Equal molar bilateral alanine-based hydrazides 4 (0.67 g, 2 mmol) and 1,3-phenylene diisothiocyanate 5 (0.38 g, 2 mmol) were dissolved in 50 mL CH₃CN and stirred at 80°C for 12 h (Figure 1A). Removing the solvent through evaporation in vacuum, the crude product showed a set of well-resolved ¹H NMR signals (Supplementary Figure 1), indicating the formation of a main product with well-defined structure and of high purity and high yield. This product was suggested to be the cyclic azapeptide 1 according to the MALDI-TOF mass spectrum that shows a sharp peak of 528.03 (Supplementary Figure 2, calcd for [1+H]⁺ [C₂₂H₂₄N₉O₄S₂]⁺: 529.14). After recrystallizing in CH₃CN, pure product 1 of high isolated yield (72%) was obtained. Generally, the two-point reaction could result in diverse oligomers, polymers, and cyclic compounds, which have not been found during the synthesis of 1. The meta-position of the two benzene linkers and the folded β-turn structures are expected to contribute to the accessible synthesis of 1. Acyclic analogs 2 and 3 (Figure 1B) were synthesized according to the procedures described in Scheme S1 in Supplementary Material.

β-Turn structure has been well-identified in alanine-based N-amidothiourea in our previous works. The calculated structure of cyclic azapeptide 1 features two β-turns as shown in Figure 2A. ¹H NMR spectra of 1 were recorded in 90:10 (v/v) CD₃CN–DMSO-d₆ mixtures at variable temperatures (Supplementary Figure 3), from which the temperature coefficients of the chemical shift of –NH protons were obtained (Figure 2B). Different from –NH_a (−7.66 ppb/°C), –NH_b (−7.21 ppb/°C), and –NH_c (−10.91 ppb/°C), which have very negative values, a much more positive temperature coefficient of thioureido –NH_d (−1.94 ppb/°C) was observed, suggesting the protection of –NH_d by intramolecular interaction, ascribed to the intramolecular 10-membered ring hydrogen bonds that maintain the two β-turns in 1 (Figure 2A) (Cao et al., 2017; Yan et al., 2019; Zhang et al., 2020). The obvious NOE signals of H_e-H_f and H_d-H_f again support the folded β-turn structures in cyclic azapeptide 1 (Supplementary Figure 4 and Figure 2C).

Folded β-turn structures also exist in acyclic control compounds 2 and 3, suggested by the DFT calculated structures (Supplementary Figure 5), as well as the temperature coefficients of the chemical shift of the –NH protons (Supplementary Figures 6–9). Given the more positive temperature coefficient of the –NH_d in 1 (-1.94 ppb/°C) than those in 2 and 3 (−2.65 and −2.64 ppb/°C, respectively, Table 1), it is believed that the intramolecular hydrogen bonds in 1 are more stable than those in 2 and 3, owing to the rigid cyclic conformation that stabilizes the β-turn structures. Meanwhile, CD spectra in CH₃CN show an obvious CD signal at 270 nm of 1 and weak signals for 2 and 3, suggesting the more efficient intramolecular chirality transfer from chiral alanine residues to achiral phenylthiourea chromophore in 1, deduced from the absorption spectra (Figure 2D). Meanwhile, the cyclic conformation also leads to higher lipophilicity of 1, supported by their retention times in reverse-phase HPLC (Supplementary Table 1). This could be a potential advantage of 1 for anion transmembrane transport (Valkenier et al., 2014).

The binding capacity of 1 toward halide anions in CH₃CN was next examined. Despite no perceivable binding to Cl⁻, Br⁻, and I⁻ (Supplementary Figures 10–12), a substantial change in absorption and CD spectra of 1 was observed in the presence of F⁻ (Figure 3 and Supplementary Figure 13), undergoing multiple processes since the two binding groups, thiourea moieties, in 1 are in close proximity. Upon the addition of F⁻ from 0.0 to 1.0 eq, the absorbance at 272 nm that from the thiourea chromophore is increased, along with the development of a new band at 315 nm, which is assigned to the charge transfer band of the anion–thiourea complex (Figure 3A). Significantly, a new CD band develops at 315 nm, suggesting the occurrence of intermolecular chirality transfer. With increasing F⁻ concentration from 1.0 to 2.0 eq, the absorption at 272 nm blue shifts to 263 nm, while that at 315 nm shifts to 295 nm. While the CD signals at 272 and 315 nm become more negative, the positive signals at 238 and 215 nm are also enhanced, leading to bisignate and coupled Cotton effects at 272 and 238 nm (Figure 3B). From 2.0 to 5.0 eq F⁻, both absorption and CD spectra are changed slightly (Figure 3C). Thus, a final 1:2 binding complex is suggested for 1 in the presence of F⁻. The optimized structure of 1·2F⁻ complex shows that one F⁻ interacts with 1 through three N–H···F⁻ hydrogen bonds in the cavity of each side, breaking the β-turn structures. Therefore, one 1 molecule binds two F⁻ ions through six hydrogen bonds (Supplementary Figure 14 and Supplementary Table 2), resulting in a high binding energy up to−158.12 kcal mol⁻¹ or−79.06 kcal mol⁻¹ per F⁻ ion. Moreover, the calculated CD spectrum performed by TD-DFT is similar to the experimental spectrum of 1 in the presence of 2.0 eq F⁻ (Supplementary Figure 15), indicating the reliability of the calculated binding model.

Acyclic compounds 2 and 3 can also bind with fluoride in CH₃CN by employing two thiourea groups. In 2, the two thiourea moieties are distant, so only one equal binding process is shown in the titration spectra, in which the absorbance at 292 nm can be assigned to the F⁻ binding complex of charge transfer transition (Supplementary Figure 16). Thereby, intermolecular chirality transfer is indicated by the emerged Cotton effects at 292 nm. Differently, the two thiourea groups in 3 are in close proximity, and the resultant binding complexes exhibit enhanced absorption and CD peaks at 266 nm, while the CD signals at 242 nm from the alanine residues are decreased (Supplementary Figure 17). The binding constants of acyclic 2 (K = 2.2 × 10⁵ M⁻¹, Supplementary Figure 16) and 3 (K₁₁ = 5.9 × 10⁵ M⁻¹, K₁₂ = 8.5 × 10³ M⁻¹, Supplementary Figure 17) with F⁻ are fitted to be much lower than those of cyclic 1 with F⁻ (K₂₁ = 1.7 × 10⁵ M⁻¹, K₁₁ = 1.9 × 10⁸ M⁻¹, K₁₂ = 8.6 × 10⁵ M⁻¹, Supplementary Figure 13). No binding events are observed for 2 and 3 in the presence of Cl⁻, Br⁻, and I⁻ (Supplementary Figures 18–23).

The selective binding of 1–3 to fluoride implies that they might be candidates for selective transport of fluoride. We thus studied the fluoride transport activities of 1–3 according to osmotic assay as shown in Figure 4A (Clarke et al., 2016). Large unilamellar vesicles (LUVs, 400 nm), composed of 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC), were loaded with a buffered KF solution (300 mM) and suspended in a buffered K⁺ gluconate solution (300 mM). Transporters were added in the presence and absence of valinomycin (Vln, 0.5 μM, 0.1 mol%) to initiate the transport process (Figure 4A). The large hydrophilic anion gluconate was used as the extravesicular anion, which rules out an anion exchange mechanism. Without Vln, a fluoride transporter will create a membrane potential under the concentration gradient of fluoride, preventing fluoride efflux. Then Vln, a natural potassium carrier, can dissipate the membrane potential by transporting potassium leading to observable fluoride efflux. The overall KF cotransport will give rise to an osmotic gradient and drive the efflux of water from vesicles, leading to vesicle shrinkage. This process can be monitored by a fluorometer as an increase of 90° light scattering that results from the vesicle shrinkage. In the absence of Vln, the compounds 1–3 show no response from osmotic assay (Supplementary Figure 24), suggesting that the transporters cannot facilitate K⁺/F⁻ symport or F⁻/gluconate antiport. Then Vln was added before the measurement began, and compound 1 leads to a significant increase in light scattering intensity, which was not observed for 2 and 3 (Figure 4B and Supplementary Figure 25), implying that transporter 1 is capable of transporting fluoride via an electrogenic anion uniport mechanism. The transport activity was quantified by employing transporter 1 at different concentrations (Figure 4C). Fitting the data with Hill equation, EC₅₀ (the effective concentration to reach 50% of maximum transport at 300 s) and n (Hill coefficient) are obtained and shown in Figure 4D. The concentration of lipids in osmotic assay is 500 μM; the EC₅₀ value (53.12 μM) corresponds to a transporter/lipid ratio of 10.62 mol%. The corresponding Hill coefficient n (2.24) is attributed to F⁻ being transported as a 2:1 (transporter:anion) complex, which presumably leads to more effective shielding of the negative charge of F⁻ compared with 1:1 and 1:2 complexes. Although acyclic counterparts 2 and 3 can bind fluoride as shown by absorption and CD titrations, they show no fluoride transport activity from the osmotic assay, even in the presence of Vln (Supplementary Figure 25), ascribed to their much lower lipophilicity (Supplementary Table 1), as well as lower F⁻ affinity.

Ion-selective electrode (ISE) assay was next applied to directly measure the fluoride transport and confirm the transport mechanism (Wu et al., 2016). POPC LUVs (mean diameter 200 nm) were prepared and used in this section. In the absence of Vln, negligible efflux of fluoride was detected, even when the concentration of transporter 1 is high up to 40 mol%. In the presence of Vln, the KF efflux could reach 47% in the presence of 40 mol% 1 and was increased to be 85% when the concentration of transporter 1 was up to 80 mol% (Figure 5). The transport activity of 1 depends on the presence or absence of Vln, confirming the electrogenic anion transport mechanism. Again, acyclic compounds 2–3 show negative response in the ISE assay (Supplementary Figure 26).

The transport mechanism (carrier or channel) was next investigated by employing lipids composed of dipalmitoylphosphatidylcholine (DPPC), which feature a phase transition at 41°C (Wu et al., 2015). The decreased fluidity of lipids below 41°C could suppress the transport activity of ion carriers but not ion channels. The amount of F⁻ transport of 1 at 300 s decreased from 47% at 43°C to 15% at 37°C in the osmotic assay (Figure 6), supporting the carrier transport mechanism.

It is noteworthy that HF is membrane permeable, and this could lead to an artifact in F⁻ transport studies when the transporter facilitates H⁺ (OH⁻) transport (Clarke et al., 2016). In the F⁻ transport artifact, the pH gradient generated by the HF efflux is dissipated by the H⁺ influx (or OH⁻ efflux) facilitated by the test transporter coupled to the K⁺ efflux facilitated by Vln, leading to the net KF efflux. Therefore, a H⁺ or OH⁻ transporter (e.g., FCCP) can generate a positive response in a F⁻ transport assay even if the transporter does not facilitate “true” F⁻ transport by reversibly binding F⁻ ions. KOAc-loaded vesicles were thus utilized to rule out H⁺ and OH⁻ transport, under which condition membrane-permeable acetic acid forms and would lead to the KOAc efflux in the presence of Vln and a H⁺ (OH⁻) transporter. Alternatively, acetate ion transport facilitated by a transporter would also lead to a response in the presence of Vln. If a transporter does not facilitate H⁺ (OH⁻) transport or acetate transport, acetate will stay inside the LUVs in the form of membrane-impermeable acetate ion. In our experiment, no detectable efflux of acetate was found (Supplementary Figure 27), confirming that transporter 1 is unable to transport H⁺ (OH⁻) or acetate. Thus, the KOAc-loaded osmotic assay, together with the KF-loaded osmotic assay, provides unambiguous evidence that cyclic compound 1 is capable of transporting fluoride via an electrogenic anion uniport mechanism. The integrity of the lipid bilayer can be evidenced by the calcein leakage assay (Wu et al., 2015). Calcein was encapsulated inside POPC LUVs; its fluorescence intensity was monitored to study whether pores are formed in lipids. Only negligible leakage of calcein occurred during the measurement (Supplementary Figure 28), confirming that cyclic compound 1 is functioning as a carrier rather than destroying the vesicle structure or forming large pores.

In view of the fact that the reported fluoride transporters are also active for chloride transport (Clarke et al., 2016; Park et al., 2019; Park and Gabbai, 2020), it is necessary to investigate the selectivity of fluoride over chloride. Hence, LUVs loaded with KCl (300 mM) were prepared to evaluate the transport activity toward chloride (Supplementary Figure 29A). Gratifyingly, no positive response could be observed in the Vln-coupled KCl efflux (Supplementary Figure 29B), suggesting the inactivity of 1–3 toward chloride transport, and thereby a transport selectivity of 1 toward fluoride over chloride. This highlights the importance of a small cyclic structure for selective ion transport.

Isophthalic acid, benzoic acid, L-alanine, 1,3-phenylene diisothiocyanate, phenyl isothiocyanate, acetonitrile for reaction, and acetonitrile for spectroscopy were purchased from Sigma Aldrich Co. and used directly without purification. Acetonitrile-D₃ and dimethyl sulfoxide-D₆ were purchased from Cambridge Isotope Laboratories Inc. All other starting materials were obtained from Sinopharm Chemical Reagent Ltd.

¹H and ¹³C NMR spectra were recorded on a Bruker AV600MHz or AV850MHz spectrometer. High-resolution mass spectra (HRMS) were acquired on a Bruker micro TOF-Q-II mass spectrometer. Circular dichroism (CD) spectra were recorded on Jasco J-1500. Absorption spectra were recorded on a Shimadzu UV-2700 UV-Vis spectrophotometer. Fluorescence spectra were obtained on a Horiba Fluorolog-3 spectrometer or a Hitachi F-7000 spectrometer. ISE assay was conducted by an Orion Dual Star.

Cyclic compound 1 and reference linear compound 2–3 were synthesized according to reported literatures with little modification (Yan et al., 2019). Isophthaloyl dichloride was synthesized by refluxing isophthalic acid (10.0 mmol) with SOCl₂ (15.0 mmol) in CH₂Cl₂ (50.0 mL) overnight, evaporating solvent and unreacted SOCl₂ on a rotary evaporator to obtain white solid without further purification. Isophthaloyl dichloride was mixed with AOEt·HCl (2.0 g, 13.0 mmol) and Et₃N (3.0 mL) in CHCl₃ (50.0 mL) and stirred overnight. The mixture was washed successively with NH₃ · H₂O (1%, 50.0 mL), HCl (1%, 50.0 mL), and saturated NaCl solutions (50.0 mL). Then pure product 6 (80% yield averagely) was obtained by drying the solution over anhydrous Na₂SO₄ and concentrating in vacuo. To synthesize product 4, excess aqueous hydrazine (85%, 6.0 mL) was added to 6 in EtOH (50.0 mL) and refluxed the mixture for 24 h. The solvent was removed by evaporation to obtain a crude product and washed with CH₃CN several times to afford pure white product 6 (65% yield averagely).

Linear compound 2 was synthesized by refluxing 4 (1.0 mmol) with an excess amount of Ph(NCS) (2.2 mmol) in CH₃CN for 24 h. Pure product 2 (90% yield averagely) precipitated from the solution and could be used directly without further purification.

8 was synthesized and purified under the similar way as 4, then 3 can be acquired by refluxing excess 8 (2.2 mmol) with 5 (1.0 mmol) in CH3CN for 24 h. After filtrating, pure solid product 3 was obtained (90% yield averagely).

1: ¹H NMR (600 MHz, DMSO-d₆): δ (ppm) 10.44 (s, 2H), 9.81 (s, 2H), 9.28 (d, J = 5.3 Hz, 2H), 9.23 (s, 2H), 8.61 (s, 1H), 8.19 (dd, J = 7.8, 1.7 Hz, 2H), 7.86 (dd, J = 8.1, 1.9 Hz, 2H), 7.65 (t, J = 7.7 Hz, 1H), 7.36 (t, J = 8.1 Hz, 1H), 6.91 (t, J = 1.9 Hz, 1H), 4.32–4.25 (m, 2H), 1.38 (d, J = 7.1 Hz, 6H); ¹³C NMR (214 MHz, DMSO): δ (ppm) 180.98, 171.88, 167.21, 139.44, 133.88, 130.71, 128.77, 128.55, 127.48, 123.84, 120.30, 49.69, 16.75; HRMS (ESI): calcd for [C₂₂H₂₄N₈O₄S₂Na]⁺: 551.1254, found: 551.1245.

2: ¹H NMR (600 MHz, DMSO-d₆): δ (ppm) 10.43 (s, 2H), 9.75 (s, 2H), 9.30 (s, 2H), 9.04 (s, 2H), 8.43 (s, 1H), 8.06 (d, J = 7.4 Hz, 2H), 7.65 (d, J = 25.5 Hz, 4H), 7.61 (t, J = 7.6 Hz, 1H), 7.33 (t, J = 7.3 Hz, 4H), 7.14 (t, J = 6.9 Hz, 2H), 4.40 (s, 2H), 1.43 (d, J = 6.3 Hz, 6H); ¹³C NMR (151 MHz, DMSO): δ (ppm) 180.74, 172.28, 167.48, 139.48, 134.19, 131.04, 128.74, 128.62, 127.69, 125.26, 124.62, 49.52, 17.19.; HRMS (ESI): calcd for [C₂₈H₃₀N₈O₄S₂Na]⁺: 629.1724, found: 629.1726.

3: ¹H NMR (600 MHz, DMSO-d₆): δ (ppm); 10.43 (s, 2H), 9.76 (s, 2H), 9.38 (s, 2H), 8.92 (s, 2H), 8.07 (s, 1H), 7.93 (d, J = 7.3 Hz, 4H), 7.54 (t, J = 7.3 Hz, 2H), 7.49 (t, J = 7.4 Hz, 4H), 7.41 (d, J = 7.7 Hz, 2H), 7.32 (t, J = 8.0 Hz, 1H), 4.34 (s, 2H), 1.41 (d, J = 6.7 Hz, 6H); ¹³C NMR (214 MHz, DMSO): δ (ppm) 180.71, 172.37, 167.97, 139.38, 133.66, 132.09, 128.75, 128.21, 127.94, 121.84, 121.68, 49.58, 17.06; HRMS (ESI): calcd for [C₂₈H₃₀N₈O₄S₂Na]⁺: 629.1724, found: 551. 629.1713.

The large unilamellar vesicles (LUVs, mean diameter 400 nm) of POPC were prepared by freeze/thaw cycles; 30.4 mg POPC dissolved in CHCl₃ (50 mL) was slowly evaporated using a rotary evaporator and dried under high vacuum at room temperature for 4 h. The obtained lipid film was rehydrated by mixing with 4 mL saline solution (300 mM KF, 10 mM HEPES, pH 7.2) and vortexing for 2 min. Then the mixture was subjected to 10 freeze–thaw cycles and extruded through a 0.45-μm polycarbonate membrane for at least five times. Then the suspension (10 mM LUVs) can be used without further treatment.

In each run, 100 μL lipids stock (300 mM KF, 10 mM HEPES, pH 7.2, 10 mM LUVs) was added to 1.99 mL external solution (300 mM Glc-K, 10 mM HEPES, pH 7.2, 10 mM LUVs). DMSO (40 μL) solutions of transporters or DMSO (40 μL) was added at time 0. The light scattering intensity at 600 nm was monitored by a fluorimeter (λ_ex = 600 nm, λ_em = 600 nm). In this section, none of the three compounds show any response in osmotic assay, so valinomycin (Vln, 0.5 μM, 0.1 mol%) was added to accelerate the outflow of K⁺. If the outflow of F⁻ accompanied the outflow of K⁺, an osmotic gradient will be formed, causing vesicle shrinkage and increasing the amount of 90° light scattering. At 300 s, FCCP (25 μM, 5 mol%) and Vln (0.5 μM, 0.1 mol%) (if not added previously) were added to accelerate the transport process to the end for calibration. The fractional light scattering intensity (I_f) was calculated based on the following equation:

In the equation, R_t is the light scattering intensity at time t, R_f is the final light scattering intensity obtained by the addition of FCCP and Vln, and R₀ is the light scattering intensity at time 0. Different concentrations of the transporter molecule were added to obtain a series of fractional light scattering intensity (I_f). Fitting I_f vs. molecule concentration was performed using the following equation:

In the equation, y is the value of I_f corresponding to the carrier molecule loaded at concentration x, y₀ is the I_f value measured at no compound added, y_max is the maximum I_f value, n is the Hill coefficient, and K is the EC₅₀ value.

The large unilamellar vesicles (LUVs, mean diameter 200 nm) of POPC were prepared by freeze/thaw cycles; 50 mg POPC dissolved in CHCl₃ (50 mL) was slowly evaporated using a rotary evaporator and dried under high vacuum at room temperature for 4 h. The obtained lipid film was rehydrated by mixing with a 5 mL saline solution (300 mM KF, 10 mM HEPES, pH 7.2) and vortexing for 2 min. Then the mixture was subjected to 10 freeze–thaw cycles and extruded through a 0.22-μm polycarbonate membrane for at least five times. The obtained lipid suspension (3.8 mL, 0.05 mmol) was transferred to size exclusion chromatography (stationary phase Sephadex G-50; mobile phase 300 mM Glc-K, 10 mM HEPES, pH 7.2) and diluted with the mobile phase to acquire 10 mL of 5 mM lipids stock solution.

In the F⁻ transport study, 1 mL of the lipid stock solution was added to 4 mL Glc-K (300 mM buffered to pH 7.2 using 10 mM HEPES) to generate a solution with a concentration gradient. As the former osmotic assay demonstrates that the transporter may work in an electrogenic way, Vln was added before the measurement. Then DMSO (40 μL) solutions of transporters or DMSO (40 μL) was added at time 0. The fluoride efflux was monitored by a fluoride selective electrode. At 300 s, Triton X-100 [100 μL, 20% (v/v)] was added as a detergent to lysate the LUVs for calibration.

DPPC LUVs (mean diameter 400 nm) were prepared as follows. A chloroform solution (30.4 mg DPPC) was evaporated slowly by a rotary evaporator and dried under high vacuum at room temperature for 4 h. After drying, the lipid film was rehydrated by vortexing with a 4 mL saline solution (300 mM KF, 10 mM HEPES, pH 7.2) and vortexing for 2 min at 55°C. The mixture was subjected to 10 freeze–thaw cycles (water bath was maintained at 55°C) and extruded through a 0.45-μm polycarbonate membrane for at least five times at 55°C. Then the suspension (10 mM LUVs) can be used without further treatment.

The transport of F⁻ was performed by the same way as above-mentioned. The lipid solution was stirred and thermostated in a polystyrene cuvette at 37 or 43°C.

POPC LUVs (mean diameter 200 nm) were prepared as follows. A chloroform solution (10 mg POPC) was evaporated slowly by a rotary evaporator and dried under high vacuum at room temperature for 4 h. After drying, the lipid film was rehydrated by vortexing with a 1 mL calcein solution (100 mM NaCl, 10 mM HEPES, 100 mM calcein, pH 7.0) and vortexing for 2 min. The mixture was subjected to 10 freeze–thaw cycles and extruded through a 0.22-μm polycarbonate membrane for at least five times. The unentrapped calcein was separated by size exclusion chromatography (stationary phase Sephadex G-50; mobile phase 100 mM NaCl, 10 mM HEPES, pH 7.0) and diluted with the mobile phase to acquire 5 mL of 2 mM lipid stock solution.

For each run, 0.1 mL of the lipid stock solution was added to a 1.88 mL buffer solution to a final lipid concentration of 100 μM. DMSO (20 μL) solutions of transporters or DMSO (20 μL) was added at time 0. The fluorescence emission at time t (λ_ex = 495 nm, λ_em = 515 nm) was recorded over 30 min. At 30 min, Triton X-100 [10 μL, 20% (v/v)] was added as a detergent to lysate the LUVs for calibration. The fractional fluorescence intensity (I_f) was calculated based on the following equation:

In the equation, R_t is the fluorescence intensity at time t, R_f is the final fluorescence intensity obtained by the addition of detergent, and R₀ is the fluorescence intensity at time 0.