A synthetic lipopeptide targeting top-priority multidrug-resistant Gram-negative pathogens

The emergence of multidrug-resistant (MDR) Gram-negative pathogens is an urgent global medical challenge. The old polymyxin lipopeptide antibiotics (polymyxin B and colistin) are often the only therapeutic option due to resistance to all other classes of antibiotics and the lean antibiotic drug development pipeline. However, polymyxin B and colistin suffer from major issues in safety (dose-limiting nephrotoxicity, acute toxicity), pharmacokinetics (poor exposure in the lungs) and efficacy (negligible activity against pulmonary infections) that have severely limited their clinical utility. Here we employ chemical biology to systematically optimize multiple non-conserved positions in the polymyxin scaffold, and successfully disconnect the therapeutic efficacy from the toxicity to develop a new synthetic lipopeptide, structurally and pharmacologically distinct from polymyxin B and colistin. This resulted in the clinical candidate F365 (QPX9003) with superior safety and efficacy against lung infections caused by top-priority MDR pathogens Pseudomonas aeruginosa, Acinetobacter baumannii and Klebsiella pneumoniae.

A ntimicrobial resistance is a major global health crisis requiring urgent attention 1,2 . Notably, hospital-acquired pneumonia and ventilator-associated pneumonia caused by multidrug-resistant (MDR) Pseudomonas aeruginosa, Acinetobacter baumannii, and Enterobacterales (e.g. Klebsiella pneumoniae) are major causes of morbidity and mortality [3][4][5] . Carbapenem-resistant isolates of these pathogens have been identified by the World Health Organization (WHO) as the top three priority pathogens to be targeted for antibiotic drug development 2 . The current dry drug development pipeline for new antibiotics has forced clinicians to resort to using the old polymyxin antibiotics (polymyxin B and colistin) as a last-line therapy against these difficult-to-treat Gram-negative pathogens [6][7][8] . First discovered in the 1940s, the polymyxins are a family of naturally occurring polybasic cyclic lipopeptides produced as non-ribosomal secondary metabolites by the soil bacterium Paenibacillus polymyxa ( Fig. 1a and Supplementary Table 1) 9 . Polymyxin B and colistin (as its pro-drug colistin methansulfonate) were first introduced into clinical practice in the late 1950s 8,9 . They have very similar chemical structures, differing only by a single amino acid at position 6 in the cyclic ring of the polymyxin scaffold (Fig. 1a). Remarkably, since their introduction over 60 years ago, no new polymyxins have been approved for clinical use 9 .
The effective use of polymyxin B and colistin in the clinic is severely hampered by several major pharmacological issues. Firstly, their therapeutic window is very narrow due to significant dose-limiting nephrotoxicity, which can occur in up to 60% of patients receiving intravenous (IV) polymyxins 10,11 . Secondly, acute toxicity severely limits the maximal dose of intravenous administration 12 . Thirdly, limited exposure in the lungs following intravenous administration 13,14 and binding to lung surfactant 15 render them often inefficacious for pulmonary infections. Fourthly, polymyxin B and colistin products are still manufactured by fermentation as multicomponent mixtures (Fig. 1a), presenting pharmaceutical quality issues 16 . Therefore, there is an urgent need to develop new polymyxins with a much-improved therapeutic window and efficacy, in particular against pneumonia [17][18][19][20][21] . Here we present for the first time, a rational drug design strategy focused on concomitantly optimizing polymyxin structure-activity relationship (SAR), structure-toxicity relationship (STR), and structure-pharmacokinetic relationship (SPR, focusing on lung exposure) through modification of multiple non-conserved positions within the polymyxin scaffold (Fig. 1b). This led to the development of a synthetic lipopeptide, which has a wider therapeutic window, reduced nephrotoxicity and acute toxicity, improved drug exposure, and efficacy against lung infections compared to polymyxin B and colistin.

Results
Lipopeptide design strategy. Previously, we established an SARbased mechanistic model of polymyxin antimicrobial activity built on the interaction of the polymyxin molecule with the lipid A component of LPS, the primary target of the polymyxins in the outer membrane of Gram-negative bacteria 19,21 . Here, binding of the polymyxin molecule to lipid A is characterized by a unique folded conformation wherein its N-terminal octanoyl fatty-acyl group and D-Phe 6 and Leu 7 make hydrophobic contacts with the hydrocarbon tails of lipid A; while the positively charged L-2,4diaminobutyric acid (Dab) side-chains make ionic contact to the phosphates of lipid A and the ketodeoxyoctonic acid (Kdo) group. This ultimately results in displacement of divalent cations and disruption of membrane lipid packing leading to permeabilization of the outer membrane (Fig. 1b). We used this SAR-based mechanistic model to identify amino acid positions within the polymyxin scaffold that would be amenable to structural modification to modulate toxicity and lung pharmacokinetics without reducing antimicrobial activity. To this end, we identified the N-terminal fatty-acyl group, and positions 3, 6, and 7 (Fig. 1b, c), as being potential positions that can be targeted.
The amino side chain groups of Dab 1 , Dab 5 , Dab 8 , and Dab 9 form critical ionic interactions with the 1-and 4′-phosphate groups of lipid A (Fig. 1b). These electrostatic interactions are the primary driver for initiating the interaction of the polymyxin molecule with lipid A 21 . Therefore, these four Dab residues, which are conserved residues in all native polymyxins (Supplementary Table 1), appear to be essential for antimicrobial activity and not tolerant to modification 19,21 . The hydrophobic N-terminal fatty-acyl group and residues at positions 6 and 7 form hydrophobic interactions with the fatty-acyl chains of lipid A. The Dab residue at position 3 appears to make hydrophilic contact with the Kdo sugars of lipid A (Fig. 1b) 21 . Importantly, these positions in the polymyxin scaffold are not structurally conserved in the native polymyxins (Supplementary Table 1), which potentially made them attractive sites to target for modification. At the N-terminus and positions 6 and 7, our strategy was to reduce the hydrophobicity at these sites (Fig. 1c). Conceivably, this would reduce membrane interactions with kidney tubular cells decreasing nephrotoxicity, lessen binding to lung surfactants and reduce acute toxicity. Previously, it has been shown that removing the hydrophobic N-terminal fatty-acyl group altogether from the polymyxin scaffold significantly reduces nephrotoxicity and acute toxicity 19 , but the loss of hydrophobicity significantly changes the interaction with LPS 21 and is detrimental to antimicrobial activity 19,21 . Furthermore, while reducing the hydrophobicity of the N-terminal fatty-acyl group alone can maintain potency it does not overcome the nephrotoxicity 19 . Therefore, a possible solution would be to target a more balanced reduction of the hydrophobicity at the N-terminus as well as the hydrophobic motif at positions 6 and 7 in the polymyxin scaffold (Fig. 1c). Modeling suggested that this design strategy would potentially generate lipopeptides that retain the ability to adopt a folded conformation similar to polymyxin B upon interacting with the LPS and retain antibacterial activity (Fig. 1d). Furthermore, native polymyxin M, which contains the same N-terminal fatty-acyl groups as polymyxin B and colistin but has a less hydrophobic motif (D-Leu 6 -Thr 7 ) at positions 6 and 7 (Supplementary Table 1), displayed similar LPS binding and antimicrobial potency against Gram-negative isolates to polymyxin B 21 . At position 3, we looked at removing the positive charge, changing the stereochemistry, and increasing or reducing the length of the residue side chain 22 (Fig. 1c, d).
Simultaneous optimization of the SAR, STR, and SPR. We concomitantly optimized for multiple parameters, including antimicrobial activity (against a panel of clinical MDR and ATCC reference strains, including carbapenem-resistant isolates of P. aeruginosa, A. baumannii, and K. pneumoniae), nephrotoxicity 16 , acute toxicity, and lung surfactant binding (Table 1, Fig. 2). An iterative design process was utilized, first looking at modifications to positions 6 and 7, followed by optimizing combinations of modifications to positions 3, 6, and 7, then the N-terminus, positions 3, 6, and 7 (Table 1). At positions 3, 6, and 7 we explored both proteogenic and non-proteogenic α-amino acid residues, while at the N-terminus we explored non-branched fatty acids and substituted aromatic-acyl groups. To generate our lipopeptide candidates a total synthesis approach was utilized using our efficient synthesis platform ( Supplementary Fig. 1), which afforded lipopeptides in sufficient yields to enable         Table 1. Examining modifications to the hydrophobic motif at positions 6 and 7 revealed that decreasing the hydrophobicity of the residue at position 7 generally decreased potency relative to polymyxin B and colistin, particularly against P. aeruginosa and A. baumannii isolates; with valine (F225) or L-2-aminobutyric acid (Abu) (F319) having the least impact on potency (Table 1). Reducing the hydrophobicity of position 6 alone (F228) or in combination with decreased hydrophobicity at position 7 (F229 and F230) could result in an even greater decrease in potency relative to polymyxin B and colistin (Table 1). However, incorporation of less hydrophobic residues at position 6 (e.g. F228) or 7 (e.g. F100, F124, F224, F225, F226, F227, F319) alone or in combination (e.g. F229, F230) showed a remarkable reduction of nephrotoxicity relative to polymyxin B and colistin in our mouse model, with no nephrotoxicity being observed in most cases within the tolerated doses (Table 1, Fig. 2a). Examination of modifications to position 3 in combination with D-Leu at position 6 and less hydrophobic modifications (Thr or Abu) to position 7 (Table 1), revealed that antimicrobial activity and nephrotoxicity could be modulated by structural changes at this position. Shortening the length of the side chain of Dab 3 through the incorporation of L-2,3-diaminopropionic acid (Dap), improved antimicrobial activity (e.g. F100 vs F251 and F319 vs F287), consistent with previously reported data 22 . Increasing the length of the side chain at position 3 by one carbon (F252) did not impact potency but appeared to increase nephrotoxicity slightly (e.g. F100 vs F252; Table 1). Changing the stereochemistry at position 3 from the Lto the D-configuration did not have a significant impact on potency overall against the panel of bacterial isolates, but substantially increased nephrotoxicity (e.g. F100 vs F085 and F251 vs F300), showing that the stereochemistry of position 3 is a key structural feature of the polymyxin scaffold that influences nephrotoxicity (Table 1). This explains the nephrotoxicity of polymyxin A, the first native polymyxin to be discovered, which has a D-Dab residue at position 3 and a D-Leu 6 -Thr 7 motif at positions 6 and 7 (Supplementary Table 1). It was never introduced into clinical practice because it was believed to be more nephrotoxic than polymyxin B and colistin 9 . Compared with polymyxin B and colistin, F287 was the best lipopeptide of this series having comparable potency against the panel of bacterial isolates and no nephrotoxicity at a dose up to 72 mg/kg/ d in our mouse model (Table 1, Fig. 2a). Finally, we optimized the N-terminus of F287. Replacement of the octanoyl fatty-acyl group of F287 with a less hydrophobic C6-alkyl chain (F342) slightly decreased potency against several A. baumannii isolates (Table 1); whereas replacement with an unsubstituted phenyl group (F379) significantly decreased potency (>8 fold) against most isolates relative to F287 (Table 1). Increasing the hydrophobicity of F379 with the incorporation of an additional phenyl ring at the para position (F477) did improve potency against all isolates compared to F379; however, F477 was not more active than F287 and also resulted in slightly increased nephrotoxicity (Table 1). Monochloro substitution on the phenyl group of F379 afforded lipopeptides with improved potency (e.g. F371 and F378) relative to F379 (Table 1). Dichloro-substitution on the phenyl ring could yield more potent lipopeptides (e.g. F365, F381) than the mono-chlorinated analogs (Table 1). F365 and F381 were also the most active of all the lipopeptides tested against P. aeruginosa FADDI-PA025 (Table 1). Notably, with the chloro-substituted analogs, the potency, nephrotoxicity, and acute toxicity were dependent on the positions of chloro-substitution (e.g. F365 vs F360 and F365 vs F381) (Table 1 and Fig. 2c). Substitution of a chlorine atom on the phenyl ring of F365 with fluorine at either the 2 or 4-position (e.g. F448 and F449) appeared to decrease potency relative to F365 against several isolates of P. aeruginosa and A. baumannii (Table 1). Overall, F365 was the best of this series of N-terminal analogs having similar potency to F287 (slightly better than polymyxin B and colistin) and no nephrotoxicity even at a dose up to 72 mg/kg/d in our mouse model (Fig. 2a). Subsequently, we evaluated the efficacy of the potential lead lipopeptides in a neutropenic mouse bloodstream infection model developed by our group 16 to allow for rapid screening of large numbers of lipopeptides (Table 1). Several of our lipopeptides (e.g. F365, F371, F381, and F477) showed a greater reduction in the bacterial burden in blood than polymyxin B and colistin (Table 1).
To predict the in vivo efficacy against lung infections, the in vitro antibacterial activity of our lipopeptides was further examined in the presence of Survanta ® , a natural bovine lung surfactant extract that has been used to examine the potential lung surfactant binding of lipopeptides in antibiotic development 23 . MICs of both polymyxin B and colistin increased 8-fold in the presence of the lung surfactant; whereas the MICs of our synthetic lipopeptides were not impacted (e.g. F365) or slightly impacted (e.g. F251) in the presence of the lung surfactant (Fig. 2b). It appears that position 7 is a driver of lung surfactant binding of polymyxin B and colistin, as substitution with a less hydrophobic residue at position 7 helped to restore activity in the presence of lung surfactant (e.g. polymyxin B/ colistin vs F319 in Fig. 2b). The N-terminus also had an impact on lung surfactant binding, which appeared to be dependent on the hydrophobicity of the N-terminal group, e.g. F477 (Nterminal biphenyl [C12], D-Leu 6 -Abu 7 ) vs F342 (N-terminal hexanoyl [C6], D-Leu 6 -Abu 7 ). Modification to position 3 did not appear to have any significant influence on binding to lung surfactant (e.g. F287 vs F319 and F183 vs F251, Fig. 2b). Together, the results here indicate that the combined overall hydrophobicity of the position 7 residue and N-terminal group is the primary driver of lung surfactant binding. In a mouse, acute toxicity model, both polymyxin B, and colistin had an average maximum tolerated dose (MTD) of 5.4 and 5.3 mg/kg respectively, while our lipopeptides achieved up to a 4-fold increase in the MTD relative to polymyxin B and colistin (Fig. 2c). Our results clearly show that the N-terminal fatty-acyl group is the primary structural component of the polymyxin scaffold that drives the acute toxicity (e.g. F287 vs F342 and F379, Fig. 2c), with position 3 and positions 6/7 having less impact (e.g. F287 vs polymyxin B, Fig. 2c). Using the ratio of MTD/MIC as an indicator of the relative safety ( Fig. 2d), it is evident that these lipopeptides have the potential for much better safety than polymyxin B and colistin, with F365 being a stand-out candidate.
Selection of a lead candidate with a significantly improved safety profile. The therapeutic potential of F287 and F365 was further evaluated to select a lead candidate to move forward. To this end, extended MIC screening was conducted against a large panel (n = 600) of MDR clinical isolates, including 407 carbapenem-resistant isolates ( Fig. 2e-g). Specifically, against P. aeruginosa (n = 213), F365 (MIC 50 /MIC 90 , 0.5/1 μg/mL) was 2-fold more potent than polymyxin B (MIC 50 /MIC 90 , 1/2 μg/mL) and slightly more potent than F287 (MIC 50 /MIC 90 , 1/1 μg/mL). Against A. baumannii (n = 210), F287 (MIC 50 /MIC 90 , 0.25/ 0.5 μg/mL) and F365 (MIC 50 /MIC 90 , 0.25/1 μg/mL) were up to 8-fold and 4-fold more potent than polymyxin B (MIC 50 /MIC 90 , 1/4 μg/mL), respectively. In addition, a 7-day in vitro passaging study showed that resistance to polymyxin B, colistin, F287, and F365 at 4× MIC emerged within 3 days; while at 16× MIC, regrowth was observed from day 3 and 4 for polymyxin B and colistin respectively; but not for F287 and F365 ( Supplementary  Fig. 2), indicating a lower potential of resistance for both lipopeptides than polymyxin B and colistin. No growth was observed over 7 days with polymyxin B, colistin, F287, or F365 at 32× MIC. Overall, F287 and F365 had very similar in vitro antimicrobial activity profiles. To gauge their potential therapeutic windows, F287 and F365 were administered intraperitoneally (IP) in mice at a total dose of 120 mg/kg (as 6 divided doses every 2 h). At 20 h after the last dose, F365 showed no damage to the kidneys, similar to the saline control, whereas mice treated with F287 at the same dose, showed minor signs of damage to the kidneys (Supplementary Table 2). The nephrotoxicity profile of F365 was further evaluated in mice at a higher total dose of 150 mg/kg IP with treated mice showing no significant damage to the kidneys, similar to the saline control ( Supplementary Fig. 3). Considering the in vitro potency and the higher dose achieved without observing nephrotoxicity, it was evident that F365 had a greater relative safety than F287 and therefore was selected as the lead candidate to progress forward.
Efficacy and pharmacokinetics of F365. The in vivo efficacy of F365 was firstly examined in a neutropenic mouse pneumonia model against polymyxin-susceptible MDR clinical isolates of P. aeruginosa, A. baumannii, and K. pneumoniae, which also included carbapenem-resistant isolates ( Fig. 3a-c). At 45 mg/kg/ d, the highest IP dose that could be safely administered for polymyxin B, both polymyxin B and F365 did not show any significant killing at 24 h (compared to the pre-treatment bacterial inoculum) against the polymyxin-susceptible MDR clinical isolates ( Fig. 3a-c). However, the improved acute tolerability of F365 allowed us to safely increase the IP dose to 90 mg/kg/d to achieve bacterial killing at 24 h against P. aeruginosa, A. baumannii, and K. pneumoniae (Fig. 3a-c). In a neutropenic mouse thigh infection model against a polymyxin-susceptible carbapenem-resistant A. baumannii clinical isolate FADDI-AB30, F365 also showed >2.0 log 10 reduction in CFU/thigh than polymyxin B ( Supplementary Fig. 4). Considering the superior safety profile of F365 and its lack of binding to lung surfactant, we further examined its efficacy against polymyxin-resistant isolates of P. aeruginosa, A. baumannii, and mcr-1 positive K. pneumoniae in the neutropenic mouse pneumonia model ( Fig. 3d-f; MICs are shown in Supplementary Table 3). Pleasingly, F365 (240 mg/kg/d subcutaneously) displayed significant bacterial killing (up to 4.65 log 10 reduction in CFU/lung) against the polymyxin-resistant isolates, highlighting the advantage of the increased therapeutic index and lower lung surfactant binding of F365 to achieve efficacious drug exposure at the infection site.
Protein binding of F365 in mouse (36%), rat (33%), and human (45%) plasma were similar and substantially lower than polymyxin B (95%, 80%, and 68%, respectively). The pharmacokinetics of F365 and polymyxin B in both mice and rats showed that F365 had a higher total body clearance and shorter half-life than polymyxin B (Fig. 4a, b and Supplementary Table 4). However, F365 had a higher free (unbound) drug exposure (fAUC) in plasma than polymyxin B in mice at 5 mg/kg and also in rats at 1 mg/kg. In rats we observed higher (>4-fold) urinary recovery for F365 than polymyxin B, suggesting decreased reabsorption of F365 by the kidneys (Supplementary Table 4). We further examined the pharmacokinetics of F365 and polymyxin B in mouse pulmonary epithelial lining fluid (ELF) (Supplementary Table 4 and Fig. 4c). After a single subcutaneous dose of 40 mg/kg, F365 and polymyxin B had very different concentration-time profiles, with F365 having a 3-fold higher C max, than polymyxin B. While the AUC ELF of F365 (86.2 mg·h/L) was comparable to that of polymyxin B (80.0 mg·h/L), based on the marked loss of antibacterial activity (87.5%) of polymyxin B due to binding to lung surfactant (Fig. 2b), the estimated fAUC ELF of F365 would be approximately 8-fold higher than that for polymyxin B (Supplementary Table 4). Collectively, our results demonstrated significantly different pharmacokinetics in the lungs, which explains the much-improved efficacy of F365 in the mouse pneumonia model, including against polymyxinresistant isolates with polymyxin B MICs of 4-8 μg/mL.

Mechanisms of improved activity and reduced nephrotoxicity.
Polymyxins exert their antibacterial effect by disrupting the Gram-negative outer membrane 21 . We demonstrated that F365 was able to permeabilize the outer membrane of P. aeruginosa and inhibit efflux to a greater extent than polymyxin B (Supplementary Fig. 5). We also conducted transcriptomics analyses to examine the mechanism underlying improved potency of F365. Treatments of A. baumannii AB5075 with F365, F287, F319, and colistin induced significant perturbations to bacterial gene expression in a concentration-and time-dependent manner ( Fig. 5a, b). A low concentration (2 μg/mL) of all lipopeptides resulted in enhanced cell envelope biogenesis, expression of membrane transporters, fatty acid degradation, and arginine metabolism, and reduced trehalose biosynthesis ( Fig. 5c). At the high concentration (8 μg/mL), all lipopeptides caused extensive transcriptomic perturbations, with approximately 33% and 50% of the genome differentially expressed at 1 and 4 h, respectively (Fig. 5b); particularly upregulated biosynthesis of major cell constituents, carbohydrate, and energy metabolism; and downregulated lipid A biosynthesis (lpxC, lpxH, lpxB). Notably, at 8 μg/ mL F365 caused the highest upregulation of fatty acid biosynthesis genes (e.g. fabG, fabZ, fabD) (Fig. 5c), indicating that F365 caused greater membrane destabilization consistent with its increased potency (Table 1 Polymyxin nephrotoxicity is associated with high levels of accumulation in renal proximal tubular cells in which transporters (megalin/PEPT/OCTN2) and endocytosis play key roles 24 . We have demonstrated that intracellular accumulation of polymyxin B and colistin results in oxidative stress and apoptosis 25 . Here we examined the impact of the synthetic lipopeptides F365, F287, F319, polymyxin B, and colistin on the transcriptomes of human HK-2 cells ( Fig. 5d-f). At 100 μM, polymyxin B, colistin, F287, and F319 significantly changed the expression of 1576, 1282, 1076, and 763 genes, respectively, while F365 only changed the expression of 70 genes (Fig. 6a). Gene expression following treatment with polymyxin B, colistin, F287, or F319 are close to each other, but are well segregated from F365 and the untreated control in the Principal Component Analysis plot (Fig. 5d). Consistently, F365 had a relatively low similarity (0.78-0.82, semantic analysis) of perturbed gene ontology compared with polymyxin B, colistin, F287, and F319. Polymyxin B and colistin caused the most significant transcriptomic perturbations (Fig. 5e). Notably, all lipopeptides except F365 altered the expression of potassium channels (Fig. 5f), including KCNJ16 (inwardly rectifying potassium channel Kir5.1). Furthermore, F365 had no effect on cell viability of HK-2 cells or KCNJ16 knockout, while KCNJ16 knockout protected HK-2 cells against 24 h treatment with all other lipopeptides (Fig. 6b). These results are consistent with our recent CRISPR screen findings where KCNJ16 was top-ranked in mediating polymyxin-induced toxicity in HK-2 cells 26 . All lipopeptides except F365 repressed the expression of key endocytic genes (Fig. 5f), but upregulated HSC70 chaperone genes associated with clathrin coat disassembly. Overall, F365 caused minimum perturbations to HK-2 transcriptome compared to polymyxin B, colistin, F287, and F319 (Fig. 6c), consistent with its significantly reduced nephrotoxicity.
Matrix enhanced-surface-assisted laser desorption/ionization mass spectrometry (ME-SALDI-MS) imaging revealed very different distribution, accumulation, and metabolism of F365 compared to polymyxin B and colistin in mouse kidney tissues (Fig. 7). MS signals corresponding to the [M + K] + and [M + Na] + molecular ions for each lipopeptide were observed in the kidney tissue samples. Polymyxin B 1 /B 2 and colistin A/B had a similar level of distribution in both cortical and medullary regions; whereas F365 was primarily in the cortex with very little medullary distribution (Fig. 7a, b). Concentrations of polymyxin B 1 /B 2 (38.2 and 70.9 µM, respectively) and colistin A/B (86.7 and 140.8 µM) in the cortex region were~2-7 fold higher than F365 (20.8 µM) (Fig. 7b). Ion signals were observed for putative metabolites (M1, M2, M3) of polymyxin B 1 , colistin, and F365 that lacked the N-terminal region (Fig. 7c, Supplementary Figs. 6,  7). Polymyxin B 2 and colistin B appeared to undergo less metabolism than polymyxin B 1 and colistin A, respectively ( Supplementary Fig. 7). This potentially explains the differences in the concentrations observed here for polymyxin B 1 /B 2 and colistin A/B in the kidney tissue. Metabolism of F365 in the kidneys differed from polymyxin B and colistin, with M3 being the major metabolite and no M1 metabolite detected ( Fig. 7c and Supplementary Fig. 7). Collectively, the pharmacokinetic and mechanistic results show that the very low potential for nephrotoxicity of F365 is a combination of less kidney reabsorption (and accumulation in tubular cells) and less intracellular toxicity.
F365 safety in a primate model. To further evaluate the safety of F365, a 14-day repeat-dose GLP toxicology study was undertaken in cynomolgus monkeys (Supplementary Table 5). F365 was well tolerated with no adverse clinical signs, including signs of acute toxicity, observed at any dose level. Kidney histopathological results revealed minimal tubular degeneration in 2 of 8 monkeys at the dose of 20 mg/kg/day; and minimal (4 of 8 monkeys), mild (1 of 8 monkeys), and moderate (1 of 8 monkeys) tubular degeneration at the highest examined dose of 50 mg/kg/day. Histopathological changes correlated with elevated serum creatinine and urea in 2 of 8 monkeys that received 50 mg/kg/day for 14 days. From these findings above, the No-Observed-Adverse-Effect-Level (NOAEL) for F365 was considered to be at least 20 mg/kg/day. It has been previously reported that polymyxin B administered to female cynomolgus monkeys at 3.8 mg/kg/12 h (7.6 mg/kg/day) for a shorter time period of 7 days, showed increased histological damage in all test animals along with mild increases in serum creatinine and blood urea nitrogen concentrations 27 . In patients, the maximum recommended dose for intravenous polymyxin B is 2.5 mg/kg/day, with nephrotoxicity and acute toxicity being observed at these doses 12,[28][29][30] . Based on allometric scaling and the No-Observed-Adverse-Effect-Level (NOAEL) 31 in monkeys of 20 mg/kg/day from our 14-day GLP toxicity study of F365, it is estimated that F365 can be safely administered to patients at a dose of at least 6.5 mg/kg/day without any acute toxicity or nephrotoxicity.

Discussion
Developing safer, potent polymyxin lipopeptides is very challenging due to the complex interrelationships between structure and antibacterial activity, nephrotoxicity, acute toxicity, lung drug exposure, and lung surfactant binding 19 . There have been efforts to address their pharmacological liabilities with the development of new polymyxin analogs derived from polymyxin B and colistin 8,[17][18][19][20][21][22][32][33][34] . While many novel polymyxin B and colistin derivatives have been generated using this approach, very few have entered clinical evaluation 19,35 . The polymyxin B analog SPR206 has successfully completed a Phase 1 study of both singleand multipleascending dose cohorts 35 . While no nephrotoxicity was observed at up to 100 mg q8h in the multiple-ascending dose cohort, neurotoxicity at single doses exceeding 200 mg was observed 35 . These drug development efforts have primarily focused on modification of the N-terminus of the polymyxin B and colistin scaffold with very limited exploration of the remaining lipopeptide scaffold, especially the cyclic ring 19,21 . Furthermore, previous polymyxin drug discovery programs have traditionally relied on cell-based assays using human kidney cells to screen for nephrotoxicity. However, there is a growing consensus that the results from these in vitro cytotoxicity assays do not translate to in vivo models and are not able to reliably differentiate the nephrotoxicity of different polymyxins 16,22,32 . To address this, our group was the first to develop a mouse model to screen for potential nephrotoxicity 16 , which was also utilized in this work. Finally, while previous polymyxin drug discovery efforts have focused on trying to ameliorate nephrotoxicity, this has been done without also targeting acute toxicity and the high lung surfactant binding. Our approach of systematically modifying multiple positions throughout the polymyxin scaffold, while optimizing for multiple parameters (antibacterial activity, nephrotoxicity, acute toxicity, lung drug exposure, and lung surfactant binding) concomitantly, represents a different strategy for the discovery of new polymyxin lipopeptides. This is the first time that positions 6 and 7 in the polymyxin scaffold have been specifically targeted to modulate their pharmacological properties. As a result, we identified several new structural hotspots driving antimicrobial activity (positions 6 and 7), nephrotoxicity (positions 3, 6, and 7), acute toxicity (N-terminal fatty-acyl group), and lung surfactant . Additionally, our results highlight how narrow the chemical space can be for exploring modifications to the polymyxin scaffold when optimizing for multiple parameters simultaneously. Optimizing the combination of modifications at all the positions targeted (N-terminus, positions 3, 6, and 7) was critical for providing lipopeptides with the best balance between potency and toxicity. Subtle structural changes to individual positions of the lipopeptide scaffold could cause pronounced changes to activity and toxicity. This was nicely exemplified by simply changing the positions of the chloro-substituents in the phenyl ring at the N-terminus, whereby going from the 2,4-dichloro substitution (F365) to the 2,6-dichloro substitution (F360), resulted in a significant decrease in potency, but did not impact nephrotoxicity (F360 vs F365 in Table 1). However, going from the 2,4-dichloro substitution (F365) to the 3,5-dichloro substitution (F381) maintained potency but increased nephrotoxicity and acute toxicity (F365 vs F381 Table 1 and Fig. 2c). Similarly, switching of the stereochemistry of a Dab or Dap residue at position 3 from the L-to D-stereoisomer led to increased nephrotoxicity, without impacting potency (e.g. F085 vs F100 and F251 vs F300 in Table 1). The findings here indicate that along with hydrophobicity, molecular conformation may also play an important role in polymyxin activity and toxicities.
Importantly, for the first time, we have been able to structurally disconnect the antimicrobial activity from nephrotoxicity, acute toxicity, and lung surfactant binding to develop a synthetic lipopeptide that is pharmacologically distinct from the existing polymyxin drugs. The improved activity, safety, and minimal lung-surfactant binding of F365 allowed us to achieve sufficient drug exposure following parenteral administration to target lung infections, even those caused by polymyxin-resistant Gramnegative isolates. Considering the increased potency and efficacy, decreased acute toxicity and nephrotoxicity, and the higher unbound lung exposure, F365 represents a promising synthetic lipopeptide drug candidate for targeting problematic carbapenem-resistant pulmonary infections in humans. To this end, F365, now known as QPX9003, has completed INDenabling studies and has commenced Phase 1 clinical development (NCT04808414).       MT1G  MT1H  MT1B  MT1M  MT1F  MT1X  MT2A  MT1E   ATF3  GADD45G  JUN  DUSP10  FOSB  GADD45B  JUNB  TRAF1  CTSW  DDIT3  ERN1  TNFRSF10A  GADD45A  BBC3  JUND  FOS  NR4A1  BIRC3  PTPN13  CTSK  CTSO  TNFSF10  STAT1   ATP6V0D2  ND4  ND2  ND1  COX1  COX3  ND6  CYTB  ND5  ND3  COX2  ATP6  ND4L  ATP8  LHPP   HSPA6 HSPA8

Methods
Materials. Polymyxin B sulfate and colistin sulfate were obtained from BetaPharm (Shanghai, China). The isolation of purified polymyxin B 1 and polymyxin B 2 , colistin A and B from commercial preparations of polymyxin B and colistin, respectively, was conducted as previously described 16 .
Modeling of the polymyxin B and F365-Lipid A complex. The methodology used for the construction of the models of polymyxin B and F365 in complex with Kdolipid has been previously described by our group 36 .
Lipopeptide synthesis. The general methodology used for the preparation of the synthetic lipopeptides has been previously reported by our group 37 . A full description of the materials used, synthesis protocol, and analytical data obtained are presented in the Supplementary Information.

Measurements of minimum inhibitory concentrations (MICs).
MICs were determined in cation-adjusted Mueller-Hinton broth (CaMHB) by the broth micro-dilution method against ATCC strains and clinical isolates 16 . MICs in the presence of Survanta ® were determined as described above against P. aeruginosa ATCC 27853. In the 100 μL of fresh CaMHB containing polymyxins, 10% Survanta ® (v/v, 2.5 mg/mL phospholipids) was added. MICs in the presence and absence of Survanta ® were compared to determine the effect of binding to lung surfactant on the antibacterial activity of polymyxins.
In vitro permeabilization and efflux inhibition assays. Bacterial strains were seeded in 50 mL of Luria-Bertani (LB) broth at OD 600 0.05 and grown at 37°C for 3-4 h until an OD 600 of 0.6-0.8 was achieved. Bacterial cells were washed twice with 25 mL of buffer (50 mM sodium phosphate, 0.5% glucose, pH 7, with or without 1 mM Mg 2+ ), resuspended in 10 mL of the buffer, and then adjusted to an OD 600 of 0.4. F365 and polymyxin B (n = 3) were titrated in 2-fold dilutions starting with 40 μg/mL in 50 μL of buffer; 50 μL of the substrate (4 × nitrocefin or Leu-Nap) at 256 μg/mL in buffer was added to each concertation followed dilution with 50 μL of  Histological examination revealed that all the kidneys exposed to polymyxin B 1 , polymyxin B 2 , colistin A, and colistin B had severe microscopic damage with SQS scores up to +5; whereas the kidneys exposed to F365 showed no significant microscopic damage and were not graded for damage. a Representative cryo-images of mouse kidney tissue sections after treatment with polymyxin B 1 , colistin A, F365, and saline control; and the corresponding ME-SALDI-MS image of the tissue sections after scanning for the parent ions [M + K] + for polymyxin B 1 , colistin A, F365, respectively. b Representative ME-SALDI-MS images highlighting the distribution and accumulation of polymyxin B 1 , polymyxin B 2 , colistin A, colistin B, and F365 in mouse kidney tissue. c Representative ME-SALDI-MS images showing the distribution and accumulation of F365 and its metabolites M2 and M3 in mouse kidney tissue after treatment. All ME-SALDI-MS images were normalized to total ion count based on the highest intensity peak across each tissue section and its corresponding concentration curve. Selected lipopeptide-related ions are displayed with a scale between 0 and 40% relative to the highest intensity peak in the summed spectrum of the whole tissue. A default color intensity gradient is used to visualize and differentiate low concentrations (purple to blue), mid-range concentrations (green to yellow), and high concentrations (orange to red) of lipopeptide disposition.
the buffer. Bacterial suspension (50 μL, OD 600 0.4) in buffer was added to the lipopeptide solutions and the rate of substrate hydrolysis was measured at 490 nm. Nephrotoxicity in mice. The experimental methodology has been previously reported by our group 16 . Briefly, individual lipopeptide solutions were prepared in sterile saline (6 mg/mL). Swiss mice (female, 7-week-old, 22-28 g) were subcutaneously administered at 12 mg/kg in, 6 doses every 2 h to reach a total dose of 72 mg/kg (n = 3). For high-dose nephrotoxicity experiments, mice were intraperitoneally administered at 20 or 25 mg/kg, in 6 doses every 2 h to reach a total dose of 120 mg/kg (n = 6) or 150 mg/kg (n = 3). At 20 h after the last dose, the right kidney from each mouse was collected and placed in 10% buffered (pH 7.4) formalin (Sigma-Aldrich, Castle Hill, NSW, Australia). The kidneys were subjected to histological examination at the Australian Phenomics Network-Histopathology and Organ Pathology (University of Melbourne, Parkville, VIC, Australia), who was blind to the treatment groups 16 .
Neutropenic mouse bloodstream infection model. The experimental methodology for the neutropenic mouse bloodstream infection model was described previously 16 . Briefly, 7-week-old female Swiss mice (22-28 g) were rendered neutropenic by intraperitoneal administration of cyclophosphamide, 4 days (150 mg/ kg) and 1 day (100 mg/kg) prior to inoculation. Bloodstream infection was established by intravenous administration of a 50 µL bolus of early log-phase bacterial suspension (4 × 10 8 CFU/mL). Lipopeptide solutions were prepared at 1 mg/mL (free base) in sterile saline. At 2 h after inoculation, mice (n = 3) were injected intravenously with lipopeptide solution at 4 mg/kg (free base) or saline control. At 0 and 4 h after administration, blood was collected, diluted serially in sterile saline, and plated on nutrient agar plates. Agar plates were incubated at 37°C overnight for bacterial colony counting and the bacterial load in the blood (log 10 CFU/mL) in each mouse was calculated. In vivo efficacy was calculated as the difference of the log 10 CFU/mL blood values between the treated mice and the saline-treated mice at 4 h post dosing (Δlog 10 CFU/mL blood = log 10 [treated] CFU/mL blood − log 10 [control] CFU/mL blood).
Neutropenic mouse pneumonia model. The experimental methodology for the neutropenic mouse lung infection model was previously described in detail 13 . Briefly, Swiss mice (female, 7-week-old, 22-28 g) were rendered neutropenic by intraperitoneal administration of cyclophosphamide, 4 days (150 mg/kg) and 1 day (100 mg/kg) prior to inoculation. Lung infection was established by the administration of 25 µL of early log-phase bacterial suspension (polymyxin-susceptible or resistant isolate, 4 × 10 7 CFU/mL), directly into the lungs via the trachea using a MicroSprayer™ device. Lipopeptide solutions were prepared in sterile saline. At 2 h after inoculation, mice (n = 3 or 4) were administered lipopeptide (45-240 mg/kg/d free base) or control (sterile saline) via an intraperitoneal or subcutaneous bolus injection in three divided doses every 8 h. At 0 or 24 h after the administration of lipopeptide solution or saline (control), animals were euthanized and lungs were collected and homogenized under sterile conditions. Homogenate was filtered and the filtrate was collected, serially diluted in sterile saline, and spread on nutrient agar plates. Agar plates were incubated at 37°C overnight for bacterial colony counting and the bacterial load of the lung (log 10 CFU/lung) in each mouse was calculated. In vivo efficacy was calculated as the difference of the log 10 CFU/lung values between the treated mice at 24 h and the pre-treatment (control) at 0 h (Δlog 10 CFU/ lung = log 10 [treated] CFU/lung − log 10 [pre-treatment] CFU/lung).
Neutropenic mouse thigh infection model. The experimental methodology for the neutropenic mouse thigh infection model was previously described in detail 38 . Briefly, Swiss mice (female, 7-week-old, 22-28 g) were rendered neutropenic by intraperitoneal administration of cyclophosphamide 4 days (150 mg/kg) and 1 day (100 mg/kg) prior to inoculation. Thigh infection was established by the administration of a 50 µL bolus of early log-phase bacterial suspension (4 × 10 8 CFU/mL) directly into each thigh of two mice in each group (n = 4 thighs); this method is well accepted in antibiotic pharmacokinetic/ pharmacodynamic studies to minimize animal usage. Lipopeptide solutions were prepared in sterile saline. At 2 h after inoculation, mice were administered a lipopeptide or control solution via an intraperitoneal bolus injection at 15 mg/kg, three doses every 8 h. At 0 and 24 h after the administration, animals were euthanized and each thigh was collected and homogenized under sterile conditions. Homogenate was filtered and the filtrate was collected, serially diluted in sterile saline, and spread on nutrient agar plates. Agar plates were incubated at 37°C overnight for bacterial colony counting and the bacterial load of both thighs (log 10 CFU/thigh) in each mouse was calculated. In vivo efficacy was calculated as the difference of the log 10  Distribution, accumulation, and metabolism in mouse kidneys. Individual solutions of polymyxin B, colistin, or F365 were subcutaneously (SC) administered to Swiss mice (female, 7-week-old, 22-28 g) at 10 μmol/kg, 6 doses on Day 1 and 3 doses on Day 2 at every 2 h (n = 3). At 2 h after the last dose on Day 2, mice were rapidly euthanized. The left kidney was collected, flash-frozen in liquid nitrogen, and stored at -80˚C for MS imaging analysis. The right kidney was fixed in 10% formalin for histological examination (Histopathology and Organ Pathology Unit at the University of Melbourne). The level of histological damage to the kidneys was graded as previously described 16 . The distribution of lipopeptides in kidney tissues from dosed animals was analyzed by ME-SALDI-MSI on a Shimadzu MALDI-7090, tandem time-of-flight (TOF)-MS instrument in reflectron positive mode at a spatial resolution of 50 µm. Validation of the detected lipopeptides was performed comparing MS/MS analysis results with lipopeptide standards of polymyxin B 1 , polymyxin B 2 , colistin A, colistin B, and F365. MS analysis was conducted using 5 mg/mL α-cyano-4-hydroxycinnamic acid (α-CHCA) dissolved in a mixture of 70% acetonitrile and 30% water with 0.1% trifluoroacetic acid (TFA), mixed 1:1 with purified lipopeptide standards. A total of 1 µL was spotted onto electrochemically etched, porous-silicon semiconductor substrates for desorption/ionization on silicon (DIOS) 39 . MS/MS fragmentation experiments were conducted using an ion selection window of ±1.5 Da and helium as the collision gas to generate collision-induced dissociation (CID) spectra. For lipopeptide distribution in mouse kidneys, serially sectioned 8-µm frozen slices of the kidney were loaded onto DIOS substrates and coated with a thin layer of α-CHCA using a matrix sprayer (TM-Sprayer, HTX) for MSI analysis. A thin α-CHCA layer was applied using two spray passes of 5 mg/mL, dissolved as above, at a flow rate of 0.12 mL/min and velocity of 1200 mm/min at 75°C. Mounted tissue was scanned for imaging alignment (Epson Perfection V370) and calibrated as previously described. Resulting polymyxin maps were visualized and normalized against the total ion current (TIC) of each section using IonView software (Shimadzu). Lipopeptide accumulation in mouse kidneys was determined by simultaneous ME-SALDI-MS analysis of kidney tissue from treated animals and serial dilutions of the drug deposited on control kidney tissue. Two kidney tissue sections (one for the control and one from the treated kidney sample) were placed on the same DIOS