Neutralizing the Impact of the Virulence Factor LecA from Pseudomonas aeruginosa on Human Cells with New Glycomimetic Inhibitors

Abstract Bacterial adhesion, biofilm formation and host cell invasion of the ESKAPE pathogen Pseudomonas aeruginosa require the tetravalent lectins LecA and LecB, which are therefore drug targets to fight these infections. Recently, we have reported highly potent divalent galactosides as specific LecA inhibitors. However, they suffered from very low solubility and an intrinsic chemical instability due to two acylhydrazone motifs, which precluded further biological evaluation. Here, we isosterically substituted the acylhydrazones and systematically varied linker identity and length between the two galactosides necessary for LecA binding. The optimized divalent LecA ligands showed improved stability and were up to 1000‐fold more soluble. Importantly, these properties now enabled their biological characterization. The lead compound L2 potently inhibited LecA binding to lung epithelial cells, restored wound closure in a scratch assay and reduced the invasiveness of P. aeruginosa into host cells.


Introduction
Carbohydrate-protein interactions are essential recognition codes in many biological processes, including bacterial and viral infections. Therefore, lectins of pathogenic origin involved in host-cell recognition, adhesion and/or biofilm formation are being recognized as new therapeutic targets. [1,2] Pseudomonas aeruginosa is a Gram-negative opportunistic bacterium that belongs to the group of highly drug resistant ESKAPE pathogens. [3] According to the World Health Organization (WHO), the pathogen ranks as critical priority 1 due to its high level of resistance to antibiotics and the lack of new treatment options to combat it. Consequently, several new therapeutic strategies aim at an anti-virulence therapy which circumvents the selection pressure of antibiotic treatment, thus preventing the emergence of new resistances. [4] Antimicrobial resistance of P. aeruginosa is further enhanced by the bacterium's ability to grow in biofilms-the causative mechanism for chronic infections. [5] To this end, P. aeruginosa adhesion, biofilm formation and virulence depend on the tetravalent lectins LecA and LecB, both of which are encoded in its conserved core genome and are functionally conserved across clinical isolates. [6][7][8][9] Furthermore, it was shown that LecA acts as a lipid zipper upon binding to its cellular receptor, [10,11] the glycosphingolipid globotriaosylceramide (Gb3), and induces the formation of a plasma membrane nanodomain, [12,13] cellular signaling [14] and ultimately leads to bacterial invasion. [10] Thus, the inhibition of LecA and LecB is a promising path forward to counteract P. aeruginosa's pathogenicity and break its antimicrobial resistance. [4] Encouragingly, it was also demonstrated that the inhalation of aerosols containing D-galactose and L-fucose, the monosaccharide ligands of LecA and LecB, respectively, reduced the bacterial burden and improved the clinical situation in cystic fibrosis patients. [15] Consequently, numerous glycomimetics have been developed based on D-galactose (K d = 88 μM) [16] for LecA and L-fucose/D-mannose (K d = 430 nM for methyl α-L-fucoside, K d = 71 μM methyl α-D-mannoside) [17] for LecB. [2,4,18] Recently, we reported small molecules based on C-glycosides targeting LecB which showed potent inhibition of LecB in the low micromolar range, prevented biofilm formation and exhibited beneficial pharmacokinetic properties. [19][20][21] For LecA, numerous monovalent glycomimetic inhibitors have been reported, [22][23][24] demonstrating the importance of an aromatic aglycon attached to the β-galactoside to reach lower micromolar affinity. To alter the galactoside moiety itself, we have recently introduced the first covalent lectin inhibitor targeting a surface exposed cysteine residue in LecA. [25] Furthermore, we also identified the first noncarbohydrate lectin inhibitor for a bacterial lectin. [26] Despite numerous attempts to increase potency, all monovalent LecA inhibitors reached at best only moderate potencies in the 5 to 50 μM range.
In contrast to LecB, the quaternary structure of LecA [27] displays two adjacent binding sites that are optimally oriented in space for the simultaneous binding of two galactose moieties of a single divalent ligand, which can result in an increased binding affinity. Notably, Pieters and co-workers developed low nanomolar LecA inhibitors with K d s down to 12 nM by connecting two galactosides through a linker containing several copies of rigid glucose-triazole linkers. [28,29] Despite the efficient click chemistry applied for the final assembly of the divalent ligands, in total 17 synthetic steps were required to prepare the azide and alkyne building blocks and final assembly. Replacing one of the glucose-bistriazolyl spacers with a cyclohexyl bisthiourea moiety in a 9-step synthesis yielded one compound with 30 nM affinity. [30] Similarly, more complex divalent LecA ligands with peptide-based linkers were reported by Novoa et al. (K d = 82 nM) and Huang et al. (K d = 71 nM). [31,32] Inspired by Pieters' approach, we recently reported a series of highly active divalent LecA inhibitors with acylhydrazone based linkers (K d = 11-81 nM) that simplified the synthesis to only four chemical steps. [33] While these molecules showed the highest potency among reported divalent LecA inhibitors, they suffer from their intrinsic hydrolytic lability of the acylhydrazone bond and a very low aqueous solubility, despite the presence of the two hydro-philic carbohydrates. Furthermore, acylhydrazones undergo hydrolysis already at weak acidic pH resulting in toxic aldehyde and hydrazide degradation products that may also be formed in vivo.
Here, we report the optimization of these divalent molecules by replacing the acylhydrazone motif and varying the linker to balance potency, solubility and stability. The lead compound was then successfully analyzed in in vitro assays and showed excellent properties to block LecAmediated host cell binding, restore wound healing and decrease bacterial invasion into human cells.

Results and Discussion
We substituted the acylhydrazone motif present in our previous compounds with a more stable amide bond and varied linker identity and length in order to increase solubility and stability (Figure 1a). Two galactoside building blocks bearing coumaric acid (1, Figure 1b) or hydroxyphenyl propionic acid (2) as aglycons were selected to investigate the effect of the rigid olefin in 1, comparable to the parent acylhydrazones, versus the flexible alkyl motif in 2. These galactosylated carboxylic acids were then coupled to various bisanilines to yield the desired divalent LecA inhibitors. Since an optimal length and geometry of the divalent ligand is important to simultaneously bind to the two neighboring LecA sites, linker length was varied by stepwise introduction of methylene units. The aromatic moieties of the linker and its length were also varied and bisaniline linkers B-F and their monovalent control A (Figure 1c) were used to mimic the previously used bisbenzaldehyde structures. [33] Furthermore, our rational solubility optimization included the introduction of hydrogen-bonding polar groups or ionizable moieties into the divalent ligands using bisaminopyridine linkers H-J or sulfonated linker L and their monovalent controls G and K, respectively.
Synthesis of the two galactoside building blocks 1 and 2 started with β-selective glycosylation of benzyl coumarate or methyl 3-(4-hydroxyphenyl)propanoate with β-D-galactose pentaacetate (3) under Lewis acid catalysis (Scheme 1). β-Glycosides 4 and 5 were obtained in 76 % and 86 % yields, respectively. Saponification of the esters with aqueous NaOH resulted in galactosides 1 and 2 in good yields. Synthesis of coumarate 1 was initially attempted using the methyl ester under identical glycosylation conditions as for compound 2, but this transformation was unsuccessful most probably due to its poor solubility in dichloromethane and only poor yields were achieved using more polar chloroform as a solvent instead. Changing from methyl to benzyl coumarate improved solubility in those solvents and the glycosylation gave compound 4 in good yield (76 %).
Final assembly of the divalent LecA inhibitors was achieved by coupling of the amino-substituted linkers A-L with carboxylate-containing galactosides 1 or 2 using HBTU or PyBOP as peptide coupling reagents (Scheme 3). High turnovers were observed for all coupling reactions, but purification difficulties caused varying yields: the lower solubility was responsible for isolated yields in the benzene series (A1-F1 and A2-F2), whereas side product formation was observed in the pyridine series (H1-I1 and G2-J2). The sulfonated ligands K2 and L2 were obtained as ammonium salts since their chromatography required buffered eluents to avoid acid-catalyzed self-hydrolysis of the glyosidic linkage upon solvent removal and concentration.
Next, we quantified the solubility of at least one representative of each new class and of one parent bisacylhydrazone in aqueous media (Table 1). All tested new derivatives showed improved solubility compared to the previous bisacylhydrazone 14. The very low kinetic solubility of bisacylhydrazone 14 (S = 1.6 � 0.4 μM) was increased twofold in its amide analogues D1 (S = 3.3 � 1.3 μM) and D2 (S = 3.1 � 2.1 μM). Substitution of the benzene ring with a pyridine moiety further increased solubility twofold in case of α,β-unsaturated amide analogue H1 (S = 8.1 � 2.2 μM) and almost fivefold in the saturated analogue H2 (S = 14.8 � 1.5 μM) compared to D1 and D2, respectively. Surprisingly, another fourfold increase in solubility was observed for I2 (S = 63.3 � 4.2 μM) compared to H2, despite only a small structural difference of one additional methylene in the linker. The longest pyridine ligand J2 showed again almost identical solubility to H2. Hence, the linker has a strong impact on solubility in the pyridine amide series, which could result from the conformational preferences of an ethylene glycol compared to alkyl chains with different numbers of methylene units that might affect ligand selfaggregation and thus their solubility. However, excellent aqueous solubility was finally achieved with sulfonated divalent ligand L2, which was fully dissolved from its solid form in an aqueous buffer (S > 1.5 mM).
Then, we analyzed the stability of the compounds in plasma. Acylhydrazone 14 was degraded faster (t 1/2 = 48.9 min) in mouse plasma whereas a higher stability was observed in human plasma (t 1/2 = 123.5 min). The tested representatives of the new ligands exhibited high stability in both mouse and human plasma (t 1/2 � 180 min), with exception of the coumarates: pyridine H2 was only moderately stable in mouse plasma (t 1/2 = 81.1 min) and benzene D2 showed a slow degradation in human plasma (t 1/2 =  132.5 min). The observed increase in plasma stability of the tested amide derivatives compared to bisacylhydrazone 14 supports our design hypothesis for the isosteric replacement of the bisacylhydrazone linking motif. In addition, all tested compounds showed very high mouse and human plasma protein binding (PPB) with lowest PPB of 97 % for saturated pyridyl amide H2. In contrast, sulfonated L2 showed a remarkably low PPB in both species (mouse 14 %, human 38 %).
Replacement of the benzene ring with a pyridine in G2 (IC 50 = 14.3 � 7.2 μM) or addition of the sulfonate in K2 (IC 50 = 14.4 � 3.6 μM) had only minor effects on LecA inhibition. In contrast to the monovalent controls and similar to our previous observations for the bisacylhydrazones, titrations with all divalent LecA ligands exhibited very steep Hill slopes with fitted IC 50 values between 3.2-8.0 μM. These data are indicative for the high potency of divalent compounds and suggest that the lower assay limit was reached. [33] To overcome the assay limit for these highly potent inhibitors, direct binding to immobilized LecA was quantified using surface plasmon resonance (SPR, Figure 2, Figures S2-S4). In agreement with the competitive binding assay, the binding affinity determined by SPR for the monovalent acrylamide ligand A1 (K d = 5.21 � 0.60 μM) and    propanamide ligand A2 (K d = 5.38 � 0.09 μM) were comparable to their IC 50 values. Interestingly, a striking difference was observed among divalent inhibitors. Within the benzene series, the acrylamides B1-F1 showed K d values in the nanomolar to micromolar range while their propanamide analogues B2-F2 were two-to threefold more active when attached to the shorter linkers B-D and 100-to 200-fold more potent for the longest linkers E and F. With respect to linker length, in the acrylamide series linker C containing one methylene unit exhibited strongest binding affinity (C1 K d = 37.7 � 11 nM). Increasing to four methylene units in F1 led to a complete loss of divalent binding affinity boost (K d = 2.25 � 0.3 μM). In contrast, all divalent ligands carrying the propanamide motif (B2-F2) showed very high binding affinity to LecA between 15 and 23 nM. K d values were oscillating with the number of methylene units present in the linker, presumably linked to the zig-zag geometry of the hydrocarbon chain. These observations could be explained by an increased rigidity of the acrylamides, that does not allow the ligand to simultaneously bind to two adjacent binding sites when the linker is too long.
Monovalent ligands from the pyridine series (G2) and sulfonated K2 showed slightly enhanced binding affinity to LecA (G2 K d = 2.97 � 0.08 μM, K2 K d = 3.79 � 0.10 μM) when compared to A2. Substitution of the benzene ring with a pyridine was also favored for divalent ligands: three-and four-fold increase in binding affinity was observed for acrylamide-based ligand H1 (K d = 13.7 � 1.6 nM) and I1 (K d = 40.2 nM) when compared to their benzene analogues D1 and E1, respectively. Among the propanamide derivatives, the increase in binding affinity was less pronounced: H2 (K d = 9.9 � 0.1 nM) and I2 (K d = 30.8 � 1.5 nM) were twice more active, while the longest pyridine ligand J2 (K d = 25.1 � 1.9 nM) was equipotent to its benzene analog F2. The observed affinity increase could result from additional interactions of the pyridine rings with the protein surface. Furthermore, divalent sulfonate L2 (K d = 32.8 � 1.5 nM) also reached low nanomolar affinity for LecA, indicating that an ether function in the linkers is not essential and that the negatively charged sulfonate solubility tags were tolerated.
In general, divalent ligands containing propanamides (B2-F2, H2-J2 and L2) showed faster association kinetics (k on ) compared to their acrylamide analogues (B1-F1, H1-I1), indicating the importance of a certain degree of flexibility to allow simultaneous binding to two LecA binding sites.
Next, we investigated the two most promising compounds H2 and L2 for their efficacy to inhibit LecA binding to H1299 human lung epithelial cells using flow cytometry and confocal microscopy. To this end, LecA was fluorescently labelled with Alexa Fluor 488 NHS ester (LecA-AF488), incubated with varying concentrations of the inhibitors (100 nM to 10 μM) and subsequently added to the cells.
For flow cytometry analysis, inhibitors were preincubated with 0.16 μM LecA prior addition to the H1299 cells. After incubation for 30 min, fluorescence intensity was recorded at the flow cytometer. The sulfonated divalent inhibitor L2 decreased LecA binding to H1299 cells by more than 30 % at 250 nM, whereas divalent pyridine H2 only showed 16 % inhibition (Figure 4a, b). On the other hand, H2 was slightly more potent at the concentration range from 0.5 to 2.5 μM, consistent with its higher affinity measured by SPR. At 1 μM, H2 and L2 decreased LecA binding to the host cells by 50 % and at 7.5 μM and above, a complete inhibition of lectin binding to the cells was achieved. In contrast, 10 μM of the corresponding monovalent ligands G2 and K2 were required for 50 % inhibition ( Figure S10). Thus, both divalent inhibitors, H2 and L2, showed high efficacy to prevent LecA binding to host cells and revealed a significant increase in potency compared to the monovalent inhibitors G2 and K2. When mean fluorescence intensity (MFI) of samples treated with LecA-AF488 and divalent inhibitors were normalized to the negative control and plotted as a function of inhibitor concentration ( Figure S11), H2 and L2 demonstrated a dose-dependent inhibition of lectin binding to the host cells and determined IC 50 values were 0.54 μM for H2 and 0.56 μM for L2 (Figure 4c). On the other hand, the IC 50 values of the monovalent ligands were in the micromolar range (G2 IC 50 = 2.8 μM, K2 IC 50 = 1.9 μM), thus further confirming the benefit of divalent ligands.
The ability of H2 and L2 to diminish LecA binding to host receptors at the plasma membrane and its subsequent internalization into H1299 cells was further analyzed by confocal fluorescence microscopy. Here, the applied LecA concentration was increased to improve signal to noise and consequently image quality and inhibitors were preincubated with 0.5 μM LecA before addition to the cells. After 60 min at 37°C, the cells were imaged: in inhibitor-treated cells, a remarkably low fluorescence signal for LecA was observed at 10 μM divalent ligands compared to those without inhibitor (Figure 4d). Sulfonated inhibitor L2 showed a stronger inhibition of lectin binding than H2 from the pyridine series. A complete inhibition of LecA binding and cellular uptake in the presence of divalent ligands was achieved at 7.5 μM, corresponding to a 10-fold potency increase compared to the monovalent inhibitors G2 and K2, which required 75 μM to block LecA binding ( Figure S11). Confocal microscopy images thus confirmed that the divalent inhibitors can prevent LecA binding to the cell surface and reduce lectin uptake in a concentration-dependent manner. Overall, these data suggest that the divalent inhibitors H2 and L2 have a high potential to block LecA and prevent recognition of the glycosphingolipid Gb3 on host cells, possibly impairing the invasiveness of P. aeruginosa.
The opportunistic bacterium P. aeruginosa depends on favorable circumstances to infect the host, such as a suppressed immunity or wounded skin. [34,35] Chronic infection of wounds with P. aeruginosa increases the risk of nosocomial spread, are difficult to treat and after systemic dissemination can result in a fatal outcome in patients. In a previous study, we investigated the effect of the fucosebinding P. aeruginosa lectin LecB on epithelial wound healing. [36] Cell migration was found to be strongly inhibited in the presence of LecB in a dose-dependent manner, and could be fully blocked with 50 μg mL À 1 (4.3 μM) LecB. Here, we investigated whether LecA also impairs the same physiological process. To this end, H1299 cell monolayers were grown, and after scratching with a pipette tip, wound closure was monitored microscopically (Figure 5a). Cells incubated with 3.9 μM LecA showed a visible decrease in cell migration, compared to full closure of the wound after 24 h in untreated cells. Furthermore, cell detachment in the close proximity of the wound was observed, suggesting that the lectin LecA can significantly hinder cell migration, adhesion and wound closure. The mode of action of LecA in the wound healing process remains elusive. However, in the presence of the monovalent inhibitor K2 (100 μM), wound healing was partially restored to about 50 % after 24 h compared to samples treated solely with LecA ( Figure S13). Importantly, when LecA was added to the cells together with L2 (10 μM and 100 μM, respectively), the wound healed at a similar rate to the control in absence of LecA. Remarkably, in presence of 10 μM L2, cell migration was restored to about 70 % compared to the negative control, and the wound closed by 80 % after 24 h of incubation with 100 μM L2 (Figure 5b). In conclusion, the detrimental effect of LecA in the wound-scratch assay could be reversed by addition of our inhibitors and wound healing was restored to a physiological closure rate. Thus, these data demonstrate that LecA binds to host cells to impair cell migration and, furthermore, unravel the potential of these inhibitors in restoring such a process, an important step towards the treatment and eradication of chronic infections.
P. aeruginosa is able to invade non-phagocytic cells via a lipid zipper mechanism. [10] Binding of LecA to the glycosphingolipid Gb3 followed by LecA-induced receptor clustering plays a crucial role in this process. The LecA ligand 4nitrophenyl α-D-galactopyranoside (PNPG) has been previously shown to interfere with these events at a high concentration of 10 mM leading to a reduction of bacterial invasiveness to 30-60 % as compared to untreated bacteria (100 % invasiveness). [10,31] For invasion assays, H1299 cells were incubated with P. aeruginosa in presence and absence of LecA inhibitors, followed by killing of extracellular bacteria with the antibiotic amikacin. Then, all intracellular bacteria were released by lysis of the human cells and viable bacteria were quantified by plating. To assess the impact of LecA inhibitors on the invasiveness, the influence of the divalent LecA inhibitor L2 on bacterial invasion into host cells was evaluated and compared to PNPG. Compared to untreated bacteria (100 % invasion), invasiveness was slightly reduced to 89 % in presence of 100 μM PNPG and to 44 % with 10 mM PNPG, which is comparable to literature data. [10,31] In contrast, the divalent LecA ligand L2 achieved an invasion decrease to 44 % already at 100 μM ( Figure 5c). L2 was thus able to reach a similar reduction of invasion at a 100-fold lower concentration than the one needed for PNPG.
Additionally, the cytotoxicity of sulfonated ligands K2 and L2 on H1299 cells was evaluated prior to the invasion assay ( Figure S12). K2 and L2 did not exhibit significant toxicity during 24 h in ranges from 10-100 μM and cellular viability was preserved. However, > 50 % inhibition of cell proliferation was observed for higher inhibitor concentrations (500 and 750 μM), indicating a potential toxicity window for the tested compounds.
Encouraged by these promising in vitro results, an in vivo pharmacokinetic study was conducted for H2 and L2 in mice (1 mg kg À 1 , i. v.) and in rats for L2 (10 mg kg À 1 , i. v.). Compound concentrations were monitored in plasma and urine ( Figure 6). H2 and L2 were cleared from mouse plasma within 1 h. Both compounds rapidly reached high concentrations in urine, several orders of magnitude above their respective in vitro LecA binding affinity. In fact, concentrations above the K d s were detected over the entire observation period of 5 h. The overall lower urine exposure of L2 compared to H2 (c urine max = 63.0 μM and 13.4 μM for H2 and L2, respectively) could result from first pass metabolism and a lower hepatic stability of L2 in vitro (Table 2). Interestingly, H2 showed higher renal elimination at 5 h than at 3 h, possibly a result of redistribution from another compartment. In rats, higher amounts of L2 were detected both in plasma and urine, not solely corresponding to the 10-fold increased dose. L2 showed a half-life of 37 � 12 min, a clearance of 9.84 � 5.0 mL kg À 1 min À 1 and a mean residence time 41 � 18 min in rat plasma in vivo, exceeding the concentration of its on-target affinity for at least 6 h. As a result, the overall drug exposure in rat plasma was higher than in mice with an area under the curve (AUC) of 19.7 � 12.8 μM h. In agreement with its strong hydrophilicity, L2 showed a rather low distribution volume (V D = 0.46 L kg À 1 ). Importantly, high levels of L2 were present in rat urine even 24 h after administration.