Dynamic bond-driven encapsulation of enzymes in metal–organic frameworks beyond pore size constraints

Abstract

Immobilizing enzymes in a metal–organic framework (MOF) is an effective approach to improve their stability and reusability, but the small pore sizes of most MOFs exclude many industrially relevant enzymes. Here, we present a dynamic bond-mediated approach that enables enzyme encapsulation beyond pore size constraints. We constructed a series of mesoporous MOFs by integrating robust trivalent metal–carboxylate clusters with dynamic divalent metal–pyridyl units. Systematic variation of metal combinations and linker lengths precisely tunes the framework stability and dynamics, enabling reversible dissociation and reformation of metal–pyridyl bonds. These dynamic bonds function as molecular gates, permitting enzymes larger than the intrinsic pores to infiltrate while preserving framework integrity. The strategy was applied to encapsulate diverse enzymes, preserving high enzymatic activity and enhancing operational stability. Furthermore, it supports the co-immobilization of multi-enzyme systems, such as NahK and GlmU, for efficient cascade synthesis of high-value glycosylated donors.

Metal–organic frameworks with dynamic metal–ligand bonds form reversible gates that admit enzymes larger than the pores, enabling stable encapsulation and efficient multi-enzyme cascade catalysis for glycan precursor synthesis.

Introduction

Enzymes are important biological catalysts with remarkable efficiency, regioselectivity, and stereoselectivity¹. However, their direct use in industrial processes is often hindered by limited stability and reusability^2,3. Immobilizing enzymes on solid supports has been demonstrated as an effective strategy to address these challenges by improving operational stability, recyclability, and storage lifetime while preserving catalytic efficiency and substrate specificity. Among the various host materials, metal–organic frameworks (MOFs), a class of crystalline porous solids constructed from metal nodes and organic linkers, have emerged as attractive platforms. MOFs combine high crystallinity, large surface areas, tunable porosity, and chemical modularity, enabling high enzyme loadings and providing structural confinement effects that enhance enzyme robustness^4–10. Despite these advantages, conventional enzyme immobilization strategies face significant limitations. Surface attachment via physical adsorption or covalent linkage is simple and versatile¹¹, but suffers from low loading capacity, poor enzyme retention, and potential activity loss. In situ enzyme encapsulation during MOF crystallization can accommodate large enzymes within small-pore frameworks^12,13. For example, several biomimetic mineralization, de novo water-based, and mechanochemical approaches have successfully achieved enzyme immobilization by enabling enzyme-friendly synthesis environments, demonstrating good universality and applicability^14–16. However, this strategy is limited to a narrow set of MOFs that can be synthesized under mild conditions¹⁷. Post-synthetic diffusion of enzymes into preformed MOFs with suitable pore sizes offers mild, biocompatible loading while preserving both MOF crystallinity and enzyme activity^7,18,19. However, this approach is constrained by pore size requirements: few MOFs possess pore apertures larger than 3 nm, while many enzymes exceed this size range. Collectively, the harsh synthetic conditions of stable MOFs coupled with the size mismatch between enzyme dimensions and MOF cavities have restricted the broader application of MOFs for enzyme immobilization.

The dynamic behaviors of MOFs, involving transient dissociation and reformation of coordination bonds or cooperative framework rearrangements, offer a potential route toward overcoming size restrictions. For example, exploiting linker dissociation-association dynamics, guest molecules larger than the intrinsic pore apertures have been successfully encapsulated into ZIF-8 and UiO-66 nanocrystals^20,21. More recently, enzymes have been shown to act as macro-ligands, competing with linkers to generate reversible defects that enable their entry into robust UiO-66¹⁸. However, this approach is limited by the low intrinsic dynamics of stable MOFs, which require long incubation times and nanosized MOF particles to achieve adequate enzyme loading. Thus, although these studies highlight the potential of framework dynamics for bypassing aperture limits, leveraging this behavior for the efficient encapsulation of large and unstable enzymes remains largely unexplored.

In this work, we present a dynamic coordination bond-driven strategy that enables the encapsulation of enzymes in MOFs beyond their intrinsic pore size limits. A family of mesoporous MOFs with tunable dynamic behavior was constructed by integrating robust trivalent metal–carboxylate clusters (M = Cr³⁺, Fe³⁺, and Al³⁺) with labile divalent metal–bipyridyl coordination units (M′ = Cu²⁺, Co²⁺, Ni²⁺, and Pd²⁺), which can be obtained as large single crystals suitable for direct structural determination by Single-crystal X-ray diffraction (SCXRD). These reversible coordination bonds can reversibly dissociate and reassociate, allowing the gradual diffusion and encapsulation of large enzymes exceeding the framework aperture, while maintaining framework crystallinity. By systematically varying the metal combinations and the length of organic linkers, we precisely modulate coordination bond dynamics and MOF stability to optimize enzyme loading. Time-resolved confocal laser scanning microscopy directly visualizes enzyme ingress, revealing diffusion pathways distinct from the intrinsic MOF channels. Spectroscopic and crystallographic analyses, in combination with density functional theory (DFT) calculations, confirm both the occurrence of dynamic bond reorganization and the preservation of the framework structure. The strategy is compatible with a diverse range of enzymes, including cytochrome C (Cyt C), horseradish peroxidase (HRP), lipase, β-glucosidase (BGL), and nitroreductase (NTR). Furthermore, the platform enables the co-immobilization of N-acetylhexosamine 1-phosphate kinase (NahK) and N-acetylglucosamine 1-phosphate uridyltransferase (GlmU) for multi-enzyme cascade synthesis of high-value glycosylated donors.

Results and discussion

Construction of MOFs with dynamic bonds

To enable enzyme encapsulation beyond pore size constraints, we developed a tunable MOF platform (M-L-M′) that integrates three key design elements: (i) robust high-valent metal–carboxylate clusters (M = Cr³⁺, Fe³⁺, and Al³⁺) to impart chemical stability, (ii) labile divalent metal–bipyridyl coordination site (M′ = Cu²⁺, Co²⁺, Ni²⁺, and Pd²⁺) to introduce reversible dynamic bonding, and (iii) heteroditopic organic linkers of varying lengths (L = L1, L2, and L3, Fig. 1) to tune pore dimensions and framework flexibility. Together, these elements allow systematic modulation of dynamic behavior while maintaining overall framework integrity.

Fig. 1.

Illustration of a dynamic bond-driven enzyme encapsulation strategy.

As a representative example, the solvothermal reaction of Cr³⁺, L1, and Cu²⁺ afforded single crystals of Cr-L1-Cu (Fig. 2b). SCXRD revealed a hexagonal P6/mmm framework composed of Cr₃O(OOC)₆ clusters, Cu²⁺ ions, and L1 linkers. Each Cu²⁺ center is coordinated by four pyridyl groups from L1 and two hydroxyl ligands. Each Cr₃O(OOC)₆(OH)(H₂O)₂ cluster comprises three Cr³⁺ ions bridged by one μ₃-O and six carboxylates, terminated by two water and one hydroxyl ligand (Fig. 2a). The resulting framework features hexagonal channels of 1.8 nm. Topologically, the Cu(L1)₄ and Cr₃O(OOC)₆ units serve as four-connected and six-connected nodes, generating a network with stp topology (Fig. 2c–e)^22,23.

Fig. 2. Single-crystal structures of Cr-L-M′.

a Coordination structures of Cu(L1)₄, Cu(L2)₄, Cu(L3)₄, and the Cr₃O(OOC)₆ cluster. b Single-crystal structures of Cr-L1-Cu, Cr-L2-Cu, and Cr-L3-Cu. c–e Topological representation of four-connected and six-connected nodes, and the resulting STP network. f–h PXRD patterns before and after metal exchange for Cr-L1-M′, Cr-L2-M′, and Cr-L3-M′ (M′ = Cu²⁺, Co²⁺, Ni²⁺, and Pd²⁺). Insets show crystal images.

Replacing L1 with longer L2 and L3 affords isostructural frameworks. The pore sizes were extended to 2.2 nm and 3.0 nm for Cr-L2-Cu and Cr-L3-Cu, respectively (Fig. 2b). Furthermore, M-L-M′ can be synthesized from different high-valent metal clusters. Fe- and Al-based analogues were synthesized under similar conditions as Cr-L1-Cu, using Fe(NO₃)₃ or Al(NO₃)₃ in place of Cr(NO₃)₃ (Figs. S1–5). For example, single crystal structures of Fe-L1-Cu revealed a framework nearly identical to Cr-L1-Cu, with a slightly longer a-axis length (22.648 Å for Fe-L1-Cu, 22.297 Å for Cr-L1-Cu) due to the longer Fe–O bonds (Table S1)

Cr-L-M′ (L = L1, L2, and L3) with different divalent metals (M′ = Co²⁺, Ni²⁺, or Pd²⁺) were synthesized by the post-synthetic metal exchange of Cr-L-Cu (Fig. 2f–h). Interestingly, single crystal structures of Cr-L1-Pd and Cr-L3-Pd (Table S2) revealed that the Pd²⁺ exchange resulted in the loss of the two equatorial ligands, transforming the six-coordinate CuN₄O₂ octahedral center into a four-coordinate PdN₄ square-planar center. This geometry is consistent with the d⁸ configuration of Pd²⁺ complexes, which favor square-planar coordination due to the large ligand-field splitting that elevates the $d_{{\text{x}}^2-{\text{y}}^2}$ orbital energy, allowing all eight d-electrons to occupy lower-energy orbitals for maximum stabilization. The counterions of the PdN₄ center could not be refined crystallographically, but were identified as Cl⁻ by SEM-EDX mapping (Figs. S6–8).

Phase purity of all the M-L-M′ series (M = Cr³⁺, Fe³⁺, and Al³⁺; L = L1, L2, and L3; M′ = Cu²⁺, Co²⁺, Ni²⁺, and Pd²⁺) was confirmed by powder X-ray diffraction (PXRD, Fig. 2f–h, and Tables S3–S6), which matched simulated patterns from SCXRD-derived models. Inductively coupled plasma optical emission spectroscopy (ICP-OES) of the digested MOF determined the compositions of the Cr-L-M′ series from metal exchange (Tables S7–S9). SEM imaging showed preserved morphology, and EDX mapping confirmed uniform element distribution throughout each crystal (Figs. S5–8).

Due to the large pore sizes and flexible frameworks, the M-L-M′ series collapsed after solvent removal, even when activated using supercritical CO₂, making N₂ adsorption measurements unsuitable for porosity evaluation. Instead, their porous nature was assessed using methylene blue (MB) adsorption experiments (Figs. S9–11)²⁴. The total MB uptake increased with linker elongation and showed a linear correlation with the porosity calculated from single-crystal structures. MOFs combining high-valent and low-valent metal centers with heteroditopic linkers have been documented previously^25–28, but our platform uniquely enables systematic tuning of the cluster type, divalent metal, and linker length. This modular design allows us to precisely control the balance between framework stability and dynamic bond behavior.

Tuning the stability and dynamic behavior of MOFs

To evaluate stability and dynamic behavior, M-L-M′ samples were incubated under enzyme-relevant conditions and across a range of pH values. The solid phases were examined by PXRD to assess crystallinity, while the supernatants were analyzed using ICP-OES, proton nuclear magnetic resonance spectroscopy (¹H NMR), and high-performance liquid chromatography (HPLC) for metal/linker release.

We first examined the effect of the trivalent metal cluster on the framework stability. PXRD (Fig. 3a, b) indicates that Cr-L1-Cu retains crystallinity in Tris-HCl buffer (pH 7) after 72 h, whereas Fe-L1-Cu and Al-L1-Cu rapidly decompose. The high stability of Cr-L1-Cu arises from the strong Cr–carboxylate bonds in the Cr₃O(OOC)₆ cluster. The d³ electronic configuration of Cr³⁺ half-fills the t_2g orbitals without e_g antibonding electrons, yielding stronger and more kinetically inert coordination bonds compared to Fe³⁺ (d⁵) or Al³⁺ (d⁰). Supernatant analysis further revealed the complete dissolution of Fe-L1-Cu within 45 min, releasing all the Fe, Cu, and L1 (Fig. 3c, d). In contrast, Cu²⁺ concentration in supernatant peaked after 36 h and then plateaued for Cr-L1-Cu, corresponding to ~2% of Cu²⁺ released from Cr-L1-Cu solid into the buffer solution. Meanwhile, no detectable Cr³⁺ or L1 release was observed. The selective Cu²⁺ release and preserved PXRD patterns indicate a reversible Cu–N dissociation/reformation equilibrium without framework collapse (Figs. S12–19).

Fig. 3. Stability and dynamic behavior of M-L-M′.

a Schematic illustration of the dynamic dissociation of MOFs in Tris-HCl buffer (pH = 7). b PXRD of Al-L1-Cu, Fe-L1-Cu, and Cr-L1-Cu after soaking in Tris-HCl buffer for 24 h. c Time-dependent leaching of Cu²⁺, Cr³⁺, Fe³⁺, and L1 into the supernatant during immersion of MOFs in Tris-HCl buffer. Leaching profiles of metals and linkers from M-L-M′ in Tris-HCl buffer, tuned by different high-valent metal nodes (d), low-valent metal nodes (e), and heteroditopic linkers (f). g–i Stability window of Cr-L1-M′, Cr-L2-M′, and Cr-L3-M′ determined by PXRD; Created with BioRender.com.

We then fine-tuned stability and dynamics by varying the divalent metals (Cu²⁺, Co²⁺, Ni²⁺, and Pd²⁺) and linker lengths (L1, L2, and L3) in Cr-L-M′ (Fig. 3e, f). While PXRD patterns confirmed crystallinity across all the Cr-L-M′ samples, supernatant analysis revealed clear differences in divalent metal leaching (Fig. S20). The divalent metal leaching followed the order Co > Ni > Cu, with no detectable Pd release. This trend is consistent with the intrinsic stability of metal–pyridyl coordination described by the Irving–Williams series, in which the stability of complexes formed by divalent first-row transition metal ions increases across the period to a maximum stability at Cu²⁺, driven by progressively enhanced ligand-field stabilization and metal–ligand covalent character. Consequently, Cu²⁺ (d⁹) and Ni²⁺ (d⁸) form shorter and more robust M–N bonds than Co²⁺ (high-spin d⁷), with Cu²⁺ further benefiting from Jahn–Teller–stabilized equatorial coordination. Pd²⁺ (second-row d⁸) exhibits even higher metal–pyridyl coordination stability owing to its stronger metal–ligand covalency, which results in a robust framework. The Pd-based MOFs (Cr-L-Pd, L = L1, L2, and L3) therefore showed no metal or linker loss, suggesting their high stability due to the strong Pd–N bonds. Divalent metal leaching rates increased with longer linkers (Fig. 3f). For example, the Cu²⁺ release increases from 2% in Cr-L1-Cu to 5% in Cr-L3-Cu. Meanwhile, the leaching of Cr³⁺ and L is much lower than that of divalent metals in all cases. These results indicate that the dynamic behavior of Cr-L-M′ arises primarily from partial dissociation of M′–pyridine bonds, while the framework remains intact.

We further examined the stability of Cr-L-M′ in aqueous solutions across a wide pH range (Fig. 3g–i). Across Cr-L-M′ with different divalent metals, the stability order followed Pd > Ni ≈ Cu > Co, consistent with ligand exchange rate constants, where Pd–N bonds are the most inert²⁹. Linker elongation from L1 to L2 and L3 slightly reduced the pH stability window, with L2 and L3-based MOFs showing comparable stability ranges. Together, these results demonstrate that the stability and dynamic behavior of M-L-M′ frameworks can be systematically tuned by varying the high-valent cluster, the divalent metal center, and the linker length.

Mechanistic study of dynamic behavior

The mechanism of dynamic behavior was investigated using Cr-L3-Cu as a representative example. X-ray photoelectron spectroscopy (XPS) verified that the oxidation states and coordination environments of the metal centers were preserved after treatment in Tris-HCl buffer (pH 7), while the increased proportion of uncoordinated pyridinic N indicated the partial breaking of Cu–pyridyl bonds (Fig. S21a). Raman spectroscopy provided complementary evidence, where the pyridine N–Cu vibrational band from Cr-L3-Cu decreased in intensity after buffer treatment, indicating partial bond dissociation. In contrast, the corresponding pyridine N–Pd band in Cr-L3-Pd remained unchanged, consistent with the relative stability differences (Fig. S21b). Infrared (IR) spectra before and after treatment showed no significant changes in coordination bond signatures or linker-associated vibrations. In particular, the band at 3446 cm⁻¹, corresponding to coordinated H₂O, exhibited no increase in intensity, confirming the intact Cr–carboxylate bonds (Fig. S21c). The breaking of Cr–carboxylate bonds in water would be expected to replace the carboxylate with terminal Cr–H₂O/OH, leading to an enhanced IR peak at 3446 cm^{−1 30}. PXRD patterns remained unchanged, indicating retention of the crystalline structure (Fig. S21d). Consistently, methylene blue adsorption showed comparable uptake before and after treatment, excluding the formation of noticeable defects (Fig. S10). Thermogravimetric analysis (TGA) also revealed no shift in the major weight-loss step at ~350 °C, assigned to linker decomposition, confirming that the overall metal–linker composition was maintained and that no nanoparticle formation occurred during treatment (Fig. S21e). Taken together, these results demonstrate that the dynamic behavior originates from reversible Cu–N bond dissociation without disrupting Cr–carboxylate bonds or the framework intactness.

To mechanistically understand the stability trend of Cr–carboxylate and Cu–pyridine sites in water, DFT calculations were performed using the Quantum ESPRESSO package^31–33. Water substitution at the Cr–carboxylate sites required a high energy input (1.19 eV), consistent with robust Cr–O coordination. In contrast, water substitution at Cu–pyridine sites was thermodynamically favorable (–0.50 eV, Fig. S21f). The large energy difference between Cr–carboxylate and Cu–pyridine dissociation also arises from the confinement of the framework. Replacing a Cr–carboxylate bond would distort adjacent Cu–N coordination in the linker, which is highly directional and depends on σ-orbital overlap between pyridyl N and Cu d-orbitals. By contrast, the more ionic and less directional Cr–O bonds provide greater structural flexibility, accommodating the distortion and thereby stabilizing the water-bound Cu complex (Tables S10–12).

Combining spectroscopic, crystallographic, and computational results, we propose that the dynamic behavior of Cr-L-Cu originates from selective Cu–N bond dissociation. In aqueous buffer, water as a competing ligand substitutes at Cu–pyridine sites, causing partial Cu²⁺ release. The process is reversible, as pyridyl groups readily re-coordinate, restoring the structure. Meanwhile, water substitution at Cr–carboxylate sites requires much higher energy (1.19 eV) and is therefore suppressed, ensuring the Cr–O backbone remains intact. Solvent-dependent tests show that dynamic Cu–N dissociation occurs only in coordinating aqueous media, whereas organic solvents give negligible metal release; this behavior reflects the water-substitution mechanism at Cu–pyridine sites and indicates that solvent coordination strength dominates the dynamics, whereas pH and temperature mainly influence the overall stability of the MOFs (Figs. S22, 23).

Dynamic bond-driven enzyme encapsulation

Building on the stability and dynamic nature of the Cr-L-M′ series, we investigated their enzyme adsorption behavior. Cyt C (2.6 × 3.2 × 3.3 nm, Fig. S24), which is larger than the intrinsic pore apertures of these MOFs, was selected as a model enzyme to probe the immobilization process. The loading capacity of Cyt C strongly depended on both divalent metal centers and linker lengths. Pd²⁺-based frameworks (Cr-L1-Pd, Cr-L2-Pd, and Cr-L3-Pd) showed no detectable enzyme uptake, consistent with their absence of dynamic behavior. In contrast, Cu²⁺, Co²⁺, and Ni²⁺ based MOFs displayed noticeable Cyt C loading (Fig. 4a). The linker elongation also increases the Cyt C uptake, as evidenced by the higher Cyt C loading of Cr-L2-Cu and Cr-L3-Cu than that of Cr-L1-Cu (Figs. S25, 26). Consequently, Cr-L3-M′ (M′ = Cu²⁺, Co²⁺, and Ni²⁺) represents an ideal platform for enzyme immobilization, achieving an optimal balance between structural stability and framework dynamics that results in high enzyme loading capacity.

Fig. 4. Enzyme loading behavior and characterization in Cr-L3-M (M = Cu, Co, Ni, Pd) frameworks.

a A time-dependent loading chart of Cyt C in Cr-L3-M (M = Cu, Co, Ni, Pd). b Schematic diagram of the temporal distribution of Cyt C@Cr-L3-M and CLSM images of Cyt C@Cr-L3-M′ fixed at various incubation times, Scale bar, 50 μm. c FT-IR spectra of Cyt C@Cr-L3-M compared to those of Cyt C and Cr-L3-M. Plots of the loading capacities of HRP (d), Lipase (e), and NTR (f) in Cr-L3-M. Error bars represent the standard deviations of three independent measurements.

We employed confocal laser scanning microscopy (CLSM) and fluorescein isothiocyanate (FITC)-labeled enzymes (Fig. S27) to monitor the time-dependent diffusion of enzymes into the MOF crystals. CLSM images revealed that enzymes diffused uniformly from the exterior surface toward the center of the needle-shaped crystals (Figs. 4b, S28–31). Since the hexagonal channels of the MOF are aligned parallel to the long axis of the crystals, the observed diffusion pattern in CLSM images suggests a diffusion pathway distinct from the intrinsic pore network. This behavior contrasts with previous reports on hexagonal channel MOFs such as NU-1000 and PCN-600, where enzyme diffusion occurred predominantly from both ends toward the center of the needle-shaped crystals, in alignment with the channel direction³⁴. The uniform diffusion observed in our work indicates an alternative permeation mechanism independent of the inherent channel structure. We propose a encapsulation mechanism enabled by the dynamic dissociation of the divalent metal–pyridyl coordination bonds (M′ = Cu²⁺, Co²⁺, and Ni²⁺), which transiently generates localized structural apertures sufficiently large to permit enzyme entry. The reformation of metal–pyridyl bonds subsequently “seals” the enzyme within the framework.

PXRD confirmed that Cr-L3-M′ (M′ = Cu²⁺, Co²⁺, and Ni²⁺) frameworks retained their crystallinity after enzyme encapsulation (Fig. S32). Fourier transform infrared spectroscopy (FT-IR) spectra (Figs. 4c, S33), TGA (Fig. S34), and MB adsorption measurements (Fig. S35) collectively verify the immobilization of enzymes within MOFs. To evaluate the generality of this strategy, other large enzymes with diverse isoelectric points, stabilities, and catalytic functions were examined, including HRP (4.4 × 4.6 × 6.8 nm), lipase (6.9 × 5.1 × 6.8 nm), BGL (6.0 × 5.0 × 6.4 nm), and NTR (5.2 × 4.5 × 2.9 nm). All of these enzymes exhibited trends similar to Cyt C (Figs. 4d–f, S36–45).

Catalytic performance of enzyme@dynamic MOFs

The catalytic performance of Cyt C, HRP, lipase, BGL, and NTR immobilized in Cr-L3-M was evaluated, where the activity trend followed Cr-L3-Ni > Cr-L3-Cu > Cr-L3-Co > Cr-L3-Pd (Figs. S46–53). The superior performance of Cr-L3-Ni is tentatively attributed to strong coordination between Ni²⁺ and histidine (His) residues/6* His-tags of enzymes, which promote stable enzyme binding. In contrast, the rigid Cr-L3-Pd prevented enzyme penetration into the cavities, leading to low loading and poor catalytic activity.

Given its strict reliance on the bulky nicotinamide adenine dinucleotide (NADH) cofactor and its high sensitivity to temperature, pH, and organic solvents, NTR represents an ideal model for probing the protective advantages of dynamic MOF encapsulation^35–37. These characteristics impose stringent demands on mass transport and environmental stability, making NTR particularly suitable for evaluating whether the Cr-L3-Ni framework can enhance enzyme robustness under challenging conditions. To evaluate the protective effects, the activities of NTR@Cr-L3-Ni and free NTR were compared under conditions including organic solvents, alkali, and high temperatures. Free NTR exhibited a significant decrease in activity (Fig. 5a–d), whereas NTR@Cr-L3-Ni maintained most of its initial activity, demonstrating the effective protective capabilities of Cr-L3-Ni. Moreover, the NTR@Cr-L3-Ni consistently showed high catalytic efficiency over five cycles (Fig. 5e). PXRD analysis further confirmed that NTR@Cr-L3-Ni maintained its crystalline structure after multiple catalytic cycles (Fig. S54). Collectively, these findings highlight the effectiveness of the dynamic bond-based enzyme immobilization strategy, enabling versatile enzyme loading while preserving activity and enhancing operational stability.

Fig. 5. Biocatalysis, protection effect, and reusability of NTR@Cr-L3-Ni.

a A schematic illustration of the enzyme preparation and immobilization process. b–d Stability of NTR@Cr-L3-Ni under organic solvents/additives: (Ethanolamine, Sodium dodecyl sulfate, Tetrahydrofuran, DMF, and Urea) (b), alkalinity (c), temperature (d); e Cyclic capacity of NTR@ Cr-L3-Ni; f The reduction reaction of nitroarenes in the presence of NTR or NTR@Cr-L3-Ni; g The conversion rate of the reduction reaction of nitroarenes catalyzed by NTR and NTR@Cr-L3-Ni. Error bars represent the standard deviations of three independent measurements. Created with BioRender.com.

To further assess the substrate scope of NTR@Cr-L3-Ni, we examined nitroaromatic compounds with electron-donating, electron-withdrawing, or bulky substituents. The immobilized NTR@Cr-L3-Ni delivered good conversions comparable to free NTR (Fig. 5f, g), underscoring its broad applicability.

Multi-enzyme cascade catalysis

N-acetyl-D-glucosamine (GlcNAc) is a key component of numerous bioactive carbohydrates, including heparin, hyaluronan, and chitosan^38–40. For incorporation into glycan biosynthetic pathways, GlcNAc must be activated into its corresponding sugar nucleotide form, uridine 5′-diphospho-N-acetyl-D-glucosamine (UDP-GlcNAc), which serves as a valuable substrate for carbohydrate synthesis, a potent enzyme inhibitor, and an essential assay reagent. Efficient in vitro preparation of UDP-GlcNAc is therefore highly desirable, but the instability of natural enzymes limits large-scale production^41,42. We employed the Cr-L3-Cu/Co/Ni platform to co-immobilize NahK and GlmU (denoted as NahK-GlmU, Fig. 6a, b) for cascade synthesis of UDP-GlcNAc. Notably, both NahK and GlmU are large, metal-sensitive, and intrinsically unstable glycosyltransferases that are notoriously difficult to immobilize by conventional methods, further underscoring the universal capability of the dynamic MOF matrix to accommodate and stabilize labile enzymes. FT-IR, TGA, Zeta potential measurements, and PXRD patterns (Figs. 6e, S55–58) collectively confirmed the successful incorporation of the enzymes into MOFs. CLSM imaging of FITC-tagged NahK and Rhodamine B (RhB)-tagged GlmU confirmed that two enzymes were co-immobilized within the pores of Cr-L3-Cu/Co/Ni (Fig. 6c). The resulting biocatalyst exhibited superior catalytic activity compared with free enzymes (Fig. 6d–f), owing to shortened diffusion pathways between two enzymatically active sites. It also retained >80% activity after five cycles (Fig. 6g) and preserved function under harsh conditions, whereas free enzymes rapidly deactivated (Figs. S59, 60). These results establish Cr-L3-Cu/Co/Ni as a robust and versatile platform for multi-enzyme cascade biocatalysis.

Fig. 6. Encapsulation and catalytic activity of NahK-GlmU in Cr-L3-Cu/Co/Ni.

Size analysis of NahK (a) and GlmU (b). c CLSM images of FITC-NahK@Cr-L3-Cu/Co/Ni and RhB-GlmU@ Cr-L3-Cu/Co/Ni, Scale bar, 50 μm. d Scheme of the synthesis of UDP-GlcNAc catalyzed by NahK-GlmU. e PXRD patterns of calculated Cr-L3-Cu/Co/Ni-simulation and NahK-GlmU@Cr-L3-Cu/Co/Ni. f The synthesis of UDP-GlcNAc catalyzed by NahK and GlmU and NahK-GlmU@Cr-L3-Cu/Co/Ni. g The cyclic capacity of NahK-GlmU@Cr-L3-Ni. Error bars represent the standard deviations of three independent measurements.

Overall, we have developed a tunable MOF platform that combines robust high-valent metal–carboxylate clusters, labile divalent metal–pyridyl coordination sites, and heteroditopic linkers of varying lengths to achieve a controllable balance between structural stability and framework dynamics. This design enables reversible coordination bond dissociation and reformation for the encapsulation of enzymes larger than the intrinsic pore sizes of the framework. This strategy enables the immobilization of diverse enzymes at high loading capacities while maintaining their catalytic activity and improving operational stability. This dynamic bond-driven approach offers a general design principle for size-independent encapsulation of biomacromolecules, overcoming long-standing pore-size limitations and paving the way for advanced biocatalysis and biomedical applications.

Methods

Materials and instrumentation

All the reagents and solvents were commercially available and used as received. Deuterated solvents were purchased from Cambridge Isotope Laboratory (Andover, MA). Single-crystal X-ray diffraction intensity data for Cr-L-Cu (L = L1, L2, and L3) were collected on a Bruker D8 Venture and Liquid diffractometers. ¹H NMR and ¹³C NMR were recorded on a Bruker Avance III 400 NMR spectrometer. FT-IR spectra were recorded on a Vector 27 Bruker Spectrophotometer by transmission through KBr pellets containing ground crystals in the 4000–400 cm⁻¹ range. TGA data were obtained on a TGA 4000 thermal analysis system at a heating rate of 5 °C min⁻¹ under an air atmosphere. The PXRD patterns were collected at room temperature using a scan speed of 0.1 s/step on a Bruker Advance D8 (40 kV, 40 mA) diffractometer equipped with Cu radiation. Simulated PXRD patterns were generated from single-crystal data using Mercury 3.0. UV-vis absorption spectra were recorded on a Shimadzu UV-3600. Metal content analyses were obtained by inductively coupled plasma optical emission spectroscopy (ICP-OES) using the Jarrell-Ash 1100 + 2000 instrument. The CLSM images were obtained by Nikon AX with the excitation at 488 nm. HPLC analysis was performed on an Agilent 1260 Infinity II LC system.

Synthesis of Cr-L1-Cu

Cr(NO₃)₃ ∙ 9H₂O (60 mg, 0.15 mmol), L1 (73.8 mg, 0.6 mmol), Cu (NO₃)₂ ∙ 2.5H₂O (46.5 mg, 0.2 mmol), and DMF (10 mL) were combined in a 20 mL Pyrex vial and sonicated for 5 min. The resulting suspension was sealed and heated at 85 °C for 12 h, affording blue-green hexagonal crystals, which were isolated by filtration and washed with DMF (37.1 mg).

Synthesis of Cr-L1-M′ (M′ = Co, Ni, and Pd)

Cr-L1-Cu crystals (37 mg) were immersed in a DMF solution of CoCl₂·6H₂O (40 mg in 4 mL), NiCl₂·6H₂O (40 mg in 4 mL), or PdCl₂ (10 mg in 4 mL), respectively. For Co and Ni exchange, samples were heated at 75 °C; for Pd, 60 °C. The supernatant was refreshed every 6 h. After 24 h, crystals were collected by filtration and washed with DMF (3×). Yields: Cr-L1-Co (35.6 mg), Cr-L1-Ni (32.5 mg), Cr-L1-Pd (41.3 mg).

Synthesis of Fe-L1-Cu

Fe(NO₃)₃ ∙ 9H₂O (60.6 mg, 0.15 mmol), L1 (73.8 mg, 0.6 mmol), Cu (NO₃)₂ ∙ 2.5H₂O (46.5 mg, 0.2 mmol), and DMF (10 mL) were combined, sonicated (5 min), sealed, and heated at 85 °C for 12 h. Brown hexagonal crystals were isolated by filtration and washed with DMF (40.1 mg).

Synthesis of Al-L1-Ni

Al(NO₃)₃ ∙ 9H₂O (56.3 mg, 0.15 mmol), L1 (73.8 mg, 0.6 mmol), NiCl₂ ∙ 6H₂O (35.7 mg, 0.05 mmol) were dissolved in DMF (10 mL), sonicated for 5 min, and heated at 85 °C for 24 h. Bluish hexagonal crystals were obtained (44.4 mg) after isolation and washing.

Synthesis of Al-L1-Cu

Al-L1-Ni crystals (44 mg) were immersed in DMF solutions of CuCl₂·2H₂O (40 mg in 4 mL). The supernatant was refreshed every 6 h. After 24 h at room temperature, Al-L1-Cu crystals were collected and washed with DMF (38.6 mg).

Synthesis of Cr-L2-Cu

Cr(NO₃)₃∙9H₂O (40 mg, 0.1 mmol), 4-pyridyl acrylic acid (L2, 59.7 mg, 0.4 mmol), Cu (NO₃)₂ ∙ 2.5H₂O (46.5 mg, 0.2 mmol), and DMF (10 mL) were combined, sonicated (5 min), sealed, and heated at 85 °C for 24 h. Yellow-green hexagonal crystals were isolated (19.8 mg).

Synthesis of Cr-L2-M′ (M′ = Co, Ni, Pd)

Cr-L2-Cu crystals (20 mg) were immersed in DMF solutions of MCl₂ (M = Co, Ni, Pd; 40 mg CoCl₂·6H₂O, 40 mg NiCl₂·6H₂O, or 10 mg PdCl₂ in 4 mL DMF) at 75 °C. The supernatant was refreshed every 6 h. After 24 h, crystals were collected and washed with DMF, yielding Cr-L2-Co (18.6 mg), Cr-L2-Ni (19.5 mg), and Cr-L2-Pd (20.3 mg).

Synthesis of Fe-L2-Co

Fe(NO₃)₃∙9H₂O (40.4 mg, 0.1 mmol), L2 (59.7 mg, 0.4 mmol), and CoCl₂•6H₂O (47.6 mg, 0.2 mmol) were dissolved in DMF (10 mL), sonicated for 5 min, sealed, and heated at 85 °C for 24 h. After cooling and standing at room temperature for another 24 h, yellow-green crystals were isolated (19.3 mg).

Synthesis of Al-L2-Cu

Al(NO₃)₃ ∙ 9H₂O (37.5 mg, 0.1 mmol), L2 (59.7 mg, 0.4 mmol), Cu (NO₃)₂ ∙ 2.5H₂O (46.5 mg, 0.2 mmol), and DMF (10 mL) were combined, sonicated for 5 min, sealed, and heated at 85 °C for 24 h. Bluish crystals were collected and washed with DMF (17.6 mg).

Synthesis of Cr-L3-Cu

Cr(NO₃)₃ ∙ 9H₂O (20 mg, 0.05 mmol), 4-Pyrid-4-ylbenzoic acid (L3, 40 mg, 0.2 mmol), Cu (NO₃)₂ ∙ 2.5H₂O (18.8 mg, 0.1 mmol), and DMF (10 mL) were combined, sonicated (5 min), sealed, and heated at 85 °C for 24 h. Green hexagonal crystals were obtained (13.4 mg).

Synthesis of Cr-L3-Cu-small

Cr(NO₃)₃ ∙ 9H₂O (80 mg, 0.2 mmol), 4-Pyrid-4-ylbenzoic acid (L3, 160 mg, 0.8 mmol), Cu (NO₃)₂ ∙ 2.5H₂O (75 mg, 0.4 mmol), DMF (10 mL), HNO₃ (300 μL)were combined, sonicated (5 min), sealed, and heated at 85 °C for 24 h. Blue-green hexagonal crystals of 1–5 μm were obtained (20.1 mg).

Synthesis of Cr-L3-M′ (M′ = Co, Ni, and Pd)

Cr-L3-Cu crystals (13 mg) were immersed in DMF solutions of MCl₂ (M = Co, Ni, Pd; 20 mg CoCl₂·6H₂O, 20 mg NiCl₂·6H₂O, or 10 mg PdCl₂ in 4 mL DMF at room temperature. The supernatant was refreshed every 6 h. After 24 h, exchanged crystals were isolated and washed with DMF, affording Cr-L3-Co (10.3 mg), Cr-L3-Ni (10.8 mg), and Cr-L3-Pd (12.5 mg).

Synthesis of Fe-L3-Cu

Fe(NO₃)₃ ∙ 9H₂O (20 mg, 0.05 mmol), L3 (39.9 mg, 0.2 mmol), Cu (NO₃)₂ ∙ 2.5H₂O (18.8 mg, 0.1 mmol), and DMF (10 mL) were combined, sonicated, sealed, and heated at 85 °C for 48 h. Yellow-green hexagonal crystals were obtained (28.4 mg).

Synthesis of Al-L3-Ni

Al(NO₃)₃ ∙ 9H₂O (18.8 mg, 0.15 mmol), L3 (39.9 mg, 0.2 mmol), Ni (NO₃)₂ ∙ 6H₂O (29.1 mg, 0.1 mmol), and DMF (10 mL) were treated as above, yielding bluish crystals after 48 h (22.4 mg).

Synthesis of enzyme@Cr-L3-M′ (M′ = Cu, Co, Ni, and Pd)

Cr-L3-M crystals (10 mg) were washed with ultrapure water and incubated with the enzyme (8 mg) in Tris-HCl buffer (pH 7.6, 2 mL) in a 4 mL polyethylene tube under stirring at room temperature for 48 h. The crystals were centrifuged at 8000 × g for 3 min, and the supernatant was retained for loading efficiency analysis. The precipitate was washed three times with ultrapure water to remove unbound enzyme.

Stability test

The as-synthesized MOFs (30 mg) were placed in 4 mL aqueous solutions (pH = 1–13, adjusted by HCl and NaOH) at room temperature for 24 h. Then, the sample was collected by filtration and washed with water for further PXRD investigations.

Methods for fitting and calculating parameters of adsorption isotherms

The Langmuir isotherm equation is expressed as follows:

$$q_e=\frac{q_m{K}_L{C}_e}{1+K_L{C}_e}$$ 1

where q_e (mg g⁻¹) is the adsorption capacity of dye at equilibrium, C_e (mg L⁻¹) is the concentration of dye in solution at equilibrium, q_m (mg g⁻¹) is the Langmuir maximum adsorption capacity, K_L(dm³ mg⁻¹) is the Langmuir constant, K_F ((mg g⁻¹) (L mg⁻¹) ^1/n) is the Freundlich constant, and n indicates the Freundlich exponent.

Measurement of the protein loading content

The loading of Cyt C and HRP in L3-Cr-M (M = Cu, Co, and Ni) was examined by examining the concentration of enzymes in the supernatant at different incubation times after the centrifugation of a mixed solution of enzymes and MOFs. Standard curves of Cyt C and HRP were constructed in the range from 0 to 4 mg/mL (Fig. S37a, b). Typically, a 200 μL CytC/HRP sample was added to 96-well plates and detected by a UV-vis spectrophotometer at 550 nm and 403 nm, respectively.

Expression and purification of enzymes

All recombinant strains harboring corresponding genes were cloned and stored in our laboratory (Table S13). Briefly, Escherichia coli BL21 (DE3) strains containing the target plasmid were cultured in Luria-Bertani (LB) medium with 100 μg/ mL antibiotic (ampicillin, kanamycin, or carbenicillin) at 37 °C under vigorous shaking at 8.9 g. When OD600 reached 0.6–0.8, isopropyl β-D-1-thiogalactopyranoside (IPTG) with a final concentration of 0.1 mM was added and cultured at 16 °C overnight. Cells were harvested by centrifuging at 8944 g for 15 min. Protein purification was performed by utilizing Ni-NTA Sepharose affinity resin. The Ni²⁺ column was then washed with buffer A (50 mM Tris-HCl, pH 7.9, 500 mM NaCl, 10 mM imidazole), and the enzyme was eluted with buffer B (50 mM Tris-HCl, pH 7.9, 500 mM NaCl, 300 mM imidazole) in sequence. The purified fraction was desalted by centrifugal filter (Millipore, 10 kDa) to remove imidazole and other components, and protein concentration was determined by Bradford protein assay kit.

Fluorescein-tagged enzymes

Fluorescein isothiocyanate (FITC, 4 mg) and enzyme (CytC, HRP, NTR, Nark, and GlmU, 20 mg) were dissolved in carbonate-bicarbonate aqueous buffer solution (0.1 M, pH 9.2, 2 mL) and left for 2 h in darkness at room temperature under gentle stirring. The fluorescein-tagged enzyme was dialyzed against dialysis buffer (1 L, 50 mM Tris-HCl containing 500 mM NaCl, pH 7.9) for 72 h at 4 °C in the dark, and the dialysis buffer was changed every 4 h. Transfer the dialyzed enzyme into a fresh tube and centrifuge it at 8944c g for 10 min at 4 °C to remove precipitate, if any. The supernatant was then concentrated with Amicon® Ultra-15 Centrifugal Filter Unit (30 kDa) to 1 ml by centrifugation (3420 g). Similarly, rhodamine B isothiocyanate-labeled GlmU was obtained.

Measurement of the enzymatic activity of immobilized enzyme and free enzyme

The specific activities for Cr-L3-M (M = Cu, Co, Ni, and Pd) immobilized CytC/HRP and free CytC/HRP were measured with 1 mM ABTS, 2 mM H₂O₂, 1 μg enzyme in 200 μL NaOAc/HOAc (50 mM, pH 4) at 25 °С for 10 min. The catalytic efficiency was evaluated by UV-vis measurement at 420 nm. The specific activities for Cr-L3-M immobilized NTR and free NTR were measured with 5 mM p-nitrophenyl, 5 mM NADH, and 1 μg enzyme in 200 μL Tris-HCl buffer (200 mM, pH 7.4) at 37 °С for 1 h. The catalytic efficiency was evaluated by UV-vis measurement at 340 nm. The specific activities for Cr-L3-M immobilized β-glucosidase (BGL) and free BGL were measured using 5 mM p-nitrophenyl β-D-glucopyranoside (pNPG) as the substrate. Reaction mixtures contained pNPG and either free BGL (1 μg) or an equivalent amount of BGL@Cr-L3-M (based on the loading amount) in 200 μL sodium acetate buffer (50 mM, pH 5.0). The reactions were incubated at 37 °C for 10 min and then quenched with 1 mol/L Na₂CO₃. After centrifugation to remove MOF particles, the amount of released p-nitrophenol was quantified by UV-vis measurement at 400 nm. The specific activities for Cr-L3-Cu/Co/Ni immobilized NahK-GlmU and free NahK-GlmU were measured with 10 mM GlcNAc, 15 mM ATP, 15 mM UTP, 10 mM MgCl₂, 0.5 μg NahK, 0.5 μg GlmU in 200 μL Tris-HCl buffer (50 mM, pH 7.9) at 37 °С for 12 h.

HPLC detected formed sugar nucleotides. HPLC analysis was performed on an Agilent 1260 Infinity II LC system. The column YMC-Pack Polyamine II column (250 × 4.6 mm i.d., S-5 μm) was used to monitor UDP-sugar at 35 °C. Eluent A was a 20 mM sodium phosphate buffer, pH 5.8, and eluent B was methanol. Eluent (80% A, 20% B) was flowed (1 mL/min) through the column for 30 min. The column effluent was continuously detected by PDA at 262 nm.

Proteinase K treatment

In general, free enzymes, enzyme@Cr-L3-M (M = Cu, Co, Ni, and Pd), were prepared as aqueous solutions with an enzyme concentration of 4 mg/mL. 50 μl of each solution was added to a 1.5 mL centrifuge tube, followed by the addition of 50 μL of Proteinase K solution (20 mg/mL) to the tubes, while the control group received 50 μL of DI water. After incubating at 40 °C for 2 h, enzyme activity tests were conducted on all samples.

SDS-PAGE analysis

Before the protein analysis assay, 4.0 mg of Enzyme@Cr-L3-M (M = Cu, Co, Ni, and Pd) was dissolved in 0.2 M HCl (aq). Following a 20-min incubation, the solution underwent centrifugation (14,000 g) for a brief period. Subsequently, 15 µL of the supernatant was combined with 1.5 µL of loading buffer, and the mixtures were subjected to electrophoresis on a sodium dodecyl sulfate polyacrylamide gel (SDS-PAGE) consisting of a 4% polyacrylamide stacking gel and a 12% polyacrylamide resolving gel. Electrophoresis was carried out at 100 V under reducing conditions. Subsequently, the gel was subjected to Coomassie Blue Fast Staining following the provided instruction manual.

Determination of apparent rate constants

The apparent rate constants (k_obs) for Cyt C, HRP, NTR, and lipase encapsulated in Cr-L3-M frameworks were determined by monitoring the time-dependent absorbance changes of their corresponding chromogenic substrates. All kinetic assays were carried out under pseudo-first-order conditions, with the substrate concentration kept in large excess relative to the enzyme.

For Cyt C and HRP, the oxidation of ABTS was monitored at 420 nm. Because the absorbance increases as ABTS•⁻ accumulates, the kinetic traces were linearized using the expression:

$$\ln \left(A_{\infty }-A_t\right)=-k_{obs}t+C$$ 2

where A_t is the absorbance at time t, and A_∞ is the maximum absorbance obtained after the reaction reaches a steady state.

For NTR, the consumption of NADH was followed at 340 nm, where NADH exhibits a characteristic absorption band. In this system, the absorbance decreases with time, and the apparent rate constant was obtained by fitting:

$$\ln A_t=-k_{obs}+C$$ 3

which is equivalent to the first-order decay of NADH under excess-substrate conditions.

For lipase, the hydrolysis of p-nitrophenyl palmitate (PNPP) was quantified by monitoring the formation of p-nitrophenolate at 406 nm. Since the absorbance increases during product formation, k_obs was extracted by fitting:

$$\ln \left(A_{\infty }-A_t\right)=-k_{obs}t+C$$ 4

In all cases, linear regions of the plots (typically the initial 300–900 s, depending on the system) were used for regression. The values of k_obs were obtained from the slopes of the fitted lines. All measurements were performed in triplicate, and the reported rate constants represent mean values with standard deviations.

Recycling performance testing

The cyclic stability of NTR@Cr-L3-Ni was assessed by quantifying p-nitrophenol formation using a UV-vis assay at 340 nm. For each cycle, NTR@Cr-L3-Ni (1 mg) was added to 200 μL Tris-HCl buffer (200 mM, pH 7.4) containing 5 mM p-nitrophenyl substrate and 5 mM nicotinamide adenine dinucleotide (NADH), followed by incubation at 37 °C for 1 h. After centrifugation, the catalyst was recovered, washed twice with deionized water (7829 g, 2 min), and resuspended in a fresh reaction solution of identical composition to initiate the next cycle. Catalytic activity in each cycle was determined from the amount of p-nitrophenol produced, and normalized to the maximum activity observed within each batch (defined as 100%).

The recycling performance of NahK-GlmU@Cr-L3-Ni was evaluated by monitoring UDP-sugar formation via HPLC. For each cycle, Cr-L3-Ni (1 mg) containing immobilized NahK and GlmU was incubated in 200 μL Tris-HCl buffer (50 mM, pH 7.9) supplemented with 10 mM GlcNAc, 15 mM ATP, 15 mM UTP, 10 mM MgCl₂, 0.5 μg NahK, and 0.5 μg GlmU at 37 °C for 12 h. After centrifugation, the supernatant was analyzed on an Agilent 1260 Infinity II LC system equipped with a YMC-Pack Polyamine II column (250 × 4.6 mm, S-5 μm) at 35 °C using an isocratic eluent of 80% 20 mM sodium phosphate buffer (pH 5.8) and 20% methanol at 1 mL/min, with detection at 262 nm. The recovered solid was washed twice (7826 g, 2 min) and reintroduced into a fresh reaction solution to initiate the next cycle. UDP-sugar formation in each cycle was quantified and normalized to the highest yield within each batch (defined as 100%).

Single-crystal X-ray crystallography

Single-crystal X-ray diffraction intensity data for Cr-L1-Cu, Fe-L1-Cu, and Cr-L3-Pd were collected on a Bruker D8 Venture diffractometer fitted with a PHOTON-100 CMOS detector, monochromatized microfocus Mo Kα radiation (λ = 0.71073 Å), and a nitrogen flow controlled by a KRYOFLEX II low-temperature attachment operating at 193 K, 253 K, and 296 K. Raw data collection and reduction were controlled using APEX3 software⁴³. The structures were solved by direct methods and refined by full-matrix least squares. Absorption corrections were applied using the SADABS routine. ast-squares on F² using the SHELXTL software package⁴⁴.

Single-crystal X-ray diffraction intensity data for Cr-L1-Pd were collected on a Bruker D8 Liquid diffractometer fitted with a PHOTON-100 CMOS detector, monochromatized microfocus Ga Kα radiation (λ = 1.34139 Å), and a nitrogen flow controlled by a KRYOFLEX II low-temperature attachment operating at 223 K. Raw data collection and reduction were controlled using APEX3 software⁴³. Absorption corrections were applied using the SADABS routine. The structures were solved by direct methods and refined by full-matrix least-squares on F² using the SHELXTL software package⁴⁴.

Theoretical calculations

All first-principles calculations were performed with Perdew-Burke-Ernzerhof exchange-correlation functional⁴⁵ (PBE) using the PWscf code of Quantum ESPRESSO^31–33 (version 7.3.1). Optimized norm-conserving Vanderbilt (ONCV) pseudopotentials⁴⁶ from the PseudoDojo library⁴⁷ were employed, with a kinetic energy cutoff of 40 Ry for the plane-wave basis set and 160 Ry for the charge density and potential. The semiempirical Grimme’s DFT-D3 method⁴⁸ is used to account for the van der Waals correction. We found that only Γ point sampling was enough for the total energy to converge within 1 meV/atom in self-consistent calculation with a self-consistency convergence threshold of 1 × 10⁻⁸ Ry. All structure models were fully relaxed with a convergence criterion of 1 × 10⁻³ a.u for forces and 1 × 10⁻⁴ a.u for total energy. Initial magnetic moments of the Cr₃O cluster were set in an antiferromagnetic configuration. The start magnization setting for the Cr₃O cluster is antiferromagnetic. The relaxed crystal structure was visualized using the VESTA package.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

Y.C.L. and M.Q. contributed equally to this work. S.Y., J.-L.Z., and X.Z. conceived the original idea. Y.C.L. and M.Q. performed the synthesis. Y.C.L. and M.Q. performed the comprehensive structural characterization, property measurements, and data analysis. L.G. conducted the DFT calculations. Y.X.Z. performed the ICP-OES measurements. Y.Y. and J.Z. performed the CLSM measurements. Y.C.L., M.Q., S.Y., J.-L.Z., and X.Z. drafted the manuscript. All authors contributed to the revision of the manuscript.

Peer review

Peer review information

Nature Communications thanks Lien-Yang Chou, Fa-Kuen Shieh and the other anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

Previously published protein structures used in this study were obtained from the Protein Data Bank (PDB) under the following accession codes: 3CP5 (Cyt C) 1W4Y (HRP) 7×32 [10.2210/pdb7X32/pdb] (NTR) 1TRH (Lipase) 3VIK (BGL) 4OCO (NahK) 1FWY (GlmU) Crystallographic data for the structures reported in this Article have been deposited at the Cambridge Crystallographic Data Center, under deposition numbers CCDC 2489486 (Cr-L1-Cu), 2489487 (Cr-L1-Pd),2489488 (Fe-L1-Cu), and 2489489 (Cr-L3-Pd). Copies of the data can be obtained free of charge via https://www.ccdc.cam.ac.uk/structures/. All other data supporting the findings of this study are available within the article and its supplementary information. Source data are provided with this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-70249-x.