Receptor design and preliminary experiments
This work represents the first steps in a long-term program aimed at binding 3 and 4, and thus generating new families of antibiotics. Our initial aim was to establish a carboxylate-binding core that can be developed into three-dimensional structures with clefts or cavities providing interactions with groups R’ in R’CO2- targets. Charge-neutral designs were preferred partly for practical reasons, but also encouraged by the neutral binding site of vancomycin (see Fig. 1b). An attractive starting point was the tetralactam family of anion receptors 7 (Fig. 2) studied by Jurczak, Chmielewski and coworkers35,36,37. These macrocycles showed good affinities to carboxylates in DMSO, and possessed sp3 CH groups which could be substituted to generate cleft-type architectures, e.g. by using diamino acid starting materials. Groups R could be modified to engender solubility in water. Because of uncertainty regarding the best solubilising groups to use, we decided to prepare alkynyl functionalised aromatic spacers and use Cu-catalysed azide cycloadditions38,39 for late-stage modifications.
Initial designs are exemplified by macrocycles 8–10 (Fig. 2), prepared from amino acid derivatives 11 and 12 as described in the Supplementary Information. As expected, the organic-soluble 8a-10a were found to be effective anion receptors in DMSO. For example 1H NMR titration of 9a with Bu4N+AcO- caused signal movements up to 0.3 ppm, analysed using the programme Bindfit40 to yield Ka (1:1) = 6200 m‑1 (see Supplementary Fig. 46). To achieve water-solubility we attached a number of charge-neutral solubilising groups, but none could prevent self-association at concentrations realistic for NMR studies, and we therefore resorted to the tricarboxylate unit shown. 1H NMR titrations of macrocycles 8b-10b against acetate and chloride in water resulted in small signal movements (typically ~0.02 ppm), suggestive of binding. The data for 8b did not give satisfactory analyses using any binding model, but the titrations with 9b and 10b were consistent with 1:1 binding with very low affinities (e.g. 8 and 13 m‑1 for 9b with acetate and chloride respectively; see Supplementary Figs. 48 and 49). These preliminary experiments suggested that (a) we should focus on the larger 20-membered macrocycles, which seemed to be more promising; (b) there was little difference between macrocycles containing benzene and pyridine units, so both should be pursued; and (c) negative charges did not preclude measurable anion binding. On the other hand the low affinities, especially to acetate, further emphasised the difficulty of binding hydrophilic anions in water.
Tricyclic octalactams 5 and 6
Our next aim was to raise affinities and control substrate-selectivity by introducing further interactions. An obvious approach was to link the anilino groups in 9b and 10b, creating a bridge with a hydrophobic inner surface and generating an amphiphilic three-dimensional binding site. The diamine 13 is commercially available, inexpensive, and seemed a good candidate for a bridging unit, as in 19. We therefore embarked on the sequences outlined in Fig. 3. On reaching intermediates 18, the expectation was that treatment with base to expose the NH2 groups would be followed by cyclisation to 19. In fact, the major products isolated were the tricycles 20 and 21 in 50% and 60% yields respectively, derived from initial reaction between two molecules of 18 followed by cyclisations at either end. This unintended outcome presumably relates to the conformational properties of the bis(phenoxy)phenyl bridge in 18, which may resist bending into the curved shape required for cyclisation. For 18 (X = CH), the cyclisation was inconveniently slow but could be accelerated by adding chloride ions as templates.
To achieve solubility in water, 20 and 21 were treated with azide 22 under copper catalysed “click/CuAAC” conditions to give 5 and 6. After purification by chromatography, water (or D2O) was then added followed by NaOH to give solutions at pH = ~7.4 for NMR studies. Dilution studies showed small signal movements down to 22 μm for 5 and 74 μm for 6, but no further changes so the tricycles were taken to be monomeric below these concentrations (Supplementary Figs. 36 and 37). Similar cycloadditions were used to place smaller tricarboxylate groups, as used for 8b-10b, at the four positions, but NMR dilution studies implied aggregation even at the lowest concentrations (Supplementary Fig. 40).
Tricycles 20, 21, 5 and 6 are essentially dimeric versions of 9 and 10 (ignoring the change in water-solubilising groups), with identical binding sites at either end. The binding sites are well-spaced, but as they are connected via a macropolycyclic architecture they may not behave independently, so that analysis according to the simple statistical 1:2 model (the divalent equivalent of 1:1 stoichiometry)41 may not be appropriate. Indeed, 1H NMR titrations of 20 and 21 against anions in DMSO-d6 gave very poor fits to the statistical model, with signals often changing their direction of movement part way through the titration (see Supplementary Figs. 54-59). Alternative models, available in Bindfit, are 1:2 non-cooperative, which assumes that affinities of the two binding sites are the same but that chemical shifts of protons in the 1:1 and 1:2 complexes are independent of each other, and 1:2 cooperative which allows for different affinities36,37. Of these two, the 1:2 non-cooperative model gave better fits with lower errors, so was adopted for this work. The results implied that “dimerisation” made little difference to the tetralactams’ binding ability in DMSO. For example, titration of 20 with Bu4N+AcO- in DMSO-d6 yielded Ka = 7560 m‑1, very similar to the value of 6600 m‑1 measured for 9a.
For studies of 5 and 6 in water, a first question was whether the side-chain carboxylates might bind in an intramolecular fashion, potentially preventing the molecules acting as receptors. Examination of models suggested that this should be possible, though not necessarily strongly favoured. A 1H NMR pH titration on 5 was performed, revealing downfield movements of ~0.14 ppm between pH 7.7 and 9.0 for the isophthalamide CHd and amide NHa and NHb signals (see Supplementary Fig. 41). These protons should make contact with bound anions, so the movements are consistent with side-chain carboxylate binding. If this explanation is correct, the data also imply that empty binding sites were available at pH 7.7. We have previously established that the dendrimeric nonacarboxylate solubilising groups are not fully deprotonated at neutral pH42, so more internal binding might be expected at increasing pH. Another factor might be the interaction between -CO2H and -CO2- groups within the partially ionised dendrimers. If both groups are present they might bind to each other, making the carboxylate less available for the tetralactam binding site. In any case, the pH titration indicated that carboxylate recognition should be possible at near-neutral pH, even if affinities might be lowered by intramolecular competition.
Tricycle 5 was then titrated against a number of carboxylate anions as sodium salts. Given the sensitivity of the spectra to changes in pH, all titrations were performed at a steady pH in the range 7.4-7.6. Where necessary, both D2O and 9:1 H2O/D2O were employed as solvents, allowing all protons to be followed. Spectra from a titration of 5 (10 μm) with sodium acetate are shown in Fig. 4a (for signal labels, see Fig. 4c). Significant movements are observed for several receptor signals. Consistent with the pH titration (see above) isophthalamide CHd and amide NHa and NHb moved downfield by ~0.16 ppm. The signals for NHc, CHi, CHj and CHk showed smaller up-field movements ranging from 0.03 ppm to 0.08 ppm. Notably the signal movements were generally much larger than those for 9b + acetate, suggesting a qualitative difference between monomeric and dimeric water-soluble receptors. As observed previously for 20, the 1:2 statistical model was clearly inappropriate due to bidirectional signal movements, and the 1:2 non-cooperative model gave lower errors than 1:2 cooperative. The 1:2 cooperative model, as implemented in Bindfit, was therefore employed for all analyses. In the case of acetate (Fig. 4b), the resulting affinity was ~270 m‑1, roughly thirty times that of 9b for acetate, and remarkably high considering the challenge of binding carboxylates in water and the net negative charge on the receptor. The affinity is all the more impressive given that some interference from side-chain carboxylates is likely at the near-neutral pH.
Fig. 4d shows the anions tested as substrates for receptor 5 and Fig 4e lists the measured binding constants. Some data are also given for tricycle 6, but as this receptor lacks the relatively sharp and mobile1H NMR signal for CHd, it was more difficult to study and received less attention. In general, results for 5 and 6 were fairly similar. An interesting aspect of the results in Fig. 4e is the lack of selectivity between different carboxylates. By extending to polycyclic structures, we had expected to reinforce binding with hydrophobic interactions. However, the fact that formate, acetate, propionate and benzoate gave such similar results suggests that polar interactions are dominant. It seems that incorporation of the tetralactam unit in the tricyclic structure generates a binding site for carboxylates in general, sufficiently powerful to be effective in water. Some enantioselectivity might be expected given the chiral structure of 5, and indeed D-lactate 27 was bound ~1.5 times more strongly than L-lactate 28. The three amino acids tested caused almost no movements of receptor signals, implying very low affinities at best. Inorganic anions were not extensively investigated, but chloride gave signal movements generally similar to carboxylates with roughly 2-fold lower affinity.
Structural insights
To throw light on these results, the structures of the receptors and complexes were investigated by molecular modelling and NMR. The core tetralactam macrocycle 9 (R = H) was subjected to Monte Carlo Molecular Mechanics (MCMM), which yielded a variety of conformations. However MCMM in the presence of formate confirmed that favourable binding geometries are feasible, with all NH groups involved in H-bond donation to the anion (e.g. Supplementary Fig. 81). Dimeric structure 5 is too large for comprehensive studies, but models were elaborated from 9 and minimised with water GB/SA solvation to give an indication of the system’s behaviour. Notably, all the resulting conformations featured folded arrangements whereby the tetralactams are twisted relative to each other, bringing the two bis(phenoxy)phenyl bridges together. Fig. 5a shows one such example (see also Supplementary Fig. 82). Contact between the bridges should be driven by hydrophobic interactions, so these conformations are not unexpected in water. Support for inter-bridge contacts could not be obtained through NOE spectroscopy because of the symmetry of the system, but major differences between the spectra of 20 in DMSO and 5 in water suggested a conformational change on moving between solvents. In particular, the signal due to CHi shifts upfield by ~0.4 ppm on transferring from DMSO to water, while CHk is almost stationary (see Supplementary Fig. 78). This could suggest a folded structure in water promoting inter-bridge shielding. The solvophobic interactions pulling the bridges together should be far weaker in DMSO than in water, so one might expect a shift from flexible to folded conformation on moving between these solvents.
Rigorous modelling of the interaction of 5 with carboxylates was again unrealistic, but we noted that conformations with multiple NH-O- hydrogen bonds and good interbridge interactions could be generated if the bridges were allowed to slide relative to each other. The conformation in Fig. 5b, with propionate guests, is illustrative (see also Supplementary Fig. 83). After energy minimisation, the binding sites at each end of the complex form six hydrogen bonds to carboxylate oxygens, ranging from 1.9 to 2.4 Å. Support for structures of this type was obtained through 2D NOESY investigations of the 5 + propionate complex. Unfortunately the spectra required concentrations of 5 around 1 mM, at which self-association was expected to be considerable. Accordingly, the spectra were dominated by through-space contacts between core and side-chain protons. However, as shown in Fig. 5c, a clear and distinctive cross-peak was observed between the sharp quartet of the propionate and Hi, the bridge proton nearest to the tetralactam core. Modelling based on the conformation in Fig. 5b suggests that these protons can approach within 3 Å of each other in the complex. On the other hand, the model accounts for the promiscuity of receptor 5. Space is available for a variety of groups R in RCO2-, while contact between R and receptor is marginal and unlikely to lead to much affinity enhancement.
These structural studies suggest an explanation for the affinity variations between monomer and dimer structures, and between DMSO and aqueous media. Monocyclic structures 9 and 10 are quite flexible and are fairly effective in DMSO (Ka ~ 103-104 m‑1) but much less so in the far more challenging medium of water. Tricyclic organic-soluble receptors 20 and 21 are presumably more constrained, but not in a manner that enhances binding greatly in DMSO. In contrast, tricyclic water-soluble receptors 5 and 6 are firmly held in a closed twisted state by hydrophobic interactions, and this folded structure is well preorganised for binding. If this is correct, comparisons can be drawn with natural (protein) receptors which denature in organic solvents but fold into active conformations through hydrophobic interactions in water.
In conclusion, we have shown for the first time that synthetic receptors with charge-neutral binding sites can bind carboxylates in water in the absence of organic cosolvent. Selectivity data imply that it is the carboxylate group itself which is complexed, as opposed to some other part of the substrate. Our receptors achieve respectable affinities of ~300 m‑1, despite the use of negatively charged solubilising groups with potential for competitive inhibition. The success of these dendrimeric polycarboxylate solubilising groups is perhaps surprising but also encouraging, in that the study of anion recognition in water becomes much easier if solubility issues can be resolved. The architecture of 5 and 6, featuring macrocyclic binding sites with sp3-hybridised carbon linkage points, is well adapted for elaboration into structures with enclosed, functionalised cavities. The design can therefore serve as a platform for the development of selective C-terminal peptide receptors, and perhaps ultimately for alternatives to the glycopeptide antibiotics.