Selective Bacterial Targeting and Infection‐Triggered Release of Antibiotic Colistin Conjugates

Abstract In order to render potent, but toxic antibiotics more selective, we have explored a novel conjugation strategy that includes drug accumulation followed by infection‐triggered release of the drug. Bacterial targeting was achieved using a modified fragment of the human antimicrobial peptide ubiquicidin, as demonstrated by fluorophore‐tagged variants. To limit the release of the effector colistin only to infection‐related situations, we introduced a linker that was cleaved by neutrophil elastase (NE), an enzyme secreted by neutrophil granulocytes at infection sites. The linker carried an optimized sequence of amino acids that was required to assure sufficient cleavage efficiency. The antibacterial activity of five regioisomeric conjugates prepared by total synthesis was masked, but was released upon exposure to recombinant NE when the linker was attached to amino acids at the 1‐ or the 3‐position of colistin. A proof‐of‐concept was achieved in co‐cultures of primary human neutrophils and Escherichia coli that induced the secretion of NE, the release of free colistin, and an antibacterial efficacy that was equal to that of free colistin.


Introduction
Infections by multidrug-resistant bacteria have been recognized as aglobal threat to human health. [1] In particular, the WHO has emphasized that the arsenal of antibiotics against resistant Gram-negative pathogens has become alarmingly thin. [2] This led to the decision to reactivate the natural lipopeptide colistin 1 ( Figure 1a,the shown colistin B, also known as polymyxin E2, was used throughout in this study) [3,4] as a" last resort" treatment of infections,a fter its systemic use has been banned for decades due to its toxicity. Thestrong nephrotoxicity of colistin, leading to acute kidney injury (AKI) in over 30 %o fc ases, [5] as well as adverse neurological effects make use of this compound ad elicate balance between benefit and harm for the patients. [6] However,t he potent, broad spectrum activity and the (yet) low level of resistance [7] make colistin an attractive antibiotic scaffold. In consequence,t he synthesis of colistin derivatives with improved drug properties or of colistin-related scaffolds [8] has recently become at opic of intense research activities,with most efforts focused on the empirical replacement of the basic amino acids by other chemical moieties. [9] Binary conjugates of colistin with dextrin, [10] alginate-oligosaccharide [11] and poly(ethylene glycol) (PEG) derivatives [12] have also been reported, that lead to aliberation of colistin by the action of a-amylase, [10] esterase or alginate lyase activity, [11] or hydrolysis. [12] At argeting for these macromolecules to inflammation sites can be achieved due to an enhanced leakiness of vessels in inflamed tissue. [13] Herein, we would like to present an ovel concept to improve the therapeutic window of highly potent, but also toxic antibiotics like colistin by selectively enhancing their concentration at the pathogen, while minimizing the (toxic) exposure to host cells.F or this purpose,adrug conjugation approach [14] was applied to introduce two additional functionalities to 1: -Atargeting moiety should direct the antibiotic effector specifically to bacterial cells -T he release of the effector should be triggered selectively at the site of infection We decided to reduce the concept to practice with components accessible by peptide chemistry (Figure 1b): Due to the advantages outlined above,c olistin was the effector antibiotic of choice.F or bacterial targeting, the 13mer fragment Ubi 29-41 of the human antimicrobial peptide ubiquicidin [15] was selected, because its ability to bind to anionic phospholipid head groups of Gram-negative bacteria with high affinity and specificity,while showing little binding to mammalian cells,h as been validated with al arge body of animal and even human imaging data. [16] To protect the targeting peptide from proteolytic degradation under in vivo conditions,t he original Ubi 29-41 was replaced with its all-denantiomer. [17] Ac orresponding conjugate with aU bi 29-41 peptide made of l-amino acids was also synthesized for comparison. In order to limit the release of the effector only to infection-related situations,w ew anted to take advantage of the fact that neutrophil granulocytesa re rapidly recruited at infection sites, [18] concomitant with al ocal release of neutrophil elastase (NE) by the immune cells to destroy bacteria ( Figure 1c). Thus,aproteolytic cleavage site recognized by NE was incorporated in the linker connecting Ubi 29-41 and colistin. In order to assure ahigh cleavage fidelity, an optimized substrate peptide composed of the four nonproteinogenic amino acids Bpa-Met(O 2 )-Oic-Abu (Bpa = 4benzoyl phenylalanine,M et(O 2 ) = l-methionine sulfone, Oic = octahydro-1H indole-2-carboxylic acid, Abu = 2-aminobutyrate), that is cleaved more than 7000 times faster than the commonly used Ala-Ala-Pro-Val sequence, [19] was incorporated into the conjugate.Inaddition, we expected the nonproteinogenic amino acids to be resistant to proteolytic degradation by other plasma enzymes.T he fact that unmodified colistin is released upon proteolysis by NE is seen as another advantage,b ecause the antibiotic properties of colistin (e.g. its high potency) are not impaired by residual linker parts.W hile prodrug-protected cationic antimicrobial peptides have been reported before, [20] the concept of trifunctional antibiotic conjugates is,t ot he best of our knowledge, new-and quite complex, as it requires an interplay of at least four components,t hat is,t he trifunctional conjugate,n eutro- phil elastase,n eutrophil granulocytes and bacteria (Figure 1c). This paper describes the realization and validation of the concept.

Results and Discussion
Thes ynthesis of the conjugates was achieved by two different strategies.T os atisfy the material demand for the biological characterization experiments,asemisynthetic approach was chosen. Because all attempts to connect Ubi 29-41 as aC -terminal thioester to ac ysteine-modified elastase substrate by native chemical ligation failed (data not shown), ac opper-catalyzed alkyne-azide cycloaddition reaction was applied for coupling.T he Eastern part of the molecule was assembled by N-terminally extending the elastase substrate Bpa-Met(O 2 )-Oic-Abu, obtained by peptide synthesis,with 4pentinoic acid to give 2.The alkyne-modified substrate 2 was subsequently reacted with the amino group(s) of 1 that was isolated from ac ommercial fermentation product ( Figure 2). Themonosubstituted isomers 3 were separated from 1 and the higher substitution variants,b ut no attempt was made to further resolve the positional isomers that eluted with similar retention times from the HPLC column (see the Supporting Information). In parallel, as hort spacer consisting of balanine and aC -terminal azido-lysine was added to the Cterminus of the all-d-enantiomer of Ubi 29-41 4,o rt oi ts all-lenantiomer 5,r espectively,t ob uild the Western part of the conjugate.F inally,the fully functionalized conjugates 6 and 7 were prepared by coupling 3 to 4 and 5 via acopper-catalyzed click reaction, respectively,a nd characterized by HPLC (purities > 95 %) and high resolution mass spectrometry.
In the second strategy,also the colistin part was obtained by organic synthesis in order to fully control the position of the linker attachment to any of the five primary amino groups. We devised two different synthetic schemes to derivatize either the two extracyclicdiaminobutyric acid (Dab) residues at positions 1and 3, or the three intracyclic ones at positions 5, 8, and 9(for atom numbering,see Figure 1a). Thecompounds were prepared using solid-phase peptide synthesis (SPPS) on 2-chlorotrityl chloride (CTC) resins and afluorenylmethoxycarbonyl (Fmoc) protocol. Thes ide chain amino groups of Dab were protected as tert-butyloxycarbonyl (Boc), the side chain hydroxy group of Thra stert-butyl (tBu), while the g- amino groups of the Dab involved in cyclization (position 4) or branching carried a1 -(4,4-dimethyl-2,6-dioxocyclohex-1ylidene)ethyl (Dde) moiety that is orthogonal to Boc and tBu. Fort he two conjugates modified at extracyclic residues, ac onvergent approach was selected in which the extracyclic chain was synthesized in parallel to the cyclic core,f ollowed by fragment condensation. Thesynthesis of 22,the conjugate modified at position 1, started with the loading of g-Bocprotected Dab onto chlorotrityl resin ( Figure 3). After assembly of the extracyclic colistin sequence,the gDde group at Dab-1 was cleaved, the elastase substrate peptide with Nterminal 4-pentinoyl group was coupled, and the protected peptide 16 with af ree C-terminus was cleaved from the support by mild acidic treatment with hexafluoro-2-propanol (HFIP). Forthe cyclic portion of colistin, alinear peptide was obtained by SPPS from Thr( position 10) to the gDdeprotected Dab residue at position 4. After Dde removal and cleavage from the resin, the peptide was cyclized and Fmoc-deprotected, thereby offering as ingle amino group at the aposition of residue 4. Thecyclopeptide 19 was then coupled to the free carboxy group of 16.After aglobal deprotection, the triple bond of the 4-pentinoyl moiety was "clicked" to the azido group of the ubiquicidin derivative 4,affording the final compound 22.Conjugate 28,modified at position 3of1,w as obtained in asimilar manner (Supplementary Figure S1). For the synthesis of three conjugates at intracyclic residues,Fmoc-Thr(tBu) was attached to 2-CTC resin, and linear peptide chains with the elastase sequences attached to positions 5, 8, or 9, respectively,were assembled by sequential deprotection/ couplings (Supplementary Figures S2, S3, and S4). The branching positions for the elastase substrate peptide were a-Dde protected, and after completion of the elastase substrate synthesis at the g-amino group of the respective Dab,the Dde group was removed, and the remaining colistin sequence was assembled. After cleavage from the resin, the peptides were cyclizedi ns olution and globally deprotected. Finally,the 4-pentinoyl moiety was coupled by cycloaddition to the ubiquicidin derivative 4,affording the final compounds 34, 40,a nd 46.F inal conjugates 22, 28, 34, 40,a nd 46 were characterized by 1D and 2D NMR spectroscopy,h igh resolution mass spectrometry,a nd HPLC.T he structure elucidation by NMR spectroscopy was based on ac omparative analysis with peptide 50 (Supplementary Figure S18 and S19) and af ull assignment of 1 (Supplementary Table S1, Figures S20-S26), which, to the best of our knowledge,h as not been reported so far.E ven though 1 HNMR and fragment data are available for aw ide range of natural and synthetic derivatives, [4,21] only the synthetic octapeptin representative FADDI-118 was completely assigned. [8] Corroboration of the linker attachment positions was achieved by detailed analysis of specific 1 Hand 13 Cchemical shifts and HMBC correlations to the corresponding Dab moiety (Supplementary Tables S2-S6, Figures S27-S46). Overall, the final products were obtained in amounts of 4.9-6.9 mg using the fully synthetic approach, which enabled the determination of structureactivity relationships in ab asic microbiological experiment (see below). However,conjugates required in larger amounts were prepared by semisynthesis.
In order to verify the claim that the Ubi 29-41 -carrying conjugate indeed accumulates at Gram-negative bacteria, we prepared an analogue of 6 that was further functionalized with af luorescent dye for bacterial imaging (Supplementary Figure S5). Fort his purpose, 4 was C-terminally elongated with cysteine to 8,and subsequently reacted with fluorescein maleimide to give 10.T he intermediate 10 was then coupled to 3 to yield the tetrafunctional conjugate 12.A na nalogous procedure was conducted with the all-l-peptide 9,y ielding analogue 13.
Thebinding properties of 12, 13,and controls were probed with the three medically relevant Gram-negative pathogens Escherichia coli (E. coli), Pseudomonas aeruginosa (P. aeruginosa), and Acinetobacter baumannii (A. baumannii)b y quantifying the fluorescence of bacteria following a1 h compound incubation and subsequent washing steps.W hile 5(6)-carboxyfluorescein did hardly show binding above background, both 12 and 13 led to fluorescence that was 17-100 fold above background (Figure 4). For E. coli and A. baumannii the bacterial binding of the all-l-Ubi 29-41 conjugate 13 was higher than that of 12,w hereas P. aeruginosa preferentially bound the all-d-Ubi 29-41 variant 12.However,the signal strength for both conjugates suggests that chiral recognition plays alimited role for the binding of the peptide to the lipid membrane of bacteria. Thefluorescence signals obtained for the all-l-Ubi 29-41 -fluorescein and all-d-Ubi 29-41 -fluorescein conjugates 10 and 11 were lower compared to 12 and 13, which demonstrates that also colistin (and not only ubiquicidin) contributed to binding.I nf luorescence microscopy images of smears of whole blood after addition of E. coli and 12 or 13,the fluorescence signal was co-localized with bacteria (with enhanced concentration at the bacterial septum), but not with human erythrocytes, thrombocytes,o rw hite blood cells (Figure 4b,c). This indicates that the conjugates may indeed target bacteria selectively.
As mentioned above,t he all-d-analogue of Ubi 29-41 was incorporated in 6 to achieve ab etter proteolytic stability of the peptide.T ov erify this assumption, the stability of 6 was tested in phosphate-buffered saline (PBS) as well as in human plasma at 37 8 8Ci nc omparison to 7,t he epimer of 6 that carried all-l amino acids in the Ubi 29-41 part (Supplementary Figure S6). Thechemical stability in PBS was comparable for 6 and 7 (40 %and 53 %after 24 h, respectively). In plasma, 6 showed astability of 44 %after 24 h, avalue that is similar to the one obtained in PBS.I ns tark contrast, we observed that the plasma stability of 7 was only 5% after 24 h. We suggest that this difference is due to plasma proteases that specifically recognize l-amino acids,b ut not d-amino acids.F or this reason, all subsequent experiments were continued with analogue 6.
As af irst proof of concept for antibiotic efficacy, the activity of 6 was tested in two strains of E. coli (K12 and DSM 1116) in the presence or absence of neutrophil elastase.A s ac ontrol, free 1 was applied. Conjugate 6 showed reduced antimicrobial activity with minimal inhibitory concentration (MIC) values of 4and 16 mm for the two strains compared to 1,w hich had MICs of 0.063 mm against both strains (Table 1 and Supplementary Figure S7). However, the activity of colistin was almost completely restored upon addition of NE at 10 mU mL À1 to 6.N eutrophil elastase alone did not display any antimicrobial activity.T he microbiological data were supported by LC-MS tracking of the colistin release from 6 upon treatment with NE in RPMI 1640 medium. After  Figure S8) to generate 1.T he experiments demonstrate that the antibacterial activity of 6 is indeed shielded, but can be liberated by pure,recombinant NE. The role of the optimized NE cleavage peptide was investigated by ac omparison with 49,aconjugate that comprised the commonly used Ala-Ala-Pro-Val peptide substrate for NE (Supplementary Figure S9). Thea ntimicrobial activity of 49 was low (> 32 mm)both in the presence and in the absence of neutrophil elastase (Supplementary Figure S10). Also the intermediate 48,t hat was devoid of d-Ubi 29-41 ,s howed low (MIC = 16 mm)a ntimicrobial activity.T hese findings demonstrate that utilizing the optimized NE substrate peptide was mandatory for the concept.
In order to probe whether the attachment position of the elastase-cleavable linker to colistin impacts the cleavage efficacyo ft he conjugates,t he five structurally defined regioisomers 22, 28, 34, 40,and 46 were tested against E. coli K12 in comparison to free 1.R emarkably,t he intracyclic derivatives 34, 40,and 46 showed no activity (MIC > 32 mm)in the absence of NE (Table 2a nd Supplementary Figure S11). Following activation by elastase,weobserved that the microbiological potencyo ft he isomers differed markedly:T he highest activity was observed for 22 and 28,h aving the elastase substrate at extracyclic positions 1and 3, respectively. 34,b earing the elastase substrate at position 5, showed enhancement to am oderate activity (MIC = 4 mm). Isomers 40 and 46,that carried the NE substrate at positions 8and 9, respectively,s howed even lower activities.B ecause the microbiological activity is mostly due to free 1,t his rank order reflects differences in NE cleavage efficiency. We conclude that the most favorable Dab residue for linker attachment are the ones at positions 1and 3(as in 22 and 28, respectively).
Next, we addressed the question whether human neutrophil granulocytes (polymorphonuclear neutrophils,P MN) release sufficient amounts of neutrophil elastase upon stimulation for the cleavage of the linker in 6.A ccording to previous investigations,n eutrophil granulocytes contain an average amount of 1.59 pg neutrophil elastase per cell, [22] which corresponds to 1.59 mgmL À1 for 1 10 6 cells mL À1 ,o r to 100 mU mL À1 . [23] Thus,i fa bout 10 %o ft he NE was released or functionally active on the surface of the cells upon stimulation of the granulocytes,s ufficient activity should be available for am easureable cleavage of the conjugates.T o probe this experimentally,P MN were isolated from fresh whole blood, taking into account that the cells are very sensitive and readily activated by different stimuli, including their physical handling. We applied am ild, so-called untouched isolation procedure that removes contaminating cells out of suspension by iron-conjugated antibodies and magnets. [24] Three sources for the isolation of the PMN were compared:i )buffy coat, ii)the filter content from ap lateletpheresis instrument, and iii)fresh venous whole blood (Supplementary Figure S12). Fresh whole blood was found to yield the highest quality PMNs with ap urity of > 98 % according to FACS analysis after labeling with anti-CD15-PE and anti-CD16-APC antibodies;t herefore,t his source was used in further studies.T he addition of the chemotactic and neutrophil granulocytes activating peptide formyl-methionylleucyl-phenylalanine (fMLP,c oncentration 0.1 mm)l ed to a2 .5-fold increase of the activation state marker CD66b, which is in good agreement with previous data (Supplementary Figure S13). [24] Thea mount of elastase released by neutrophil granulocytesi nR PMI 1640 medium was quantified by an enzyme-linked immunosorbent assay (ELISA). A basal amount of 39 ng/1 10 6 cells mL À1 was detected, which increased to 84 ng mL À1 upon stimulation with 0.1 mm fMLP, corresponding to 2.5 and 5.5 mU mL À1 ,r espectively (Figure 5a). Since ac onsiderable part of the active NE is not secreted but remains associated with the cells, [25] these values are considered sufficient to exert cleavage activities in the range of 10 mU mL À1 ,a su sed in the antibiotic assays with recombinant NE (Tables 1a nd 2). In order to address the question whether NE is not only secreted but also active,we carried out afunctional assay which monitored the release of p-nitroaniline at 410 nm following the cleavage of the peptidic substrate MeO-Suc-AAPV-pNA by NE ( Supplementary  Figures S14 and S15). [26] We observed substrate cleavage from both the supernatant or ac ell suspension of PMN in RPMI 1640/HEPES medium, indicating elastase activity in the in vivo situation;i nl ine with the literature,t he cleavage was stronger (2.5-fold under unstimulated and 1.8-fold under stimulated conditions) in the cell suspension as compared to the supernatant.
Another potential confounder was that colistin itself might influence the activity of NE. While Jones et al. [27] reported that colistin had an enhancing effect on the activity of NE in sputum samples,a sw ell as on purified NE in vitro, the study of Hector et al. [28] contradicted these findings and postulated that colistin has an inhibiting effect on the activity of NE in vitro.S ince this aspect was important for our concept, we probed the influence of free colistin 1 as well as   (Figure 5b). The inhibitory effect of 6 might be due to its competition with the assay substrate peptide MeO-SucAAV-pNAf or the occupation of the NE active site.O n the other hand, the increase of activity by colistin itself was considered to be favorable for our approach, since it may lead to an auto-enhancement of the liberation of 1 from the conjugate.N ext, we checked whether the exposure of PMN to E. coli bacteria could trigger the release of colistin from the conjugate 6 ( Figure 5c). In RPMI 1640, cleavage efficiencies of 74 %, 80 %, and 50 % were found at concentrations of 6 of 0.1, 1.0, and 10 mm, respectively.T he addition of fMLP had no significant effect. Thec leavage efficiencyo ft he two lower concentrations correlates well with the one obtained with pure protein (Supplementary Figure S8);t he reduced rate for 10 mm substrate may reflect al imiting concentration of the enzyme.I np lasma, free 1 could not be detected under the experimental conditions (for ad iscussion see below). Overall, we conclude from the ELISA-based quantification of NE and from the activity-based functional assays that PMN can indeed provide as ufficient amount of enzyme to cleave NE-sensitive drug conjugates and to release colistin. Finally,t he crucial proofof-concept experiment addressed the question whether the conjugate 6 can reduce the number of viable bacteria in ac o-culture of PMN with E. coli. Fort his purpose,b ac- Figure 5. Cellular proof of concept for triggered antibiotic release from 6.a)Release of neutrophil elastase (NE) from neutrophil granulocytes. PMN were isolated and suspended in RPMI 1640. Foractivation, 0.1 mm fMLP was added.T he samples were incubated for 1hat 37 8 8C, centrifuged to remove cells, and the NE concentration was determined in the supernatant by ELISA. Values normalizedt o 1 10 6 cells mL À1 .Data are presented as the mean of 4separate experiments. *p < 0.03 compared to unactivated. b) Activity of NE after co-cultivation of PMN with E. coli K12 in presence of 1 or 6. 5 10 5 cells mL À1 of PMN were co-cultivated with 8 10 6 cells mL À1 of E. coli K12 (ratio of bacteria :PMN 16:1) for 1hat 37 8 8CinR PMI 1640/HEPES-NaCl 1:1. Data are presented as the mean of two separate experiments. c) Release of colistin from 6.5 10 5 cells mL À1 of PMN suspended in RPMI 1640 were cocultivated with 8 10 6 cells mL À1 of E. coli K12 for 1hat 37 8 8C, and 6 was added at three concentrations. After 1hcolistin concentrations were determined by HPLC-MS. Data are presented as the mean of two separate experiments. d,e) Antimicrobial activity of 6 against E. coli K12 during co-cultivation of PMN with the bacteria in RPMI 1640 (d) and in human blood plasma (e). 5 10 5 cells mL À1 of PMN suspended in RPMI 1640 or human blood plasma were co-cultivated with 8 10 6 cells mL À1 of E. coli K12 for 1hat 37 8 8C with shaking (ratio bacteria to PMN 16:1). Colistin 1 and the conjugate 6 were added at three concentrations (0.1 mm,1mm,a nd 10 mm). Data are presented as the mean of two separate experiments. p < 0.05 for all data points compared to control (unpaired Mann-Whitney test). Whiskers present standard deviation. teria and PMN in aratio of 16:1 were co-cultivated for 1hin RPMI 1640 and in blood plasma. We were pleased to observe that 0.1 mm and 1.0 mm of 6 reduced bacterial growth by 53 % and 83 %, respectively,a nd abolished growth completely at 10 mm in RPMI 1640 (Figure 5d). This efficacy was equal to that of the positive control 1.A gain, the addition of 0.1 mm fMLP had no effect. To mimic ar elevant in vivo matrix, equivalent experiments were performed in blood plasma. Again, 6 reduced growth by 29 %and 95 %at0.1 and 1.0 mm, respectively,w hich was comparable to 74 %a nd 97 % reductions by colistin at equal concentrations (Figure 5e). Both compounds abolished growth completely at 10 mm.
Despite the fact that we did not detect free 1 in plasma upon addition of 6 to aco-culture of PMN and E. coli K12, the conjugate clearly reduced the number of viable bacteria according to the CFU test. We hypothesize that the released amount of 1 was sufficient to exhibit antibacterial activity,but fell below the limit of detection for colistin in plasma (LOD = 50 nm,Supplementary Figure S16).
Ther eleased colistin will be eliminated from the body mainly by renal clearance,a nd the same renal toxicity mechanisms as for unconjugated colistin are operative. [6a] But due to the targeting of the drug to the bacterial membrane,w ee xpect that the systemic dose required to achieve an effective dose at the site of infection can be lowered;i na ddition, the fraction of conjugate not reaching the infection site is eliminated in ap rotected prodrug form; both factors should contribute to lower renal exposure to free colistin and lower renal toxicity.However, the proof of these conceptual considerations needs to be given in further in vivo investigations.

Conclusion
We have validated anew concept for antibiotic conjugates that are activated by the pathogen at the site of infection. Mechanistic experiments demonstrate that 6 can initiate and accomplish the complex sequence of events outlined in Figure 1c:d irection of the antibiotic conjugate to bacterial pathogens,s ecretion of ap roteolytic enzyme from immune cells upon exposure to the pathogen, enzymatic release of antibiotic from the conjugate,a nd finally bacterial killing. Compared to previous work with bifunctional constructs [10][11][12] our trifunctional approach principally enables amore directed targeting to the bacterial surface and more focused release of the drug at the infection site;i na ddition, the compounds represent defined, single molecular entities.T he large molecular size would restrict an application of the conjugates to parenteral modes of administration. However, such antibiotics would be used in hospital settings for the treatment of life-threatening infections,w here the highly polar peptide conjugates are well-suited for intravenous administrations that constitute the preferred clinical standard.
In future studies,w ea im at expanding the concept to additional Gram-negative,c linically important pathogens, and to probe its in vivo efficacy.