exploration of Physicochemical Properties.

Each of the 54 compounds (1–54) are compared against each of the remaining 53 compounds for differences in potency (Y-axis both left and right panel) and molecular representation (Physicochemical Properties: X-axis left panel; Radial: X-axis right panel). Each pair is represented by a dot. In this way a pair of compounds with similar activity potencies and physicochemical properties will be shown by a dot in the upper right hand quadrant of the left panel. The dots are colored by activity of the most potent compound in a pair, using a continuous color of: Grey (no activity), Yellow (low activity), Orange (moderate activity), and Red (high activity). Shown below the panels are structures for two such pairs. The pair in the left location on both panels (19–35) is identified by open blue circles, whilst the pair in the right location (2–32) is indicated by open black circles. Under each structure is the total activity value used for each compound, as well as the three physicochemical values (MW, AlogP, and RB) associated with a given agent.

The disparity between the SAS maps is a strong argument for the exploration of dense portions of the chemical space; here, representative compounds based on physicochemical properties would have been ill-equipped to properly characterize the active compounds found. Indeed, because of the high structural density of positional scanning libraries,²⁰ they are inherently very rich in SAR information, and well suited to assess the presence of activity cliffs. ^{10 21 30} The SAR information from the positional scanning libraries was thus not only able to communicate that in general the active compounds (1–27) are more hydrophobic and have more rotatable bonds than the inactive compounds (28–54), but was also able to capture the specific atom connectivity, as described by radial fingerprints, that plays a key role in the activity of the molecules.

Exploring the antibacterial activity of frontrunner agents using a library of ESKAPE pathogen isolates.

Thus far, all data was derived using individual, albeit highly drug resistant, isolates. To assess the full antibacterial potency of front runner agents, additional data was collected using a panel of clinical ESKAPE isolates (Supplemental Table 5–6). We determined that all Gram positive strains (E. faecium and S. aureus), as well as those isolates of the Gram negative organism A. baumannii, were sensitive to frontrunner agents at 2 μM, with absolutely no variation. Furthermore, the growth of 95% of all isolates (regardless of species) was inhibited by the five lead bis-cyclic guanidines at concentrations of ≤ 10 μM. K. pneumoniae and E. cloacae strains displayed slight variation in MIC values, with agents 2 and 19 inhibiting the growth of 90% of isolates for both species at 5 μM, and 70% of strains at 2 μM. Lead agent 16 had an MIC of 2 μM against 90% of E. cloacae strains, and 5 μM against 90% of K. pneumoniae strains. Lead agent 7, was found to be similar to 16 in activity towards K. pneumonia, inhibiting 90% of strains at 5 μM, and 90% of E. cloacae strains at 2 μM. Compound 1 had the most variation in MIC against K. pneumoniae and E. cloacae, with growth of 90% of clinical isolates for each pathogen inhibited at 10 μM. Against the P. aeruginosa panel of strains, the lead bis-cyclic guanidines had slightly higher MIC values. Lead agents 2 and 16 inhibited 90% of isolates at 5 μM, whilst agents 1, 7, and 19 inhibited 90% at 10 μM.

Given the minor variations observed in MIC for three of the Gram-negative organisms, and that these particular bacteria are renowned for efflux detoxification of antibacterial agents, we hypothesized that the differences observed likely relate to variation in efflux pump activity between strains. To test this contention, we reanalyzed MICs for all five front runner agents against our full panel of K. pneumoniae, P. aeruginosa or E. cloacae isolates in the presence of sub-inhibitory concentrations of the known efflux inhibitor, reserpine.³¹ We determined that, whilst 100 μM of reserpine or 2 μM of compound 1 individually had no effect on the growth of P. aeruginosa isolate 1420, the two combined strongly inhibited growth of this strain (Supplemental Figure 1A). Similarly, when using P. aeruginosa strain 1414 and frontrunner 19, we observed complete inhibition of growth when this agent was paired with reserpine (Supplemental Figure 1B). The effects observed appear to be universal, regardless of strain or compound tested. For example, K. pneumoniae strain 1441, when used with compound 16 (Supplemental Figure 1C), or E. cloacae strain 1446 when tested with compound 7 (Supplemental Figure 1D), resulted in complete inhibition of growth in combination with 100 μM of reserpine. It should be noted that the data presented herein represents a worst-case scenario. For example, compound 1 was the least active of any agent against P. aeruginosa strain 1420; the same is true for all other pairings presented. Similar data was returned for all front runner agents, against all 10 isolates of the three Gram-negative organisms (data not shown). These findings support the hypothesis that inherent efflux mechanisms of certain Gram-negative organisms result in MIC variations for the lead bis-cyclic guanidines between clinical isolates. As such, we suggest that any minor decrease in activity for these agents can be restored by the use of a known efflux pump inhibitor.

Assessing bactericidal characteristics.

We next set out to perform a thorough in vitro and in vivo characterization of these five lead agents, to assess their antimicrobial activities. To do this, we first used a minimal bactericidal concentration (MBC) assay to distinguish whether these compounds were bactericidal or bacteriostatic in nature. Upon analysis, all compounds were found to be bactericidal at concentrations close to their MICs (Table 1). Compounds 1 and 2 proved to be the most bactericidal, with the former agent having MBC₉₀ values ranging from 1.9 μM (against E. faecium) to 4.0 μM (against P. aeruginosa); whilst the latter had MBC₉₀ values ranging from 1.7 μM (against E. faecium) to 4.2 μM (against E. cloacae). Similarly, compound 19 was also strongly bactericidal, with MBC₉₀ values ranging from 2.6 μM (against K. pneumoniae) to 4.7 μM (against A. baumannii). For the most part, compound 16 was significantly bactericidal in effects, with MBC₉₀ values ranging from 2.8 μM (against A. baumannii) to 4.5 μM (against K. pneumoniae); however it’s MBC₉₀ against E. cloacae was slightly higher at 9.6 μM. Finally, compound 7 was strongly bactericidal against the first four ESKAPE pathogens, with MBC₉₀ values ranging from 2.2 μM to 3.3 μM; however this number rose to 6.4 μM against E. cloacae, and 14 μM against P. aeruginosa. As such, all compounds displayed effective bactericidal activity, with many proving so even at very low concentrations.

Considering the strong bactericidal nature of the bis-cyclic guanidines, we next assessed the ability of front runner agents to lyse bacterial cells.As such, a time kill assay was performed using all five lead agents against exponentially growing MRSA cells (Figure 6a). Alongside we also used positive controls agents, including sodium dodecyl sulfate (SDS), Lysostaphin (Lys) (a S. aureus specific lytic agent), benzalkonium chloride (BA) and benzethonium chloride (BC). These latter two agents are cationic detergents, and were included because the bis-cyclic guanidines have the potential to be cationic in nature at physiological pH. For our lead agents at MIC we observed limited change in bacterial density over the 2h period. Such findings were similar to our negative control, doxycycline (a translation inhibitor). By way of comparison, we recovered only 52% of cells upon exposure SDS. More profoundly, we achieved >50% lysis of MRSA cells within only 10 minutes of exposure to the positive control agent lysostaphin, with viability continuing to decrease over time. Finally, both cationic detergents proved highly lytic towards MRSA, with only 29% and 33% of cells surviving exposure to benzalkonium chloride or benzethonium chloride, respectively. At 120 minutes, cultures from these tests were serial diluted and cell viability assessed by CFU mL⁻¹. The bacteriostatic control antibiotic doxycycline displayed a 92% recovery of cells once the antibiotic was washed out. Incubation with the lytic control agents (BA, BC, SDS, and Lys) resulted in 0% cell recovery after the 120 minute assay. With regards to the bis-cyclic guanidines, we observed a 2.5-log reduction in viability after the 2h period (0.2% recovery). As such, it would appear that although our front runner compounds result in significant bacterial death during initial incubation, this is not the result of bacterial lysis. Accordingly, these data effectively demonstrate that although the bis-cyclic guanidines are strongly bactericidal, their mode of action does not appear to be via bacterial cell lysis; unlike that of simple cationic detergents.

Figure 6. Bis-cyclic guanidines are bactericidal but not bacteriolytic.

A. Time kill studies were performed using MRSA and the front runner agents (at MIC concentrations), alongside positive (4 μM lysostaphin, 0.001% Benzalkonium chloride (BA), 0.001% Benzethonium chloride (BC), and 2.0% Sodium dodecyl sulfate (SDS)), and negative (200 μM Doxycyline (Doxy)) control agents. Shown is the optical density of cells relative to starting values from three independent experiments. Error bars are shown ±SEM. B: Cell viability of all samples after the 120 min experiment. Compounds were removed by centrifugation and washing of cells, followed by serial dilution and enumeration. Percent recovery was determined by comparison to no drug (ND) controls.

Determining the antibiofilm capacity of lead agents.

Biofilm formation is a common feature for all of the ESKAPE pathogens, and has profound influence of disease severity and mortality.³² Biofilms form on implanted devices, as well as on bone, and in the heart, and are innately resistant to antimicrobial intervention.³³ As such, we next set out to assess whether our front runner compounds displayed antibacterial activity. These were performed using minimum biofilm eradication assays (MBEC), as described previously.³⁴ Whilst the MBEC₅₀ values for the lead compounds were found to be in excess of MIC and MBC₉₀ data, we did observe some highly promising anti-biofilm effects with each agent (Table 1). Compound 19 proved to be our most effective in this regard, having MBEC₅₀ values ≤ 8.6 μM against S. aureus, K. pneumoniae and P. aeruginosa, and between 13 μM and 24 μM for the remaining organisms. For 16, we determined MBEC₅₀ values of 6.3 μM, and 7.6 μM for K. pneumoniae and S. aureus, respectively; and 13 μM to 34 μM against the other four species. The remaining three agents (1, 2 and 7) had MBEC₅₀ values that were typically higher than this; however, 2 and 7 were strongly active against K. pneumoniae (4.7 and 4.8 μM), and S. aureus (2.2 and 4.4 μM) biofilms. As such, it appears that bis-cyclic guanidines not only have strong potential as broad spectrum antibacterial agents, but also have the capacity to limit biofilms formed by each of the ESKAPE pathogens.

Exploring the potential for front-runner toxicity towards human cells.

Ensuring selectivity for prokaryotic over eukaryotic cells is of primary importance during the development of antimicrobial agents. As such, we next performed cytotoxicity testing for the five lead bis-cyclic guanidines using human A549 adenocarcinomic alveolar basal epithelial cells. The screening of these five lead agents revealed remarkably low toxicity (Figure 7 and Table 1). Specifically, compounds 2 (Figure 7b) and 16 (Figure 7d) allowed for > 65% recovery at concentrations up to 100 μM, and > 50% at 225 μM. Compound 1 (Figure 7a) and 19 (Figure 7e) allowed for > 63% and 53% recovery at concentrations up to 125 μM respectively, with a slight decline to around 40% and 35% at 225 μM. Compound 7 (Figure 7c) yielded less favorable results, but still displayed limited toxicity, with > 60% recovery at 45 μM; a concentration that is 25 × the MIC. After this concentration, A549 recovery was consistently ≥ 33% at concentrations up to 225 μM. The cytoxicity data was used to determine IC₅₀ values where possible, as well as Activity Indices (AI = IC₅₀ / MIC), to gain a sense of therapeutic window and selectivity (Table 1). Importantly, compound 16 never resulted in 50% human cell toxicity, meaning that it has an AI value far in excess of 139. For compounds 1, 19, and 2, we obtained IC₅₀s of 164 μM, 146 μM, and 125 μM, which resulted in selectivity windows of AI = 100, AI = 88, and AI = 82, respectively. Finally, even compound 7, which had slightly more toxic effects, had an IC₅₀ of 66 μM and an AI = 43. As such, each of our front-runner compounds appears to have excellent specificity for bacterial cells over their eukaryotic counterparts.

Figure 7. Cytotoxicity of lead agents.

Shown is the survival of A549 cells measured using an MTT assay with all five lead agents (A-E). Data is presented as percent recovery compared to vehicle only controls. Error bars are shown ±SEM, from at least three independent experiments; MICs are denoted by grey coloring. A solid black line is shown for IC50 value determination. Hemolytic capacity towards human erythrocytes was also measured using the lead agents (F). Data is shown as percent hemolysis compared to positive (1% Triton-X100 (T), 100% hemolysis) controls. Lead agents were added at a concentration of 10 μM. Error bars are shown ±SEM, from at least three independent experiments. A solid black line is shown at 1% hemolysis.

To ensure that these findings were not specific to the cell line used, we next assessed the tendency of lead bis-cyclic guanidines to lyse human red blood cells (hRBCs). In agreement with data from A549 cells, hemolysis assays reveal that bis-cyclic guanidine have little to no apparent toxicity towards human cells; demonstrating no effective capacity to lyse hRBCs (Figure 7f). Using all lead agents at 10 μM (>5 × MIC for each molecule) we observed hemolysis levels ranging from 0.34% to 0.5%, which clearly demonstrates that lead agents have limited ability to lyse red blood cells. This is placed in context when one compares these values to that of the positive control (Triton-X100, 100% hemolysis). The inability to lyse hRBCs in addition to the lack of toxicity towards A549 cells reveal a high selectivity of bis-cyclic guanidines towards bacterial cells over human counterparts, and therefore suggests that bis-cyclic guanidines have very strong potential for development as new antibacterial agents.

Exploring the potential for ESKAPE pathogen resistance to front runner bis-cyclic guanidines.

An important attribute of potential antimicrobial agents is that the development of resistance to their action is not readily attained. Thus, we determined the spontaneous mutation frequencies for each of our five frontrunner agents. Despite numerous attempts using agar containing compounds at concentrations ranging from 2–10 × MIC, we could not generate spontaneous mutants for any of the ESKAPE pathogens (> 1 × 10¹¹ CFU collectively tested for each organism). This is in good agreement with work by Rideout et al.,^13a and their study of agents chemically related to the bis-cyclic guanidines, where spontaneous mutants could also not be generated. In the absence of spontaneous mutants, we next performed stepwise resistance assays, by serially passage of ESKAPE organisms in liquid media over 8 separate cycles (1 per day). For each passage, the concentration of front-runner compound was increased two fold; alongside a control agent (E. faecium and A. baumannii = tetracycline; P. aeruginosa and E. cloacae = ciprofloxacin; S. aureus = vancomycin; K. pneumoniae = rifampin). Against S. aureus, lead compounds 1 and 2 displayed the smallest increase in MIC, with only a two-fold decrease in sensitivity observed (Figure 8). We also observed limited resistance for 7, where a four-fold increase in MIC was noted after 8 passages. Finally, 16 and 19 both led to a 16-fold reduction in susceptibility, which, whilst higher than our other compounds, was significantly less than that of the control, vancomycin. For this latter agent, we noted a continued doubling of the MIC for every passage up to 128 fold increase in MIC. The control agents for each of the other five pathogens behaved similarly, with continued doubling up to 128 fold of the original MIC. However, in each case, the bis-cyclic guanidines outperformed the existing, approved, control agents. Lead agents 2, 7, 16, and 19 were remarkably effective at limiting resistance development in the Gram negative species K. pneumoniae, A. baumannii, and E. cloacae. Testing with these agents revealed a sensitivity limit of ≤ 8-fold, with concentrations higher resulting in complete inhibition of bacterial growth. Against P. aeruginosa, lead agents 16 and 19 had the smallest increase in sensitivity at 8-fold, a promising observation for a pathogen known to readily develop resistance to antimicrobial agents. As such, there appears to be very limited potential for resistance to our front-runner agents, with no-spontaneous mutation seemingly apparent, and limited room for adaptive tolerance to their affects.

Figure 8. Exploring Adaptive Tolerance by ESKAPE Pathogens to Front Runner Agents.

ESKAPE pathogens were serially passaged for eight days in fresh liquid media (changed every 24h), with the concentration of compound increased 2-fold each day. Shown are the increases in MIC observed over time. Ef = E. faecium; Ec = E. cloacae.

Lead bis-cyclic guanidines are efficacious during in vivo infection.

As a final measure of the suitability of our lead compounds to serve as anti-bacterial agents, we studied the in vivo efficacy in mice. Using MRSA as a representative ESKAPE organism, we infected mice with 1 × 10⁸ bacterial cells in 5% mucin via intraperitoneal injection. At 1h post-infection, mice were then I.V. injected with either vancomycin (positive control), or I.M. with our front runner compounds. Each group of mice was compared to a negative control group receiving only vehicle (45% w/v (2-hydroxypropyl)-β-cyclodextrin in water). At 2xMIC for compound 1, all mice survived the 5 day infection period (Figure 9). Similarly, compounds 2, 16 and 19 also proved highly efficacious, with only a single mouse succumbing to infection after the first day, and the rest surviving through day 5. Finally, compound 7 was only marginally less effective, with 1 mouse lost on each of days 2 and 4, which still resulted in statistically significant protection compared to vehicle only controls. When using our control agent vancomycin, we observed 50% protectivity at 5xMIC (not significant), and 100% protectivity at 10 × MIC. Based on these encouraging results, we suggest that our lead bis-cyclic guanidines have excellent in vivo activity, even at very low doses.

Figure 9. Front runner bis-cyclic guanidines are efficacious during in vivo infection.

Mice were I.P. infected with a lethal dose of S. aureus. After 1h, they were then injected with either front-runner bis-cyclic guanidines (at 2 × MIC), vancomycin (positive control, at 5 × MIC and 10 × MIC) or vehicle alone (negative control). Mice were then monitored for five days, and the significance of mortality measured using a log rank and chi square test with 1-degree of freedom. * = p > 0.05, ** = p > 0.01.

Conclusions

The combinatorial scaffold libraries in this study allowed for the assessment of > 6 million compounds for antibacterial activity against the ESKAPE pathogens. The screening ultimately identified a bis-cyclic guanidine scaffold with broad spectrum activity towards each of these organisms. The utilization of a positional scanning library (PSL) was crucial in identifying the most effective functional groups at each of the three variant positions of the core scaffold. The PSL data guided synthesis of 27 individual compounds with significantly increased activity towards all 6 ESKAPE pathogens. The five most promising individual compounds were chosen as lead agents for further characterization of antibacterial activity (1, 2, 7, 16, and 19). These lead agents proved to be strongly bactericidal (but not bacteriolytic), had promising abilities to eradicate biofilms created by each of the ESKAPE pathogens, and demonstrated little capacity for the development of resistance. Moreover, the bis-cyclic guanidines proved to be highly selective towards bacteria, revealed by low toxicity towards human lung epithelial cells and erythrocytes. Finally, using a murine model of lethal peritonitis we observed in vivo efficacy of the bis-cyclic guanidines. Taken together we present the discovery of a novel class of bis-cyclic guanidines that have high specificity toward the ESKAPE pathogens in vitro and in vivo, and which display significant promise for development as antibacterial agents.

Materials and Methods

Synthesis of Library 2157 and Individual Compounds and Construction of Scaffold Ranking Plate:

General Synthesis of bis-cyclic guanidines (Scheme 1): Library 2157 as well as the individual compounds reported herein (1–54) were synthesized following the same synthetic scheme (Scheme 1).^{35 36 17} Utilizing the “tea-bag” methodology³⁷, 100 mg of p-methylbenzhydrylamine (MBHA) resin (1.1mmol/g, 100–200 mesh) was sealed in a mesh “tea-bag,” neutralized with 5% diisopropylethylamine (DIEA) in dichloromethane (DCM), and subsequently swelled with additional DCM washes. Fmoc-L-Lys(Boc)-OH was coupled in Dimethylformamide (0.1M DMF) for 120 minutes in the presence of Diisopropylcarbodiimide (DIC, 6 equiv.) and 1-Hydroxybenzotriazole hydrate (HOBt, 6 equiv.) (1, Scheme 1). The Fmoc protecting group was removed with 20% piperidine in DMF for 20 minutes and the R₁ carboxylic acids was coupled (10 equiv) in the presence of DIC (10 equiv) and HOBt (10 equiv) in DMF (0.1M) for 120 minutes (2, Scheme 1). The Boc protecting group was then removed with Trifluoroacetic Acid (TFA) in DCM for 30 minutes and subsequently neutralized with 5% DIEA/DCM (3 x). Boc-Amino Acids (R₂) were coupled utilizing standard coupling procedures (6 equiv.) with DIC (6 equiv.) and HOBt (6 equiv.) in DMF (0.1M) for 120 minutes. The Boc group was removed with 55% TFA/DCM for 30 minutes and subsequently neutralized with 5% DIEA/DCM (3 x). Carboxylic acids (R₃) were coupled (10 equiv) in the presence of DIC (10 equiv) and HOBt (10 equiv) in DMF (0.1M) for 120 minutes (3, Scheme 1). All coupling reactions were monitored for completion by the Ninhydrin test. Reductions were performed in a 4000mL Wilmad LabGlass vessel under nitrogen. Tetrahydrofuran (THF, 1.0M) borane complex solution was used in 40 fold excess for each amide bond. The vessel was heated to 65 °C and maintained at this temperature for 96 hours. The solution was then removed and the bags washed with THF and methanol (MeOH). Once completely dry, bags were treated overnight with piperidine at 65 °C and washed several times with DMF, DCM, and methanol (4, Scheme 1). Before proceeding, the completion of reduction was monitored by LCMS analysis of a control compound (4, Scheme 1) that was cleaved from solid support (HF, anisole, 0°C 7hr). Cyclization (5, Scheme 1) was performed with a 5-fold excess (for each cyclization) of cyanogen bromide (CNBr) in a 0.1M anhydrous DCM solution overnight. Following the cyclization, the bags were rinsed with DMF and DCM. The resin was cleaved with HF in the presence of anisole in an ice bath at 0 °C for 90 minutes (6, Scheme 1). After removal of the HF by gaseous N₂ the products were then extracted from the vessels with 95% acetic acid in water, transferred to scintillation vials, frozen and lyophilized. Compounds were then reconstituted in 50% acetonitrile and water, frozen and lyophilized three more times. For initial screening (data shown in section “Screening of individual compounds”) the individual compounds were tested as crude material in case the activity was driven by a side reaction that was also present in the original positional scanning library. After initial screening, the 5 front runner compounds, 1, 2, 7, 16 and 19 were selected for purification and all data reported in section “Exploring the antibacterial activity of frontrunner agents using a library of ESKAPE pathogen isolates” and beyond is from the purified stock of these five compounds. All chirality was generated from the corresponding amino acids. As previously reported by our group and others, the reduction of polyamides with borane is free of racemization^{38 38b 39}. For those compounds with multiple chiral centers, a single diastereomer was obtained.

LCMS analysis of crude material:

Purity and identity of initial crude compounds was verified using a Shimadzu 2010 LCMS system, consisting of a LC-20AD binary solvent pumps, a DGU-20A degasser unit, a CTO-20A column oven, and a SIL-20A HT auto sampler. A Shimadzu SPD-M20A diode array detector was used for detections. A full spectra range of 190–600nm was obtained during analysis. Chromatographic separations were obtained using a Phenomenex Luna C18 analytical column (5μm, 50 × 4.6mm i.d.). The column was protected by a Phenomenex C18 column guard (5μm, 4 × 3.0mm i.d.). All equipment was controlled and integrated by Shimadzu LCMS solutions software version 3. Mobile phases for LCMS analysis were HPLC grade or LCMS grade obtained from Sigma Aldrich and Fisher Scientific. The mobile phases consisted of a mixture of LCMS grade Acetonitrile /water (both with 0.1% formic acid for a pH of 2.7). The initial setting for analysis was 5% Acetonitrile (v/v), then linearly increased to 95% Acetonitrile over 6 minutes. The gradient was then held at 95% Acetonitrile for 2 minutes before being linearly decreased to 5% over 0.1 minutes and held until stop for an additional 1.9 minutes. The total run time was equal to 12 minutes, the total flow rate was 0.5mL/minute. The column oven and flow cell temperature for the diode array detector was 30°C. The auto sampler temperature was held at 15°C, and 5uL was injected for analysis.

HPLC Purification (compounds 1, 2, 7, 16 and 19):

All purifications were performed on a Shimadzu Prominence preparative HPLC system, consisting of LC-8A binary solvent pumps, a SCL-10A system controller, a SIL-10AP auto sampler, and a FRC-10A fraction collector. A Shimadzu SPD-20A UV detector was used for detection. The wavelength was set at 214nm during analysis. Chromatographic separations were obtained using a Phenomenex Luna C18 preparative column (5μm, 150 × 21.5mm i.d.). The column was protected by a Phenomenex C18 column guard (5μm, 15 × 21.2mm i.d.). Prominence prep software was used to set all detection and collection parameters. The mobile phases for HPLC purification were HPLC grade obtained from Sigma Aldrich and Fisher Scientific. The mobile phase consisted of a mixture of Acetonitrile/water (both with 0.1% formic acid). The initial setting for separation was 2% Acetonitrile, which was held for 2 minutes, then the gradient was linearly increased to 20% Acetonitrile over 4 minutes. The gradient was then linearly increased to 55% Acetonitrile over 36minutes. The HPLC system was set to automatically flush and re-equilibrate the column after each run for a total of 4 column volumes. The total flow rate was set to 12mL/min and the total injection volume was set to 3900uL. The fraction collector was set to collect from 6 to 40 minutes. The corresponding fractions were then combined and lyophilized.

LCMS analysis of purified compounds:

The purity and identity of purified compounds 1, 2, 7, 16 and 19 (all data reported from section “Exploring the antibacterial activity of frontrunner agents using a library of ESKAPE pathogen isolates” through to the end of the manuscript was generated with purified compounds) were carried out using a Shimadzu 2020 LCMS system, consisting of a LC-20AD binary solvent pumps, a DGU-20A degasser unit, a CTO-20A column oven and a SIL-20A HT auto sampler. A Shimadzu SPD-M20A diode array detector was used for detections. A full spectra range of 190–460nm was obtained during analysis. Chromatographic separations were obtained using a Phenomenex Gemini C18 analytical column (5μm, 250 × 2mm i.d.). The column was protected by a Phenomenex C18 column guard (5μm, 4 × 2mm i.d.). All equipment was controlled and integrated by Shimadzu Lab Solutions software version 5.53 SP3. Three different sets of conditions were used for analysis. Condition 1 (Acetonitrile/water pH 2.7): The mobile phase consisted of a mixture of LCMS grade Acetonitrile/water (both with 0.1% formic acid for a pH of 2.7) with initial settings for analysis of 5% organic mobile phase (v/v), which was linearly increased to 95% organic mobile phase over 38 minutes. The gradient was then held at 95% organic mobile phase for 4 minutes, then linearly decreased to 5% over 2 minutes and held until stop for an additional 1 minute. The total run time was equal to 46 minutes. Condition 2 (Methanol/water pH 7.4): The mobile phase consisted of LCMS grade Methanol/water containing 10mM Ammonium Bicarbonate (adjusted pH 7.4 with formic acid). The initial setting for analysis was 5% organic mobile phase (v/v), which was linearly increased to 95% organic mobile phase over 38 minutes. The gradient was then held at 95% organic mobile phase for 4 minutes, then linearly decreased to 5% over 2 minutes and held until stop for an additional 1 minute. The total run time was equal to 46 minutes. Condition 3 (Methanol/water pH 5.14): The mobile phase consisted of LCMS grade Methanol/water containing 50mM Ammonium Formate (adjusted pH 5.14 with formic acid). The initial setting for analysis was 60% Methanol (v/v), which was linearly increased to 80% Methanol over 10 minutes, before the gradient was linearly increased to 83% Methanol over 25 minutes. The gradient was again linearly increased to 95% Methanol over 3 minutes and held at 95% for an additional 4 minutes. Then the gradient was linearly decreased to 60% Methanol over 2 minutes and held until stop for a total run time of 46 minutes.

NMR analysis of purified compounds:

¹H and ¹³C NMR spectra were obtained utilizing the Bruker 400 Ascend (400 and 100 MHz, respectively). NMR chemical shifts were reported in δ (ppm) using the δ 7.26 signal of CDCl₃ (¹H NMR) and the δ 77.16 signal of CDCl₃ (¹³C NMR) as internal standards.

(S)-4-butyl-3-hexyl-1-(4-((S)-2-imino-3-(3-(trifluoromethyl)phenethyl)imidazolidin-4-yl)butyl)imidazolidin-2-imine (1)

Using the synthetic approach described in Scheme 1 for the synthesis of compound 1 was synthesized using the following reagents: (α-α-α-Trifluoro-m-Tolyl) acetic acid (R₁), Boc-L-Norleucine (R₂), Heptanoic Acid (R₃). Final crude product was purified by HPLC as described above. 1H NMR (400 MHz, CHLOROFORM-d): δ 8.7 (br. s., 2 H) 7.5 – 7.6 (m, 2 H) 7.4 – 7.5 (m, 2 H) 4.0 – 4.1 (m, 1 H) 3.8 – 3.9 (m, 1 H) 3.5 – 3.7 (m, 5 H) 3.3 – 3.5 (m, 2 H) 3.1 – 3.3 (m, 3 H) 2.9 – 3.1 (m, 2 H) 1.7 – 1.9 (m, 1 H) 1.5 – 1.6 (m, 5 H) 1.5 (d, J=8.1 Hz, 2 H) 1.2 – 1.4 (m, 13 H) 0.9 – 1.0 (m, 6 H); 13C NMR (100 MHz, CHLOROFORM-d) δ 169.0, 159.5, 157.2, 139.4, 132.7, 129.2, 56.4, 51.0, 46.1, 45.1, 43.0, 42.8, 33.5, 31.6, 31.5, 31.4, 29.0, 26.8, 26.5, 26.4, 22.5, 22.4, 20.5, 14.0, 13.9; LCMS (ESI+) Calcd for C₃₀H₄₉F₃N₆: 551.40, found [M+H]+: 551.35. LCMS retention time (214nm) Condition 1 (Acetonitrile/water pH 2.7): 17.168 min. Condition 2 (Methanol/water pH 7.4) 33.528 min. Condition 3 (Methanol/water pH 5.14): 9.254 min. All three conditions showed the desired compound accounting for 100% peak area and peak height.

(S)-4-butyl-3-(4-cyclohexylbutyl)-1-(4-((S)-2-imino-3-(3-(trifluoromethyl)phenethyl)imidazolidin-4-yl)butyl)imidazolidin-2-imine (2)

Using the synthetic approach described in Scheme 1 for the synthesis of compound 2 was synthesized using the following reagents: (α-α-α-Trifluoro-m-Tolyl) acetic acid (R₁), Boc-L-Norleucine (R₂), Cyclohexanebutyric Acid (R₃). Final crude product was purified by HPLC as described above. 1H NMR (400 MHz, CHLOROFORM-d): δ 8.7 (br. s., 2 H) 7.5 – 7.6 (m, 2 H) 7.4 – 7.5 (m, 2 H) 4.0 – 4.1 (m, 1 H) 3.8 (d, J=9.9 Hz, 1 H) 3.5 – 3.7 (m, 3 H) 3.5 (br. s., 1 H) 3.3 – 3.5 (m, 2 H) 3.1 – 3.30 (m, 3 H) 2.9 – 3.0 (m, 2 H) 2.2 (br. s., 3 H) 2.0 (s, 1 H) 1.6 – 1.9 (m, 6 H) 1.5 – 1.60 (m, 3 H) 1.4 – 1.5 (m, 2 H) 1.2 – 1.4 (m, 12 H) 0.8 – 1.0 (m, 4 H); 13C NMR (100 MHz, CHLOROFORM-d) δ 159.5, 157.2, 139.4, 132.7, 129.2, 123.5, 58.6, 56.4, 50.1, 46.1, 45.1, 43.0, 42.8, 37.5, 37.1, 33.5, 33.4, 33.3, 31.5, 31.4, 27.1, 26.6, 26.4, 26.3, 23.9, 22.5, 20.5, 13.9; LCMS (ESI+) Calcd for C₃₃H₅₃F₃N₆: 591.43, found [M+H]+: 591.45. LCMS retention time (214nm) Condition 1 (Acetonitrile/water pH 2.7): 18.363 min. Condition 2 (Methanol/water pH 7.4) 34.487 min. Condition 3 (Methanol/water pH 5.14): 12.048 min. All three conditions showed the desired compound accounting for 100% peak area and peak height.

(S)-4-(cyclohexylmethyl)-3-hexyl-1-(4-((S)-2-imino-3-(3-(trifluoromethyl)phenethyl)imidazolidin-4-yl)butyl)imidazolidin-2-imine (7)

Using the synthetic approach described in Scheme 1 for the synthesis of compound 7 was synthesized using the following reagents: (α-α-α-Trifluoro-m-Tolyl) acetic acid (R₁), Boc-L-Cyclohexylalanine (R₂), Heptanoic Acid (R₃). Final crude product was purified by HPLC as described above. 1H NMR (400 MHz, CHLOROFORM-d): δ 8.7 (br. s., 2 H) 7.5 – 7.6 (m, 2 H) 7.4 – 7.5 (m, 2 H) 4.0 – 4.1 (m, 1 H) 3.8–3.9 (m, 1 H) 3.5 – 3.7 (m, 4 H) 3.3 – 3.5 (m, 3 H) 3.1 – 3.3 (m, 3 H) 2.8 – 3.1 (m, 3 H) 1.6 – 1.8 (m, 8 H) 1.5 – 1.6 (m, 4 H) 1.2 – 1.4 (m, 14 H) 0.9 – 1.1 (m, 5 H); 13C NMR (100 MHz, CHLOROFORM-d) δ 159.5, 157.2, 139.4, 132.7, 129.2, 77.2, 58.6, 54.7, 51.8, 46.2, 45.2, 43.0, 42.8, 39.8, 34.3, 33.5, 32.5, 31.6, 31.4, 28.9, 26.7, 26.5, 26.3, 26.2, 26.1, 25.9, 22.5, 20.5, 14.0; LCMS (ESI+) Calcd for C₃₃H₅₃F₃N₆: 591.43, found [M+H]+: 591.45. LCMS retention time (214nm) Condition 1 (Acetonitrile/water pH 2.7): 18.380 min. Condition 2 (Methanol/water pH 7.4) 35.664 min. Condition 3 (Methanol/water pH 5.14): 12.373 min. All three conditions showed the desired compound accounting for 100% peak area and peak height.

(S)-1-(4-((S)-3-(4-cyclohexylbutyl)-2-iminoimidazolidin-4-yl)butyl)-4-(cyclohexylmethyl)-3-hexylimidazolidin-2-imine (16)

Using the synthetic approach described Scheme 1 for the synthesis of compound 16 was synthesized using the following reagents: Cyclohexanebutyric Acid (R₁), Boc-L-Cyclohexylalanine (R₂), Heptanoic Acid (R₃). Final crude product was purified by HPLC as described above. 1H NMR (400 MHz, CHLOROFORM-d): δ 8.7 (br. s., 2 H) 3.8–3.9 (m, 2 H) 3.6 – 3.8 (m, 5 H) 3.4 3.5 (m, 1 H) 3.3 (dd, J=9.72, 5.81 Hz, 1 H) 3.1 – 3.2 (m, 3 H) 1.5 – 1.8 (m, 17 H) 1.2 – 1.4 (m, 22 H) 1.0 1.2 (m, 2 H) 0.8 – 1.0 (m, 6 H); 13C NMR (100 MHz, CHLOROFORM-d) δ 169.1, 159.5, 157.2, 57.8, 54.7, 51.8, 46.1, 45.3, 42.8, 41.9, 39.8, 37.5, 37.2, 34.3, 33.4, 33.3, 32.5, 31.6, 31.4, 28.9, 27.5, 26.8, 26.7, 26.6, 26.5, 26.4, 26.2, 26.1, 25.9, 23.9, 22.6, 20.5, 14.0; LCMS (ESI+) Calcd for C₃₄H₆₄N₆: 557.52, found [M+H]+: 557.50. LCMS retention time (214nm) Condition 1 (Acetonitrile/water pH 2.7): 19.706 min. Condition 2 (Methanol/water pH 7.4) 37.568 min. Condition 3 (Methanol/water pH 5.14): 14.248 min. All three conditions showed the desired compound accounting for 100% peak area and peak height.

(S)-1-(4-((S)-3-(2-((3S,5S,7S)-adamantan-1-yl)ethyl)-2-iminoimidazolidin-4-yl)butyl)-4-butyl-3-hexylimidazolidin-2-imine (19)

Using the synthetic approach described in Scheme 1 for the synthesis of compound 19 was synthesized using the following reagents:1-Adamantaneacetic Acid (R₁), Boc-L-Norleucine (R₂), Heptanoic Acid (R₃). Final crude product was purified by HPLC as described above. 1H NMR (400 MHz, CHLOROFORM-d): δ 8.7 (br. s., 2 H), 3.9 (dd, J=9.6, 4.7 Hz, 1 H), 3.8 – 3.9 (m, 1 H), 3.6 – 3.8 (m, 5 H), 3.4 – 3.5 (m, 1 H), 3.3 (dd, J=9.7, 5.6 Hz, 1 H), 3.0 – 3.3 (m, 3 H), 2.0 (br. s., 4 H), 1.5 – 1.8 (m, 18 H), 1.2 – 1.4 (m, 15 H), 0.9 – 1.0 (m, 6 H); 13C NMR (100 MHz, CHLOROFORM-d) δ 169.2, 159.5, 157.3, 57.4, 56.4, 51.1, 46.0, 45.4, 42.9, 42.1, 40.5, 36.9, 31.7, 31.6, 31.5, 29.0, 28.5, 26.8, 26.7, 26.6, 26.4, 22.6, 22.5, 20.5, 14.0, 13.9; LCMS (ESI+) Calcd for C₃₃H₆₀N₆: 541.49, found [M+H]+: 541.45. LCMS retention time (214nm) Condition 1 (Acetonitrile/water pH 2.7): 18.270 min. Condition 2 (Methanol/water pH 7.4) 36.038 min. Condition 3 (Methanol/water pH 5.14): 12.692 min. All three conditions showed the desired compound accounting for 100% peak area and peak height.

Positional Scanning Library 2157:

Positional scanning library 2157 was synthesized using the general Scheme 1. The positional scanning library incorporates both individual and mixtures of amino acids (R₂) and carboxylic acids (R₁ and R₃). The synthetic technique facilitates the generation of information regarding the likely activity of individual compounds from screening of the library.^{14 40 41} Equimolar isokinetic ratios have previously been determined and calculated for each of the amino and carboxylic acids utilized for the respective mixtures.^{42 43} The bis-cyclic guanidine library 2157 has a total diversity of 45,864 compounds (42×26×42=45,864). The R₁ and R₃ positions as shown in Scheme 1 (6) each consist of 42 carboxylic acids and the R₂ contains 26 amino acids.

Scaffold Ranking Library:

The scaffold ranking library contains one sample for each of the 37 positional scanning libraries tested. Each of these samples contains an approximate equal molar amount of each compound in that library. So, for example, the sample 2157 in the scaffold ranking library contains 45,864 compounds in approximately equal molar amounts. These samples can be prepared by mixing the cleaved products of the complete positional scanning library, as was the case for sample 2157, or they can be synthesized directly as a single mixture.^{14 6}

Bacterial strains and growth conditions.

For this study we used a representative panel of multi-drug resistant clinical ESKAPE pathogen isolates (Supplemental Table S6).^{44 45 3a, 46 47} Gram-positive organisms were grown in tryptic soy broth media (TSB), whilst Gram-negative organisms were grown in lysogeny Broth (LB), as described by us previously. ⁴⁵

MIC and MBC determination assays.

The minimum inhibitory concentration (MIC) for the combinatorial libraries, deconvolved 2157 library, and individual compounds were determined as follows. Broth cultures of ESKAPE strains were grown overnight before being diluted 1 in 1,000 in fresh media. Sterile 96-well plates were loaded with culture and compounds (in DMF) were added at decreasing concentrations to equal a total volume of 200 μl per well. Care was taken to not add more than 2.0% DMF to any well. Plates were then incubated at 37°C, and MICs determined after 24 hours by visual inspection for a lack of turbidity in wells. All assays were performed in triplicate with identical results obtained. For both the scaffold ranking and positional scanning samples, relative broad-spectrum activity was determined via stacked scores:

$$StackedScore=\sum _{i=E,S,K,A,P,E}\frac{100}{MICof{i}^{\text{th}}Pathogen}$$

Minimal bactericidal concentrations (MBC) were determined for 1, 2, 7, 16, and 19 using MIC cultures. Briefly, compound was washed by centrifugation and serial dilution, before plating on Tryptic Soy Agar (TSA). Plates were incubated for 24 hours at 37°C and cell viability assessed by determining CFU/ml at each concentration, for every compound. Percent recovery was then determined compared to CFU/ml from no drug controls. All concentrations and controls were tested using three biological replicates, alongside two technical replicates for each data point.

Time kill assay.

Time kill assays were performed in a 96-well microtiter plate using a BioTek Synergy2 plate reader. To prepare bacterial cultures, stationary phase MRSA cells were inoculated into fresh TSB and grown for 3 hours. After this time, cultures were inoculated into a 96-well microtiter plate at an OD₆₀₀ of 0.5, followed by the addition of test agent at MIC concentrations. In parallel, 2.0% sodium dodecyl sulfate (SDS), 0.001% benzalkonium chloride, 0.001% benzethonium chloride, and 4 μM lysostaphin were used as positive controls. Doxycycline (200 μM), a bacteriostatic translation inhibitor that does not result in cell lysis was used as a negative control. Assays were performed in triplicate over the span of 130 minutes, with OD₆₀₀ readings taken every 10 minutes.

MBEC determination assays.

The minimum biofilm eradication concentration (MBEC) was determined in 96-well microtiter plates as follows. Broth cultures of ESKAPE strains were grown using the conditions described above. Biofilms for each of the ESKAPE pathogens were generated from these as we have previously described for S. aureus, however human serum was not used for non-staphylococcal organisms.^{46, 48} For all organisms, biofilms were developed by standardizing an overnight culture into fresh media to an OD₆₀₀ of 0.5 and adding 150 μl into each well of a 96-well microtiter plate. Biofilms were allowed to develop for 24 hours, before the media was carefully removed and 200 μl of fresh media added containing a range of front-runner agent (above and below MIC). These cultures were incubated at 37°C overnight alongside no drug controls. After 24 hours, the media was removed from wells and the biofilm resuspended in phosphate buffered saline (PBS). Cultures were mixed by vigorous pipetting, before being serially diluted in PBS, and plated in duplicate on relevant agar. Plates were incubated at 37°C for 24 hours, and CFUs determined by enumeration. Each analysis was performed using three technical replicates, and antibiofilm activity was determined by comparing treated to untreated samples.

Cytotoxicity assay.

Cytotoxicity assays were performed using human A549 cells (adenocarcinomic human alveolar epithelial cells), as described by us previously.⁴⁶ Briefly, cells were cultured in F-12K Nutrient mixture (Kaighn’s Modification) media containing L-glutamine, supplemented with 10% fetal bovine serum and 1% penicillin-streptomycin for 3 days at 37° and 5% CO₂. Cells were then diluted to 1.0 × 10⁵ ml⁻¹ using F-12K supplemented media, and added to 96 well tissue culture plates at a volume of 100 μl.. Plates were incubated for 24 hours at 37°C and 5% CO₂, allowing the cells to adhere to the plastic. After this time, media was carefully removed and 200 μl fresh F-12 added with test compounds at concentrations ranging from 1–125 μM. Plates were then incubated for 48 hours at 37^oC and 5% CO₂.After 48 hours the media was removed, and new media added containing MTT (3-(4, 5-dimethylthiazol-2-yl)-2, 5-diphenyltetrazolium bromide), followed by incubation for four hours at 37°C and 5% CO₂. After 4h, 50 μl of media was removed, replaced with DMSO and incubated for ten minutes at 37^oC in order to solubilize any formazan produced. A Biotek plate reader was used to measure the absorbance of formazan production at 540nM, and IC₅₀ values were determined for each of the five compounds. Front runners were solvated in 45% w/v (2-hydroxypropyl)-β-cyclodextrin in water for these studies; which was also used alone as a negative control. IC₅₀ values were determined for each compound by comparison to vehicle only controls, to assess toxicity to human cells.

Hemolysis assay.

A hemolysis assay was performed using whole human blood (Bioreclamation), as described previously,⁴⁹ with the following modifications. Human red blood cells (hRBCs) were resuspended 20% v/v in 1X HA buffer (4.25 ml 10% NaCl; 1ml CaCl₂ in 50 ml sterile water), and lead agents were added at a concentration of 10 μM, to a final volume of 100 μl. Cells were incubated for 15 minutes at 37°C before being centrifuged at 5,500 g for 1 minute to pellet non-lysed hRBCs. The supernatant was removed, added to a 96-well microtiter plate and the OD₅₄₃ read using a Biotek synergy2 plate reader. The negative control was vehicle only (DMF), and the positive control was 1.0% triton X-100. Assays were performed in triplicate, with data displayed as percent hemolysis compared to controls, defined as: Percent Hemolysis = (OD₅₄₃ test sample – OD₅₄₃ no drug control) / (OD₅₄₃ triton X-100 – OD₅₄₃ no drug control) × 100.

Resistance Assays.

In order to test potential resistance towards the lead agents, a serial passage assay was performed alongside control compounds (E. faecium and A. baumannii = tetracycline; P. aeruginosa and E. cloacae = ciprofloxacin; S. aureus = vancomycin; K. pneumoniae = rifampin). ESKAPE pathogens were grown overnight in liquid media at 37 °C. These cultures were then diluted 1:100 in fresh media, and seeded into a 96-well plate. Lead bis-cyclic guanidines or control agents were added to respective wells at half MIC concentrations. Plates were then incubated for 24 hours at 37 °C, with bacteria removed from these cultures on the following day, to inoculate fresh media (1:100 dilution) containing compounds at a 2-fold higher concentrations. These were then grown overnight, and the procedure repeated for a total of eight days. The cultures were observed for a lack of growth, indicating strains were no longer able to resist the action of a given compound. Each experiment was performed in triplicate, yielding identical results.

Assessing efficacy during bacterial infection in vivo.

A murine model of lethal peritonitis was used to demonstrate the effectiveness of the bis-cyclic guanidines to clear bacterial infections, as described by us previously.⁴⁵ All animal studies received written approval after review by the Institutional Animal Care & Use Committee in the Division of Comparative Medicine & Division of Research Integrity & Compliance at the University of South Florida. Six mice per group were injected with 1 × 10⁸ CFU ml⁻¹ of Staphylococcus aureus (USA300 strain FPR3757) in PBS containing 5% mucin. After 1h, mice were inoculated with either 5xMIC (4 nM) or 10xMIC (8 nM) of vancomycin (I.V., positive control); 2xMIC (2 μM) of front runner agents (I.M., test group); or vehicle alone (I.V.; 45% w/v (2-hydroxypropyl)-β-cyclodextrin in water; negative control). Mice were monitored twice daily for five days to assess mortality. The clinical endpoint of this study was when the mice reached a pre-moribund state. Characteristics of pre-moribund state include: hunched posture, rapid, shallow and/or labored breathing, ruffled fur, lethargy, failure to respond to stimuli, soiled anogenital area, paralysis, paresis, head tilt, circling, vocalizations, and non-purposeful movements and/or were unable to eat or drink. Those mice reaching this state prior to the completion of the 5 day infection period were euthanized. The number of mice surviving between control and treatment groups was compared and analyzed for statistical significance using a log rank test.

SAS Maps:

SAS maps were generated following a standard and well-validated protocol. ^{7 20} Briefly, for each pair of compounds ith and jth, potency differences were determined as the absolute difference between their pIC₅₀ activity values. On a relative scale, the potency similarity (PS_i,j) was measured with the expression:

$$P{S}_{i,j}=1-\frac{\left|A_i-A_j\right|}{max-min}$$

where Ai and Aj are the activity values of the ith and jth molecules, and max-min indicates the range of activities in the data set. Pairwise structural similarities were computed using the Tanimoto coefficient⁵⁰ with radial fingerprints as implemented in Canvas.²⁸ Property similarities were computed with 6 continuous coordinates: MW, PSA, AlogP, RB, HBA and HBD.

Properties were auto-scaled with mean centering using the equation:⁵¹

$$p_{ki}=\frac{P_{ki}-\overline{P_k}}{{\sigma }_{P_k}}$$

where 𝑝_ki denotes the scaled version of the 𝑘th property for the 𝑖th molecule, 𝑃_ki denotes the unscaled value, and $\overline{P_k}$ and ${\sigma }_{P_k}$ denote, respectively, the mean and standard deviation of the 𝑘th property over all molecules in the study. The Euclidean distance between a pair of molecules in the property space was then computed with the expression:

$$d_{ij}={\left[\sum _{k=1}^K{\left(p_{ki}-p_{kj}\right)}^2\right]}^{1/2}$$

where 𝑑_ij, denotes the Euclidean distance between the 𝑖th and 𝑗th molecules; 𝑃_ki, and 𝑃_kj, denote the value of the scaled property 𝑘 of the 𝑖th and 𝑗th molecules, respectively. In this work 𝐾=6 for the four physicochemical properties. Then, Euclidean distances were scaled from 0 to 1 as follows:

$$s{d}_{ij}=\frac{d_{ij}-\min d_{ij}}{\max d_{ij}-\min d_{ij}}$$

where 𝑠𝑑_ij, is the scaled distance, and max 𝑑_ij, and min 𝑑_ij, indicate the range of distances in the data set. Pairwise property similarities were measured with the expression:

$$P{S}_{ij}=1-s{d}_{ij}$$

where 𝑃𝑆_ij, represents the molecular similarity using four continuous descriptors of the 𝑖th and 𝑗th molecules, and 𝑠𝑑_ij, is the scaled distance.