Living organisms synthesize single crystals of biominerals through controlled nucleation and growth assisted by functional biomolecules such as proteins bearing glycine in water at near room temperature1,2. The crystallization process typically involves the sequestering and concentrating of ions from water within a matrix or compartment, in which a family of disordered solid precursors are formed, accumulated, stabilized, temporarily stored, and further utilized as starting materials for crystallization without resorting to low levels of supersaturation3-5. Adapting the biomineralization path to synthesize crystals with high production efficiency and structure diversity remains challenging. Here, we mimic this strategy to synthesize single crystals of an important class of organic materials - covalent organic frameworks (COFs) in water under ambient conditions.
COFs, constructed from organic monomers termed as knots and linkers through covalent bonds, are polymeric materials with permanent porosity6. Their synthesis typically involves volatile organic solvents, high temperature, high pressure and inert atmosphere or vacuum7,8. During the synthesis, amorphous polymeric materials forms spontaneously, aggregates and precipitates rapidly, and depends on random error-corrections offered by microscopic reversibility of the dynamic covalent bonds to crystallize into network structures9. The free polymerization and crystallization process naturally requires trial and error, chemical intuition and large-scale screening to find an experimental condition to crystallize a specific structure10, and prevents the formation of single crystal COFs with controlled size and their large-scale synthesis11,12.
Our strategy to creating single-crystal COFs explores amphiphilic molecules of palmitoylglycine (C16-GlyA) with long hydrophobic chains as dynamic barriers to separate knots and linkers in water and oil phases (C16-GlyA phase) to regulate the polymerization and crystallization processes (Fig. 1a, b, c). C16-GlyA was synthesized from palmitoyl chloride and glycine (see Supplementary Information Section 1.1 for details). It can assemble into vesicles with a layered supramolecular structure of around 331 nm in water (Extended Data Fig. 1a, b), affording a hydrophobic compartment13.
We started with COF-301 that was in previous work only obtained as poorly crystalline aggregates with a low yield and no structure elucidation could be enabled despite of its high promise in different areas14, 15 by condensation of 2,5-dihydroxy-1,4-benzenedicarboxaldehyde (linker-A) with one equivalent of tetrakis(4-aminophenyl)methane (TAM) under ambient conditions. Solution nuclear magnetic resonance (NMR) and electrospray ionization high-resolution mass spectrometry indicated that TAM protonated with p-toluenesulfonic acid (PTSA) mainly dissolved in water while linker-A was in the C16-GlyA phase when added individually (Extended Data Fig. 1a, b and Extended Data Fig. 2). When powders of linker-A were added into the C16-GlyA emulsion, it tuned from white to yellow. The size of vesicles grew to ~ 503 nm and no precipitate or vesicles of other size was detected (Extended Data Fig. 1a), furthering confirming the dissolution of the monomer in the C16-GlyA phase. Subsequent addition of PTSA-protonated TAM led to gel-like mixtures within 10 minutes. After thorough rinsing with water and tetrahydrofuran, disordered solid phase with a yield of 91% was obtained (Fig. 1d and Extended Data Fig. 3a, b). No monomers were detected in the solid phase by high-resolution mass spectrometry, and the broad differential thermal gravity (DTG) peak suggested existence of a mixture of materials (Fig. 2a), and Fourier transform infrared (FTIR) spectra indicated they were polyimines containing large amounts of free aldehyde groups (Extended Data Fig. 3c).
We then evaluated the yield and crystallinity of the isolated solid phase at different reaction time. The yield showed no obvious time dependence and fluctuated between 91% and 95% (Fig. 2b). Bragg peaks of (2 0 0), (2 1 1) and (1 1 4) of COF-301 emerged after two days and single-crystal COF-301 could be clearly observed by optical microscopy (Extended Data Fig. 3a and Fig. 2c), suggesting a gradual amorphous-to-crystalline transition. As the reaction proceeded, the intensities of Bragg peaks increased, demonstrating gradual transformation of the disordered phase to crystals (Fig. 2b and Extended Data Fig. 3a, b). As also visualized by optical microscopy (Fig. 2c), 2 μm-sized crystals occurred on day two and more appeared with the increase of time. Pure single-crystal COF-301 was obtained after 14 days (Fig. 1e). They are uniform cuboids with typical length and width of around 2 μm and 0.6 μm (length/width = 3.3) at a molar concentration of linker-A of 0.05 mol L-1, respectively. Mono-dispersed crystals were obtained despite the variation of the concentration of the monomers by five orders of values from 0.005 mmol L-1 to 0.5 mol L-1. Notably, the size of the crystal increased from ~ 0.6 μm × 0.08 μm at 0.005 mmol L-1 to ~ 20 µm × 10 µm at 0.5 mol L-1 (Fig. 1f, g). The increase of the crystal size with the increase of the concentrations of the reactants in solution is contrary to the correlation in normal crystal growth16. Meanwhile, policies such as the usage of poor solvent to decrease the solubility of reactants, addition of reaction inhibitors/competitors and slow addition of reactants were employed to keep low concentrations of reactants to decrease amorphous parts and increase crystal size of COFs in contemporary methods10-12.
Transmission electron microscopy (TEM) revealed that there was ~ 30 nm of adsorbed layer on the surface of the crystals (Fig. 2d, left), which could be removed by rinsing with tetrahydrofuran (Fig. 2d, right). Characteristic peaks of C16-GlyA in confocal Raman spectroscopy disappeared after rinsing, indicating the main component of the adsorbed layer was the amino-acid derivative (Extended Data Fig. 3d). Note that the weak interactions (mainly hydrophobic-hydrophobic and hydrogen bonding interactions) among C16-GlyA were highly dynamic, which facilitated the diffusion of TAM and PTSA. Deprotonated TAMs entered the microenvironment and reacted with linker-A, trigging the nucleation and growth of the crystal, while PTSA-protonated TAMs diffused to water due to solubility (Extended Data Figs. 1 and 2).
We selected an as-synthesized COF-301 (termed as COF-301-S) to prove single crystallinity and resolve the atomic structure with single-crystal XRD (see Supplementary Information Section 2 for details). COF-301-S was single crystal with a tetragonal space group of I41/a (a = 26.434(4) Å, c = 7.5876(15) Å) and 7-fold interpenetration. The single network exhibits a diamond-type topology in which each TAM molecule is connected to four linker-As (Fig. 3a). The refined data resolution was ∼ 0.9 Å, and all non-hydrogen atoms in the COF-301-S were identified. The selected area electron diffraction revealed that the interpenetration direction, i.e. along the longest diagonal of the unit cage (Fig. 2c and Fig. 3a, black arrow indicates the direction), was parallel to the elongation direction of the cuboids. The COF-301-S has one-dimensional (1D) straight channels along the interpenetration direction. A notable framework distortion was observed upon removal of C16-GlyA by rinsing with tetrahydrofuran (Fig. 3b). The topology and interpenetration degree remained, while the pore size changed from 9.6 Å to 3.1 Å. The washed COF-301 (termed as COF-301-W) exhibited a different phase with a symmetry of I2/c (a = 20.276(8) Å, b = 8.7098(18) Å, c = 20.212(4) Å, β = 99.308(12) °) (see Supplementary Information Section 3 for details).
We then investigated the surface morphology evolution of COF-301 to gain an insight into the crystal growth model. Intentionally aggregated crystals were prepared by dip coating on newly cleaved mica for characterization with atomic force microscopy (AFM). Disparate surface morphologies on the top and side facets of the cuboids have been observed (Fig. 2e, f). The (010) plane exhibits a higher density of growth sites or kinks than (100)/(001) planes (Fig. 2c, e and f), which is a characteristic of adhesive growth in which the building units stochastically integrate into the growing surface17. The (100)/(001) planes, on the other hand, show a train of terraces indicating layer-by-layer growth in which the building units preferentially integrate into the kink sites along the terrace edge (Fig. 2f). Due to the limited number of kink sites, the growth rate of the layer-by-layer mode is lower than that of the adhesive mode18, 19, leading to growth anisotropy. The line scans in Fig. 2e present the typical step sizes on the top surface, in which 0.22 nm, i.e. the height difference between interpenetrated networks, and its multiples have been identified (Fig. 2g). The terrace height on the (100)/(001) facets has four typical values (Fig. 2f), i.e. 1.0 nm (one TAM + one linker-A), 1.3 nm (one TAM + two linker-A), 1.7 nm (two TAM + one linker-A), and 2.0 nm (two TAM + two linker-A) (Fig. 2h). The sub-unit-cell step heights on both the top and side facets suggest that monomers and dimers are the main fundamental building units during the crystal growth. The step-by-step crystal growth from solution involving disordered solids differs from known crystallization path either from a saturated solution and or a melt in crystal growth20. It also differs from the error-correction mechanism found in contemporary synthesis methods for COFs and two-dimensional (2D) polymers9, 21, 22.
To understand the effects of different parameters on the isolated yield of the solid phase and its conversion ratio to single-crystal COFs, the length of the hydrophobic chain of the amphiphilic molecules of amino-acid derivatives and their head groups and added catalysts were investigated. We first performed control experiments without C16-GlyA, no solid was obtained even over a month. We then synthesized octanoylglycine (C8-GlyA), decanoylglycine (C10-GlyA), dodecanoylglycine (C12-GlyA), tetradecanoylglycine (C14-GlyA) and stearoylglycine (C18-GlyA) (see Supplementary Information Section 1.1 for details) to investigate the effects of the length of the hydrophobic chain of the glycine derivatives (Extended Data Fig. 4a). Little solid was obtained for C8-GlyA, C10-GlyA, and C12-GlyA, the yield of solid phase and crystals was 87% and 79% for C14-GlyA, and then both reached maximum for C16-GlyA. With C18-GlyA, the yield of solid phase decreased to 79% (Extended Data Fig. 4a). The results indicated that the hydrophobicity and order of the assembled structure of the derivatives played important roles in regulating the diffusion, polymerization and crystallization process for the formation of COF-301. C8-GlyA, C10-GlyA, and C12-GlyA were packed too loosely to form micelles, and therefore unable to regulate these processes23. With the increase of the chain length, and therefore the enhancement of the hydrophobicity and order of the assembled structures, larger micelles were formed and higher yields were achieved (Extended Data Fig. 1a). The hydrophobicity of C18-GlyA hindered its efficiency to be dispersed in water and therefore resulted in a low yield for the formation of COF-301. We also utilized palmitic acid (CH3(CH2)14COOH) for the synthesis. It led to clusters with a yield of 81% (Extended Data Fig. 4b), indicating the glycine group played a crucial role in keeping COF-301 in well-defined shapes and as mono-dispersed particles. We further synthesized palmitoyl-L-alanine (C16-L-AlaA), palmitoyl-L-phenylalanine (C16-L-PhalaA), palmitoyl-L-valine (C16-L-ValA) and (S)-3,3-dimethyl-2-palmitamidobutanoic acid (C16-L-tLeuA) (see Supplementary Information Section 1.1 for details) to investigate the effects of steric hindrance near the polar groups of the amino-acid derivatives on the morphology, size and yield, and no significant difference was observed (Extended Data Fig. 4c).
Control experiment without adding PTSA led to disordered materials with a yield of 5% (Extended Data Fig. 5). The yield of single-crystal COF-301 increased with the increase of molar concentration of PTSA, reached maximum at six equivalent molar amounts to linker-A, and slightly fluctuated thereafter. Meanwhile, the time needed for achieving pure crystals decreased and shortened to 3 days at 9 and 10 molar equivalents of PTSA (Extended Data Fig. 5). Acetic acid and hydrogen chloride were also utilized as catalysts, and they provided lower yield (Extended Data Fig. 6). Interestingly, hydrogen chloride could offer larger crystals than that with PTSA. This can be attributed to the stronger acidity of hydrogen chloride, which facilitates the hydrolysis of the disordered polyimines into monomers and dimers as building units for subsequent crystal growth.
This biomimetic synthesis strategy shows general applicability. We chose monomers with parent core (terephthalaldehyde (linker-B)), electron-accepting (2,3,5,6-tetrafluoroterephthalaldehyde (linker-C)), heterocyclic (pyridine-2,5-dicarbaldehyde (linker-D)) besides the electron-donating (linker-A) substitutions near aldehyde groups and monomers with different length (4,4'-biphenyldicarboxaldehyde, linker-E) to gain single-crystal COF-300, SYSU-8, SYSU-9 and COF-320 under the same crystallization condition as the synthesis of COF-301 (Methods). Single-crystal and mono-dispersed COFs were obtained in all cases (Fig. 1h, i and Extended Data Fig. 7a, b), though the substitutions can notably affect the reaction activity of aldehyde groups and the crystallization process11. This is in sharp contrast to the contemporary synthesis methods for COFs, such as the solvothermal method, which requires extensive refinement of optimal reaction conditions on minor change in the monomer structure10 (see Supplementary Information Section 1.2 for details). All non-hydrogen atoms of the crystals were identified (Fig. 3c-3f and see Supplementary Information Section 2, 3 for details). The typical size of as-synthesized SYSU-9 (termed as SYSU-9-S) was ~ 20 μm × 3 μm (Fig. 1h), and its structure was determined by single-crystal XRD with a resolution of 1.1 Å (Fig.3e). SYSU-9-S crystallized in the space group I41 (a = 26.461(4) Å, c = 7.4600(15) Å). The size of SYSU-8 was ~ 15 µm × 3 µm (Fig. 1i). COF-320 crystallized in the space group I41/a (a = 23.360 Å, c = 8.430 Å) and an interpenetration degree of 9-fold (Fig. 3f). A new crystal phase was obtained, its symmetry was higher than that synthesized by solvothermal method 24, and allowed atomic structure by continuous rotation electron diffraction with an increased resolution of ∼ 0.9 Å (see Supplementary Information Section 2 for details). The size of COF-300 could be down to 600 nm (see Supplementary Information Section 1.3 for details) which made it processable as colloids, for example, into flexible films (Extended Data Fig. 8). All the synthesized COFs have one-dimensional (1D) straight channels in which the substitutions distributed spirally, offering varied channel surfaces with different polarity and pore size (COF-301-W: 3.1 Å; COF-300: 5.5 Å; SYSU-8: 5.0 Å; SYSU-9-W: 6.8 Å, COF-320: 6.7 Å) (Fig. 3b-f). We then chose 2DCOF-TTA-TBA as an illustrative example to extend the biomimetic synthesis methodology to synthesize organic 2D materials due to their great potential to go beyond inorganic analogues such as graphene and MoS2 and great potential as key materials for next-generation technologies25. AA stacked nanosheets with the space group P-6 (a = 17.875 (8) Å, c = 3.4652 (11) Å) were obtained (Extended Data Fig. 7c and see Supplementary Information Section 3 for details). Its thickness was about 50 nm, and could be easily dispersed well in solvents such as water and filtered into thin films (Extended Data Fig. 8). Importantly, the sheets could also be exfoliated by sonication into nanosheets with a thickness down to around 0.7 nm, which corresponded to around 2 layers (Extended Data Fig. 8e). The formation of the imine bonds in all the crystals were also confirmed by solid-state 13C cross-polarization magic angle spinning NMR, FTIR spectroscopy and thermogravimetric analysis (TGA), and permanent porosity of the COFs was confirmed by gas adsorption (see Supplementary Information Section 4-7 for details). By comparing the 77 K nitrogen adsorption of known COFs, COF-300 and COF-TTA-BTA, the biomimetic strategy offered materials (COF-300: 1320 cm2 g-1; COF-TTA-BTA: 1438 cm2 g-1) with Brunauer-Emmett-Teller surface area slightly higher than best reported results from COFs obtained with contemporary methods.26, 27 (Extended Data Fig. 9).
The C16-L-tLeuA could also be spread on the water surface to induce the crystallization of COF-366 (Methods), a typical 2D COF 28. Initially, amorphous films were formed and stabilized underneath C16-L-tLeuA (Extended Data Fig. 9), and gradually transformed into crystalline film with micrometer-sized single-crystal domains. The high crystallinity of the domain enabled near-atomic structure observation by aberration-corrected high-resolution TEM with a resolution down to ~ 2 Å (Extended Data Fig. 10).
We further investigated the role of the pore structure and pore wall structure in tuning the porosity of the single-crystal COFs with adsorption of CO2 at 195K and 273K (Fig. 4a, 4b). In the uptake of CO2, ‘gate type’ sorption profile occurred at 195 K29, and disappeared at 273K. The gate opening pressure at which an abrupt increase in adsorption after a threshold pressure and the uptake of CO2 could be tuned by varying the pore surface or adsorption temperature, allowing systematic controlling of the uptake. Among them, COF-300 offered a CO2 uptake of 480 and 8.5 cm3 g-1 (P/P0 = 0.95) at 195K and 273K, respectively (Fig. 4a, b). Interestingly, the substitutions applied to COF-300 (P/P0 = 0.95) significantly decreased the uptake at 195K, while increased by 294%, 384%, 624%, and 729% with COF-320 (25 cm3 g-1), SYSU-8 (30 cm3 g-1), COF-301 (53 cm3 g-1), and SYSU-9 (62 cm3 g-1) at 273K, respectively.
A key obstacle in exploring the application of single-crystal COFs is mass production, which remains challenging to be resolved30. We demonstrated the capability of scale-up of the biomimetic strategy with gram-scale production of single crystals of COF-301 (yield = 92%), COF-300 (yield = 95%), SYSU-8 (yield = 94%), SYSU-9 (yield = 93%), COF-320 (yield = 94%) and COF-TTA-BTA (yield = 92%) (Fig. 4c, d). In contrast, the yield of single-crystal COF-300 has been reported, which amounted to 49.8% 12. Importantly, the amino-acid derivatives could be recovered by recrystallization with a recovering rate of ≥ 92%.
In summary, we developed a new path and strategy for crystal growth of organic polymers, revealed the formation mechanism and crystal growth model, and presented the production of six types of single-crystals COFs with controlled size, compositions and properties at the gram-scale besides a two-dimensional COF in water under ambient conditions.