A family of antibiotics that evades resistance by binding polyprenyl phosphates

Abstract

Cilagicin is a Gram-positive active antibiotic that has a dual polyprenyl phosphate binding mechanism which impedes resistance development. Here we bioinformatically screened predicted non-ribosomal polypeptide synthetase encoded structures to search for antibiotics that might similarly avoid resistance development. Synthesis and bioactivity screening of the predicted structures we identified led to three antibiotics that are active against multidrug-resistant Gram-positive pathogens, two of which, paenilagicin and virgilagicin, did not develop resistance even after prolonged antibiotic exposure.

Graphical Abstract

A schematic illustrating the discovery approach used in this study. NRPS BGCs acquired from JGI and GenBank were bioinformatically queried with the linear polypeptide sequence of cilagicin, those BGCs predicted to share >50% of amino acids in the same positions were pursued for synBNP. Three BGCs met the criteria of this search and were bioinformatically predicted and built using total synthesis. The resultant synBNP molecules were assayed for antibiotic activity and antibiotic resistance profile and we found that paenilagicin and virgilagicin evaded resistance development while bacilagicin did not.

Antibiotic resistant infections are a growing public health threat and therefore new antibiotics capable of circumventing existing resistance mechanisms are essential to combat these infections.¹ Bacterial natural products (NPs) have historically been a rich source of antibiotics with diverse modes of actions.² Unfortunately, accessing microbial NPs from their native producers is hampered by a lack of expression of most biosynthetic gene clusters (BGCs) under laboratory conditions.^{3, 4} To address this bottleneck, we have developed a biology-free discovery approach where BGC products are bioinformatically predicted and their structures are produced by total chemical synthesis. We have termed these molecules synthetic bioinformatic NPs (synBNPs).⁵ This method provides access to bioactive small molecules inspired by BGCs whose products are otherwise inaccessible via culture-dependent methods.⁶

Using a synBNP approach to study non-ribosomal polypeptide synthetase (NRPS) BGCs in sequenced bacterial genomes led to our discovery of the antibiotic cilagicin (1). Cilagicin is a dodeca-lipodepsipeptide with potent activity against a number of clinically relevant multidrug resistant Gram-positive pathogens.⁷ Cilagicin’s antibacterial activity results from its ability to bind both undecaprenyl phosphate (C55:P) and undecaprenyl pyrophosphate (C55:PP). These two polyprenyl phosphates are essential chaperones involved in moving glycopeptide subunits to the outside of the cell where they are polymerized to build the cell wall.⁸ The absence of observed resistance to cilagicin in both clinical isolates and laboratory experiments is likely due to its ability to bind to two distinct molecular targets. Cilgacin’s ability to evade antibiotic resistance makes this class of antibiotics appealing to explore in more detail for clinical development purposes. Here, we sought to identify additional synBNP antibiotics that evade resistance by first bioinformatically screening and then synthesizing cilagicin-like structures predicted from NRPS BGCs found in sequenced bacterial genomes. Our efforts led to the discovery of three antibiotics, two of which (paenilagicin (2) and virgilagicin (4)) sequester both C55:P and C55:PP and do not develop resistance even after prolonged antibiotic exposure.

The cilagicin (cil) BGC was originally selected as a synBNP target based on a phylogenetic analysis of condensation starter (Cs) domains. The cil Cs domain was related to domains from known antibiotic producing BGCs but was associated with a clade that did not contain sequences from any previously characterized BGCs (Figure 1). To identify structure predictions that could serve as synBNP targets to produce antibiotics that do not develop resistance, we searched bioinformatically predicted NRPS-derived NPs for structures related to cilagicin. NRPS biosynthesis takes place in an assembly line fashion involving distinct modules containing sets of domains that build a product one amino acid at a time. A canonical NRPS extender module contains a minimum of three domains: a thiolation (T) domain that passes the growing polymer from one module to the next, an adenylation (A) domain that selects and activates a specific amino acid substrate and a condensation (C) domain that catalyzes the formation of an amide bond between the new amino acid and the previously assembled portion of the peptide.⁹ The amino acid used by each adenylation (A) domain can be predicted based on 10 amino acid residues that line the substrate binding pocket.¹⁰

Figure 1:

Cilagicin (1), a dual polyprenyl phosphate binding antibiotic: A) Molecular structure of cilagicin and cilagicin (cil) BGC. Substrate predictions for each A-domain in the cil BGC are shown. B) Structures of polyprenyl phosphates C55:P and C55:PP.

For this study, we used complete NRPS BGCs collected from sequenced bacterial genomes found in the Joint Genome Institute and GenBank databases.^{11, 12} A-domain substrate binding pockets found in these NRPS systems were compared to a manually curated list of signature sequences collected from characterized NRPS BGCs to generate linear peptide product predictions from each BGC. The resulting database of predicted peptide sequences was queried with the cilagicin linear peptide sequence and hits were ranked based on the number of positionally identical residues (PIRs) shared with the query sequence (Figure S1). Two predicted products were identical to cilagicin, both of which were predicted from BGCs in Paenibacillus mucilaginosus genomes. Only three additional predicted NRPS structures shared 7 or more (i.e. >50%) PIRs with cilagicin. There were no predicted products that shared 6 PIRs with cilagicin, while there was a large diverse collection of sequences that shared 5 or fewer PIRs with cilagicin. We hypothesized that the small number of structures sharing >50% PIRs with cilagicin could potentially share a mode of action with cilagicin and we therefore focused on these structures for this study. The BGCs from which the three potential antibiotics were predicted were found in the sequenced genomes of Paenibacillus puerhi (NZ_WUWM01000006.1), Bacillus cereus (CP068135.1), and Virgibacillus sp. Bac332 (NZ_CP033046.1) (Figure 2).

Figure 2:

Search of predicted NRPS structures for antibiotics that evade resistance. A) Summary of bioinformatic approach used to identify BGCs of interest. Sequenced NRPS BGCs were collected from publicly available genome databases. Clusters without clearly defined start domains (Cs and CAL) and termination domains (TE) were removed because both are required for cilagicin-like structures. Linear peptide sequences from each complete NRPS BGCs were predicted using A-Domain signatures. Resultant predictions were ranked by their identity to the linear sequence of cilagicin. B) Three BGCs that were predicted to encode linear NRPS that are >50% identical to the sequence of cilagicin.

The three predicted BGCs we identified contain either 11 or 12 NRPS modules and are expected to encode unique undeca- and dodeca- peptides. Each linear peptide contains a Thr residue at either the first or second position. As seen in cilagicin, we predicted that the NPs encoded by these BGCs are cyclized through their C-terminal carboxylates and this conserved N-terminal Thr (Figure 3). While the encoded linear peptides are different lengths, cyclization through the Thr would generate an 11 amino acid cilagicin-like macrocycle in each product. In the case of cilagicin, the Cs domain present in the cilC NRPS protein is predicted to add a long chain fatty acid to the N-terminus. Notably, the three BGCs identified in this study contain CoA-ligase (CAL) domains in place of a Cs domain (Table S2–S4). Similar to Cs domains, CAL domains are predicted to append an N-terminal lipid onto the NRPS encoded polypeptides. The absence of a Cs domain in these three BGCs would explain why they were not identified in the original Cs domain phylogenetic analysis that uncovered the cil BGC. NRPS-derived lipopeptides are often produced with a range of different lipids and bioinformatically predicting the exact lipid(s) used in their biosynthesis remains a challenge. In our structure prediction analysis, we used myristic acid because it is one of the most common lipids found in lipopeptide secondary metabolites and it displayed potent activity in our original cilagicin study. Based on these bioinformatic arguments, the undeca- and dodeca- lipodepsipeptides predicted to arise from the three NRPS BGCs we identified are shown in Figure 3. In reference to the organisms in which these BGCs are found, we have named these structure predictions paenilagicin (2), bacilagicin (3) and virgilagicin (4), respectively.

Figure 3:

NP structure predictions. A) Domain composition and A-domain analysis [CoA-ligase (CAL), phosphopantetheine-binding (PP), ketosynthase (KS), acyl transferase (AT), adenylation (A), condensation (C), thiolation (T), epimerization (E), thioesterase (TE)]. B) Final structures predicted to arise from each BGC. Variable (red) and conserved (green) regions are highlighted.

We generated a synBNP based on each predicted structure. Bioinformatically predicted linear peptides were synthesized using Fmoc-based solid phase peptide synthesis ending with a myristic acid on the N-terminus. Following linear assembly, ester bonds were formed on resin between the threonine side chain and the predicted amino acid from the last module in each BGC. Branched linear peptides were released from solid support by HFIP cleavage. Each ring structure was completed in solution via amide coupling between the free amine of the branched amino acid and the carboxylic acid formerly bound to the resin. Cyclized lipodepsipeptides were deprotected in 95% TFA and HPLC purified to yield the final predicted molecular product. All structures were confirmed by HRMS (Table S5, Figures S1–S4) and by 1H and 13C NMR (Figures S5–S10).

Each synBNP structure was tested for antibiotic activity against Gram-positive and Gram-negative bacteria as well as human cells (Table 1, Table S6). As seen with cilagicin, all three synBNPs exhibited activity against clinically relevant gram-positive pathogens. Against most strains, the new compounds showed a slight reduction in potency compared to cilagicin; however, paenilagicin and virgilagicin were slightly more active than cilagicin against Clostridium difficile. Generally, they did not have Gram-negative activity, although paenilagicin, like cilagicin, showed mild activity against Acinetobacter baumannii. No synBNPs inhibited the growth of human cells at the highest concentration tested.

Table 1:

MIC activity (μg/mL) of synBNPs

	Cilagicin (1)	Paenilagicin (2)	Bacilagicin (3)	Virgilagicin (4)
Gram-Positive
Staphylococcus aureus	1	2	2	4
Enterococcus faecium	2	2	4	8
Enterococcus faecalis	1	2	2	4
Clostridium difficile	4	2	4	2
Streptococcus agalactiae	1	2	4	4
Gram-Negative
Acinetobacter baumannii	4	8	>64	>64
Escherichia coli	>64	>64	>64	>64
Human Cells
HEK293 (IC50)	>64	>64	>64	>64

Cilagicin is a bi-functional antibiotic that is able to sequester both C55:P and C55:PP. Addition of excess C55:P or C55:PP to culture media suppresses cilagicin’s activity by sequestering the antibiotic away from the cell wall of target bacteria. We explored the role of C55:P and C55:PP in the activity of each synBNP by determining its MIC against Staphylococcus aureus in culture broth supplemented with varying ratios of antibiotic and polyprenyl phosphate (Figure 4). Like cilagicin, the activity of paenilagicin and virgilagicin were suppressed by C55:P and C55:PP in a dose dependent manner, suggesting these structures retain both polyprenyl phosphates as molecular targets. The activity of bacilagicin was only suppressed by C55:P. In the case of C55:PP, the MIC of bacilagicin remained largely unchanged even when five-fold molar excess of C55:PP was added to the assay media, suggesting that bacilagicin does not sequester C55:PP (Figure 4A). Cilagicin’s antibacterial activity was completely suppressed at less than a 2-fold molar excess of C55:P or C55:PP. When suppression of antibacterial activity was observed for the new synBNPs it required ~3-fold molar excess of a polyprenyl phosphate. This difference in polyprenyl phosphate affinity could explain the lower potency seen for these antibiotics.

Figure 4:

Suppression of antibiotic activity by polyprenyl phosphates and development of resistance in serially passaged cultures. A) MICs of antibiotic against S. aureus in the presence of different molar ratios of C55:P or C55:PP. The highest concentration of peptide tested was 64 ug/mL. Average of two replicate experiments with error bars representing standard deviation. B) Fold change in MIC from Day 0 to Day 14 of each synBNP against serially passed S. aureus cultures exposed 0.5x MIC of the same antibiotic.

The ability to bind two molecular targets (C55:P and C55:PP) is what we believe enables cilagicin to avoid the development of antibiotic resistance even after prolonged exposure. Other antibiotics that bind a single polyprenyl phosphate molecule (e.g., amphomycin and bacitracin) typically develop resistance quickly.^{13, 14} As such, we expected that bacilagicin would no longer be able to avoid resistance development during long-term exposure to a pathogen. To test this, we attempted to raise S. aureus antibiotic resistant mutants by daily serial passage for 14 days in the presence of sub-MIC (0.5x MIC) levels of paenilagicin, bacilagicin, virgilagicin, cilagicin or amphomycin. As anticipated, cultures exposed to paenilagicin or virgilagicin, like those exposed to cilagicin, did not develop antibiotic resistance. By contrast, cultures exposed to bacilagicin, quickly showed a 4-fold increase in MIC, following a similar pattern to amphomycin (Figure 4B). These results reinforce our hypothesis that sequestration of both C55:P and C55:PP provides a unique antibacterial mechanism against which it is difficult for pathogens to develop resistance.

The structures of these polyprenyl phosphate binding antibiotics show the most variability in the region surrounding the site of cyclization (Figure 3B). This variable region includes the three C-terminal residues of each peptide and the residues adjacent to the conserved Thr. The central region of each macrocycle is more highly conserved. In fact, these 10-membered macrocycles all contain a “DGnxGY” motif that we predict is important for target engagement and therefore potentially useful for guiding the discovery of additional polyprenyl phosphate binding antibiotics in the future. A key difference between bacilagicin, which only binds C55:P, and the other structures in this family that bind both C55:P and C55:PP is the absence of the positively charged residue at position 11. While a more extensive SAR study would be required to determine the role of this residue in target engagement, the extra positive charge may be important for binding the additional negative charge found on the pyrophosphate in C55:PP. We did not find any known antibiotics that share the “DGnxGY” motif. The closest match we found was in locillomycin; however, the stereochemistry of multiple residues is inverted in this structure.¹⁵ Locillomycin also contains a 9 membered macrocycle in place of the 11 membered ring seen in the polyprenyl phosphate binding synBNPs we have identified. Not only does locillomycin differ in structure from these antibiotics, but biosynthetically it is predicted to arise from the repetitive use of some NRPS modules making the BGC much smaller than those described here. To the best of our knowledge the molecular target of locillomycin has not been reported and therefore whether it binds one or both polyprenyl phosphates, or if it has a different molecular target, remains to be determined.

While cilagicin is a promising candidate for the development of antibiotics that can overcome resistance mechanisms plaguing our current arsenal of approved drugs, as with any preclinical class of therapeutics, many unforeseen issues could arise during the development of the compound. With this in mind, we sought to expand the available structural diversity within the cilagicin family by bioinformatically screening sequenced bacterial genomes for BGCs predicted to encode cilagicin-like structures. This search led us to synthesize three additional members of this mechanistically novel class of antibiotics. As seen with cilagicin, two of these structures do not develop resistance even after extended antibiotic exposure. These structures should provide alternative drug development candidates should they be needed in the future. We believe that coupling synBNP methods with the targeted search of databases comprised of bioinformatically predicted BGC product structures is now a straightforward and broadly applicable approach for identifying bioactive small molecules with specific desirable features.¹⁶

Methods

General Experimental Procedures and Materials:

All reagents and solvents were purchased from commercial sources and used without further purification. Solvents used for chromatography were HPLC grade or higher. Preparative HPLC was performed on an CombiFlash EZ Prep purification system with UV detection and equipped with a Phenomenex Luna 5μm C18 prepHPLC column using a dual solvent system (A/B: water/acetonitrile, supplemented with 0.1% (v/v) formic acid). HRMS and MS/MS data were acquired on a SCIEX ExionLC UPLC coupled to an X500R QTOF mass spectrometer, equipped with a Phenomonex Kinetex PS C18 100 Å column (2.1 × 50 mm, 2.6 µm) and operated by SCIEXOS software. 1H NMR and 13C NMR spectra were acquired at room temperature on a Bruker Avance DMX 600 MHz spectrometer (The Rockefeller University, New York, NY) equipped with cryogenic probes and spectra were analyzed using MestReNova software (version 14.3.0–30573). Chemical shift values were reported in ppm and referenced to residual solvent signals, for 1H NMR: DMSO-d6 = 2.54 ppm; for 13C NMR: DMSO-d6 = 40.45 ppm.

Identification and bioinformatic analysis of natural cilagicin biosynthetic gene clusters (BGC):

Sequenced nonribosomal peptide synthetase (NRPS) BGCs were collected from the bacterial genome databases JGI and GenBank. BGCs without clearly defined starting (condensation start (Cs) or CoA Ligase (CAL)) and ending (thioesterase (TE)) domains were removed from the collection. For the remaining complete sequenced NRPS BGCs, the 10 amino acids that make up each adenylation (A)-domain binding pocket (i.e., amino acids 235, 236, 239, 278, 299, 301, 322, 330, 331, and 517) were identified using a curated list of A-Domain substrate signatures to predict the substrate of each BGC A-domain. These A-Domain signatures allowed us to make a linear polypeptide sequence prediction for each NRPS BGC in the collection. Using the linear polypeptide sequence of cilagicin as a query term, we ranked NRPS BGCs by their linear polypeptide sequence similarity to cilagicin. BGCs in the resultant list that shared ≥50% polypeptide sequence similarity to cilagicin we deemed congeners. This search yielded two BGCs from Paenibacillus mucilaginosus genomes that shared 100% polypeptide sequence identity to cilagicin. Three BGCs were found to share 7 or more (≥58%) of the 12 amino acid positions in cilagicin. The remaining BGCs from the search shared 5 or fewer (≤42%) amino acid positions with cilagicin. The three BGCs with ≥50% polypeptide sequence similarity to cilagicin were carried forward as the congener BGCs investigated in this study.

Solid Phase Peptide Synthesis:

Natural cilagicin analogs characterized in this study were synthesized using standard Fmoc-based solid-phase peptide synthesis (SPPS) methods on 2-chlorotrityl chloride resin. All peptides were synthesized starting from the penultimate module’s amino acid, for paenilagicin this was ornithine, for bacilagicin this was serine, and for and virgilagicin this was arginine. 2-cholorotrityl resin pre-loaded with the appropriate amino acid was swollen in DCM for 30 minutes at room temperature then drained and washed with DMF (3 mL, 3x). Subsequent couplings were carried out using Fmoc-protected amino acids (or a fatty acid) (3 equiv. relative to resin loading) mixed with HATU (3 equiv.) and DIPEA (3 equiv.) in DMF (5 mL). Each coupling reaction was carried out for 45 minutes at room temperature then washed with DMF (5 mL). Fmoc deprotection was carried out by treating resin-bound peptide with 20% piperidine in DMF (5 mL) for 5 minutes (2x). After deprotection, the resin was then washed with DMF (5 mL, 2x), DCM (5 mL, 2x), and DMF (5 mL, 2x). These steps were repeated for each amino acid and fatty acid to construct the linear peptides.

Ester bond formation:

Ester bonds were formed between the unprotected threonine hydroxyl and the carboxylic acid of the final module’s amino acid. For paenilagicin and bacilagicin this amino acid is tyrosine. For virgilagicin this amino acid is valine. The resin-bound peptide with a free hydroxyl group was mixed with the appropriate Fmoc-AA (15 equiv.) and DIC (15 equiv.) in 7mL DMF. DMAP (0.5 equiv.) was added to the solution, and gently shaken for ~16 hours at room temperature.

Peptide cyclization:

Resin-bound linear peptides were cleaved by treating with 20% hexafluroisopropanol (HFIP) in DCM for 1 hour (2x). Crude linear peptides were then collected by filtration and dried under reduced pressure. The cleaved linear peptides were cyclized without purification by resuspending in DMF to 0.002M and then mixing with PyAOP (7 equiv.) and DIPEA (20 equiv.). After 2 hours, reaction was transferred to a separatory funnel and ethyl acetate (2.5x volume of DMF) was added. This organic layer was washed with saturated brine (4x), then dried over sodium sulfate. Dried organic layers were filtered and concentrated under reduced pressure to yield crude cyclized peptide.

Bulk deprotection:

Peptides were dissolved in 6 mL of cleavage cocktail (95% (v/v) TFA, 2.5% (v/v) triisopropylsilane and 2.5% (v/v) water) for 1.25 hours. Cleavage cocktail was evaporated under air flow to yield crude deprotected peptides.

Peptide purification:

Crude peptides were purified on an Phenomenex Luna 5μm C18 prepHPLC column attached to a CombiFlash EZ Prep purification system using a dual solvent system (A/B: water/acetonitrile, supplemented with 0.1% (v/v) formic acid). Peptide purity and identity were confirmed by UPLC, HRMS, and NMR.

Minimum inhibitory concentration (MIC) assay:

MIC assays were conducted using the protocol recommended by the Clinical and Laboratory Standards Institute.¹⁷ Culture conditions (temperature, medium) are detailed in Supplementary Table S6. All compounds were dissolved in sterile DMSO (ATCC, USA) to give a concentration of 6.4 mg/mL. Tested compounds were serially diluted 2-fold in DMSO from a maximum stock concentration of 6.4 mg/mL to 0.006 mg/mL. In a 96-well plate filled with 49 μL fresh growth medium, 1 μL of compound stock dilution was added across wells in a row. An overnight culture of an assay strain was diluted 5,000-fold in fresh medium. 50 μL of this inoculum dilution was added into each well, giving a final volume of 100 μL per well. Final assayed concentrations of test compounds ranged from 64 μg/mL to 0.06 μg/mL. MIC values were recorded as the minimum concentration at which no bacterial growth appeared, based on visual inspection, after 16 hours of static incubation at 37 °C. Clostridium difficile plates were statically incubated under anaerobic conditions (Vinyl anaerobic chamber, 37 °C, 5% H2, 5% CO2, 90% N2). MICs were performed in technical duplicate (n=2) and repeated three independent times (n=3).

Cytotoxicity assay:

The cytotoxicity of natural cilagicin analogs were tested using an MTT (3-(4,5-Dimethyl-2-thiazolyl)-2,5-diphenyl-2H-tetrazolium bromide) assay. HEK293 cells were seeded in a 96-well plate with a density of 5,000 cells/well and cultured in Dulbecco’s Modified Eagle Medium (DMEM) without phenol red and supplemented with 10% fetal bovine serum, 1% Pen/Strep and 1% glutamate for 24 hours at 37 °C with 5% CO₂. Serially diluted compounds were added into each well at a final concentration ranging from 64 μg/mL to 0.06 μg/mL. After 48 hours of incubation, the media was removed and 15 μL of freshly prepared MTT solution (5 mg/mL in DPBS) was added to each well. The plates were incubated for 3 hours at 37 °C with 5% CO₂ after which the MTT solution was removed by aspiration. Precipitated formazan crystals were dissolved by addition of 100 μL of solubilization solution (40% DMF, 16% SDS and 2% acetic acid in H2O). The absorbance of each well was measured at OD570nm using a microplate reader (Infinite 200 PRO, Tecan). All experiments were performed in duplicate (n=2) and repeated three independent times (n=3).

Undecaprenyl phosphate feeding assay:

The effect of cell wall phospholipids undecaprenyl phosphate (C55:P) and undecaprenyl pyrophosphate (C55:PP) on natural cilagicin analogs’ antibacterial activity was evaluated by co-drying peptide and lipid at molar ratios from 0x to 5x, at 0.5x increments, in plastic tubes in vacuo for 2 hours to completely remove all organic solvent. After drying, compounds were resuspended in scant 2.5 μL methanol, then in 50 μL fresh LB to bring the peptide concentration to 128 μg/mL, followed by vigorous sonication and vortexing. 25 μL of this solution was transferred in duplicate to a 384 well plate and serially diluted 2-fold in LB medium from 128 μg/mL to 0.012 μg/mL. An overnight culture of S. aureus USA300 was diluted 5,000-fold in fresh LB medium. 12.5 μL of this inoculum dilution was added to each well, giving a final volume of 25 μL per well. Final assayed concentrations of test compounds ranged from 64 μg/mL to 0.06 μg/mL. MIC values were recorded as the minimum concentration at which no bacterial growth appeared, based on visual inspection, after 16 hours of static incubation at 37°C. All assays were run in duplicate (n=2) and repeated two independent times (n=2).

Evaluating antibiotic resistance by serial passage in liquid broth:

A single colony of S. aureus USA300 was inoculated in 5 mL LB broth and grown overnight at 37 °C with continuous shaking (200 rpm). The overnight culture was then diluted 1:5,000 into fresh LB medium. 50 μL aliquots of dilute cells were transferred into individual wells of 96-well plates containing 50 μL of serially diluted cilagicin, natural cilagicin analogs, and amphomycin (Cayman Chemical Company, USA), in accordance with the standard MIC assay set up described above. Note: stock dilutions of test compounds were prepared fresh daily. Plates were statically incubated at 37 °C. After 24 hours, the MIC was recorded. For the next round of assays, an aliquot from the culture well at half of the MIC from the previous day’s MIC plate was diluted 5000-fold in fresh LB and mixed with serially diluted antibiotics. The MIC was determined as described above. This process was repeated daily for 14 days. For amphomycin, LB medium was supplemented with 100 μg/mL CaCl2-2H2O. Experiments were performed three independent times (n=3).

Data Availability Statement

The data underlying this study are available in the published article and its Supporting Information.