Design of bifunctional linkers

To facilitate the assembly of CA and KDGA, we designed bifunctional linkers composed of three parts: BzS, ethylene glycol, and maleimide (Fig. 1b). Ethylene glycol provides solubility and an adjustable length to connect the two proteins. The bifunctional linkers were prepared through a ten-step synthesis (Scheme 1, details are listed in the ESI^†), allowing for the substitution of BzS and adjustment of the ethylene glycol length with other protein components in future studies. Since the active site of CA is buried 15 Å from the surface, the linker required at least three repeating units of the ethylene glycol groups. Thus, we synthesized two bifunctional linkers with five (BzS5) or three (BzS3) ethylene glycol repeats, which provided either sufficient or close-fitting distances between the two proteins.

Open in a separate window

Scheme 1

Synthetic procedure of bifunctional linkers, BzS5 (b; n = 5) and BzS3 (a; n = 3). (i): Toluene, Ar, 90 °C; (ii): DCC, dry DMF, 0 °C to r.t.; (iii): p-TsCl, dry THF, 0 °C to r.t.; (iv): NaN₃, abs. EtOH, 0 °C to reflux; (v): PPh₃, THF/H₂O, 0 °C to r.t.; (vi): Boc₂O, DCM, 0 °C to r.t.; (vii): 1, DIAD, PPh₃, dry THF, 0 °C to r.t.; (viii): TFA, DCM, r.t.; (ix): 2, DIPEA, dry DMF, 0 °C to r.t.; (x): dry anisole/dry MeCN, reflux.

KDGA bioconjugation with bifunctional linkers

We examined the X-ray crystal structure of KDGA to determine the optimal position for introducing a cysteine residue. Although pre-existing ones might not entirely participate in thiol–ene bioconjugation, we prepared a double mutant (C120A/C150A or KDGA*) to eliminate any potential interference (Fig. S2e^†). We anticipated that E288C mutation would be suitable for attaching BzSn (n = 5 or 3) not to generate steric clash upon the binding of two CA proteins to each side of KDGA (Fig. 1d). High-resolution electrospray ionization mass spectrometry (HR-ESI-MS) and thiol assay results indicated that all four cysteine residues in a tetramer were successfully conjugated with the bifunctional linkers, yielding KDGA*-E288C|BzS5 (Fig. S3 and Table S5^†).

Self-assembly of CA and KDGA enzymes

We investigated the self-assembly of CA and KDGA using dynamic light scattering (DLS) (Fig. 2). The hydrodynamic diameters of the wild-type proteins, CA and KDGA, in TAPS buffer were determined to be 5.1 ± 0.30 nm and 9.7 ± 0.94 nm, respectively (Fig. 2a). These sizes are comparable to the calculated values (4.7 nm and 8.3 nm) using the X-ray crystal structures and WinHydroPro10.⁴³ The hydrodynamic diameter of KDGA*-E288C|BzS5 (10 ± 1.2 nm) was similar to that of the wild-type protein, indicating that attaching a flexible linker marginally altered the overall protein size.

Open in a separate window

Fig. 2

Dynamic light scattering (DLS) of heterooligomeric protein assembly. (a) CA, KDGA, and KDGA*-E288C|BzS5 alone (10 μM monomer). Before and after mixing CA and (b) KDGA or (c) KDGA*-E288C|BzS5. (d) The addition of CBzS (10 μM) to the CA/KDGA assembly in 50 mM borate, pH 8.0 buffer. The protein samples in (a)–(c) were prepared in 50 mM TAPS, pH 8.0, 100 mM Na₂SO₄ buffer.

Mixing the wild-type proteins CA and KDGA overnight at room temperature without the linker did not induce any spectral changes (Fig. 2b). However, mixing KDGA*-E288C|BzS5 with CA (CA/KDGA) in equal monomeric ratios immediately generated an opaque solution with white precipitates. When we mixed the samples at 4 °C for 15 min to reduce nonspecific aggregation, the solution was transparent but yielded protein complexes with an increased hydrodynamic diameter of 13 ± 0.60 nm (Fig. 2c). The size was comparable to the simulated value (11 nm) using the programs AutoDock Vina,⁴⁴ HADDOCK,⁴⁵ and WinHydroPro10, assuming that all added CA proteins bound to a homotetramer of KDGA*-E288C (Fig. 1d and S4^†). Therefore, these results indicate that CA and KDGA, two structurally and functionally unrelated proteins, form heterooligomers via specific interactions between BzS5 and the Zn cofactor in CA.

We investigated whether protein assembly could be reversed by adding CA inhibitors, such as 4-carboxybenzenesulfonamide (CBzS). Notably, we used borate buffer (pH 8.0) because CBzS and CA alone created small yet detectable aggregates in TAPS, but not in borate buffer (Fig. S5^†); heteroassembly was similarly observed in borate buffer with a slight size increase (14 ± 0.11 nm), possibly due to a buffer-specific effect in determining hydrodynamic diameters⁴⁶ (Fig. 2d). The addition of CBzS reverted the size of the pre-assembled heterooligomers back to 9.2 ± 0.21 nm, reporting that CBzS competes with the BzS5 linker for CA binding and reversibly releases the KDGA component. The reversibility of heterooligomer self-assembly and disassembly are akin to how small molecules transmit chemical signals to macromolecules and regulate intermolecular interactions in biological systems.

Characterization of CA/KDGA heterooligomeric enzymes

The molecular interactions between CA and its inhibitor function as the primary driving force for protein self-assembly and determine the fractions of bound CA in the heteroassembly. Thus, the hydrolytic activity of unbound CA with p-nitrophenyl acetate (pNPA)⁴⁷ could be a sensitive and accurate readout to quantitatively analyse how many protein components participate in self-assembly and how strongly these PPIs occur (Fig. 3a). The specific activity and catalytic efficiency of the wild-type protein CA towards pNPA were measured to be 5.3 ± 0.34 μmol min⁻¹ mg⁻¹ and 2500 ± 290 M⁻¹ s⁻¹, respectively (Fig. S6^†), consistent with previous reports.⁴⁸ The addition of unmodified wild-type KDGA protein did not affect the catalytic activity of CA.

Open in a separate window

Fig. 3

Quantitative analysis of heterooligomeric protein self-assembly. (a) A reaction scheme of CA with pNPA before (catalytically on/inhibitor-unbound state) and after (off/bound) the addition of BzS-containing molecules. (b) The specific activity of CA (1.0 μM) with pNPA upon the addition of CBzS or KDGA*-E288C-conjugated with BzS5 or BzS3 linkers. (c) The specific activity of CA or 1:1 mixing of the CA/KDGA assemblies at various concentrations. (d) Temperature-dependent circular dichroism analysis of CA, KDGA*-E288C, and the 1:1 ratio of the assembled species (5 μM monomer in total).

When KDGA*-E288C|BzS5 (1 μM) was added at a 1:1 monomer ratio to CA, esterase activity was reduced by 54% (Fig. 3b). The incomplete shutdown of CA activity can be attributed to the modest binding affinity of the BzS and 10-fold lower protein concentration than in DLS experiments described above. A 5-fold excess of KDGA*-E288C|BzS5 to CA led to 90% inactivation of the esterase activity, which led to the estimated K_i value of BzS5 attached to the KDGA protein to be 370 ± 53 nM, using the Dynafit.⁴⁹ This number is only marginally altered from the K_i value of CBzS (170 ± 37 nm), suggesting that the molecular functionalization and attachment to a macromolecule barely perturbed the binding affinity to the CA as shown previously.^50–52 Based on these quantitative analyses, we estimated that 94% of the CA and KDGA*-E288C|BzS5 proteins would participate in the heteroassembly when 10 μM of each monomer was mixed for the DLS experiments shown in Fig. 2c.

Next, we replaced the CA with other isoforms to determine whether the binding affinity of BzS can be quantitatively translated to protein oligomerization. CA1 or CA12 have 60% and 36% sequence identity to CA, respectively, and exhibit significantly different binding affinities (K_i values) to a BzS analog, p-azo benzenesulfonamide: 1015 nM, 92 nM, and 87 nM in CA1, CA, and CA12, respectively.⁵³ The mixing of KDGA*-E288C|BzS5 protein with CA12, having a comparable binding affinity to the BzS-like molecule, yielded a similar size growth to CA. In contrast, CA1, with at least a 10-fold weaker binding affinity than CA, led to a negligible alteration in the hydrodynamic diameter of the protein upon mixing with KDGA*-E288C|BzS5 (Fig S7^†). These results indicate that the binding affinity of the BzS moiety serves as the primary driving force for protein self-assembly. These data also suggest that heteroassembly occurs with high selectivity for protein building blocks. When we switched to the shorter linker (BzS3), KDGA*-E288C|BzS3 exhibited a lower level of CA inhibition, resulting in an estimated K_i value of 1.2 ± 0.06 μM (Fig. 3b). These data suggest that BzS5 exhibits the optimal length to accommodate the formation of an asymmetric CA–KDGA interface for self-assembly. In addition, we measured the hydrolytic activities of the CA/KDGA assembly after serial dilutions (Fig. 3c). Presumably due to the reversibility of BzS-driven PPIs, the net catalytic activities were inversely proportional to the final protein concentrations and consistent with the simulation using the K_i value. These results indicate that the association and dissociation of the heteroassembly are reversible and kinetically labile.

Furthermore, we assessed the secondary structures and thermal stabilities of the protein components and self-assembled heterooligomers by circular dichroism (CD) spectroscopy. Each protein component (CA and KDGA*-E288C) showed α-helical structures, and the spectrum was retained after protein heterooligomerization (Fig. S8a^†), suggesting that the flexible linker enforced the two proteins in proximity with no significant structural perturbation. Temperature-dependent CD spectral changes at 210 nm demonstrated that CA and KDGA showed melting temperatures (T_m) at 54.5 °C and 96.7 °C, respectively, consistent with the reported values (Fig. 3d).^54,55 The mixing of CA and KDGA*-E288C|BzS5, where 68% of the BzS is expected to be bound to the CA, revealed two discrete T_m values, 54.7 °C and 100 °C. Because DLS data showed no temperature dependence below the T_m of CA (Fig. S8b^†), these data suggest that each component protein retained its native thermal stability and behaved independently while being physically connected. These results contrast with other designed protein assemblies with increased thermal stability.^56–58 The distinction could be attributed to the discrete assembly mechanism; BzS5 containing ethylene glycol chains provides structural flexibility and sufficient distance between the two component proteins. Thus, this unique feature could benefit the formation of self-assembled protein architectures where their individual biochemical properties need to be preserved in an additive manner.

Self-assembly of dimeric CA and KDGA enzymes

To diversify protein heterooligomers, we designed a dimer of CA (dCA) by placing a cysteine residue at the E187 position (E187C) while removing the pre-existing cysteine (C206S) (Fig. 4a).⁵⁹ The double mutant was obtained as a mixture of monomers and dimers (Fig. S9b and c^†), and treatment with 5 molar equiv. of cupric chloride^60,61 to the CA monomer, followed by size exclusion chromatography led to the isolation of dCA (Fig. S9^†). It showed the hydrodynamic diameter of 8.5 ± 0.52 nm (Fig. 4b), comparable with the calculated value from the manually docked dimer structure using the PyMOL program (7.5 nm).

Open in a separate window

Fig. 4

Redesign of dimeric CA (dCA) and its heterooligomer with BzS5-conjugated KDGA. (a) A model of dCA and KDGA*-E288C|BzS5 assembly. DLS data of (b) dCA (10 μM) and (c) dCA/KDGA assembly with KDGA*-E288C|BzS5 in an equimolar ratio (10 μM monomer) before and after mixing for 2–6 min. (d) Buffer-dependent dCA/KDGA assembly. (e) dCA/KDGA assembly in TAPS buffer (pH 8.0) upon the addition of citrate at various time points (final pH = 7.9). (f) Disassembly of dCA/KDGA assembly upon adding DTT (10 mM).

Next, we mixed dCA and KDGA*-E288C|BzS5 (dCA/KDGA) in TAPS pH 8.0 buffer at an equivalent monomer ratio, which immediately generated large species with a hydrodynamic diameter of 1200 ± 140 nm (Fig. 4c). Prolonged incubation ultimately led to the formation of white protein precipitates. The analogous dCA/KDGA assembly was observed in pH 7.4 buffer (Fig. 4d), but not in pH 6.2 citrate and Bis–Tris buffers; the size of the assembled products was reduced to 20 ± 0.64 nm, which roughly corresponds to the size of two dCA binding to one KDGA*-E288C|BzS5 tetramer. These data suggest that the pH-dependent BzS binding affinity to CA⁶² is operative as the primary force for PPI formation.

We further investigated the dCA/KDGA assembled products by measuring the hydrolytic activities with pNPA (Fig. S6 and S10^†); dCA alone showed comparable activity to that of the wild-type monomeric protein, and the addition of KDGA without the linker did not perturb the esterase activity of dCA. However, KDGA*-E288C|BzS5 reduced the activity of dCA, with its binding affinity calculated to be 290 ± 18 nM. The value was comparable to that of monomeric CA, supporting that PPIs between dCA and KDGA occur through a direct coordination of the BzS group to the zinc ion with similar degrees of thermodynamic driving forces to those with CA/KDGA.

Previous studies reported that a metal-chelator, reducing agent, or protease can trigger protein disassembly as external stimuli.^63–65 Similarly, we explored whether such changes could regulate the heteroassembly. The addition of 13 mM citrate (26% v/v) prior to or during dCA/KDGA assembly in TAPS buffer significantly perturbed the PPI formation (Fig. 4e). Because the net pH change was minimal (pH 7.9), these results were attributed to citrate acting as a competing metal-coordinating ligand.⁶⁶ Thus, citrate presumably functioned both as an acidic buffer component and a metal chelator, when we initiated the self-assembly in 50 mM citrate, pH 6.2, 100 mM Na₂SO₄ buffer (Fig. 4d). In contrast, no size reduction was observed, when citrate was added to the fully assembled product in TAPS buffer (Fig. 4e). It is possible that the heteroassembly might subsequently form nonspecific hierarchical structures and lower the accessibility of citrate into the active site pockets of CA.

In contrast, reducing agents, such as dithiothreitol (DTT), can reduce the size of the pre-mixed dCA and KDGA*-E288C|BzS5 in TAPS buffer (Fig. 4f). The fraction of large species grown for 4 min decreased significantly and dissociated into each component protein, suggesting that the disulfide bond in dCA or thiol–ene bioconjugation was reverted, possibly reflecting the reducing environments. This dynamic and responsive feature is also reminiscent of how chemical stimuli regulate natural macromolecular assembly and disassembly.

This work was supported by the Creative-pioneering re-searchers program from Seoul National University (SNU), National Research Foundation (NRF) from Korea government (NRF-2022R1A2C4001207). The authors are also grateful to Prof. Byeong-Hyeok Sohn and Saero Kim for the guidance in TEM experiments and Prof. Hyun-Woo Rhee and Chang Ryul Choi for HR-ESI-MS experiments.