Our synthetic strategy is based on subcomponent self-assembly38,39. As illustrated in Fig. 1b. elementary units A and B serve as the basic organic building blocks, forming covalent backbones via dynamic imine bonds. Coordinatively labile metal ion C acts as the metal template, leveraging its labile nature for adaptive coordination40,41. Counter anion D functions as a directional anchor, guiding the packing arrangement of adjacent layers through weak C-D-C bridging. Consequently, reversible covalent, coordination, and metal-ionic bonds were formed in situ around metal template C in a “one-pot” reaction. This results in a hierarchical assembly of organic and inorganic subcomponents through a synergistic multilevel process: (1) Units A, B, and C form covalent metallacycles; (2) anion D chelates the adjacent metal template C, directing the layering of covalent metallacycles to form tubular structures; (3) as the tubes grow continuously in the longitudinal direction, a symmetry-determined arrangement of tubes occurs along the perpendicular direction; eventually, (4) a well-ordered framework emerges as the most thermodynamically stable structure. Following this process, the combination of A, B, C, and D leads to the formation of a SNF material; while substituting bitopic unit E for A results in a CNF material. Due to the structural independence of covalent backbones, metal ion C and anion D within CNFs could be easily removed, leading to the formation of layered COFs. These COFs could be remetalated with C, and the layers can be re-bridged with D, effectively restoring them to the original CNF state (Fig. 1c).
Supramolecular nanotube frameworks
Under solvothermal conditions, SNF-LIFM1 was synthesized by using a mixture of A1, B1, and Zn(NO3)2 with a molar ratio of (1:1:1.6) in a solution of CH3CN, mesitylene, and deionized water (5/5/3, v/v/v) at 110°C for two days. The obtained orange solid is insoluble in common polar and nonpolar organic solvents. To achieve high crystallinity, a combination of Zn2+ and NO3‾ is critical. Screening tests of metal ions like Fe2+, Co2+, and Ni2+ as template, and counter anions of Cl−, SO42−, BF4− and SO3CF3− as directing agent, respectively, failed in yielding the desired crystalline products, suggesting the highly labile d10 electronic configuration of Zn2+ ion42 and suitable anion bridging between metal centres43 are essential for synergistic assembly. Moreover, an excess amount of Zn(NO3)2 (1.6 equiv.) is needed, which acts as a Lewis acid to catalyze imine condensation44.
The high crystallinity of SNF-LIFM1 was identified using powder X-ray diffraction (PXRD) (Supplementary Fig. 1). The formation of imine tridentate ligands and the corresponding Zn2+ complex was confirmed using multiple techniques. Fourier-transform infrared spectroscopy (FT-IR) (Supplementary Figs. 2 and 3) and solid-state nuclear magnetic resonance (SSNMR) spectra (Supplementary Fig. 4) provided initial evidence, while further confirmation was obtained through the examination of the single-crystal structures of the model compounds M1 and M2 (Supplementary Figs. 5, 6 and Supplementary Tables 1, 2). Due to the very narrow pore size of SNF-LIFM1, nitrogen (N2) adsorption measurements at 77 K failed to provide the pore information. Instead, carbon dioxide (CO2) adsorption at 195 K was conducted, revealing a type-I gas adsorption isotherm with rapid gas uptake at low relative pressures, indicative of its high microporosity (Fig. 2h)45. The Brunauer–Emmett–Teller (BET) surface area was determined to be 347 m2g− 1 (Supplementary Fig. 8).
The crystal structure of SNF-LIFM1 was determined by combining 3D electron diffraction (ED)46, STEM imaging, and PXRD. The 3D ED reciprocal lattice (Fig. 2g) indicated that SNF-LIFM1 was crystallized in a hexagonal unit cell of a = b = 32.4 Å, c = 4.1 Å, and γ = 120°. STEM images were obtained along the [001] direction using both annular dark-field (ADF) (Fig. 2c) and integrated differential phase contrast (iDPC) signals (Fig. 2e)47. STEM imaging provides a versatile tool box to characterize atomic structures48. The ADF signal, being more sensitive to Z-number, enabled the localization of Zn ions, whereas the iDPC contrast provided a more uniform representation of high and low Z elements, thus revealing the organic component as well49. By utilizing the periodic information, lattice averaged projected potential maps could be obtained with improved signal, revealing a hexagonal arrangement of triangular covalent metallacycles (Figs. 2d and 2f). Combining the 3D ED data and STEM images, the crystal structure of SNF-LIFM1 was built in space group P63. The solved structural model aligns well with the distribution of Zn ions in the ADF signal (Fig. 2d), the triangular rings in the iDPC signal (Fig. 2f), and the iDPC-STEM image acquired along [100] (Supplementary Fig. 10), giving simulated PXRD pattern in good agreement with the experimental data of SNF-LIFM1 (Supplementary Fig. 7).
Analysis of the crystal structure of SNF-LIFM1 reveals that the triangular metallacycles are pillared by NO3‾ along the c-axis, resulting in the formation of 1D nanotubes (Fig. 2j). These nanotubes are further perfectly arranged in a hexagonal pattern in the crystallographic ab-plane, thereby creating extended supramolecular frameworks (Fig. 2a). Notably, the symmetry determined assembly of SNF-LIFM1 on microscopic level is well transmitted to the mesoscopic level, leading to directional aggregation of a uniform bundle-like morphology (Fig. 2i and Supplementary Fig. 11).
Covalently-linked nanotubular frameworks
The hexagonal arrangement of triangular nanotubes in SNF-LIFM1 presents a fascinating structure, which inspired our curiosity whether it is possible to connect the vertexes of adjacent triangular macrocycles using organic connectors to build covalently-linked extended frameworks (Fig. 3a). To achieve this goal, we performed molecular engineering by designing bitopic building units that join two organic units of A1 head-to-head through para-phenylene (-Ar-) connectors, offering three linear building blocks E1-E3 that were functionalized with four acetals (Fig. 3b). To increase the solubility of the bitopic building blocks, alkoxy groups (-OC6H13) were introduced on phenylene rings. Additionally, the lengths of the -Ar units were systematically varied from one phenylene ring (E1), three phenylene rings (E2), to five phenylene rings (E3) to fine-tune the gaps between adjacent triangular nanotubes (Fig. 3b).
By modifying the solvothermal conditions used for constructing SNF-LIFM1 and using linear building blocks E1, E2, and E3 instead of building unit A1, three covalently-linked extended frameworks were synthesized, namely CNF-LIFM1, CNF-LIFM2, and CNF-LIFM3 (Fig. 3b). The synthetic and characterization details are described in the Supplementary Materials (Supplementary Figs. 12–35). It is worthy of noting that no solid products was obtained without the addition of Zn(NO3)2, emphasizing its critical role in the hierarchical assembly process of the tubular frameworks. The formation of covalently-linked structures was fully confirmed by FT-IR (Supplementary Figs. 12, 20, and 28) and SSNMR spectra (Supplementary Figs. 14, 22, and 30). The narrow full widths at half maximum (FWHM) of PXRD patterns confirm that all CNF-LIFMs are highly crystalline (Supplementary Figs. 13, 21, and 29). The initial intense low-angle diffraction peaks of CNF-LIFM1, CNF-LIFM2, and CNF-LIFM3 are situated at 2θ = 2.01°, 1.60°, and 1.28°, respectively. The gradual shift of the position of first peak correlates well with the increase in the length of the linear building blocks E1, E2, and E3 (Figs. 4a, 4d, and 4g).
The crystal structures of CNF-LIFMs were determined with a combination of PXRD analysis, STEM imaging, and model building. The corresponding simulated PXRD patterns from each of the structural models exhibit good agreement with the experimental PXRD patterns (Figs. 4a, 4d, and 4g). ADF-STEM images were obtained along the [001] direction for all CNF-LIFMs (Figs. 4b, 4c for CNF-LIFM1, Figs. 4e, 4f for CNF-LIFM2, and Figs. 4h, 4i for CNF-LIFM3). The images reveal the C6-symmetric alignments of 1D hexagonal channels with the triangular nanotubes in each node. The lattice averaged maps of the ADF-STEM images clearly show the position of the Zn-pillars of the nanotubes (Figs. 4c, 4f, and 4i). Additionally, Fast Fourier Transform (FFT) analysis confirm the periodic structure of each material, yielding d-spacings of 42 Å, 56 Å, and 70 Å for CNF-LIFM1, CNF-LIFM2, and CNF-LIFM3 respectively, in close resemblance with the d-spacings obtained from PXRD data, which were 44 Å, 55 Å, and 69 Å, respectively. The STEM imaging results manifest that the present covalent-coordination assembly strategy assisted by dynamic Zn-coordination not only give rise to high crystallinity of 3D CNFs, but also benefit the atomic-resolved structural characterization.
All CNF-LIFMs adopt a honeycomb-type structure, with triangular nanotubes at each corner covalently linked by the -Ar- connectors head-to-head (Figs. 4j, 4k, and 4l). The precise arrangement of nanotubes creates 1D hexagonal channels, making CNF-LIFMs highly porous. N2 adsorption measurements at 77 K indicate that all CNF-LIFMs exhibit a type-IV isotherm characteristic of mesoporous materials (Supplementary Figs. 16, 23, and 31). The BET surface areas of CNF-LIFM1, CNF-LIFM2, and CNF-LIFM3 were determined to be 275, 760, and 670 m2g− 1, respectively (Supplementary Figs. 17, 24, and 32). Using quenched solid density functional theory (QSDFT) fitting of the adsorption isotherms, the diameters of the 1D channels were estimated to be approximately 36 Å for CNF-LIFM1, 46 Å for CNF-LIFM2, and 57 Å for CNF-LIFM3, closely matching the proposed structural models (Figs. 4j, 4k, and 4l). These porosity analyses clearly suggest that uniform tubular frameworks are being formed. The variation of the organic connectors successfully expands the 1D hexagonal channels, accounting for the creation of variable mesoporosities50. This, in turn, results in the formation of hierarchical microporous and mesoporous tubular frameworks.
Structural transformations
According to the well identified structures of SNF-LIFM1 and CNF-LIFMs, the covalent backbones are self-linked to be independent of the metal templates, which means that demetalation should be possible without destruction of the covalent backbones (Supplementary Figs. 36–49). To remove the Zn centres, the sample was treated in an ethanol solution of ammonia for 0.5-2 h, resulting in demetalated materials with approximately 95% of the Zn2+ removed as testified by inductively coupled plasma (ICP) analysis (Supplementary Table 4). FT-IR spectra indicated that the covalent backbones of all materials remained intact (Supplementary Figs. 36, 39, 42, and 47).
For SNF-LIFM1, the colour changed from orange to pale yellow after demetalation. The insoluble SNF-LIFM1 in common organic solvents becomes soluble in THF, which allows crystal growth of the demetalated material. Single-crystal structural analysis revealed that the directions of imine bonds are altered, producing a nearly circular macrocycle with low ring stain (Fig. 5a, Supplementary Fig. 38, and Supplementary Table 3). Without the support of Zn-NO3 pillars, the covalent macrocycles tend to disperse. This explains why the crystalline SNF-LIFM1 turned amorphous after demetallation (Supplementary Fig. 37).
In contrast, since the triangular metallocycles in CNF-LIFMs are interconnected by -Ar- connectors, the process of demetallation will not lead to dispersion of the covalent macrocycles. As a results, well-ordered layered covalent-organic frameworks (COFs) are produced with an eclipsed stacking arrangement (Fig. 5b). The narrow FWHM values observed in the PXRD patterns indicate all resulting COF-LIFMs have good crystallinity (Supplementary Figs. 40, 43, and 48). The configuration changes of the covalent macrocycles induce a transformation in the crystal lattice to cause a shift in the peak positions of COF-LIFMs. Take CNF-LIFM2 as an example, Fig. 5c showcases the comparison of PXRD patterns before and after demetallation, and Fig. 5d presents the crystal structure model of demetalated material (COF-LIFM2) which is deduced from the crystal structures of CNF-LIFM2 and the unconnected covalent macrocycles. N2 adsorption measurements at 77 K revealed the high porosity of COF-LIFM2 (Supplementary Fig. 45 inset) with a BET surface of 570 m2g− 1 (Supplementary Fig. 46). COF-LIFM2 exhibits two channels with average pore diameters estimated to be approximately 13 Å and 42 Å by using QSDFT fitting (Supplementary Fig. 45). These values align precisely with the theoretical values of 12 Å and 43 Å. Notably, COF-LIFMs could be remetalated by Zn2+ and the layers could be re-bridged by NO3‾ by immersing in a CH3CN solution of Zn(NO3)2 to revert to CNF-LIFMs. PXRD analysis confirmed that the remetalated materials retained comparable crystallinity with the as-synthesized CNF-LIFMs, with no discernible difference in peak intensities and positions (Supplementary Figs. 40, 43, and 48). The highly controlled transformation and reformation of CNF-LIFMs and COF-LIFMs attest to the versatility of the cooperative dynamic bonding process for fabricating fascinating structures.