Human defensin-inspired discovery of peptidomimetic antibiotics

Significance

We report the development of peptidomimetic antibiotics derived from a natural antimicrobial peptide, human α-defensin 5. By engaging multiple bacterial targets, the lead compound is efficacious in vitro and in vivo against bacteria with highly inducible antibiotic resistance, promising a useful therapeutic agent for the treatment of infections caused by antibiotic-resistant bacteria.

Abstract

Antibiotics with multiple mechanisms of action and broad-spectrum are urgently required to combat the growing health threat posed by resistant pathogenic microorganisms. Combining computational and medicinal chemistry tools, we used the structure of human α-defensin 5 (HD5) to design a class of peptidomimetic antibiotics with improved activity against both gram-negative and gram-positive bacteria. The most promising lead, compound 10, showed potent killing of multiple drug-resistant gram-negative bacteria isolated from patients. Compound 10 exhibited a multiplex mechanism of action through targeting membrane components—outer membrane protein A and lipopolysaccharide, as well as a potential intracellular target—70S ribosome, thus causing membrane perturbation and inhibition of protein synthesis. In vivo efficacy, stability, and safety of compound 10 were also validated. This human defensin-inspired synthetic peptidomimetic could help solve the serious problem of drug resistance to conventional antibiotics.

Antimicrobial resistance is a serious threat to human health. By 2050, an estimated 10 million people are projected to die annually of infections complicated by antimicrobial resistance (1–3). Novel antimicrobials with de novo structures, broad-spectrum activity, and multimodal mechanisms are urgently needed (4). However, exhaustion of traditional antimicrobial resources has seriously challenged conventional screening approaches because of the high frequency of rediscovering known compounds, which thus have slowed the development of alternative antibiotics that possess de novo structural scaffolds, novel mechanisms of action (MoA), and extended antimicrobial spectra (5).

The human body is a valuable chemical resource that harbors host-protective cationic antimicrobial peptides (CAMPs), which have been investigated as a class of promising templates for antibiotic development (6). However, natural CAMPs are often unsuitable for direct clinical translation for various reasons, including suboptimal antimicrobial activity, poor pharmacokinetic properties, low proteolytic stability, and high costs for production (7, 8). To improve the pharmacological viability of natural CAMPs, a number of peptidomimetic chemistries have been successfully developed, among which are “foldamers” of unnatural amino acids mimicking the structure and function of naturally occurring CAMPs (7–12), small molecule-based peptidomimetics, such as the flavonoid scaffold decorated with cationic residues/moieties (13–15), and protegrin-derived β-hairpin peptidomimetics (5, 16).

Our group has exploited a different peptidomimetic approach using human endogenous enteric defensins. Human α-defensin 5 (HD5), a natural CAMP secreted by intestinal Paneth cells (17, 18), is bactericidal to both gram-negative and gram-positive bacteria through a well-established MoA of membrane perturbation (19). The development of HD5 is hampered by unfavorable pharmacological properties along three major axes: 1) insufficient activity to meet requirement of clinical therapeutics; 2) inhibition of activity by serum and physiological ionic microenvironment; and 3) a relatively large size and complex structure that challenges chemical modification and synthesis. To date, no approach has resolved these critical issues ideally. From our perspective, overcoming these challenges necessitates simplification of full-length HD5 into a simple mini-HD5 that allows highly accessible synthesis and subsequent peptidomimetic efforts to improve drug-like properties, including bactericidal activity, physicochemical/biological stability, and cost of synthesis. Guided by these concepts, we set out to simplify and redesign HD5 into an antimicrobial compound that overcomes the shortcomings of the parent defensin HD5.

Here, we successfully redesigned HD5 as a class of peptidomimetic antibiotics by combining computational and medicinal chemistry approaches. A structure–activity relationship (SAR) study identified peptidomimetic antibiotic 10 as the “best-in-class” candidate that showed a synthesis-friendly linear structure, powerful bactericidal action, and favorable therapeutic index. Further efforts found 10 interrupted the membrane integrity and preserved the membrane activity of its parent defensin HD5 through targeting outer membrane protein A (OmpA) and lipopolysaccharide (LPS) on the surface of gram-negative outer membrane (OM). In addition, we also found that 10 targeted bacterial 70S ribosomes. Finally, in vivo efficacy and toxicity were validated in infected and healthy mice, respectively. The in vivo stability was assessed through pharmacokinetics. The flow-process pipeline for this rational design of HD5 is summarized in Fig. 1A.

Fig. 1.

Structure-guided discovery of pharmacophores and lead scaffolds in HD5. (A) Schematic illustration of the pipeline for discovering/designing HD5-inspired peptidomimetic antibiotics. (B) Cartoon models of HD5 highlighting its potential pharmacophore motifs. Pharmacophores 1 and 2 are colored green and cyan, respectively. Zoomed views are presented to indicate the biointerfaces and structures of pharmacophores in which residues are labeled. The 2F_o–F_c electron density maps are shown around the pharmacophores (gray mesh), and electron density map information was retrieved from the Protein Data Bank (http://www.rcsb.org/pdb/explore/explore.do?structureId=1ZMP). (C) Design and synthesis of lead peptides based on the pharmacophore templates. (D) In vitro bactericidal activity of lead peptides against members of ESKAPE pathogens. Data are presented as mean ± SD. The EC₅₀ values of the two lead peptides are indicated for each pathogen, which are fitted by using a four-parameter dose–response model. The experiment was performed in triplicate.

Materials and Methods

All peptidomimetic chemicals were synthesized by solid-phase peptide synthesis and purchased from Shanghai Apeptide. Computational studies and biological assays, including molecular docking, molecular dynamics simulations, bacterial culture, bactericidal assays, and cell viability assays, were performed by using standard approaches, with all methods detailed in SI Appendix. Animal studies were approved and guided by the Animal Advisory Committee at Zhejiang University.

Results

Structure-Guided Identification of Pharmacophores and Lead Compounds.

Our pipeline to design defensin-inspired antibiotics started with the identification of pharmacophores that form the HD5 interactive biointerface with multiple bacterial targets, including bacterial envelopes, LPS, OmpA (e.g., Acinetobacter baumannii), and lipid II (e.g., Staphylococcus aureus) (Fig. 1A and SI Appendix, Fig. S1) (17, 20–24). Previous research has demonstrated that residues A¹, T², R⁶, R¹³, and R³² are crucial for HD5 to adsorb to the LPS-containing OM, and basic residues (R⁶, R⁹, R¹³, R²⁵, R²⁸) for HD5 to electrostatically interact with the inner membrane (IM) (25, 26). To fully map out pharmacophore biointerfaces, we modeled HD5 interactions with A. baumannii OmpA and S. aureus lipid II using computational biology approaches (SI Appendix, Figs. S2 and S3). After molecular dynamic analysis, the contact residues were conclusively identified (SI Appendix, SI Materials and Methods, and Figs. S2 and S3, and Movie S1). Our structure-guided analysis in combination with the previous findings by others identified motifs T²-C³-Y⁴-C⁵-R⁶-T⁷-G⁸ in the β1 strand and R²⁵-L²⁶-Y²⁷-R²⁸ in the β3 strand as potential pharmacophores responsible for the antimicrobial activity of HD5 (Fig. 1B). We synthesized lead peptides 1 and 2 corresponding to the two identified pharmacophores (Fig. 1C), and both peptides exhibited bactericidal activity (Fig. 1D). Lead peptide 1 exhibited considerable activity against S. aureus in vitro (EC₅₀ ∼7.043 ± 0.2440 μM) but 10-fold higher EC₅₀ against methicillin S. aureas (MRSA), whereas lead peptide 2 had comparable activity against both pathogens (EC₅₀ ∼15.72 ± 1.764 μM and 17.01 ± 1.104 μM). Lead peptide 2 was more active than 1 against Escherichia coli (EC₅₀ ∼12.08 ± 1.033 μM vs. 28.25 ± 3.271 μM), but both exhibited similar activity against Klebsiella pneumoniae. Notably, lead peptide 1 did not show any bactericidal activity against Pseudomonas aeruginosa and A. baumannii within the range of concentrations used (EC₅₀ > 100 μM), while lead peptide 2 demonstrated concentration-dependent killing of the two pathogens, with EC₅₀ values of ∼17.57 ± 0.6533 μM and ∼21.54 ± 1.683 μM, respectively. Since the two lead peptides were less effective in bacterial killing than their parent defensin HD5 (SI Appendix, Fig. S4), the use of medicinal chemistry tools was warranted for functional improvement. Considering its bactericidal effect and broad-spectrum activity, we selected peptide 2 as the lead compound for structural optimization.

Peptidomimetic Chemistry-Guided Structural Optimization and SAR.

The tetra-peptide backbone R²⁵-L²⁶-Y²⁷-R²⁸ of lead compound 2 consists of positively charged, hydrophobic, and aromatic groups that could support access to bacterial envelopes. We maintained this core structure in the next round of design as it likely contributed to the superior bactericidal activity of lead compound 2. The R₁/R₂ moieties located at the N and C termini were used for chemical functionalization and subsequent structural optimization (Fig. 2A). We employed three peptidomimetic strategies, including appending potent electron-donating or hydrophobic moieties (e.g., long alkyl groups), substitution by aromatic groups, and replacement by heteroaromatic groups (Fig. 2A), which generated a family of peptidomimetic antibiotics (identifier 3∼10) (Fig. 2B).

Fig. 2.

Structural optimization of the lead scaffold by peptidomimetic chemistry. (A) Proposed peptidomimetic-chemistry pipeline for structurally reprogramming the parent scaffold of the lead compound. (B) Design cycle and chemical structures of a new class of peptidomimetic antibiotics generated from the peptidomimetic strategy.

Improved bactericidal activity was achieved via an N-terminal cap of a long alkyl group, including palmityl, lauryl, or myristyl, which generated antibiotics 3, 4, and 5, respectively. All three peptidomimetic antibiotics had significantly stronger bactericidal activity than 2 (Fig. 1D and SI Appendix, Fig. S5). Notably, 3 is at least 200-fold more potent than HD5 against K. pneumoniae (EC₅₀ ∼0.2298 ± 0.01212 μM vs. 26.43 ± 2.415 μM) and P. aeruginosa (EC₅₀ ∼ < 0.195 μM vs. 12.53 ± 1.507 μM). Interestingly, shortening the alkyl group in 3 to a lauryl group, resulting in compound 4 (Fig. 2B), increased the compound’s performance against gram-positive pathogens (S. aureus and MRSA); however, this modification attenuated the activity against gram-negative pathogens, though with the benefit of decreasing its cytotoxicity to mammalian cells (SI Appendix, Fig. S6 A and B). Replacing palmityl with myristyl yielded 5, which improved the activity against S. aureus and MRSA, while maintaining the activity against gram-negative ESKAPE members (Enterococcus faecium, S. aureus, K. pneumoniae, A. baumannii, P. aeruginosa, Enterobacter spp.), including E. coli, K. pneumoniae, and A. baumannii. Antibiotic 5 also showed lower cytotoxicity than 3 and lower concentration for killing 99% bacteria (LC99) for these six bacteria than did 3, 4, and HD5 (SI Appendix, Figs. S6C and S11).

Replacing the long alkyl chain with 2-(4-(tert-butyl)phenyl)-4-methylpyrimidine-5-carboxylic acid, a heteroaromatic group previously used to replace the lipopeptide tail in arylomycin optimization (27), led to a decrease in bactericidal activity (SI Appendix, Fig. S7A). Nevertheless, the resultant compound 6 still exhibited stronger activity than 2 (Fig. 1D and SI Appendix, Fig. S7A), especially against S. aureus and E. coli, with ∼30-fold enhancement of activity. Importantly, compound 6 did not show any cytotoxicity to the three mammalian cell lines at the range of concentrations tested, suggesting a favorable safety profile (SI Appendix, Fig. S8A). To further improve bactericidal activity, we appended the heteroaromatic group to the C terminus of 5 and generated compound 7, which showed systematic decreases in EC₅₀ compared with HD5 and compounds 2 and 5 (Fig. 1D and SI Appendix, Figs. S4, S5C, and S7B). However, 7 exerted higher cytotoxicity to the three mammalian cells than all of the above compounds (SI Appendix, Fig. S8B).

To overcome the safety shortcomings observed with 7, we tested aromatic substitutions. Replacing the heteroaromatic group with p-aminobenzenesulfonyl dramatically alleviated the cytotoxicity of 7 (SI Appendix, Fig. S9A). However, the resulting compound 8 showed at least a threefold decrease in bactericidal activity against MRSA, E. coli, K. pneumoniae, and P. aeruginosa, while comparable with compound 7 against S. aureus and A. baumannii (SI Appendix, Fig. S10A). Swapping the N- and C-terminal modifications of 8 resulted in 9, which exhibited a favorable safety profile with peritoneal macrophages and hepatocyte BNL CL.2 cell line, but an eightfold enhancement in cytotoxicity to the fibroblast 3T3 cell line (SI Appendix, Fig. S9B). The antimicrobial activity of 9 decreased substantially compared with that of 8 (SI Appendix, Fig. S10B). Unexpectedly, substituting the myristyl group of 9 with palmityl yielded a structural analog termed 10 that showed the highest activity against ESKAPE pathogens (SI Appendix, Fig. S10C) with relatively low cytotoxicity and the lowest LC99 values among other family members (SI Appendix, Figs. S9C and S11). Considering its potency and a wide therapeutic window (EC₅₀ < 0.50 μM and CC50 > 20 μM, safety index > 40) (Fig. 3A), antibiotic 10 was selected as the “best-in-class” antibiotic candidate for in-depth investigation.

Fig. 3.

In vitro validation and SAR studies. (A) EC₅₀-CC₅₀ plots present the EC₅₀ values of peptidomimetic antibiotics against members of ESKAPE pathogens and the CC₅₀ values of peptidomimetic antibiotics to mammalian cells. Red points represent peptidomimetic antibiotics with a high therapeutic index. (B) In vitro antimicrobial activity of 5, 10, and Imipenem Cilastatin against clinically isolated bacteria. Data are presented as mean ± SD. The EC₅₀ values of the tested compounds are indicated for each pathogen, which are fitted by using a four-parameter dose–response model. All experiments were performed in triplicate.

Peptidomimetic 10 Is an Effective Broad-Spectrum Antibiotic against High-Risk Clinical Isolates.

SAR-directed drug screening clearly indicated 10 is the most powerful one among these defensin-inspired peptidomimetics against ESKAPE pathogens with favorable therapeutic index (Fig. 3A). To further assess its clinical potential, we tested the bactericidal activity of the compound against clinically isolated pathogens with high risk and lethality (Fig. 3B). Both 10 and its homologous compound 5 exhibited potent killing of clinical isolated Listeria monocytogenes, S. aureus, and Staphylococcus epidermidis, in which 10 shows significantly superior activity, suggesting it is also best-in-class in clinical isolates. Nevertheless, compound 10 did not exhibit significant advantages relative to Imipenem Cilastatin in combatting the three isolates, especially for S. aureus and S. epidermidis. For Salmonella typhimurium and other clinically isolated drug-resistant bacteria, including carbapenem-resistant (CR) K. pneumoniae, CR-P. aeruginosa, pandrug-resistant (PDR) Providencia rettgeri and PDR-A. baumannii, Imipenem Cilastatin showed very limited activity, while 10 still maintained high-performance killing comparable to the last-resort antibiotic polymyxin B (PMB) (Fig. 3B and SI Appendix, Fig. S12). These results all demonstrated that 10 was a promising candidate for further development.

Peptidomimetic Antibiotic 10 Retains the Membrane Activity of Defensin.

Given the broad-spectrum and potent activity against gram-negative bacteria that are difficult-to-treat in clinic, we sought to uncover the MoA of 10 by using bacterial model of E. coli ATCC25922. Membrane perturbation represents the most common mechanism against gram-negative bacteria for defensins. Thus, we examined if 10 retains the membrane activity of its parent defensin HD5 to penetrate the membrane barrier of gram-negative bacteria. As expected, addition of 10 to E. coli indeed resulted in irregularly shaped holes in the bacterial membrane barrier, similar to those generated by HD5 and PMB (a membrane-activity lipopeptide), while E. coli cells treated by kanamycin (a nonmembrane-activity antibiotic) preserved integral membrane architecture (Fig. 4A). Fluorescent probe 1-N-phenylnaphtylamine (NPN) assay indicated that treatment of 10, HD5, or PMB increased the permeability of bacterial OM in a dose-dependent manner, whereas kanamycin did not (Fig. 4 B–E). These findings all indicated that 10 retains the membrane activity of its parent defensin HD5.

Fig. 4.

Peptidomimetic antibiotic 10 exhibits membrane-perturbance activity to gram-negative OM. (A) Scanning electron microscopy of E. coli ATCC25922 untreated or incubated with different types of antimicrobials including 10, HD5, PMB, and kanamycin. (Scale bars, Left, 2 μm.) Red frames highlight the high-magnification fields. (Scale bars, 500 nm.) (B–E) OM permeability of E. coli ATCC25922 probed with NPN after treating with different antimicrobials including 10 (B), HD5 (C), PMB (D), and kanamycin (E) for 1 h. Permeability of OM is positively corelated with the fluorescent intensity (FI) generated from NPN. The fluorescence data of NPN were normalized to relative FI. Data are presented as mean ± SD. All experiments were performed in triplicate.

Peptidomimetic Antibiotic 10 Exerts Membrane Activity by Interacting with Surface Targets: OmpA and LPS.

To further elucidate the molecular targets of 10 responsible for the membrane perturbation, a transcriptomic assay was performed to reveal the alterations of E. coli in response to 10. Results demonstrated that the transcriptome alterations included multiple envelope stress pathway-related genes, indicating the disruption of envelope homeostasis (Fig. 5A) (28). Intriguingly, we found the expression of genes related to OM proteins (OMPs) and LPS pathway was also altered significantly (Fig. 5B), corresponding to the surface targets of HD5: OmpA and LPS. We initially focused on OmpA, the major OMP of E. coli contributing to maintenance of OM integrity (29). Full-length OmpA comprises two independent folded structures including N-terminal β-barrel domain and C-terminal periplasmic domain (30). To assess the possibility of OmpA as one of membrane targets of 10, we respectively expressed the two domains of E. coli OmpA in a prokaryotic expression system. Adding purified OmpA β-barrel domain significantly antagonized the bactericidal activity of low-dosage 10 in a concentration-dependent manner (Fig. 5C and SI Appendix, Fig. S13 A and B). In contrast, addition of OmpA periplasmic domain failed to affect 10’s activity (Fig. 5D and SI Appendix, Fig. S13 C and D). More importantly, surface plasmon resonance (SPR) assays proved that 10 could effectively bind to OmpA β-barrel domain with a high affinity (K_D ∼96 nM), similar to its parent defensin HD5 (K_D ∼201 nM) (Fig. 5 E and F). However, OmpA periplasmic domain was not strongly bound by 10 and HD5, as indicated by their rapid dissociation (SI Appendix, Fig. S14).

Fig. 5.

Peptidomimetic antibiotic 10 exerts membrane activity by targeting OmpA and LPS. (A) RNA-seq profile showing the alterations of bacterial envelope stress-related genes in E. coli treated with 10; data are presented using heatmap, each group was repeated in triplicate. (B) RNA-seq profile uncovering the significant alterations of genes related to OMPs pathway and LPS pathway, data are presented using heatmap, each group was repeated in triplicate. (C) In vitro bactericidal activity of 10 against E. coli in presence of different concentrations of recombinant OmpA β-barrel domain. Addition of OmpA β-barrel domain antagonized activity of low-concentration 10 in a concentration-dependent manner. Data are mean ± SD with triplicate experiments. (D) In vitro bactericidal activity of 10 against E. coli in the presence of different concentrations of recombinant OmpA periplasmic domain. Addition of OmpA periplasmic domain did not affect activity of low-concentration 10. Data are mean ± SD with triplicate experiments. (E) SPR analysis of interaction between 10/HD5 and OmpA β-barrel domain. (F) Table recording the interactive parameters calculated from SPR assay, including the equilibrium dissociation constant (K_D), association constant (K_a), and dissociation constant (K_b). (G) OM permeability of wild-type E. coli (E. coli^WT) and OmpA-deficient E. coli (E. coli^ΔOmpA) probed with NPN, after treating with 10 for 1 h. Permeability of OM is positively corelated with the FI generated from NPN. The fluorescence data of NPN were normalized to relative FI. (H) In vitro bactericidal activity of 10 against wild-type E. coli (E. coli^WT) and OmpA-deficient E. coli (E. coli^ΔOmpA), the experiments were repeated nine times. (I) In vitro virtual-colony count assay determining fold-change in MBCs of 10, PMB, and kanamycin against E. coli in the presence of different concentrations of LPS. Addition of LPS significantly improved the MBCs of 10 and PMB, but not kanamycin’s MBC. (J) Fluorescent quenching analysis of interaction between 10 and LPS. Fluorescent intensities of 10 were significantly quenched by adding LPS in a concentration-dependent manner. Experiments were repeated in triplicate. (K) Table recording the binding parameters calculated from fluorescent quenching assay, including the quenching constant (K_SV), association constant (K_a), dissociation constant (K_b), and equilibrium dissociation constant (K_D). (L) Limulus amebocyte lysate assay detection of LPS neutralization of compound 10. (M) OM permeability of E. coli probed with NPN, after treating with 10 in absence or presence of Mg²⁺ (50 μg/mL) for 1 h. Permeability of the OM is positively corelated with the FI generated from NPN. The fluorescence data of NPN were normalized to relative FI. (N) In vitro bactericidal activity of 10 against E. coli in the absence or presence of Mg²⁺ (50 μg/mL). (O) Fold-change in MBC of 10 against E. coli in the absence or presence of Mg²⁺ (50 μg/mL). Data are mean ± SD with triplicate experiments. Statistical significance was calculated by one/two-way analysis of variance with the Bonferroni correction for multiple comparisons. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001; ns, not significant.

We then investigated the OM permeability on OmpA-knockout E. coli (E. coli^ΔOmpA). Results showed that the enhancement of OM permeability resulted from 10 challenge significantly decreased on E. coli^ΔOmpA as compared with wild-type E. coli (E. coli^WT) (Fig. 5G). Additionally, deleting OmpA from E. coli indeed attenuated the bactericidal activity of low-dose 10 with approximately twofold increasement in EC₅₀ (Fig. 5H and SI Appendix, Fig. S15A). These findings verified OmpA as one of the surface targets of 10, in which the 10-binding domain was the N-terminal β-barrel structure. Nevertheless, comparing bacterial cells with/without OmpA showed insignificant change in minimal bactericidal concentration (MBC) of 10 (SI Appendix, Fig. S15B), suggesting that 10 is likely to have a second, OmpA-independent MoA to disrupt OM.

Thus, we next tested if 10 also targeted LPS similarly to other CAMPs’ membrane activity. Adding purified LPS potently antagonized bactericidal activity of 10 in a concentration-dependent manner (SI Appendix, Fig. S16A). This inhibitory effect to LPS was dramatic, as LPS at concentration of 25 μg/mL increased approximately sixfold the MBC of 10 and fourfold the MBC of PMB, but showed no effect on antimicrobial activity of kanamycin, suggesting LPS might be an important surface target of 10 (Fig. 5I and SI Appendix, Fig. S16). We further assessed the molecular interaction between 10 and LPS by a fluorescent quenching assay. LPS effectively quenched the fluorescence of compound 10 in a concentration-dependent manner with a high quenching constant (K_SV) (Fig. 5 J and K) that reported strong affinity between LPS and 10 (K_D ∼0.793 nM) (Fig. 5 J and K). To test the ability of 10 to inactivate LPS, we performed a limulus amebocyte lysate assay that showed LPS was dose-proportionally neutralized by 10 (Fig. 5L). Structurally, LPS is composed of an O-antigen, a core polysaccharide, and a lipid A domain, in which lipid A represents the toxic core of LPS and usually is the binding domain of LPS-targeting antibiotics (e.g., PMB) (31, 32). Consequently, we next examined whether 10 neutralizes LPS through acting on lipid A domain. It is reported that divalent ions (e.g., Mg²⁺) can bridge and neutralize the negatively charged lipid A domain in LPS by electrostatic interaction, thereby contributing to tight packing and low permeability (33, 34). NPN fluorescent probe assay demonstrated that incorporation of Mg²⁺ indeed protected the integrity of E. coli OM from the attack of 10 (Fig. 5M). We thus investigated the effect of Mg²⁺ on the bactericidal activity of compound 10. As expected, incorporation of Mg²⁺ significantly diminished 10’s activity with fourfold increase of MBC (Fig. 5 N and O). These results revealed that both LPS and OmpA were targets for 10 to exert membrane activity.

In summary, our defensin-inspired antibiotic 10 perfectly preserved the membrane activity of its parent HD5, with a similar MoA of targeting the OmpA β-barrel domain and LPS lipid A domain.

Bacterial 70S Ribosome Is a Potential Intracellular Target of Peptidomimetic Antibiotic 10.

Although we uncovered the MoA of membrane activity, we found the RNA-sequencing (RNA-seq) profile of E. coli treated by 10 was significantly different from that of HD5 and PMB (SI Appendix, Fig. S17). Hence, we asked if there might be another distinct MoA contributing to 10’s antimicrobial action, in addition to membrane perturbation. Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis revealed that treatment by 10 led to large up-regulation of numerous genes encoding 70S ribosomal subunits that all were enriched in the ribosome pathway (Fig. 6 A and B). Such an enrichment map was also distinct from that of HD5 and PMB (Fig. 6B and SI Appendix, Fig. S18). Considering that bacterial ribosome has already been identified as a mature and highly druggable target for antibiotics (e.g., kanamycin) (35), we rationally speculated 70S ribosome might be an independent intracellular target of 10. To verify this, we performed an antimicrobial block assay to test if exogeneous E. coli 70S ribosome could block the bactericidal effect of compound 10. Surprisingly, incorporating E. coli 70S ribosome largely inhibited the activity of 10 at an extremely low concentration, with approximately eightfold enhancement of MBC value, this inhibitory action was stronger than that of LPS and OmpA (Fig. 6C and SI Appendix, Fig. S19A). As expected, we found E. coli 70S ribosome antagonized kanamycin but failed to affect activity of PMB (Fig. 6C and SI Appendix, Fig. S19 B and C). To investigate in-depth molecular interaction between 10 and E. coli 70S ribosome, we performed a fluorescent quenching assay. Results showed 70S ribosome also dramatically quenched the fluorescence of 10 in a concentration-dependent manner, with a K_sv ∼1.11 × 10⁷ M⁻¹ that improved ∼30-fold compared with that of LPS (Fig. 6 D and E). In addition, 10 also showed stronger affinity to 70S ribosome than that of LPS (K_D ∼0.140 nM) (Fig. 6E). Additional in vitro translation assay proved compound 10 was able to potently inhibit protein synthesis of E. coli 70S ribosome, with a nanomolar IC₅₀ (∼19.51 ± 2.771 nM) (Fig. 6F). Collectively, these findings implicated bacterial 70S ribosome as a potential intracellular target of compound 10 that might confer the antimicrobial activity by impairing protein synthesis.

Fig. 6.

Peptidomimetic antibiotic 10 targets bacterial 70S ribosome and inhibits protein synthesis. (A) RNA-seq profile showing the alterations of genes encoding ribosomal subunits in E. coli treated with 10, data are presented using heatmap, each group was repeated in triplicate. (B) KEGG enrichment analysis of differentially expressed genes of RNA-seq profile in E. coli treated with 10. (C) In vitro virtual-colony count assay determining fold change in MBCs of 10, PMB and kanamycin against E. coli in presence of E. coli 70S ribosome. Addition of 70S ribosome significantly improved the MBCs of 10 and kanamycin, but not PMB’s MBC. Data are mean ± SD with triplicate experiments. **P < 0.01, ****P < 0.0001; ns, not significant. (D) Fluorescent quenching analysis of interaction between 10 and E. coli 70S ribosome. FIs of 10 were significantly quenched by adding 70S ribosome in a concentration-dependent manner. (E) Table recording the binding parameters calculated from fluorescent quenching assay, including the quenching constant (K_SV), association constant (K_a), dissociation constant (K_b), and equilibrium dissociation constant (K_D). (F) In vitro translation assay demonstrating the inhibition of protein synthesis ability of 70S ribosome with the treatment of 10. Data are mean ± SD with triplicate experiments. Statistical significance was calculated by one/two-way analysis of variance with the Bonferroni correction for multiple comparisons.

In Vivo Biological Profiles.

To assess the in vivo therapeutic performance of 10, animal models of sepsis and bacterial pneumonia were established. In E. coli-septic mice, untreated controls all died within 12 h, while administration of 10 manifested therapeutic efficacy comparable to the last-resort antibiotic PMB (Fig. 7A). Compound 10 considerably decreased the bacterial burden in the blood and organs (Fig. 7 B and C) and, consequently, attenuated multiorgan injury (SI Appendix, Fig. S20). Mice with K. pneumoniae-induced pneumonia displayed high mortality with 100% lethality within 48 h. Treatment with 10 resulted in an 80% survival rate; PMB treatment resulted in a 100% survival rate (SI Appendix, Fig. S21A) but PMB was ineffective in S. aureus-induced pneumonia (SI Appendix, Fig. S21B), where 10 showed 100% survival rates comparable to vancomycin (SI Appendix, Fig. S21B). As expected, the bacterial burden in bronchoalveolar lavage fluid was in keeping with the results of survival rates (SI Appendix, Fig. S21 C and D). The total cell count and protein concentration in bronchoalveolar lavage fluid were also significantly reduced in animals treated with 10 (SI Appendix, Fig. S21 E and F), indicating decreased damage to pulmonary barriers.

Fig. 7.

Protection of mice against E. coli-challenged septicemia. (A) Kaplan–Meier percent survival analysis of C57BL/6 mice challenged with E. coli and concomitantly treated with compound 10 or PMB. n = 6 per group. (B) Determination of bacterial burden in blood from E. coli-challenged mice with/without treatment of compound 10 or PMB. n = 6 per group. Dashed line indicates the limit of detection. (C) Determination of bacterial burden in organs, including liver, spleen, lung, and kidney, from E. coli-challenged mice with/without treatment of compound 10 or PMB. n = 6 per group. Dashed line indicates the limit of detection. Statistical significance was calculated by one-way analysis of variance with the Bonferroni correction for multiple comparisons. *P < 0.05, **P < 0.01, ***P < 0.001; ns, not significant.

To assess the in vivo safety of 10, healthy C57BL/6 mice were intraperitoneally injected with high doses of 10. Mice showed 100% survival after challenges with extremely high doses of 64, 128, and 264 mg/kg of 10 (SI Appendix, Fig. S22A), suggesting its pharmacologically acceptable safety profile. We then evaluated the in vivo stability through a single-dose pharmacokinetic study. Injection of 10 generated a peak plasma concentration of 3,455 ng/mL and a half-life of 2.35 h (SI Appendix, Fig. S22B), indicating a favorable in vivo stability.

Discussion

Resistance to antimicrobials is an increasing threat to global healthcare due to the emergence of drug resistance and limited development of new antibiotics (1, 3, 5). Valuable new antibiotics now include highly drug-like peptide antimicrobials (e.g., teixobactin and darobactin) from bacterial resources based on biosynthetic gene clusters (28, 36, 37). Application of artificial intelligence methods has advanced antibiotic discovery, as exemplified by introducing known compounds used in other fields as antibiotics through deep learning (38). Incorporating modular design and chimeric chemistry can transform “old” antibiotics to “new” antibiotics, thereby mitigating antimicrobial resistance (39, 40). Despite these significant advances, the huge demand for novel antibiotics necessitates continuous discovery/design of alternatives. A potential resource is the antimicrobial chemistry of human endogenous antibiotics, such as HD5 (6, 17), as these CAMPs often contain regions of cationicity and hydrophobicity (e.g., amphiphilicity) segregated in a stable structure and are thought to kill bacteria by primarily targeting their negatively charged membrane, thus being less likely to induce antibiotic resistance (6, 41). However, intrinsic shortcomings of CAMPs, such as susceptibility to proteolysis, inadequate bactericidal activity, and difficult chemical synthesis and modification hinder direct translation to the clinic (6, 7, 41).

Several peptidomimetic strategies, such as foldamers and flavonoid-based mimics, have alleviated some of the intrinsic shortcomings of these CAMPs (7, 9, 11, 13). The “foldamers” approach incorporates different types of unnatural residues/moieties to restrict backbone flexibility, thus stabilizing specific conformations and enhancing metabolic stability (7–9). By fine-tuning the relative positions of cationic and hydrophobic residues on the peptide backbone, improved antimicrobial activity and enhanced selectivity could be achieved (42). This strategy led to the design of magainin foldamers, such as helical peptoid mimics and rationally stapled magainin II (10–12). Similarly, the small-molecule–based approach mainly considers the factors of scaffold rigidity, cationicity, and hydrophilicity, as exemplified by rigid and hydrophobic flavonoids decorated with cationic groups (13–15). Of note, although unnatural peptides comprising β-amino acids (i.e., β-peptides) are resistant to proteolysis (8, 43, 44), they do not mimic helical conformations adopted by some parent CAMPs (44, 45), in contrast to α/β-peptides that contain both α- and β-residues (44, 46). Srinivas et al. (16) introduced a rigid β-turn, D-Pro-L-Pro, to mimic the β-hairpin structure of protegrin I, and significantly improved peptide stability. The resultant product Murepavadin and its inhaled formulation entered clinical trials for the treatment of P. aeruginosa pneumonia (5, 16).

Structural classes of antimicrobial peptides include highly disordered linear peptides that adopt an α-helical structure upon membrane binding (e.g., magainin), β-hairpin peptides stabilized by two intramolecular disulfides (e.g., protegrins), and mammalian defensins that fold into a three-stranded β-sheet conformation stabilized by three intramolecular disulfide bridges (17, 47). Compared with defensins, magainin, and protegrins are relatively simple at the structural level and their peptidomimetic strategies are understandably focused on localized modifications to improve backbone rigidity and, thus, proteolytic stability and to modulate amphiphilicity for enhanced bioactivity and reduced cytotoxicity (7, 13, 42). By contrast, human α-defensins are only moderately cationic and, in fact, cationicity is not the most important functional determinant for their functions (48).

Furthermore, while human α-defensins kill gram-negative bacteria through membrane disruption in a largely structure-independent fashion, they kill gram-positive strains by sequestering lipid II to inhibit bacterial cell wall synthesis (24, 49), which is structure-dependent. Extensive mutational studies of the prototypic human α-defensin HNP1 and HD5 elucidated hydrophobic residues at the C terminus, selective Arg residues, disulfide-bonding and dimerization as functionally critical elements (50–53). In particular, hydrophobicity and the ability of human α-defensins to dimerize, oligomerize, and multimerize upon target binding are thought to be an overriding determinant for their functional pleiotropism (54), as exemplified by a multiplicity of molecular targets of α-defensins, including lipids and microbial membranes, bacterial cell wall components, bacterial toxins and virulence factors, viral proteins, carbohydrates and glycoproteins, cellular receptors and host proteins, and nucleic acids (48). In light of the substantive differences between defensins and other classes of CAMPs at the structural, functional, and mechanistic levels, we chose a global peptidomimetic strategy, where structure-guided molecular dissection of defensin-target interactions allowed us to make novel contributions to the fertile field of rationally designed antibiotic mimics.

Inspired by the pharmacophore concept in small-molecule drug design (55, 56), we mapped the interactions between HD5 and its potential targets using computational chemistry approaches to identify key pharmacophore fragments in HD5, resulting in the simple tetra-peptide scaffold R²⁵-L²⁶-Y²⁷-R²⁸ for functionalization and optimization. In contrast to foldamers and flavonoid-based mimics whose backbones are rigidified, our core scaffold is flexible to ensue adaptive conformations for specific and high-affinity molecular interactions. Since hydrophobicity dominates α-defensin functionality, we reasoned that terminal modification of the scaffold by nonpolar moieties, such as long alkyl chains and aromatic groups, would not only enhance peptide killing of bacteria in general, but also modulate its selectivity toward gram-positive and gram-negative bacteria and possibly its cytotoxicity profile as well. This approach was validated by our in vitro SAR and in vivo efficacy studies. Of note, terminal modifications of the tetra-peptide scaffold by hydrophobic moieties did not cause functionally adverse reduction in solubility and likely contributed to enhanced proteolytic resistance and improved metabolic stability of peptide compounds, as verified by pharmacokinetic studies. Since the scaffold of compound 2 is sufficiently small, the addition of bulky structural elements may lead to a different class of compounds with significantly altered activity profiles and modes of action, an important lesson for the design of defensin-derived peptidomimetics. Our core scaffold may be a useful template for the design of other classes of peptidomimetic antibiotics as well.

Despite previous efforts in designing HD5 derivatives through residue substitution and nanotechnology (20, 57), their pharmacokinetic properties and efficacy in combatting high-risk CR-/PDR-clinical isolates have yet to be demonstrated. Of the newly designed antibiotics, 10 exhibited high activity and a wide therapeutic window. Importantly, 10 was also effective against a series of clinically isolated CR- and PDR-superbugs, against which the first-line antibiotic Imipenem Cilastatin is ineffective. Notably, the CR-K. pneumoniae and PDR-P. rettgeri were isolated from two intensive care unit patients, both of whom suffered bacterial infection-induced critical illnesses, such as multiorgan dysfunction syndrome and septic shock (for details, see SI Appendix, Tables S1 and S2). Our peptidomimetic antibiotics may offer a possibility to manage these critical conditions in the future. Compound 10 has a unique structure, consisting of an N-terminal p-aminobenzenesulfonyl group, a core peptide scaffold, and a palmityl chain. Despite its origin from HD5, 10 is quite different from the parent defensin in both biological activity profiles and MoAs. Additional studies are warranted to elucidate the molecular basis underlying the functional and mechanistic differences between 10 and HD5. It is conceivable that our concept for redesigning HD5 can be generalized into a standard protocol to guide redesign of other defensins from hits to lead compounds to drug candidates.

Against clinically “difficult-to-treat” pathogens, especially for gram-negative bacteria, multimodal antibiotics with double MoAs or multiple targets may avoid the high-frequency antimicrobial resistance/persistence manifested by evolving gram-negative bacteria (58, 59). Specifically, gram-negative bacteria possess a tight membrane barrier that contains a lipidic IM and an LPS-coated OM (60). Because of the existence of OM, compounds with hydrophobicity or a relatively large size (∼ >600 Da) are hindered from entering bacterial cells (5). Hydrophilic compounds with low molecular weight can penetrate the OM; however, they are hindered by the lipidic IM (5). The double-membrane barrier equips gram-negative bacteria with a stronger tolerance to antimicrobials, and impedes the development of antibiotics against gram-negative bacteria (5). Our lead candidate, 10, is a relatively large molecule with molecular mass of 1184.57 Da that is larger than the cutoff for compounds to penetrate the OM (∼600 Da). Remarkably, 10 showed broad-panel activity against a series of gram-negative bacteria, including multidrug-resistant strains. We found that LPS and OmpA were the surface targets of 10 that mediated its membrane activity, and we also identified bacterial 70S ribosome as a potential intracellular target of 10. The double mechanisms of membrane perturbation and inhibition of protein synthesis might synergistically contribute to the effective killing of 10 to gram-negative bacteria. Genetic knockout of OmpA could significantly attenuate the OM permeabilization caused by 10, but it could not abolish 10’s bactericidal activity, suggesting that single-target mutation may not generate antimicrobial resistance to 10. Nevertheless, detailed structural basis of 10 in complex with its molecular targets remains unclear. Future refinement of the MoAs of 10 should facilitate the discovery, design, and evaluation of new classes of peptide antibiotics to confront the challenge of antimicrobial resistance. Finally, although we identified 70S ribosome as a potential target that warrants in-depth investigation in the future, the possibility of the existence of other molecular targets responsible for the antibacterial action of 10 cannot be formally excluded.

Data Availability

The raw sequencing data reported in this paper have been deposited in the National Center for Biotechnology Information Sequence Read Archive, https://www.ncbi.nlm.nih.gov/sra/PRJNA789103. All other data are available in the main text and supporting information.