Introduction

Access to safe freshwater is paramount for human survival, agriculture, and sustained economic development1. However, the projected 22–34% increase in the global population by mid-century will inevitably escalate freshwater demands, directly conflicting with the diminishing availability of clean water2. This scarcity is further exacerbated by climate disruptions and pervasive anthropogenic contamination. Consequently, water desalination and recycling technologies have become increasingly critical for long-term water sustainability3,4. Reverse osmosis (RO) has emerged as the leading membrane technology for potable water production5. Early advancements focused on polyamide thin-film composite (TFC) membranes, which continue to serve as the industrial standard6,7. However, challenges in achieving ordered nanostructure and uniform pore size distribution in these membranes make it difficult to surpass the established permeability-selectivity trade-off8,9.

Covalent organic frameworks (COFs) are crystalline porous materials synthesized via the directional covalent assembly of organic monomers, intrinsically offering aligned pore channels, uniform pore sizes, and high porosity10,11,12,13. Despite these advantages, critical challenges for large-area membrane configuration lie in their intrinsically oversized pores (0.8–5 nm)14 and often-insufficient crystallinity. These structural limitations restrict COF membrane applications predominantly to organic solvent nanofiltration15,16, macromolecular dye separation17, and multivalent ion sieving18,19. Notably, achieving effective desalination with COF membranes remains a grand challenge, as the majority of sub-nanometer monovalent ions (e.g., 1–3 Å for naked ions and 6.5–8 Å for hydrated ions)20 typically bypass their pores, resulting in limited exploration of their potential in low-pressure RO desalination. Recently, Liu et al. introduced light-controlled COF membranes for pressure-driven RO seawater desalination. Despite achieving a notable 91.2% NaCl rejection, this figure falls short of the ≥ 95% industrial benchmark for practical RO systems.

Current strategies for engineering COF pore dimensions primarily involve chemical and physical modification21. Chemical approaches involve either pre- or post-synthetic modification, adjusting pore structure before or after membrane formation. Physical modification, or stacking engineering, manipulates pore aperture by altering the COF stacking mode. However, altering monomeric structure with pendant groups or side chains inevitably perturbs reaction kinetics, hindering initial COF nucleation and often leading to synthesis failure due to increased steric hindrance22,23. Moreover, precise control over synthesis parameters—including temperature, solvent, and interfacial conditions—is paramount, as these critically dictate the microstructure and crystallinity of COF membranes. Notably, such modifications frequently result in low-crystallinity COF membranes, which can diminish water permeability ascribed to reduced porosity and compromise selectivity caused by heterogeneous pore size distributions24,25.

Notably, hydrogen bonding plays a pivotal role in regulating COF architecture and functionality. By introducing hydrogen-bond donors or acceptors into the framework, it is possible to enhance inter- and intra-layer interactions, thereby refining pore geometry and improving structural order. Exploiting hydrogen bonding enables precise control of COF pore apertures for efficient molecule and ion separations. Zhu et al. achieved ultramicropores in hydrophilic aliphatic linker-based COFs using engineered hydrogen-bonding networks26. Similarly, Ma et al. synthesized hydrazone-linked COFs with reduced pore size by modifying monomer hydroxyl sites27. Current strategies to engineer these networks primarily employ amide, acylhydrazone, urea, or β-ketoenamine linkages28,29,30,31. β-ketoenamine configurations are particularly promising, establishing extensive hydrogen-bond connectivity while ensuring structural integrity. This combination optimizes crystalline order and stability, potentially mitigating the stability-crystallinity trade-off32. The high hydrogen bond density in β-ketoenamine-linked COFs is crucial for enhanced performance.

To this end, we introduce a facile hydrogen-bond fortification strategy to fabricate ultramicroporous, high-crystalline COF membranes for high-performance water desalination. This approach significantly advances our fundamental understanding of how optimizing the integration of inter- and intra-layer hydrogen bonds—which constrain molecular rotation—can simultaneously mitigate defect formation and modulate COF stacking patterns. The key to this strategy lies in substituting 1,3,5-triformylbenzene (TFB) with 1,3,5-triformylphloroglucinol (Tp) monomers featuring phenolic hydroxyls reacted with benzene-1,3,5-tricarbohydrazide (Bth). This substitution enables the hydrazone linkages to undergo enol-keto tautomerism, leading to the formation of thermodynamically stable β-ketoenamine configurations. The concerted effects of extensive intra-layer hydrogen bonding networks and inter-layer hydrogen bonding interactions effectively restrict molecular rotation, thereby producing COFs with enhanced crystallinity. This structural transformation induces a crystallographic transition from the hydrogen bond-deficient AA-stacking observed in TFB-Bth COF to a highly desirable AB-stacking in Tp-Bth COF, which features well-aligned ultramicropores and crystallinity. The resultant Tp-Bth COF membranes demonstrate exceptional desalination performance, achieving a 99.6% NaCl rejection and favorable water permeability, thereby far surpassing the desalting performance of advanced COF membranes. Furthermore, the abundant hydrogen-bonding networks intrinsic to the β-ketoenamine Tp-Bth COF provide enhanced structural robustness and stability in acidic conditions, underscoring its potential for sustainable water purification technologies.

Results

Design and synthesis of ultramicroporous COF nanofilms

Figure 1a illustrates the chemical structure of TFB and Tp molecules, in which three hydroxy groups were inserted to the adjacent phenolic aldehyde. This facile hydroxyl-insertion strategy enables the formation of extensive hydrogen-bonding networks within the Tp-Bth COF film. The resultant intra- and interlayer hydrogen bonds play a critical role in confining molecular rotation, which in turn markedly improves the long-range crystalline ordering of Tp-Bth COF films. Conversely, acylhydrazone-linked TFB-Bth COF films exhibit diminished crystallinity, largely attributed to topological imperfections stemming from weak hydrogen-bonding interactions. Furthermore, the distinctive adjacent hydroxyl configuration of the Tp monomer facilitates enol-keto tautomerism upon Schiff base reaction with hydrazide, yielding a thermodynamically stable Tp-Bth COF characterized by a β-ketoenamine configuration. This structural evolution endows the Tp-Bth membrane with an AB stacking mode and precisely narrowed, uniform ultramicroporous channels, leading to a substantially improved desalination performance compared to the TFB-Bth COF counterpart.

Fig. 1: Demonstrating hydrogen-bond regulated dual-configurational COF membranes.
Fig. 1: Demonstrating hydrogen-bond regulated dual-configurational COF membranes.The alternative text for this image may have been generated using AI.
Full size image

a Schematic explanation of the hydrogen-bond-regulated strategy. b Left panel: Schematic illustration of NaCl rejection by the imine-linked TFB-Bth COF membrane with low rotational energy barriers. Right panel: Schematic illustration of NaCl rejection by the β-ketoenamine-linked Tp-Bth COF membrane with high rotational energy barriers.

To overcome the inherent water insolubility of Bth monomers, our strategy for fabricating free-standing COF films incorporates a facile preliminary dissolution step into the conventional interfacial polymerization (IP) process. In specific, a dichloromethane (DCM) solution containing either Tp or TFB monomers was introduced into a glass Petri dish. An acetic aqueous solution containing Bth monomers was then gently layered atop the DCM phase, which initiated the Schiff base condensation at the free interface. The inclusion of the acetic acid solution resolved the solubility issue of Bth monomers while also catalyzing the reaction. Both the resulting yellow Tp-Bth COF and white TFB-Bth COF films exhibited uninterrupted continuity and mechanical resilience, allowing for facile retrieval using a wire loop (Fig. 2a, b). These films were then systematically transferred to various substrates—including copper mesh, porous alumina, and silicon wafers—to comprehensively assess their structural and morphological characteristics.

Fig. 2: Structural and functional characterization of free-standing COF nanofilms.
Fig. 2: Structural and functional characterization of free-standing COF nanofilms.The alternative text for this image may have been generated using AI.
Full size image

The Optical photograph of the freestanding film at the free two-phase interface (Image of the film captured by a wire loop), experimental, Pawley refined, and simulated PXRD profiles of the Tp-Bth film with difference (Rwp = 5.83% and Rp = 4.29%) and the TFB-Bth film with difference (Rwp = 16.74% and Rp = 12.9%), GIWAXS data, and HRTEM image (the lower inset is SAED pattern) of Tp-Bth COF (a, c, e, g) and TFB-Bth COF (b, d, f, h). i Variable temperature IR spectra of Tp-Bth COF nanofilm; (j) Solid-state 13 C NMR spectrum of Tp-Bth COF (top) and TFB-Bth COF (bottom) (400 MHz).

Transmission electron microscopy (TEM) images revealed both transparent and continuous COF films devoid of identifiable defects or slits. This high integrity was further corroborated by scanning electron microscopy (SEM) images, which showed smooth, ultrathin film conformity on porous supports (Supplementary Figs. 1, 2). Elemental mapping via energy-dispersive X-ray spectroscopy (EDS) confirms the homogeneous dispersion of carbon, nitrogen, and oxygen constituents across the film surface. Height profile data via atomic force microscopy (AFM) quantify the nanofilm thicknesses at 120.8 ± 0.3 nm for Tp-Bth COF and 40.4 ± 0.8 nm for TFB-Bth COF, respectively (Supplementary Figs. 3, 4). The diminished thickness observed in the TFB-Bth COF nanofilm is attributable to the higher hydrophilicity of the TFB monomer (n-octanol/water partition coefficient: 4.85 for Tp vs. 1.32 for TFB). This characteristic facilitates the facile diffusion of TFB into the aqueous phase, promoting bulk phase-transfer polymerization rather than a strictly interfacial reaction33. Conversely, the more hydrophobic nature of Tp monomers restricts their reaction to the interface, thereby leading to the formation of a comparatively thicker Tp-Bth COF nanofilm. Fourier transform infrared (FTIR) spectroscopy provides crucial insights into the COF linkages (Supplementary Fig. 5). The TFB-Bth COF exhibits a distinct vibrational signature at 1674 cm−1, signifying C = N bond formation and corresponding to the imine-linked framework. In contrast, a prominent peak at 1584 cm−1 in the Tp-Bth COF confirms the formation of C = C bonds, demonstrating the formation of irreversible β-ketoenamine linkage.

To elucidate the crystalline structural disparities between the two COFs, we conducted powder X-ray diffraction (PXRD) analyses on both two nanofilms. The Tp-Bth COF exhibited distinct diffraction peaks at 6.9o and 26.8o, whereas the TFB-Bth COF showed broadened and less intense diffraction peaks, suggesting a weakly ordered structure due to structural defects. Pawley refinement confirmed that the Tp-Bth COF adopts an AB stacking arrangement. The observed diffraction peaks at 6.9o and 26.8o correspond to the (100) and (002) crystal planes, respectively, showing favorable agreement with the simulated pattern for AB stacking (Fig. 2c). However, the limited hydrogen-bonding network in TFB–Bth COF yields inherent low crystallinity. This results in closely approximated goodness-of-fit residuals (Rwp, Rp) for both the AA- and AB-stacking models, preventing an unambiguous distinction based on these metrics alone. Crucially, however, a direct visual comparison between the experimental XRD pattern and the simulated profiles (AA- and AB-stacking; Fig. 2d and Supplementary Fig. 7) exhibited a substantial match with the AA-stacking simulation. Consequently, we assign the AA-stacking mode to TFB–Bth COF as the best-supported structural interpretation available from the current data. Further elucidation of crystalline anisotropy using two-dimensional grazing-incidence wide-angle X-ray scattering (2D-GIWAXS) revealed a marked contrast between the two COFs. The Tp-Bth COF exhibited a highly oriented (100) crystallographic plane with vertically aligned nanochannels, as evidenced by its pronounced scattering intensity at Qx,y = 0.49 Å−1 (Fig. 2e). In comparison, the TFB-Bth COF (Fig. 2f) displayed a relatively random orientation. This obvious distinction is primarily attributed to the extensive interlayer hydrogen-bonding networks within the Tp-Bth COF. These networks impose supramolecular confinement effects, establishing long-range crystallographic order that markedly enhances channel regularity and minimizes defect formation. High-resolution TEM (HRTEM) imaging of Tp-Bth COFs clearly revealed long-range ordered lattice fringes with a characteristic spacing of 0.30 nm, which is indexed to the (002) crystallographic plane (Fig. 2g). Complementary selected area electron diffraction (SAED) patterns, exhibiting bright diffraction spots, further corroborated the crystallinity of the Tp-Bth COF film. In sharp contrast, the HRTEM image of the TFB-Bth COF (Fig. 2h) shows a weakly ordered structure, with the corresponding SAED pattern (inset) displaying diffuse rings, characteristic of low crystallinity. Notably, HRTEM images acquired from multiple regions across the Tp-Bth COF (Supplementary Figs. 8, 9) consistently illustrated that crystalline domains predominantly occupied the entire prepared COF nanofilms, highlighting their uniform, high-quality microstructure.

It is reported that robust hydrogen-bonding interactions often induce spectral displacements in the IR spectrum of their constituent molecular units28. To gain real-time spectroscopic evidence of these dynamic interactions under thermal perturbation, we performed variable-temperature FT-IR (VT-IR) experiments (Fig. 2i). As predicted, hydrogen bonding was disrupted at elevated temperatures, resulting in blue shifts in the stretching frequencies of the β-ketoenamine carbonyl and N–H bonds. This spectral change directly signifies an increase in the bond orders of these moieties33,34,35. For each VT-IR experiment, spectra were collected at temperatures ranging from room temperature up to 150 °C. In specific, the β-ketoenamine N–H peak, initially at 3066 cm−1 at room temperature, gradually shifted to 3079 cm−1 at 150 °C, an overall shift of 13 cm−1. This thermal response upon heating directly indicates the disruption of the hydrogen bonding and a resulting enhancement in the N–H bond order arising from the formation of unbound amine groups28. Similarly, the β-ketoenamine carbonyl peak at 1677 cm−1 blue-shifted by 9 cm−1 to 1686 cm−1 when heated to 150 °C. This shift indicates an increased bond order due to the formation of free amide carbonyl moieties at higher temperatures.

To further determine the structural configurations of COF nanofilms, solid-state 13C nuclear magnetic resonance (13C NMR) spectroscopy was conducted (Fig. 2j). The TFB-Bth-COF spectrum showed a clear resonance at 146.59 ppm, corresponding to the imine carbon (C = N) derived from Schiff base reaction between TFB and Bth monomers, thereby confirming its successful framework construction. Furthermore, characteristic chemical shifts at 180.17 ppm and 99.85 ppm are assigned to carbonyl (C = O) and olefinic (C = C) moieties, respectively, validating the formation of β-ketoenamine linkages in Tp-Bth COF. Supplementary Fig. 12 shows the temporal evolution of Tp-Bth COF film surface architectures under controlled synthesis durations. As the reaction time increased, the resulting films exhibited progressively denser microstructures, indicative of enhanced structural consolidation during the growth process. This structural refinement correlates with reaction time-dependent crystallization kinetics, where prolonged monomer assembly enables defect annihilation via hydrogen-bonding modulated self-correction.

Morphologies and structural characteristics of COF membranes

Building upon the synthesis of free-standing COF nanofilms, we engineered two distinct COF variants in situ atop a porous Kevlar support via IP reaction in a custom-built diffusion cell (Supplementary Figs. 13, 14). The resultant Tp-Bth COF membrane presented a distinct yellow color, differentiating it from the pale-yellow Kevlar substrate, while the TFB-Bth COF membrane exhibited a grayish-white appearance (Supplementary Fig. 15). SEM images (Fig. 3a, b) showed continuous and compact surface morphologies of both COF membranes in contrast to porous Kevlar support (Supplementary Fig. 16). This validates the formation of defect-free COF layers with good mechanical flexibility.

Fig. 3: Morphology characterization and physicochemical properties of the membranes.
Fig. 3: Morphology characterization and physicochemical properties of the membranes.The alternative text for this image may have been generated using AI.
Full size image

a–e Surface SEM, Cross-sectional TEM, and AFM images of Tp-Bth (a, c) and TFB-Bth (b, d) COF membranes (insert: digital images of COF membranes with a diameter of 4.5 cm). f, g XPS spectra and Zeta potentials of Kevlar, Tp-Bth, and TFB-Bth COF membranes. h, i N2 adsorption-desorption isotherm and pore size distributions for the Tp-Bth and TFB-Bth nanofilms. j The rejection performance of uncharged organic solutes (the feed concentration is 1000 ppm).

The impact of monomeric concentration on the surface morphological structure of as-fabricated COF membranes was conducted through characterization by SEM and AFM. Evidently, elevating either Bth or Tp concentration led to a significant enhancement of nucleation densities and film compactness (Supplementary Figs. 17–19). Higher monomeric concentrations contributed to accelerated reaction kinetics that favor heterogeneous nucleation, leading to the formation of more polymeric nodules that coalesce into densely packed surface36. Similarly, prolonged reaction duration facilitated the subsequent growth and crystallization, thereby leading to the formation of well-structured and compact COF selective layers (Supplementary Fig. 20). AFM analyses further revealed a marked decrease from 19.7 nm to 8.9 nm in surface roughness when increasing the reaction time from 6 to 24 h (Supplementary Fig. 21). This decrease primarily correlates with particle growth and coalescence, which led to smoother and thicker COF layers. Likewise, the TFB-Bth membrane exhibited a more densely packed surface as increasing monomeric concentrations (Supplementary Figs. 22–24). The presence of larger polymeric nodules on the surface is likely related to higher hydrophilicity and reactivity of TFB molecules, leading to uncontrolled phase-change polymerization that yielded randomly distributed polymeric particles.

Furthermore, the as-fabricated COF membranes exhibited relatively hydrophobic surfaces, characterized by water contact angles ranging from 80o to 90o (Supplementary Figs. 25, 26), showing minor dependence on monomer concentrations. This surface hydrophobicity is primarily attributed to the abundance of non-polar aromatic skeletons and the low-polarity imine linkages integrated within the framework structure. Crucially, however, the extensive hydrogen-bonding networks inherent to the COF structure can function as preferential pathways for water molecules. Overall, the observed surface hydrophobicity is unlikely to significantly impede their RO application in water desalination. Cross-sectional TEM images (Fig. 3c, d) revealed the COF layer thickness of 92.3 nm (Tp-Bth) and 88.3 nm (TFB-Bth), consistent with cross-sectional SEM data (Supplementary Fig. 27). Furthermore, cross-sectional TEM imaging was conducted on Tp-Bth COF membranes synthesized using varying reaction times. As visualized in Supplementary Fig. 28, the membrane thickness exhibits a progressive increase corresponding to the extension of the reaction time from 3 h to 12 h. Furthermore, our engineered Tp-Bth COF membranes exhibited significantly lower surface roughness (Ra = 7.26 nm) compared to Kevlar supports (Ra = 23.5 nm, Supplementary Fig. 16). This distinct contrast in nanoscale smoothness not only provides evidence for the formation of a smooth COF layer but also demonstrates its potentially high fouling resistance.

The elemental composition of the hydrogel support and the two COF membranes was determined by XPS analyses, with results detailed in Fig. 3f and Table S1. Tp-Bth COF membranes exhibited atomic ratios of 64.50% carbon, 5.69% nitrogen, and 29.81% oxygen, which is characteristic of the pre-designed oxygen-enriched framework structure. In contrast, TFB-Bth COF showed a distinct distribution: 69.35% carbon, 17.25% nitrogen, and 13.4% oxygen, reflecting the substantial architectural divergence between the two COF membranes. Furthermore, high-resolution deconvoluted XPS spectra provided additional confirmation for the presence of β-ketoenamine linkages in Tp-Bth and imine bonds in TFB-Bth, aligning with our previous FT-IR and 13C NMR analyses (Supplementary Fig. 29). In specific, N 1 s peaks at 400.9 eV (N–H) and 400.1 eV (N–C), along with O 1 s peaks at 532.8 eV (O = C) and 532.2 eV (O-C), confirmed the β-ketoenamine formation within Tp-Bth COF membranes (Supplementary Figs. 29d–f). Conversely, the C 1 s spectrum of TFB-Bth COF membranes showed a distinct peak at 285.2 eV, indicative of imine bond formation, which is further corroborated by corresponding peaks in the N 1 s and O 1 s spectra (Supplementary Figs. 29g–i). Intriguingly, Zeta potential measurements revealed that the presence of the Tp-Bth COF layer substantially increased the isoelectric points of the composite membranes compared to both the TFB-Bth membrane and the hydrogel support (Fig. 3g). This charge alteration mainly stems from the protonation of the densely populated amine (-NH) groups within the crystalline Tp-Bth framework, a process facilitated by the presence of acetic acid. The resulting local electropositivity is critical, as it promotes the strong electrostatic repulsion of monovalent cations, thereby enhancing the overall water desalination efficacy.

Given the critical role of COF layer pore size in membrane desalination, their intrinsically specific surface area and pore size distribution were analyzed using Brunauer-Emmett-Teller (BET) characterizations. The nitrogen adsorption and desorption isotherms (Fig. 3h–i) for both Tp-Bth and TFB-Bth nanofilms consistently display a classic Type I isotherm, marked by a pronounced low-pressure adsorption surge (P/P₀ < 0.1). This characteristic profile establishes the prevalence of microporous structures within both COF films. BET data revealed a substantially higher specific surface area for the Tp-Bth film (60.5 m2 g−1) compared to that of the TFB-Bth film (40.5 m2 g−1). Notably, the Tp-Bth COF film exhibited a smaller pore size of 0.64 nm and a distinctly narrower ultramicroporous distribution in contrast to its TFB-Bth counterpart. This distinct difference in pore architecture provides strong evidence for their disparate stacking configurations and indicates structural regularity along with a narrower pore size distribution for the Tp–Bth film. As shown in Supplementary Fig. 30, the intrinsic pore size of TFB–Bth COF is measured to be 12.1 Å, which aligns closely with the pore size determined by BET measurements, further corroborating its AA-stacking mode. Conversely, the notable discrepancy observed between the intrinsic pore size and the experimentally measured pore size in Tp–Bth COF validates its AB-stacking configuration. We attribute this structure to the high crystallinity of the Tp-Bth film, a characteristic that simultaneously facilitates the formation of a long-range ordered pore architecture and yields the narrower, more uniform ultramicropores essential for high-selectivity desalination.

To assess pore characteristics in an aqueous environment, we performed molecular weight cut-off (MWCO) measurements on both COF membranes, using uncharged alcohols as model solutes (Table S2). The determined MWCO values were 70.4 Da for Tp-Bth and 384 Da for TFB-Bth membranes, respectively (Fig. 3j), aligning with the BET characterization results. Furthermore, probability density distribution profiles were derived from these rejection curves, as depicted in the Supplementary Fig. 31. A comparative analysis revealed a distinction between the hydrodynamic diameter calculations via MWCO (0.35 nm) and the BET-derived geometric pore size (0.64 nm) for Tp-Bth COF membranes. We presume this dimensional mismatch arises from the formation of hydrogen-bonded hydration layers, where oxygen/nitrogen moieties within the framework establish hydrogen-bonding interactions with adjacent water molecules, reducing accessible pore dimensions37. The abundance of ultramicroporous pores within high-crystalline Tp-Bth COF films is expected to enable a high desalination efficiency.

Desalination performance of ultramicroporous COF membranes

The desalination performance of as-fabricated COF membranes was assessed using a 2000 ppm NaCl solution as the feed, operating at 15 bar and room temperature. As plotted in Fig. 4a, elevating the Bth concentration from 0.5 to 1.25 mM (Tp = 1.75 mM) led to a consistent increase in NaCl rejection from 99.2% to 99.6%, with a minor trade-off in water permeability (1.7 L m−2 h−1 bar−1). Moreover, the as-synthesized membranes demonstrated good reproducibility, with the performance error bars remaining within an acceptable range. This observed trend is primarily ascribed to the formation of a more densely packed membrane surface at higher Bth concentrations, supported by SEM and AFM analyses. However, at Bth concentrations exceeding 1.5 mM, a double volume of acetic acid was required to ensure a homogeneous aqueous phase. The presence of this excessive acid negatively impacted film formation and compromised the intrinsic hydrogen-bond density, leading to a noticeable decline in desalination performance. This hypothesis is further corroborated by the lower desalination performance observed in Supplementary Fig. 32 across varying Tp concentrations (0.5–2 mM) at a fixed 1.5 mM Bth concentration. On the other hand, low monomeric concentrations (Tp ≤ 1.25 mM, Bth ≤ 1 mM) led to relatively loose crosslinked networks in the COF layer, resulting in salt rejection below 99.0% (Supplementary Fig. 33).

Fig. 4: Performance and application of Tp-Bth and TFB-Bth COF membranes.
Fig. 4: Performance and application of Tp-Bth and TFB-Bth COF membranes.The alternative text for this image may have been generated using AI.
Full size image

a, b Water permeance and NaCl rejection of Tp-Bth COF membranes under different Bth content and reaction time. c Separation performance of the TFB-Bth, Tp-Bth, and SW30 membranes toward a 2000 ppm NaCl and 5 ppm boron feed solution. d, e Digital photographs, SEM images (scale bar 1 μm) of Tp-Bth membrane acid immersion process. f Performance of the long-term pH stability (pH = 3) of the Tp-Bth membrane at a feed pressure of 15 bar and room temperature (2000 ppm NaCl). g Evaluation of antifouling performance: Normalized water flux versus operation time for Tp-Bth COF membrane using BSA and SA as model foulants. The fouling tests comprise three stages: (1) collecting baseline of water flux using a feed solution of 2000 ppm NaCl, (2) recording flux when fouled by 200 ppm model foulants in a feed solution of 2,000 ppm NaCl and (3) evaluating flux recovery using a feed solution of 2,000 ppm NaCl after cleaning the membrane via deionized water flushing. h flux recovery ratio (FRR), flux decline Ratio (FDR), reversible flux decline rate (DRr) and irreversible flux decline rate (DRir) for Tp-Bth COF membrane. i Radar plot comparing the physicochemical properties of Tp-Bth and TFB-Bth COF membranes. j Long-term operational stability performance of Tp-Bth membrane. k Membranes performance in this work compared to other membranes in the literature. Data are presented as mean ± SD (n = 3 for a–c).

Figure 4b illustrates a non-monotonic relationship between NaCl rejection and reaction time: initial increases are followed by subsequent declines. This trend arises because extended reaction duration facilitates epitaxial growth and crosslinking, elevating water/salt selectivity, but then causes over-aggregation and structural defects that compromise performance38. Following reaction parameter optimization, our engineered Tp-Bth COF membrane exhibited an exceptionally high NaCl rejection (99.6%) and good water permeability of 1.7 L m−2 h−1 bar−1. We attribute this performance primarily to its narrow ultramicroporous pore size distribution (centered at 0.64 nm) resulting from H-bond mediated high crystallinity, as well as an elevated isoelectric point that promotes effective sodium ion rejection. Using the identical methodology, the optimal desalination performance of TFB-Bth membranes was also explored (Supplementary Figs. 35–37). In contrast, TFB-Bth membranes showed a substantially lower NaCl rejection of 92.7%. This diminished performance correlates with their larger average pore size and higher MWCO, likely due to a comparatively low crystallinity. Furthermore, we investigated the effect of salt concentration on the rejection performance. Notably, the membrane exhibited minimal loss in rejection rate when the NaCl concentration was raised from 2000 ppm to 10,000 ppm (Supplementary Fig. 38). This demonstrates its exceptional stability under high osmotic pressure and confirms its strong potential for use in seawater desalination.

The substantial difference in NaCl rejection (92.7% vs. 99.6%) between the two COF membranes is attributed to the following key factors: (i) Hydrogen-bond-stabilized β-ketoenamine structure: The presence of adjacent phenolic hydroxyl groups on the Tp monomer enables enol-keto tautomerism, forming a stable β-ketoenamine linkage. This oxygen-rich structure promotes the formation of dense hydrogen-bond networks contributing to enhanced salt rejection; (ii) Enhanced crystallinity via molecular restriction: Extensive hydrogen-bonding networks restrict molecular rotation and suppress disordered stacking, promoting higher crystallinity and long-range ordered pore channels. Moreover, these networks largely strengthen interlayer binding energy and reduce interlayer spacing; (iii) Precise stacking transition mediated by hydrogen bonds: Rich intralayer hydrogen-bonds minimize stacking energy discrepancies during COF growth, enabling a controlled transition from the AA-stacking in TFB-Bth COF to the AB-stacking in Tp-Bth COF; (iv) Defect suppression and ultramicropore formation: This crystallinity consequently inhibits defect formation. The AB-stacking mode provides ultra-microporous channels ( ~ 0.64 nm), smaller than the hydrated sodium ion diameter ( ~ 0.72 nm), enabling highly efficient size exclusion. In contrast, the hydrogen-bond-deficient TFB-Bth COF membrane exhibits disordered stacking and a relatively large, non-uniform pore size distribution, leading to a lower desalination efficiency.

Given the exceptional desalination performance of Tp-Bth COF membranes, we further investigated their efficiency in boron removal from seawater. Boron, a prevalent natural element, predominantly exists as boric acid in aquatic environments, with typical seawater concentrations ranging from 4 to 6 ppm39. Boron removal poses a significant challenge, generally requiring multi-stage RO to meet drinking water standards40. Notably, in a single pass, Tp-Bth membranes effectively diminished boron concentration from 5 ppm to 1.14–1.25 ppm (75.9 ± 1.1% rejection, Fig. 4c), thus meeting stringent drinking water standards (1.0–2.4 ppm)41,42. Furthermore, a comparative analysis was conducted between Tp-Bth COF membranes and a state-of-the-art commercial brackish water RO membrane (BW30, DuPont). As illustrated in Fig. 4c, the Tp-Bth membrane presented distinctly higher NaCl and boron rejection than the BW30. In specific, the Tp-Bth membrane demonstrated a water permeability of 1.7 ± 0.09 L m−2 h−1 bar−1, with 99.6 ± 0.05% NaCl rejection and 75.9 ± 0.11% boron rejection, compared with 3.24 ± 0.16 L m−2 h−1 bar−1, 96.6 ± 0.6% and 32.3 ± 0.18% for BW3039,43. The demonstrated improvement of the Tp-Bth membrane in both salt and boron rejection establishes it as a highly promising candidate, directly enabled by our hydrogen-bond fortification strategy. To further evaluate the practical utility of our fabricated Tp-Bth membranes for water desalination, we subsequently conducted systematic investigations into their antifouling performance, acid stability, and operational stability.

To address the inherent acid sensitivity of conventional imine-linked COFs, we engineered β-ketoenamine-linked COF membranes to overcome this persistent limitation. Following a 30-day immersion in a pH 3 acidic solution, SEM confirmed the membranes retained their structural integrity (Fig. 4d, e), and their water permeability and salt rejection remained consistently stable (Fig. 4f). ATR-FTIR analysis (Supplementary Fig. 39) showed unchanged spectra of the Tp-Bth COF membrane after prolonged acid exposure, confirming robust acid stability, mainly due to their stable β-ketoenamine linkages. The non-reversible nature of the enol-to-keto tautomerization equilibrium and the robust hydrogen-bonding interactions between adjacent layers play a dual role in conferring this enhanced stability in acidic environments44. SEM and AFM images (Supplementary Figs. 40, 41) further supported this, revealing negligible structural morphological changes after 60 days of acid immersion, which correlated with only a slight decrease in desalination performance (Table S3).

To assess the anti-fouling properties of Tp-Bth COF membranes, bovine serum albumin (BSA) and sodium alginate (SA) were utilized as model organic foulants. Normalized water flux trends in response to 200 ppm foulant exposure and subsequent water rinsing were monitored (Fig. 4g). All membranes showed a reduction in flux when exposed to BSA or SA, followed by flux restoration after water flushing. Our comprehensive anti-fouling assessment utilized four key performance indicators: flux recovery ratio (FRR), flux decline ratio (FDR), reversible flux decline rate (DRr), and irreversible flux decline rate (DRir). The fouling resistance of the Tp-Bth membrane against both BSA and SA model foulants is evident in Fig. 4h, with FRR/FDR/DRr/DRir values of 95.0/11.5/6.5/5.0% and 98.3/18.6/2.0/16.6%, respectively. The absence of a consolidated cake layer on the fouled Tp-Bth COF membrane in SEM images (Supplementary Fig. 42) further supports its antifouling capability. We attribute this low fouling propensity to the low surface roughness of the Tp-Bth COF selective layer, coupled with the introduced H-bond networks. These networks may facilitate the effective immobilization of a significant number of water molecules, creating a stable hydrated layer on the membrane surface that serves as a highly effective antifouling barrier.

Long-term stability is of paramount significance for assessing the industrial viability of RO membranes. As displayed in Fig. 4j, the Tp-Bth membrane exhibits significant durability, sustaining continuous operation at 15 bar pressure for 80 h with NaCl rejection consistently at 99.5%. Despite a slight flux decrease, the membrane retained a high desalination performance, underscoring its potential for practical use. Concurrently, this provides additional evidence for the unparalleled stability inherent to the β-ketoenamine architecture. As demonstrated in Fig. 4k, the desalination performance of our Tp-Bth COF membrane sets a benchmark for COF membranes and outperforms many conventional alternatives, establishing its competitiveness in water purification technologies.

Mechanistic insights into hydrogen-bond fortification

To understand the critical role of hydrogen-bonding interactions in COF membrane crystallization, we conducted systematic density functional theory (DFT) simulations. These calculations examined charge changes during hydrogen-bond formation and the energy relationships of different conformational states. The comparative analysis reveals that Tp-Bth COF, featuring its extensive hydrogen-bonding network, offers improved interlayer interactions, higher structural stability, and boosted solvent affinity relative to the TFB-Bth COF.

Differential charge densities in Fig. 5a, b reveal charge redistribution localized predominantly within individual layers of the TFB-Bth COF. The consistent electron loss and gain patterns across adjacent layers demonstrate electrostatic repulsion, thus confirming the van der Waals (vdW) interactions as the primary mode of interlayer bonding for TFB-Bth COF45. In contrast, Fig. 5b shows a pronounced charge density decrease at the amine-associated hydrogen atoms (light blue shade), coupled with electron density accumulation at oxygen and nitrogen centers within adjacent molecular layers (yellow shade), indicative of hydrogen-bonding interactions.

Fig. 5: Mechanistic insights into hydrogen bond formation and performance enhancement.
Fig. 5: Mechanistic insights into hydrogen bond formation and performance enhancement.The alternative text for this image may have been generated using AI.
Full size image

a Differential charge density of TFB-Bth COF. b Differential charge density of Tp-Bth COF (blue, decrease in electron density; yellow, increase in electron density). c DFT calculation of the interlayer binding energy of TFB-Bth COF and Tp-Bth COF. d DFT analysis of torsional energy barriers in single-layer covalent organic frameworks (COFs), with dihedral angles referenced to the central N-N bond within the molecular framework. Both TFB-Bth and Tp-Bth systems exhibit initial torsion angles of −60 °. The horizontal axis represents the torsional angle variation spanning from −60 ° (minimum) to 60 ° (maximum). The figure illustrates the molecular conformations at dihedral angles of −60 °, 0 °, and 60 °. e Interaction energy obtained from DFT calculations and the growth of the COF membranes.

A direct comparative analysis of interlayer binding energies between the two COF systems (Fig. 5c) reveals substantially augmented interfacial cohesion in the hydrogen-bond-rich Tp-Bth COF compared to its TFB-Bth counterpart. Furthermore, structural characterization shows a decreased interlayer distance of 3.2 Å for Tp-Bth COF, aligning with our previous TEM measurements and distinctly lower than that of TFB-Bth COF (3.7 Å). The intralayer hydrogen-bonding geometries for both COF systems were further quantified. As depicted in Supplementary Fig. 43, Tp-Bth COF exhibits significantly shorter hydrogen-bond distances compared to TFB-Bth COF (1.7 vs. 2.3 Å), demonstrating stronger interaction energies for Tp-Bth COF according to Jeffrey’s classification criteria46. The combined charge density distribution analysis, bond length measurements, and binding energy calculations reveal that N–H···O hydrogen bonding interactions are crucial for substantially enhancing interlayer cohesion within COF structures. These collective findings provide multiscale evidence for the indispensable role of robust hydrogen-bonding networks in fine-tuning interlayer interactions and stabilizing the framework architecture.

Using the central N-N dihedral angle within COF monolayers as a structural probe, we comparatively analyzed bond rotation energetics in TFB-Bth and Tp-Bth systems (Fig. 5d). In the weakly hydrogen-bonded TFB-Bth COF, the dihedral angle exhibits unrestricted rotation across a broad angular range, primarily constrained by the steric interactions between adjacent functional groups. This considerable rotational freedom drives the formation of a three-dimensional hyperbranched architecture, while simultaneously leading to out-of-plane side-chain deviations that disrupt periodic stacking alignment along the z-axis. In contrast, Tp-Bth COF exhibits rotational constraints enforced by hydrogen bonds, wherein energy barriers sharply increase with angular deviation from equilibrium. This inherent molecular locking effectively suppresses stacking defects and spatial propagation errors during framework assembly through precise bond orientation control29. The fundamental difference in rotational dynamics—molecular confinement regulated by hydrogen bonds versus free rotation primarily limited by steric hindrance effects—dictates their distinct structural stabilities and functional performance.

Organic solvents critically control COF crystallization, molecular alignment, and pore structure via their physicochemical properties (viscosity, polarity)47,48,49,50. Computational analyses reveal distinct solvent-framework interactions: TFB-Bth COF shows a large intermolecular energy difference (0.44 eV) between water and dichloromethane, leading to random molecular orientation and heterogeneous pores51. Conversely, Tp-Bth COF exhibits a much smaller energy difference (0.07 eV) due to balanced interfacial forces. This near-equilibrium affinity promotes directional alignment during assembly, enhancing crystallization and yielding uniform pores. Overall, DFT calculations reveal that the synergy between hydrogen bonding interactions and reduced solvent affinity differences yields a more ordered crystal structure, resulting in more uniform pores.

Discussion

In summary, we successfully engineered two distinct COF membranes, demonstrating a hydrogen bond-regulated strategy that endows the Tp-Bth COF membrane with unprecedented crystallinity and an exceptional water desalination performance. This breakthrough performance significantly surpasses that of state-of-the-art COF membranes. The attributes of Tp-Bth are meticulously linked to three synergistic effects of hydrogen bonding: the formation of dense inter- and intra-layer hydrogen-bonding networks via a stable β-ketoenamine configuration, the induction of a highly ordered AB-stacking arrangement, and the suppression of structural defects during synthesis. Beyond its separation capabilities, the Tp-Bth membrane exhibits acid resistance and antifouling properties, coupled with validated long-term stability crucial for industrial applications. This work not only provides a powerful strategy for designing highly ordered COF membranes but also establishes a benchmark for high-performance desalination membranes for sustainable water management.

Methods

Fabrication of Kevlar hydrogel support membrane

Kevlar hydrogel support membranes were fabricated via wet phase inversion52,53,54. A clarified solution was first prepared by sonicating deionized water (1.2 mL) and KOH (1.2 g), followed by adding Kevlar fiber (1.2 g) and DMSO (60 mL) with 48 h of magnetic stirring. The resulting dark red solution was centrifuged at 6578 x g to remove bubbles and precipitates. The dope solution was manually cast (250 μm thickness) onto PP/PE non-woven fabric fixed to a glass plate. Phase inversion was achieved by immersion in 25 °C water, where DMSO/water exchange formed a transparent 3D porous hydrogel. After 10 min, the Kevlar hydrogel was transferred to a freshwater bath for over 6 h. The resulting supports exhibited water permeation flux of 500−600 LMH bar−1.

Fabrication of COF free-standing nanofilm and membrane

Free-standing COF nanofilms were synthesized at ambient conditions via interfacial polymerization. The organic phase contained Tp (1.75 mM, 22.05 mg) in DCM (15 mL), while the aqueous phase dissolved Bth (1.25 mM, 4.7 mg) and acetic acid (0.1 M) in water (15 mL). After layering the aqueous solution over the DCM phase in a Petri dish to form a static interface, the sealed system was reacted undisturbed for 12 h.

Tp-Bth/TFB-Bth thin films were fabricated in situ on Kevlar hydrogel supports via interfacial Schiff-base reactions. A 4.5 cm diameter Kevlar membrane was vertically mounted in a U-shaped diffusion cell (dual 18 mL chambers). The organic and aqueous phases were loaded into opposing chambers, followed by sealed reaction at 29 °C/30% RH for 12 h. Membranes were rinsed with water, soaked in ultrapure water, and characterized. TFB-Bth membranes were synthesized identically.