Single-atom cobalt-hydroxyl modification of polymeric carbon nitride for highly enhanced photocatalytic water oxidation: ball milling increased single atom loading

Expediting the oxygen evolution reaction (OER) is the key to achieving efficient photocatalytic overall water splitting. Herein, single-atom Co–OH modified polymeric carbon nitride (Co-PCN) was synthesized with single-atom loading increased by ∼37 times with the assistance of ball milling that formed ultrathin nanosheets. The single-atom Co-N₄OH structure was confirmed experimentally and theoretically and was verified to enhance optical absorption and charge separation and work as the active site for the OER. Co-PCN exhibits the highest OER rate of 37.3 μmol h⁻¹ under visible light irradiation, ∼28-fold higher than that of common PCN/CoO_x, with the highest apparent quantum yields reaching 4.69, 2.06, and 0.46% at 400, 420, and 500 nm, respectively, and is among the best OER photocatalysts reported so far. This work provides an effective way to synthesize efficient OER photocatalysts.

Single-atom Co^II-OH modified polymeric carbon nitride synthesized with increased single-atom loading under the assistance of ball milling exhibits high photocatalytic water oxidation activity with Co-N₄OH as the highly active site.

Massive fuel energy consumption induced environmental and ecological problems, especially the greenhouse effect, and the resultant extreme climates and rise in sea level are threatening human life.¹ As a potential substitution for fuel energy, hydrogen energy conversion from solar energy via photocatalytic water splitting attracts great attention from scientists.^2–5 However, the photocatalytic hydrogen evolution efficiency from overall water splitting is still restricted by the sluggish oxygen evolution reaction (OER) that involves energy absorption, four-electron transfer, breakage of O–H bonds, and formation of O–O bonds,^6,7 and thus efficient OER photocatalysts become the key to achieving efficient overall water splitting. Though numerous hydrogen evolution photocatalysts have been reported, research on OER photocatalysts is mainly around a few semiconductors including BiVO₄, WO₃, Ag₃PO₄, α-Fe₂O₃, etc.8–11 and their activity is not high enough yet for practical applications. Therefore, exploring high-efficiency OER photocatalysts is still necessary.Polymeric carbon nitride (PCN) was first reported in 2009 (ref. 12) as a photocatalyst with a layered melon-type orthorhombic structure,¹³ and thereafter quickly became a “star” photocatalyst thanks to its advantages of being visible-light responsive and metal-free, non-toxic, and low cost, and its relatively high chemical stability.¹⁴ Because of several self-deficiencies including fast photogenerated charge recombination and a narrow optical absorption spectrum, PCN exhibits relatively low photocatalytic activity.¹⁵ Then, a series of strategies were put forward successively to enhance the photoactivity of PCN, such as enhancement of crystallinity,¹⁶ morphological control,¹⁷ structural modification¹⁸ (including extensively researched single atom modification in recent years^19,20), exfoliation,²¹ construction of hetero-(homo-)junctions,²² and loading of noble metals.²³ Though photocatalytic water splitting on PCN was extensively researched in the past, the research was mainly around the hydrogen evolution half-reaction used for exploring properties and the catalytic mechanism of photocatalysts, and little research was focused on the industrially useable overall water splitting process owing to the sluggish OER.¹⁵ Therefore, enhancing the photocatalytic OER activity of PCN becomes the key to practical applications.To increase OER rates of PCN, several kinds of methods were proposed, such as rational design of compound cocatalysts (e.g., CoO_x, IrO₂, CoP, CoPi, RhO_x, RuO_x, PtO_x, MnO_x, Co(OH)₂, Ni(OH)₂, and CoAl₂O₄ (ref. 24–30)), modification of carbon dots and carbon rings,^31,32 fabrication of special architectures of PCN (e.g., PCN quantum dot stacked nanowires³³), and single-atom (e.g., B, Co, and Mn^34–36) modification. For instance, Zhao and coauthors prepared B and N-vacancy comodified PCN that exhibits the highest OER rate of ∼28 μmol h⁻¹ (ref. 36) and recently their group further used these B doped PCN ultrathin nanosheets to fabricate a Z-scheme heterojunction for overall water splitting with a solar-to-hydrogen efficiency reaching ∼1.2%.³⁷ Comparatively, PCN loaded with compound cocatalysts can only enhance OER activity to a limited degree and there are finite methods for carbon modification and special architecture fabrication. Single-atom modification shows a bright prospect, on account of metal atoms capable of being inserted into the framework of PCN and effectively increasing the OER activity. However, reported single metal atom modification routes are all based on direct ion adsorption on PCN or calcination of mixtures of metal salts and PCN feedstocks.^34,35,38 New routes need be explored to increase effective loading of single atoms in PCN. Besides, the metal-OH structure is considered efficient for the OER,^30,39,40 and a single metal atom-OH structure has never been reported for modification of PCN, though Mn–OH was thought to play a key role in the OER process.³⁴Ball milling is an extensively used versatile and scalable way for preparation of heterogeneous catalysts and even single-atom catalysts,^41,42 but was rarely used in synthesis of PCN-based single-atom photocatalysts. In this work, we synthesized single-atom Co–OH modified PCN (Co-PCN) with the single-atom content in PCN highly increased with the assistance of ball milling. The simple synthetic route is shown in Fig. 1a. PCN was ball-milled to obtain BM-PCN that then adsorbed Co²⁺ till saturation to form BM-PCN/Co which was calcined to obtain BM-PCN/Co-c (Co-PCN). For comparison, PCN was directly used to adsorb Co²⁺ till saturation to form PCN/Co which was calcined to obtain PCN/Co-c. PCN mainly comprises large blocks with the size of several micrometers (Fig. S1^†), while BM-PCN contains massive irregular particles with the size reduced to several hundreds of nanometers (Fig. S2^†), indicative of high efficacy of ball milling. BM-PCN/Co-c exhibits a similar morphology as BM-PCN (Fig. 1b and S3^†) and PCN/Co-c exhibits a similar morphology to PCN (Fig. S4^†), but the surface area and mesopore volume of BM-PCN and BM-PCN/Co-c are not higher than those of PCN and PCN/Co-c (Fig. S5^†), manifesting that ball-milling and subsequent calcination did not form massive mesopores, which accords well with the particle size variation from several micrometers (before ball milling) to several hundreds of nanometers (after ball milling). However, the Co content in BM-PCN/Co-c, BM-PCN/Co, PCN/Co-c, and PCN/Co was measured to be 0.75, 0.50, 0.02, and ∼0.02 wt%, respectively, by inductively coupled plasma mass spectrometry (ICP-MS). The ∼37 times higher Co content in BM-PCN/Co-c than in PCN/Co-c suggests the ball-milling enhanced adsorption of Co²⁺ on surfaces of BM-PCN, which should arise mainly from the ball-milling induced increase of surface energy and adsorption sites.⁴³Open in a separate windowFig. 1(a) Schematic illustration for synthesis of single-atom Co^II-OH modified PCN (BM-PCN/Co-c); and (b) SEM, (inset in b) TEM, (c) AFM, (d) EDS elemental mapping, and (e) HAADF-STEM images of BM-PCN/Co-c.The TEM image shows the existence of small and ultrathin nanosheets in BM-PCN/Co-c (inset in Fig. 1b) which can also be observed in the atomic force microscopy (AFM) image with a thickness of ∼7–10 nm and lateral size of <70 nm (Fig. 1c), and formation of these ultrathin nanosheets results from the ball milling of PCN.⁴⁴ It should be noted that most formed ultrathin nanosheets with high surface energy may stack into compact particles upon ball milling, leading to no increase of the total surface area. Energy dispersive X-ray spectroscopy (EDS) elemental mapping images of BM-PCN/Co-c indicate homogeneous distribution of C, N, O, and Co elements in the sample (Fig. 1d). The high-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) image of BM-PCN/Co-c shows massive white spots (marked by circles) with a mean size of <1 Å dispersed in the sample (Fig. 1e and S6^†), which should correspond to single-atom Co.To further verify the single-atom Co structure in BM-PCN/Co-c, Co K-edge X-ray absorption near-edge structure spectroscopy (XANES) and extended X-ray absorption fine structure (EXAFS) analysis were performed. As shown in Fig. 2a, the absorption-edge position of BM-PCN/Co-c is quite close to that of CoO and their peak positions are similar and far from those of other reference samples, indicating that the valence of Co in BM-PCN/Co-c is about +2. The bonding structure around Co was determined by Fourier transformed (FT) k³-weighted EXAFS analysis. As shown in Fig. 2b, a distinct single Co-ligand peak at ∼1.6 Å for BM-PCN/Co-c is observed, which prominently differs from the Co–Co coordination peak at ∼2.2 Å for Co foil and the Co^II–O coordination peak at ∼1.7 Å for CoO. The wavelet transform (WT) contour plot of BM-PCN/Co-c shows only one intensity maximum (Fig. S7^†), and the Cl 2p core-level XPS spectrum of BM-PCN/Co-c reveals no residue of Cl (Fig. S8^†). These further indicate the single-atom dispersion of Co species. Apparently, the Co-ligand peak is almost consistent with the Co^II–N peak for Co porphyrin, suggesting that the single-atom Co in BM-PCN/Co-c mainly coordinates with N. Least-square EXAFS curve fitting was performed to confirm quantitative structural parameters of Co^II in BM-PCN/Co-c, as shown in Fig. 2c, S9, and S10 and Table S1.^† Simple Co–N single-shell fitting of BM-PCN/Co-c (Fig. S10^†) gave a coordination number of 5.6 ± 0.4 (Table S1^†), that is, Co^II coordinates with five atoms. Considering that the PCN monolayer provides four appropriate N coordination sites at most,⁴⁵ Co^II likely coordinates with four N atoms and one OH atom. Thus, we further performed Co–N₄/Co–O double-shell fitting (Fig. 2c) and the obtained R-factor (0.0011) remarkably reduces relative to that from Co–N single-shell fitting (0.0035), indicative of rationality of the proposed Co^II–N₄OH structure. Confirmed Co–N and Co–O bond lengths are 2.04 and 2.15 Å, respectively (Table S1^†).Open in a separate windowFig. 2(a) Co K-edge XANES and (b) EXAFS spectra of Co foil, Co porphyrin (Copr), CoO, Co₃O₄, Co₂O₃, and BM-PCN/Co-c; EXAFS (c) R space-fitting and (inset in c) K space-fitting curves of BM-PCN/Co-c; (d) optimized structure of PCN and Co-doped PCN with different doping configurations and calculated formation energies (e) of Co doped PCN; and (e) Co 2p and (f) O 1s core-level XPS spectra of samples.To further confirm rationality of the Co–N₄OH coordination structure, density functional theory (DFT) calculations were conducted. As shown in Fig. 2d, three possible Co^II coordination structures in the PCN monolayer were explored. The Co–N₄OH structure without removal of H from PCN exhibits a much lower formation energy (∼0.15 eV) than Co–N₄ and Co–N₃ structures with removal of two H atoms from PCN (∼2.51 and 3.55 eV), demonstrating a high probability of existence of the Co–N₄OH structure in BM-PCN/Co-c. This structure can also be evidenced by X-ray photoelectron spectroscopy (XPS). As shown in Fig. 2e, the Co 2p core-level XPS spectrum of BM-PCN/Co-c shows two distinct peaks at binding energies of 796.8 and 781.4 eV beside satellite peaks, corresponding to Co 2p_1/2 and 2p_3/2 of Co^II ions.⁴⁶ The spectrum of BM-PCN/Co also shows two Co 2p peaks but at a binding energy ∼1.1 eV higher, suggesting variation of the Co coordination structure from BM-PCN/Co to BM-PCN/Co-c. PCN/Co-c exhibits no peaks because of its low Co content. Fig. 2f shows O 1s core-level spectra of PCN, BM-PCN, BM-PCN/Co-c, and PCN/Co-c. All the samples exhibit one peak at a binding energy of ∼532.0 eV, ascribed to surface hydroxyl species,⁴⁷ but an additional peak could be obtained for BM-PCN or BM-PCN/Co-c after deconvolution. The peak at a binding energy of ∼531.3 eV for BM-PCN should be assigned to adsorbed H₂O at new active adsorption sites generated by ball milling. This peak can also be observed in the spectrum of BM-PCN/Co, but with a ∼0.1 eV shift to a higher binding energy (Fig. S11^†) owing to the influence of adsorbed Co^II ions. The peak at ∼531.2 eV for BM-PCN/Co-c should be assigned to Co–OH,⁴⁸ given that there is only one O 1s peak for BM-PCN-c (synthesized by direct calcination of BM-PCN) (Fig. S11^†). The calculated Co/O(–Co) molar ratio, based on the XPS data, is ∼1.07 (Table S2^†), close to 1, consistent with the Co–N₄OH coordination structure.In C 1s and N 1s core-level XPS spectra, BM-PCN, BM-PCN/Co-c, PCN/Co-c, and BM-PCN/Co exhibit similar peaks to PCN (Fig. S12a–d^†), indicative of their similar framework structure which can also be evidenced by their similar N/C molar ratios, 1.53 (Table S3^†), but the N–H peak of BM-PCN shifts ∼0.2 eV to a lower binding energy relative to that of PCN, likely arising from the ball-milling induced destruction of intralayer hydrogen bonds (Fig. S13^†). The Co content in BM-PCN/Co, BM-PCN/Co-c, and PCN/Co-c is too low to cause detectable variation of C 1s and N 1s peaks. Similar FT-IR absorption bands of the samples (Fig. S14a and b^†) also indicate their basic frame structure, but in enlarged spectra (Fig. S14c^†), ν(C–N) and ν(C N) absorption bands of BM-PCN shift 16 cm⁻¹ to a higher wavenumber and 19 cm⁻¹ to a lower wavenumber, respectively, relative to those of PCN at 1242 and 1640 cm⁻¹,⁴⁹ likely resulting from the ball-milling induced hydrogen bond destruction, and the shift of these two absorption bands turns smaller for BM-PCN/Co-c, suggesting calcination-induced reforming of the destroyed hydrogen bonds, which is consistent with the XPS results (Fig. S12c^†). Besides, BM-PCN exhibits a wider and relatively stronger ν(N–H)/ν(O–H) absorption band than PCN (Fig. S14a^†), probably owing to the hydrogen bond destruction and new adsorbed H₂O, while this absorption band for BM-PCN/Co-c becomes much weaker, suggesting hydrogen bond reforming and loss of new adsorbed H₂O (Fig. 2f). Zeta potentials of the samples dispersed in water reflect variation of surface adsorbed hydroxyl species. As shown in Fig. S15a,^† all the samples exhibit negative zeta potentials because of dissociation of surface hydroxyl species. The zeta potentials, following the order PCN (−24 mV) > BM-PCN (−41 mV) < BM-PCN/Co-c (−30 mV) ≈ PCN/Co-c (−28 mV), suggest the ball milling-induced increase of surface hydroxyls in BM-PCN and calcination-induced decrease in BM-PCN/Co-c, consistent with the FT-IR results.Solid-state ¹³C magic-angle-spinning nuclear magnetic resonance (NMR) spectra of PCN, BM-PCN, BM-PCN/Co-c, and PCN/Co-c show two similar peaks at chemical shifts of ∼164 and 156 ppm (Fig. S15b^†), ascribed to C−NH_x and N C–N, respectively,⁵⁰ indicating their similar molecular framework, but in enlarged spectra, BM-PCN exhibits ∼0.3° movement of the N C–N peak to a lower chemical shift compared with PCN, because of the ball-milling induced hydrogen bond destruction, and the C−NH_x peak of BM-PCN/Co-c moves ∼0.2° to a lower chemical shift, likely owing to formation of the C–N–Co structure whose peak lies close to the C−NH_x peak.⁵¹ The XRD patterns of the samples are shown in Fig. S15c.^† PCN and PCN/Co-c exhibit typical diffraction peaks of melon-type carbon nitride with a layered orthorhombic structure and peaks at 13.1° and 27.6° correspond to (210) and (002) facets, respectively,^13,52 but BM-PCN reveals remarkably decreased peak intensity and ∼0.2° shift of the (002) peak to a lower 2θ (indicative of the increased interlayer distance) relative to PCN, demonstrating the ball-milling induced hydrogen bond destruction and substantial decrease of crystallinity. The remarkable decrease of crystallinity and almost no change of the surface area of BM-PCN, compared with those of PCN, further suggest that ball milling may form massive thin nanosheets (Fig. 1c) most of which stack into compact particles (Fig. 1b) owing to their high surface energy. In comparison with BM-PCN, BM-PCN/Co-c exhibits a narrower (002) peak, suggesting enhanced crystallinity owing to the calcination-induced hydrogen bond reforming, consistent with the FT-IR results. On the whole, it is likely the ball-milling induced destruction of hydrogen bonds that contributes largely to the increase of surface energy and new active adsorption centers and thus Co²⁺ adsorption on BM-PCN.Optical absorption capability of samples was investigated by UV-vis diffuse reflectance spectroscopy (DRS). As shown in Fig. 3a, BM-PCN/Co-c, BM-PCN, and PCN/Co-c exhibit considerably higher, lower, and similar optical absorption than/to PCN, respectively. For BM-PCN/Co-c, the optical absorption enhancement at a wavelength of <400 nm may benefit from the electron-rich Co that enhances π–π* transitions in heptazine rings,⁵³ and the Urbach tail absorption should arise from the Co–OH doping.^54,55 Bandgaps (E_g) of PCN, BM-PCN, BM-PCN/Co-c, and PCN/Co-c were roughly confirmed as 2.70, 2.81, 2.56, and 2.73 eV, respectively, via the formula E_g/eV = 1240/(λ_ed/nm)⁵⁶ where λ_ed is the absorption edge determined by solid lines in the spectra. The wider E_g of BM-PCN probably results from the quantum size effect of massive ultrathin crystal nanosheets (Fig. 1c) formed by ball milling, and the narrower E_g of BM-PCN/Co-c arises from the Co–OH doping that was then verified by DFT calculations. As shown in Fig. S16,^† the calculated E_g of BM-PCN/Co-c, ∼1.90 eV, is much smaller than that of PCN (2.57 eV), in accordance with the experimental results. For PCN, the conduction band (CB) is contributed by C 2p and N 2p orbitals and the valence band (VB) mainly by N 2p orbitals, while for BM-PCN/Co-c, the CB is contributed by Co 3d, C 2p, and N 2p orbitals and the VB mainly by Co 3d and N 2p orbitals (Fig. S16c and d^†), effectively manifesting that the narrowing of E_g of BM-PCN/Co-c results from the Co–OH doping. In addition, there are prominent doping levels (E_d) in the bandgap of BM-PCN/Co-c, mainly contributed by Co 3d and O 2p orbitals (Fig. S16d^†), effectively proving the Co–OH doping effect in BM-PCN/Co-c. Similar calculation results have been reported for Pt–OH modified carbon nitride.⁵⁷ Given that the experimental Co content (0.75 wt%) is much lower than the theoretical (6.71 wt%), practical doping levels in the bandgap may approach more to the VB. CB edges of the samples (E_CB) could be roughly determined by using Mott-Schottky plots (Fig. S17^†) and their Fermi levels (E_f) were subsequently confirmed based on VB-XPS spectra (Fig. S18^†). Energy band levels of the samples are shown in Fig. 3b, and it seems that ball milling causes a slight downshift of the VB edge (E_VB) of BM-PCN, favorable for photocatalytic water splitting, but the Co–OH doping causes a slight downshift of E_CB and upshift of E_VB of BM-PCN/Co-c. It is noteworthy that the E_d close to the VB edge (E_VB) can capture photogenerated holes⁵⁸ and thus the single-atom Co–OH works as the active site for the OER (Fig. 3b).Open in a separate windowFig. 3(a) UV-vis diffuse reflectance spectra of PCN, BM-PCN, BM-PCN/Co-c, and PCN/Co-c; (b) energy band levels of the samples and schematic illustration for water oxidation on BM-PCN/Co-c; (c) photoluminescence spectra, (d) time-resolved fluorescence spectra, and (e) anodic photocurrent (J_a) response of the samples; and (f) EPR spectra of the samples in the dark and under visible light irradiation. Data in (d) are the results of fitting decay curves to a tri-exponential model. Dark J_a in (e) was set as zero for distinct comparison.Spectroscopy and photoelectrochemical tests were conducted to evaluate photogenerated charge separation and transfer performance. As shown in Fig. 3c, photoluminescence (PL) spectra of all the samples show one emission peak, basically corresponding to their bandgap emission. BM-PCN exhibits weaker PL intensity than PCN, revealing a decreased photogenerated charge recombination efficiency, which originates from faster charge transfer from the inside to the surface of ultrathin nanosheets (Fig. S19^†) and trapped by surface states.⁵⁹ BM-PCN/Co-c exhibits the lowest PL intensity and the PL intensity of PCN/Co-c is lower than that of PCN, which arises from the E_d capturing photogenerated holes to reduce their direct recombination with electrons beside the ultrathin nanosheet effect in BM-PCN/Co-c. Fig. 3d shows time-resolved fluorescence spectra of the samples. Decay curves were well fitted to a tri-exponential model (S3) and the obtained results are shown in Fig. 3d. Three lifetimes (τ₁–τ₃) and their mean lifetime (τ_m, 89.2 ns) of BM-PCN are all much longer than those of PCN (τ_m = 17.9 ns), further suggesting the faster charge transfer from the inside to the surface of ultrathin nanosheets in BM-PCN, decreasing the direct charge recombination efficiency, but with subsequent surface radiative recombination.⁶⁰ Interestingly, the τ₁–τ₃ and τ_m (10.8 ns) of BM-PCN/Co-c are much shorter than those of PCN, which should result from faster transfer of holes to E_d that effectively decreases the charge recombination efficiency, with subsequent nonradiative energy transformation.⁶¹ The Co–OH doping effect also makes PCN/Co-c exhibit shorter τ₁–τ₃ and τ_m (16.5 ns) than PCN. Fig. 3e shows the photocurrent response of the samples. Their anodic photocurrent density follows the order PCN < PCN/Co-c < BM-PCN < BM-PCN/Co-c, indicating gradually increased photogenerated charge separation efficiencies,⁶² basically consistent with the PL results. The relatively high photocurrent response of BM-PCN benefits from the applied bias that effectively inhibits surface recombination of photogenerated charge carriers.To assess charge mobility of the samples, their electrochemical impedance spectroscopy (EIS) spectra were tested with high-frequency data simply fitted to an equivalent circuit (Fig. S20^†). The obtained charge transfer resistance (R_ct) follows the order PCN (26 Ω) > BM-PCN (18 Ω) ≈ PCN/Co-c (19 Ω) > BM-PCN/Co-c (13 Ω). Apparently, BM-PCN/Co-c exhibits smaller R_ct than BM-PCN and PCN/Co-c, and PCN/Co-c exhibits smaller R_ct than PCN, indicating the highest charge transfer performance of BM-PCN/Co-c⁶³ which originates from the single-atom Co modification⁶⁴ that may increase the electron density to facilitate charge transport. The smaller R_ct of BM-PCN than that of PCN indicates the additional favorable effect of ultrathin nanosheets.⁶⁵Fig. 3f shows electron paramagnetic resonance (EPR) spectra of the samples. All reveal one single Lorentzian line centered at a g of 2.0039, attributed to unpaired electrons in heptazine rings.⁶⁶ In the dark, the EPR signal intensity follows the order PCN < BM-PCN < PCN/Co-c < BM-PCN/Co-c, and the stronger signal of BM-PCN than that of PCN results from formation of ultrathin nanosheets that enhances delocalization of unpaired electrons, while the stronger signal of BM-PCN/Co-c and PCN/Co-c mainly benefits from the Co doping that increases the delocalized electron density.⁶⁷ Under visible light irradiation, the samples exhibit remarkable signal enhancement, following the sequence PCN < BM-PCN < PCN/Co-c < BM-PCN/Co-c, similar to that of the signal intensity in the dark, suggesting that the increase in the delocalized electron density facilitates charge photoexcitation. The high delocalized electron density favors charge transport, consistent with the EIS results, and the high photoexcited charge density benefits enhancement of photocatalytic activity.Photocatalytic OER activity of various samples was well evaluated using Ag⁺ as the sacrificial agent (Fig. S21^†). The Co content in BM-PCN/Co-c was optimized according to the photocatalytic OER rates and BM-PCN-c exhibits no detectable OER activity (Fig. S22^†), indicating indispensability of the Co–OH structure for the OER. The influence of the calcination temperature (T_c °C) of BM-PCN/Co on OER rates of BM-PCN/Co-c (T_c = 460) and BM-PCN/Co-cT_c was investigated and BM-PCN/Co-c exhibits the highest photoactivity (Fig. 4a), manifesting that the optimal calcination temperature is 460 °C. Under both simulated solar light and visible light irradiation (λ ≥ 420 nm), BM-PCN/Co-c exhibits substantially higher OER activity than PCN/Co-c (Fig. 4b), further suggesting the significance of the single-atom Co loading amount, and remarkably higher activity than common PCN/CoO_x (with 0.75 wt% Co, obtained via photodeposition) and BM-PCN-c/Co(OH)₂ (with 0.75 wt% Co), demonstrating the high efficacy of the single-atom distribution of Co–OH in BM-PCN/Co-c. Besides, urea was used as the feedstock to synthesize carbon nitride (marked as PCN-urea) with a larger surface area (76 m² g⁻¹ (ref. 68)) than PCN, and PCN-urea was further used to synthesize PCN-urea/Co-c similar to the synthesis of BM-PCN/Co-c. The OER activity of BM-PCN/Co-c is prominently higher than that of PCN-urea/Co-c (with the optimized Co content and Co single atom distribution, Fig. S23^†), suggesting the significant role of ball milling in fabricating the single-atom Co–N₄OH structure. To quantitively compare photoactivity of the samples, their mean OER rates under visible light illumination for 2 h are shown in Fig. 4c. The OER rate of BM-PCN/Co-c can reach ∼37.3 μmol h⁻¹, about 13.8, 28.7, 2.6, and 2.0 times those of PCN/Co-c, PCN/CoO_x, BM-PCN-c/Co(OH)₂, and PCN-urea/Co-c, respectively. Comparatively, less N₂ was generated for BM-PCN/Co-c (Fig. S24^†), further demonstrating the significance of single-atom Co–OH modification.Open in a separate windowFig. 4(a) The influence of the calcination temperature (T_c °C) of BM-PCN/Co on photocatalytic OER activity of BM-PCN/Co-c (T_c = 460) and BM-PCN/Co-cT_c, under Xe-lamp illumination, with AgNo₃ as the sacrificial agent; (b) photocatalytic oxygen evolution on various samples under Xe-lamp illumination with or without using a 420-nm filter; (c) corresponding OER rates of the samples in 2 h; (d) photocatalytic OER rates of BM-PCN/Co-c under irradiation with various monochromatic light sources for 12 h; (e) apparent quantum yields (AQYs) of BM-PCN/Co-c at different wavelengths and reaction times and the highest AQY at every wavelength, along with the UV-DRS spectrum; and (f) proposed mechanism for photocatalytic water oxidation on the single-atom Co^II-OH structure.Photocatalytic oxygen evolution on BM-PCN/Co-c was also tested under monochromatic light irradiation (Fig. S25^†). Apparently, BM-PCN/Co-c can exhibit OER activity even at a wavelength of 500 nm. The mean OER rate in 12 h decreases from 1.85 to 0.54 μmol h⁻¹ with increasing wavelengths from 400 to 500 nm (Fig. 4d), independent of light intensity of the Xe lamp and is mainly dependent on optical absorption capability of BM-PCN/Co-c at various wavelengths (Fig. 3a). Fig. 4e shows apparent quantum yields (AQYs) of BM-PCN/Co-c at different reaction times and wavelengths. Basically, there are maxima of AQYs with increasing reaction time at every wavelength, suggesting the adverse effect of excessive photodeposited Ag on surfaces of samples. These maxima are shown in Fig. 4e and accord well with the UV-vis DRS spectrum with increasing wavelengths. The maxima of AQYs at 400, 420, 450, and 500 nm can reach 4.69, 2.06, 1.07, and 0.46%, respectively. Compared with the reported photocatalytic OER results for PCN (Table S4^†), BM-PCN/Co-c exhibits the top-class performance.To investigate chemical stability of BM-PCN/Co-c, the cyclic OER experiment was conducted. After five consecutive runs, OER rates of BM-PCN/Co-c decrease less (Fig. S26a^†), with the morphology similar to the original (Fig. S26b^†). Co single atoms in the sample could still be distinctly observed by HAADF-STEM (Fig. S26c and d^†). In addition, N 1s core-level XPS spectra of BM-PCN/Co-c are almost similar before and after the cyclic experiment (Fig. S26e^†). These indicate the high stability of the basic framework structure of the sample. However, Co 2p core-level spectra show remarkable differences before and after the experiment, not only the Co^II peak shift, probably owing to ion (e.g., IO₄⁻) adsorption, but also formation of a large amount of Co^III (Fig. S26f^†). Coexistence of Co^II/Co^III may suggest the photocatalytic OER mechanism.The proposed OER mechanism based on the Co–OH structure is shown in Fig. 4f, according to the reported results in Mn doped PCN.³⁴ Four holes are needed to complete four oxidation steps and obtain one O₂ molecule. The first step starting with one hole may involve formation of the Co^III O bond. The Co–N₄OH structure should facilitate the water oxidation more compared with that of Co–N₄ without OH coordination, by leaving out the initial adsorption process of H₂O molecules.³⁴ On the whole, the high photocatalytic OER activity of Co-PCN benefits from the Co–N₄OH structure that not only effectively enhances optical absorption, and charge separation and transport, but also works as the highly active site for the OER.     

A Paal–Knorr agent for chemoproteomic profiling of targets of isoketals in cells

Natural systems produce various γ-dicarbonyl-bearing compounds that can covalently modify lysine in protein targets via the classic Paal–Knorr reaction. Among them is a unique class of lipid-derived electrophiles – isoketals that exhibit high chemical reactivity and critical biological functions. However, their target selectivity and profiles in complex proteomes remain unknown. Here we report a Paal–Knorr agent, 4-oxonon-8-ynal (herein termed ONAyne), for surveying the reactivity and selectivity of the γ-dicarbonyl warhead in biological systems. Using an unbiased open-search strategy, we demonstrated the lysine specificity of ONAyne on a proteome-wide scale and characterized six probe-derived modifications, including the initial pyrrole adduct and its oxidative products (i.e., lactam and hydroxylactam adducts), an enlactam adduct from dehydration of hydroxylactam, and two chemotypes formed in the presence of endogenous formaldehyde (i.e., fulvene and aldehyde adducts). Furthermore, combined with quantitative chemoproteomics in a competitive format, ONAyne permitted global, in situ, and site-specific profiling of targeted lysine residues of two specific isomers of isoketals, levuglandin (LG) D2 and E2. The functional analyses reveal that LG-derived adduction drives inhibition of malate dehydrogenase MDH2 and exhibits a crosstalk with two epigenetic marks on histone H2B in macrophages. Our approach should be broadly useful for target profiling of bioactive γ-dicarbonyls in diverse biological contexts.

Natural systems produce various γ-dicarbonyl-bearing compounds that can covalently modify lysine in protein targets via the classic Paal–Knorr reaction.

Synthetic chemistry methods have been increasingly underscored by their potential to be repurposed as biocompatible methods for both chemical biology and drug discovery. The most-known examples of such a repurposing approach include the Staudinger ligation¹ and the Huisgen-based click chemistry.² Moreover, bioconjugation of cysteine and lysine can be built upon facile chemical processes,³ while chemoselective labelling of other polar residues (e.g., histidine,⁴ methionine,⁵ tyrosine,⁶ aspartic and glutamic acids^7,8) requires more elaborate chemistry, thereby offering a powerful means to study the structure and function of proteins, even at a proteome-wide scale.The classical Paal–Knorr reaction has been reported for a single-step pyrrole synthesis in 1884.^9,10 The reaction involves the condensation of γ-dicarbonyl with a primary amine under mild conditions (e.g., room temperature, mild acid) to give pyrrole through the intermediary hemiaminals followed by rapid dehydration of highly unstable pyrrolidine adducts (Fig. S1^†).Interestingly, we and others have recently demonstrated that the Paal–Knorr reaction can also readily take place in native biological systems.^11–13 More importantly, the Paal–Knorr precursor γ-dicarbonyl resides on many endogenous metabolites and bioactive natural products.¹⁴ Among them of particular interest are isoketals¹⁵ (IsoKs, also known as γ-ketoaldehydes) which are a unique class of lipid derived electrophiles (LDEs) formed from lipid peroxidation (Fig. S2^†)¹⁶ that has emerged as an important mechanism for cells to regulate redox signalling and inflammatory responses,¹⁷ and drive ferroptosis,¹⁸ and this field has exponentially grown over the past few years. It has been well documented that the γ-dicarbonyl group of IsoKs can rapidly and predominantly react with lysine via the Paal–Knorr reaction to form a pyrrole adduct in vitro (Fig. 1).¹⁵ Further, the pyrrole formed by IsoKs can be easily oxidized to yield lactam and hydroxylactam products in the presence of molecular oxygen (Fig. 1). These rapid reactions are essentially irreversible. Hence, IsoKs react with protein approximately two orders of magnitude faster than the most-studied LDE 4-hydoxynonenal (4-HNE) that contains α,β-unsaturated carbonyl to generally adduct protein cysteines by Michael addition (Fig. S3^†).¹⁵ Due to this unique adduction chemistry and rapid reactivity, IsoKs exhibit intriguing biological activities, including inhibition of the nucleosome complex formation,¹⁹ high-density lipoprotein function,²⁰ mitochondrial respiration and calcium homeostasis,²¹ as well as activation of hepatic stellate cells.²² Furthermore, increases in IsoK-protein adducts have been identified in many major diseases,²³ such as atherosclerosis, Alzheimer''s disease, hypertension and so on.Open in a separate windowFig. 1The Paal–Knorr precursor γ-dicarbonyl reacts with the lysine residue on proteins to form diverse chemotypes via two pathways. The red arrow shows the oxidation pathway, while the blue one shows the formaldehyde pathway.Despite the chemical uniqueness, biological significance, and pathophysiological relevance of IsoKs, their residue selectivity and target profiles in complex proteomes remain unknown, hampering the studies of their mechanisms of action (MoAs). Pioneered by the Cravatt group, the competitive ABPP (activity-based protein profiling) has been the method of choice to analyse the molecular interactions between electrophiles (e.g., LDEs,²⁴ oncometabolites,²⁵ natural products,^26,27 covalent ligands and drugs^28–30) and nucleophilic amino acids across complex proteomes. In this regard, many residue-specific chemistry methods and probes have been developed for such studies. For example, several lysine-specific probes based on the activated ester warheads (e.g., sulfotetrafluorophenyl, STP;³¹N-hydroxysuccinimide, NHS³²) have recently been developed to analyse electrophile–lysine interactions at a proteome-wide scale in human tumour cells, which provides rich resources of ligandable sites for covalent probes and potential therapeutics. Although these approaches can also be presumably leveraged to globally and site-specifically profile lysine-specific targets IsoKs, the reaction kinetics and target preference of activated ester-based probes likely differ from those of γ-dicarbonyls, possibly resulting in misinterpretation of ABPP competition results. Ideally, a lysine profiling probe used for a competitive ABPP analysis of IsoKs should therefore possess the same, or at least a similar, warhead moiety. Furthermore, due to the lack of reactive carbonyl groups on IsoK-derived protein adducts, several recently developed carbonyl-directed ligation probes for studying LDE-adductions are also not suitable for target profiling of IsoKs.^33,34Towards this end, we sought to design a “clickable” γ-dicarbonyl probe for profiling lysine residues and, in combination with the competitive ABPP strategy, for analysing IsoK adductions in native proteomes. Considering that the diversity of various regio- and stereo- IsoK isomers¹⁵ (a total of 64, Fig. S2^†) in chemical reactivity and bioactivities is likely attributed to the substitution of γ-dicarbonyls at positions 2 and 3, the “clickable” alkyne handle needs to be rationally implemented onto the 4-methyl group in order to minimize the biases when competing with IsoKs in target engagement. Interestingly, we reasoned that 4-oxonon-8-ynal, a previously reported Paal–Knorr agent used as an intermediate for synthesizing fatty acid probes³⁵ or oxa-tricyclic compounds,³⁶ could be repurposed for the γ-dicarbonyl-directed ABPP application. With this chemical in hand (herein termed ONAyne, Fig. 2A), we first used western blotting to detect its utility in labelling proteins, allowing visualization of a dose-dependent labelling of the proteome in situ (Fig. S4^†). Next, we set up to incorporate this probe into a well-established chemoproteomic workflow for site-specific lysine profiling in situ (Fig. 2A). Specifically, intact cells were labelled with ONAyne in situ (200 μM, 2 h, 37 °C, a condition showing little cytotoxicity, Fig. S5^†), and the probe-labelled proteome was harvested and processed into tryptic peptides. The resulting probe-labelled peptides were conjugated with both light and heavy azido-UV-cleavable-biotin reagents (1 : 1) via Cu^I-catalyzed azide–alkyne cycloaddition reaction (CuAAC, also known as click chemistry). The biotinylated peptides were enriched with streptavidin beads and photoreleased for LC-MS/MS-based proteomics. The ONAyne-labelled peptides covalently conjugated with light and heavy tags would yield an isotopic signature. We considered only those modified peptide assignments whose MS1 data reflected a light/heavy ratio close to 1.0, thereby increasing the accuracy of these peptide identifications. Using this criterium, we applied a targeted database search to profile three expected probe-derived modifications (PDMs), including 13 pyrrole peptide adducts (Δ273.15), 77 lactam peptide adducts (Δ289.14), and 557 hydroxylactam peptide adducts (Δ305.14), comprising 585 lysine residues on 299 proteins (Fig. S6 and S7^†). Among them, the hydroxylactam adducts were present predominately, since the pyrrole formed by this probe, the same as IsoKs, can be easily oxidized when being exposed to O₂. This finding was in accordance with a previous report where the pyrrole adducts formed by the reaction between IsoK and free lysine could not be detected, but rather their oxidized forms.³⁷ Regardless, all three types of adducts were found in one lysine site of EF1A1 (K387, Fig. S8^†), further confirming the intrinsic relationship among those adductions in situ.Open in a separate windowFig. 2Adduct profile and proteome-wide selectivity of the γ-dicarbonyl probe ONAyne. (A) Chemical structure of ONAyne and schematic workflow for identifying ONAyne-adducted sites across the proteome. (B) Bar chart showing the distribution of six types of ONAyne-derived modifications formed in situ and in vitro (note: before probe labelling, small molecules in cell lysates were filtered out through desalting columns).State-of-the-art blind search can offer an opportunity to explore unexpected chemotypes (i.e., modifications) derived from a chemical probe and to unbiasedly assess its proteome-wide residue selectivity.^38,39 We therefore sought to use one of such tools termed pChem³⁸ to re-analyse the MS data (see Methods, ESI^†). Surprisingly, the pChem search identified three new and abundant PDMs (Fig. 1 and Table S1^†), which dramatically expand the ONAyne-profiled lysinome (2305 sites versus 585 sites). Overall, these newly identified PDMs accounted for 74.6% of all identifications (Fig. 2B and Table S2^†). Among them, the PDM of Δ287.13 (Fig. 1 and S7^†) might be an enlactam product via dehydration of the probe-derived hydroxylactam adduct. The other two might be explained by the plausible mechanism as follows (Fig. 1). The endogenous formaldehyde (FA, produced in substantial quantities in biological systems) reacts with the probe-derived pyrrole adduct via nucleophilic addition to form a carbinol intermediate, followed by rapid dehydration to a fulvene (Δ285.15, Fig. S7^†) and immediate oxidation to an aldehyde (Δ301.14, Fig. S7^†). In line with this mechanism, the amount of FA-derived PDMs was largely eliminated when the in vitro ONAyne labelling was performed in the FA-less cell lysates (Fig. 2B and Table S3^†). Undoubtedly, the detailed mechanisms underlying the formation of these unexpected PDMs require further investigation, and so does the reaction kinetics. Regardless, all main PDMs from ONAyne predominantly target the lysine residue with an average localization probability of 0.77, demonstrating their proteome-wide selectivity (Fig. S9^†).Next, we adapted an ABPP approach to globally and site-specifically quantify the reactivity of lysine towards the γ-dicarbonyl warhead through a dose-dependent labelling strategy (Fig. 3A) that has been proved to be successful for other lysine-specific probes (e.g., STP alkyne).³¹ Specifically, MDA-MB-231 cell lysates were treated with low versus high concentrations of ONAyne (1 mM versus 0.1 mM) for 1 h. Probe-labelled proteomes were digested into tryptic peptides that were then conjugated to isotopically labelled biotin tags via CuAAC for enrichment, identification and quantification. In principle, hyperreactive lysine would saturate labelling at the low probe concentration, whereas less reactive ones would show concentration-dependent increases in labelling. For fair comparison, the STP alkyne-based lysine profiling data were generated by using the same chemoproteomic workflow. Although 77.5% (3207) ONAyne-adducted lysine sites can also be profiled by STP alkyne-based analysis, the former indeed has its distinct target-profile with 930 lysine sites newly identified (Fig. S10 and Table S4^†). Interestingly, sequence motif analysis with pLogo⁴⁰ revealed a significant difference in consensus motifs between ONAyne- and STP alkyne-targeting lysines (Fig. S11^†).Open in a separate windowFig. 3ONAyne-based quantitative reactivity profiling of proteomic lysines. (A) Schematic workflow for quantitative profiling of ONAyne–lysine reactions using the dose-dependent ABPP strategy (B) Box plots showing the distribution of R_10:1 values quantified in ONAyne- and STP alkyne-based ABPP analyses, respectively. Red lines showing the median values. ***p ≤ 0.001 two-tailed Student''s t-test. (C) Representative extracted ion chromatograms (XICs) showing changes in the EF1A1 peptide bearing K273 that is adducted as indicated, with the profiles for light and heavy-labelled peptides in blue and red, respectively.Moreover, we quantified the ratio (R_{1 mM:0.1 mM}) for a total of 2439 ONAyne-tagged lysines (on 922 proteins) and 17904 STP alkyne-tagged lysines (on 4447 proteins) across three biological replicates (Fig. S12 and Table S5^†). Strikingly, only 26.7% (651) of quantified sites exhibited nearly dose-dependent increases (R_{1 mM:0.1 mM} > 5.0) in reactivity with ONAyne, an indicative of dose saturation (Fig. 3B and C). In contrast, such dose-dependent labelling events accounted for >69.1% of all quantified lysine sites in the STP alkyne-based ABPP analysis.³¹ This finding is in accordance with the extremely fast kinetics of reaction between lysine and γ-dicarbonyls (prone to saturation). Nonetheless, by applying 10-fold lower probe concentrations, overall 1628 (80.2%) detected lysines could be labelled in a fully concentration-dependent manner with the median R_10:1 value of 8.1 (Fig. 3B, C, S12 and Table S5^†). Next, we asked whether the dose-depending quantitation data (100 μM versus 10 μM) can be harnessed to predict functionality. By retrieving the functional information for all quantified lysines from the UniProt Knowledgebase, we found that those hyper-reactive lysines could not be significantly over-represented with annotation (Fig. S12^†). Nonetheless, among all quantified lysines, 509 (25.1%) possess functional annotations, while merely 2.5% of the human lysinome can be annotated. Moreover, 381 (74.8%) ONAyne-labelled sites are known targets of various enzymatic post-translational modifications (PTMs), such as acetylation, succinylation, methylation and so on (Fig. S13^†). In contrast, all known PTM sites accounted for only 59.6% of the annotated human lysinome. These findings therefore highlight the intrinsic reactivity of ONAyne towards the ‘hot spots’ of endogenous lysine PTMs.The aforementioned results validate ONAyne as a fit-for-purpose lysine-specific chemoproteomic probe for competitive isoTOP-ABPP application of γ-dicarbonyl target profiling. Inspired by this, we next applied ONAyne-based chemoproteomics in an in situ competitive format (Fig. 4A) to globally profile lysine sites targeted by a mixture of levuglandin (LG) D₂ and E₂, two specific isomers of IsoKs that can be synthesized conveniently from prostaglandin H₂ (ref. 41) (Fig. S2^†). Specifically, mouse macrophage RAW264.7 cells (a well-established model cell line to study LDE-induced inflammatory effects) were treated with 2 μM LGs or vehicle (DMSO) for 2 h, followed by ONAyne labelling for an additional 2 h. The probe-labelled proteomes were processed as mentioned above. For each lysine detected in this analysis, we calculated a control/treatment ratio (R_C/T). Adduction of a lysine site by LGs would reduce its accessibility to the ONAyne probe, and thus a higher R_C/T indicates increased adduction. In total, we quantified 2000 lysine sites on 834 proteins across five biological replicates. Among them, 102 (5.1%) sites exhibited decreases of reactivity towards LGs treatment (P < 0.05, Table S6^†), thereby being considered as potential targets of LGs. Notably, we found that different lysines on the same proteins showed varying sensitivity towards LGs (e.g., LGs targeted K3 of thioredoxin but not K8, K85 and K94, Table S6^†), an indicative of changes in reactivity, though we could not formally exclude the effects of changes in protein expression on the quantified competition ratios. Regardless, to the best of our knowledge, the proteome-wide identification of potential protein targets by IsoKs/LGs has not been possible until this work.Open in a separate windowFig. 4ONAyne-based in situ competitive ABPP uncovers functional targets of LGs in macrophages. (A) Schematic workflow for profiling LGs–lysine interactions using ONAyne-based in situ competitive ABPP. (B) Volcano plot showing the log₂ values of the ratio between the control (heavy) and LGs-treated (light) channels and the −log₁₀(P) of the statistical significance in a two-sample t-test for all quantified lysines. Potential targets of LGs are shown in blue (R_C/T>1.2, P < 0.05), with the validated ones in red. (C) Bar chart showing the inhibitory effect of 2 μM LGs on the cellular enzymatic activity of MDH2. Data represent means ± standard deviation (n = 3). Statistical significance was calculated with two-tailed Student''s t-tests. (D) Pretreatment of LGs dose-dependently blocked ONAyne-labelling of MDH2 in RAW264.7 cells, as measured by western blotting-based ABPP. (E and F) LGs dose-dependently decreased the H2BK5 acetylation level in RAW 264.7 cells, as measured either by western blotting (E) or by immunofluorescence imaging (F). n = 3. For G, nuclei were visualized using DAPI (blue).We initially evaluated MDH2 (malate dehydrogenase, mitochondrial, also known as MDHM), an important metabolic enzyme that possesses four previously uncharacterized liganded lysine sites (K157, K239, K301 and K329, Fig. 4B) that are far from the active site (Fig. S14^†). We found that LGs dramatically reduced the catalytic activity of MDH2 in RAW264.7 cells (Fig. 4C), suggesting a potentially allosteric effect. We next turned our attention to the targeted sites residing on histone proteins, which happen to be modified by functionally important acetylation, including H2BK5ac (Fig. 4B) that can regulate both stemness and epithelial–mesenchymal transition of trophoblast stem cells.⁴² We therefore hypothesized that rapid adduction by LGs competes with the enzymatic formation of this epigenetic mark. Immunoblotting-based competitive ABPP confirmed that LGs dose-dependently blocked probe labelling of H2B (Fig. 4D). Further, both western blots and immunofluorescence assays revealed that LG treatment decreased the level of acetylation of H2BK5 (average R_C/T = 1.3, P = 0.007) in a concentration-dependent manner (Fig. 4E and F). Likewise, a similar competitive crosstalk was observed between acetylation and LG-adduction on H2BK20 (average R_C/T = 1.2, P = 0.01) that is required for chromatin assembly⁴³ and/or gene regulation⁴⁴ (Fig. 4B and S15^†). Notably, these findings, together with several previous reports by us and others about histone lysine ketoamide adduction by another important LDE, 4-oxo-2-noenal,^11,45,46 highlight again the potentially important link between lipid peroxidation and epigenetic regulation. In addition to the targets validated as above, many other leads also merit functional studies considering diverse biological or physiologic effects of LGs in macrophages.     

Supramolecular encapsulation of redox-active monomers to enable free-radical polymerisation

Extended polymeric structures based on redox-active species are of great interest in emerging technologies related to energy conversion and storage. However, redox-active monomers tend to inhibit radical polymerisation processes and hence, increase polydispersity and reduce the average molecular weight of the resultant polymers. Here, we demonstrate that styrenic viologens, which do not undergo radical polymerisation effectively on their own, can be readily copolymerised in the presence of cucurbit[n]uril (CB[n]) macrocycles. The presented strategy relies on pre-encapsulation of the viologen monomers within the molecular cavities of the CB[n] macrocycle. Upon polymerisation, the molecular weight of the resultant polymer was found to be an order of magnitude higher and the polydispersity reduced 5-fold. The mechanism responsible for this enhancement was unveiled through comprehensive spectroscopic and electrochemical studies. A combination of solubilisation/stabilisation of reduced viologen species as well as protection of the parent viologens against reduction gives rise to the higher molar masses and reduced polydispersities. The presented study highlights the potential of CB[n]-based host–guest chemistry to control both the redox behavior of monomers as well as the kinetics of their radical polymerisation, which will open up new opportunities across myriad fields.

Extended polymeric structures based on redox-active species are of great interest in emerging technologies related to energy conversion and storage.

Polyviologens are redox-active polymers based on N-substituted bipyridinium derivatives which have emerged as promising materials for energy conversion and storage.^1–5 Their physicochemical properties can be adjusted through copolymerisation of the redox-active viologen monomers.^6–8 The resultant materials are stable, water soluble and exhibit fast electron transfer kinetics. Polyviologens have been commonly fabricated through step-growth polymerisation in linear and dendritic architectures,^9–13 as supramolecular polymers,^14–16 networks,^6,17,18 and covalent organic frameworks.^19,20 Alternatively, anionic/cationic or metathesis-based polymerisations are used to avoid interference of radical-stabilising monomers with the radical initiators, however, these techniques are highly water- and/or oxygen-sensitive.^21,22 When free-radical polymerisation (FRP) is conducted in the presence of viologen species, its reduction can cause a depletion of active radicals and thus disruption of the polymerisation process. Despite varying solvents, comonomers and initiator loadings, the direct FRP of viologen-containing monomers remains therefore limited to molar masses of 30 kDa.^23–25 Accessing higher molar masses has been possible via post-polymerisation modification,^26–28 which has impacted the electrochemical properties of the resultant materials.^29,30 Alternative strategies to access higher molar masses of redox-active polymers and control their polymerisation are highly desirable.Incorporation of cucurbit[n]uril (CB[n]) macrocycles have lead to a variety of functional materials through host–guest chemistry.^31–34 Moreover, the redox chemistry of viologens can be modulated through complexation with CB[n].^35–38 Specifically, CB[n] (n = 7, 8) can tune the redox potential of pristine viologens and efficiently sequester monoreduced viologen radical cations, avoiding precipitation in aqueous environments. Further to this, we recently demonstrated that the viologen radical cation is stabilised by −20 kcal mol⁻¹ when encapsulated in CB[7].³⁹Consequently, we envisioned that incorporating CB[n]s as additives prior to polymerisation could (i) overcome current limits in accessible molar masses, (ii) increase control over FRP of viologen-based monomers through encapsulation and (iii) enable separation of radical species avoiding aggregation.Here, we demonstrate a new approach to control FRP of redox-active monomers leading to high molar masses and decreased dispersity of the resultant polymers. In absence of CB[n], co-polymerisation of the N-styryl-N′-phenyl viologen monomer 1²⁺ and N,N-dimethylacrylamide (DMAAm) only occurs at high initiator loadings (>0.5 mol%, Fig. 1a), leading to low molecular weights and high polydispersity. Using our synthetic approach, 1²⁺ is efficiently copolymerised with DMAAm in the presence of CB[n] (n = 7, 8) macrocycles resulting in control of the polymer molar mass across a broad range, 4–500 kDa (Fig. 1b). Finally, CB[n] are successfully removed from the polymer via competitive host–guest binding and dialysis. Spectroscopic and electrochemical studies revealed that solubilisation/stabilisation of the reduced species and/or shielding of the redox-active monomers from electron transfer processes was responsible for this enhancement.Open in a separate windowFig. 1Schematic representation of the investigated polymerisation. (a) Conventional free radical polymerisation either completely fails to copolymerise redox-active monomers (low initiator loading) or delivers copolymers with limited molar masses and high dispersities (high initiator loading). (b) CB[n]-mediated protection suppresses interference of viologen monomers with radicals formed through the initiation process facilitating copolymerisation. The molar mass of the resulting copolymers is readily tunable via the amount of present CB[n] macrocycles and the CB[n] is post-synthetically removed via competitive binding to yield the final copolymer with desired molar mass. Cl⁻ counter-ions are omitted for clarity.Recent studies on symmetric aryl viologens demonstrated 2 : 2 binding modes with CB[8] and high binding constants (up to K_a ∼ 10¹¹ M⁻²).^40,41 Incorporation of polymerisable vinyl moieties, in combination with the relatively static structure of their CB[n] host–guest complexes, was postulated to allow polymerisation without unfavorable side reactions. The asymmetric N-styryl-N′-phenyl viologen monomer 1²⁺ prepared for this study (Fig. S1a and S2–S13^†) displays a linear geometry and was predicted to bind CB[n] (n = 7, 8) in a 2 : 1 and 2 : 2 binding fashion (Fig. S1b^†).^40,42 Binding modes between CB[n] (n = 7, 8) and 1²⁺ were investigated through titration experiments (¹H NMR and ITC) which confirmed the formation of 1·(CB[7])₂ and (1)₂·(CB[8])₂ (see Fig. S25 and S26^†). ¹H NMR titration of CB[7] with 1²⁺ demonstrates encapsulation of both aryl moieties (including the vinyl group) through upfield chemical shifts of the respective signals (Fig. 2a). Similar upfield shifts were observed for CB[8] (Fig. 2c). Different para-aryl substituents (vinyl vs. hydrogen) resulted in either head-to-tail or head-to-head (1)₂·(CB[8])₂ dimers (Fig. S1b and S26^†), a previously reported phenomenon.⁴³ Nonetheless, the reversible nature of the complex renders the vinyl group temporarily available for copolymerisation. In the presence of CB[8], 1²⁺ yields polymer molar masses of up to 500 kDa as its complexation is more robust. ITC data confirmed binding stoichiometry, with binding constants of K_a = 2.64 × 10⁶ M⁻¹ for 1·(CB[7])₂ and K_a = 9.02 × 10¹⁰ M⁻² for (1)₂·(CB[8])₂ (Table S2, Fig. S29a and b^†).Open in a separate windowFig. 2Supramolecular complexation of 1²⁺ and CB[n]. ¹H NMR spectra of 1²⁺ at (a) χ_CB[7] = 2, (b) χ_CB[n] = 0 and (c) χ_CB[8] = 1 in D₂O. Cl⁻ counter-ions are omitted for clarity.The free radical copolymerisation of 1²⁺ and DMAAm ([M] = 2 M), in the absence of CB[n], was based on optimised DMAAm homopolymerisations (Fig. S14 and S15^†) and full conversion was confirmed by ¹H NMR spectroscopy (Table S1 and Fig. S16^†). 1²⁺ was maintained at 1 mol% relative to DMAAm and by varying the radical initiator concentration molar masses of up to 30 kDa with broad dispersities (Đ = 11.4) were obtained (Fig. S17^†). Lower initiator concentrations (<0.25 mol%) limited polymerisation (M_n = 3.7 kDa) and size exclusion chromatography elution peaks exhibited extensive tailing, suggesting that 1²⁺ engages in radical transfer processes.To verify our hypothesis that CB[n] macrocycles can modulate the redox behavior of 1²⁺, FRP of 1²⁺ and DMAAm was conducted with varying amounts of CB[n] (n = 7, 8) (Fig. 3, S18 and S20^†). Full conversion of all monomers including their successful incorporation into the polymer was verified via¹H NMR spectroscopy and SEC (Fig. S18 and S21–S23^†). Using CB[7], the molar mass of the copolymers was tunable between M_n = 3.7–160 kDa (Fig. 3b and S21a^†). Importantly, in the presence of CB[8], a broad range of molar masses M_n = 3.7–500 kDa were accessible for 0 < χ_CB[8] < 1.2 (Fig. S20 and S21b^†). Increasing the CB[n] (n = 7, 8) concentration caused dispersity values to converge to Đ = 2.2 (χ_CB[8] = 1.2, χ is the ratio of CB[n] to the redox-active monomer, Fig. S20^†). The copolymers were purified by addition of adamantylamine (competitive binder) prior to dialysis to deliver CB[n]-free redox-active copolymers (Fig. S23^†).Open in a separate windowFig. 3(a) In situ copolymerisation of DMAAm with 1²⁺ and CB[7]. (b) Molar mass and dispersity vs. amount of CB[7] in the system. Fitted curve is drawn to guide the eye. Cl⁻ counter-ions are omitted for clarity.The range of molar masses obtainable through addition of CB[n] (n = 7, 8) correlated with the measured K_a (Fig. 3b and S20^†). Binding of 1²⁺ to CB[8] was stronger and therefore lower concentrations of CB[8] were required to shift the binding equilibrium and mitigate disruption of the polymerisation. Dispersity values reached a maximum at χ_CB[7] = 0.6 or χ_CB[8] = 0.3, suggesting 1⁺˙ is only partially encapsulated. Consequently, higher CB[n] concentrations can enable FRP with lower initiator concentrations (0.10 mol%, Fig. S19^†), which demonstrates the major role of complexation to modulate electron accepting properties of 1²⁺.The redox-active monomer 1²⁺ can engage with propagating primary radicals (P_m˙) to either be incorporated into the growing polymer chain (P_m–1²⁺˙) or to abstract an electron deactivating it (P_m). This deactivation likely occurs through oxidative termination producing 1⁺˙ (energetic sink), inactive oligo- and/or polymer chains (P_m) and a proton H⁺, causing retardation of the overall polymerisation. Oxidative terminations have been previously observed in aqueous polymerisations of methyl methacrylate, styrenes and acrylonitriles that make use of redox initiator systems.^44–47 Another example by Das et al. investigated the use of methylene blue as a retarder, with the primary radical being transferred to a methylene blue electron acceptor via oxidative termination, altogether supporting the outlined mechanism of our system (extended discussion see ESI, Section 1.4^†).⁴⁸The process of retardation can, however, be successfully suppressed, when monomer 1²⁺ is encapsulated within CB[n] macrocycles. Herein the formation of 1·(CB[7])₂ or (1)₂·(CB[8])₂ results in shielding of the redox-active component of 1²⁺ from other radicals within the system, hampering other electron transfer reactions. This inhibits termination and results in extended polymerisation processes leading to higher molar mass polymers through mitigation of radical transfer reactions. Moreover, suppressing the formation of 1⁺˙ through supramolecular encapsulation minimises both π and σ dimerisation of the emerging viologen radical species,³⁹ preventing any further reactions that could impact the molar mass or polydispersity of the resulting polymers.Cyclic voltammetry (CV) and UV-Vis titration experiments were conducted to provide insight into the impact of CB[n] on the redox behavior and control over FRP of 1²⁺. Excess of CB[n] (n = 7, 8) towards 1²⁺ resulted in a complete suppression of electron transfer processes (Fig. S31 and S32^†). Initially, 1²⁺ shows a quasi-reversible reduction wave at −0.44 V forming 1⁺˙ (Fig. 4a). Increasing χ_CB[7], this reduction peak decreases and shifts towards more negative potentials (−0.51 V, χ_CB[7] = 1) accompanied by the formation of 1²⁺·(CB[7])₁. A second cathodic peak emerges at −0.75 V due to the increased formation of 1²⁺·(CB[7])₂. At χ_CB[7] = 2, this peak shifts to −0.80 V, where it reaches maximum intensity, once 1²⁺·(CB[7])₂ is the dominating species in solution. When 2 < χ_CB[7] < 4, the intensity of the reduction peak decreases and the complexation equilibrium is shifted towards the bound state, complete suppression of the reduction peak occurs at χ_CB[7] = 4. Similarly, the oxidation wave intensity is reduced by 95% at χ_CB[7] = 4 causing suppression of potential oxidative radical transfer processes (Fig. 4c).Open in a separate windowFig. 4Mechanism of the CB[n]-mediated (n = 7, 8) strategy for the controlled copolymerisation of redox-active monomer 1²⁺. (a) Cyclic voltammogram with varying amounts of CB[7]. (b) UV-Vis titration of 1²⁺ with varying amounts of CB[7]. (c) Intensity decay of the oxidation peak at −0.27 V and change in absorption maximum of 1⁺˙ at 590 nm vs. χ_CB[7]. (d) Electron transfer processes of 1²⁺ to generate 1⁺˙ and 1⁰. (e) Reduction of 1²⁺ resulting in precipitation of 1⁰. (f) Stabilisation of 1⁺˙ through encapsulation with CB[7]. (g) Protection of 1²⁺ from redox processes through CB[7]-mediated encapsulation.The concentration of 1⁺˙ can be monitored using UV-Vis (Fig. 4b and S34^†).⁴⁹ Absorbance at 590 nm (λ_max) vs. χ_CB[7] was plotted and the concentration of 1⁺˙ increases, reaching a maximum at χ_CB[7] = 4 (Fig. 4c). When χ_CB[7] > 4, a decrease in concentration of 1⁺˙ was observed. We postulate the following mechanism: at χ_CB[7] = 0, 1²⁺ is reduced to produce high concentrations of 1⁺˙ that partially disproportionates to form 1⁰, which precipitates (Fig. 4e and S34^†). When 0 < χ_CB[7] < 4, increasing amounts of green 1⁺˙ are stabilised through encapsulation within CB[7] suppressing disproportionation (Fig. 4c (cuvette pictures), Fig. 4f). For χ_CB[7] > 4, 1²⁺ is protected from reduction through encapsulation (Fig. 4g).To further demonstrate applicability of this strategy, we chose another viologen-based monomer 2²⁺ for copolymerisation (Fig. 5a). As opposed to 1²⁺, CB binds predominantly to the styryl moiety of 2²⁺ (Fig. S27 and S28^†).⁵⁰ ITC data showed that 2²⁺ binds CB[7] in a 1 : 1 fashion with a binding affinity of K_a = 2.32 × 10⁶ M⁻¹ (Fig. S30 and Table S2^†). Monomer 2²⁺ was also analysed via CV and showed three reversible reduction waves at −0.91 V, −0.61 V (viologen) and 0.40 V (styrene). Similar to 1²⁺, excess CB[7] selectively protects the molecule from redox processes, while the vinyl moiety remains accessible (Fig. 5a, S33c and d^†). For CB[8], only partial suppression of electron transfer processes was observed (Fig. S33e and f^†). We therefore chose CB[7] as an additive to increase control over FRP of 2²⁺ (Fig. 5b). Copolymerisation of 2²⁺ (1 mol%) and DMAAm ([M] = 2 M) at χ_CB[7] = 0 resulted in M_n = 28 kDa. When χ_CB[7] = 0.1, 0.2 or 0.3, M_n increased gradually from 124 to 230 and 313 kDa, respectively, demonstrating the potential of this strategy for FRP of redox-active monomers. Higher percentages of CB[7] led to copolymers with presumably higher molar masses causing a drastic decrease in solubility that prevented further analysis. Investigations on a broader spectrum of such copolymers, including those with higher contents of 2²⁺ are currently ongoing.Open in a separate windowFig. 5(a) Cyclic voltammogram of viologen-containing monomer 2²⁺ and its complexation with CB[n] (n = 7, 8) at a concentration of 1 mM using a scan rate of 10 mV s⁻¹ in 0.1 mM NaCl solution. (b) Molar mass and dispersity of 2²⁺-containing copolymers vs. equivalents of CB[7]. Cl⁻ counter-ions are omitted for clarity.In conclusion, we report a supramolecular strategy to induce control over the free radical polymerisation of redox-active building blocks, unlocking high molar masses and reducing polydispersity of the resulting polymers. Through the use of CB[n] macrocycles (n = 7, 8) for the copolymerisation of styrenic viologen 1²⁺, a broad range of molar masses between 3.7–500 kDa becomes accessible. Our mechanistic investigations elucidated that the redox behavior of monomer 1²⁺ is dominated by either CB[n]-mediated stabilisation of monoradical cationic species or protection of the encapsulated pyridinium species from reduction. In the stabilisation regime (χ_CB[7] < 4), 1²⁺ is reduced to form the radical cation 1⁺˙, which is subsequently stabilised through CB[7] encapsulation. Upon reaching a critical concentration of CB[7] (χ_CB[7] > 4), the system enters a protection-dominated regime, where reduction of 1²⁺ is suppressed and the concentration of 1⁺˙ diminishes. The resulting copolymers can be purified by use of a competitive binder to remove CB[n] macrocycles from the product. This strategy was successfully translated to a structurally different redox-active monomer that suffered similar limitations. We believe that the reported strategy of copolymerisation of redox-active monomers will open new avenues in the synthesis of functional materials for energy conversion and storage as well as for applications in electrochromic devices and (nano)electronics.     

Exciton interactions in helical crystals of a hydrogen-bonded eumelanin monomer

Eumelanin, a naturally occurring group of heterogeneous polymers/aggregates providing photoprotection to living organisms, consist of 5,6-dihydroxyindole (DHI) and 5,6-dihydroxyindole-2-carboxylic acid (DHICA) building blocks. Despite their prevalence in the animal world, the structure and therefore the mechanism behind the photoprotective broadband absorption and non-radiative decay of eumelanin remain largely unknown. As a small step towards solving the incessant mystery, DHI is crystallized in a non-protic solvent environment to obtain DHI crystals having a helical packing motif. The present approach reflects the solitary directional effect of hydrogen bonds between the DHI chromophores for generating the crystalline assembly and filters out any involvement of the surrounding solvent environment. The DHI single crystals having an atypical chiral packing motif (P2₁2₁2₁ Sohncke space group) incorporate enantiomeric zig-zag helical stacks arranged in a herringbone fashion with respect to each other. Each of the zig-zag helical stacks originates from a bifurcated hydrogen bonding interaction between the hydroxyl substituents in adjacent DHI chromophores which act as the backbone structure for the helical assembly. Fragment-based excited state analysis performed on the DHI crystalline assembly demonstrates exciton delocalization along the DHI units that connect each enantiomeric helical stack while, within each stack, the excitons remain localized. Fascinatingly, over the time evolution for generation of single-crystals of the DHI-monomer, mesoscopic double-helical crystals are formed, possibly attributed to the presence of covalently connected DHI trimers in chloroform solution. The oligomeric DHI (in line with the chemical disorder model) along with the characteristic crystalline packing observed for DHI provides insights into the broadband absorption feature exhibited by the chromophore.

Single crystals of DHI monomer, a eumelanin precursor, adopt an atypical chiral packing arrangement incorporating enantiomeric zig-zag helical stacks while its covalently connected DHI trimer forms double-helical crystals in the mesoscopic scale.

Eumelanin, which represents a broad class of natural pigments found in the animal kingdom, acts as a biological shield for protecting the skin cells against harsh UV radiation.¹ Eumelanin, a black coloured pigment, obtained from the oxidative polymerization of 5,6-dihydroxyindoles (DHIs) and 5,6-dihydroxyindole carboxylic acid (DHICA) is one of the extensively explored archetypes of the melanin family (Fig. 1a).² Eumelanin has synergistic merits of possessing broadband UV absorption and proficient dissipation of the excessive electronic energy via non-radiative deactivation of the excited states, thereby resulting in the photoprotective nature of the pigment.³ Apart from the photoprotective behaviour, melanin possesses exceptional antioxidant activity via its free radical scavenging traits.^4–7 However, much less has been understood about the fundamental photophysics and structural features of eumelanin due to the enormous heterogeneity in the molecular framework^8,9 coupled with poor solubility in common solvents.¹⁰ Recent years have witnessed a growing interest towards unravelling the excited state processes occurring in the eumelanin pigment upon interaction with light.^11,12 A better correlation between the structure–property relationship and photoexcited state processes in eumelanin can guide the development of inspired functional materials for potential application in biomedical and dermo-cosmetic fields.^1,13–15Open in a separate windowFig. 1(a) Chemical diagrams of 5,6-dihydroxyindole (DHI) and 5,6-dihydroxyindole carboxylic acid (DHICA). (b) Various unconventional hydrogen bonding interactions identified in DHI crystals.In the natural world, the chromophoric architecture of eumelanin can symbolize an organized and efficient organic system for photoprotection that nature developed through evolution. The outcome of the research done so far indicates the presence of continuous π-stacks of oligomers in eumelanin which induce different levels of aggregation to construct the eumelanin framework.¹⁶ Furthermore, eumelanin has been reported to exhibit weak fluorescence which indicates the presence of competing non-radiative channels that provide efficient de-excitation pathways for repopulating the ground states.^17–20 The broadband absorption of eumelanin has theoretically and experimentally been evaluated, in part, to be a result of π-stacking interactions between the oligomers of DHI/DHICA in multiple oxidation states.^21–24 While dealing with biomacromolecules, non-covalent interactions such as hydrogen bonding and π–π stacking take the centre stage in controlling the supramolecular architecture, especially in the 3-dimensional structures of proteins and DNA. Modulating the balance between each of these noncovalent interactions over another will produce significant changes not only in the structure but also in the functional properties.Hydrogen bonding is simultaneously both ubiquitous and diverse and therefore its significance in biochemical systems comes as no surprise particularly due to the surrounding aqueous environment. Apart from the classical hydrogen bonding interactions, an array of hydrogen bond-like weak interactions which include a delocalized π-system acting as the acceptor group to the X–H hydrogen donor (X = O, N, C) is identified to provide additional contributions for stabilizing the biomolecular structure and controlling intrinsic functions (Fig. 1b).^25,26 Investigations aimed at identification (using X-ray crystallography) followed by energetic quantification of the major stabilizing interactions such as those with the aromatic π-rings in biological complexes are of paramount importance for developments in diverse areas including drug design. There has been extensive research conducted on eumelanin building blocks showcasing their ability to form hydrogen bonds through the –OH and –NH functional groups.^27,28 Most reports almost exclusively focus on the hydrogen bonding with the solvent environment surrounding the eumelanin monomer units.²⁹ Findings from these theoretical studies have demonstrated the role of several deactivation pathways in the presence of a protic solvent, namely –OH and –NH bond elongation and 5-/6-membered ring puckerings.¹⁸Chemical and spectral evidence from the eumelanin polymeric structures identified so far points to five main levels of chemical disorder leading to the supramolecular structure, which includes (i) disorder from the simultaneous presence of different building blocks; (ii) molecular size disorder; (iii) disorder from the position of coupling; (iv) electronic/redox disorder of the constituent units and (v) supramolecular disorder.³⁰ Given the complex structure of melanin, a bottom-up approach using the building blocks or basic constituent molecules of eumelanin is a pertinent strategy for the mechanistic study of the photoproperties of eumelanin. This can be followed by understanding the more intricate structures of melanin formed from the constituents with less complex approaches. Due to the abundant presence of water in the natural media, the corresponding solute–solvent interactions can have profound significance in driving the fast polymerization and the consequent heterogeneity of natural melanin. The tedious task of extracting melanin from natural sources and the lack of solubility of the polymeric melanin material in organic/aqueous solvents have called for basic model systems to understand the complex eumelanin architecture. In this regard, we have adopted a facile approach to decode the perpetual puzzle by single crystal X-ray crystallographic and spectroscopic analyses of DHI crystalline aggregates derived from a non-aqueous environment (chloroform). Due to the highly autooxidative³¹ nature of eumelanin precursors even in the slight presence of protic solvents, the simple model implemented here precludes the contribution of solute–solvent hydrogen bonding interaction towards the formation and resultant structure of DHI crystalline aggregates. In chloroform, each DHI molecule experiences weak interactions solely from the neighbouring DHI chromophores thereby leading to helical aggregation.Our efforts towards recognizing and monitoring the photogenerated excitons and charge-transfer dynamics in crystalline and contorted polyaromatic assemblies^32–36 prompted us to explore the structure–optical property relationship in the eumelanin precursor molecule DHI. Unlike the commonly understood π-stacking in eumelanin derivatives, the single crystals of DHI arrange in a helical zig-zag fashion with a completely edge-to-face aggregate structure driven by both conventional and unconventional hydrogen bonds (Fig. 1b). The structural heterogeneity imposed by the different hydrogen bonds has led to varied levels of exciton delocalization between the neighbouring chromophores in the crystalline DHI aggregates. Along with the diffracting single crystals of monomeric DHI, covalently connected trimeric units of DHI are also identified in chloroform solution, which form double-helical crystals in the mesoscopic scale. Such double-helical architectures are omnipresent in nature as exemplified by the DNA structure.DHI was synthesized by following a previously reported procedure having l-dopa as the starting material (Scheme S1, Appendix C1–C4 and C7, ESI^†).³⁷ Slow evaporation from dry chloroform solution of DHI produced colourless diamond shaped single crystals of DHI (Fig. S1a^†). Interestingly, the DHI molecule with no chiral centre atypically crystallized in the Sohncke space group, P2₁2₁2₁ (Table S1^†). Single-crystal X-ray structure analysis revealed the presence of conventional and unconventional hydrogen bonds directing the crystalline self-assembly of DHI chromophores about a zig-zag helical backbone. Fig. 2 presents the four different types of non-covalent dimers (D1–4) distinguished within the DHI crystal. The zig-zag helical stacks proceed along the crystallographic a-axis (Fig. S2^†) and are fabricated by bifurcated O–H⋯O hydrogen bonds (d_O1⋯H = 2.25 Å, d_O2⋯H = 2.34 Å, <O1–H–O2 = 67.89°) between the two hydroxyl substituents in the DHI chromophore (D3 in Fig. 2). Such bifurcated hydrogen bonded assemblies are prevalent in the secondary and tertiary structures of proteins.³⁸ Interestingly, in the DHI crystal, enantiomeric helical stacks (Fig. S2^†) that are arranged with respect to different screw axes are observed, wherein each stack aligns in a herringbone fashion to the other zig-zag helix (as represented by the dimers D1, D2 and D4). The stacks are interconnected majorly through the unconventional hydrogen bonds such as C–H⋯π (d_C–H⋯π = 2.66–2.99 Å), O–H⋯π (d_O–H⋯π = 2.59 Å), C–H⋯N (d_C–H⋯N = 2.72 Å), C–H⋯O (d_C–H⋯O = 2.66 Å) and the classical N–H⋯O (d_N–H⋯O = 2.72 Å) hydrogen bonds. The absence of π–π stacking interaction is validated by the Hirshfeld surface analysis wherein the formation of the DHI crystalline assembly is majorly stabilized by the C⋯H (40.5%), H⋯H (29.7%), N⋯H (4.3%) and O⋯H (25.4%) noncovalent interactions (Fig. S3 and Table S2^†).Open in a separate windowFig. 2Different orientations of DHI (D1–4) and the directing hydrogen bonds observed in the single crystal.Detailed examination of the interchromophoric interactions supporting the zig-zag helical stacks in DHI crystals using Bader''s quantum theory of atoms in molecules (QTAIM) analysis revealed the presence of supramolecular synthons in the crystal system (Fig. S4^†). This is evidenced by the (3,+1) ring critical points in each of the representative dimers. For a molecular self-assembly to occur efficiently, recognition between the intermolecular functionalities is important, which often culminates in the formation of smaller repetitive units or supramolecular synthons.³⁹ The recognition information which is then carried by these units forms the kernel of self-aggregation or crystallization processes. In the case of DHI crystals, all the dimer assemblies D1–D4 display synthon formation with the D1 and D3 synthons showing greater energetic stabilities. The dimer unit representing the helical backbone, D3, forms two supramolecular synthons orchestrated by the bifurcated O–H⋯O hydrogen bonds and a weak C–H⋯O interaction (Fig. S4^†). The five- and six-membered rings so formed fabricate the helical zig-zag backbone of the DHI crystal. Similarly, the dimer D1 also forms two synthons through a classical N–H⋯O interaction along with the weak C–H⋯π and O–H⋯π interactions. D2 and D4 dimers are stabilized by one synthon each, materialized by C–H⋯π and C–H⋯N interactions in D2 and N–H⋯O and C–H⋯O interactions in D4.The synthon formation and the concomitant electron delocalization involving the π-rings in dimers D1 and D2 have resulted in the aromatic stabilization of the π-rings upon comparison with the monomer DHI. The nucleus independent chemical shift (NICS(1)) values evaluated for D1–4 and monomer DHI in the ground state indicate the aromatic stabilization of the π-surface leading to the favourable alternate stacking of the enantiomeric zig-zag helices facilitated by the unconventional hydrogen bonds. The negative NICS(1) values for the six- and five-membered (6C, 5C) rings of molecule A (Fig. S5 and Table S3^†) increased to −27.22 ppm (6C) and −29.13 ppm (5C) when compared to the monomer DHI (6C: −26.14 ppm, 5C: −28.03 ppm). Similarly, in D2, the π-surface of molecule B that is involved in the weak interaction undergoes aromatic stabilization (6C: −29.43 ppm, 5C: −31.48 ppm). Truncated symmetry adapted perturbation theory (SAPT(0)) analysis⁴⁰ performed on the DHI dimers shows higher stabilization for D1 (E^SAPT_int = −9.70 kcal mol⁻¹) and D3 (E^SAPT_int = −6.57 kcal mol⁻¹) dimers, which could be attributed to the two supporting supramolecular synthons in both the dimers (Table S4^†). The total stabilization of D1 and D3 orientations is facilitated by the higher contributions of electrostatic energy (E⁽¹⁾_elc = −6.17 to −6.14 kcal mol⁻¹) and induction energy (E⁽¹⁾_elc = −1.60 to −1.38 kcal mol⁻¹) towards the total SAPT energy. The dominant role of the stronger classical hydrogen bonds in the fabrication of D1 and D3 synthons when compared to the other DHI orientations (having equal contribution from the weak unconventional interactions) explains the observed energy distribution in the SAPT(0) analysis.The crystalline architecture of the DHI precursor molecule identified herein could provide a sound model for understanding the inherent nature of the excited energy states leading to the characteristic photo-function of the eumelanin pigment. Several experimental and theoretical investigations on eumelanin aggregates revealed the occurrence of excitation energy transfer within the aggregates.^41,42 The extent of energy delocalization within the four dimer orientations in the DHI crystal structure is determined using the fragment-based excited state analysis developed by Plasser, executed in the TheoDORE package.^43–45 The expanse of excitation delocalization amongst the fragment units is described by the value of participation ratio (PR). The contribution towards the exciton delocalization from the fragments is defined by the mean position or the POS value, wherein the indicated number shows the involvement of one or more units. The charge-transfer or Frenkel character of the excited states is defined by the CT number which assumes values closer to one for pure CT states and closer to zero for pure Frenkel states. In dimers D1 and D2, the first singlet excited state of highest oscillator strength (S2) shows Frenkel exciton character with effective delocalization of the excitons between the individual constituting units during the excitation process (Tables S5 and S6^†). The low CT values of the S2 states in D1 (PR = 1.79, POS = 1.33, CT = 0.08) and D2 (PR = 1.98, POS = 1.45, CT = 0.02) along with a PR close to a value of two indicated the delocalization of the Frenkel excitons on the two monomers. Hence, there exists a possible excitation energy delocalization along the adjacent enantiomeric stacks in the DHI crystal. However, for D3 with the bifurcated hydrogen bonding interaction between the monomeric units, both S1 and S2 states have significant oscillator strength. The Frenkel excitons in S1 and S2 states remain localized on only one fragment of D3 (Table S7^†), while in D4, there exists a partial delocalization of the Frenkel excitons in the S1 state (Table S8^†). Thus, it is understood that within each enantiomeric stack the initial Frenkel excitons remain localized on one fragment. The natural transition orbitals (NTOs) of dimers D1–4 give an idea about the nature of the dominant orbital transitions for the allowed electronic excitations. The allowed electronic excitation in all the dimers could only be well represented by taking two distinct orbital transitions with significant coefficients into consideration. The absence of a single dominant orbital transition in the dimers proposed the need for fragment-based hole–electron analysis for the better understanding of the electronic energy delocalization. The hole–electron isosurface analysis provides a pictorial representation of the corresponding delocalized and localized nature of the initial Frenkel excitons in the four DHI orientations (Fig. 3). The delocalization of the hole and electron density on both the fragments puts forward the possibility of effective energy transfer in dimers D1, D2 and D4.⁴⁶ In dimer D3, a very weak delocalization of the electron–hole densities was observed suggesting a localized exciton formation.Open in a separate windowFig. 3Hole–electron isosurface plots of the DHI orientations in the crystal. (a) D1, (b) D2, (c) D3 and (d) D4.Several factors including slight exposure to air,⁴⁷ humidity and/or light have been observed to cause the autooxidative polymerization of DHI, wherein lowered temperatures decrease the kinetics of this solid-state polymerization. Although, a major fraction of the solid oligomer mixture remains as the DHI monomer (≥80%), oligomeric units up to DHI-hexamers have been identified with varying solubilities in different solvents. MALDI-MS spectra (Appendix C5, ESI^†) sequentially collected for the oligomer mixture in CHCl₃ and DMSO indicated the major presence of DHI trimers (in CHCl₃; m/z = 442.138) and DHI-hexamers (in DMSO; m/z = 882.382) along with the smaller counterparts. Interestingly, over the time course for nucleation and the subsequent growth mechanism of the single crystals of monomeric DHI in chloroform, oligomerization of DHI is found to occur concurrently to form the covalently connected DHI trimer (DHI-T). Along with the diffracting single crystals of DHI, small right-handed double-helical crystals (non-diffracting, Fig. S1b–f and S6^†) are observed for the first time, which could be attributed to the self-assembled morphology of DHI-T (Fig. 4a). Observed only in chloroform, we speculate that the prolonged exposure of the chlorinated solvent plays a significant role in the chemical transformation of the DHI monomer to the covalent trimer, possibly through a radical initiated reaction.⁴⁸ The ¹H-NMR spectrum obtained for the bulk crystalline sample dissolved in CDCl₃ along with the observed MALDI-MS data of the DHI trimer evidenced the formation of DHI-T, although in very low yields when compared to the monomer (Appendix C5 and C6^†).Open in a separate windowFig. 4(a) Optical microscopy images of the right-handed double-helical crystals of the DHI trimer. (b) Normalized absorbance and excitation spectra showing DHI-T formation in chloroform solution and the Kubelka–Munk transformed diffuse reflectance spectrum of the DHI bulk crystal. (c) Optimized structure of DHI-T at the CAM-B3LYP/6-311g+(d,p) level of theory.With the understanding that DHI readily aggregates/crystallizes in chloroform, the directionality of the hydrogen bonding interactions (to the π-ring) within the definite spatial arrangements of D1 and D2 orientations hints towards the mode/position of coupling for the associated generation of DHI-T (Appendix C9^†). The molecular structure of the DHI trimer that best fits the observed characterization data (Fig. S7a and Appendix C8, ESI^†) is in line with the chemical disorder model having semiquinone and catechol units connected covalently as an extension of the D1 and D2 noncovalent interactions. Geometry optimization of the predicted structure of the DHI-T performed using the CAM-B3LYP/6-311g+(d,p) level of theory in Gaussian 16 suite shows a twisted conformation having the possibility of forming intramolecular hydrogen bonds from the –OH and –NH functionalities (Fig. 4 and S7b^†). Separation of DHI-T from DHI is a real challenge since exposure of DHI to the adsorbent in column chromatography accelerates the oxidative polymerization of DHI, resulting in a black, insoluble material difficult to characterize. Also, the presence of higher oligomer units of DHI (hexamers etc.) was not identified in the multiple data sets collected for the DHI sample dissolved in CHCl₃.Solvent-dependent steady-state UV-vis absorption and fluorescence emission measurements of DHI were performed and the line shapes of the absorption spectra of monomeric DHI in different solvents match consistently (Fig. S8^†). Two major absorption bands at λ_a1 ∼ 270 nm and λ_a2 ∼ 300 nm form the characteristic absorption spectrum of the DHI monomer. The fluorescence emission of monomeric DHI exhibits a single broad spectral feature peaking at λ_em ∼ 330 nm in a majority of the solvents. The relative fluorescence quantum yields are found to be exceptionally low in chloroform, dichloromethane, THF and water (Table S9^†) indicating the presence of non-radiative decay channels for dissipating the excitation energy. In chloroform, the emergence of a red-shifted tail in the absorption spectrum of the DHI-monomer is observed over time possibly signifying the onset of DHI-T formation. The fluorescence emission in CHCl₃ also shows a new band peaking at λ_em ∼ 460 nm along with the emission band at λ_em ∼ 335 nm (Fig. S9^†). A broad red-shifted band arising at 370 nm in the excitation spectrum of DHI solution collected in chloroform at λ_em ∼ 460 nm (Fig. 4b) evidenced the presence of the DHI trimer. A similar decrease in fluorescence quantum yield in chloroform and the concomitant emergence of new bands in the fluorescence emission profiles have been noted previously in tryptophan and other indole species synthesized for eumelanin investigation.^49,50 In such cases, the photoionization of excited indole leads to the ejection of a solvated electron which attacks the chloroform molecule, releasing a chloride ion, and further undergoes reactions to yield photoproducts.Spectroscopic investigation of the crystalline DHI (containing both monomer single crystals and covalent trimer crystals) sample showed broad, red-shifted absorption bands spanning from 210 to 560 nm (Fig. 4b and S10^†). The solid-state absorption spectrum shows two prominent bands at λ_a1 ∼ 280 nm and λ_a2 ∼ 305 nm which could be attributed to the red-shifted absorption bands of the crystalline DHI monomer (compared to the monomer absorption bands in the solution state). The observed red-shift in the absorption band of the DHI crystal arises from the nonplanar packing motif and the ensuing intermolecular interactions in the solid state. The presence of a broad shoulder band centred at λ_a3 ∼ 375 nm in the absorption spectrum could be assigned to the double-helical crystal of the covalent DHI trimer. The crystalline state fluorescence emission spectrum of DHI spans from 390 to 490 nm (Fig. S10^†). The excitation energies and the allowed vertical transitions of the monomer and DHI-T have been computed at the CAM-B3LYP/6-311g+(d,p) level of theory. Unlike the precursor DHI monomer which undergoes higher energy electronic transitions (at 270 nm and 300 nm), the favorable transition in the covalent trimer is red-shifted with the S₀ → S₁ electronic excitation occurring at ∼434 nm. Hence, the spectroscopic and theoretical investigation of bulk crystalline DHI indicates that the broad absorption profile of the eumelanin precursor could be ascribed to the combined effects of the non-planar chromophore stacking and the presence of covalent DHI trimers that exist as double-helical aggregates. The photoprotective nature of eumelanin arises from the signature broadband absorption of eumelanin which spans throughout the UV and visible region tailing around 800 nm as explained by the chemical disorder model. In line with this understanding, the spectroscopic data of the DHI crystal also exhibit broadband absorption which expands up to 600 nm unlike the DHI monomer. Moreover, the low relative fluorescence quantum yields of DHI suggest the presence of non-radiative decay pathways within the DHI units. The oligomeric trimer which in itself shows structural heterogeneity aggregates as double helical structures and shows a red-shifted absorption band which is comparable to the computed TDDFT vertical excitation energies. Thus, our report on the characterization of DHI and the oligomeric trimer could possibly be beneficial in advancing melanin structure characterization and elucidating the photoprotective function of eumelanin.The solid-state CD spectrum of crystalline DHI (in KBr, Fig. S11a^†) showed the signatures for the presence of helical packing^51–55 (possibly from the zig-zag helical motif along with the double helical arrangement). However, the basis of the CD couplet of significant intensity spanning from ∼250 to 600 nm (including ranges outside of the absorption maxima) could not be exclusively assigned to chiral absorption from the chromophoric packing.⁵⁶ In the case of macromolecular systems having long-range organization, differential scattering of incident left and right circularly polarized light can provide significant contributions to the observed circular dichroism.^57–59 The occurrence of broad CD bands outside the absorption bands of the macromolecule can signify the possible role of differential scattering in the circular dichroism.⁶⁰ Although, for the DHI sample, the characteristic CD spectrum has been found to be concurrent for the different data sets collected using freshly prepared crystalline samples on different days (Fig. S11a^†), the ratios of the intensities of the positive and negative bands have been observed to vary. Such a heterogeneity in the ellipticity values of the positive and negative bands could be attributed to the possible presence of different enantiomeric assemblies that exhibit varying abilities to undergo chiral absorption and differential scattering.⁵⁷ The possibility of having linear dichroism (LD) artifacts in the CD data was evaluated for the DHI sample (Fig. S11b^†).⁶¹ The LD artifact fell within the error bar of the order of 10⁻³–10⁻⁴ mdeg and hence, the contribution of LD to the strong CD signal of DHI could be ignored.^62,63 Also, the idea of having a chiral nucleation centre, probably from any conformationally chiral DHI oligomer units, leading to the double-helical aggregation and the consequent mesoscopic chirality could not be ignored while assessing the origin of the observed CD spectrum. The existing uncertainties in solving the source of the double-helical aggregation of the DHI chromophores and identifying the intermolecular forces acting behind the same remain a challenge that requires detailed examination in future studies.     

0D–1D hybrid nanoarchitectonics: tailored design of FeCo@N–C yolk–shell nanoreactors with dual sites for excellent Fenton-like catalysis

Heterogeneous Fenton-like processes are very promising methods of treating organic pollutants through the generation of reactive oxygen containing radicals. Herein, we report novel 0D–1D hybrid nanoarchitectonics (necklace-like structures) consisting of FeCo@N–C yolk–shell nanoreactors as advanced catalysts for Fenton-like reactions. Each FeCo@N–C unit possesses a yolk–shell structure like a nanoreactor, which can accelerate the diffusion of reactive oxygen species and guard the active sites of FeCo. Furthermore, all the nanoreactors are threaded along carbon fibers, providing a highway for electron transport. FeCo@N–C nano-necklaces thereby exhibit excellent performance for pollutant removal via activation of peroxymonosulfate, achieving 100% bisphenol A (k = 0.8308 min⁻¹) degradation in 10 min with good cycling stability. The experiments and density-functional theory calculations reveal that FeCo dual sites are beneficial for activation of O–O, which is crucial for enhancing Fenton-like processes.

Novel 0D–1D hybrid nanoarchitectonics consisting of FeCo@N–C yolk–shell nanoreactors are developed for Fenton-like reaction. With the multilevel advantages of this design, FeCo@N–C nano-necklaces exhibit excellent performance for BPA removal.

Advanced oxidation processes (AOPs) are one of the most promising strategies to eliminate organic contaminants, sustainably generating reactive oxygen species (ROS) to ideally destroy all non-biodegradable, recalcitrant, toxic, or membrane-permeable organic impurities.^1–4 Among these AOPs, sulfate radical (SO₄˙⁻)-based Fenton-like processes have gained increasing attention as a water treatment strategy because of the strong oxidation potential of SO₄˙⁻ (3.1 V vs. normal hydrogen electrode) at wider pH ranges. SO₄˙⁻ is mainly produced by physical or chemical methods for activation of persulfate salts, such as peroxymonosulfate (PMS) and persulfate.^5–9 Over the past two decades, heterogeneous catalysis has emerged as the most effective approach to water treatment, with much effort dedicated to developing better catalysts, including transition metal-based and carbonaceous materials.^10,11 Unfortunately, most metal-based catalysts suffer from leaching of toxic metal ions, which can thwart their practical application,^12,13 and although carbonaceous catalysts produce no secondary pollution, their cycle performance is always depressed.¹⁴ There is therefore an urgent need to find robust catalysts with adequate activity and stability for Fenton-like processes.To achieve superior performance, an ideal Fenton-like catalyst should contain oxidants with favorably reactive centers for cleavage of peroxyl bonds (O–O), have structure optimized for target pollutant attraction, and have chainmail to protect the vulnerable active sites for long periods.^15–17 Recent studies have demonstrated Co–N–C active sites prefer to activate the O–O of PMS.¹⁸ Furthermore, introducing Fe-doping into the Co–N–C system not only suppresses Co²⁺ leaching, but also modulates the pyrrolic-N content, which is the adsorption site for capture of bisphenol A (BPA).¹⁹ We previously discovered that Co@C yolk–shell nanoreactors could enhance the catalytic activity because of the confinement effect in the nano-spaces between the core and shell, while the carbon shell acted like a chainmail protecting the Co active sites, keeping them highly reactive after five cycles.^20,21Combining different kinds of materials to generate novel hybrid material interfaces can enable the creation of new kinds of chemical and physical functionalities that do not currently exist. However, one cannot simply mix these materials in an uncontrolled manner, because the ensemble of interfaces created by random mixing tends to favour thermodynamically stable interfaces that are functionally less active. Therefore, to prepare new materials with high functionality, it is necessary to carefully control the hybridization of components in interfacial regions with nanometric or atomic precision. By further hybridization of different components e.g., zero to one dimension (0D–1D) hybrid structures, we can prepare the structure to increase not only the specific surface area but also the interfacial region between different materials.In this work, we report novel 0D–1D hybrid nanoarchitectonics (necklace-like structures) consisting of FeCo@N–C yolk–shell nanoreactors as a PMS activator for Fenton-like processes. This catalyst has multilevel advantages: (i) each FeCo@N–C unit is a well-formed yolk–shell nanoreactor, which can guarantee sufficient contact of reactants and active sites, as well as defend them for good durability; (ii) all single nanoreactors are threaded along the carbon fibers, providing a highway for electron transport; and (iii) all the carbon fibers constructed into a thin film with macroscopic structure, which overcomes the complex recyclability of powder catalysts. Benefiting from favorable composition and unique structure, the FeCo@N–C catalyst delivers excellent performance for BPA removal via activation of PMS accompanied with good stability.The synthesis processes of necklace-like nanoarchitecture containing FeCo@N–C yolk–shell nanoreactors are illustrated in Fig. 1a. First, uniform Fe–Co Prussian blue analogue (Fe–Co PBA) nanocubes with an average size of 800–900 nm (Fig. 1b) are encapsulated in polyacrylonitrile (PAN) nanofibers by electrospinning. The obtained necklace-like FeCo PBA–PAN fibers (Fig. 1c) are then pyrolyzed at 800 °C in N₂ atmosphere to produce FeCo@N–C nano-necklaces. The scanning electron microscopy (SEM) image (Fig. 1d) of the FeCo@N–C shows this necklace-like morphology with its large aspect ratio, with the FeCo@N–C particles strung along the PAN-derived carbon fibers. A broken particle (Fig. 1e) shows that the FeCo@N–C has a yolk–shell architecture, which is also identified by transmission electron microscopy (TEM). Fig. 1f and g show the well-defined space between the inner yolk and outer shell, which is attributed to the volume shrinkage of the original Fe–Co PBAs. During pyrolysis, Fe–Co PBA is reduced to FeCo (inner yolk) and PAN is carbonized (outer carbon shell), resulting in the unique necklace-like nanoarchitecture.^22–24 The high-resolution TEM in Fig. 1h shows a lattice fringe of 0.20 nm, which matches well with the (110) plane of FeCo alloy.²⁵ The scanning transmission electron microscopy (STEM) image (Fig. 1i) and corresponding elemental map (Fig. 1j) indicate that FeCo nanocrystals are well distributed in the inner core with some small FeCo nanocrystals located on external carbon shells. Furthermore, the control samples of Fe@N–C and Co@N–C nano-necklaces, prepared by only replacing the Fe–Co PBA nanocubes with Fe–Fe PB and Co–Co PBA (Fig. S1^†), also demonstrate the versatility of this synthetic strategy. The formation of hierarchical porous structure, beneficial to the PMS transportation on the surface of catalysts, could be determined by N₂ adsorption–desorption isotherms and corresponding pore volume analysis (Fig. S2 and Table S1^†).Open in a separate windowFig. 1(a) Preparation of FeCo@N–C necklace-like nanoarchitecture. SEM images of (b) Fe–Co PBA cubic particles and (c) the electrospun FeCo PBA–PAN fibers. (d and e) SEM, (f and g) TEM, and (h) high-resolution TEM images of FeCo@N–C nano-necklaces. (i) STEM and (j) the corresponding elemental mappings of C, N, Fe, and Co.The X-ray diffraction patterns of the as-prepared products are depicted in Fig. S3,^† with one prominent diffraction peak centered at 44.8° corresponding to the (110) lattice plane of FeCo alloy. All the products also have a characteristic signal at 26°, implying that graphite carbon is formed during pyrolysis. Raman spectroscopy further analyzed the crystal structures and defects of the FeCo@N–C nano-necklaces (Fig. S4^†), where peaks found at 1349 cm⁻¹ and 1585 cm⁻¹ index the disordered (D band) and graphitic carbon (G band), respectively.²⁶ X-ray photoelectron spectroscopy investigated the composition and valence band spectra of FeCo@N–C nano-necklaces. The survey spectrum (Fig. S5a^†) reveals the presence of Fe (1.4%), Co (1.2%), C (86.4%), N (4.5%), and O (6.5%) in the composite. The high-resolution N 1s spectrum (Fig. S5b^†) exhibits broad peaks at 398.1, 401.1, and 407.4 eV, corresponding to the pyridinic-N, graphitic-N, and σ* excitation of C–N, respectively.²⁷ The high-resolution Fe 2p spectrum (Fig. S5c^†) shows a broad peak at 707.4 eV, attributed to Fe⁰. Similarly, the 777.5 eV peak observed in the Co 2p spectrum (Fig. S5d^†) corresponds to Co⁰, implying that FeCo dual sites have formed.²⁸ The oxidation state of these sites was investigated by ⁵⁷Fe Mössbauer spectroscopy, which found a sextet in the Mössbauer spectrum of the FeCo@N–C nano-necklaces attributed to FeCo dual sites (Fig. 2a and Table S2^†).²⁹ The coordination environment of the FeCo dual sites was also verified by X-ray absorption fine structure (XAFS) spectroscopy. Fig. 2b shows that the X-ray absorption near-edge structure (XANES) spectra of the Fe K-edge, which demonstrates a similar near-edge structure to that of Fe foil, illustrating that the main valence state of Fe in FeCo@N–C nano-necklaces is Fe⁰. Furthermore, the extended-XAFS (EXAFS) spectra (Fig. 2c) displays a peak at 1.7 Å, which is ascribed to the Fe–N bond, and a remarkable peak at approximately 2.25 Å corresponding to the metal–metal band.^10,30 The Co K-edge and EXAFS spectra (Fig. S6^†) also confirm the presence of Co–N and the metal–metal band. These results provide a potential structure of the FeCo dual sites in the FeCo@N–C nano-necklaces, as illustrated in Fig. 2d.Open in a separate windowFig. 2(a) ⁵⁷Fe Mössbauer spectra of FeCo@N–C nano-necklaces at 298 K. (b) Fe K-edge XANES spectra of FeCo@N–C nano-necklaces and Fe foil. (c) Corresponding Fourier transformed k³-weighted of the EXAFS spectra for Fe K-edge. (d) Possible structure of the FeCo dual sites.This dual-metal center and necklace-like structure may be beneficial to enhance catalytic performance. Fig. 3a shows the Fenton-like performance for BPA degradation compared to Fe@N–C nano-necklaces, Co@N–C nano-necklaces, and FeCo@N–C particles (Fe–Co PBA directly carbonized without electrospinning). Here, the FeCo@N–C nano-necklaces display a higher catalytic performance, with BPA completely removed in 7 min. To clearly compare their catalytic behavior, the kinetics of BPA degradation was fitted by the first-order reaction. As shown in Fig. 3b, FeCo@N–C nano-necklaces exhibit the highest apparent rate constant (k = 0.83 min⁻¹), which is approximately 6.4, 2.6, and 1.2 times that of FeCo@N–C particles, Fe@N–C nano-necklaces, and Co@N–C nano-necklaces, respectively. The significantly enhanced performance of FeCo@N–C nano-necklaces suggests that the FeCo dual sites and necklace-like nanoarchitecture are crucial. Furthermore, the concentration of BPA and PMS in the solution is higher than that in yolk–shell nanoreactor, resulting a concentration gradient which helps to accelerate the diffusion rates of reactants (Fig. 3c).^31,32 For these nano-necklaces, the carbon shell acts like a chainmail protecting the FeCo active sites from attack by molecules and ions, and all the nanoreactors are threaded along the carbon fibers, providing a highway for electron transport, which is important for SO₄˙⁻ generation (SO₄˙⁻ production as eqn, HSO₅⁻ + e⁻ → SO₄˙⁻ + OH⁻). Electrochemical impedance spectroscopy further confirms the good conductivity of the FeCo@N–C nano-necklaces (Fig. 3d). In addition, the concentration of metal-ion leaching and cycling performance (Fig. 3e and f) reveal the high reusability of FeCo@N–C nano-necklaces, with 95% BPA removal in 20 min after five cycles, which is also proved by the SEM and TEM characterization (Fig. S7^†). The effect of other reaction parameters on the BPA degradation, such as pH, reaction temperature, PMS or catalysts dosage, and common anions, were investigated in detail (Fig. S8–S11^†). All the results demonstrate that FeCo@N–C nano-necklaces deliver a better performance for PMS catalysis. In addition, the turnover frequency (TOF) value of FeCo@N–C nano-necklaces is 5.5 min⁻¹ for BPA degradation, which is higher than many previously reported catalysts (detailed catalytic performance comparison as shown in Table S3^†).Open in a separate windowFig. 3(a) BPA degradation efficiency in different reaction systems and (b) the corresponding reaction rate constants. (c) Schematic illustration of PMS activation in FeCo@N–C nano-necklaces. (d) Nyquist plots of the catalysts. (e) The metal leaching in different reaction systems. (f) Cycling performance of FeCo@N–C nano-necklaces for BPA removal. Reaction conditions: [catalyst] = 0.15 g L⁻¹, [BPA] = 20 mg L⁻¹, [PMS] = 0.5 g L⁻¹, T = 298 K, and initial pH = 7.0.To examine the enhanced catalytic activity, radical quenching experiments were conducted. As shown in Fig. 4a, when NaN₃ is added to the reaction solution as a scavenger for ¹O₂, there is no significant reduction of BPA decomposition, implying that non-radicals are not the dominant reactive species. By comparison, when tert-butanol (TBA) (radical scavenger for ˙OH) is added, there is a slight (2.8%) decrease in BPA removal. However, if methanol (radical scavenger for SO₄˙⁻ and ˙OH) is added, the efficiency of BPA degradation declines by up to 59.2%, indicating that the major radicals generated from the PMS activation are SO₄˙⁻;³³ the presence of these radicals is also verified by electron paramagnetic resonance (EPR) (Fig. 4b). Furthermore, the significant inhibition ratio can be observed when KI (quencher for the surface) is added, demonstrating that BPA degradation is mainly attributed to reactions with SO₄˙⁻, which is produced by a surface catalytic process.³⁴Open in a separate windowFig. 4(a) Effects of the radical scavengers on BPA degradation. (b) EPR spectra of SO₄˙⁻ and ˙OH. (c) The energy profiles of PMS on FeCo@N–C nano-necklaces surface. (d) Optimized configurations of PMS adsorbed on FeCo@N–C nano-necklaces.Density-functional theory was applied to calculate the surface energy of PMS activation at FeCo dual sites (Fig. 4c, d and S12^†). The dissociation barrier of PMS into SO₄˙⁻ and OH⁻ is −2.25 eV, which is much lower than that on an Fe or Co single site, suggesting that cleavage of O–O bonds of PMS occurs more easily on FeCo dual sites. This is because FeCo dual sites provide two anchoring sites for the dissociated O atoms, leading to more efficient activation of O–O. The FeCo@N–C nano-necklaces can reduce the energy barrier of O–O bond breaking, which results in high activity for PMS activation and thus high productivity of SO₄˙⁻.     

Monitoring single Au38 nanocluster reactions via electrochemiluminescence

Herein, we report for the first time single Au₃₈ nanocluster reaction events of highly efficient electrochemiluminescence (ECL) with tri-n-propylamine radicals as a reductive co-reactant at the surface of an ultramicroelectrode (UME). The statistical analyses of individual reactions confirm stochastic single ones influenced by the applied potential.

Herein, we report for the first time single Au₃₈ nanocluster reaction events of highly efficient electrochemiluminescence (ECL) with tri-n-propylamine radicals as a reductive co-reactant at the surface of a Pt ultramicroelectrode (UME).

Single entity measurements have been introduced by Bard and Wightman based on the collisions/reactions of individual nanoparticles and molecules at an ultramicroelectrode (UME).^1–9 Since then, the field of single entity electrochemistry has gradually attracted several research groups and has become a frontier field of nanoelectrochemistry and electroanalytical chemistry.^8,10–14 For instance, it has been shown that the chemistry of the electrode surface plays an important role in the collision/reaction events and the kinetics of reaction processes.^15–21 Dasari et al. reported that hydrazine oxidation and proton reduction can be detected using single Pt nanoparticles on the surface of a mercury or bismuth modified Pt UME, and the material of the electrode was found to affect the shape of current–time transients.^22,23 Fast scan cyclic voltammetry provides better chemical information about transient electrode–nanoparticle interactions, which is otherwise difficult to obtain with constant-potential techniques.²⁴ There are only a few reports on photoelectrochemical systems including semiconductor nanoparticles designed to detect single nanoparticles in the course of photocatalysis processes.^25–28 More importantly, owing to the nature of stochastic processes of single entity reactions, statistical analyses have shown substantial influences on the understanding of the underlying processes.Electrochemiluminescence or electrogenerated chemiluminescence (ECL),²⁹ as a background-free technique,^30–32 was also utilized to detect individual chemical reactions and single Pt nanoparticle collisions based on the reaction between the Ru(bpy)₃²⁺ complex and tri-n-propylamine (TPrA) radicals on the surface of an ITO electrode.^2,33,34 It was found that the size of the nanoparticles, the origin of the interaction between particles and the electrode surface, the concentration of species generation, and the lifetime of individual electrogenerated nanocluster species (i.e., Au₃₈²⁺, Au₃₈³⁺, and Au₃₈⁴⁺) in conjunction with the reactivity of those oxidized species with co-reactant radical intermediates (i.e., TPrA radical) play crucial roles in the frequency of the ECL reaction events leading to individual ECL responses. More strikingly, a higher ECL reaction frequency is directly proportional to the amount of collected ECL light.²¹ Chen and co-workers also employed ECL to study stationary single gold-platinum nanoparticle reactivity on the surface of an ITO electrode.³⁵ Lin and co-workers monitored the hydrogen evolution reaction in the course of “ON” and “OFF” ECL signals.³⁶ Recently, we performed a systematic and mechanistic ECL study of a series of gold nanoclusters, with the general formula of Au_n(SC₂H₄Ph)_m^z (n = 25, 38, 144, m = 18, 24, 60 and z = −1, 0, +1), where near-infrared (NIR) ECL emission was observed.³⁷ There are several enhancement factors, such as catalytic loops^38,39 that improve the signal to noise ratio. The Wightman group was able to report single ECL reactions based on the capability of ECL.⁷ Furthermore, thus far, we have explored ECL mechanisms and reported the ECL efficiency of five different gold nanoclusters i.e., Au₂₅(SR)₁₈¹⁻, Au₂₅(SR)₁₈⁰, Au₂₅(SR)₁₈¹⁺, Au₃₈(SR)₂₄⁰ and Au₁₄₄(SR)₆₀⁰, among which the Au₃₈(SR)₂₄⁰/TPrA system revealed outstanding ECL efficiency, ca. 3.5 times higher than that of Ru(bpy)₃²⁺/TPrA as a gold standard. Therefore, we decided to focus on the Au₃₈ (SR)₂₄⁰/TPrA system. It was discovered that the ECL emission of these nanomaterials can be tuned through varying the applied potential and local concentration of the desired co-reactant.Herein, for the first time we report on ECL via a single Au₃₈(SC₂H₄Ph)₂₄ nanocluster (hereafter denoted as Au₃₈ NC) reaction (eq. (1)) in the vicinity of an UME in the presence of TPrA radicals as a reductive co-reactant.1where x is the oxidation number that can be either 0, 1, 2, 3 or 4. Single ECL spikes (Fig. 1A) along with ECL spectroscopy were used for elucidating individual reaction events. Indeed, each single ECL spike demonstrates a single Au₃₈^(x−1)* reaction product. Au₃₈ NCs were synthesized according to procedures reported by us and others, and fully characterized using UV-Visible-NIR, photoluminescence, ¹HNMR spectroscopy and MALDI mass spectrometry to confirm the Au₃₈ nanocluster synthesis (details are provided in ESI, Sections 1–3, Fig. S1–S4^†).^38,40,41Fig. 2 (left) shows a differential pulse voltammogram (DPV) in an anodic scan of a 2 mm Pt disc electrode immersed in 0.1 mM Au₃₈ acetonitrile/benzene solution containing 0.1 M TBAPF₆ as the supporting electrolyte. There are five discrete electrochemical peaks at which Au₃₈⁰ was oxidized to Au₃₈⁺ (E°′ = 0.39 V), Au₃₈²⁺ (E°′ = 0.60 V), and Au₃₈^3+/4+ (E°′ = 0.99 V) and reduced to Au₃₈⁻ (E°′ = −0.76 V) and Au₃₈²⁻ (E°′ = −1.01 V).^38,40,41Open in a separate windowFig. 1(A) An example of the reaction event transient of 10 μM Au₃₈ in benzene/acetonitrile (1 : 1) containing 0.1 M TBAPF₆ in the presence of 20 mM TPrA at 0.9 V vs. SCE, acquired at 15 ms time intervals using a 10 μm Pt UME. The white dashed-line indicates the threshold to identify single ECL spikes. (B) Illustration of a single nanocluster ECL spike. (C) ECL instrumentation with an inset showing ECL spike generation in the vicinity of the Pt UME.Open in a separate windowFig. 2Anodic DPV for Au₃₈ (left), reaction energy diagram of Au₃₈²⁺ and TPrA· (middle) along with the ECL–voltage curve (right) in an anodic potential scan at a 2 mm Pt disk electrode immersed in a solution of 10 μM Au₃₈ with 20 mM TPrA.The rich electrochemistry of Au₃₈ NCs is well-matched with that of co-reactants such as TPrA to generate near infrared-ECL (NIR-ECL), and the ECL emission efficiency of the Au₃₈/TPrA system is 3.5 times larger than that of the Ru(bpy)₃²⁺/TPrA co-reactant ECL system.²⁷Thus, it is of utmost interest to investigate the ECL generation of the above co-reactant system in single reactions, which improves the ECL signal detection sensitivity. To perform the ECL experiment a solution of 10 μM Au₃₈ NC with 20 mM TPrA was prepared. We first confirmed the ECL light generation of such solution along with its blank solution containing only TPrA using a typical 2 mm diameter Pt disk electrode (Fig. 2, S5 and S6^†).A 10 μm Pt UME electrode, which is electrochemically inert (Fig. S7^†), was utilized to investigate the ECL of single NC reactions under potentiostatic conditions, at which a specific positive bias potential was applied to oxidize both Au₃₈ and TPrA. Fig. 1A shows a typical ECL–time transient current curve (ECL intensity versus time) at 0.90 V vs. SCE, which was acquired using a photomultiplier tube (PMT, R928) for a duration of 1800 s at data acquisition time intervals of 15 ms (Fig. 1C and ESI, Section 3^†). Fig. 1B represents an exemplary event of a single ECL spike with a sharp increase followed by a decay in the ECL intensity. It is observed from the many spikes in Fig. 1B that this process can reoccur with a high probability in the vicinity of the UME, probably due to an electrocatalytic reaction loop (Fig. 1C). Indeed, ECL intensity was enhanced in this way as an already relaxed species, i.e., Au₃₈^z+1*, participates in an oxidation step to regenerate Au₃₈^z+1 to react with the TPrA radical (TPrA˙).Once photons resulting from the excited state relaxation in the vicinity of the UME are captured by the PMT, individual reaction events can be observed (Fig. 1A with the instrumentation schematic shown in Fig. 1C). As shown in Fig. 3A, there are many ECL spikes during 1800 s of measurement, each of which represents an individual ECL generation reaction in the vicinity of the UME surface. It is worth noting that there are several spikes with various intensities. This is most likely due to the Brownian motion which is random movement due to the diffusion of individual nanocluster species such as Au₃₈⁰, Au₃₈¹⁺, Au₃₈²⁺, etc., electrogenerated at the local applied potentials. Long and co-workers⁴² proposed that silver nanoparticle collision on the surface of a gold electrode follows Brownian motion, leading to several types of surface-nanoparticle response peak shapes. In fact, the observed ECL spikes, shown in Fig. 1C, with a rise and an exponential decay suggested that Au₃₈ nanocluster species diffuse directly through the electrode double-layer, move towards the tunneling region of the electrode surface, collide⁴² and become oxidized, react with TPrA radicals thereafter to produce excited states, and emit ECL. It is worth emphasizing that this path could be partially different for each individual nanocluster owing to the angle and direction relative to the electrode surface. The single Au₃₈ NC reaction behaviour at various bias potentials was investigated following the electrochemical energy diagram shown in Fig. 2, middle. For example, at a bias potential of 0.70 V (the green spot on the DPV in Fig. 2), Au₃₈⁰ undergoes two successive oxidation reactions to Au₃₈²⁺ and TPrA oxidation and deprotonation start to generate TPrA·. In fact, at a very close oxidation potential to Au₃₈²⁺, TPrA is also oxidized to its corresponding cation radical (ca. 0.80 V vs. SCE) Fig. S6,^† followed by deprotonation to form the TPrA radical.³⁸ The TPrA· with a very high reduction power (E°′ = −1.7 eV)⁴³ injects one electron to the LUMO orbital of the nanocluster and forms excited state Au₃₈⁺*, as illustrated in the reaction energy diagram in Fig. 2, middle.³⁸ Then, Au₃₈⁺* emits ECL light while relaxing to the ground state. For another instance, at 1.10 V vs. SCE (the red spot on the DPV in Fig. 2), Au₃₈⁰ is oxidized to Au₃₈^3/4+ feasibly. At this potential, the TPrA radical is generated massively in the vicinity of the electrode. The efficient electron transfer between the TPrA radical and Au₃₈^3/4+ generates both Au₃₈²⁺* and Au₃₈³⁺* that emit light at the same wavelength of 930 nm.³⁸ The results of such interactions produced a transient composed of many ECL events (Fig. 3A), which is an indication of bias potential enforcement on the nanocluster light emission.Open in a separate windowFig. 3Single-nanocluster ECL photoelectron spectroscopy of Au₃₈. ECL–time transients (A), statistics of the number of photons (B), histogram of the single reaction time between sequential spikes (C) and accumulated ECL spectrum (D) for a 10 μm Pt UME at 1.1 V immersed in a 10 μM Au₃₈ nanocluster solution in benzene/acetonitrile (1 : 1) containing 0.1 M TBAPF₆ in the presence of 20 mM TPrA. (E)–(H) The counterpart plots to (A)–(D) for the UME biased at 0.7 V. # represents the number.We further tried to collect the current–time traces of such events; however, owing to the high background current originating from the high concentration of TPrA relative to that of the nanocluster, no noticeable spikes in the current were observed.In order to study the photochemistry and understand deeply the single nanocluster reactions, ECL–time transients were collected at different applied potentials (i.e., 0.7, 0.8, 0.9 and 1.1 V vs. SCE) as labelled in green, brown, purple, and red on the DPV in Fig. 2, respectively. The transients were further analysed using our home-written MATLAB algorithm adapted from that for nanopore electrochemistry.⁴⁴ The population of individual events was identified by applying an appropriate threshold to discriminate ECL spikes from the noise as demonstrated in Fig. 1A. In fact, the applied algorithm also assisted us to learn about the raising time and intensity of each spike, as well as photons of individual spikes. For instance, Fig. 3A shows another typical transit for 1800 s at an UME potential bias of 1.1 V for the ECL events. Indeed, the integrated area of each peak, the charge of the photoelectrons at the PMT, is directly proportional to the number of photons emitted from individual reactions (see ESI, Section 5^†). Basically, the PMT amplifies the collected single photon emitted in the course of light-to-photoelectron conversion (see ESI, Section 6 and Fig. S8^†) and translates a single photon into photoelectrons. The extracted charge of each ECL reaction, Q_ECL, was then converted to the corresponding number of photons by dividing by the gain factor, g, which is 1.55 × 10⁶ (Fig. S8^†), following eqn (2):2The histograms of the number of photons show a Gaussian distribution (Fig. 3B) with a reaction frequency of 53.5 ± 2.9 at E = 1.1 V, whereas at a lower potential of 0.7 V the reaction frequency drops to 18.5 ± 1.7 (Fig. 3F). This indicates that there is a three-fold lower reaction occurrence at the lower potential. The integration of the Gaussian fitting at 1.1 V and 0.7 V also reveals a three-fold drop from 3.3 × 10⁵ to 1.2 × 10⁵ photons over 1800 s.To further explore the effect of electrode potential bias on the single Au₃₈ NCs ECL reaction, potentials lower than 1.1 and higher than 0.7 V, ca. 0.8 and 0.9 V (brown and purple labels in Fig. 2), were applied. In fact, the resulting ECL–time transients show a lower population of single spikes (Fig. S12A and ESI,^†). The integrated Gaussian curve values support the ECL–time transient observations with ∼4.1 × 10⁴ and ∼6.5 × 10⁴ photons, respectively. In fact, it is unlikely that the PMT would get more than two events in the duration, owing to the following reasons: (i) it has been shown that only 5.5% of incoming photons can be effectively converted to photoelectron signals by our R928 PMT during our absolute efficiency calibration, ESI Section 6 and Fig. S8–S19^†;⁴⁵ (ii) spherical ECL emission is proven to be detected for a substantial small part upon examination of our detection system for the absolute ECL quantum efficiency;⁴⁵ (iii) Au₃₈ nanocluster ECL emissions occur at 930 nm, which is almost at the wavelength detection limit of our PMT response curve.^38,45In addition, we evaluated the stochastic (a series of random events at various probability distributions) nature of the observed events and extracted the reaction time interval (τ) at various potentials. The resulting graph shows an exponential decay (Fig. 3C) as expressed in eqn (3):3where frequency (λ) gives the mean rate of the event and A represents the fitting amplitude. One can expect to obtain the distribution of the number of emitted photons and spatial brightness function. In fact, the exponential decay is a clear indication of random single reaction events as Whiteman and co-workers described for a 9,10-diphenylanthracene (DPA) ECL system in the annihilation pathway.^7,46 At a potential of 1.1 V, λ and A are found to be 4.98 ± 0.02 ms⁻¹ and 80.4 ± 3.2, whereas at 0.7 V, λ and A turned out to be 32.9 ± 1.6 ms⁻¹ and 9.5 ± 0.1 (Fig. 3C and G). Indeed, the lower potential of 0.70 V vs. SCE is high enough to generate the TPrA radical along with Au₃₈²⁺, thereby leading to excited Au₃₈⁺*, Fig. 3E. One can conclude that at the applied potentials of 0.7 V and 1.1 V, Au₃₈⁰ is oxidized to Au₃₈²⁺ and Au₃₈⁴⁺, resulting in the generation of Au₃₈⁺* and Au₃₈³⁺* under static conditions. Thus, there are higher populations of ECL spikes with no discrepancy in the number of collected photon distributions. However, at two intermediate potentials, i.e., 0.8 and 0.9 V, a dynamic behaviour which is due to the mixed oxidation of Au₃₈ species, in the vicinity of the UME, is observed. In fact, at these two applied potentials, the local concentration of the corresponding gold nanoclusters (i.e., Au₃₈³⁺ and Au₃₈⁴⁺) is not sufficient to produce significant ECL spikes. We also attempted to collect the ECL spectrum using a charge-coupled device (CCD) camera, which is relatively more sensitive in the NIR region (e.g., λ > 900 nm, Fig. S16^†). Fig. 3D and H display an accumulated spectrum at 1.1 and 0.7 V vs. SCE, which is collected for 30 minutes. The fitted accumulated ECL spectrum indicates an ECL peak emission at 930 nm and supports higher reactivity at 1.1 V than that at 0.7 V.³⁸ To confirm that the observed ECL spikes and accumulated spectra are generated based on the oxidation of Au₃₈ nanoclusters in the presence of TPrA radicals, ECL–time transients were recorded upon holding an applied potential at which no faradaic process occurs. Fig. S11^† represents ECL–time curves and accumulated ECL spectra at 0.0 V and 0.4 V. One can notice that no appreciable ECL signal can be observed.In addition, we investigate the Pearson cross-correlation (ρ) between the intensities of ECL spikes with τ as shown in Fig. S14^† in which there is a positive correlation at 0.7 and 1.0 V and a negative correlation at 0.8 and 0.9 V. In fact, ρ evaluates whether there is a stationary random process between the two defined parameters (see ESI, Section 6^†). Interestingly, the frequency of the reaction at different applied potentials revealed decay from 0.7 to 0.8 V, followed by an upward trend to 0.9 and 1.1 V vs. SCE (Fig. S15^†). This could be additional support for the transition stage at 0.8 and 0.9 V, where the applied potential as the major driving force to generate oxidized forms (e.g., Au₃₈³⁺ and Au₃₈⁴⁺) governs the flux of the nanocluster species that reach the vicinity of the electrode. Furthermore, the effectiveness of electron transfer reaction kinetics between the radical species, i.e., Au₃₈^z+1 and TPrA radical, competes with the flow of the incoming nanoclusters. It is worth mentioning that each of the ECL single event experiments was repeated three times, and very similar results were obtained. Moreover, lower (5 μM) and higher (20 μM) concentrations of Au₃₈ in the presence of 20 mM were tested. In fact, the former shows a smaller number of single reactions; however the later revealed a larger number of multiple reactions (Fig. S13^†).In summary, in this communication we demonstrated that Au₃₈ NC ECL at the single reaction level can be monitored using a simple photoelectrochemical setup following a straightforward protocol. Indeed, we have rich basic knowledge about the ECL mechanisms of various gold nanoclusters with different charge states (Au₂₅(SR)₁₈¹⁺, Au₂₅(SR)₁₈⁰, Au₂₅(SR)₁₈¹⁻) and various sizes (Au₂₅(SR)₁₈⁰, Au₃₈(SR)₂₄⁰, Au₁₄₄(SR)₆₀⁰) in fine detail. Thus, the ECL emission mechanisms of gold clusters, including the contribution of each charge state and influence of various concentrations of co-reactants, are well known. For instance, in our previous studies^{38,39,47–49} we clearly identified three charge states of an Au₂₅(SR)₁₈¹⁻/TPrA system and we discovered that at a high concentration of TPrA the reduction in the bulk solution of gold nanoclusters influences the ECL emission wavelength. We also have learnt that the Au₃₈/TPrA system is a co-reactant independent of co-reactant concentration. Furthermore, an extensively higher concentration of TPrA provides a dominant reaction over any unknown decomposition reaction at higher oxidation states of Au₃₈. It was discovered that the population of ECL reactions is directly governed by the applied bias potential on a Pt UME. This work is a strong indication of the high sensitivity of the ECL technique in detecting single ECL reactions in a simple solution, which complements those reported by the Bard group using rubrene, for instance, embedded in an organic emulsion in the presence of TPrA or oxalate as a co-reactant.^50,51 These systems needed a substantial ECL enhancement in the presence of an ionic liquid as the supporting electrolyte and emulsifier. The current approach can be further extended to investigate other molecules and nanomaterials'' electrocatalytic processes at the single reaction level.     

Photostable polymorphic organic cages for targeted live cell imaging

Fluorescent microscopy is a powerful tool for studying the cellular dynamics of biological systems. Small-molecule organic fluorophores are the most commonly used for live cell imaging; however, they often suffer from low solubility, limited photostability and variable targetability. Herein, we demonstrate that a tautomeric organic cage, OC1, has high cell permeability, photostability and selectivity towards the mitochondria. We further performed a structure–activity study to investigate the role of the keto–enol tautomerization, which affords strong and consistent fluorescence in dilute solutions through supramolecular self-assembly. Significantly, OC1 can passively diffuse through the cell membrane directly targeting the mitochondria without going through the endosomes or the lysosomes. We envisage that designing highly stable and biocompatible self-assembled fluorophores that can passively diffuse through the cell membrane while selectively targeting specific organelles will push the boundaries of fluorescent microscopy to visualize intricate cellular processes at the single molecule level in live samples.

In this article, we demonstrate the relatively unexplored potential of organic cages for use in targeted live cell imaging and highlight the importance of inter- and intramolecular interactions to stabilize and improve the performance of fluorophores.

Organic fluorophores are of great interest in many research fields especially live cell imaging, which allows for noninvasive observation and monitoring of biological processes. Small molecule fluorophores can be functionalized and tuned with high precision to tag a diverse variety of biological targets; however, limited water solubility, cell-permeability, photostability and targetability have halted the biomedical translation of many promising fluorescent molecules.¹ Ultimately, new generations of fluorescent tags were designed to include new functionalities that can improve their physical properties and cell penetration such as long aliphatic side chains and cell penetrating peptides.^2–6 In addition, various strategies have been employed to improve the photostability of these compounds by the addition of electron withdrawing groups, conformational changes, encapsulation or incorporation within host macrocycles.^7–13Major efforts in the organic fluorophores development have been directed towards designing stable probes targeting the mitochondria as it plays a central role in the generation of adenosine triphosphate (ATP), central metabolism, and apoptosis.^14,15 Successful examples of mitochondria targeting fluorophores include triphenyl phosphonium cations, heterocyclic aromatic cations, macrocyclic amphiphiles, BODIPY derivatives and mitochondria targeted peptides that have limited solubility.^16–21 Commercially available dyes for mitochondrial imaging such as Mito Tracker Red have a high photobleaching tendency and can form covalent bonds with the mitochondrial respiratory chain complex I resulting in increased cytotoxicity. Consequently, various organic, inorganic and hybrid platforms were developed ranging from small nanoparticles to self-assembled frameworks on the quest for improved mitochondrial imaging agents.^22–27Porous organic cages have received major attention over the last decade as they are easily prepared, soluble in a range of solvents and are intrinsically porous.^28–31 They have shown excellent applicability in molecular recognition, sensing and hydrocarbon separation but their biomedical imaging applications have never been investigated.^32–35 Recently, self-assembled organic imine stacks showed impressive electroluminescence and live cell imaging.³⁶ Moreover, an intracellular targeted macrocyclic nanohoop showed fast cell uptake and two-photon live cell fluorescence imaging.³⁷ In this work, a tautomeric organic cage (OC1) that shows unprecedented photostability for live-cell imaging with high biocompatibility, cell permeability and mitochondrial targetability is reported (Fig. 1). We hypothesize that the keto–enol tautomerism is playing a major role in promoting the stability and sustaining the fluorescence of this platform. Such effect has been previously reported in other organic dyes such as oxyluciferin and HDFL.^38,39Open in a separate windowFig. 1Tautomeric forms of OC1 and chemical structures of OC2 and OC3.Investigating other organic cages with less hydroxyl groups (OC2) or replacing the hydroxyl groups with methoxy substituents (OC3) resulted in the loss of cell stability, retention and targetability (Fig. 1). The presence of the –OH groups in OC1 not only enables keto–enol tautomerization but also promotes H-bonding interactions, which ultimately improves solubility and cell permeability. We envision that understanding the nature of the intra- and intermolecular interactions in organic probes can lead to a profound enhancement in the photobleaching efficacy in addition to cell permeability. OC1 can effectively image the mitochondria over 3 h with high selectivity and without any loss of efficacy (no photo bleaching), which is comparable to the commercially used MitoTracker RedFM.OC1, a [2 + 3] cage, was synthesized following a reported procedure and its structure was confirmed by ¹H Nuclear Magnetic Resonance (NMR) and TOF-Mass analysis (Fig. S1–S4^†).³² Dynamic light scattering (DLS) of OC1 revealed an average size of 30–50 nm in DMSO–H₂O (2–5%) with a negative zeta potential of −5 mV (Fig. S5^†). Single crystal X-ray analysis revealed that OC1 crystallizes in a triclinic crystal system with chiral P1 space group and the asymmetric unit consists of two molecules of [2 + 3] cages (Fig. S6^†). Crystal structure analysis suggests that the imine cage crystallizes as the keto tautomeric form (Fig. 2a). This results in the formation of intramolecular N–H⋯O hydrogen bonding between the imine (N–H) hydrogen and the keto (C O) oxygen (d_{C–OH actual} = 1.36 Å; d_{C O actual} = 1.25–1.26 Å vs. d_{C O for OC1} = 1.260 Å and d_{C N actual} = 1.25 Å vs. d_{C–NH for OC1} = 1.30–1.35 Å) (Fig. S6^†). Three different windows of each cage are connected with three different cages in the same plane through weak C–H⋯O hydrogen bonding (d = 2.414 Å, and θ = 60°), which creates isolated pores at the center of the cage (Fig. 2b). Furthermore, the packing diagram suggests the formation of one-dimensional infinite extrinsic pores along the b-axis (Fig. 2c and S6^†). OC1 is stable in aqueous solutions even at low pH as well as in cell media (Fig. S7–S10^†).Open in a separate windowFig. 2(a) SCXRD and crystal image of OC1. (b) OC1 self-assembled through hydrogen bonding interactions. (c) Packing of OC1 along b-axis.To test our original proposal that organic cages can act as effective imaging agents, we first investigated the UV-visible spectrum of OC1 in DMSO, which exhibited two characteristic peaks at λ_max = 286 and 330 nm that can be assigned to π–π* and n–π* transitions, respectively (Fig. S11^†). Fluorescence emission spectra showed that OC1 exhibits two major peaks at around 411 and 497 nm upon excitation at 330–350 nm in DMSO, while excitation at 400 and 480 nm showed characteristic emission peaks at 503 and 538 nm, respectively (Fig. S12^†). This behavior can be explained by the presence of two emissive forms due to the keto–enol tautomerization. The quantum yield (QY) of OC1 in DMSO (0.25) and chloroform (0.4) was also tested. Molar extinction coefficient of OC1 was estimated as 46 806 M⁻¹ cm⁻¹ in DMSO (Fig. S13–S14^†). The quantum yield of OC1 in DMSO are about half that of in chloroform that is significantly matched with the reported intracellular targeted macrocyclic nanohoop structure.³⁷ The fluorescent properties of OC1 were maintained at different pH values (4, 7 and 10) (Fig. S15^†). We hypothesize that this mode of assembly, as supported by SCXRD, promotes stability and prevents aggregation-induced quenching that is very common in small molecule dye fluorophores. The keto–enol tautomerism is boosting intra- and intermolecular hydrogen bonding interactions that can be contributing to the emissive characteristics of this system. In fact, the presence of the phenolic OH groups have been reported to have critical effects on the structural and spectroscopic properties of organic cages.^31,40We then proceeded to test OC1 for cell imaging by first testing the cytotoxicity of OC1 in MCF-7 cells (Fig. S16a^†). The cells were treated with a solution of concentrations ranging from 3.9–500 μg ml⁻¹ for 24 h. Cell death was measured using MTT assay demonstrating the excellent biocompatibility of OC1 with an IC50 of > 500 μg ml⁻¹. MCF-7 cells were then either co-incubated with OC1 and LysoTracker Red or MitoTracker RedFM, the cells are then fixed with 4% formaldehyde (w/v) and imaged on a confocal laser-scanning microscope (CLSM). Interestingly, OC1 showed strong colocalization in the mitochondria with a Pearson''s correlation coefficient of 0.9751 (Fig. S16b and c^†). Real time imaging of MCF-7 cells at different time intervals after co-internalization of OC1 and MitoTracker showed a clear signal overlap on a subcellular level (Fig. 3 and MOV1), which ultimately supports that OC1 can selectively target the mitochondria.Open in a separate windowFig. 3Live cell imaging of MCF-7 cells incubated with 250 μg ml⁻¹ of OC1 for 3 h showing the uptake and colocalization of OC1 (abs/em ∼495/519 nm) with the commercial probe MitoTracker RedFM (abs/em ∼581/644 nm) in the mitochondria (scale bar at 50 μm).These encouraging results piqued our interest to investigate other organic cages in order to have a better understanding of the mechanistic factors in play. As the tautomeric activity is well reported to affect the photophysical properties of organic compounds, we compared OC1 to an organic cage with less phenolic OHs (OC2) and an organic cage with only methoxy substituents (OC3). OC2 and OC3 were synthesized and characterized according to reported procedures (Fig. S17–S25^†).^32,41 Absorbance, emission properties and stabilities of OC2 and OC3 in DMSO were also verified. OC2 showed one major characteristic emission peak at around 547 nm upon excitation at 350 and 440 nm. The quantum yield (QY) of OC2 in DMSO (6.7%) and chloroform (11.1%) was also tested (Fig. S13 and S14^†). OC3 showed emission at 494 and 537 nm upon excitation at 320 nm and 480 nm upon excitation at 420 nm (Fig. S26^†). The quantum yield (QY) and molar extinction coefficient of OC3 in DMSO is negligible. Cytotoxicity and biocompatibility studies of OC2 and OC3 showed a relatively higher cytotoxicity compared to OC1 (Fig. S27^†). As to targetability and photostability, OC2 and OC3 showed less selectivity with colocalization in both mitochondria and lysosome where the fluorescent signal slowly disappeared, which implies the OC2 and OC3 are less stable than OC1 (Fig. S28^†). Moreover, FACS data revealed that OC1 showed the highest uptake percentage in comparison to OC2, and OC3 with an uptake percentage of 86.4%, 78%, and 69%, respectively (Fig. S29^†). Based on this data, we can conclude that indeed the hydroxyl groups, and their prompted keto–enol tautomerism, are playing a major role in stabilizing the overall structure through hydrogen bonding while simultaneously improving photostability and targetability. As a control, we tested the aldehyde starting material, 1,3,5-triformylphloroglucinol, to verify that OC1 is being uptaken as a whole and not as the aldehyde building block or a possible decomposition product. The experiment clearly demonstrates that 1,3,5-triformylphloroglucinol as a fluorophore had no selectivity towards the mitochondria and is initially distributed in the cytoplasm and finally localizes in the lysosomes for degradation (Fig. S30^†).As for the uptake mechanism, we initially hypothesized that OC1 will be taken up by cells through endocytosis, which is the typical biological pathway of nanoparticles and nanoassemblies.^42,43 Surprisingly, we discovered that OC1 can passively diffuse rather than being actively transported through the cell membrane. Passive transport is a type of membrane transport that does not require energy to move substances across cell membranes as it mainly relies on the second law of thermodynamics. We tested OC1 uptake in the presence of different endocytic inhibitors including chlorpromazine (CPZ), filipin (FIL), and amiloride to inhibit clathrin-mediated, caveolae mediated, and macropinocytotic endocytosis pathways, respectively.⁴⁴ All cells were treated with the inhibitors for 30 min, followed by 3 h incubation with OC1. The cellular uptake data showed that there was no significant reduction in the uptake using the active transport inhibitors (Fig. 4a). Moreover, the passive transport was tested by incubating OC1 (125 μg ml⁻¹) with MCF-7 for 3 h at different temperatures (4, 27, or 37 °C). The uptake was monitored by confocal microscopy (Fig. 4b) and quantified by FACS (Fig. 4c). Interestingly, no obvious reduction of uptake was observed in all the cases.Open in a separate windowFig. 4Cellular uptake mechanism of OC1. (a) Cell uptake of OC1 in the presence of endocytosis inhibitors to investigate endocytic uptake mechanism. (b) MCF-7 cell uptake CLSM of OC1 at 37 °C, 25 °C, and 4 °C, respectively (scale bar at 30 μm). (c) Quantification of cell fluorescence relative to 37 °C.After verifying the targetability and the uptake mechanism of OC1, we focused our attention on photostability. Since the commercially available mitochondrion tracker dyes normally have moderate photostability, it is critical to evaluate the photostability of OC1 in live cells. We first compared the photostability of OC1 with the commercially available MitoTracker RedFM. Since the excitation/emission wavelengths of OC1 (abs/em ∼495/519 nm) and MitoTracker RedFM (abs/em ∼581/644 nm) are distant from each other, we performed simultaneous scans for both dyes for 100 scans at 1.5 s per scan by incubating OC1 and MitoTracker RedFM with MCF-7 cells at 37 °C for 48 h. The results showed that at a laser power density of P ∼ 0.4132 W cm⁻²OC1 (ε = 46 806 M⁻¹ cm⁻¹) had higher retention and photostability than that of the MitoTracker RedFM where OC1 retained 54% of its initial fluorescence intensity while the fluorescence of MitoTracker RedFM dropped below 15% (Fig. 5a). Moreover, OC1 showed high photostability even at 72 h post-staining (Fig. 5b). Finally, it was noted that throughout the 72 h of incubation of OC1 with MCF-7, over three generations occurred in the cell culture and showed fluorescent labeling of the mitochondria suggesting that OC1 can be passed through to daughter cells upon cell division.Open in a separate windowFig. 5Intracellular retention and photostability of OC1. (a) Photostability of OC1 in MCF-7 cells co incubated with MitoTracker RedFM for 48 h. Plot of the sum intensity of each frame normalized to time 0 (frame 1) and plotted against time (seconds). Intensity data analysis was done using Python and plotted on Excel. (b) CLSM images (72 h post incubation) showing OC1 distribution while most of the MitoTracker RedFM signal was lost (Scale bar at 30 μm).     

Separation of pyrrolidine from tetrahydrofuran by using pillar[6]arene-based nonporous adaptive crystals

Pyrrolidine, an important feedstock in the chemical industry, is commonly produced via vapor-phase catalytic ammoniation of tetrahydrofuran (THF). Obtaining pyrrolidine with high purity and low energy cost has extremely high economic and environmental values. Here we offer a rapid and energy-saving method for adsorptive separation of pyrrolidine and THF by using nonporous adaptive crystals of per-ethyl pillar[6]arene (EtP6). EtP6 crystals show a superior preference towards pyrrolidine in 50 : 50 (v/v) pyrrolidine/THF mixture vapor, resulting in rapid separation. The purity of pyrrolidine reaches 95% in 15 min of separation, and after 2 h, the purity is found to be 99.9%. Single-crystal structures demonstrate that the selectivity is based on the stability difference of host–guest structures after uptake of THF or pyrrolidine and non-covalent interactions in the crystals. Besides, EtP6 crystals can be recycled efficiently after the separation process owing to reversible transformations between the guest-free and guest-loaded EtP6.

Here we offer a rapid and energy-saving method for adsorptive separation of pyrrolidine and tetrahydrofuran by using nonporous adaptive crystals of per-ethyl pillar[6]arene.

Pyrrolidine is an important feedstock in the chemical industry that has been widely used in the production of food, pesticides, daily chemicals, coatings, textiles, and other materials.¹ Particularly, pyrrolidine is a raw material for organic synthesis of medicines such as buflomedil, pyrrocaine, and prolintane.² Moreover, pyrrolidine is also used as a solvent in the semi-synthetic process of simvastatin, one of the best-selling cardiovascular drugs.³ In the chemical industry, there are many preparation methods for pyrrolidine. The most common way to obtain pyrrolidine is the gas-phase catalytic method using tetrahydrofuran (THF) and ammonia as raw materials;⁴ this is carried out at high temperature under catalysis by solid acids. However, separating pyrrolidine from the crude product is difficult because of similar molecular weights and structures between pyrrolidine (b.p. 360 K and saturated vapor pressure = 1.8 kPa at 298 K) and THF (b.p. 339 K and saturated vapor pressure = 19.3 kPa at 298 K), which result in complicated processes and large energy consumption.⁵ Therefore, it is worthwhile to find energy-efficient and simple methods to separate pyrrolidine from THF.Many techniques and materials, including porous zeolites, metal–organic frameworks (MOFs), and porous polymers, have facilitated energy-efficient separations of important petrochemicals and feedstocks, including THF and pyrrolidine.^6,7 However, some drawbacks of these materials cannot be ignored.⁸ For example, the relatively low thermal and moisture stabilities of MOFs limit their practical applications. Therefore, the development of new materials with satisfactory chemical and thermal stabilities for pyrrolidine/THF separation is of high significance.In the past decade, pillararenes have been widely studied in supramolecular chemistry.⁹ Owing to their unique pillar structures and diverse host–guest recognitions, pillararenes have been used in the construction of numerous supramolecular systems.¹⁰ Recently, nonporous adaptive crystals (NACs) of macrocycles, which have shown extraordinary performance in adsorption and separation, have been developed by our group as a new type of adsorption and separation materials.¹¹ Unlike MOFs, covalent-organic frameworks (COFs), and other materials with pre-existing pores, NACs do not have “pores“ in the guest-free form, whereas they adsorb guest vapors through cavities of macrocycles and spaces between macrocycles. NACs have been applied in separations of many significant chemicals such as alkane isomers, aromatics, and halohydrocarbon isomers.¹² However, such materials have never been used to separate pyrrolidine and THF. Herein, we utilized pillararene crystals as a separation material and realized the selective separation of pyrrolidine from a mixture of pyrrolidine and THF. We found that nonporous crystals of per-ethyl pillar[6]arene (EtP6) exhibited a shape-sorting ability at the molecular level towards pyrrolidine with an excellent preference, while crystals of per-ethyl pillar[5]arene (EtP5) did not (Scheme 1). In-depth investigations revealed that the separation was driven by the host–guest complexation between pyrrolidine and EtP6, which resulted in the formation of a more stable structure upon adsorption of pyrrolidine vapor in the crystalline state. EtP6 crystals can also adsorb THF. However, when these two chemicals simultaneously exist as the vapor of a 50 : 50 (v/v) mixture, EtP6 prefers pyrrolidine as an adsorption target. Compared with previously reported NAC-based separation, this separation took place rapidly. 95% purity was achieved in 15 min, and the purity increased to 99.9% after 2 h of separation. Moreover, pyrrolidine was removed upon heating, along with the structural transformation of EtP6 back to its original state, endowing EtP6 with excellent recyclability.Open in a separate windowScheme 1Chemical structures and cartoon representations: (a) EtP5 and EtP6; (b) THF and pyrrolidine.EtP5 and EtP6 were prepared as previously described and then a pretreatment process was carried out to obtain guest-free EtP5 and EtP6 (Fig. S1–S4†).¹³ According to powder X-ray diffraction (PXRD) patterns, activated EtP5 and EtP6 (denoted as EtP5α and EtP6β, respectively) were crystalline, and the patterns matched previous reports (Fig. S5 and S6^†).¹⁴ Studies from our group indicated that EtP5α and EtP6β crystals were nonporous, presumably due to their dense packing modes.We first investigated the adsorption capabilities of EtP5α and EtP6β towards pyrrolidine and THF vapors. Based on time-dependent solid–vapor adsorption procedures, both EtP5α and EtP6β showed good ability to adsorb pyrrolidine and THF vapors. As shown in Fig. 1a, the adsorption amount of THF in EtP5α was higher than that of pyrrolidine. It took 6 hours for EtP5α to reach saturation points for adsorption of both pyrrolidine and THF vapors. The final storage of THF in EtP5α was 2 : 1 (molar ratio to the host), whereas the storage of pyrrolidine was 1 : 1. It seemed that the THF vapor was favored to occupy EtP5α, which was ascribed to the relatively lower boiling point of THF. A similar phenomenon was found for EtP6β. Time-dependent solid–vapor adsorption experiments for pyrrolidine demonstrated that it took just 1 hour to reach the saturation point, while it took 4 hours for the THF vapor (Fig. 1b). The adsorption amount of THF vapor was twice that of pyrrolidine. ¹H NMR spectra and thermogravimetric analyses (TGA) further confirmed the adsorption and storage of THF and pyrrolidine in both hosts (Fig. S7–S16†). Meanwhile, in the desorption process, adsorbed pyrrolidine and THF in EtP6β were easily released under reduced pressure and heating. Based on these data, it was clear that pyrrolidine could be adsorbed rapidly by both EtP5α and EtP6β in molar ratios = 1 : 1, while THF could be captured in a relatively slow process. Structural changes after adsorption of these two vapors were analyzed via PXRD experiments, in which varying degrees of changes before and after adsorption were observed, evidencing the appearance of new crystal structures (Fig. 1c and d). Nevertheless, only slight differences were observed in the PXRD patterns after the adsorption of THF or pyrrolidine, which might be ascribed to the structural similarity of the two molecules.Open in a separate windowFig. 1Time-dependent solid–vapor adsorption plots of (a) EtP5α and (b) EtP6β for single-component pyrrolidine and THF vapors. PXRD patterns of (c) EtP5α and (d) EtP6β: (I) original activated crystals; (II) after adsorption of THF vapor; (III) after adsorption of pyrrolidine vapor.To study the mechanism of adsorption, guest-loaded single crystals were obtained by slowly evaporating either THF or pyrrolidine solutions of pillararenes (Tables S2 and S3^†). In the crystal structure of THF-loaded EtP5 (2THF@EtP5, Fig. 2a and S17^†),^11a two THF molecules are in the cavity of one EtP5 molecule driven by multiple C–H⋯O hydrogen bonds and C–H⋯π bonds. EtP5 assembles into honeycomb-like infinite edge-to-edge 1D channels. In the crystal structure of pyrrolidine-loaded EtP5 (pyrrolidine@EtP5, Fig. 2b and S19^†), one pyrrolidine molecule, stabilized by C–H⋯π interactions and C–H⋯O hydrogen bonds between hydrogen atoms on pyrrolidine and oxygen atoms on EtP5, is found in the cavity of EtP5. It''s worth mentioning that a hydrogen atom which is linked with the N atom of pyrrolidine also forms a strong hydrogen bond with an oxygen atom on the ethoxy group of EtP5. EtP5 forms imperfect 1D channels because of partial distortion of orientation. The PXRD patterns simulated from these crystal structures matched well with the experimental results (Fig. S18 and S20^†), which verified that the uptake of vapors transformed EtP5α into pyrrolidine-loaded EtP5.Open in a separate windowFig. 2Single crystal structures: (a) 2THF@EtP5; (b) pyrrolidine@EtP5.In the crystal structure of THF-loaded EtP6 (2THF@EtP6, Fig. 3a and S21^†), one EtP6 molecule encapsulated two THF molecules in its cavity with C–H⋯O interactions, forming a 1 : 2 host–guest complex. Although 1D channels are observed, EtP6 adopts a slightly different conformation, caused by the presence of THF. Moreover, the PXRD pattern of EtP6β after adsorption of THF vapor matches well with that simulated from 2THF@EtP6, which is evidence for the structural transformation upon adsorption. In the crystal structure of pyrrolidine-loaded EtP6 (pyrrolidine@EtP6, Fig. 3b and S23^†), a 1 : 1 host–guest complex with pyrrolidine is found. Driven by C–H⋯π interactions and C–H⋯O hydrogen bonds formed by hydrogen atoms on pyrrolidine and oxygen atoms on EtP6, one pyrrolidine molecule is in the cavity of EtP6 with the nitrogen atom inside the cavity. The window-to-window packing mode of hexagonal EtP6 molecules in pyrrolidine@EtP6 contributes to the formation of honeycomb-like infinite edge-to-edge 1D channels, favorable for guest adsorption. Likewise, the PXRD result of EtP6β after adsorption of pyrrolidine is in line with the simulated pattern of pyrrolidine@EtP6, indicating that EtP6β transformed into pyrrolidine@EtP6 in the presence of pyrrolidine (Fig. S22 and S24^†).Open in a separate windowFig. 3Single crystal structures: (a) 2THF@EtP6; (b) pyrrolidine@EtP6.According to the adsorption ability and different crystal structures after adsorption of guest vapors, we wondered whether EtP5α or EtP6β could separate mixtures of THF and pyrrolidine. We first evaluated separation by EtP5α. GC analysis indicated that the adsorption ratios of THF and pyrrolidine were 65.7% and 34.3%, respectively, when EtP5α was exposed to 50 : 50 (v/v) pyrrolidine/THF mixture vapor (Fig. 4a and S25^†). Such adsorption was also illustrated by ¹H NMR (Fig. S26^†). Although EtP5α showed a preference for THF, the selectivity is not satisfactory and cannot be applied to industrial separation. The less satisfactory selectivity may be ascribed to the similar crystal structures of EtP5 after adsorption of THF or pyrrolidine and insufficient strong stabilizing interactions. The PXRD pattern of EtP5α after adsorption of the 50 : 50 (v/v) pyrrolidine/THF mixture vapor exhibited minor differences compared with that simulated from either 2THF@EtP5 or pyrrolidine@EtP5, due to poor selectivity (Fig. 4b).Open in a separate windowFig. 4(a)Time-dependent solid–vapor adsorption plot for EtP5α in the presence of 50 : 50 (v/v) pyrrolidine/THF mixture vapor. (b) PXRD patterns of EtP5α: (I) original EtP5α; (II) after adsorption of THF vapor; (III) after adsorption of pyrrolidine vapor; (IV) after adsorption of pyrrolidine/THF mixture vapor; (V) simulated from the single crystal structure of pyrrolidine@EtP5α; (VI) simulated from the single crystal structure of 2THF@EtP5α. (c) Time-dependent solid–vapor adsorption plot for EtP6β in the presence of 50 : 50 (v/v) pyrrolidine/THF mixture vapor. (d) PXRD patterns of EtP6β: (I) original EtP6β; (II) after adsorption of THF vapor; (III) after adsorption of pyrrolidine vapor; (IV) after adsorption of pyrrolidine/THF mixture vapor; (V) simulated from the single crystal structure of pyrrolidine@EtP6β; (VI) simulated from the single crystal structure of 2THF@EtP6β.Nevertheless, selective separation of THF and pyrrolidine was achieved with EtP6β. As shown in Fig. 4c, time-dependent solid–vapor adsorption experiments for a 50 : 50 (v/v) pyrrolidine/THF mixture were conducted. Unlike the phenomenon in single-component adsorption experiments, uptake of pyrrolidine by EtP6β increased and reached the saturation point rapidly (less than 2 hours), while capture of THF was negligible. According to the NMR and GC results (Fig. S27 and S28^†), the purity of pyrrolidine was determined to be 99.9% after 2 hours of adsorption, which indicates the remarkable selectivity of EtP6β for pyrrolidine. The PXRD pattern of EtP6β after adsorption of the mixture was consistent with that from single-component adsorption, indicating the structural transformation in the crystalline state upon selective capture of pyrrolidine from the mixture. Although THF and pyrrolidine have similar molecular structures, their non-covalent interactions with EtP6 are different. We assume that the hydrogen bond between N–H and the oxygen atom on EtP6 stabilizes pyrrolidine and leads to such selectivity. More importantly, compared with previous adsorption processes using NACs reported by our group, the selective separation of pyrrolidine was completed rapidly. According to the GC results, the purity of pyrrolidine reached around 95% in the initial 15 min, while it usually takes hours for selective separations of other substrates using NACs. Increasing the adsorption time to 2 h improves the purity to over 99%. The rapid separation of pyrrolidine with high purity using EtP6β shows great potential in industrial applications.Apart from selectivity, recyclability is also an important parameter for an adsorbent. Consequently, recycling experiments were carried out by heating pyrrolidine@EtP6 under vacuum at 100 °C to remove adsorbed pyrrolidine. According to TGA and PXRD analysis, the recycled EtP6 solid maintained crystallinity and structural integrity that were the same as those of activated EtP6 crystals (Fig. S29 and S30^†). Besides, it is worth mentioning that the recycled EtP6 solids were still capable of separating mixtures of pyrrolidine and THF without loss of performance after being recycled five times (Fig. S31^†).In conclusion, we explored the separation of pyrrolidine/THF mixtures using NACs of EtP5 and EtP6. Pyrrolidine was purified using EtP6 from a 50 : 50 (v/v) pyrrolidine/THF mixture with a purity of 99.9%, but EtP5 exhibited selectivity towards THF. Moreover, the separation of pyrrolidine by EtP6 was extremely fast so that over 95% purity was determined within 15 min of adsorption. The rapid separation is unique among NAC-based separations. Single-crystal structures revealed that the selectivity depended on the stability of the new structures after adsorption of the guests and the non-covalent interactions in the host–guest complexes. PXRD patterns indicated that the structures of the host crystals changed into the host–guest complexes after adsorption. Additionally, the NACs of EtP6 exhibited excellent recyclability over at least five runs; this endows EtP6 with great potential as an alternative adsorbent for rapid purification of pyrrolidine that can be applied in practical industry. The fast separation with such simple NACs in this work also reveals that minor structural differences can cause significant changes in properties, which should provide perspectives on designs of adsorbents or substrates with specifically tailored binding sites.     

A therapeutic keypad lock decoded in drug resistant cancer cells

A molecular keypad lock that displays photodynamic activity when exposed to glutathione (GSH), esterase and light in the given order, is fabricated and its efficacy in drug resistant MCF7 cancer cells is investigated. The first two inputs are common drug resistant tumor markers. GSH reacts with the agent and shifts the absorption wavelength. Esterase separates the quencher from the structure, further activating the agent. After these sequential exposures, the molecular keypad lock is exposed to light and produces cytotoxic singlet oxygen. Among many possible combinations, only one ‘key’ can activate the agent, and initiate a photodynamic response. Paclitaxel resistant MCF7 cells are selectively killed. This work presents the first ever biological application of small molecular keypad locks.

Information processing therapeutics with an implemented keypad lock logic gate selects input order for activation in drug resistant cancer cells.

The complex nature of diseases such as cancer necessitates smarter drugs that can discriminate each disease state or regulate drug efficacy spatially and/or temporally. With this intention, activatable drugs, drugs with on demand release properties are developed with promising selectivity.^1–4 Information processing therapeutics which are based on molecular logic gate operations are another approach to solve this problem.^5–7 Molecular logic gates are small compounds using Boolean logic operations to process inputs (i.e. the analyte concentration), and give an output as a result (fluorescence, and therapeutic activity etc.).⁸ Selective drug activation, release, multiple-analyte sensing and theranostic applications of these devices have been explored by us and others.^5,9–19Among the operations that can be carried out using small molecules, keypad locks provide an alternative application in information security.²⁰ This logic operation can give a specific output when the inputs are given in the correct form and correct sequence. For the device, each input is considered as an AND logic operation where the history of the process is also considered. A pioneering example was reported by Margulies and Shanzer in 2007 where energy transfer is modulated by chelation of Fe³⁺ in a pH dependent manner.²¹ Later, various other devices were introduced with advanced properties such as more than 2 input responsiveness and error detection capability.^22–24 All-photonic logic gates to address chemical waste production is extensively studied by Gust, Andréasson and Pischel.^25,26 Beside small molecule keypad locks, enzymes, antibodies, and DNA hybrids are used to achieve the same goal.^27–30 Although their potential use in molecular cryptology is highlighted, so far, there is no solid biological application of small molecule keypad locks.In the research presented here, a molecular keypad lock is developed which displays a photodynamic therapeutic output when a molecule is exposed to analytes in the correct order and type (PS3, Fig. 1). Two inputs of the system are chosen to be the common markers of drug resistant tumours: glutathione (GSH) and esterase enzyme (E). Cancer cells develop resistance to traditional chemotherapy in time by changing the protein expression or metabolite content of the cell. This adaptation of cancer cells is an obstacle for their treatment and needs to be addressed. Glutathione is a tripeptide used in reductive biochemical synthesis and it is known to be present in elevated levels in rapidly dividing cells such as cancer cells.³¹ A high GSH level is reported to contribute to drug resistance, since GSH adducts of the drugs are exported out of the cell much more rapidly.^32,33 Likewise, esterase enzyme activity is known to be associated with drug detoxification as this enzyme contributes to the chemical conversion of the drug.^34,35 Glutathione and esterase enzyme are chosen to be the first two inputs of the molecular keypad lock, the first two digits of the password. In the research, light is used as the final input. Although trivial, light is essential for photodynamic activity and spatiotemporal control of irradiation, further improving selectivity of the therapy.Open in a separate windowFig. 1Chemical structures of model photosensitizers (PS1 and PS2) and a molecular keypad lock (PS3). Ester bonds (red) are prone to hydrolysis by the esterase enzyme. Distyryl sites of the photosensitizers (blue) can react with thiol nucleophile provided that it is bound to an electron deficient group (i.e. pyri-dinium).Keypad lock PS3 is a photodynamic therapy (PDT) agent. PDT is a non-invasive method used for the treatment of surface cancers and certain other diseases ranging from atherosclerosis to macular degeneration.^36–39 In this therapy, a photosensitizer is excited with light, and produces cytotoxic singlet oxygen (¹O₂) thereby triggering apoptosis or necrosis of the cell, initiating an immune response and blocking microvasculature.⁴⁰ In the research, a boradiazaindecene (BODIPY) photosensitizer is used to benefit from versatile chemistry and spectroscopic properties.^41–45Near-IR absorbing PS3 shown in Fig. 1 is the molecular keypad lock and it is synthesized in 13 steps (Scheme S1^†). PS3 and model compound PS2 have heavy atoms on the structure to favour intersystem crossing required for transition to the triplet state and hence ¹O₂ generation occurs.⁴³ Ester bonds on the structure of PS3 are prone to cleavage by esterase enzyme. Distryryl bonds on the PS3 (blue) tend to reduce or form an adduct with thiol nucleophiles when it is activated by the pyridinium electron withdrawing group.⁴⁶ This property lies at the heart of sequential operation of esterase and GSH. When GSH reacts with electron poor double bonds, the extended conjugated structure is broken and PS3-a is generated (Fig. 2). This structure has absorption below 550 nm, like brominated core BODIPY molecules (compound 8, Scheme S1^†), and therefore can be excited with a green light. A quencher (green) is attached to ensure that photodynamic activity is OFF until esterase cleaves the ester bond. This is because of the energy transfer from the photosensitizer to this module, until esterase separates the photosensitizer. Since PS3 lacks absorption around the 500–550 nm region, it is inactive until GSH reacts with the compound. However, the GSH reacted photosensitizer does show absorption in this region; so, in order to avoid full activation just by GSH, a quencher module is attached. Spectral overlap between the BODIPY core (see the structure of compound 8 in the ESI,^† similar to that of PS3-a in terms of conjugation) and quencher (Q) can be seen from UV-Vis absorption and fluorescence spectra (Fig. 3 and S1^†). By this way, the photosensitizer is chemically modulated by GSH to ensure excitation, and then esterase enzyme inhibits energy transfer by removing the quencher. Lastly a green light is used to excite the photosensitizer leading to generation of photodynamic action. Since light is necessary for the final excitation of the molecule, it should always be the last input. If the order of esterase and GSH changes, as shown in Fig. 2, activation is not expected to take place since cleavage of the ester bonds generates 4-hydroxybenzyl derivative on PS3, which spontaneously faces 1,4-elimination to generate pyridine (Fig. S2^†).⁴⁷ Pyridine on its own is not sufficiently electron withdrawing to favour nucleophilic attack of double bonds by GSH and to activate it as demonstrated below. Therefore, the photosensitizer preserves extended conjugation and essentially lacks absorption at the wavelength of excitation.Open in a separate windowFig. 2Sequential operation of GSH and esterase. GSH can only react with BODIPY distyryl units when the structure has electron withdrawing pyridinium, either reducing it or forming an adduct. Esterase enzyme cleaves ester bonds, liberating the photosensitizer from the quencher module (green). Initial esterase activity converts the pyridinium unit to pyridine, thereby decreasing the reactivity of double bonds with GSH.Open in a separate windowFig. 3Normalized UV-Vis absorption and fluorescence spectra of PS1–3 in 2% water in THF (a and b). Samples are excited at 600 nm. Spectral changes of PS3 (10 μM) alone (black) or PS3 upon exposure to 0.5 mM GSH (c) and 10U esterase (d) for 90 min and 60 min at 37 °C, in 2% water in THF, respectively. A new peak at 544 nm appears upon incubation with GSH which is attributed to reduced PS3 and/orthe GSH-adduct. Esterase treatment increases the relative intensity of the shoulder peak around 600 nm.In order to understand the response of the PS3 to GSH, a molecule is incubated with 0.5 mM of GSH at 37 °C for 90 min. A new peak at 544 nm appears in UV-Vis absorption spectra consistent with the hypothesis (Fig. 3c, S1 and S9^†). The formation of the GSH adduct (PS3-a) is demonstrated by Liquid Chromatography Mass spectrometry analysis (Fig. S3^†). When control module PS1 is exposed to the same conditions, this new peak is not detected indicating that the pyridine bearing structure is neither activated enough for the nucleophilic substitution by GSH nor did it display PDT activity (Fig. S4 and S5^†). On the other hand, GSH treated pyridinium bearing PS2 immediately displayed a colour change indicative of broken conjugation (Fig. S6^†). When PS3 is incubated with esterase for 1 h, a small hypsochromic shift in the absorption peak is detected as a shoulder to the parent peak which is attributed to the conversion of pyridinium to pyridine (PS3-c, Fig. 3d). The control PS3 sample which is incubated under the same conditions but lacks esterase does not show an enhancement of this peak (Fig. 3d, black). High Resolution Mass Spectrometry analysis of the esterase treated PS2 samples confirm the hydrolysis of the ester and subsequent formation of the pyridine compound (Fig. S7^†). Esterase treated samples display an increase in the emission intensity when excited at 620 nm (Fig. S8^†). This is attributed to the initial quenching of the quencher module by the pyridinium photosensitizer. Analysis of the absorption and emission spectra suggest that the quencher module of PS3 can induce energy transfer to the pyridinium photosensitizer (Fig. 3). Once separated by esterase, fluorescence of the quencher module increases. In the case of GSH treated sample, a small enhancement in emission upon excitation at 500 nm is observed (Fig. S9^†). Note that the GSH adduct (or PS3 with reduced double bonds) has higher absorption at this wavelength, which would be the reason for the increase in emission intensity. In the spectral analysis organic solvents with a low water content are used to monitor the formation of water-insoluble, neutral, pyridine-bearing intermediate species.In the project, the molecular keypad lock is aimed to unlock in the presence of drug resistant tumour markers and get activated. Activation cannot take place when the input order differs. To demonstrate this, photodynamic action in the presence of all three inputs in a different order is investigated. ¹O₂ production can be followed by using trap molecule, 1,3-diphenylisobenzofuran (DPBF).⁴⁸ This molecule reacts with ¹O₂ and loses its absorption at 418 nm. The effect of different input combinations on the PDT action are given in Fig. 4. In the first 15 min, all samples are kept in the dark. Under such conditions no ¹O₂ generation is detected, which indicates lack of dark activity. DPBF is exposed to light from a LED source (peak 505 nm) under the same experimental conditions and no decrease in the absorption is detected. This control experiment eliminates the photodegradation of DPBF in the absence of a photosensitizer. Upon irradiation before the activation of the photosensitizer by GSH and esterase, no ¹O₂ generation is observed as expected. The results show that ¹O₂ generation, and the subsequent decrease in DPBF absorption, are significantly more in the input order of glutathione, esterase enzyme and light, consistent with the proposed mode of activation.Open in a separate windowFig. 4 ¹O₂ generation ability of PS3 (0.1 μM) when three inputs are given in a different order. All samples contain 50 μM of ¹O₂ trap molecule DPBF. In the first 15 minutes samples are kept in the dark. GSH is added in 0.5 mM concentration and incubated for 90 min at 37 °C. Samples are incubated with 10U esterase for 1 h at 37 °C. An LED light is irradiated from a 30 cm distance for 45 min.To analyse the effect of PDT action in the cell, a drug resistant cell line is generated. MCF7 cells are exposed to an increased dose of traditional cancer therapeutic agent paclitaxel as described in the literature.⁴⁹ When the spindle-shaped morphology is obtained following maximum drug dose application, cells are reported to have drug resistance. At this stage, PS3 is applied to both normal and drug resistant cells. When cell viabilities at various concentrations are analysed, it has been found that the light toxicity of PS3 is significantly enhanced in drug resistant cells (Fig. 5). The IC50 values of irradiated samples are calculated to be 124.8 μM for MCF7 cells. This value is reduced to 52.5 μM in paclitaxel resistant MCF7 (Pac-MCF7) indicating improved cytotoxicity in these cells. Efficient induction of apoptosis is also proved by Annexin V and PI staining (Fig. 6). Under dark conditions, cells do not have significant loss of viability. Upon irradiation, resistant cells are more prone to apoptosis by the photosensitizer. Relative singlet oxygen generation abilities and results of cell culture experiments altogether confirm selective activation in drug resistant cells.Open in a separate windowFig. 5Change in the cell viability of normal and paclitaxel resistant MCF7 cells (Pac-MCF7) in the presence of PS3 at various concentrations. For each group, cell viability is analysed both after incubation in the dark or after irradiation with a 505 nm LED light source from a distance of 10 cm. Average values of three independent experiments are used.Open in a separate windowFig. 6Apoptosis induction by PS3 (25 μM) in normal and paclitaxel resistant MCF7 cancer cells under dark conditions and upon irradiation with a 505 nm LED light from 10 cm distance. Scale bars: 50 μm.     

Do carbon nanotubes catalyse bromine/bromide redox chemistry?

The redox chemistries of both the bromide oxidation and bromine reduction reactions are studied at single multi-walled carbon nanotubes (MWCNTs) as a function of their electrical potential allowing inference of the electron transfer kinetics of the Br₂/Br⁻ redox couple, widely used in batteries. The nanotubes are shown to be mildly catalytic compared to a glassy carbon surface but much less as inferred from conventional voltammetry on porous ensembles of MWCNTs where the mixed transport regime masks the true catalytic response.

Schematic of a carbon nanotube impact in bromide solution.

The bromine–bromide redox couple plays an essential role in diverse energy storage devices including hydrogen–bromine, zinc–bromine, quinone–bromine, vanadium–bromide and bromide–polysulphide flow batteries.^1–5 The Br₂/Br⁻ redox couple is attractive as a cathode reaction due to its high standard potential, large solubility of both reagents, high power density and cost efficiency.⁶ The performance of such devices is generically limited by the thermodynamics and kinetics of the redox couple comprising the battery with fast (‘reversible’) electron transfer is essential. In many cases, including the Br₂/Br⁻ couple the electrode reaction involves more than one electron as given in the stoichiometric reaction:2Br⁻ − 2e⁻ ⇄ Br₂; E⁰ = 1.08 V vs. SHEwith, at high bromide concentrations, the possibility of the follow up chemical reaction⁷Br₂ + Br⁻ ⇄ Br₃⁻Since electrons are usually transferred sequentially this implies that the mechanism is multistep with any of the individual mechanistic steps in principle being rate limiting. For this reason catalysts are commonly required to enhance the electrode kinetics at otherwise favourable electrode materials. One type of catalyst which has seen wide usage, including for the Br₂/Br⁻ couple^8,9 are carbon nanotubes (CNTs) with suggested advantages which include high surface area and the inherent porosity of CNT composites.¹⁰ The deployment of CNTs as a porous composite presents a further level of complexity to the electrode reaction beyond its multistep character because of the ill-defined mass transport within the porous layer. In particular ascertaining the intrinsic electron transfer kinetics and hence the level of catalysis, if any, is essentially impossible since these are masked in the voltammetric response by diffusional mass transport effects.^11–14 Specifically the transport within the porous structure of CNT layers is dominated by thin-layer and other^15,16 effects which give the illusion of electrochemical reversibility. In order to unscramble possible electro-catalysis of the bromine/bromide couple a different approach is needed.In the following we study both the electro-oxidation of bromide (BOR) and the electro-reduction of bromine (BRR) at single MWCNTs via ‘nano-impact (aka ‘single entity’) electrochemistry’^17–20 in aqueous solution. In this approach a micro-wire electrode at a fixed potential is inserted in a suspension of CNTs in the solution of interest. From time to time a single CNT impacts the electrode, adopts the potential of the latter for the duration of the impact which in the case of CNTs can vary from 1–100 of seconds^21–23 and sustained catalytic currents flow if the oxidation/reduction of interest is faster at the nanotube in comparison with the micro-wire electrode. The catalytic currents are studied as a function of potential revealing the electron transfer kinetics. Fig. 1 shows the concept of the experiment.Open in a separate windowFig. 1Schematic representation of ‘nano-impact’ electrochemistry on a carbon micro wire electrode for the oxidation of aqueous bromide from which the kinetics of the BOR are inferred. Analogous experiments but showing negative impact currents allow the inference of the kinetics of the BRR.The BOR and BRR were studied first, however, voltammetrically at an unmodified glassy carbon (GC) electrode as shown in Fig. 2 (black line) using 5.0 mM solutions of either NaBr or Br₂ in 0.1 M HNO₃. The midpoint potential was 0.82 V versus the saturated calomel electrode (SCE) consistent with the literature values for the formal potential of the Br₂/Br⁻ couple.²⁴ The voltammograms were analysed to give transfer coefficients of 0.45 ± 0.01 and 0.33 ± 0.01 (ESI, Section 2^†) for the BOR and BRR respectively. Both processes were inferred to be diffusional and the diffusion coefficients D_Br⁻ and D_Br2 were calculated to be 2.05 (±0.04) × 10⁻⁵ cm² s⁻¹ and 1.50 (±0.04) × 10⁻⁵ cm² s⁻¹ (ESI, Section 3^†) using the Randles–Ševčík equation for an irreversible reaction the values are consistent with literature reports.²⁴ Then the electrodes were modified with 30 μg of MWCNTs consisting of ca. 125 monolayers (the calculation is given in the ESI, Section 9^†) of MWCNTs assuming that they are closely packed across the area of the GC electrode, and the resulting voltammograms are shown in Fig. 2 (red line). In comparison with the unmodified electrode, enhanced currents are seen for the Br₂/Br⁻ couple which partly reflects the enhanced capacitance of the interface reflecting in turn the large surface area of the deposited nanotubes (ca. 60–120 cm²). Larger signals are also seen indicating a thin layer contribution from the material occluded within the porous layer which also leads to the apparently quasi-reversible shape of the voltammograms obtained for both reactions. A log–log plot of peak current (I_p) vs. scan rate (ν) showed a gradient value of 0.68 (±0.01) and 0.66 (±0.03) for the BOR and BRR (ESI, Section 4^†) confirming a mixed mass transport regime^12,14 with a combination of semi-infinite diffusion and thin layer behaviour. The transition from the fully irreversible to the apparent quasi-reversible character is sometimes confused with electro-catalysis attributed to the CNTs rather than thin-layer diffusion. In order to ascertain the true catalytic response, single entity electrochemistry was measured to obtain the BOR and BRR responses at single CNTs.Open in a separate windowFig. 2Cyclic voltammograms at pristine GC (black line) and 30 μg MWCNTs dropcast on GC (red line) at a scan rate of 0.05 V s⁻¹ (a) for the bromide oxidation reaction (BOR) in 5.0 mM NaBr in 0.1 M HNO₃, (b) for the bromine reduction reaction (BRR) in 5.0 mM bromine in 0.1 M HNO₃.For single entity measurements, a clean carbon wire (CWE, length 1 mm and diameter 7 μm) working electrode was used. Chronoamperograms were recorded at a constant applied potential of 0.2 V vs. SCE and 1.3 V vs. SCE for the BOR and BRR respectively (5.0 mM solutions). These values were selected in the light of Fig. 2 to provide a large overpotential for each reaction. Clear oxidative and reductive current steps were observed (Fig. 3). These were ascribed to the arrival of a MWCNT at the electrode surface and the resulting catalytic electron transfer for the duration of the impact. No steps were observed in the absence of MWCNTs (ESI, Fig. S4^†). The average residence time of the MWCNT was 1.2 (±0.5) seconds and the frequency of the collisions was 0.3 (±0.1) impacts per second. The average impact current for the BOR at 1.3 V vs. SCE was 2.8 (±0.2) nA (65 impacts) and for the BRR at 0.2 V vs. SCE it was 3.8 (±0.1) nA (70 impacts). The impact currents were assumed to be entirely faradaic since control experiments in 0.1 M HNO₃ solution in the presence of 100 μg of MWCNTs (in the absence of Br⁻ and Br₂) showed no obvious impacts as shown in ESI Section 10.^†Open in a separate windowFig. 3Chronoamperograms showing the impact step current (a) for the BOR in 5.0 mM NaBr in 0.1 M HNO₃ at 1.3 V vs. SCE, (b) for the BRR in 5.0 mM bromine in 0.1 M HNO₃ at 0.2 V vs. SCE.Further, impacts for both the BOR and BRR were observed at various potentials (ESI, Section 11^†) and analysed to obtain the average faradaic current at each potential. The average impact step current was plotted against the applied potential (Fig. 4). Two sigmoidal curves were obtained reflecting the current–potential response for either the bromide oxidation (BOR) or the bromine reduction (BRR). The curves reflect the average voltammograms (current–potential characteristics) for the Br₂/Br⁻ redox reaction at single carbon nanotubes. The shape of the two sigmoidal curves reflects the onset of electrolysis followed by a diffusion controlled plateau at high over-potentials.²⁵ Mass transport corrected Tafel analysis (Fig. 4; inset) showed the transfer coefficients β to be ca. 0.42 and α to be ca. 0.20 from the impacts for the BOR and BRR respectively (ESI, Section 6^†). The length distribution of the MWCNTs was calculated (ESI, Section 6^†) from the currents recorded at potentials corresponding to the plateau in Fig. 4 assuming that the reactions are (Fickian) diffusion controlled at the potentials used and by modelling the CNTs as cylindrical electrodes²¹ assuming a nanotube radius of 15 (±5) nm and the diffusion coefficients reported above. Chronoamperometry was also conducted for the BOR and BRR in the absence of MWCNTs at 1.3 V and 0.2 V vs. SCE respectively to confirm that no impact currents were contributed by the redox species in the electrolyte (ESI, Section 5^†). Alongside, chronoamperograms in 0.1 M HNO₃ and 100 μg show that the impact current was contributed only by the Br⁻ and Br₂ redox reaction and the results are shown in the ESI, Section 10.^†Open in a separate windowFig. 4Average step currents observed as a function of applied potential (a) for the BOR in 5.0 mM NaBr in 0.1 M HNO₃ at, (b) for the BRR in 5.0 mM Bromine in 0.1 M HNO₃; insets in both the cases show mass transport corrected Tafel analyses.The lengths were found to be 5.4 (±3.4) μm (BOR) and 5.9 (±1.3) μm (BRR) and are given in Fig. 5 (see ESI, Section 7^† for calculations). These values were compared with previously reported dark-field optical microscopy data and good agreement was observed with the literature value of 5.3 (±2.1) μm.²⁶ The observed consistency provides strong support for the choice of modelling the single entity voltammetry by analogy with that of a cylindrical electrode.Open in a separate windowFig. 5The length of MWCNTs calculated from the impact currents for the BOR (at 1.3 V vs. SCE) and BRR (at 0.2 V vs. SCE).It is evident that the single entity measurements allow a clear analysis of the catalytic behaviour of the carbon nanotubes by providing a well-defined diffusional regime conducive to the extraction of the electrode kinetics of both the bromide oxidation and the bromine reduction process. In contrast, electrodes were formed by ensembles of carbon nanotubes in the form of a porous layer where the mixed transport regime is not amenable to ready modelling and the dissection of thin-layer effects from the measured voltammetry. The electron transfer kinetics for both the BOR and BRR at single MWCNTs was then obtained via full simulation of the two single entity ‘voltammograms’ using the above measured diffusion coefficients and again treating the impacted MWCNT as a cylindrical electrode with uniform diffusional access and further assuming Butler–Volmer kinetics. For the BOR, one electron transfer was considered as given below,For the BRR the two electron transfer was modelled as,Br₂ + 2e⁻ → 2Br⁻The set of parameters used for the analysis are given in the ESI, Section 8.^† By using the transfer coefficients deduced from Fig. 4, the only unknown is the standard electrochemical rate constant k which is determined by fitting the impact voltammogram measured relative to a formal potential for the Br₂/Br⁻ couple of 0.82 V vs. SCE obtained from the voltammogram at pristine GC. Fig. 6 shows the fitting for the BOR and the BRR with rate constants k_BOR of 1.0 (±0.1) × 10⁻³ cm s⁻¹ and k_BRR of 5.0 (±0.1) × 10⁻⁴ cm s⁻¹ respectively. The transfer coefficients and rate constants obtained from impacts were compared to the voltammograms obtained at pristine GC for the BOR and BRR and are given in Open in a separate windowFig. 6DIGISIM simulated curves (black line) for average impact currents obtained at different potentials (red circles) (a) for the BOR with a rate constant (k_BOR) of 1.0 (±0.1) × 10⁻³ cm s⁻¹; (b) BRR with a k_BRR of 5.0 (±0.1) × 10⁻⁴ cm s⁻¹.Transfer coefficients and rate constants for the BOR in 5.0 mM NaBr in 0.1 M HNO₃ and the BRR in 5.0 mM bromine in 0.1 M HNO₃ obtained at the glassy carbon macroelectrode GC, and single MWCNT impact current

Simplifying and expanding the scope of boron imidazolate framework (BIF) synthesis using mechanochemistry

Mechanochemistry enables rapid access to boron imidazolate frameworks (BIFs), including ultralight materials based on Li and Cu(i) nodes, as well as new, previously unexplored systems based on Ag(i) nodes. Compared to solution methods, mechanochemistry is faster, provides materials with improved porosity, and replaces harsh reactants (e.g. n-butylithium) with simpler and safer oxides, carbonates or hydroxides. Periodic density-functional theory (DFT) calculations on polymorphic pairs of BIFs based on Li⁺, Cu⁺ and Ag⁺ nodes reveals that heavy-atom nodes increase the stability of the open SOD-framework relative to the non-porous dia-polymorph.

Mechanochemistry enables rapid access to boron imidazolate frameworks (BIFs), including ultralight materials based on Li and Cu(i) nodes, as well as new, previously unexplored systems based on Ag(i) nodes.

Mechanochemistry^1–7 has emerged as a versatile methodology for the synthesis and discovery of advanced materials, including nanoparticle systems^8–10 and metal–organic frameworks (MOFs),^11–15 giving rise to materials that are challenging to obtain using conventional solution-based techniques.^16–18 Mechanochemical techniques such as ball milling, twin screw extrusion¹⁹ and acoustic mixing^20,21 have simplified and advanced the synthesis of a wide range of MOFs, permitting the use of simple starting materials such as metal oxides, hydroxides or carbonates,^22,23 at room temperature and without bulk solvents, yielding products of comparable stability and, after activation, higher surface areas than solution-generated counterparts.^24–29 The efficiency of mechanochemistry in MOF synthesis was recently highlighted by accessing zeolitic imidazolate frameworks (ZIFs)^30,31 that were theoretically predicted, but not accessible under conventional solution-based conditions.¹⁷The advantages of mechanochemistry in MOF chemistry led us to address the possibility of synthesizing boron imidazolate frameworks (BIFs),^32–34 an intriguing but poorly developed class of microporous materials analogous to ZIFs, comprising equimolar combinations of tetrahedrally coordinated boron(iii) and monovalent Li⁺ or Cu⁺ cations as nodes (Fig. 1A–C). Although BIFs offer an attractive opportunity to access microporous MOFs with lower molecular weights, particularly in the case of “ultralight” systems based on Li⁺ and B(iii) centers, this family of materials has remained largely unexplored – potentially due to the need for harsh synthetic conditions, including the use of n-butyllithium in a solvothermal environment.^32–34Open in a separate windowFig. 1Structures of previously reported BIFs with: (A) zni-, (B) dia-, or (C) SOD-topology (M = Li, Cu); (D) tetrakis(imidazolyl)boric acids used herein for mechanochemical BIF synthesis; and (E) schematic representation of the herein developed mechanosynthesis of dia- and SOD BIF polymorphs based on Li, Cu or Ag metal nodes.We now show how switching to the mechanochemical environment enables lithium- and copper(i)-based BIFs to be prepared rapidly (i.e., within 60–90 minutes), without elevated temperatures or bulk solvents, and from readily accessible solid reactants, such as hydroxides and oxides (Fig. 1D and E). While the mechanochemically-prepared BIFs exhibit significantly higher surface areas than the solvothermally-prepared counterparts, mechanochemistry allows for expanding this class of materials towards previously not reported Ag⁺ nodes. The introduction of BIFs isostructural with those based on Li⁺ or Cu⁺ but comprising of Ag⁺ ions, enables a periodic density-functional theory (DFT) evaluation of their stability. This reveals that switching to heavier elements as tetrahedral nodes improves the stability of sodalite topology (SOD) open BIFs with respect to close-packed diamondoid (dia) topology polymorphs.As a first attempt at mechanochemically synthesis of BIFs, we targeted the synthesis of previously reported zni-topology LiB(Im)₄ and CuB(Im)₄ frameworks (Li-BIF-1 and Cu-BIF-1, respectively, Fig. 1A) using a salt exchange reaction between LiCl or CuCl with commercially available sodium tetrakis(imidazolyl)borate (Na[B(Im)₄]) (Fig. 2A). Milling of LiCl and Na[B(Im)₄] in a 1 : 1 stoichiometric ratio for up to 60 minutes led to the appearance of Bragg reflections consistent with the target Li-BIF-1 (CSD MOXJEP) and the anticipated NaCl byproduct. The reaction was, however, incomplete, as seen by X-ray reflections of Na[B(Im)₄] starting material. In order to improve reactant conversion, we explored liquid-assisted grinding (LAG), i.e. milling in the presence of a small amount of a liquid phase (measured by the liquid-to-solid ratio η³⁵ in the range of ca. 0–2 μL mg⁻¹). Using LAG conditions with acetonitrile (MeCN, 120 μL, η = 0.5 μL mg⁻¹) led to the complete disappearance of reactant X-ray reflections, concomitant with the formation of Li-BIF-1 alongside NaCl within 60 minutes.Open in a separate windowFig. 2(A) Reaction scheme for the mechanochemical synthesis of Li-BIF-1 by a salt metathesis strategy. Selected PXRD patterns for: (B) Na[B(Im)₄] (C) LiCl, (D) simulated Li-BIF-1 (CSD MOXJPEP) and (E) synthesized BIF-1-Li by LAG for 60 minutes with MeCN (η = 0.5 μL mg⁻¹), (F) CuCl, (G) simulated Cu-BIF-1 (CSD MOXJIT), and (H) synthesized BIF-1-Cu by LAG for 60 minutes with MeOH (η = 0.50 μL mg⁻¹). Asterisks denote NaCl, a byproduct of the metathesis reaction. (Fig. 2B–E, also see ESI^†). The copper-based zni-CuB(Im)₄ (Cu-BIF-1) was readily obtained from CuCl within 60 minutes using similar LAG conditions. We also explored LAG with methanol (MeOH), revealing that the exchange reaction to form NaCl took place with both LiCl and CuCl starting materials. With LiCl, however, the PXRD pattern of the product could not be matched to known phases involving Li⁺ and B(Im)₄⁻ (see ESI^†). With CuCl as a reactant, LAG with MeOH (η = 0.5 μL mg⁻¹) cleanly produced Cu-BIF-1 alongside NaCl (see ESI^†).Next, we explored an alternative synthesis approach, analogous to that previously used to form ZIFs and other MOFs: an acid–base reaction between a metal oxide or hydroxide and the acid form of the linker: tetrakis(imidazolato)boric acid, HB(Im)₄ (Fig. 3A).^36–40 Neat milling LiOH with one equivalent of HB(Im)₄ in a stainless steel milling assembly led to the partial formation of Li-BIF-1, as evidenced by PXRD analysis (see ESI^†). Complete conversion of reactants into Li-BIF-1 was achieved in 60 minutes by LAG with MeCN (η = 0.25 μL mg⁻¹), as indicated by PXRD analysis (Fig. 3B–E), Fourier transform infrared attenuated total reflectance spectroscopy (FTIR-ATR), thermogravimetric analysis (TGA) in air, and analysis of metal content by inductively-coupled plasma mass spectrometry (ICP-MS) (see ESI^†).Open in a separate windowFig. 3(A) Reaction scheme for the mechanochemical synthesis of Li-BIF-1 using the acid–base strategy. Selected PXRD patterns for: (B) H[B(Im)₄] (C) LiOH, (D) simulated Li-BIF-1 (CSD MOXJPEP), (E) synthesized BIF-1-Li by LAG for 60 minutes with MeCN (η = 0.25 μL mg⁻¹), (F) Cu₂O, (G) simulated Cu-BIF-1 (CSD MOXJIT), and (H) synthesized Cu-BIF-1 by ILAG for 60 minutes with MeOH (η = 0.50 μL mg⁻¹) and NH₄NO₃ additive (5% by weight).Neat milling of HB(Im)₄ with Cu₂O under similar conditions gave a largely non-crystalline material, as evidenced by PXRD (see ESI^†). Switching to the ion- and liquid-assisted grinding (ILAG) methodology, in which the reactivity of a metal oxide is enhanced by a small amount of a weakly acidic ammonium salt, and which was introduced to prepare zinc and cadmium ZIFs from respective oxides,^37–40 enabled the synthesis of Cu-BIF-1 from Cu₂O. Specifically, PXRD analysis revealed complete disappearance of the oxide in samples obtained by ILAG with either MeOH or MeCN (η = 0.5 μL mg⁻¹) in the presence of NH₄NO₃ additive (5% by weight, see ESI^†). Notably, achieving complete disappearance of Cu₂O reactant signals also required switching from stainless steel to a zirconia-based milling assembly, presumably due to more efficient energy delivery.⁴¹ After washing with MeOH, the material was characterized by FTIR-ATR, TGA in air, and analysis of metal content by ICP-MS (see ESI^†).Whereas both the metathesis and acid–base approaches can be used to mechanochemically generate Li- and Cu-BIF-1, the latter approach has a clear advantage of circumventing the formation of the NaCl byproduct. Consequently, in order to further the development of mechanochemical routes to other BIFs, we focused on the acid–base strategy. As next targets, we turned to MOFs based on tetrakis(2-methylimidazole)boric acid H[B(Meim)₄],³⁶ previously reported³² to adopt either a non-porous diamondoid (dia) topology (BIF-2) or a microporous sodalite (SOD) topology (BIF-3) with either Li⁺ or Cu⁺ as nodes (Fig. 4). Attempts to selectively synthesize either Li-BIF-2 or Li-BIF-3 by neat milling or LAG (using MeOH or MeCN as liquid additives) with LiOH and a stoichiometric amount of HB(Meim)₄ were not successful. Exploration of different milling times and η-values produced only mixtures of residual reactants with Li-BIF-2, Li-BIF-3, and/or not yet identified phases (see ESI^†). Consequently, we explored milling in the presence of 2-aminobutanol (amb), which is a ubiquitous component of solvent systems used in the solvothermal syntheses of BIFs.^32,33 Gratifyingly, using a mixture of amb and MeCN in a 1 : 3 ratio by volume as the milling liquid led to an effective strategy for the selective synthesis of both the dia-topology Li-BIF-2 (CSD code MOXKUG), and the SOD-topology Li-BIF-3 (CSD code MUCLOM).^‡ The selective formation of phase-pure samples of Li-BIF-2 and Li-BIF-3 was confirmed by PXRD analysis, which revealed an excellent match to diffractograms simulated based on the previously reported structures (Fig. 4B–G). Systematic exploration of reaction conditions, including time (between 15 and 60 minutes) and η value (between 0.25 and 1 μL mg⁻¹) revealed that the open framework Li-BIF-3 is readily obtained at η either 0.75 or 1 μL mg⁻¹ after milling for 45 minutes or longer (Fig. 4B–G, also see ESI^†).^§ Lower η-values of 0.25 and 0.5 μL mg⁻¹ preferred the formation of the dia-topology Li-BIF-2, which was obtained as a phase-pure material upon 60 minutes milling at η = 0.5 μL mg⁻¹, following the initial appearance of a yet unidentified intermediate. The preferred formation of Li-BIF-2 at lower η-values is consistent with our previous observations that lower amounts of liquid promote mechanochemical formation of denser MOF polymorphs.³⁷Open in a separate windowFig. 4(A) Reaction scheme for the mechanochemical synthesis of Li-BIF-3. Comparison of selected PXRD patterns for the synthesis of Li-BIF-2 and Li-BIF-3: (B) H[B(Meim)₄] reactant; (C) LiOH reactant; (D) simulated for Li-BIF-3 (CSD MUCLOM); (E) simulated for Li-BIF-2 (CSD MOXKUG); (F) Li-BIF-3 mechanochemically synthesized by LAG for 60 minutes with a 1 : 3 by volume mixture of amb and MeCN (η = 1 μL mg⁻¹); and (G) Li-BIF-2 mechanochemically synthesized by LAG for 60 minutes with a 1 : 3 by volume mixture of amb and MeCN (η = 0.5 μL mg⁻¹). Comparison of selected PXRD patterns for the synthesis of Cu-BIF-2 and Li-BIF-3: (H) Cu₂O; (I) Cu-BIF-3 (CSD MOXJOZ); (J) Cu-BIF-2 (CSD MUCLIG); (K) Cu-BIF-3 mechanochemically synthesised by ILAG for 60 minutes using NH₄NO₃ ionic additive (5% by weight) and MeOH (η = 1 μL mg⁻¹); and (L) mechanochemically synthesised Cu-BIF-2 by ILAG for 90 minutes using NH₄NO₃ ionic additive (5% by weight) and MeOH (η = 0.5 μL mg⁻¹).Samples of both Li-BIF-2 and Li-BIF-3 after washing with MeCN were further characterized by FTIR-ATR, TGA in air, and analysis of metal content by ICP-MS (see ESI^†). Nitrogen sorption measurement on the mechanochemically obtained Li-BIF-3, after washing with MeCN and evacuation at 85 °C, revealed a highly microporous material with a Brunauer–Emmett–Teller (BET) surface area of 1010 m² g⁻¹ (Fig. 5A), which is close to the value expected from the crystal structure of the material (1200 m² g⁻¹, 32 For direct comparison with previous work,³² we also calculated the Langmuir surface area, revealing an almost 40% increase (1060 m² g⁻¹) compared to samples made solvothermally (762.5 m² g⁻¹) (Fig. 5A, inset).Experimental Brunauer–Emmett–Teller (BET) and Langmuir surface area (in m² g⁻¹) of mechanochemically synthesized SOD-topology BIFs, compared to previously measured and theoretically calculated values, along with average particle sizes (in nm) established by SEM and calculated energies (in eV) for all Li-, Cu-, and Ag-BIF polymorphs. The difference between calculated energies for SOD- and dia-polymorphs in each system is given as ΔE (in kJ mol⁻¹)

Operando acoustic analysis: a valuable method for investigating reaction mechanisms in mechanochemistry

We present a new operando approach for following reactions taking place in mechanochemistry, relying on the analysis of the evolution of the sound during milling. We show that differences in sound can be directly correlated to (physico)chemical changes in the reactor, making this technique highly attractive and complementary to others for monitoring mechanochemical reactions. Most notably, it can provide unique information on the actual movements of the beads within the milling jars, which opens new avenues for helping rationalize mechanochemical processes.

We present a new operando approach for following reactions taking place in mechanochemistry, relying on the analysis of the evolution of the sound during milling.

The growing attention to mechanochemistry has led to an impressive range of research areas to which this “green” synthetic method can be applied, ranging from pharmacy to catalysis as well as crystal engineering or materials science.^1,2 Although widely used for synthesizing elaborate compounds, the details of grinding and more specifically of the physical and chemical mechanisms taking place are still often seen as unfathomable. In the case of ball-milling, this is mainly due to the opaque and closed environment of the milling reactors in which most reactions take place.In light of the great interest in mechanochemistry, several research groups have thus strongly focused their attention on developing in situ as well as operando analytical methods in order to monitor reactions and observe evolutions in the structure, crystallinity and/or texture of compounds involved during the milling processes.^3–8 Except for temperature, these methods usually require the use of transparent reactor materials (for Raman or X-ray diffraction monitoring, for example),^4,9 or the adaptation of the geometry and size of the reactor itself (in particular in the case of NMR analyses).^10–16 However, several studies have shown that the kinetics and/or reaction mechanisms in ball-milling can be greatly affected by the size and materials composing the reactor and beads, meaning that the latter methods cannot be readily applied to different reactions without re-investigating the synthetic conditions.^17,18 Hence, developing new operando approaches allowing the evolution of reaction media to be followed, but without requiring any modification of the initial set of experimental parameters, would be greatly beneficial for the mechanochemistry community.Several mechanical behaviours involving the beads and reactors have been shown to come into play during a ball-mill synthesis.^19–22 The reacting particles undergo direct impacts from the beads, as well as shearing forces, which both can lead to particle size reduction and the creation of “fresh reactive surfaces”, and thereby enable the chemical reactions to take place. Moreover, under some milling conditions, specific trajectories of the beads have also been identified, such as their direct back and forth movements from one wall of the reactor to another, or their rolling along the sides of the milling reactor. Such processes are generally difficult to predict, and depend on numerous experimental parameters such as the type of beads (size, number and material), the type of reactors (volume, geometry, and material), the milling parameters (frequency, time), not to mention the type of movements imposed to the reactors by the mill (e.g., planetary, oscillating vertically or horizontally), but also the physical and chemical properties of the milled reagents. Various sounds can sometimes be noticed during the milling of compounds, which attest of changes in the beads'' motions in the reactor. Yet, although a few previous attempts to follow the sound during ball-milling experiments had been reported, notably in 2004 by Cuadrado-Laborde et al.,²³ to the best of our knowledge, only sounds resulting from the impact of a bead on the reactor were identified and analysed. No systematic investigation was made at the time on how the sound recorded could inform on other movements of the bead (aside from the simple impact), nor, more importantly, on how it could be used to directly follow different physico-chemical changes occurring in the milling reactors.²³ In this context, we decided to record the sound produced by a vertical mixer mill upon milling, in order to try to decipher the acoustic variations arising from different movements of the beads, and how they can be related to changes in the composition or physico–chemical properties of the medium. Moreover, as detailed below, we investigated how this method compares to other known operando techniques, i.e. Raman spectroscopy and temperature measurements, for following reaction mechanisms taking place during ball-milling (see Fig. 1).Open in a separate windowFig. 1Top: schematic representation of the set-up used for the operando monitoring of mechanochemical reactions, using the combination of Raman spectroscopy, thermal imaging, and the recording of an AF spectrogram. Bottom: photo of the instrumental set-up used in this work.As a first system of study, we decided to follow the mechanochemical formation of a 1 : 1 cocrystal of terephthalic acid (H₂BDC) and 1,4-diazabicyclo[2.2.2]octane (DABCO) (see Fig. 2a). The sound was recorded through the use of a microphone situated close to the milling reactor. This recording was then used to generate an audio-frequency spectrogram (AF spectrogram), which allows the Fourier transform of the signal to be monitored as a function of time, and different sound harmonics to be observed. In parallel to the sound recording, Raman analyses and temperature measurements were also carried out to follow the evolution of the reaction medium (Fig. 1). These three operando methods were adapted to a vertical mixer mill (Pulverisette 23, Fritsch) using a Perspex reactor with two ZrO₂ beads (Ø 10 mm). Here, the experiments were conducted at a frequency of 50 Hz with a reaction time of 60 min, for a total mass of reactants ∼500 mg (see ESI^† for more details).Open in a separate windowFig. 2(a) Dry milling reaction between H₂BDC (1.0 eq) and DABCO (1.0 eq) during 60 min at 50 Hz, inducing the formation of a cocrystal (see Fig. S1 and S2^† for ex situ pXRD and Raman analyses of the cocrystal). (b). Left: intensity of the 3^rd harmonic of the AF spectrogram as a function of time, obtained by integration between 145 and 155 Hz. Centre: temperature profile of the reactor during the milling reaction presented in (a). Right: time-resolved operando Raman spectra for a selected area (between 950 and 1050 cm⁻¹); the intensity scale goes from low (blue) to high (red and then white). The red shaded rectangle emphasizes the presence of the intermediate species, while purple ones highlight the erratic rolling motions.During the mechanochemical formation of the H₂BDC–DABCO cocrystal (Fig. 2a),²⁴ various sounds were noticed. Careful analysis of the AF spectrogram revealed contrasts in signal intensity in the 105–145 Hz and 145–155 Hz regions (see Fig. S3b^†), i.e. close to the 3^rd harmonic of the milling frequency (50 Hz in this case). Notably, the acoustic intensity between 145 and 155 Hz was found to significantly increase between ∼1000 and 1500 s of milling (Fig. 2b, red shaded zone). This experimental observation could be reproduced in the same time-frame by repeating the experiment (see Fig. S4^†). Interestingly, when looking at the measurements recorded using the other operando methods, an increase of the reactor''s temperature by about 10–12 °C was simultaneously observed using the thermal imaging camera (Fig. 2b). Moreover, the Raman spectra recorded over the same period of time also revealed shifts in the DABCO and H₂BDC vibration bands (Fig. 2b). Hence, these first results suggest that changes in sound can be directly correlated to chemical reactions occurring in the reactor.Further analysis of the Raman spectra was performed (Fig. S5^†), focusing notably on the zone between 950 and 1050 cm⁻¹, in which the DABCO ν_CC/ω_CH2 vibration bands appear.²⁵ The latter vibration mode was first found to shift from 970.8 cm⁻¹ to 984.3 cm⁻¹ after ∼1000 s of milling, and then to further shift to 1013.5 cm⁻¹ (after ca. 500 s), which corresponds to the vibration frequency expected for the final cocrystal (Fig. 2b). Although the exact nature of the intermediate has not yet been unambiguously identified, it can be tentatively assigned to an {H₂BDC–DABCO} species, in which only one of the nitrogen atoms of DABCO becomes protonated (based on the shift observed in Raman spectroscopy),²⁵ leading to a single N–H⋯O hydrogen-bonding motif (in contrast with the final cocrystal which contains two). Remarkably, the previously noticed variation of sound around 150 Hz (3^rd harmonic) happens precisely during the short lifespan of this intermediate, i.e. between ∼1000 and 1500 s. To the best of our knowledge, it is the first time that evidence of the presence of reaction intermediates in mechanochemistry is accounted for by analysing the sound occurring during the milling.In order to help rationalize the origin of the changes in sound during the formation of the intermediate, the movements of the beads across the transparent Perspex milling reactors were visualized using a digital camera. In doing so, it was found that when the intermediate appears (thus when the intensity of the sound is maximum in the 145–155 Hz frequency range, i.e. in the red-shaded area on Fig. 2b), the beads undergo a rolling type of movement on the wall of the reactor. Such motion can be opposed to the erratic impact of the beads which occurs during most of the remaining time, and can be identified here on the AF spectrogram by the more intense inharmonic signals between 105 and 145 Hz (see ESI Fig. S3^†). It is possible that the predominance of rolling between 1000 and 1500 s may be caused by the significant changes in texture of the medium upon formation of the intermediate. More generally speaking, to the best of our knowledge, it is the first time that an acoustic signature characteristic of rolling events in mechanochemistry is identified through sound measurements.Beyond the possibility of directly observing the formation and/or disappearance of a reaction intermediate, we also found that the analysis of the AF spectrogram actually helps understand deeper some of the details of the data recorded using Raman spectroscopy and the thermal imaging camera. For this reaction system, the “rolling” vs. “impact” sound signatures were found to largely appear in specific regions, as illustrated in Fig. S6.^† Indeed, the impact is appearing in the 105–145 Hz region, while rolling is visible in the 145–155 Hz one. Yet, interestingly, the signature characteristic of rolling motions was observed not only during the life-span of the intermediate, but also intermittently before and after its appearance (see for example the purple-shaded rectangles in Fig. 2b), with a simultaneous decrease in the Raman signals intensities and increase in temperature. Concerning the Raman data, this can be explained by the fact that when rolling occurs, the beads are more likely to remove the powder from the point of focus of the Raman laser, implying that less compound is then analysed during the Raman acquisition time, hence the decrease in signal intensity. Regarding temperature, the small increase may be caused by a more significant frictional heating of the beads against the reactor''s walls during the rolling events.²⁶Overall, throughout this first example, we have been able to show how sound modifications can attest of a chemical evolution of the medium inside the milling reactor in a complementary way to two other operando techniques (Raman analyses and thermal imaging measurements), and how they can also explain some of the smaller features observed during the acquisition of the operando data (notably when rolling motions occur). In light of these first results, we decided to study on one hand whether other changes in the physical aspect of a reaction medium could be correlated to variations in sound, and on the other hand how sound measurements may actually inform on chemical events taking place in the milling reactors, in situations when temperature measurements or Raman operando analyses can be ambiguous, poorly informative, or not even applicable.In order to investigate how changes in physical aspect of a reaction medium can be followed through the recording of AF spectrograms, we decided to study the hydrolysis of SiO₂ (initially introduced in the reactor as “fumed silica”), upon progressive addition of small amounts of water during ball-milling. Indeed, as shown in Fig. 3, this system appeared to us as ideal for such investigations, because it goes through many of the physical states commonly encountered in ball-milling reactions. First, the medium is under the form of a light fluffy powder, which upon addition of water forms more agglomerated particles (leading to a “heavy powder”), which then further agglomerate and stick to the beads (leading to a so-called “snow-ball” effect).^7,8,27,28 The subsequent addition of water enables the medium to re-spread onto the walls of the reactor (the phase then having the aspect of a “sticky gel” or “gel”), and finally evolves towards a slurry type of texture. These changes were thus followed using a similar set-up as for the study of the H₂BDC–DABCO cocrystal (P23 vertical mixer mill operated at 50 Hz, using a Perspex reactor with two zirconia beads), and after each addition of water, the milling was performed for 5 minutes. Here, in addition to recording the AF spectrogram, the changes in temperature of the reactor were monitored using a thermal-imaging camera, and the movements of the beads inside the reactor were also filmed at specific time-points using a digital camera. It is worth noting that the observations described below were found to be essentially reproducible (see Fig. S7^† for comparison of AF spectrograms).Open in a separate windowFig. 3(a) Scheme of the hydrolysis of SiO₂ in a Perspex reactor using two zirconia beads. After each addition of 10 μL of H₂O, the mixture was milled during 5 min at 50 Hz. (b) Sound intensity variation of the 3^rd harmonic (145–155 Hz) according to the physical state of SiO₂ upon addition of H₂O. Associated pictures are taken just before the water addition. The red dashed line is given as a guideline to show the differences in sound intensities for motions which are predominantly impact. The blue arrows indicate the 5 minutes periods of milling when impact motions were predominant. (c) Temperature profile of the reactor. Blue shaded rectangles emphasize the impact motions regions. The green arrows highlight the regions which show similar temperature profiles (but different sound signatures). (d) Details of the three different bead motions detected during the experiment (on the left: pictures of the reactor, on the right: schemes of the beads motion): impact, rolling under snow-ball effect and rolling.When looking at the full AF spectrogram, it is clear that it is around the third harmonic (∼150 Hz) that the contrast is the most significant and the intensity the highest (see Fig. S8 in ESI^† for changes in sound intensity in other spectral regions). As for the H₂BDC–DABCO system, it was found that high intensity signals around this frequency can attest of rolling-types of motions, as shown for example by analysing the results in the zone corresponding to the addition of 180–300 μL of water, in which the presence of rolling is clearly visible when visualising the movement of the beads and looking at the aspect of the medium after opening of the reactor. Concomitantly, the events with the lowest intensity signals at ∼150 Hz (i.e. after addition of 140 and 170 μL) could be corroborated with more “impact-like” movements (see blue arrows in Fig. 3b, zone corresponding to the “sticky gel” aspect of the mixture). These events were also found to correspond to a simultaneous increase in signal intensity in the 105–145 Hz region, in line with previous observations made for the H₂BDC–DABCO system (see Fig. S9^†). However, it is important to highlight that in the case of the hydrolysis of silica, a lower intensity signal at ∼150 Hz did not systematically correlate with a higher intensity sound in the 105–145 Hz region (see ESI, Fig. S8^†), in contrast with what had been seen previously for the cocrystal (see below for further discussion). This shows that in order to help interpret the subtleties of the movements of the beads within the reactor and of the physico–chemical processes taking place, the investigation of several zones of the AF spectrograms is actually needed.When analysing in more detail the overall plot of the changes in sound intensity around the 3^rd harmonic as a function of the amount of water added (Fig. 3), it is clear that the most significant increase occurs in the zone corresponding to the formation of the “snow-ball” (i.e. after addition of 90–130 μL of water – red-shaded zone), when the beads start rolling against the reactor''s walls and accumulating powder at their surface. Most interestingly, when looking at the movement of the beads in this regime using a digital camera, a rolling of both beads in separate zones of the reactor could be observed, as tentatively illustrated in Fig. 3d. A possible explanation to the related increase in sound would be that upon the formation of the snow-balls, the walls of the reactor then become mostly free of powder, inducing such sound intensification. A contrario, when the powder is homogeneously spread over the walls of the reactor, as is the case upon formation of the “heavy powder” (zones highlighted in light blue in Fig. 3, which correspond essentially to rolling motions), the sound intensity is much more dampened (see Fig. 3, S7 and S9^†).The importance of being able to follow the sound during the milling further appears when comparing the results of the AF spectrogram analyses to those of the temperature measurements. In several situations, just like for the H₂BDC–DABCO system, the rolling events were found to be associated to temperature increases inside the milling reactor, as shown in particular after addition of 150 μL and of 180 to 250 μL of water. However, changes in sound intensity around 150 Hz did not systematically correlate in the same way to changes in temperature. For example, while the temperature profiles after addition of 70 and 110 μL of water were essentially the same (see green arrows on Fig. 3c), the sound intensity at ∼150 Hz over that same period of time was very different, in one case corresponding to the milling of a “heavy powder”, while in the other to a snowball. Such results clearly demonstrate the complementarity of acoustic and temperature measurements in trying to decipher the evolution in the physico–chemical properties of the medium during the milling. Considering the increasing number of studies in which temperature measurements are reported,^26,29,30 and related to endo- or exothermal chemical reactions occurring in the milling reactors, being able to take into consideration that changes in beads motions can also be correlated to changes in temperature is another important feature to bear in mind.Overall, although many other aspects related to the milling of silica in presence of water would deserve to be investigated, these results on the hydrolysis of silica clearly highlight that (i) beyond the identification of rolling or erratic impact movements of the beads, acoustic measurements can attest of other phenomena occurring in the reactor, such as the formation of snow-balls (during which a different type of rolling movements take place), or the evolution of the aspect of the mixture (e.g. formation of a “sticky gel” inducing impact motions); and (ii) sound measurements are highly complementary to temperature measurements, as they can bring unique evidence of phenomena taking place during the milling.Finally, in order to demonstrate the utility of sound measurements for the optimization of ball-milling reaction conditions, we decided to study the hydrolysis of an activated fatty acid, lauric acid, using this operando method. Indeed, during our previous work on the ¹⁷O/¹⁸O isotopic enrichment of fatty acids,^31,32 we had found that for saturated fatty acids, the hydrolysis time of the activated acyl-imidazole intermediates had to be systematically optimized by repeating the experiments with various milling times. Given that this hydrolysis is accompanied by a change in texture of the reaction medium, which switches from a powdery aspect to a more oily one, we suspected that acoustic measurements could be used to determine the time after which the hydrolysis is complete (Fig. 4a).Open in a separate windowFig. 4(a) Reaction corresponding to the mechanochemical hydrolysis of activated lauric acid (CH₃–(CH₂)₁₀–COOH), which was followed here using operando analyses. (b) Operando experiments conducted in a Raman laser-transparent Perspex reactor (see picture in the orange rectangle, left) and in an opaque stainless steel reactor (see picture in the blue rectangle, right) with: temperature profiles of the reactor during the milling reactions, intensity of the 5^th harmonic of the AF spectrograms as a function of time obtained by an integration between 205 and 245 Hz, and Raman analyses (for the experiment in a Perspex reactor). Dotted red lines are used as visual helps for indicating the completion of the reaction.Two experimental conditions were compared: the first hydrolysis experiment was performed in a Perspex reactor with two zirconia beads (10 mm Ø) in order to allow the Raman signals to be recorded, while the second one was performed in a stainless steel reactor with two stainless steel beads (10 mm Ø), and can be seen as a “blind” experiment (as it cannot be followed by Raman spectroscopy). The changes in sound, temperature, and, in the former case, in Raman signatures, during the milling are compared in Fig. 4.For these milling experiments, carried out at a frequency of 50 Hz, it was found that the highest contrast on the AF spectrograms appears close to the 5^th harmonic, and more specifically in the 205–245 Hz frequency region (see ESI Fig. S10–S12^†). For both types of milling reactors, after an initial period during which significant variations in the sound intensity are observed, there is a stabilisation of the sound intensity (see dashed red vertical line in Fig. 4b), suggesting that a final physical state of the reaction medium has been reached. When looking at the simultaneous temperature measurements, it appears that this time also corresponds to the threshold after which no strong changes in reactor''s temperature are noticed. Moreover, when looking at the operando Raman data recorded for the experiment carried out in a Perspex reactor, it is clear that this threshold also corresponds to the period after which no further change in the vibration frequencies occurs, as illustrated by focusing on the 1100–1200 cm⁻¹ spectral region. Using ex situ analyses by FTIR spectroscopy, it was verified that after stabilization of the sound the hydrolysis is indeed complete, for reactions carried out both in Perspex and stainless steel reactors (see Fig. S12^†). Overall, this demonstrates the utility of performing sound measurements in operando mode, in order to determine the time at which experiments are over, especially in situations when simultaneous Raman analyses are not possible, for example when working in a stainless steel reactor, or when temperature measurements may not be possible, for example in cases when dedicated temperature-regulated milling systems are used for the experiments.In conclusion, we have presented here a new approach based on the study of the sound for following mechanochemical syntheses using operando methods. The potential of this approach was illustrated in three different examples, demonstrating how it can be used to draw attention to the formation of synthetic intermediates, to know when a reaction is complete, and also to follow changes in the physical state of a reaction medium. Moreover, it was shown that sound analyses can shed light on different types of movements of the beads within the milling reactors, meaning that it is highly interesting to use such analyses to decipher the different phenomena taking place within the milling jars. Overall, this makes sound measurements a promising addition to the toolbox of methods which mechanochemists can now use to try to understand and control the evolution of reaction media during ball-milling.     

Ba4Ca(B2O5)2F2: π-conjugation of B2O5 in the planar pentagonal layer achieving large second harmonic generation of pyro-borate

The nonlinear optical (NLO) crystals that can expand the wavelength of the laser to the deep-ultraviolet (DUV) region by the cascaded second harmonic generation (SHG) are of current research interest. It is well known that borates are the most ideal material class for the design of new DUV NLO crystals owing to the presence of good NLO genes, e.g., BO₃ or B₃O₆ groups. However, the NLO pyro-borates with the B₂O₅ dimers as the sole basic building units are still rarely reported owing to their small SHG responses. In this communication, by constructing a planar pentagonal [Ca(B₂O₅)]_∞ layer, the NLO pyro-borate Ba₄Ca(B₂O₅)₂F₂ with a large SHG response (∼2.2 × KDP, or ∼7 × α-Li₄B₂O₅) and a DUV transparent window has been designed and synthesized. The first-principles calculations show that the large SHG response of Ba₄Ca(B₂O₅)₂F₂ mainly originates from the better π-conjugation of the coplanar B₂O₅ dimers in the [Ca(B₂O₅)]_∞ layer. In addition, the planar pentagonal pattern in the [Ca(B₂O₅)]_∞ layer provides an ideal template for designing the new DUV NLO crystals, apart from those in known DUV borates, e.g., the [Be₂BO₃F₂]_∞ layer in KBe₂BO₃F₂ (KBBF).

A new deep-UV NLO pyro-borate Ba₄Ca(B₂O₅)₂F₂ was synthesized by solid-state reactions. The better π-conjugation of B₂O₅ dimers in the planar pentagonal layer achieves a large SHG response (∼2.2 × KDP), which is the largest among all the known DUV transparent borates with B₂O₅ units.

Deep-ultraviolet (DUV, λ < 200 nm) coherent lights with high photon energy, high spatial resolution, and a small heat-affected zone are of significance for applications in photolithography, high-resolution spectroscopy, laser cooling, and scientific equipment.^1–4 However, it is difficult or well-nigh impossible for solid-state lasers to directly radiate the DUV coherent lights. In contrast, relying on the process of second harmonic generation (SHG) of nonlinear optical (NLO) crystals is a more effective way to generate the DUV coherent lights and causes much attention.^5,6 Therefore, the NLO crystal has become an important material basis of solid-state lasers, which seriously affects the development of all-solid-state laser technology. However, it is still a great challenge to rationally design and synthesize DUV NLO crystals because of the extremely rigorous requirements of structural symmetry and properties.^7–10 Structurally, the DUV NLO crystals must crystallize in the noncentrosymmetric (NCS) space groups which are the prerequisite for the materials to exhibit SHG responses. Moreover, it should possess a broad transparency window, a largely effective NLO coefficient (d_eff ≥ 0.39 pm V⁻¹), and a moderate birefringence (0.05–0.10@1064 nm) to achieve the phase-matching (PM) conditions in the DUV region.¹⁰ Based on these requirements, borates have been considered as the ideal material class for DUV NLO crystals because of their special structure and properties'' virtues, including the rich acentric structural types, large band gaps, and stable physical and chemical properties.⁸ To date, the commercialized borate-based UV NLO crystals consist of β-BaB₂O₄ (BBO), LiB₃O₅ (LBO), CsLiB₆O₁₀ (CLBO),^9,10 and the practical DUV NLO crystal KBe₂BO₃F₂ (KBBF). Especially for KBBF, it has become the sole material that can generate DUV coherent laser light (177.3 nm) by a direct SHG method.⁷ Other excellent borate-based UV NLO crystals also consist of K₃B₆O₁₀Cl,¹¹ SrB₅O₇F₃,¹² Li₂B₆O₉F₂,⁵ CsAlB₃O₆F,¹³ M₂B₁₀O₁₄F₆ (M = Ca, Sr),¹⁴ NH₄B₄O₆F,¹⁵ NaSr₃Be₃B₃O₉F₄,¹⁶ AB₄O₆F (A = K, Rb, and Cs),¹⁷etc.The above borate-based materials have achieved great success as UV and DUV NLO crystals, which are mainly attributed to the ability of boron atoms to coordinate with three or four oxygen anions forming trigonal-planar or tetrahedral building blocks.^18,19 For example, the first borate-based NLO crystal, KB₅O₈·4H₂O (KB₅), has the basic building units (BBUs) of [B₅O₁₀], while the BBUs of β-BBO, LBO, and KBBF are [B₃O₆], [B₃O₇], and isolated [BO₃], respectively.^7,8 Remarkably, although various borate crystals with different types of borate groups have been explored during the past decades, the pyro-borate NLO crystals with B₂O₅ groups as the sole BBUs are rarely reported owing to their weak SHG responses.^20–23 For example, the SHG response of the DUV transparent α-Li₄B₂O₅ (ref. 23) is only ∼0.3 × KDP, which is far smaller than the expected value (0.39 pm V⁻¹, 1 × KDP).Actually, the flexible B₂O₅ groups which are composed of two π-conjugated BO₃ units through corner-sharing may also be capable of generating excellent optical performance if they have benign arrangements. In recent research, Pan''s group has indicated that the B₂O₅ dimers are perfect for the design of DUV birefringent crystals. By the synergistic combination, they have successfully designed a potential pyro-borate birefringent crystal, Li₂Na₂B₂O₅, with a short UV cut-off edge (181 nm) and large birefringence (0.095@532 nm).²¹ And they have also grown Ca(BO₂)₂ crystals exhibiting a short UV cut-off edge and larger birefringence (169 nm; 0.2471@193 nm). Based on the analysis of the structure–property relationship of Ca(BO₂)₂, they stated that the polymerized planar B_nO_2n+1 groups, e.g., B₂O₅, could generate a larger anisotropy than isolated BO₃.²² However, their opposite arrangements of B–O groups make them crystallize in the centrosymmetric (CS) space groups, which limit their further development as NLO compounds. Thus, it is clear that pyro-borates exhibiting a large birefringence and a short UV cut-off edge would also be promising DUV NLO crystals if their SHG responses can be enhanced.Based on the above-mentioned ideas, a systematical investigation has been performed on DUV pyroborates. And finally, we successfully synthesized a new NCS pyro-borate, Ba₄Ca(B₂O₅)₂F₂, which can exhibit not only a large SHG response (∼2.2 × KDP and ∼7 × α-Li₄B₂O₅) but also a short UV cut-off edge (<190 nm). Analyzing its structure, one can find that its excellent NLO properties mainly originate from the unique planar pentagonal [Ca(B₂O₅)]_∞ layer, where the B₂O₅ groups adopt the almost coplanar configurations that favor the structure to generate large SHG response and birefringence,²¹ meanwhile the terminal O atoms of B₂O₅ groups are also linked by the Ca²⁺ cations, which eliminate the dangling bonds of B₂O₅ groups and further blue-shift the UV cut-off edge. More importantly, the adjacent [Ca(B₂O₅)]_∞ layers in Ba₄Ca(B₂O₅)₂F₂ are linked by other B₂O₅ groups to form a 3D framework, which will be favorable for the material to avoid the layer habit that KBBF suffers from. In this sense, the planar pentagonal [Ca(B₂O₅)]_∞ layer is similar to the [Be₂BO₃F₂]_∞ layer in KBBF, and it can be seen as a new structure template for the design of new DUV NLO crystals, especially for the DUV pyro-borates. Herein, we will describe the synthesis, experimental and computational characterization as well as the functional properties of the new DUV NLO material, Ba₄Ca(B₂O₅)₂F₂.A polycrystalline sample of Ba₄Ca(B₂O₅)₂F₂ was synthesized by the conventional solid-state reaction and the purity was confirmed by powder X-ray diffraction (XRD) (Fig. S1^†). With the polycrystalline sample, the thermal behavior of Ba₄Ca(B₂O₅)₂F₂ was studied by the thermogravimetric (TG) and differential scanning calorimetry (DSC) measurements. The heating DSC curve shows a sharp endothermic peak at 815 °C with no obvious weight loss in the TG curve (Fig. S2^†), suggesting that Ba₄Ca(B₂O₅)₂F₂ has good thermal stability. To further investigate the thermal behavior of Ba₄Ca(B₂O₅)₂F₂, the polycrystalline sample was calcined at 840 °C and the XRD analysis showed that the calcined sample was Ba₄Ca(B₂O₅)₂F₂, Ba₂Ca(BO₃)₂ (PDF #01-085-2268), Ba₂CaB₆O₁₂ (PDF #01-075-1401) and other unknown phases (Fig. S3^†). These results illustrate that Ba₄Ca(B₂O₅)₂F₂ melts incongruently and the suitable flux is necessary for the crystal growth.With the Na₂O–PbF₂–B₂O₃ as the flux, millimeter-sized block crystals of Ba₄Ca(B₂O₅)₂F₂ were grown for the single-crystal XRD structure determination. Ba₄Ca(B₂O₅)₂F₂ crystallizes in the NCS and polar space group, P2₁ (Table S1^†). In the asymmetric unit, there are four unique Ba, one Ca, four B, ten O, and two F atom(s), which all fully occupy the 2a Wyckoff positions (Table S2^†). All B atoms are coordinated to three oxygen atoms to form the BO₃ triangles with the B–O distances ranging from 1.312(17) to 1.460(16) Å and O–B–O angles varying from 108.0(13) to 130.2(15)°. The BO₃ triangles are further connected to form two types of B₂O₅ dimers, i.e. plane B(1,3)₂O₅ and twisted B(2,4)₂O₅, which are the BBUs of Ba₄Ca(B₂O₅)₂F₂. The Ca atoms are coordinated to six oxygen atoms to form CaO₆ octahedra with the Ca–O distances ranging from 2.285(9) to 2.325(13) Å. For the Ba²⁺ cations, they exhibit three different coordination environments, Ba(1,2)O₆F₂, Ba(3)O₈F₂, and Ba(4)O₇F₂ (Fig. S4^†) with the Ba–O distances ranging from 2.585(9) to 3.250(11) Å and the Ba–F bond lengths ranging from 2.635(8) to 2.736(8) Å. Remarkably, for the F⁻ anions, each unique fluorine atom serves as a common vertex for four Ba atoms to form the FBa₄ polyhedra (Fig. S5a^†), which could be treated as fluorine-centered secondary building units (SBUs). The Ba–F–Ba angles vary from 99.0 (2) to 120.2 (3)°. The bond valence sum (BVS) calculations show the values of 1.67–1.97, 2.45, 2.88–3.10, 1.78–2.13, and 0.95–1.09, for Ba²⁺, Ca²⁺ B³⁺, O²⁻, and F⁻, respectively (Table S2^†). The BVSs of atoms are consistent with their expected oxidation states except the one from the Ca²⁺ cations. The larger BVSs of Ca²⁺ cations can be attributed to six shorter Ca–O bond lengths, which are also observed in other Ca²⁺-containing borates, such as YCa₃(VO)₃(BO₃)₄ (2.44),²⁴ Rb₂Ca₃B₁₆O₂₈ (2.29), and Cs₂Ca₃B₁₆O₂₈ (2.30).²⁵The structure of Ba₄Ca(B₂O₅)₂F₂ is shown in Fig. 1. In the structure, the plane B(1,3)₂O₅ dimer is first connected with four CaO₆ octahedra, meanwhile, each CaO₆ octahedron is also linked by four B(1,3)₂O₅ dimers through sharing their four equatorial O atoms to form a unique planar pentagonal [Ca(B₂O₅)]_∞ layer in the b–c plane (Fig. 1a, b). Then, these [Ca(B₂O₅)]_∞ layers are further linked by the twisted B(2,4)₂O₅ dimers to construct a 3D framework with Ba²⁺ cations maintaining the charge balance (Fig. 1c). Remarkably, for the arrangements of the Ba²⁺ cations and the F⁻ anions, the fluorine-centered SBU FBa₄ polyhedra are linked to construct the 2D [F₂Ba₄] infinite layer (Fig. S5b^†) with the same orientation, which further fills the apertures in the [Ca(B₂O₅)₂] framework (Fig. S5c^†). The existence of fluorine-centered SBUs would certainly have a strong influence on the local coordinate environments, and finally on the whole structure.²⁶Open in a separate windowFig. 1(a) The [Ca(B₂O₅)] layer is composed of B₂O₅ dimers and CaO₆ octahedra. (b) The planar pentagonal topology layer. The comparison of structures between (c) Ba₄Ca(B₂O₅)₂F₂ and (d) KBBF.It is very interesting that Ba₄Ca(B₂O₅)₂F₂ contains a planar pentagonal [Ca(B₂O₅)]_∞ layer, which is similar to the [Be₂BO₃F₂]_∞ layer in KBBF. The structural evolution from KBBF to Ba₄Ca(B₂O₅)₂F₂ is also shown in Fig. 1c and d. In KBBF, the BBUs are the planar BO₃ triangles, which are connected with BeO₃F in the a–b plane by strong covalent bonds to form the [Be₂BO₃F₂]_∞ layers (Fig. S6c^†) and the [Be₂BO₃F₂]_∞ layers have achieved excellent NLO properties of the KBBF crystal.⁷ However in Ba₄Ca(B₂O₅)₂F₂, the BO₃ triangles are changed into the B₂O₅ dimers, and the BeO₃F tetrahedra are substituted by the CaO₆ polyhedra. These B₂O₅ dimers are also connected by the CaO₆ polyhedra to form the interesting planar pentagonal [Ca(B₂O₅)]_∞ layer (Fig. S6d^†). More importantly, in KBBF, the adjacent [Be₂BO₃F₂]_∞ layers are connected by the weak K⁺-F⁻ ionic bonds that results in the strong layer habit of the KBBF crystals, whereas in Ba₄Ca(B₂O₅)₂F₂, the [Ca(B₂O₅)]_∞ layers are bridged by the strong covalent B–O bonds to form a stable 3D framework, which will greatly overcome the layering tendency of the KBBF crystal and facilitate the crystal growth.In addition, we also notice that the planar pentagonal [Ca(B₂O₅)]_∞ layer maybe helpful for enhancing the SHG responses of pyro-borates because small SHG responses of pyro-borates are attributed to the typical twisted configurations of the B₂O₅ groups, which are unfavorable for forming the π-conjugation and the superposition of the microscopic SHG response. For example, α-Li₄B₂O₅, a DUV transparent pyro-borate with sole B₂O₅ units as the BBUs, has a weak SHG response, which may be derived from the twisted B₂O₅ groups and non-planar arrangements (Fig. S7a^†). However, in Ba₄Ca(B₂O₅)₂F₂, the planar configuration of the pentagonal layers can assist the B₂O₅ groups to adopt a nearly coplanar arrangement (Fig. S7b^†) and effectively enhance the π-conjugation of B₂O₅ groups. The better π-conjugation of the planar B₂O₅ groups in the planar pentagonal [Ca(B₂O₅)]_∞ layer has also been confirmed by the electron orbital calculation based on the first-principles calculations.²⁷ The calculated result is shown in Fig. 2. Clearly, the prominent conjugated interactions are observed in the nearly coplanar B(1,3)₂O₅ dimers of Ba₄Ca(B₂O₅)₂F₂ (Fig. 2a), whereas it does little in the twisted B(2,4)₂O₅ dimers of Ba₄Ca(B₂O₅)₂F₂ (Fig. 2b) and two types of twisted B₂O₅ dimers in α-Li₄B₂O₅ (Fig. 2c and d). It can be expected that the nearly coplanar B₂O₅ dimers are more conducive to the large SHG response than the twisted B₂O₅ dimers. Remarkably, the similar pentagonal layers are also observed in other pyro-phosphates, such as Ba₂NaClP₂O₇, K₂Sb(P₂O₇)F, Rb₃PbBi(P₂O₇)₂, and Rb₃BaBi(P₂O₇)₂. Clearly, as pyro-phosphates are the non-π-conjugated systems, the planar pentagonal layers are only helpful for the orientation of anion groups.^28–31 However, they cannot form the better π-conjugation. Therefore, the better π-conjugation of the nearly coplanar B₂O₅ groups in planar pentagonal layers of pyro-borate Ba₄Ca(B₂O₅)₂F₂ would have a different contributing mechanism to the SHG effect with other non-π-conjugated pyro-phosphates.Open in a separate windowFig. 2The orbitals of the nearly coplanar B(1,3)₂O₅ (a) and twisted B(2,4)₂O₅ dimers (b) in Ba₄Ca(B₂O₅)₂F₂. The orbitals of two twisted B₂O₅ dimers (c and d) in α-Li₄B₂O₅.The presence of BO₃ triangles in Ba₄Ca(B₂O₅)₂F₂ is confirmed by the IR spectral measurements (Fig. S8^†). The peaks at 1362 cm⁻¹ and 1208 cm⁻¹ can be attributed to the asymmetric stretching of BO₃ groups.³² A strong band at 1069 cm⁻¹ in the IR spectrum may be associated with the stretching vibration of B–O–B in B₂O₅.^33,34 The weak absorption bands at 950, and 810 cm⁻¹ correspond to the symmetrical stretching vibrations of BO₃ and B–O–B in B₂O₅, respectively. The peaks at 751 and 615 cm⁻¹ can be attributed to the out-of-plane bending of the BO₃ groups.³⁴ Further, the UV-vis-NIR diffuse reflectance spectrum was also measured (Fig. S9^†), which shows that Ba₄Ca(B₂O₅)₂F₂ is transparent down to the DUV region with a UV cut-off edge less than 190 nm (corresponding to a large band gap of 6.2 eV), which is comparable to the newly developed NLO-active borates, such as RbB₃O₄F₂ (<190 nm), CsZn₂BO₃X₂ (X₂ = F₂,Cl₂, and FCl)) (<190 nm) and so on.^35–38 The short cut-off edge demonstrates the potential application of Ba₄Ca(B₂O₅)₂F₂ as a DUV NLO crystal.As Ba₄Ca(B₂O₅)₂F₂ crystalizes in the NCS space group P2₁, it possesses the SHG response, which was measured by the Kurtz-Perry method with the well-known NLO material KH₂PO₄ (KDP) as a reference.³⁹ As shown in Fig. 3, the SHG intensities of Ba₄Ca(B₂O₅)₂F₂ increase with the increase of particle sizes, indicating that Ba₄Ca(B₂O₅)₂F₂ is type-I phase-matchable. The SHG intensity of Ba₄Ca(B₂O₅)₂F₂ at the particle size of 150–212 μm is about 2.2 times that of KDP, and is larger than that of KBBF (1.2 × KDP) or comparable with those newly reported UV NLO crystals, i.e. γ-Be₂BO₃F (2.3 × KDP),⁶ β-Rb₂Al₂B₂O₇ (2 × KDP),⁴⁰ Li₄Sr(BO₃)₂ (2 × KDP),⁴¹ CsB₄O₆F(∼1.9 × KDP).² In addition, as we know, the SHG response of Ba₄Ca(B₂O₅)₂F₂ is the largest among all the known DUV transparent borates with B₂O₅ units (Table S4^†). Its SHG response (2.2 × KDP) is about seven times larger than that of α-Li₄B₂O₅ (0.3 × KDP), another DUV transparent pyro-borate with sole B₂O₅ units.Open in a separate windowFig. 3(a) Phase-matching curve, i.e., particle size vs. SHG intensity, data for Ba₄Ca(B₂O₅)₂F₂ and KH₂PO₄ (KDP) as reference. The solid curve is a guide for the eye, not a fit to the data. (b) Oscilloscope traces showing SHG intensities for Ba₄Ca(B₂O₅)₂F₂ and KDP.To understand the origin of the excellent optical properties of Ba₄Ca(B₂O₅)₂F₂, we also carried out the first-principles calculations.²⁷ It shows that Ba₄Ca(B₂O₅)₂F₂ has an indirect band gap of 6.34 eV (Figures S10a^†), which is in accordance with the experimental results. The valence band maximum (VBM) of Ba₄Ca(B₂O₅)₂F₂ is mainly composed of the orbitals in Ba, and O atoms, while the conduction band minimum (CBM) is dominantly composed of the orbitals in Ba, B, and O atoms. Therefore, the band gap of Ba₄Ca(B₂O₅)₂F₂ is mainly determined by Ba atoms and B₂O₅ groups. Based on the calculated electron structure, the NLO coefficients of Ba₄Ca(B₂O₅)₂F₂ are also calculated. The largest NLO coefficient of Ba₄Ca(B₂O₅)₂F₂ is d₂₂ = −0.524 pm V⁻¹, which is about 5 times lower than that of α-Li₄B₂O₅ (d₂₄ = −0.102 pm V⁻¹) (Table S5a^†), which is matched with the experimental one. Further, the SHG-weighted density maps of Ba₄Ca(B₂O₅)₂F₂ are shown in Fig. 4. These reveal that B₂O₅ dimers make the dominant contribution (72.7%) to the total SHG effect (Table S5b^†). The band-resolved SHG analysis can also conclude that B–O orbitals in Ba₄Ca(B₂O₅)₂F₂ contribute more to the SHG response than those in α-Li₄B₂O₅ (Fig. S10b, S10c^†), indicating that the arrangements of B₂O₅ dimers in Ba₄Ca(B₂O₅)₂F₂ is more beneficial for the large SHG response. And different from α-Li₄B₂O₅, F-centered secondary building units (SBUs) exist in the structure of Ba₄Ca(B₂O₅)₂F₂, and they are further linked to construct 2D [F₂Ba₄]_∞ infinite layers, which could help B₂O₅ groups arrange in a planar pattern (Fig. S5^†).²⁶ So, based on the above analysis, we can conclude that the nearly coplanar B₂O₅ dimers in the planar pentagonal layer and the SBU FBa₄ tetrahedra make a significant contribution to the SHG response of Ba₄Ca(B₂O₅)₂F₂.Open in a separate windowFig. 4The SHG-weighted density maps of the virtual electron process (a) and virtual hole process (b) of d₂₂ for Ba₄Ca(B₂O₅)₂F₂.     

Strong non-Arrhenius behavior at low temperatures in the OH + HCl → H2O + Cl reaction due to resonance induced quantum tunneling

The OH + HCl → H₂O + Cl reaction releases Cl atoms, which can catalyze the ozone destruction reaction in the stratosphere. The measured rate coefficients for the reaction deviate substantially from the Arrhenius limit at low temperatures and become essentially independent of temperature when T < 250 K, apparently due to quantum tunneling; however, the nature of the quantum tunneling is unknown. Here, we report a time-dependent wave packet study of the reactions on two newly constructed potential energy surfaces. It is found that the OH + HCl reaction possesses many Feshbach resonances trapped in a bending/torsion excited vibrational adiabatic potential well in the entrance channel due to hydrogen bond interaction. These resonance states greatly induce quantum tunneling of a hydrogen atom through the reaction barrier, causing the reaction rates to deviate substantially from Arrhenius behavior at low temperature, as observed experimentally.

The OH + HCl reaction possesses many Feshbach resonances trapped in the hydrogen bond well in the entrance channel, which substantially enhance the reaction rates at low temperatures.

In the classical picture, a chemical reaction with an energetic barrier can only occur at collision energies higher than the barrier, which leads to the well-known Arrhenius formula for chemical reaction rates. However, chemical reactions can happen at energies below the reaction barrier through quantum tunneling,^1–3 resulting in the deviation of the reaction rates from the Arrhenius behavior at low temperatures. The effect of quantum tunneling on the reaction rates increases with decreasing reaction temperature, and hence becomes especially important in low-temperature environments such as the interstellar medium and atmospheric processes.^4,5 Reaction resonances are quasi-trapped quantum states in the transition state region with some lifetime, and can substantially promote quantum tunneling through the reaction barrier. Over the past decades, great efforts have been devoted to detecting resonances in chemical reactions and to studying their structures and dynamics.^6–14 Theoretical studies on the O(³P) + HCl → OH + H reaction using the accurate ³A′′ potential energy surface revealed that the tunneling induced by the resonances trapped in the van der Waals well in the reactant channel can substantially enhance the thermal rate constants at low temperatures.^15–17 A combined experimental and theoretical investigation discovered that resonance-induced quantum tunneling dramatically enhances the reactivity of the F + p-H₂ → HF + H reaction,¹⁴ and is fully responsible for the unusually high chemical reactivity of the reaction in the low temperature interstellar medium. Recently, a quantum dynamics calculation showed that the presence of two resonance peaks strongly influence the rotational quenching of HF (j = 1, 2) with H, leading to an up to two-fold increase in the thermal rate coefficients at the low temperatures characteristic of the interstellar medium.¹⁸ Thus, understanding quantum tunneling, and in particular, resonance-induced quantum tunneling, in chemical reactions is of general interest and fundamental importance to low-temperature chemistry.The OH + HCl → H₂O + Cl reaction is of great importance in atmospheric chemistry because it releases Cl atoms from one of the principal chlorine-containing species in the stratosphere, HCl. The Cl atoms generated from the reaction can catalyze the ozone destruction reaction in the stratosphere, which was responsible for the formation of the ozone hole over Antarctica.¹⁹ Since the chlorine-catalyzed ozone destruction is proportional to the steady-state Cl atom concentration, which is directly controlled by the rate of the reaction, extensive studies have been carried out to measure the rate coefficient for the reaction with high accuracy over a wide range of temperatures.^20–29 The measured rate coefficients exhibit a small activation energy of a few hundred K, deviate substantially from the Arrhenius limit at low temperatures and become essentially independent of temperature when T < 250 K.^25,28,29 In addition, a large H/D kinetic isotope effect has also been found.^21,22,26,28 All these observations suggest the presence of an important quantum tunneling effect in the reaction.The dynamics of this reaction and its reverse have also attracted great attention in the past decades. In particular, the endothermic Cl + H₂O → HCl + OH reaction with a late barrier has been extensively investigated as a benchmark system for mode specificity and bond selectivity chemistry.^30–35 Recently, the construction of two high-quality potential energy surfaces (PESs) in its ground electronic state using the PIP-NN method have substantially advanced the theoretical study of the dynamics and kinetics of the system. The first PES was based on a large number of ab initio data points calculated at the multi-reference configuration interaction (MRCI) level of theory by Li, Dawes, and Guo (LDG),³³ and the second one was fitted to ab initio energy points obtained using an explicitly correlated unrestricted coupled-cluster method with single, double, and perturbative triple excitations (UCCSD(T)-F12b) and the augmented correlation-consistent polarized valence triple-zeta (aug-cc-pVTZ, or AVTZ) basis set by Zuo, Zhao, Guo, and Xie (ZZGX) with a fitting error of 6.9 meV.³⁶ The static barrier height is 2.86 and 2.23 kcal mol⁻¹, respectively, for these two PESs. On both PESs, there exists a well of about 3.5 kcal mol⁻¹ in the OH + HCl entrance channel due to the hydrogen bond (HB) interaction between OH and HCl. Extensive quantum dynamics studies on the PESs have revealed many interesting features of the reaction in both directions. In particular, time-dependent wave packet calculations for the title reaction with OH in the ground and vibrational excited states found one or two broad peaks in the total reaction probabilities, which are presumed to be the signature of the resonances supported by the reactant complex well.^36,37 However, the impact of these peaks on the reaction rates has not been investigated. Ring-polymer molecular dynamics (RPMD) calculations^38,39 were also carried out on both PESs to compute the thermal rate coefficients for the reaction.^40–42 It was found that the RPMD rates on the LDG PES underestimate the experimental data, while the RPMD rates on the latest ZZGX PES agree with the experimental results much better, and do not decrease further when the temperature falls below 300 K, apparently due to quantum tunneling. Unfortunately, RPMD calculations cannot provide any clue regarding the nature of the quantum tunneling.Therefore, despite the significant progress that has been made in theoretical studies of the system, some key issues still remain to be addressed: Are the broad peaks in the total reaction probabilities obtained from the time-dependent wave packet calculations indeed the signature of the resonances in the reaction? If not, do there exist resonances in the reaction? How do resonances affect the reaction rates at low temperature? Here, we report a quantum dynamics study of the reaction on two new and more accurate PESs. Good agreement is achieved between the rate coefficients calculated on these two PESs and the experimental data. Our calculations reveal that the HB well in the entrance channel of OH + HCl supports many low energy resonance states. These resonance states substantially enhance the quantum tunneling effect and have an important impact on the reaction rates at low temperatures.In order to improve the fitting accuracy of the ZZGX PES, we constructed two new PESs using the fundamental invariant neural network (FI-NN) method, fitted to ∼70 000 ab initio energy points calculated at the UCCSD(T)-F12a and UCCSD(T)-F12b levels of theory, respectively, both with the AVTZ basis set. The fitting RMSE is 3.07 and 3.12 meV, respectively, for the F12a and F12b PES, which is about half that for the ZZGX PES.³⁶ The spin–orbit coupling of the channels of both the reagent OH and the product Cl have been included using FI-NN fitting to about 38 000 points calculated at the MRCI/aug-cc-pVTZ level of theory with a fitting error of 0.19 meV. The CASSCF wave-function with an active space of (5e, 3o) was used as a reference for MRCI. Details of the new PESs are provided in the ESI^†. The static barrier height for PESa is 0.088 eV (0.095 eV with SO correction included), and that for PESb is 0.097 eV (0.104 eV). As can be seen from Tables S1 and S2,^† the geometries and energies of all the stationary points for the F12b PES without SO correction are in good agreement with the ZZGX PES. Table S2^† also shows the corresponding complete basis set (CBS) energies for these stationary points based on AVTZ, AVQZ and AV5Z calculations. Because the F12b energies are slightly closer to the CBS results, we will present the dynamical results obtained on the F12b PES in the main text and provide those for the F12a PES in the SI.On the new PESs, we carried out potential-averaged five-dimensional (PA5D) time-dependent wave packet^43,44 calculations to obtain the total reaction probabilities for the reaction by freezing the non-reacting OH bond in its ground vibrational state. Tests revealed that the PA5D treatment is capable of providing reaction probabilities for the ground rovibrational initial state that are essentially identical to those obtained using the full six-dimensional approach, as shown in Fig. S2.^†Fig. 1(A) shows the total reaction probabilities for the HCl + OH reaction as a function of collision energy calculated on the F12b PES with both reagents in the ground rovibrational state at propagation times of 60 000, 120 000, 360 000, and 2 400 000 a.u. At high collision energies, the reaction probabilities converge quickly with respect to the propagation time, and one can barely see any difference among the four reaction probability curves, which exhibit smooth increases with collision energy. However, in the low collision energy region, large differences appear for different propagation times. At t = 60 000 a.u., the reaction probability presents a smooth curve with some small and broad oscillations, as was observed in the wave packet calculations of Guo and coworkers.³⁶Open in a separate windowFig. 1(A) Total reaction probabilities for the ground initial state of the OH + HCl → Cl + H₂O reaction on the F12b PES at wave packet propagation times of T = 60 000, 120 000, 360 000, and 2 400 000 a.u. (B) Same as (A), except showing the collision energy between 0.0 and 0.04 eV. The crosses mark the points for which the wavefunctions are shown in Fig. 2. (C) Total reaction probabilities for some partial waves J = 0, 30, 60, and 90 as a function of the collision energy. (D) Same as (C) except showing the collision energy between 0.0 and 0.04 eV.When the propagation time is increased to t = 120 000, the reaction probabilities at collision energies below 0.05 eV increase substantially. With further increasing the propagation time to 360 000 a.u., many oscillatory structures emerge at collision energies below the barrier height of 0.104 eV, in particular in the very low collision energy region as shown in Fig. 1(B). These sharp structures become fully converged after around 2 400 000 a. u. of wave packet propagation (∼58 ps). The reaction probability even at a collision energy close to zero reaches 3–4%. The convergence of the total reaction probabilities on the F12a PES is very similar to that on the F12b PES (Fig. S3^†), except that the final converged reaction probabilities for these two PESs exhibit a small shift, apparently due to slightly different barrier heights. Therefore, it is clear that reaction resonances exist in the title reaction in the very low collision energy region, and that these resonance states substantially induce quantum tunneling and enhance the reactivity. The lifetimes for these resonance states are quite long, with many being longer than 6.5 ps and having corresponding widths smaller than 0.1 meV. Fig. 1(C) presents converged (t = 2 400 000 a.u. ≈ 58 ps) total reaction probabilities for the total angular momentum J = 0, 30, 60, and 90. With increasing J, the reaction probability curve shifts to higher energy. In the low collision energy region (<0.05 eV), the total reaction probabilities for J = 30 exhibit rich oscillatory structures as in the J = 0 case (Fig. 1(D)), which are expected to have a great influence on the rate constant at low temperature. With further increase of the total angular momentum, the influence of these resonances on the total reaction fades due to the centrifugal barrier, which prevents the low-energy wave function from entering the well. They only leave a small trace in the total reaction probabilities at low energies for J = 60, and do not have any effect for J = 90.To understand the nature of these resonances, we calculated scattering wave functions at two collision energies (1.08 and 4.26 meV) with the peak reaction probabilities indicated by x in Fig. 1(B). Fig. 2(A) shows the two dimensional (2D) contour at the collision energy of 1.08 meV in the Jacobi coordinates HCl bond length (r_H–Cl) and center of mass distance between OH and HCl (R_HCl–OH), with the other coordinates integrated. As can be seen, the wave function is localized in the HB well region in the entrance channel with a few nodes in the R coordinate and no node in the r_H–Cl coordinate. Inspection of the scattering wave function for the bending and torsion coordinates reveals nodes exist in these coordinates (Fig. S4 and S5^†). The 2D contour in the coordinates of R_HCl–OH and r_H–Cl at the collision energy of 4.26 meV shown in Fig. 2(B) looks similar to that shown in Fig. 2(A), except with more nodes in the R direction. Therefore, the observed resonance states in the reaction are Feshbach resonances trapped in a bending/torsion excited vibrationally adiabatic potential (VAP) well in the reactant complex region due to the HB interaction.Open in a separate windowFig. 2Reactive scattering wave functions for the OH + HCl → Cl + H₂O reaction on the F12b PES in the two Jacobi coordinates R(HCl–OH) and r(H–Cl) with other coordinates integrated at the collision energies of 1.08 (A) and 4.26 meV (B). The contour lines are the corresponding 2D PESs along the two reactive bonds R(HCl–OH) and r(H–Cl) with other coordinates optimized. The geometries for the saddle point and the HB minimum are displayed in (A). The coordinate units in the figures are a₀. Fig. 3(A) shows the accurate rate constants for the initial ground rovibrational state, k_g, based on the probabilities for J = 0, 30, 60, 90 using the uniform J-shifting approach with a temperature-dependent shifting constant.^45–47 A test shows that the J-shifting scheme based on these four individual J values only introduces a few percent error to the rate constants in the temperature region considered here (Fig. S6^†). As can be seen from the figure, with decreasing temperature, k_g first decreases rapidly from 1000 K to 700 K, then decreases slowly. It reaches a minimum at T ≈ 260 K, and increases slowly with further decrease of the temperature. At temperatures lower than 300 K, the rate constants for the ground rovibrational initial state are larger than the measured thermal rate coefficients, with k_g being larger than k_exp by ∼70% at T = 200 K.Open in a separate windowFig. 3(A) Accurate rate constants, k_g, for the initial ground rovibrational state of the HCl + OH → H₂O + Cl reaction calculated on the F12b PES, in comparison with k^JS_g (obtained using the J-shifting approximation) and k^NR_g (based on the background reaction probabilities for J = 0 with the resonance contribution removed shown below); (B) the background reaction probabilities up to E = 0.1 eV by connecting some valleys of the reaction probabilities marked by x.Also shown in Fig. 3(A) are the rate constants for the ground rovibrational initial state, k^JS_g, obtained from the J = 0 reaction probabilities using the J-shifting approximation (see ESI^† for details). As can be seen, the J-shifting approximation works very well at high temperatures around 1000 K, but begins to overestimate the rates with decreasing temperature. At T = 500 K, k^JS_g is about 10% higher than the true rate. At temperatures below 300 K, the J-shifting approximation underestimates the reaction rate, with k^JS_g being smaller than k_g by about 16% at T = 200 K. Overall, the J-shifting approximation works fairly well for the ground rovibrational initial state, although there are numerous resonance peaks in the reaction probabilities in the low collision energy region.Now we consider the issue of the effect of the resonance structures found in the reaction on the rate constant. For a reaction system with isolated resonances, it is rather straightforward to remove the resonance contributions from the reaction probabilities by fitting the resonance peaks to some Lorentzian functions and to obtain smooth background scattering probabilities, as demonstrated in the F + HD reaction and recently in the inelastic scattering of H + HF (ref. 6 and 18). However, the OH + HCl reaction possesses numerous highly overlapped resonances in the low energy region, as shown in Fig. 1(A), and it is impractical to fit these resonance peaks as accurately as the isolated resonances. Instead, we obtained an approximate background reaction probability curve, which is shown in Fig. 3(B), by smoothly connecting some resonance valleys as shown in the figure. The rate constants k^NR_g calculated using the background curve shown in Fig. 3(B) with the J-shifting approximation were compared with the original k_g in Fig. 3(A). Given the fact that the J-shifting approximation works fairly well for k^JS_g, as shown in Fig. 3(A), it is very reasonable to expect that will work even better for the k^NR_g values obtained from the background reaction probabilities with the resonance contribution removed. As can be seen, k^NR_g exhibits rate behavior typical for systems with a low barrier with some quantum tunneling effects. At T = 1000 K, k^NR_g is essentially identical to k_g, but decreases much faster than k_g as the temperature drops. At T = 200 K, k_g is larger than k^NR_g by a factor of about 5.6 (9.8 × 10⁻¹³vs. 1.74 × 10⁻¹³ cm³ s⁻¹), indicating that the reaction probabilities in the low collision energy region due to resonances substantially enhance the rate constants for the reaction at low temperatures. It is worthwhile to note that for systems with overlapped resonances like that shown in Fig. 3(B), the reaction probabilities at the valleys must be considerably higher than the true background reaction probabilities; therefore, the background curve shown in Fig. 3(B) is the upper limit of the background reaction probabilities and the rate based on the curve shown in Fig. 3(A) is also the upper limit of the rate for the ground rovibrational initial state without the resonance contributions. Therefore, the true enhancement due to the resonances must be larger than that shown in Fig. 3(A).For reliable comparison with the measured thermal rate coefficients, we must take into account the contributions from all the thermally populated initial states of the reagents. Due to the very large number of thermally populated rotational states for this reaction even at 200 K, we opted to calculate the cumulative reaction probabilities, NE(E), (the sum of the reaction probabilities for all the initial states with a fixed total energy) for the reaction from which the thermal rate constants can be reliably evaluated.^48–52 The transition state wave packet calculations were carried out to obtain NE(E) using the details given in the ESI^†. Due to huge computational efforts required to obtain the cumulative reaction probabilities for J > 0, we only calculated NE(E) for J = 0 and employed the J–K-shifting approximation^45,49,53 to obtain the thermal rate constant. NE(J = 0, E) as a function of total energy measured with respect to the ground rovibrational energy of OH and HCl is presented in Fig. S7.^†In Fig. 4, we present thermal rate constants for the reaction calculated on both the F12a and F12b PES, together with the rate constants for the ground rovibrational state (k_g) and the previous experimental measurements.^{21,22,24–29} As can be seen, the thermal rate coefficients are smaller than k_g over the entire temperature region, indicating that reagent rotation excitations diminish the reaction rates. Overall, the thermal rate coefficients calculated on both PESs agree with the experimental results rather well, with the F12a PES slightly overestimating and the other PES slightly underestimating compared to the experimental measurements. As observed in the experiments, the thermal rate coefficients decrease quickly with decreasing temperature in the high-temperature region, but decrease much more slowly at low temperatures, in particular at T < 300 K, substantially deviating from Arrhenius behavior.Open in a separate windowFig. 4Thermal rate constants of the HCl + OH → H₂O + Cl reaction calculated on both the F12a and F12b PES, compared with rate constants for the ground rovibrational state, the RPMD rates on ZZGX PES and the previous experimental measurements.^{21,22,24–29} Experimental data are taken from Husain et al. (in circles),^21,22 Molina et al. (in diamonds),²⁴ Ravishankara et al. (in upward-pointing triangles),²⁵ Smith et al. (in squares),²⁶ Sharkey et al. (in triangle left),²⁷ Battin-Leclerc et al. (in downward-pointing triangles),²⁸ and Bryukov et al. (in leftward-pointing triangles).²⁹Also shown in Fig. 4 are the RPMD rates calculated (see ESI^† for details) on the F12a, F12b and ZZGX PESs.⁴² As shown, the RPMD rates on the F12b PES agree with those on the ZZGX PES extremely well, although the barrier heights for the F12b PES are higher by 9 meV. The present RPMD rates agree with the J–K-shifting quantum rates rather well, except at 200 K, at which the RPMD rates are overestimated by about 40%. As a result, the RPMD rate is higher than the experimental value at T = 200 K, in particular for the F12a PES. Previous studies have shown that RPMD tends to underestimate reaction rates in the strong quantum tunneling region, even for systems with resonances such as the O(³P) + HCl → OH + H reaction;⁵⁴ however, it overestimates the rates for the F + H₂ reaction with pronounced post-barrier Feshbach resonances⁵⁵ and some insertion reactions.⁵⁶ The discrepancy between the RPMD and J–K-shifting quantum rate at T = 200 K could be caused by a possible underestimation of the rate by the J–K-shifting approximation, as in the J-shifting approximation of the ground rovibrational initial state shown in Fig. 3(A). On the other hand, the rapid increase in the rate with decreasing temperature (from 250 K to 200 K) could also be a feature of bimolecular reaction rates in the high-pressure limit for reactions with a pre-reactive minimum.⁵⁷ RPMD rate theory in its original form employs a free energy calculation,^38,39 which allows thermalization in the pre-reactive minimum, sampling tunneling pathways with energies not accessible in the low-pressure limit. Therefore, RPMD models a rate process that resembles rates in the high-pressure limit, which could lead to an overestimation of the rate at low temperature. A recent study also reported a similar issue with RPMD.⁵⁸ The authors attributed it to spurious resonances in RPMD favoring energy transfer in the pre-reactive minimum, which is similar to our argument. Certainly, more quantum dynamics calculations to provide rigorous quantum rates at low temperatures to assess the accuracy of the J–K-shifting approximation and the RPMD method on this reaction system with strong HB interaction in the entrance channel would be highly desirable.Therefore, the OH + HCl reaction possesses many long-lifetime Feshbach resonances trapped in a bending excited VAP well in the entrance channel due to HB interaction. These resonance states substantially induce quantum tunneling of the hydrogen atom through the reaction barrier and enhance the reactivity in the low-collision region. Consequently, the reaction rate for the reaction becomes essentially independent of temperature in the low-temperature region and deviates substantially from Arrhenius behavior, as observed experimentally. The resonance states in the reaction are very different in location from those in the F + H₂/H₂O reactions,^{8,11,12,14,59–63} which are trapped in the VAP well in the product channel. Furthermore, they are trapped in the bending/torsion excited VAP well with HCl in the ground vibrational state, unlike those in the F + H₂/H₂O reactions with HF in vibrationally excited states.^8,63,64 In nature, these resonance states in the reaction arise from HB interaction as in the F + H₂O (v = 0) reaction,^11,62,63 but are different from those in the F + H₂/HD/HOD (v = 1) reactions^{8,10,12,14,64–66} due to chemical bond softening. Because the resonance states are trapped in the bending/torsion excited VAP well, their lifetimes are much longer than those observed in the F + H₂/H₂O reactions. As a result, they should have an important impact on the differential cross sections. More efforts, and in particular more joint efforts involving theory and experiment, should be devoted to studying these Feshbach resonances in this reaction of great atmospheric importance in detail.     

Directed evolution of a cyclodipeptide synthase with new activities via label-free mass spectrometric screening

Directed evolution is a powerful approach to engineer enzymes via iterative creation and screening of variant libraries. However, assay development for high-throughput mutant screening remains challenging, particularly for new catalytic activities. Mass spectrometry (MS) analysis is label-free and well suited for untargeted discovery of new enzyme products but is traditionally limited by slow speed. Here we report an automated workflow for directed evolution of new enzymatic activities via high-throughput library creation and label-free MS screening. For a proof of concept, we chose to engineer a cyclodipeptide synthase (CDPS) that synthesizes diketopiperazine (DKP) compounds with therapeutic potential. In recombinant Escherichia coli, site-saturation mutagenesis (SSM) and error-prone PCR (epPCR) libraries expressing CDPS mutants were automatically created and cultivated on an integrated work cell. Culture supernatants were then robotically processed for matrix-assisted laser desorption/ionization time-of-flight (MALDI-ToF) MS analysis at a rate of 5 s per sample. The resulting mass spectral data were processed via custom computational algorithms, which performed a multivariant analysis of 108 theoretical mass-to-charge (m/z) values of 190 possible DKP molecules within a mass window of 115–373 Da. An F186L CDPS mutant was isolated to produce cyclo(l-Phe–l-Val), which is undetectable in the product profile of the wild-type enzyme. This robotic, label-free MS screening approach may be generally applicable to engineering other enzymes with new activities in high throughput.

A robotic workflow for directed evolution of new enzymatic activities via high-throughput library creation and label-free MS screening.

Mimicking random mutagenesis and natural selection in the laboratory,^1–3 directed enzyme evolution is valuable in improving a variety of properties of biocatalysts (i.e., catalytic activity, stability, substrate specificity, and enantioselectivity).^4–6 While it is straightforward to generate thousands of enzyme variants using established methods such as error-prone PCR (epPCR), site-saturation mutagenesis (SSM), and DNA shuffling, phenotypic screening remains a major bottleneck due to the substrate and product versatility of biocatalysts.^7,8 Optical assays are widely applied in high-throughput screening (HTS) during directed evolution, often requiring a surrogate substrate or a coupling assay with a colorimetric or fluorogenic readout.^9,10 However, it is very challenging to apply such approaches to discover new catalytic activities. On the other hand, mass spectrometry (MS) is label-free and suitable for untargeted molecular profiling.¹¹ The high sensitivity and chemical resolution of MS further highlight its potential in discovering new products, because new catalytic activities are often weak when first emerging from enzyme promiscuity.¹²We and others have recently developed a range of high-throughput MS screening methods for directed protein evolution.^13–16 However, the label-free advantage of MS has not been fully demonstrated in engineering new enzymatic activities, possibly due to the difficulties in untargeted MS screening. Particularly, it requires careful standardization and optimization of sample preparation, MS acquisition, and data processing, which are necessary to minimize experimental noise and spot the weak signals of a new product. On the other hand, biofoundries provide an emerging infrastructure to assist the design–build–test–learn (DBTL) cycles in biological engineering via robotic standardization and parallelization.^17–19 Using an integrated biofoundry, here we report a workflow for unlabeled MS screening of recombinant libraries to rapidly isolate enzyme mutants that catalyze the formation of new products. This new workflow extends our previous matrix-assisted laser desorption/ionization time-of-flight (MALDI-ToF) MS-based screening approach from agar colonies¹⁶ to liquid cultures in industry-standard microplates for better uniformity. Unlike colony biomass, liquid culture media often contain high concentrations of nonvolatile components, which interfere with MALDI matrix crystallization and cause severe ion suppression during MS analysis. Therefore, sample preparation steps, such as liquid–liquid extraction and solid-phase extraction, need to be incorporated and automated.Cyclodipeptide synthases (CDPSs) are a family of enzymes that catalyze the formation of a diketopiperazine (DKP) from two aminoacyl-tRNA substrates (aa-tRNAs), which is an important pharmacophore for modern drug development^20–22 (Fig. 1A and S1^†). To produce new DKP derivatives for therapeutic and industrial applications, the catalytic mechanism of CDPSs has been studied^23,24 to guide engineering. AlbC (239 aa) is the first identified CDPS protein, taking phenylalanyl-tRNA^Phe (Phe-tRNA^Phe) and leucyl-tRNA^Leu (Leu-tRNA^Leu) as substrates to synthesize cyclo(l-Phe–l-Leu) (cFL) in its native host Streptomyces noursei.^25,26 Recombinant E. coli with AlbC overexpression is able to produce more cyclodipeptide derivatives in addition to cFL.²⁷ To date, eight CDPSs have been structurally characterized,^26,28–32 and mutagenesis studies reveal that the residues within the P1 and P2 catalytic pockets are the key determinants of substrate specificity. However, the detailed mechanism of substrate recognition and catalysis is still elusive, and CDPSs are recognized as recalcitrant targets for rational engineering.^33,34 To our knowledge, large-scale screening of CDPS variants has not been reported, possibly due to the lack of applicable HTS assays.Open in a separate windowFig. 1(A) Scheme of diketopiperazine (DKP) biosynthesis catalyzed by cyclodipeptide synthase (CDPS). (B) Workflow of the matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-ToF MS)-based high-throughput screening method for directed evolution of AlbC protein in E. coli.Here we develop and apply untargeted MS screening on a biofoundry for directed evolution of AlbC mutants that form new CDP products. The workflow consists of strain library creation, colony picking, microtiter cultivation, organic solvent extraction, transfer of the sample to a MALDI target and acquisition of mass spectra, followed by data processing and visualization (Fig. 1B). For library creation, AlbC variants were generated by either SSM or epPCR approaches and expressed from a pET28a plasmid under control of a T7 promoter. Upon genetic transformation of E. coli Rosetta(DE3), individual clones were transferred by a colony picker into microtiter plates. Subsequent cultivation, inducible production by addition of isopropyl β-d-1-thiogalactopyranoside (IPTG), ethyl acetate extraction, and MS sample preparation were performed using an integrated robotic workcell in the Shenzhen Synthetic Biology Infrastructure. This workcell integrates common instruments for synthetic biology, including a liquid-handling station, shaking incubators, centrifuge, plate reader, and so on. Organic extracts of the liquid cultures were spotted onto a MALDI target and overlaid with 4-CHCA matrix solution. MS targets were then manually transferred to a stand-alone MALDI mass spectrometer in this study, although a robotic configuration has been previously reported to automate this step.³⁵We first applied the workflow to analyze the wild-type (WT) AlbC-expressing strain, Rosetta(DE3)/pET28-AlbC(WT). After cell cultivation, inducible production, and sample preparation, MS analysis was performed using an Autoflex MALDI-ToF mass spectrometer in the reflector positive ion mode (Fig. 1). In the resulting MALDI mass spectra, we observed [M + H]⁺ ion peaks with m/z values corresponding to cFL (m/z 261.17), cFY (m/z 311.15), cFM (m/z 279.12), cYL (m/z 277.16), cFF(cYM) (m/z 295.15), cLL (m/z 227.16) and cMM (m/z 263.12), all of which were absent from the control strain Rosetta(DE3)/pET28 (Fig. S2^†). Production of these CDP molecules was further confirmed by examining tandem MS results using LC-MS in the multiple reaction monitoring (MRM) mode (Table S3^†), where ion fragmentation patterns were consistent with the literature.^26,27 Moreover, the observed production profile was also consistent with previous studies.^26,27 These results validated our workflow to detect CDP products from microplate cultures of AlbC-expressing E. coli using MALDI-ToF MS.To engineer AlbC mutants that synthesize new products, we first focused on the key residues in the substrate-binding pockets. AlbC employs a “ping-pong” catalytic mechanism: the initial step transfers the aminoacyl moiety of the first aa-tRNA onto a conserved serine, leading to the formation of an aminoacyl enzyme intermediate; then, the aminoacyl enzyme reacts with the aminoacyl moiety of the second aa-tRNA to form a dipeptidyl enzyme.³² Previous biochemical experiments demonstrated that two binding pockets P1 and P2 of AlbC accommodate the aminoacyl moieties of the two aa-tRNA during the biosynthesis²⁴ (Fig. 2). Only a limited number of mutations were examined on P1 and P2 residues,^25,26 possibly due to the lack of HTS assays for large-scale analysis. Instead, we applied our workflow to create and screen the SSM libraries of select residues, including 10 residues in the binding pockets (L33, V65, L119, L185, L200, M152, M159, I204, T206 and P207) and 4 residues (R99, R101, R102 and D205) outside the pockets which have interactions of the tRNA moiety reported in the literature (Fig. 2A).Open in a separate windowFig. 2(A) Structural analysis of the AlbC (PDB ID, 3OQV). Enlarged view of the catalytically active pocket. The possible catalytic residues are shown in green (pocket-1) and orange (pocket-2) and basic residues on helix α4 are labelled in cyan. (B) Heatmap of the relative activity change in the cFL production levels of mutation of 14 amino acid residues in AlbC. WT residues are labeled with a black dashed rectangle. Dark blue boxes indicate that a specific mutant was not covered in randomly picked clones (activity assigned as −1).For SSM libraries, we adopted the “22c-trick” strategy to design degenerative primers, and 94 library clones were randomly picked for each residue library to reach a >98.6% probability of full library coverage of the 20 canonical amino acids.³⁶ In addition to library variants, the WT and control strains were also included in the same 96-well plate. Overall, the abovementioned workflow (Fig. 1B) takes approximately 5 s for sample preparation and MS analysis for each mutant culture. Tentative ion peaks of cyclodipeptides were assigned based on theoretical m/z values (ESI Section 1^†) using the MetaboAnalyst webserver.³⁷ For each ion peak, one-way analysis of variance (ANOVA) was used to evaluate statistical differences between the mean peak areas of a library variant and those of the control group. AlbC gene mutations were revealed by Sanger sequencing for all 14 SSM library members.To visualize the MALDI-ToF MS screening results, a heatmap was generated based on the cFL production of library members relative to that of WT (Fig. 2B), and we observed a general consistency between our results and literature data (Fig. 3A), respectively, which was similar to a previous study.²⁶ Also, mutagenesis of basic residues (R99, R101 and R102) outside the pockets led to the decline of cFL production in the mutants, although R101 was more tolerant to mutations than R99 and R102. Furthermore, the substitution of D205 in the β6-α8 loop to alanine enhanced AlbC catalytic activity as observed previously.²⁵ On the other hand, some previously unnoticed phenomena were observed. For example, in the D205 SSM library, relative ion intensities of cFL (m/z 261.17) in many mutants are higher than that of WT (Fig. S3^†). The top 3 mutants, D205M, D205K and D205R, were subsequently analyzed by LC-MS/MS. The results not only confirmed augmented synthesis of cFL, but also revealed increased production of other leucine-containing products including cYL, cLL and cML (Fig. 3B). Also interestingly, the discovery that most mutations of the T206 residue in the P2 pocket abolished the biosynthesis of the main product cFL was unprecedented (Fig. S3 and ESI Section 2^†). When examining MALDI mass spectra in detail, we noticed that although the T206F mutation greatly reduced cFL production, cFF and cFY products were largely not affected. Replacing the small threonine side chain with a bulky aromatic phenylalanine side chain could affect the activity of WT, because the phenyl ring structure increased the steric hindrance of T206F, thus impairing binding with its substrate and causing loss of its enzyme activity for cFL (Fig. 5A). These results suggested that T206 is also a key residue for the selectivity of the second Leu-tRNA substrate binding in the pocket. Unfortunately, we did not observe any new CDP molecules produced from the 14 SSM library mutants.Impact of AlbC mutations on the main product cFL

Lithium achieves sequence selective ring-opening terpolymerisation (ROTERP) of ternary monomer mixtures

Heteroatom-containing degradable polymers have strong potential as sustainable replacements for petrochemically derived materials. However, to accelerate and broaden their uptake greater structural diversity and new synthetic methodologies are required. Here we report a sequence selective ring-opening terpolymerisation (ROTERP), in which three monomers (A, B, C) are selectively enchained into an (ABA′C)_n sequence by a simple lithium catalyst. Degradable poly(ester-alt-ester-alt-trithiocarbonate)s are obtained in a M_n range from 2.35 to 111.20 kDa which are not easily accessible via other polymerisation methodologies. The choice of alkali metal is key to achieve high activity and to control the terpolymer sequence. ROTERP is mechanistically compatible with ring-opening polymerisation (ROP) allowing switchable catalysis for blockpolymer synthesis. The ROTERP demonstrated in this study could be the first example of an entirely new family of sequence selective terpolymerisations.

A sequence selective ring-opening terpolymerisation of epoxides with CS₂ and phthalic thioanhydride yielding poly(ester-alt-ester-alt-trithiocarbonates) is reported.

Synthetic polymers are now more in demand than ever before and looking at their annually increasing production, a polymer free society is at best a vague memory rather than a vision for the future.¹ As most commodity polymers are based on chemically inert aliphatic –C–C– backbones, most polymer waste shows unappreciable degradation with respect to the polymer''s time in application.^2,3 In answer to the ever-increasing amount of plastic pollution, much effort focuses on the exploration of new heteroatom containing polymers which, because of their more polar bonds making up the backbone, are more susceptible to degradation via physical, chemical and biochemical pathways and even facilitate new chemical recycling methods.^4–8 Besides, there is also a constant demand for entirely new materials to enable technological innovation making new methodologies to synthesise heteroatom containing polymers necessary.Arguably one of the most popular methods to make such polymers is the ring-opening polymerisation (ROP) of a heterocycle A.^9–11 These polymerise under release of their associated ring strain energy to make polymers (A)_n such as poly(thio)ethers, poly(thio)ester and poly(thio)carbonates, only to name a few. Early on it has been realised that in some cases the ROP of three or four-membered heterocycles A can be coupled to the insertion of typically heteroallenes or cyclic anhydrides B to generate alternating copolymers (AB)_n.^12–14 The underlying requirements for (AB)_n polymerisations are a combination of kinetic factors, i.e. monomer A inserting orders of magnitude faster into the active catalyst growing chain-bond than monomer B, and chemoselectivities, i.e. insertion of monomer A resulting in type A active catalyst growing chain-bond that does not show appreciable reactivity with A but only with B (and vice versa). Prominent examples of this alternating (AB)_n ring-opening copolymerisation (ROCOP) include CO₂/epoxide ROCOP yielding polycarbonates and cyclic anhydride/epoxide ROCOP yielding polyesters.^15–17 Sulfur analogous are also accessible such as polythiocarbonate from CS₂/(epoxide or thiirane) and polythioesters from thioanhydride/(epoxide or thiirane) ROCOP.^18–31 Such sulfur rich polymers are attractive high-refractive index materials that can show improved crystallinity and degradability over their all-oxygen analogues and in some cases enable chemical polymer to monomer recycling.^32–39Most relevant to this study is a report by Werner and coworkers on lithium alkoxide catalysed CS₂/epoxide ROCOP yielding poly(monothio-alt-trithiocarbonate)s featuring R–O–C( S)–O–R and R–S–C( S)–S–R links (Fig. 1).^40,41 Such a polymer sequence is unexpected, as the formal product of alternating insertion would be a poly(dithiocarbonate) with R–O–C( S)–S–R links. Furthermore, the polymer shows an unusual “head-to-head-alt-tail-to-tail” selectivity meaning that the R–O–C( S)–O–R links sit adjacent to the tertiary CHR₃ positions of the ring opened epoxide (i.e. “head” position) and that the R–S–C( S)–S–R links sit adjacent to the secondary CH₂R₂ position of the ring opened epoxide (i.e. “tail” position). This sequence let the authors postulate a mechanism involving tail-selective epoxide ring-opening by a dithiocarbonate chain end which is formed by CS₂ insertion into an alkoxide intermediate. The resulting alkoxide intermediate was proposed to backbite into the adjacent dithiocarbonate link which after a rearrangement process resulted in an O/S exchange of the chain end. The rearrangement transforms the alkoxide into a thiolate chain end and the adjacent dithiocarbonate R–O–C( S)–S–R into a monothiocarbonate R–O–C( S)–O–R. CS₂ insertion of the thiolate generates a trithiocarbonate R–S–C( S)–S–R which after epoxide insertion regenerates the alkoxides. In contrast to alkoxides sitting adjacent to R–O–C( S)–S–R links, alkoxides next to R–S–C( S)–S–R links were not proposed to undergo backbiting and O/S exchange. Importantly the initial regioselectivity of the epoxide ring-opening was preserved throughout the rearrangement which explained the “head-to-head-alt-tail-to-tail” selectivity. Interestingly lithium appeared to be crucial as other alkali metal alkoxides failed to catalyse this ROCOP, while more sophisticated catalysts result in much more uncontrolled polymer sequences. Hence it appears that the Li controls which alkoxide intermediate precisely undergoes O/S exchange and to which degree this rearrangement occurs, but the reasons behind the special role of Li remains to be explored. Relatedly the ROCOP of thioanhydrides with epoxides has also been reported and similar exchange processes have been proposed as side reactions.²³ Although not directly proven, this mechanism seemed reasonable and let us hypothesise that lithium catalysts could grant a general access to control the O/S exchange process in ROCOP and even in the polymerisation of ternary monomer mixtures. Furthermore, we reasoned that the two distinct chain ends formed via O/S exchange, i.e. alkoxide and thiolate (type A vide supra), could enable discrimination between two different type B monomers and enable sequence selective terpolymerisations. It should be noted that reports exist in which mixtures of A, B and C (e.g. epoxide A, cyclic anhydride B and CO₂ C) either yield random terpolymers, (AB)_n-ran-(AC)_m poly(esters-ran-carbonate), or block-terpolymers, (AB)_n-b-(AC)_m, polyester-b-polycarbonate. In this case the monomer sequence depends on catalyst selection and reaction conditions, but sequence selective terpolymers, e.g. (ABC)_n or (ABAC)_n, are unknown.^42–48 The hypothesis of O/S exchange here led us to discover a new type of polymerisation, sequence selective ring-opening terpolymerisation (ROTERP) that we report in this contribution. ROTERP produces poly(ester-alt-ester-alt-trithiocarbonates), i.e. (ABA′C)_n sequences, from a mixture of the monosubstituted epoxides propylene oxide (PO) or butylene oxide (BO) A, phthalic thioanhydride (PTA) B and CS₂ C, employing simple lithium salts as the catalyst. Furthermore, model reactions proof the previously postulated O/S rearrangement that enable ROTERP and elucidate the role of the lithium catalyst. Finally, we employed ROTERP in so-called switchable catalysis, in which the onset of ROTERP stops the occurrence of ROP, for the construction of blockpolymers.Open in a separate windowFig. 1(Top) CS₂/epoxide ROCOP and postulated mechanism involving a central O/S exchange reaction, (middle) phthalic thioanhydride (PTA)/epoxide ROCOP and (bottom) PTA/CS₂/epoxide ROTERP reported in this study. R = Me, Et; [R_n] = polymer chain.ROTERP of mixtures of PTA, PO and CS₂ at loadings typically employed in ROCOP catalysis with lithium hexamethyldisilazide (LiHDMS) or lithium benzyloxide (LiOBn) as the catalyst at loadings in the range of 1 eq. LiX : (6.25–500 eq.) PTA : (31.25–2500 eq.) PO : (62.5–5000 eq.) CS₂ yield polymeric materials in 95% selectivity at 80 °C (see 40,47 Spectroscopic analysis of the isolated polymer reveals surprisingly clean NMR spectra given the potential for statistical terpolymerisation of these three monomers. The ¹H NMR spectrum (Fig. 2) shows two main aryl resonances for a symmetrically substituted terephthalate unit from ring opened PTA (δ = 7.72 and 7.56 ppm) as well as one main resonance for the CHMe (δ = 5.42 ppm; head position of the ring-opened epoxide) and CH₂ (δ = 3.90–3.40 ppm; tail position of the ring opened epoxide) groups respectively stemming from the ring opened PO. Correspondingly the ¹³C{¹H} NMR spectrum (ESI Fig. S1^†) reveals the almost exclusive presence of trithiocarbonate R–S–C( S)–S–R (δ = 222.8 ppm) and arylester R–C( O)–O–R (δ ∼166.6 ppm) resonances (94–98%) alongside minor thioester R–C( O)–S–R (δ = 192.7 ppm) resonances (2–6%). ¹H and 2D NMR spectra (ESI Fig. S3^†) show that trithiocarbonate units are positioned adjacent to CH₂ groups while arylesters are connected to the tertiary CHMe groups. The spectra remain unchanged after multiple precipitations from DCM : MeOH or THF : pentane and DOSY NMR shows that all ¹H NMR resonances diffuse at the same rate confirming that these are part of the same species. Furthermore, no other type of thiocarbonate R–(O/S)–C( O/S)–(O/S)–R are part of the polymer. Linkage identity was further substantiated by the ATR-IR spectrum (ESI Fig. S5^†) of these polymers showing an arylester C O stretch at = 1716 cm⁻¹ as well as a thiocarbonate C S stretch at = 1062 cm⁻¹ (ESI Fig. S5^†). Accordingly, the polymers are obtained as yellow solids due the presence of the C S chromophore (λ_abs = 435 nm, ESI Fig. S6^†). MALDI-TOF analysis unfortunately only led to decomposition of the materials and no signals could be identified as previously reported for sulfur-rich polymers.^26,38 Nevertheless ¹H NMR allows some conclusions regarding the topology as both for LiHMDS or LiOBn initiation, resonances for the HMDS and OBn groups can be identified to be part of the purified polymers. Insertion of alkalimetal alkoxides and amides into CS₂ yielding alkali dithiocarbonates and dithiocarbamates have been previously reported.^40,79 This makes initiation via CS₂ insertion likely which defines one end of the polymer and hence indicates that chains are linear rather than cyclic. As ROTERP is followed by CS₂/epoxide coupling once all PTA is consumed (vide infra) we infer that chains are terminated by CS₂ because heteroallene insertion products are generally established to be the resting states of heteroallene/heterocycle coupling reactions.¹²Data showing PTA/CS₂/epoxide ROTERP under different conditions

Mapping paratopes of nanobodies using native mass spectrometry and ultraviolet photodissociation

Following immense growth and maturity of the field in the past decade, native mass spectrometry has garnered widespread adoption for the structural characterization of macromolecular complexes. Routine analysis of biotherapeutics by this technique has become commonplace to assist in the development and quality control of immunoglobulin antibodies. Concurrently, 193 nm ultraviolet photodissociation (UVPD) has been developed as a structurally sensitive ion activation technique capable of interrogating protein conformational changes. Here, UVPD was applied to probe the paratopes of nanobodies, a class of single-domain antibodies with an expansive set of applications spanning affinity reagents, molecular imaging, and biotherapeutics. Comparing UVPD sequence fragments for the free nanobodies versus nanobody·antigen complexes empowered assignment of nanobody paratopes and intermolecular salt-bridges, elevating the capabilities of UVPD as a new strategy for characterization of nanobodies.

Ultraviolet photodissociation mass spectrometry is used to probe the paratopes of nanobodies, a class of single-domain antibodies, and to determine intersubunit salt-bridges and explore the nanobody·antigen interfaces.

Antibodies must possess high specificity toward target antigens to enable recognition and activation of an immune response. Because of this specificity, antibodies or antibody fragments are being increasingly explored for deployment in diagnostic assays, vaccine design and other therapeutic applications.¹ As such, unravelling the collection of noncovalent interactions and structural features that endow antibodies with specificity is a primary goal in biomedical research and critical for development of new biotherapeutics.² Methods to decipher antibody–antigen interactions have advanced significantly in recent years, and well-established methods including X-ray crystallography, mutagenesis techniques, cryo-EM, and hydrogen–deuterium exchange mass spectrometry remain the most versatile methods.^2–5 Another new strategy, native mass spectrometry (MS) has emerged as a powerful tool for structural biology, including analysis of macromolecular complexes and antibodies.^6–10 In this technique, rapid but gentle ionization and transfer of protein complexes into the gas-phase via electrospray ionization (ESI) of aqueous solutions of near-physiological ionic strength preserves noncovalent interactions, allowing elucidation of ligand binding and stoichiometry of complexes.^6–9 Coupled to advanced methodologies including ion mobility (IMS) or tandem mass spectrometry (MS/MS), native MS has proven an innovative strategy for interrogating protein structure, revealing topology, distinguishing conformations, identifying ligand binding sites, and determining folding thermodynamics.^6–9,11,12 As one example of a premier application in biotherapeutic development, native MS has been utilized to rapidly measure stoichiometry, heterogeneity, and stability of antibody·antigen complexes.^10,13–17Detailed structural analyses by native MS often rely on controlled dissociation or disassembly of protein complexes via ion activation in MS/MS workflows.^7,8,18–21 Collision-based dissociation methods comprise the most ubiquitous ion activation techniques and are capable of dismantling protein assemblies into subunits, detaching ligands, and facilitating determination of stoichiometries.^7,8,18 However, the slow-heating process of collision induced dissociation (CID) causes protein unfolding and less effective fragmentation of the peptide backbone, limiting characterization of native protein complexes.^7,8,18 Significantly more structural information can be acquired by employing alternative ion activation techniques.^7,8,18 For example, surface induced dissociation (SID) is a collision-based method that promotes disassembly of protein complexes into subunits through a single high energy collision that minimizes protein unfolding and enables robust characterization of native protein assemblies.^20,21 SID has proven especially versatile for the analysis of quaternary structure, as protein complexes disassemble to produce subcomplexes and subunits that reflect the native architecture.^20,21 Most impressive, the outcome of SID correlates with the magnitude of protein interfacial area (cleavage of the weakest protein:protein interfaces) and has been integrated into computational modelling workflows to enhance the accuracy of protein assembly predictions, emphasizing the value of native MS/MS for structural interrogation.^22–24 Alternatively, electron-based dissociation methods result in cleavage of the protein backbone to produce sequence fragments that are enhanced at surface exposed or flexible protein regions, informing topology (through the preservation of noncovalent interactions that prevent separation of reaction products) and degree of protein disorder.^7,8,19,25Also a method sensitive to protein structure, ultraviolet photodissociation (UVPD) is a photon-based ion activation method that has demonstrated exceptional use for structural biology. In particular, propensities of polypeptide backbone cleavages induced by 193 nm UVPD correlate with backbone flexibility and arrangement of non-covalent interactions, an outcome related to the likelihood of separation (and detection) of nascent product ions after a backbone cleavage event. Product ions enmeshed by stabilizing non-covalent interactions are less likely to separate (i.e. UVPnoD), causing an apparent suppression of backbone fragmentation. This correlation empowers characterization of conformational changes induced by point mutations, ligand binding, and protein complexation by UVPD.^26–33 Additionally, the high sequence coverage and rapid timescale of photodissociation enable detailed analyses of protein gas-phase structure, even informing proton sequestration with single residue resolution.³⁴ As native MS and UVPD increasingly gain broader utility for new protein applications, development and establishment of strategies to routinely study protein structure become imperative to cement these methodologies as cornerstones in the fields of structural biology and biotechnology that encompass development of new therapeutics, imaging agents, diagnostics and drug delivery agents.One fascinating new class of biotherapeutics are nanobodies, single domain antibodies derived from the variable domain of functional heavy chain antibodies found in camelids and certain shark species.^35,36 In contrast to conventional ∼150 kDa heterotetrameric immunoglobulin (IgG) antibodies, nanobodies feature a single peptide chain and overall reduced size of ∼15 kDa. Nanobodies offer high stability, solubility, affinity, and specificity, features that have propelled these single domain antibodies as a valuable alternative biotechnology.^37–42 Concurrently, these same features facilitate native MS analysis, for which decreased size, increased solubility and antigen affinity are favorable for rapid and routine analysis of nanobody complexes, circumventing analytical challenges and tedious sample preparations, such as proteolysis and deglycosylation steps,^14,17,43 often required for native MS of typical IgG·antigen complexes. One recent native MS study mapped the location of the epitope of influenza A hemagglutinin (HA1) bound to an antibody based on a decrease in backbone cleavages of HA1 when bound in the Ab·2HA1 complex relative to the free HA1 antigen during UVPD-MS analysis.³¹ This innovative approach motivated our interest in exploring an intriguing inverse strategy to map paratopes of nanobodies. In the present work, native MS and 193 nm UVPD are showcased as a valuable combination for determination of intersubunit salt-bridges and nanobody·antigen interfaces, ultimately localizing nanobody paratopes.Three nanobodies^44–46 with distinct proteinaceous antigens of green fluorescent protein (GFP), ribonuclease A (RNAseA), and porcine pancreatic amylase (PPA), referred to here as Gnb,⁴⁴ Rnb,⁴⁵ and Anb,⁴⁶ respectively, were targeted to evaluate native MS and UVPD for characterization of nanobody·antigen complexes. Each of these nanobodies interact with the antigen via differing contributions of the complementarity-determining regions (CDR) 1–3 and framework residues to the protein interface, which vary in surface area from 554 Ų to 683 Ų to 1062 Ų for Rnb, Gnb, and Anb, respectively. These three pairs of nanobodies and respective antigens were first ionized individually using native conditions (Fig. S1^†), and as nanobody·antigen complexes (Fig. 1). In each case, native MS produced the expected 1 : 1 complex in the full mass spectrum (MS1), in accordance with known crystal structures, in a range of charge states. A single charge state of the antigen-bound (bound state) nanobody was isolated and subjected to 193 nm UVPD (Fig. 1B, D and F), resulting in disassembly to release the free nanobody and antigen as well as sequence fragments from backbone cleavage of each protein, the latter of which will be the focus of UVPD analysis. UVPD mass spectra of each free nanobody are shown in Fig. S2.^† The corresponding backbone cleavage maps for the free nanobodies and corresponding nanobody·antigen complexes are shown in Fig. S3.^†Open in a separate windowFig. 1Native MS and UVPD of nanobody·antigen complexes. (A) MS1 of Gnb·GFP complex and (B) UVPD of the 13+ charge state. (C) MS1 of Rnb·RNAseA complex and (D) UVPD of the 11+ charge state. (E) MS1 of Anb·PPA complex and (F) UVPD of the 15+ charge state. Insets display expanded views of m/z regions populated by sequence fragments of low relative abundance. For UVPD, 1 laser pulse at 3 mJ was applied.Backbone cleavages induced by UVPD have been shown to be favored at protein regions with higher flexibility, typically ones less stabilized by networks of non-covalent interactions which might prevent separation and release of fragment ions.^26–31,47 Accordingly, comparing abundances of fragment ions produced by free and bound nanobodies should reveal regions in which backbone fragmentation is suppressed or enhanced, thus uncovering those residues involved in stabilizing interactions with the antigen and effectively localizing the paratope. Ten types of fragment ions (a, a + 1, b, x, x + 1, y, y − 1, y − 2, and z, where +1, −1, and +2 indicate the gain or loss of hydrogen atoms) commonly generated by UVPD were monitored across the three nanobodies in both the free and bound states. Among the collection of ions detected, a- and x-type ions are the most prevalent and dominant for the nanobodies and complexes, thus providing the greatest sequence coverage. Those fragment ions displaying statistically significant differences in abundances (p < 0.05, n = 5) upon complexation of the nanobody to the antigen are shown in Fig. S4–S6^† and mapped onto the backbone position cleaved to generate the fragment ions. The high diversity of fragment types characteristic of UVPD originates from competing pathways: direct dissociation from excited electronic states yields a/a + 1/x/x + 1 ions, and internal conversion to the ground state following intramolecular vibrational energy redistribution (IVR) produces b/y fragments.^32,33 IVR processes may preferentially sever weak non-covalent interactions, whereas direct dissociation from excited states occurs on a faster time-scale minimizing disruption of non-covalent interactions.^32,33 Consequently, the latter dissociation pathways and respective products, a/a + 1/x/x + 1 ions, are best suited to evaluate antigen-induced changes in nanobody topology. Variations in abundances of these four fragment types tended to be greater at the interface and CDRs, and inspection of the color-coded maps in Fig. S4–S6^† reveals that in most cases the abundances of these a/x-type ions decreased for the nanobody·antigen complexes relative to the free nanobodies, signifying suppression of backbone fragmentation.To underscore the impact of antigen binding on UVPD, significant changes in backbone fragmentation (ΔUVPD) based on differences in summed abundances of a/a + 1/x/x + 1 fragments for each free nanobody versus nanobody·antigen complex are shown in Fig. 2, along with color maps highlighting residues at the CDRs and protein interfaces. Generally, apparent suppression of backbone fragmentation (e.g., less efficient separation of nascent fragment ions owing to stabilizing non-covalent interactions) is the greatest at or adjacent to the interface residues, an outcome which is especially notable and consistent for patches of residues near CDR3 in Gnb (Fig. 2A), near CDR1 and CDR3 in Rnb (Fig. 2B), and near CDR2 and CDR3 in Anb (Fig. 2C). These patterns in suppression of backbone fragmentation upon antigen binding correlate with structural features of the respective crystal structures: main antigen contacts in Gnb are predominantly present on CDR3;⁴⁴ only CDR1 and CDR3 participate in antigen binding for Rnb;⁴⁵ CDR2 and CDR3 primarily mediate antigen binding for Anb.⁴⁶ Although nanobodies characteristically feature a disulfide bond spanning C23 to a paired cysteine directly N-terminal to the CDR3, cleavage of the disulfide bond via UVPD was sufficient to unlock the nanobody and allow release of sequence ions from other concomitant backbone cleavages within this region, including CDR1 and CDR2. Moreover, any statistically significant enhancement of backbone cleavages induced by antigen binding was sparse and remote from interaction sites. These infrequent increases in UVPD fragmentation for nanobody·antigen complexes relative to the free nanobody possibly indicate disruption of non-covalent interactions and increased flexibility in those limited regions of the nanobody. However, this sporadic enhancement of backbone cleavages upon antigen binding occurs in stark contrast to more prevalent suppression of backbone cleavages spanning larger swaths of neighboring residues in the nanobody·antigen complexes. Overall, the observed suppression of fragmentation at the protein·protein interface serves as a strong basis for the development of native MS-UVPD strategies to discern nanobody paratopes.Open in a separate windowFig. 2Suppression and enhancement of backbone cleavage sites based on abundances of UVPD fragment ions induced for each nanobody by antigen binding for (A) Gnb, (B) Rnb, and (C) Anb. ΔUVPD heat plots display significant differences (p < 0.05, n = 5) for the abundances of a- and x-type fragment ions between the free and bound nanobody. Blue and red indicate suppression and enhancement of fragment abundances, respectively, for the nanobody upon complexation. Positions that display no significant change are shown in grey, while white indicates small significant changes. Color maps highlighting interface residues and CDRs are also included for each nanobody. ΔUVPD values correspond to the fragment abundance per residue for the bound state minus the fragment abundance per residue for the free state. All product ions had a signal-to-noise ratio > 3.Contrary to classical IgGs where paratopes are primarily localized to CDRs,⁴⁸ nanobody paratopes feature greater diversity in terms of residue identity and position, and also include the involvement of framework residues as well as the potential absence of interactions from certain CDRs, obfuscating assignment of paratopes.⁴⁸ By leveraging the trends in the reduction of fragmentation observed in Fig. 2 for the three nanobody·antigen complexes relative to the free nanobodies, a strategy for the approximation of surface patches contributing to the paratope was developed. Because the most structurally significant changes of fragmentation related to antigen binding are demarcated by apparent suppression of backbone cleavages for stretches of neighboring residues, the UVPD data was analyzed by averaging abundances of fragment ions originating from backbone cleavages across every five residues (non-overlapping box-car average) prior to comparing fragment abundances for the free and bound states using Welch''s t-test (n = 5). A stringent significance cutoff of p < 0.001 was subsequently applied to limit false assignments of protein sections involved at the interface. Protein sections (5 residues long) displaying significant suppression of backbone cleavages (i.e. reduction in abundances of fragment ions originating from backbone cleavages in each 5 residue segment) according to this method are plotted onto the crystal structures in Fig. 3. Impressively, only protein sections adjacent to the protein·protein interfaces displayed significant suppression for Gnb and Anb. Similarly, Rnb mostly featured suppression adjacent to the interface, while only two sections remote from the interface (spanning residues 15–24 and 45–49) were also suppressed. Furthermore, a significant enhancement of fragmentation was only observed for Rnb near the N-terminus and C-terminus (Fig. S7^†), remote from the paratope. This Rnb complex featured the smallest interfacial area, which may lead to instability, compaction, or distortion during ion transmission, resulting in the unexpected suppression and enhancement at these regions. Regardless, UVPD suppression was predominant at the interfaces of each of the three nanobody complexes, presenting a new strategy for the approximation of nanobody residue patches that contribute to the paratope.Open in a separate windowFig. 3The a and x fragment ion abundances originating from backbone cleavages were averaged across every 5 residues for the free and bound forms of each nanobody. Sections displaying significant UVPD suppression upon complexation (p < 0.001, n = 5) were mapped onto the crystal structure as green spheres for (A) Gnb, (B) Rnb, and (C) Anb. Residue positions displaying significant suppression of backbone cleavages, interface residues, and CDR regions are shown for each nanobody as color maps. For comparative purposes, the CDR regions of each nanobody are demarcated on the crystal structures in Fig. S8.^†Tracking charge states of specifically a- and x-type ions produced by 193 nm UVPD also informs proton sequestration along the protein sequence,³⁴ a feature that is capitalized on here to identify the formation of inter-subunit salt-bridges. In this strategy, charge states for each detected a- or x-type ion are weighted based on intensity and plotted for each backbone cleavage position of the nanobody. For example, if both a₄²⁺ and a₄³⁺ have equal intensities, the weighted average charge state at backbone cleavage position 4 (between residues 4 and 5) would be 2.5. Because electrostatic interactions, such as salt-bridges, are highly stable in the gas-phase,^49,50 applying this method to monitor changes in proton sequestration across free and bound states promises to reveal the locations of inter-subunit salt bridges introduced by complexation, if charge partitioning during subunit ejection is due to heterolytic scission of salt-bridges as previously proposed.^26,51,52 Indeed, this is demonstrated in the analysis of the Rnb·RNAseA complex, for which one inter-chain salt bridge has been identified by X-ray crystallography between nanobody R107 and antigen E111. Charge site analysis of the a-ion series for free Rnb (6+ charge state) (Fig. 4A) displays discrete step-changes in charge states at residues R39, between residues 43–47 (suggesting protonation at R45), R68, Q111, and H128, indicating localized protons at these sites. Although coverage of the x-ion series is sparser, it nonetheless corroborates proton localization at Q111 as well as near the N-terminus in the span of residues Q2–L5. Protonation of side-chains is not unexpected for basic amino acids like R and K, but protonation of backbone heteroatoms is also possible for non-basic residues like Q at either the amide oxygen or the amide nitrogen of the peptide bond.^53,54 The step analysis reported here localizes all 6 protons of the 6+ charge state of free Rnb.Open in a separate windowFig. 4Weighted average charge of a-type and x-type fragment ions attributed to Rnb produced by UVPD of (A) free Rnb (6+ charge state) and (B) Rnb·RNAseA (11+ charge state), delineated based on the backbone cleavage site along the sequence of the nanobody.Charge site analysis of Rnb in the Rnb·RNAseA complex was not as comprehensive, but nonetheless many charge sites were elucidated (Fig. 4B). Specifically, R29, R46, and R68 remained protonated, while a shift occurred from protonation at Q111 in free RNAseA to protonation of the span of residues 108–106 in the Rnb·RNAseA complex, indicating proton sequestration at R107, a residue engaged in a salt-bridge with the antigen in solution. Charge migration observed upon antigen binding based on this charge-site analysis thus evinces formation of salt-bridges between binding partners. Similarly for Gnb·GFP, charge site analysis enabled localization of multiple charges in both the free and bound states (Fig. S9^†) including protonation at R36 on Gnb, which is involved in electrostatic interactions with E142 on the GFP antigen. Additionally, a charge was located at R58 for bound Gnb that is absent for free Gnb. Although slightly higher than canonical distance cutoff of 4 Å for salt bridges, Gnb R58 is within 4.5–5.5 Å of the E172 and D173 side-chains of GFP according to the crystal structure. Salt-bridge formation spanning this distance may be possible in the gas-phase, according to charge site analysis derived from the UVPD data. The companion residues, E45 and E104, of the nanobody are also engaged in putative interchain salt-bridges when bound to GFP based on the X-ray structure, however, localization of deprotonation sites of acidic residues is not possible in the positive ion mode.For Anb, coverage of a- and x-type fragments from the Anb·PPA complex was less comprehensive and precluded detailed charge site analyses (Fig. S10^†). Regardless, data for Gnb and Rnb demonstrate that basic residues engaged in inter-subunit salt-bridges maintain a proton upon disassembly of the nanobody·antigen complexes, highlighting an exceptional use of MS/MS to localize gas-phase salt-bridges between protein subunits. However, it is also noted that these changes in protonation sites may be caused by reasons other than those postulated here. Specifically, it is possible that proton migration is caused by vibrational redistribution of deposited energy from photoabsorption, leading to proton mobilization,^53,54 and is not due to heterolytic scission of salt bridges. These results are nonetheless intriguing and may guide future interpretations of charge partitioning during UVPD of protein complexes.Following incredible advances in native MS, routine and rapid analysis of biotherapeutics for quality control and drug development has become commonplace.^13–17 At the same time, UVPD has been propelled over the past decade as a premier ion activation method capable of uncovering structural features of proteins that are not revealed by other MS/MS methods.⁴⁷ As shown in this study, we extended the combination of native MS and UVPD to characterize nanobody·antigen complexes, particularly aiming to showcase UVPD for mapping the binding interface. The pattern of fragment ions generated by UVPD for nanobody complexes resulted in the discernment of inter-chain salt-bridges. Additionally, tracking trends in suppression of fragmentation enabled the localization of nanobody paratopes using only micromolar quantities of nanobodies and circumventing some of the limitations of traditional structural biology methods such as NMR and X-ray crystallography. We anticipate that further improvements in data analysis/informatics methods will further extend the level of structural detail gleaned from the very dense UVPD mass spectra generated for large macromolecular assemblies akin to nanobody·antigen complexes.     

An integrated platform approach enables discovery of potent,selective and ligand-competitive cyclic peptides targeting the GIP receptor

In any drug discovery effort, the identification of hits for further optimisation is of crucial importance. For peptide therapeutics, display technologies such as mRNA display have emerged as powerful methodologies to identify these desired de novo hit ligands against targets of interest. The diverse peptide libraries are genetically encoded in these technologies, allowing for next-generation sequencing to be used to efficiently identify the binding ligands. Despite the vast datasets that can be generated, current downstream methodologies, however, are limited by low throughput validation processes, including hit prioritisation, peptide synthesis, biochemical and biophysical assays. In this work we report a highly efficient strategy that combines bioinformatic analysis with state-of-the-art high throughput peptide synthesis to identify nanomolar cyclic peptide (CP) ligands of the human glucose-dependent insulinotropic peptide receptor (hGIP-R). Furthermore, our workflow is able to discriminate between functional and remote binding non-functional ligands. Efficient structure–activity relationship analysis (SAR) combined with advanced in silico structural studies allow deduction of a thorough and holistic binding model which informs further chemical optimisation, including efficient half-life extension. We report the identification and design of the first de novo, GIP-competitive, incretin receptor family-selective CPs, which exhibit an in vivo half-life up to 10.7 h in rats. The workflow should be generally applicable to any selection target, improving and accelerating hit identification, validation, characterisation, and prioritisation for therapeutic development.

mRNA display generates vast datasets of protein binders. Bioinformatic clustering of the sequences combined with high throughput synthesis and analysis methods allow efficient prioritisation of hits for in vivo experiments.

In recent years, peptides have emerged as powerful therapeutic agents¹ and the most prominent examples can match antibodies in selectivity and potency. This development has been greatly aided by breakthroughs in half life extension technologies by formulation or chemical modification. A key advantage of engineered peptides is their ease of production, and while antibodies are generally considered to be too large for efficient oral uptake in therapeutically relevant amounts, the small size of peptides allows the potential for oral dosing with significant progress made in recent years.^2,3 Despite these advantages, most drug development projects heavily rely on antibodies, most likely due to the ease of de novo antibody discovery by evolutionarily optimised methods. Modern in vivo or in vitro display technologies, such as phage display and mRNA display, allow for the identification of potent and specific peptide ligands through iterative screening of peptide pools containing trillions of randomised sequences.^4,5 These technologies have the potential to yield a plethora of de novo peptides which could be engineered into powerful therapeutics. However, these techniques are surprisingly scarce in clinical pipelines (compared to antibodies or small molecules) despite being known for decades and most peptides in late-stage discovery are derivatives of naturally occurring molecules.¹ Challenges in hit prioritisation and time-consuming validation could at least partly be responsible for this observation. While classical approaches would rely on Sanger sequencing of the most abundant clones (yielding very few hits), the advent of next generation sequencing (NGS) has revolutionized the information content available from de novo peptide screens.⁴ In the majority of studies, however, this information is unused and only a handful of the most abundant sequences are followed up by chemical synthesis for testing of biological activity.⁵ We rationalized that powerful bioinformatic data mining methods would best utilise the vast information contained in peptide display datasets. To prove this, we combined chemical synthesis and high throughput binding analysis on data from mRNA display. We show that these methods identify more diverse series of hits with different modes of binding as starting points for peptide drug discovery efforts, especially when a functional binder to a specific binding pocket is desired (Scheme 1).Open in a separate windowScheme 1Workflow of peptide hit identification, prioritisation, and validation.We chose the type 2 GPCR glucose-dependent insulinotropic peptide receptor (GIP-R) as a challenging and clinically relevant test case.⁶ Its natural ligand GIP and the close relative glucagon-like peptide 1 (GLP-1) are incretins, i.e., hormones which are secreted after oral nutrient uptake and which augment glucose-dependent insulin release from pancreatic beta cells.⁷ GLP-1 receptor agonists have been reported to not only have effects in metabolic disease through increasing insulin release, but also promote satiety by delaying gastric emptying, and are being investigated for their protective effects in obesity,^8,9 heart disease,¹⁰ and diabetic kidney disease.¹¹ Conversely, the biology of GIP is less clearly understood, and both GIP-R agonists¹² as well as antagonists^13,14 are under investigation as potential therapeutics in the fields of type 2 diabetes and obesity. Previous approaches to target the GIP-R thus far have made use of analogues of GIP itself,^15,16 or GIP R specific antibodies^14,17–20 and peptide display work was based on dual GIP-GLP-1 analogue libraries.^14,19,20 As such, the GIP-R represented an ideal target for our mRNA based workflow as we strived to identify potent, ligand competitive, and subfamily selective de novo binders, which could provide valuable tools to understand GIP-R biology and serve as starting points for drug discovery efforts.We employed mRNA display to identify cyclic peptide ligands towards the biotinylated extracellular domain (ECD) of the human GIP receptor (residues 22–138 of hGIP-R), which has been shown to retain binding to GIP.²¹ Initially, we established conditions for efficient cyclisation by disulphide bond formation for representative library peptides (Fig. S3^†), before conducting iterative rounds of mRNA display^5,22 using a nucleotide library encoding two conserved cysteines for macrocyclization flanking a region of 4 to 12 random amino acid sequence. Negative selections were performed against biotin-loaded streptavidin-functionalised magnetic beads, followed by positive selection against bead-immobilised hGIP-R ECD. Sufficient library enrichment was obtained for hGIP-R after five rounds of selections (percentage recovery relative to input > 0.1%) (Fig. S4a^†). NGS of enriched output cDNA was carried out and peptides were clustered by similarity for each selection round (R1–R5). All sequences exceeding six reads in R5 (the top 3160 sequences) were compared against each other using pairwise local alignments in order to derive a complete distance matrix. Single-linkage hierarchical clustering was then performed on the distance matrix, choosing a threshold for cutting the relative dendrogram into clusters based on a statistical optimality criterion.²³ For this dataset, the sequence similarity threshold for assigning a sequence to a cluster was set at 0.38, and the fewest number of members that a cluster was allowed to contain was 20. This generated 13 clusters (assigned letters A through M), where the largest cluster, A, contained 1880 sequences, while the two least populated clusters L and M contained 21 sequences each, and the unclustered sequences were assigned to a “noise” cluster (termed cluster 0), encompassing 263 sequences (Fig. 1).Open in a separate windowFig. 1High throughput pairwise clustering analysis of 3160 Round 5 output sequences selected against hGIP-R ECD. Lighter colour indicates closer similarity between individual sequences (ESI Data 1^†).In the next step, we selected peptides from each cluster (A-M, 0) based on sequence diversity and abundance in R5 for parallel solid-phase peptide synthesis (SPPS) in a 96 well format. The N-terminus was capped with an acetyl group (Ac) to mimic the fMet in mRNA-display and a C-terminal FLAG tag was added to each peptide to ensure peptide solubility in buffer by serving as a source of negative charge, analogous to C-terminal mRNA that is present during the selection screens.²⁴ Peptides were assigned identifiers with cluster ID and their rank order based on their abundance in R5 (e.g. B_5). Following synthesis and cleavage, the peptides were macrocyclized via disulphide bond formation in 20% DMSO in buffer.²⁵ Peptide identity, purity, and complete macrocyclization was confirmed by UPLC-MS and peptide concentrations were determined by UPLC-CAD (ESI Data 3^†). For the initial high throughput characterisation, peptides were used without further purification (average purity 50–70%). To establish the binding properties, we performed high-throughput single-concentration biolayer interferometry (BLI) measurements, using biotinylated hGIP-R ECD immobilised on streptavidin-functionalised biosensors. Select peptides with favourable binding potencies were resynthesized, purified and their K_d determined by multiple concentration BLI measurements (Fig. 2). The results were generally in good agreement with the single concentration data.Open in a separate windowFig. 2(a and b) Single concentration hGIP-R ECD binding K_d values and dissociation rates for the most abundant peptide clusters, and corresponding abundances/reads of peptide sequences in the final round of selection against GIP-R ECD. All peptides contain an N-terminal Ac group and C-terminal FLAG tag and are cyclised as disulphides. (c) Radiolabelled [¹²⁵I]-GIP displacement from hGIPR#5/BHK Creluc 2P cells by cyclic peptides at a single concentration of 1000 nM CPs (x axis), and 100 nM CPs (y axis) (CPM = counts per minute). All peptides were tested crude without prior purification (ESI Data 2^†).We identified several peptides (>50% of our panel) with nanomolar binding affinities toward the hGIP-R from multiple sequence clusters, particularly from the two most abundant clusters A and B (Fig. 2a and b). However, potent binders were also identified in less abundant sequence families. In particular, peptide M_46 was the second most potent binder amongst the resynthesized peptide panel, while only accumulating 0.2% of the reads compared to the most abundant peptide A_0. This highlights the benefit of following up on peptide hits independent of their ranking based on sequencing reads. Encouragingly, weak-to-no binding was observed from peptide members of cluster 0 (Fig. S8^†), corresponding to bioinformatic ‘noise’ in the NGS data, which likely consists of singleton sequences and possible sequencing errors. The FLAG tag itself did not bind to hGIP-R ECD. For members of clusters containing an internal cysteine, e.g. peptide C_24, we generated single cysteine-to-methionine mutants for each site (Met2, Met8, and Met13) and deduced the disulphide pattern from the binding data (i.e. a bond between Cys2 or Cys8 for C_24, see Fig. S8^†). In the spirit of the high throughput nature of the workflow, we did not follow up on peptides which failed during parallel synthesis, as the goal was the identification of one hit sequence. As a consequence, we suggest that the pursuit of peptides with complex folds or unpaired cysteines is only indicated if no functional hits can be identified otherwise.The non-biased selection process means that high affinity CPs can be generated against any sites on hGIP-R ECD. We hypothesized that each cluster represents peptides which bind at a specific target binding site, with the possibility of different clusters having overlapping or distinct binding sites on the receptor. In this study, we were interested in identifying inhibitory peptide families which are able to either compete or disrupt the natural ligand, whether through direct competition or allosteric binding. To investigate this, we selected representative CPs from each cluster and performed competitive displacement assays with ¹²⁵I-GIP in BHK CreLuc 2P cells stably expressing hGIP-R. Notably, this assay shows (competitive) binding to the full length GIP-R, whereas our selection and initial screen had been performed against the ECD. As shown in Fig. 2c, only peptides from clusters B and M were found to displace ¹²⁵I-GIP from hGIP-R (as measured at peptide concentrations of 100 nM and 1000 nM). Interestingly, cluster B contains a LWPF motif at the C-terminus, while cluster M has a related LPWF motif at the N-terminus, indicating a potentially conserved binding site. None of the members of other clusters were found to displace ¹²⁵I-GIP, even those which were determined to have nanomolar binding affinities to the hGIP R ECD by BLI, including the most abundant peptide family of the selection (cluster A), which covered over 50% of the whole dataset. This highlights the value of the clustering approach, as enrichment of the sequences during mRNA display is driven only by binding affinities of encoded CPs to hGIP-R ECD, but our subsequent clustering analyses served to reveal different functionalities of these CPs. Thus, the high-throughput prioritisation approach is likely to have a higher chance of identifying functional ligands, rather than those prioritised based on observed amounts of NGS reads.Encouraged by these results, we focussed our subsequent work on cluster B, namely the structurally most diverse members B_3, B_5, and B_68. B_68 had < 20% purity in the 96 well synthesis and did not reveal any binding in the initial high throughput binding assay but was included in the follow up. Further analysis of the sequence list showed another structurally diverse member of cluster B, B_1275, which we hypothesized would also exhibit potent binding and GIP competition, despite the low levels of sequencing reads. These four sequences were scaled up by traditional SPPS and further investigated in multi-concentration BLI and competitive binding assays. All purified peptides showed nanomolar binding and competition of the ¹²⁵I-GIP from BHK cells stably expressing hGIP-R (Fig. 4). Furthermore, none of these ligands were found to displace radiolabelled GLP-1 from GLP-1R, nor glucagon from GCG-R (Fig. S11^†). The parent CP sequences also do not bear any significant sequence homology with native GIP, nor any known interactors of hGIP-R, suggesting that these are the first reported examples of de novo incretin receptor selective and competitive peptides for hGIP-R.Open in a separate windowFig. 4(a) BLI binding K_d values and radiolabelled [¹²⁵I]-GIP displacement IC₅₀ data for purified peptides. Multiple concentrations of CPs were used to determine these values. (*) uncertain values due to limited solubility of CP. (b) Stability of selected peptides upon incubation with human plasma at 37 °C. (c) In vivo plasma exposure levels of selected peptides upon i.v. dosing to rats. All data are presented as mean ± SEM of three independent experiments.To optimise the biophysical and physicochemical properties of the CPs, and to identify a suitable attachment point for half-life extending moieties, such as albumin binders, we sought to establish a detailed structure–activity relationship (SAR) on cluster B. We chose the most and least abundant of our cluster B lead peptides (B_3 and B_1275) as starting points for full amino acid mutation scans.^24,26 We focussed on natural amino acids only to keep the option for (semi-)recombinant expression of the compounds, should large amounts be needed for further development stages. Single mutation variant peptides were synthesised in a 96 well format for each amino acid in the variable region contained between the two conserved cysteine residues C2 and C13. Methionine and cysteine were not included in the mutational scan due to potential oxidation, and asparagine was not included due to the risk of deamidation and isomerisation to iso-aspartate. Additionally, scrambled peptides were included in the panel. The binding affinities of these mutant peptides to hGIP-R ECD were then determined by single concentration BLI experiments.As shown in Fig. 3a and b, the consensus sequence of cluster B (positions 9–12, LWPF) was found to be the most intolerant to mutation, confirming this region to be critical for binding. Proline at position 11 was found to be completely intolerant to replacement by any other amino acid, suggesting that this residue is critical for maintaining conformationally restricted binding interface of CPs. While residues 9–12 are hydrophobic in both parent peptides, it is unlikely that the interaction of the peptides to hGIP-R ECD is unspecific, as none of the scrambled peptides (e.g. LWPF) were found to bind, including the transposed mutants of B_3 and B_1275, with interchanged L9 and W10. The data shows some general trends for both B_3 and B_1275. In general, positions 3 and 6 (both F in the parent peptides) favour aromatic residues, position 8 favours aliphatic residues, while residues 4, 5, and 7 are fairly tolerant to mutation. The single amino acid scan did not reveal any substitutions that led to remarkable improvements in binding, suggesting that these peptides may either already be optimised for binding through several rounds of mRNA-display selection, or that further binding improvements would only be achieved by synergistic action of multiple substituted residues. The latter could ideally be investigated by mRNA display based affinity maturation experiments in follow up studies.^24,26 To gain insight into the binding mode of CPs to hGIP-R on the atomistic level we employed a two-step modelling protocol that first generates multiple conformations of the complex and then selects the final conformation based on stability (for details, see ESI^†). As our peptides were demonstrated to be ¹²⁵I-GIP-competitive, we directly folded them inside the GIP binding site of hGIP-R using Rosetta (Fig. 3c and d; results for B_1275 shown). Briefly, each of the four crucial residues according to the SAR data (Leu9, Trp10, Pro11 or Phe12) was placed in the GIP binding site and the rest of the peptide was grown around it. The obtained conformations (24 000 in total) were further clustered and the representative poses of the four most populated clusters were submitted to molecular dynamics (MD) simulations (600 ns for each pose) to assess their stability. The model with the lowest L_RMSD (Root Mean Square Deviation of the ligand backbone during MD trajectory) was selected for further characterization (Fig. 3c). In this model, the C-terminal half of the peptide faces the GIP binding site forming most of the interactions with hGIP-R. This is in line with the obtained SAR data that the C-terminal region is in general less tolerant to mutagenesis compared to the N-terminus. Furthermore, the Trp10 residue overlaps with the position of Phe22 in GIP which is known to be crucial for its binding to the receptor (Fig. 3d). Finally, overall conformation of the peptide resembles that of a hairpin with the turn located on Phe6 and His7, which are the only two positions (together with the peptide termini, see below) tolerating mutagenesis to a proline.Open in a separate windowFig. 3Heat maps showing K_d (nM) values for binding of single-residue mutant peptides derived from B3 (a) and B1275 (b) to biotinylated hGIP-R ECD as determined by single-concentration BLI. The columns indicate the amino acid changes compared to the parent sequence (displayed at the top of the table). All peptides featured a C-terminal Cys and were tested in crude form. Peptides that were not tested or those where synthesis failed are marked as grey boxes. None of the scrambled mutant peptides showed any binding. (c) Atomistic model of the cyclic peptide B_1275 and GIP receptor complex suggested by molecular modelling. The LWPF sequence of the peptide is shown in red, while the rest of the molecule is coloured by the atom name (carbon in yellow, nitrogen in blue, oxygen in red, and sulphur in orange). The receptor is show in grey; the binding site residues (L35, W39, M67 and Y87) are highlighted in green. (d) Overlay of the crystal structure of GIP complexed with hGIP receptor (PDB 2QKH) and atomistic model of cyclic peptide B_1275 in the binding site of the GIP receptor. Note that the positions of W10 of the cyclic peptide and F22 of GIP overlap.Next, we wanted to use the accrued binding and SAR information to design peptides for in vivo studies. For these, peptides need to fulfil two crucial requirements. Firstly, they need to be stable in plasma and secondly, they need to be protected from renal clearance. The latter is crucial to the development of any peptide therapeutic as even protease resistant peptides usually exhibit plasma half lives in the range of a few minutes. Attachment of a fatty acid-based albumin binder (termed protractor) is one of the most employed strategies to extend peptide half-life and we opted for a 2xOEG-gGlu-C18 diacid albumin binder for our peptide hits.^3,27–29 The binding model suggested that the addition of an albumin binding moiety is tolerated at the N- or C-terminus, as these are solvent exposed. Further evidence that replacement of the Met in position 1 with Ala is possible for both B_3 and B_1275 without loss of affinity and the ease of synthesis prompted us to design four peptides with a protractor attached to the N-terminus (B_3.1, B_5.1, B_68.1 and B_1275.1). Furthermore, we synthesized linear versions of the parent peptides in order to investigate the importance of macrocyclization on binding and stability (B_3.2, B_5.2, B_68.2 and B_1275.2). Additionally, we synthesized peptides without FLAG tag to prove that binding indeed is mediated by the macrocycle and not the tag (B_1275.3). During the purification of the latter, we realized that solubility was low (as expected from the rather hydrophobic sequence), complicating purification and analysis. Thus, we utilized our SAR understanding to design variants with improved biophysical properties by reducing the hydrophobicity and increasing the charge at positions where mutation was tolerated (especially positions 1, 4 and 7). Furthermore, we replaced the Met residues to avoid oxidation of the sulphur atom. We focussed on B_1275 as we anticipated from the sequence that this peptide is most hydrophilic, as also indicated by the shortest retention time in HPLC chromatography (Table S1^†). All resulting peptides containing a FLAG tag (B_1275.7 through B_1275.11) were soluble at the relevant conditions, however, only B_1275.5 showed high solubility without the tag, as measured by nephelometry at different PEG concentrations (Fig. S14^†). Pleasingly, addition of an albumin binder (B_1275.6) did not negatively affect solubility. Analysis of the peptides by multi-concentration BLI and radio-GIP displacement showed that indeed N-terminal modification was allowed for most peptides (Fig. 4). None of the linear variants (B_3.2, B_5.2, B_68.2, B_1275.2) showed binding, suggesting that macrocyclization is crucial for the interaction. Most mutant peptides, however, retained binding, including the triple mutant B_1275.9 (M1A, T4R, H7E; K_d at 31 nM) and quadruple mutants B_1275.7 (M1A, F3W, H5E, H7E, K_d at 101 nM) and B_1275.8 (M1S, T4R, H5A, H7Q, K_d at 184 nM), showing the power of having detailed SAR information available for multi-factorial optimisation of hits. Interestingly, in contrast to the binding data obtained from F11P mutants from high throughput mutagenesis studies (Fig. 3), B_1275.10 with a F11P substitution did not show any binding, which exemplifies the limits of high throughput SAR analysis using crude peptides, as the terminal P possibly interferes with cyclisation and might lead to multi- or polymers which interfere with the BLI assay.Finally, we tested a panel of the optimised peptides for stability in human plasma and determined in vivo pharmacokinetic parameters in rats. Most cyclic peptides (B_3.1, B_1275.3, B_1275.4, B_1275.5 and B_1275.6) showed no decrease in plasma stability over the course of 5 h, including in the presence of a FLAG tag or of a protractor (Fig. 4b). While parent B_1275 had a t_1/2 of 3.5 h, the two linear peptides tested (B_3.2 and B_1275.2) exhibited low levels of degradation (t_1/2ca. 2 h, see Table S2 in ESI^† for exact values), and the biologically active linear peptides of both GIP and GLP-1 were more rapidly degraded (t_1/2ca. 35–45 min), which highlights the benefits of cyclic peptides in terms of druggability. Having established the plasma stability, we turned our attention to the in vivo half-life. We chose two protracted parent peptides (B_3.1 and B_1275.1) and the optimised B_1275 variants with FLAG tag (B_1275.4), without FLAG tag (B_1275.5) and with protractor but without FLAG tag (B_1275.6). The two non-protracted peptides exhibit very short half-lives (B_1275.4 = 3.5 min, B_1275.5 = 1.8 min), whereas the peptides carrying an albumin binder show significant plasma exposure over a long period of time (t_1/2, B_3.1 = 9.8 h,B_1275.1 = 10.7 h B_1275.6 = 4.2 h), with half-lives being increased by more than 100 fold (Fig. 4d) (as a comparison, the GLP-1 analogue semaglutide, which is dosed once-weekly in humans has a t_1/2 in rats of 7 h.)²⁸