Pathway-Directed Screen for Inhibitors of the Bacterial Cell Elongation Machinery

New antibiotics are needed to combat the growing problem of resistant bacterial infections. An attractive avenue toward the discovery of such next-generation therapies is to identify novel inhibitors of clinically validated targets, like cell wall biogenesis.

ABSTRACT

New antibiotics are needed to combat the growing problem of resistant bacterial infections. An attractive avenue toward the discovery of such next-generation therapies is to identify novel inhibitors of clinically validated targets, like cell wall biogenesis. We have therefore developed a pathway-directed whole-cell screen for small molecules that block the activity of the Rod system of Escherichia coli. This conserved multiprotein complex is required for cell elongation and the morphogenesis of rod-shaped bacteria. It is composed of cell wall synthases and membrane proteins of unknown function that are organized by filaments of the actin-like MreB protein. Our screen takes advantage of the conditional essentiality of the Rod system and the ability of the beta-lactam mecillinam (also known as amdinocillin) to cause a toxic malfunctioning of the machinery. Rod system inhibitors can therefore be identified as molecules that promote growth in the presence of mecillinam under conditions permissive for the growth of Rod^– cells. A screen of ∼690,000 compounds identified 1,300 compounds that were active against E. coli. Pathway-directed screening of a majority of this subset of compounds for Rod inhibitors successfully identified eight analogs of the MreB antagonist A22. Further characterization of the A22 analogs identified showed that their antibiotic activity under conditions where the Rod system is essential was strongly correlated with their ability to suppress mecillinam toxicity. This result combined with those from additional biological studies reinforce the notion that A22-like molecules are relatively specific for MreB and suggest that the lipoprotein transport factor LolA is unlikely to be a physiologically relevant target as previously proposed.

INTRODUCTION

New therapeutics are needed to overcome the growing problem of antibiotic-resistant bacterial infections (1). A common approach for the identification of compounds with antibacterial activity is to screen large libraries of small molecules for potential antibiotic activity. Screens involving in vitro assays of purified enzymes have successfully identified inhibitors of essential bacterial targets (2). However, molecules discovered in this way often have limited activity on cells due to their inability to penetrate the bacterial envelope and/or due to their efficient efflux by resistance pumps (3). These issues are particularly acute for Gram-negative bacteria, the outer membrane of which poses an especially difficult barrier for drugs to cross (4). Coupled with their efficient efflux systems that work synergistically with the outer membrane barrier, these organisms display a high level of intrinsic antibiotic resistance.

Alternatives to target-directed in vitro screens are those performed on growing bacterial cells to identify molecules that block their replication. These screens have the advantage that positive hits are known to have antibacterial activity from the start. However, the disadvantages are that target identification is challenging and that many of the lethal molecules identified are nonspecific poisons with metal-chelating, detergent-like, or redox-active modes of action unsuitable for drug development (2, 3).

Pathway-directed whole-cell screens combine the practicality of screens performed on growing cells with a specificity that approaches that of an in vitro assay. As with traditional whole-cell screens, this approach finds molecules with cellular activity. However, in this case, genetic logic is used in the design such that the screening readout identifies hit molecules with a mode of action that is likely limited to a particular biochemical pathway. Thus, the hits are enriched for compounds with specific on-target activity. Previously published screens for inhibitors that block wall teichoic acid (WTA) biogenesis in the Gram-positive pathogen Staphylococcus aureus and a more recent application to discover inhibitors of lipopolysaccharide biogenesis in the Gram-negative pathogen Acinetobacter baumannii are classic examples of the pathway-directed approach (5, 6).

The classic work that demonstrated the power of pathway-directed screening relied on the conditional essentiality of steps in WTA biogenesis (5). WTAs are anionic polymers attached to the peptidoglycan (PG) cell wall of Gram-positive bacteria (7). They play important roles in the surface biology of these organisms and are essential for the virulence of S. aureus (8). WTA polymers are built on the undecaprenol-phosphate (Und-P) lipid carrier at the inner face of the cytoplasmic membrane. Once their synthesis is completed, they are transported across the membrane by the flippase TarGH and then attached to the PG layer by LCP enzymes (9, 10). Blocking the late steps of WTA synthesis is lethal due to the accumulation of lipid-linked precursor molecules that deplete the cell of Und-P carrier that is also needed for PG synthesis (11). However, this lethality can be suppressed by inactivating the first enzyme in the WTA biogenesis pathway, TarO, to prevent WTA precursor accumulation (11). Thus, a screen for molecules that kill wild-type cells but not those deleted for the tarO gene should identify inhibitors of late-stage enzymes of the WTA synthesis pathway. Indeed, screens employing this logic successfully identified inhibitors that block the activity of the TarGH transporter (5).

In this report, we describe the development of a pathway-directed screen for compounds that disrupt the function of the cell elongation system of Escherichia coli. This PG biogenesis complex is called the Rod system (a.k.a., the elongasome), and it consists of five integral membrane proteins organized by filaments of the actin-like MreB protein (12) (Fig. 1A). The complex is normally essential for the growth of E. coli and other rod-shaped bacteria and is required for them to elongate their cell wall and maintain a capsule-like rod shape (13). All components of the Rod system are conserved and broadly distributed among bacteria, including penicillin-binding protein 2 (PBP2), a known beta-lactam target, and RodA, which was recently shown to function as a cell wall polymerase (14–16). Thus, the Rod system contains both proven and attractive new targets for therapeutic development.

FIG 1.

Rod system and screening rationale. (A) Shown is a cartoon of the Rod complex highlighting the individual components. The complex is organized by filaments of MreB. RodA is a PG polymerase that synthesizes the glycan strands of the PG layer (14). PBP2 is a transpeptidase that cross-links the RodA products into the existing matrix (50). The functions of MreC, MreD, and RodZ remain ill defined, but RodZ is known to interact directly with MreB (51). (B) Shown is a summary of the genetic properties of Rod system mutants and the response of cells to mecillinam treatment. Note that Rod^– cells with FtsZ^UP are resistant to mecillinam (22). This property is the basis of the pathway-directed screen. Molecules that block Rod system activity should rescue growth in the presence of mecillinam.

As part of a broader program to identify new antibacterial agents, we screened a library of ∼690,000 compounds for growth-inhibitory activity against E. coli. We then counterscreened active molecules for activity against the Rod system using our pathway-directed readouts. Eight analogs of the MreB antagonist A22 (17, 18) in the library were successfully identified, validating the effectiveness of our screening approach for identifying inhibitors of at least one component of the system. Further characterization of these inhibitors showed that their antibiotic activity under conditions where the Rod system is essential was strongly correlated with their ability to suppress the toxicity of mecillinam (also known as amdinocillin). This result combined with additional biological studies reinforce the notion that A22-like molecules are relatively specific for MreB and suggest that the lipoprotein transporter LolA is unlikely to be a physiologically relevant target of this scaffold.

RESULTS

Rationale for the screen.

The Rod system is normally essential (19). In E. coli and other rod-shaped bacteria, the depletion of any component of the system causes cells to lose their rod-shape and to develop into large spherical cells that eventually lyse (19). However, the lethality of Rod system inactivation can be suppressed in E. coli by moderate overproduction of the tubulin-like FtsZ protein (19) (Fig. 1B), filaments of which play a key role in the formation and activity of the cytokinetic ring apparatus (20). We use the term FtsZ^UP to refer to cells producing extra FtsZ from the plasmid pTB63. FtsZ^UP cells inactivated for components of the Rod system are viable and grow and divide as small spheres rather than rods. Although the mechanism by which increased FtsZ levels suppress Rod system essentiality is not known, the phenomenon has been useful for genetic analysis of the machinery and studies of the effects of antibiotics on its activity (21, 22).

Several years ago, we made the surprising observation that inhibition of the Rod system component PBP2 with a beta-lactam called mecillinam differs from genetic inactivation of the PBP2-encoding gene (22). This led us to discover that the PBP2-mecillinam complex causes the Rod system to malfunction and become toxic. Mecillinam-induced toxicity is caused by the inhibition of PG cross-linking by PBP2, while glycan chains continue being produced by RodA. Because the glycans cannot be cross-linked, they are rapidly degraded, resulting in a futile cycle of PG synthesis and breakdown by the Rod complex (22). Blocking PG synthesis by the Rod system in FtsZ^UP cells prevents the futile cycle and suppresses mecillinam toxicity (Fig. 1B). Suppression can be achieved either by genetically inactivating Rod system components or by disrupting its activity with small molecules, such as the MreB antagonist A22 (22). We therefore reasoned that other compounds that disrupt Rod system activity by targeting MreB or other members of the complex (PBP2, MreC, MreD, RodA, and RodZ) could be identified by screening for molecules that suppress the toxicity of mecillinam in an FtsZ^UP E. coli strain (Fig. 1B).

Screening for anti-Rod system compounds.

Prior to screening for Rod inhibitors, 690,000 small molecules were tested for their ability to inhibit growth of an E. coli strain in which the lptFG genes were placed under the control of the arabinose promoter and expressed at suboptimal levels (23). These genes encode components of the lipopolysaccharide (LPS) transport system such that the strain has a defective outer membrane permeability barrier, which was useful to maximize the number of lethal molecules identified. Compounds were assayed in duplicate. Endpoint absorbance (optical density at 600 nm [OD₆₀₀]) readings for the test strain were measured following 23 h of growth in the presence of compounds (Fig. 2A). Hits were identified as those molecules causing a significant reduction (−2.5σ) in OD₆₀₀ readings relative to the vehicle (dimethyl sulfoxide [DMSO]) control (Fig. 2A). All commercially available hits along with some closely related analogs were purchased and assembled into a sublibrary of 1,232 molecules to be used for the identification of inhibitors for specific targets in Gram-negative pathogens.

FIG 2.

High-throughput screening results. Shown are scatter plots of OD₆₀₀ readings for cells treated with each test compound in duplicate (replicates 1 and 2). (A) The compound collection at the ICCB-Longwood facility was screened for growth inhibition of the outer membrane-defective E. coli strain NR1099 (23). Hits were identified as any compound causing the final culture OD₆₀₀ to be less than 0.52 in either replicate. (B) A sublibrary based on the results of screen 1 was assembled and screened for the ability to suppress the toxicity of mecillinam against FtsZ^UP cells. Compounds that promoted culture growth to an OD₆₀₀ of 0.158 or better (x̅ ± 3σ) were identified as hits. (C) Hits from screen 2 were tested for their activity against FtsZ^UP cells in the absence of mecillinam. Those that allowed culture growth to an OD₆₀₀ of 0.07 or greater were identified as hits and were considered potential Rod system inhibitors. See text for details. Red dotted lines are the best fit to the replicate data, indicating strong correlation between replicate values. Gray boxes indicate compounds identified as hits.

Using this sublibrary as a starting point, the screen for Rod system inhibitors was performed in the following two steps: (i) screening the sublibrary of lethal molecules for the ability to suppress mecillinam toxicity in FtsZ^UP cells, and (ii) using additional assays to test for on-target activity against the Rod system. We screened the sublibrary for compounds capable of promoting growth of FtsZ^UP E. coli cells treated with mecillinam (2.5 μg/ml, 8× the MIC) (logic based on Fig. 1B, highlighted rows). A total of 72 compounds (5.7%) were found to suppress mecillinam toxicity (Fig. 2B), identifying them as potential Rod system inhibitors. However, prior genetic studies indicate that the inactivation of over 60 genes in E. coli beyond those encoding Rod system components is capable of suppressing mecillinam toxicity (21). Thus, a range of different targets for the hit molecules identified in the Rod screen were possible. To identify true Rod system targeting molecules among the 72 mecillinam suppressors, we performed two additional assays. The first assay is based on the observation that the Rod system is nonessential in FtsZ^UP cells (Fig. 1B). Therefore, the lethality of a molecule that targets the Rod system should be conditional and suppressed by elevated FtsZ levels. Of the 72 mecillinam suppressors, 44 compounds satisfied this screen and failed to block the growth of an FtsZ^UP E. coli strain with a tolC deletion to weaken its outer membrane permeability barrier (Fig. 2C). Finally, the 44 hits from this screen were tested for their ability to alter cell shape. A total of eight compounds were identified that induced a shape change in target cells, consistent with the inhibition of Rod system function (Fig. 1B, 3, and 4).

FIG 3.

Effect of selected hits on cell shape. Phase-contrast micrographs of the efflux-deficient E. coli strain MG1655 ΔtolC (JAB063) harboring pTB63 (FtsZ^UP) after treatment with 0.2% DMSO (A), 20 μg/ml of A22 (B), ∼10 μg/ml of compound 3 (C) or ∼10 μg/ml of compound 7 (D) at 30°C for 72 h. Scale bar = 5 μm.

FIG 4.

Chemical structures of A22 and related hit compounds. (A) Chemical structure of A22. (B) Chemical structures of positive hits from the screening procedure. Green stars highlight hits that were purchased and validated. (C) Chemical structure of MAC1324 and the products of its hydrolysis. (D) Chemical structures of A22-like compounds that were negative in the screen for Rod system inhibitors. Red stars highlight true negatives that were purchased and validated. Yellow star highlights the weakly active A22-like molecule that was the lone false-negative compound with this scaffold. (E) Structures of additional A22-like compounds that were purchased for assay validation.

Characterization of hit compounds.

All eight of the hit compounds (compounds 1 to 8, Fig. 4B) are structurally similar to the known MreB antagonist A22 (Fig. 4A). A22 was discovered in a screen for inhibitors of chromosome partitioning (17) and was later found to inhibit the Rod system through direct interaction with MreB (18, 24, 25). Compound 1 most resembles A22 (Fig. 4B), and its anti-MreB activity was confirmed (Table 1). The remaining seven compounds (compounds 2 to 8, Fig. 4B) are S-triazine-substituted analogs of the A22 scaffold reminiscent of MAC13243 (26) (Fig. 4C). This compound was initially identified as a growth inhibitor with lethal activity that could be suppressed by overexpression of the lolA gene encoding an essential component of the lipoprotein transport machinery in Gram-negative bacteria (26). However, it was subsequently found to undergo acid hydrolysis into an A22-like compound and to target MreB in Pseudomonas aeruginosa (Fig. 4C) (27–29). Consistent with these results, we found that compound 4 undergoes hydrolysis and is active, whereas the related compound 19 remains intact and has drastically reduced A22-like cellular activity (Table 1; Fig. S1 in the supplemental material). We therefore infer that hit compounds 2 to 8 (Fig. 4B) are all likely to require hydrolysis for anti-MreB activity.

TABLE 1.

Summary of compound activity

Compound	Inhibit growtha	Suppress mecb	Round cellsc	MIC90 (µg/ml)d				Min MIC suppg	Ratio MIC/mech
				E. colie	E. colie + FtsZUP	ΔmreBCD mutant + FtsZUPf	S. aureus HG003
A22	+++	Y	Y	1.6	>100	>100	>100	0.8	2
1	+++	Y	Y	0.8	>100	100	100	0.4	2
2	+++	Y	Y	NA	NA	NA	NA	NA	NA
3	+++	Y	Y	NA	NA	NA	NA	NA	NA
4	+++	Y	Y	3.1	>100	>100	100	1.6	2
5	+++	Y	Y	NA	NA	NA	NA	NA	NA
6	+++	Y	Y	NA	NA	NA	NA	NA	NA
7	+++	Y	Y	NA	NA	NA	NA	NA	NA
8	+++	Y	Y	NA	NA	NA	NA	NA	NA
9	+	Y	N	12.5	50	25	12.5	1.6	8
10	+	Yi	Yi,j	50	>100	50	100	25	2
11	+	Yi	Yi	12.5	>100	>100	>100	6.25	2
19	+	Y	Yj	100	>100	100	>100	50	4
20	+++	Y	Y	1.6	>100	>100	>100	0.8	2
21	+++	Y	Y	6.25	>100	>100	>100	3.1	2
22	+	Y	Yj	50	>100	>100	>100	25	2
23	+	Y	Y	50	>100	>100	>100	12.5	4
CBR-4830	+++	ND	Y	3.1	6.25	12.5	12.5	>100	NA

^a+++, strong growth phenotype, OD₆₀₀ reduced >85%; +, weak growth phenotype, OD₆₀₀ reduced >40%.

^bResults from the high-throughput mecillinam suppression screen using E. coli MG1655/FtsZ^UP (pTB63) and a single compound concentration (∼10 μM). Y, yes, OD₆₀₀ of ≥0.15 after 72 h at 30°C; N, no, OD₆₀₀ of <0.15 after 72 h at 30°C; ND, not detected.

^cY, yes, spherical cells observed after treatment with 20 μg/ml; N, no, spherical cells never observed.

^dMIC₉₀, MIC resulting in 90% reduced growth; NA, not assayed.

^eE. coli MG1655 ± FtsZ^UP (pTB63).

^fE. coli MG1655 ΔmreBCD (TU233) FtsZ^UP (pTB63).

^gMin MEC supp, the minimum compound concentration required to suppress 2.5 μg/ml mecillinam (i.e., OD₆₀₀ ≥0.15 after 72 h at 30°C).

^hThe ratio of compound MIC₉₀ to the minimum concentration required for mecillinam suppression.

ⁱSuppression of mecillinam and cell rounding were only observed for the purchased compounds, not the screening aliquots.

^jCell rounding required treatment with 100 μg/ml.

Efficacy of the screening procedure.

To better understand the relative effectiveness of the screen, we wanted to know how many total A22-like molecules were in the original library and whether any potential hits were missed during screening. To create a comprehensive list of all molecules resembling A22 in the initial library used for the primary screen, we performed a similarity analysis with Pipeline Pilot from SciTegic (30) using both an extended-connectivity bit string and a feature-connectivity bit string (FCFP). We applied a loose threshold of 0.4 to 0.6 to avoid false negatives (31) and identified 154 analogs (Table S1). Of these 154 compounds, 51 compounds inhibited growth (Table S1), and 14 compounds were acquired for the bioactive sublibrary (Fig. 4B and C). An additional four A22-like analogs were purchased for assembly of the sublibrary that were not present in the initial pool. Thus, the bioactive library used for the mecillinam suppression screen contained 18 molecules with similarity to A22 (compounds 1 to 18, Fig. 4B and D).

Compounds 9 to 18 were not identified as having activity against the Rod system based on the results of the sequential screening procedure (Fig. 4D). Consistent with prior structure-activity relationship (SAR) studies of A22 (28, 29, 32–34), most of these molecules lack either the isothiourea or aryl halide moieties important for bioactivity and thus would not be predicted as screening hits. However, compounds 9 to 11 resembled A22 enough to warrant further investigation of their activity. Compound 9 effectively suppressed mecillinam toxicity but did not induce cell rounding (Table 1). Compounds 10 and 11, on the other hand, had weak activity that likely contributed to their negative results in the screen. They required a concentration near or significantly above the screening concentration to suppress mecillinam toxicity and induce cell rounding (Table 1). Based on the analysis of the A22-like hits from the screen and related negatives, we conclude that the screening procedure is effective at identifying MreB antagonists and does not suffer major issues with false negatives.

Relationship between MIC and mecillinam suppression activity of A22-like molecules.

To determine the relationship between the MIC of A22-like molecules on wild-type cells and the concentration required for mecillinam suppression, we purchased compounds 20 to 23 (Fig. 4E) and assayed their activities along with those of A22 and compounds 1 and 11 (Fig. 4B and D). As expected based on prior studies of these molecules (28, 29, 32–34), they all killed wild-type E. coli and induced a rod-to-sphere shape change (Table 1 and Fig. S2). With the exception of compound 11, which had weak activity, all of the molecules suppressed mecillinam toxicity with a minimal concentration that was half of their MIC. Thus, there is good correlation between the mecillinam suppression activity and the MIC of the A22-like molecules. Because mecillinam suppression activity is reflective of MreB antagonism, this finding is consistent with the activity of A22-like molecules being largely (if not entirely) based on a disruption of Rod system function at concentrations around their MIC.

LolA is unlikely to be a physiologically relevant target for A22-like molecules.

The identification of MAC13243 as a potential inhibitor of LolA in the lipoprotein transport pathway (26) and the subsequent finding that this molecule degrades into an A22-like product have led to the suggestion that A22 and its relatives may target LolA as well as MreB (28). On the other hand, the tight correlation between the MIC and mecillinam suppression activity reported here supports a relatively specific mode of action for A22-like molecules in targeting MreB. We therefore further investigated the chemical genetic observations connecting MAC13243, A22, and LolA. The finding that LolA overproduction suppressed the lethal action of MAC13243 was the initial observation that led to the suggestion that LolA is the target for these molecules (26). It was then later shown that LolA overproduction similarly suppressed killing by A22, suggesting a connection between its activity and lipoprotein transport (28). An alternative interpretation of these results comes from the prior findings that LolA overproduction induces the Rcs envelope stress response (35), and that the Rcs stress response provides resistance to peptidoglycan synthesis inhibitors (21, 36). Thus, we suspected that it is the induction of the Rcs response by lolA overexpression that suppresses the lethal activity of A22, not the overproduction of a potential target. To investigate this possibility, we tested the effect of LolA overproduction on A22 lethality in the presence or absence of RcsF, the lipoprotein sensor responsible for triggering the Rcs response (37). As expected based on prior results, LolA overproduction was indeed capable of suppressing cell killing by A22 (Fig. 5A). However, this suppressive effect was lost in ΔrcsF mutant cells even though similar levels of LolA were produced (Fig. 5B), indicating that suppression by LolA overproduction requires the Rcs response.

FIG 5.

Rcs response is required for LolA overproduction to suppress A22. (A) Cultures of wild-type (WT) E. coli MG1655 or a ΔrcsF derivative containing plasmid pJAB107 (P_tac::lolA) were normalized for OD₆₀₀ and dilutions were prepared and spotted onto LB agar with the indicated concentrations of A22 and IPTG. Plates were grown at 30°C for 48 h and photographed. (B) The same strains with either the empty vector control (pHC800) or pJAB107 (P_tac::lolA) were grown overnight at 30°C in LB. The resulting cultures were diluted 1:100 and grown at 30°C to an OD₆₀₀ of 0.2, and as indicated, IPTG was added to 500 µM. Growth was continued, and cells from 20 ml of culture were harvested, suspended in 0.5 ml sample buffer, and boiled for 5 min. A portion (5 µl) of each extract was run on a 4 to 20% gradient gel and stained with Coomassie brilliant blue. An inducer-dependent band was observed specifically in cells with pJAB107 (P_tac::lolA) running at the expected molecular weight (MW) of mature LolA (20 kDa). The level of protein produced was unaffected by the rcsF mutation.

Rod system inactivation results in an outer membrane permeability defect.

MAC13243 was also recently identified in a high-throughput screen for molecules that disrupt the outer membrane permeability barrier of E. coli (38). The authors of that study showed that this molecule makes cells more susceptible to bulky antibiotics, like vancomycin, that are normally excluded by the outer membrane. Given the prior connection of MAC13243 with LolA (26), it was concluded that the permeability defect induced by the compound was a result of its interference with the lipoprotein transport pathway (38). However, based on our analysis described above, we reasoned that the observed effect of MAC13243 was more likely to result from its breakdown product inhibiting the Rod system. Accordingly, A22 was previously shown to sensitize cells to novobiocin, which like vancomycin is normally excluded by the outer membrane (39). To further investigate the effect of MreB inactivation on outer membrane function, we compared the sensitivity of otherwise wild-type FtsZ^UP E. coli versus an ΔmreB derivative growing as spherical cells. The mutant cells were found to be hypersensitive to vancomycin and bile salts (Fig. 6). As reported previously, they were also found to be hypersensitive to sodium dodecyl sulfate (SDS) detergent (19) (Fig. 6). We therefore conclude that cells inactivated for the Rod system have an outer membrane permeability defect, and that the previously observed effect of MAC13243 on cell permeability is most likely due to its activity against MreB, not LolA.

FIG 6.

Spherical cells have a defective outer membrane permeability barrier. Spot dilutions across a range of OD₆₀₀ values (1.0 to 1e-5) for wild-type (WT) E. coli MG1655 or ΔmreB mutant cells harboring pTB63 (FtsZ^UP) plated onto MacConkey agar containing bile salts, or LB agar containing 10 μg/ml vancomycin (Vanco), 50 μg/ml vancomycin, or 1% wt/vol sodium dodecyl sulfate (SDS). Plates were grown at 30°C for 48 h and photographed.

Other potential Rod system inhibitors and their activities.

In addition to A22 (17, 18) and MAC13243 (28, 29) that is hydrolyzed to form an A22-like molecule, there are two additional antagonists of the Rod system of known structure reported in the literature, CBR-4830 (40) and sceptrin (41). The natural product 654/A was recently identified as a possible RodA inhibitor, but its structure has yet to be revealed (42). Of the compounds of known structure, we wondered whether they or their close analogs were present in the chemical library used for screening and, if so, why they might not have been identified by the assay procedure.

Sceptrin is a natural product (43) and did not have representation in our library, but similarity analysis identified 66 potential CBR-4830 analogs in the initial library used for the lethality screen (Table S2). Of these, 22 compounds inhibited growth, and 11 compounds were included in our bioactive sublibrary (Table S2). None of the 11 CBR-4830 analogs suppressed mecillinam toxicity. To further analyze why these compounds were not identified as hits, we purchased CBR-4830 and analyzed its activity in our growth assays used for screening. Consistent with previous results (40), we found that CBR-4830 inhibited the growth of wild-type E. coli and induced cell rounding (Table 1), suggesting that CBR-4830 may indeed target the Rod system. However, CBR-4830 did not suppress mecillinam toxicity, nor was its inhibitory activity suppressed in FtsZ^UP cells (Table 1), suggesting that CBR-4830 may have significant off-target activity. Accordingly, CBR-4830 displayed potent lethality against an FtsZ^UP strain deleted for the mreBCD operon (MIC, 12.5 μg/ml), which was only about four times greater than its MIC against wild-type E. coli (3.1 μg/ml). Also, CBR-4830 inhibited the growth of S. aureus HG003 (MIC, 12.5 μg/ml), a Gram-positive bacterium that lacks MreB (Table 1). These results are in contrast to the relatively high specificity displayed by A22, which has an ∼100-fold reduced potency against ΔmreBCD FtsZ^UP mutant cells (MIC, >100 μg/ml) relative to wild-type (MIC, 1.6 μg/ml) and is inactive against S. aureus HG003 (MIC, >100 μg/ml). Thus, we conclude that CBR-4830 has anti-MreB activity in E. coli but has off-target effects that are likely to prevent the suppression of mecillinam toxicity. The potential for similar off-target activity for the CBR-4830 derivatives in our library may also have contributed to their absence among the positive hits from the screen.

DISCUSSION

In this report, we describe the development and implementation of a high-throughput screen for compounds that disrupt the activity of the Rod system. The screen takes advantage of the conditional essentiality of the complex and the ability of the beta-lactam mecillinam to make its activity toxic (19, 22). These properties allow molecules that promote growth under special circumstances to be identified that are also lethal against wild-type bacteria. Such a screening process has distinct advantages over assays for lethal activity alone because it helps rapidly eliminate nonspecific toxins from the set of screening hits. Thus, downstream efforts can be focused on the characterization of molecules that are on-target and therefore suitable for use as probes for studying cell wall biogenesis and/or use in potential drug development.

Rod system as an antibiotic target.

The Rod system contains several potential drug targets, the most attractive of which are the cell wall synthase enzymes RodA and PBP2, as well as MreB. The other components (MreC, MreD, and RodZ) are also potential targets for antibiotics, but their activities are not well defined. Therefore, the pathway for characterizing and developing molecules active against these proteins is likely to be more difficult than for those hitting the components with activities that can be assayed in vitro.

One potential problem with targeting the Rod system is that its essentiality can be relatively easily suppressed by increasing the cellular FtsZ concentration. However, the resulting spheres grow poorly in the laboratory (19) and have outer membrane barrier defects that would sensitize them to a range of insults inside the host. Indeed, coccoid cells of Shigella flexneri inactivated for the Rod system also display defects in effector secretion and the ability to invade mammalian cells (44). Thus, the activity of the Rod system is likely to be strictly essential during infection such that the suppression observed in laboratory medium is unlikely to be problematic with respect to resistance development in the host. In addition to being attractive targets on their own, RodA and PBP2 have paralogs (FtsW and PBP3) that perform essential roles within the cell division apparatus. Therefore, any molecules identified that target RodA or PBP2 have the potential to also disrupt the activity of FtsW or PBP3, respectively. Such dual targeting activity would be ideal for antibiotic development because it would significantly reduce potential problems with mutational resistance that can cause issues for the development of compounds directed against a single target (45).

Effective screening approach to identify Rod system antagonists.

The successful identification of A22-like hits validates the utility of our screening procedure for identifying antagonists of MreB. However, besides this well-studied class of molecules, no additional inhibitor scaffolds targeting MreB or other components of the Rod system were identified. Our previous analysis of mecillinam toxicity found that it could be readily suppressed by the genetic inactivation of any component of the Rod system (22). Therefore, our screen should theoretically be capable of identifying inhibitors of any member of the machinery, save for the cross-linking activity of PBP2 that is blocked by mecillinam. Why we only identified A22-like hits targeting MreB in the screen is not known. Although we favor the idea that these results reflect limitations in the chemical space covered by the library used for screening, an unexpected bias of the screen for the identification of MreB antagonists cannot be ruled out without control inhibitors that are active against other components of the system, of which none are currently known.

Target of A22 and its potential as an antibiotic.

A22 has been used as a probe to study MreB and Rod system function for some time (18). The evidence for MreB being its primary target is quite strong, including mutational resistance in the mreB gene and structures of MreB with A22 bound near its ATPase active site (18, 25, 29). However, work with the compound MAC13243 has raised questions about the specificity of A22. MAC13243 was originally described as a potential inhibitor of LolA in the lipoprotein transport pathway (26) but was subsequently shown to hydrolyze into an A22-like molecule that is the active component (28). Like MAC13243, LolA overproduction was capable of suppressing A22 lethality and LolA depletion sensitizes cells to A22, suggesting that A22 may also target LolA.

Although it was reasonable to link MAC13243 and A22 with LolA function based on the chemical genetic observations (26, 28), connections based on gene overexpression and depletion can often be indirect. In this case, LolA overproduction is known to induce the Rcs envelope stress response (35). This observation raised the possibility that it might be the Rcs response that is the underlying cause of the protection from compound treatment as opposed to target overproduction. Consistent with this idea, we show here that induction of the Rcs response is required for A22 resistance in cells overproducing LolA. Sensitization to compound treatment upon protein depletion can be similarly indirect, especially for a factor like LolA that plays a central role in envelope assembly. Thus, the hypersensitivity of LolA-depleted cells to A22, which also targets envelope assembly, does not strongly support LolA targeting by A22. A similar level of synergy was also observed between LolA depletion and the cell wall synthesis inhibitor fosfomycin (28), which targets the precursor synthase MurA, suggesting that LolA depletion is likely to sensitize cells to a number of treatments that disrupt cell wall synthesis. Finally, the available biochemical data also do not provide significant support for LolA targeting as a physiologically relevant mode of action for A22. The K_d (dissociation constant) for the interaction between LolA and A22-like molecules has been measured to be 150 to 200 μM (28), which is far greater than the 5.9 μM (1.6 μg/ml) MIC of A22. Finally, the ability of A22 to suppress mecillinam toxicity and promote growth would not make sense if it also caused significant defects in the essential lipoprotein-targeting pathway. Rather, the tight correlation of MIC and mecillinam suppression activity we observe combined with all of the available genetic and biochemical data indicate that A22-like molecules are relatively specific for MreB.

Although A22 appears to be quite specific for MreB in bacterial cells and has been a reliable probe for Rod system activity, it has not yet made it to the clinic as an antibacterial therapeutic. Problems with mammalian cell toxicity (46) as well as with high mutational resistance (2 × 10⁻⁸ for P. aeruginosa) (29) may be the issue. If these liabilities can be overcome, such A22-like derivatives would be attractive leads for future development, either as single agents or as a component of a combination therapy given that Rod defects sensitize Gram-negative bacteria to other antibiotics.

Conclusion.

Here, we report a screen for Rod system inhibitors. Although all of the hits identified were A22-like molecules, only one chemical collection was screened. Despite their seemingly large size, the chemical diversity represented in commercial collections is in actuality quite small. Genetic selections for specific missense mutants often yield “hits” at a frequency of 10⁻⁷ to 10⁻⁹. Thus, it stands to reason that even when utilizing an effective pathway-directed screen, the likelihood of a 10⁵- to 10⁶-member compound library containing even one molecule that answers a specific screen is relatively low. However, given the success of this screening procedure in identifying MreB inhibitors, screens of additional libraries using this assay are worthwhile and have the potential to identify new classes of anti-Rod system compounds that may be useful for antibiotic development.

MATERIALS AND METHODS

Media, bacterial strains, and plasmids.

Cells were grown in lysogeny broth (LB; 1% tryptone, 0.5% yeast extract, 0.5% NaCl ± 1.5% Bacto agar) at 30°C, unless indicated otherwise. The antibiotic concentrations used were tetracycline (Tet), 5 μg/ml; ampicillin (Amp), 50 μg/ml; chloramphenicol (Cam), 25 μg/ml; and kanamycin (Kan), 25 μg/ml. The bacterial strains and plasmids used in this study are listed in Tables S3 and S4, respectively. A description of their construction and a list of the primers used (Table S5) are included in the supplemental material. All deletion strains used were obtained from the Keio knockout collection (47) and confirmed by sequencing.

We purchased A22 [S-(3,4-dichlorobenzyl)isothiourea HCl] from Calbiochem. We purchased compound 1 (catalog no. PH000337; 2,4-dichlorobenzyl imidothiocarbamate), compound 9 (152811; 3,4-dichlorophenylhydrazine HCl), compound 11 (T165948; 4-fluorobenzyl imidothiocarbamate HBr), compound 20 (MP265, PH012544; 4-chlorobenzyle imidothiocarbamate HCl), and compound 22 (T282057; 2-chlorobenzyl imidothiocarbamate) from Sigma. We purchased compound 4 (DP01615; 2,4-dichlorobenzyl butylamine) and compound 21 (S01620; 3-chlorobenzyl imidothiocarbamate) from Maybridge via Fisher Scientific. We purchased compound 10 [3-(3,4-dichlorophenyl)propan-1-amine], compound 19 (Z18380660), compound 23 ([(4-nitrophenyl)methyl]sulfanyl}methanimidamide HCl), and CBR-4830 (EN300-214505) from Enamine. We verified the purchased compounds via liquid chromatography-mass spectrometry (LC-MS). All compounds were dissolved in DMSO (J. T. Baker) to 10 mg/ml or the highest obtainable concentration below 10 mg/ml and stored at −20°C.

High-throughput screening and hit validation.

All screening was conducted at the ICCB-Longwood Screening Facility using 384-well plates (Corning 3701). We performed automated pipetting with a WellMate (Agilent) or Multidrop Combi dispenser (Thermo Fisher) and pin transfer with a customized Epson E2C2515-UL Scara robot paired with an Epson C3-A601S 6-axis robot arm. We screened all compounds in duplicate plates for each condition tested.

(i) E. coli growth inhibition screen. Each compound was pin-transferred (300 nl) into 30 μl of LB with 1% NaCl for a final compound concentration of ≈25 μM. A mid-log culture of strain NR1099 (23) (30 μl) was then added to achieve a final OD₆₀₀ of 0.0002. The final concentrations of the compounds varied based on the stock concentration of the individual libraries but typically fell within 6 to 60 μM (12 to 30 μg/ml). After incubation at 37°C for ∼23 h, the OD₆₀₀ was measured with an EnVision instrument (PerkinElmer). Wells with an average OD₆₀₀ of less than 0.5 were designated hits. The data were collated and a preliminary structural analysis performed to eliminate potential panassay interference (PAINS) compounds (48). Compounds that were annotated as a known antibiotic (e.g., erythromycin) were also removed from the final hit list. All remaining hits that were commercially available were purchased, as well as close analogs when possible. Note that NR1099 only has a moderate permeability defect under the conditions used for screening. It was found to be roughly 2 to 3 times as sensitive as its parent strain to novobiocin, moenomycin, aztreonam, trimethoprim, and ciprofloxacin. Its sensitivity to rifampin was unchanged.

(ii) Mecillinam suppression screen. MG1655 wild type (WT) with pTB63 (contains ftsQAZ operon, pSC101 origin) was grown in LB-Tet overnight at 30°C. The following morning, the culture was diluted to an OD₆₀₀ of 3.3 × 10⁻⁶ in LB with added 40 mM NaNO₃ to serve as a terminal electron acceptor and enhance growth in the 384-well plate format where aeration is poor. An aliquot (30 µl) of this culture (equivalent to ∼150 cells) was added to all wells of a 384-well plate (catalog no. 3701; Corning). The bioactive sublibrary was then pin-transferred to the plates (100 nl of each compound). After incubation at room temperature for ∼30 min, 20 μl of LB containing 6.25 μg/ml mecillinam was applied to each well, resulting in a final volume of 50 μl and a final mecillinam concentration of 2.5 μg/ml. As described above, the final compound concentration varied based on the stock concentration of the individual libraries but typically fell within 2 to 30 μM (4 to 10 μg/ml). The plates were incubated for ∼72h at 30°C and the OD₆₀₀ was measured. The known Rod inhibitor A22 was used as a positive control on each plate (final OD₆₀₀, 0.42 ± 0.02; n = 128), and the compound vehicle, dimethyl sulfoxide (DMSO), was used as the negative control (final OD₆₀₀, 0.05 ± 0.03; n = 480). Hit criteria required that both replicate values be above an x̅ ± 3σ cutoff (OD₆₀₀, ≥0.158), which resulted in 72 hits (5.8%). Taking into account technical replicates, we estimate our false-positive rate to be ∼0.1%.

(iii) FtsZ^UP suppression screen. Strain JAB063 (ΔtolC mutant) containing pTB63 was grown in LB overnight. The resulting culture was then diluted to an OD₆₀₀ of 2 × 10⁻⁶ the following morning, and 50 μl was added to each well. Hit compounds from screen 2 were then pin-transferred (100 nl each) to the wells, and the plates were incubated at 30°C for ∼48h before measuring the OD₆₀₀. Hits were identified as those with replicate values above a x̅ ± 1σ cutoff (OD₆₀₀, ≥0.07), which resulted in 44 hits (61%). This loose cutoff allowed us to cast a wide net for potential rod inhibitors.

Cytological screening.

E. coli MG1655 ΔtolC (JAB063) was grown to mid-log phase in LB, back-diluted to an OD₆₀₀ of 0.05, and applied to all wells of a 96-well plate (Corning). Potential Rod system inhibitors were added in a 2-fold dilution series, with the final concentration ranging from 0.8 to 25 μg/ml. Cells were then grown with aeration at 30°C in a VersaMax microplate reader, with OD₆₀₀ measurements taken every 5 min. After treatment for 5 h, samples at the highest subinhibitory concentration were imaged as described previously (49). Morphologies were assessed qualitatively, and those compounds inducing a spherical or near-spherical cell shape were identified as the final set of hits. In addition to the 44 hits identified as putative Rod inhibitors in screen 3, the remaining 28 compounds were also assessed cytologically. All 28 compounds failed to induce cell rounding.

Compound validation.

Compounds 1, 4, 9 to 11, 19 to 24, and CBR-4830 were purchased (2 to 100 mg) and validated for growth inhibition via broth microdilution in 96- and 384-well plates (Corning) under the same growth conditions described for screen 2. We also validated mecillinam suppression and suppression by FtsZ^UP over a range of concentrations (0.2 to 100 μg/ml) using an HP D300 dispenser for precise titration and a Multidrop Combi dispenser (Thermo Fisher) for automated pipetting. Bright-field microscopy was performed on wild-type MG1655 cells after treatment with 10 to 100 μg/ml of compound for 3 to 4 h at 30°C.

Viability assays.

Wild-type E. coli MG1655 and the ΔrcsF (HC397) mutant containing either an empty vector (pHC800) or lolA expression construct (pJAB107) were grown to an OD₆₀₀ of 1.0 and serially diluted 10-fold, and 5 μl of each dilution was spotted onto LB agar containing 0, 1.25, 2.5, 5, or 10 μg/ml A22 in the presence or absence of 500 μM isopropyl thio-β-d-galactopyranoside (IPTG). Plates were incubated at 30°C for 48 h and photographed. Similarly, wild-type E. coli MG1655 and the ΔmreB mutant (TU233) expressing FtsZ^UP (pTB63) were grown to mid-log phase, diluted to an OD₆₀₀ of 0.1, serially diluted 10-fold, and spotted onto MacConkey agar or LB agar supplemented with 10 μg/ml vancomycin, 50 μg/ml vancomycin, or 1% sodium dodecyl sulfate (SDS). Plates were incubated at 30°C for 48 h and photographed.

Compound hydrolysis and LC-MS analysis.

Stocks of compounds 4 and 19 were diluted to 500 μM in water with 5% acetonitrile and 0.1% trifluoroacetic acid (TFA). This solution was allowed to sit at room temperature for 0.5 or 18 h before LC-MS analysis. LC-MS data were collected using an Agilent 6120 quadrupole LC-MS using electrospray ionization (ESI). LC-MS was conducted with the MS operating in positive-ion mode. The compound solutions (5-μl injections) were separated on a Waters Symmetry Shield RP₁₈ column (5 μm, 3.9 by 150 mm) using the following method: flow rate, 0.5 ml/min; 95% solvent A (H₂O, 0.1% formic acid) for 5 min, followed by a linear gradient of 5% solvent B (acetonitrile, 0.1% formic acid) to 100% B over 15 min.