Design of light- and chemically responsive protein assemblies through host-guest interactions

SUMMARY

Host-guest interactions have been widely used to build responsive materials and molecular machines owing to their inherently dynamic nature, interaction specificity, and responsiveness to diverse stimuli. Here we have set out to exploit these advantages of host-guest chemistry in the design of dynamic protein assemblies, using a C₄ symmetric protein, ^C98RhuA, as a building block. We show that ^C98RhuA variant individually modified with β-cyclodextrin (βCD) (host) or azobenzene (guest) functionalities can specifically pair with each other to form highly ordered 1- and 2-D assemblies. Association and dissociation ^βCDRhuA-^azoRhuA assemblies can be controlled by UV and visible light as well as by small-molecule modulators of βCD-azobenzene interactions. Kinetics analyses reveal that ^βCDRhuA-^azoRhuA nanotubes assemble without a nucleation barrier, a highly unusual occurrence for helical supramolecular systems. Taken together, our findings provide a compelling example for achieving complex structural and dynamic outcomes in protein assembly through simple chemical design.

Graphical Abstract

TOC blurb

Host-guest (HG) interactions have been integrated into protein assembly endowing it various properties barely seen in other protein assembly systems. Both light and chemical-responsiveness are achieved owing to the photoisomerization of the guest azobenzene moiety and the binding affinity differences between its two states with the complementary host β-cyclodextrin moiety. Moreover, the lability of the HG interactions enables the ligand exchange and rearrangement of the ligands, thus leading to the unusual nucleation-free growth of the nanotubular assembly.

INTRODUCTION

Essentially all cellular functions depend on the ability of proteins to alter their structures or assembly states in specific response to changes in the environment^1-3. Stimuli-dependent structural changes can not only activate biochemical processes within a protein (e.g., the chemically/physically-gated transport of ions across membranes^4-7 or allosterically regulated catalytic reactions^8,9), but also control the interactions of proteins with biological partners (e.g., signal transduction¹⁰ or gene regulation¹¹) or induce their self-assembly into extended structures (e.g., nucleotide-dependent polymerization of cytoskeletal proteins^12-14). All of these dynamic processes require proteins to access at least two or more structural states that are selectively stabilized or destabilized by external stimuli. Recent advances in deep- and machine-learning-based tools have greatly streamlined the de novo design of single-state (i.e., static) protein structures^15-17. By contrast, the design of responsive or dynamic protein assemblies is considerably more challenging due to the necessity of simultaneously considering multiple low-energy states^18-20, the transformations of which necessitate the formation and/or the reorganization of extensive networks of non-covalent interactions²¹.

As an alternative to designing responsive protein structures exclusively through non-covalent networks, the incorporation of metal coordination bonds^22-26 or reversible covalent linkages (e.g., disulfide bonds)^27-29 provides an expedient route. These bonds are stronger than non-covalent interactions and therefore require smaller design footprints to induce protein assembly or folding. Metal coordination and disulfide bonds are sensitive to different chemical inputs (e.g., presence or absence of metal ions, pH, redox agents, ligands/chelating agents) ^22-26,28, which renders the formation of protein structures formed through these bonds inherently stimuli-responsive. Furthermore, much like a fulcrum on a lever, the small footprints of these bonds can also enable the resulting protein assemblies to be flexible and accommodate different conformational states. A good case in point are two-dimensional (2D) crystalline lattices formed by _l-rhamnulose-1-phosphate aldolase (RhuA), a C₄ symmetric homotetrameric protein³⁰. A RhuA variant (^H63/H98RhuA) engineered with two histidine residues at its corner positions (residues 63 and 98) assembles into mm-scale, 2D crystalline arrays upon Zn^II or Ni^II binding in a reversible fashion (Figure 1A, top). Another variant (^C98RhuA) with only a single Cys residue in each corner (residue 98) forms similar, defect-free 2D lattices under mild oxidizing conditions that promote reversible disulfide formation near thermodynamic equilibrium. Strikingly, the single-point attachment and the flexibility of the disulfide bonds also enable the ^C98RhuA lattices to undergo coherent, large-scale opening/closing dynamics through the rotation of the individual protein building blocks (“protomers”) with respect to one another, yielding a material with the thermodynamically most negative Poisson’s ratio (−1) possible (Figure 1A, bottom). These coherent protein lattice dynamics involve a continuum of conformational states and can be mechanically or chemically actuated³¹. Importantly, such large-scale motions would be difficult–if not impossible–to obtain with larger interaction footprints between the RhuA protomers. The 2D ^C98RhuA lattices thus provide a compelling example for both stimuli-responsive self-assembly and stimuli-responsive structural dynamics.

Figure 1. Control of RhuA self-assembly with different chemical bonding strategies.

(A) Previously reported ^H63/H98RhuA and ^C98RhuA variants that self-assemble into 2D crystalline arrays via metal coordination or disulfide bonding, respectively. Surface representations for ^H63/H98RhuA and ^C98RhuA are shown on the left and cartoon illustrations metal- and disulfide-directed self-assembly are shown on the right.

(B) Current work involving ^C98RhuA variants (^azoRhuA and ^βCDRhuA) covalently modified with complementary azobenzene and β-cyclodextrin (βCD) functionalities for host-guest interactions. Top: Reaction schemes for the modification on ^C98RhuA with azobenzene and β-CD. Cysteine-reactive maleimido functional groups are drawn in red and the alkynyl and azido groups for SPAAC are drawn in green. Various linker lengths are highlighted. Bottom: Proposed scheme for the co-assembly of ^azoRhuA and ^βCDRhuA into 2D lattices that can be disassembled and reassembled through the UV- and visible-light induced photoisomerization of the azobenzene functionality.

In this work, we set out to build further on this example by implementing a different type of chemical interaction motif, namely non-covalent host-guest (HG) interactions^32,33, to construct dynamic protein assemblies on small design footprints. HG interactions have been foundational for the field of supramolecular chemistry and widely used to build responsive materials and molecular machines owing to their inherently dynamic nature, modularity and specificity^34-39. To exploit the advantages of HG chemistry for designing responsive protein materials, we chose ^C98RhuA again as our base building block. As outlined below, our goal was to establish general design strategies for HG-directed protein self-assembly, while also addressing questions that specifically pertain to the unique structural and dynamic properties of the 2D crystalline ^C98RhuA assemblies:

(1) Responsiveness to diverse stimuli:

The versatility and synthetically modular nature of HG interactions readily enable the design of supramolecular systems that are responsive to diverse chemical and physical inputs^39-41. Along these lines, we aimed to demonstrate that HG-interactions can be implemented to control protein self-assembly by light and chemical stimuli without requiring extensive (or any) reengineering of the protomers.

(2) Modulating structural/dynamic outcomes through interprotein connectivity:

A consideration of the 2D ^C98RhuA crystals suggests that not only the intrinsic flexibility but also the short lengths of the disulfide linkages may be crucial for the unbendable nature of the lattices and their coherent in-plane motions^30,31. How would the structure and the dynamics of these 2D crystals be affected with longer linkers? To answer this question–and given that there is no natural substitute for a cysteine residue with a longer thiol sidechain–we sought to replace the disulfide bonds bridging the 98 positions with longer linkers that mediate HG-interactions.

(3) Interaction specificity with minimal design:

While the small footprints of metal- and disulfide bonds offer several advantages for designing responsive protein assemblies, they cannot provide the extent of specificity of large, non-covalent protein-protein interfaces³⁰. For this reason, essentially all metal- and disulfide-directed protein assemblies constructed thus far have been homomeric (i.e., composed of multiple copies of chemically identical protomers)^22,25,30. Thus, we aimed to explore whether the specificity of HG interactions could mediate the pairing of chemically non-identical protomers, thereby allowing the formation of heteromeric assemblies.

RESULTS

Design and initial characterization of HG-mediated RhuA assemblies

There are several examples of self-assembling systems based on specific HG interactions between proteins and their natural ligands or cofactors^42-45. Following the pioneering work by Crabtree and Schreiber on chemically-induced protein dimerization⁴⁶, researchers have demonstrated that synthetic homo- or heterobifunctional linkers bearing ligand headgroups (e.g., heme, biotin, sugars) could induce the self-assembly of the corresponding host proteins into supramolecular or extended structures^42,45,47. In parallel, the ability of synthetic host molecules such as calixarenes to bind amino acid functionalities (e.g., lysines) on protein surfaces have been exploited for protein self-assembly and crystallization⁴⁸. Similarly, cucurbituril derivatives (as hosts) have been used as molecular glues to associate proteins with appropriate peptide/molecular tags (as guests) into dimeric complexes or extended arrays, but these assemblies were devoid of long-range order^49-53. Rather than relying on the specific ligand binding properties of a protein or using an external host molecule, we chose to generate two distinct variants of RhuA (^βCDRhuA and ^azoRhuA) which are individually functionalized with β–cyclodextrin (βCD) or a photoisomerizable azobenzene group to form complementary HG interactions^54-56. The advantage of such a construction is that it can in principle be applied to any protein or protein pair for self-assembly. ^azoRhuA was obtained in a one-step reaction by treating ^C98RhuA with the commercially available, thiol-reactive 4-phenylazomaleinanil group (Figures 1B, top and S1A). To synthesize ^βCDRhuA, we first modified ^C98RhuA with the bifunctional azadibenzocyclooctyne-maleimide, which then enabled the covalent attachment of an azide-functionalized βCD moiety through a strain-promoted azide-alkyne cycloaddition reaction (Figures 1B, top and S1B). As has been widely exploited in constructing light-responsive systems, βCD displays a moderately high affinity for the thermally stable E (or trans)-azobenzene but not for Z (or cis)-azobenzene, which can be transiently populated by the UV irradiation of the E isomer⁵⁶. Thus, we envisioned that ^βCDRhuA and ^azoRhuA would co-assemble under ambient conditions into 2D arrays through “corner-to-corner” HG interactions, which subsequently can be disengaged upon UV irradiation (Figure 1B, bottom).

Indeed, upon mixing equimolar ^βCDRhuA and ^azoRhuA at total concentrations above 50 mM at pH ≥7, we observed the protein solutions turning cloudy within minutes, consistent with the formation of large assemblies/aggregates (Figure S2). Dynamic light scattering (DLS) measurements confirmed the emergence of protein assemblies with hydrodynamic diameters (D_h) in the range of tens of nm’s within minutes, which evolved into larger assemblies with D_h > 100 nm over the course of several days (Figures 2A and S3). By contrast, isolated/unmixed solutions of ^βCDRhuA or ^azoRhuA remained clear under the same conditions, with no evidence of self-assembled structures (Figure S4). Similarly, mixing of ^βCDRhuA with free azobenzene molecules or mixing ^azoRhuA with free β-CD molecules did not produce any assembled structures (Figure S5) To estimate the extent of self-assembly, we monitored the light scattering of the ^βCDRhuA-^azoRhuA mixture solutions at a total protein concentration of 100 mM at the isosbestic point of E- and Z-azobenzene (425 nm), so as to reduce any optical interference due to the photoisomerization of ^azoRhuA. The yield of self-assembly clearly showed a chevron pattern with respect to the molar fraction of each species and reached a maximum at a 1:1 ^βCDRhuA:^azoRhuA ratio (Figures 2B and S6), evidencing the co-assembly of the two species. The heteromeric composition of the protein assemblies was further confirmed through mass spectrometric analyses of the aggregates, which showed the presence of both ^βCDRhuA and ^azoRhuA (Figure S7).

Figure 2. Solution and TEM characterization of ^βCDRhuA and ^azoRhuA co-assembly.

(A) DLS measurements showing the time-dependent evolution of ^βCDRhuA-^azoRhuA co-assemblies.

(B) Yield of ^βCDRhuA and ^azoRhuA co-assembly under different [^βCDRhuA]:[^azoRhuA ] ratios at constant total protein concentration, as monitored by turbidity measurements (at 425 nm) after 3 h of mixing.

(C) Negative-stain TEM images of ^βCDRhuA-^azoRhuA nanotubes at different magnifications. Power spectrum of one of the nanotubes is shown in the inset in the right panel.

(D) Distribution of ^βCDRhuA-^azoRhuA nanotubes with different diameters, determined from negative-stain TEM images.

Interestingly, negative-stain transmission electron microscopy (ns-TEM) images of the ^βCDRhuA-^azoRhuA assemblies revealed elongated, 1D structures instead of the expected 2D arrays (Figure 2C). These assemblies displayed variable lengths of up to 1 μm and a relatively narrow width distribution centered around 90 nm (Figure 2D), with dark edges that are suggestive of a tubular architecture. The power spectra of the ns-TEM images indicated the existence of two mirror-symmetric lattices (Figure 2C, right inset), which is a typical feature of helical structures²². Dry-state atomic force microscopy (AFM) measurements of the ^βCDRhuA-^azoRhuA assemblies revealed widths of >120 nm and heights of 13-16 nm (Figure S8A), consistent with near complete flattening of 1D nanotubes to give bilayers of RhuA (each RhuA protomer is 5-6 nm high). These findings suggest that the long linkers attaching the azobenzene and βCD groups to the ^C98RhuA surfaces must be flexible enough to accommodate the bending of the expected 2D arrays into 1D nanotubes.

Structural characterization of HG-mediated RhuA assemblies

To better understand the structural details of intact ^βCDRhuA-^azoRhuA nanotubes, we carried out cryogenic electron microscopy (cryo-EM) experiments (Figure 3A). Relative to the ns-TEM measurements, the nanotubes imaged by cryo-EM displayed widths with a maximum of 65-70 nm, indicating that their hollow cylindrical structures were largely preserved in vitrified ice (Figure 3B, top). We collected 13,999 movies and isolated 2.7 million single-tube fragments through filament tracing. After multiple rounds of 2D structural classification and discarding unsuitable/flattened particles, we selected three groups of nanotubes with small outer diameters (48, 52, and 58 nm) for helical reconstruction. Averaged power spectra calculated from these 2D structural classes were individually indexed, and the obtained values for helical rise and twist were used as initial inputs for reconstruction, yielding nanotubular structures with ca. 17, 19, and 21 RhuA tetramers per turn (Figure 3C). Subsequent 3D classification and CTF refinement resulted in maps with resolutions of 6.8 Å to 7.8 Å (Figure S9A). These maps were sufficiently detailed for an accurate placement of RhuA protomers and determination of important structural features. Similar to the disulfide-linked, crystalline 2D arrays of ^C98RhuA, the RhuA protomers in ^βCDRhuA-^azoRhuA nanotubes are found in an alternating packing arrangement (up-down-up-down) with respect to the tube surface (Figure 3D). AFM images collected in solutions of pre-assembled ^βCDRhuA-^azoRhuA nanotubes revealed elongated structures exhibiting checkerboard-like patterns of proteins consistent with the alternating up-down-up-down packing of ^azoRhuA and ^βCDRhuA units observed in cryo-EM data. The overall dimensions of the nanotubes also suggest that they open into flat ribbons upon adsorption to the AFM substrate (Figure S8B).

Figure 3. Structural analysis of ^βCDRhuA-^azoRhuA nanotubes.

(A) A representative cryo-EM image of co-assembled ^βCDRhuA-^azoRhuA nanotubes, showing variations in the tube diameters.

(B) Distribution of ^βCDRhuA-^azoRhuA nanotubes with different diameters, determined from cryoEM (top) and SAXS (bottom) measurements (see Methods for details).

(C) Left and middle panels: Cryo-EM helical reconstructions of co-assembled RhuA nanotubes of different diameters shown from the side and the top. Right panels: Helical nets of the corresponding nanotubes. The convention used is that the surface of the tube is unrolled and is being viewed from the outside surface of the nanotube. The positions of the dots correspond to the helical arrangement of the asymmetric units composed of a pair of two neighboring ^βCDRhuA and ^azoRhuA units. The three reconstructed tubes are generated from 10, 11, 12 protofilaments with a right-handed twist (dashed lines) in the absence of rotational point group symmetry (C₁ symmetry) or C₃ symmetry (the bottom one only).

(D) Structural model of tetrameric RhuA monomers fitted into the EM volume map of the 52-nm-diameter nanotubes, indicating that the neighboring RhuA units adopt an alternating arrangement. Position 98 of RhuA is highlighted in black.

(E) Density maps obtained from the single-particle reconstruction of a focused region on the nanotubes. Even without the applied helical symmetry, the density belonging to the host-guest pairs still maintains a pseudo-C₂ symmetry, indicating the formation of a “dimer-of-dimers” host-guest pair. The pairwise distances between C98 residues are indicated in the right panel.

Despite the apparently close packing of the RhuA protomers, there is minimal direct contact (≤100 Ų) between their surfaces. Indeed, our previous analyses of the 2D ^C98RhuA crystals indicated that such close packing is driven primarily by solvent entropy (i.e., the exclusion of ordered hydration layers)³¹, and that the alternating packing arrangement results from the long-range alignment of the large macrodipole moments (>1200 Debye) of individual RhuA protomers⁵⁷. The expected fluidity of the solvent-dependent and long-range dipolar interactions, coupled with the flexibility of the HG linkers, can clearly enable small changes in the relative orientations of neighboring RhuA protomers, ultimately giving rise to nanotubes with variable surface curvatures/diameters.

Interestingly, a close inspection of the cryoEM maps revealed densities that correspond to pairs of b-cyclodextrin-azobenzene HG complexes. These bilobal densities are centrally located in the void space surrounded by four RhuA protomers that are related to each other by local C₂ symmetry (Figure S9B). Each lobe of these densities is well accommodated by a single β-cyclodextrin-azobenzene HG complex, with the local C₂ symmetry generating a dimer of HG complexes. This geometry is highly reminiscent of a dimer-of-dimer arrangement previously observed in the crystal structure of the β-cyclodextrin-azobenzene complex⁵⁸, whereby the inter-HG interactions are stabilized by π-π stacking between the distal phenyl rings of the azobenzene groups and H-bonds between the β-cyclodextrin rings (Figure S9C). To better visualize the HG complexes, we focused on the 52-nm-diameter nanotubes and used small spherical segments (of 20 nm diameter) surrounding the HG complexes (which take into account the curvature of the nanotubes) as templates for single-particle EM reconstruction (Figure S10). This procedure improved the local resolution of the interprotomer region to 3.0 Å. At this resolution, the dimer-of-dimer arrangement of the HG complexes was clearly evident and the β-cyclodextrin rings and the azobenzene moieties could be discerned (Figure 3E, Supplemental Video S1). The densities corresponding to the linker regions between the C98 sidechains and the HG complexes were not sufficiently well-defined for accurate molecular modeling. Nevertheless, the specific locations of the azobenzene and βCD groups with respect to the four surrounding C98 residues and the different linker lengths of the azobenzene and β-cyclodextrin groups allowed us to determine the distinct arrangement of ^βCDRhuA and ^azoRhuA protomers in the nanotubes: the N-terminal face of the ^βCDRhuA facing the interior and the N-terminal face of ^azoRhuA facing the exterior. It is important to note that the dimerization interactions between the ^βCDRhuA-^azoRhuA HG complexes create connections between four RhuA protomers, which may provide additional stabilization of the nanotubes.

To further examine the self-assembly of ^βCDRhuA-^azoRhuA nanotubes in solution, we carried out small-angle X-ray scattering (SAXS) measurements. The SAXS patterns of ^βCDRhuA-^azoRhuA mixture solutions (at 100 mM total protein) obtained two days after initiation of self-assembly (i.e., under steady-state conditions) were consistent with a heterogenous mixture of nanotubes of different diameters and unassembled RhuA protomers. To fit the SAXS data, we first computed the simulated 1D SAXS profiles of RhuA protomers and various ^βCDRhuA-^azoRhuA nanotubes with distinct helical arrangements/diameters (8- to 16-start helices) using Model2SAS⁵⁹ (Figure S11). We then used a weighted fitting procedure using a combination of these simulated profiles, which yielded an excellent match to the experimental SAXS profile (Figure S12). The resulting population distribution of different components in solution reveal a maximum at ~65 nm, which closely matches the profile observed by cryo-EM (Figure 3B). The SAXS population distributions also indicate that unassembled RhuA protomers account for ~50% of the total species in solution, suggesting that the apparent dissociation constant for ^βCDRhuA-^azoRhuA nanotubes must be on the order of tens of μM, given that the concentrations of ^βCDRhuA and ^azoRhuA are 50 μM each (Figure S12C).

Responsiveness of HG-mediated RhuA self-assembly to multiple stimuli

Having structurally characterized the ^βCDRhuA-^azoRhuA nanotubes, we next investigated the control of both their assembly/disassembly and their structural reconfiguration in response to chemical and physical stimuli. We first tested the effects of chemical stimuli (i.e., pH and metal ions) that act upon the protein assembly as a whole rather than specifically on the HG interactions (Figure 4A). When we screened the co-assembly of ^βCDRhuA and ^azoRhuA at lower pH values (pH ≤6) than those that lead to nanotube formation, we observed that the solutions turned turbid within seconds. The ns-TEM images of the resulting suspensions revealed multi-layered, 2D crystalline arrays. The fast Fourier transformation (FFT) patterns obtained from these images clearly showed the overall 4-fold rotational symmetry of the arrays along with systematic absences along the 2₁ axis, consistent with p42₁2 plane group symmetry (Figure S13). This symmetry indicates an alternating arrangement of the RhuA molecules along the 2D plane with respect to the 4-fold symmetry axis (i.e., up-down-up-down) similar to that observed on the pseudo-2D surfaces of the nanotubular assemblies. RhuA has a low isoelectric point (pI ~ 5.8) and is negatively charged at neutral pH owing to an abundance of Glu and Asp residues on both the top and bottom facets of the protomer (Figure S14A). Thus, we posit that the overall neutralization of electrostatic charge on RhuA protomers at low pH (Figure S15) mitigates repulsion between HG-mediated 2D ^βCDRhuA-^azoRhuA assemblies, stabilizing them through stacking interactions in the c direction. When the pH of these suspensions was increased to ≥7.5 and incubated for 3 days, the 2D assemblies uniformly transformed into nanotubes; this process could be reversed (i.e., nanotubes → 2D arrays) upon lowering the pH back to ≤6.0 (Figures 4B and S16).

Figure 4. Stimuli-responsive behavior of ^βCDRhuA-^azoRhuA co-assemblies.

(A) Cartoon illustration describing the responsive behavior of ^βCDRhuA-^azoRhuA nanotubes.

(B) (left) Nanosheets formed after dialyzing pre-formed nanotubes in pH 6 buffer solution for 3 days. (right) Conversion of the nanosheets back to nanotubes after dialysis in pH 7.5 buffer solution for 3 days.

(C) (left) ns-TEM image of 2D nanosheets formed upon incubation of equimolar ^βCDRhuA and ^azoRhuA with 50 mM Ca²⁺ at pH 7.5. (right) Close-up view of a Ca-induced 2D ^βCDRhuA-^azoRhuA nanosheet and the corresponding fast Fourier transform (inset).

(D) Response of ^βCDRhuA-^azoRhuA nanotubes to UV- and visible-light irradiation, monitored by turbidity measurements at 425 nm and characterized by nsTEM (inset, scale bars: 200 nm). Disassembly is only observed with high-flux UV irradiation. The gradual increase in absorbance observed after ~500 s in all traces is due to slow solvent evaporation.

(E) (left) Cartoon illustration for the cyclic disassembly and reassembly of ^βCDRhuA-^azoRhuA nanotubes upon UV- and visible light irradiation and dark incubation. (right) Corresponding changes in turbidity over three cycles of UV- and visible light irradiation and dark incubation.

(F) Comparison of the dimensions of β-CD and α-CD.

(G) Increase in the rate of disassembly of ^βCDRhuA-^azoRhuA nanotubes with increasing α-CD concentrations.

(H) Decrease in the rate of the assembly of ^βCDRhuA-^azoRhuA nanotubes with increasing α-CD concentrations.

Inclusion of 0.05-0.1 M of divalent metal ions, Mg²⁺, Ca²⁺, and Ba²⁺, in solutions of ^βCDRhuA and ^azoRhuA also resulted in the formation of stacked 2D crystals, likely through a combination of direct metal-mediated interlayer interactions between the acidic residues and electrostatic charge screening (Figure 4C). Yet, in contrast to the effects of low pH, which transforms nanotubes into stacked 2D crystals, the addition of metal ions to the suspensions of pre-formed ^βCDRhuA-^azoRhuA nanotubes induced the aggregation of nanotubes (Figure S17). This observation suggests that direct metal-mediated interactions between RhuA nanotubes may be sufficiently strong to kinetically trap the aggregated state. Indeed, all-atom molecular dynamics simulations of RhuA molecules in the presence of 0.1 M Ca²⁺ indicated that there are several patches on the RhuA surface that display persistent interactions with Ca²⁺ ions (Figure S14B).

As previously mentioned, an advantage of HG interactions is the ability to control their formation through specific physical and chemical stimuli^54-56. Initially focusing on the light-responsive properties of the azobenzene-β-cyclodextrin HG pair, we first confirmed that the azobenzene moiety attached to ^azoRhuA undergoes the expected photoisomerization behavior: exposure to UV light (λ = 365 nm) was accompanied by a decrease in the intensity of the π-π* band (λ _max = 355 nm) and an increase in the n-π* band (λ _max = 440 nm) consistent with E-to-Z photoisomerization⁶⁰, with the complete conversion taking place over 5 min. The reverse (Z-to-E) process could be induced nearly fully by irradiation with visible light (λ = 470 nm) for 10 min (Figure S18), while the thermal relaxation of the Z isomer was considerably slower and not complete even after one day in the dark⁶⁰. Next, we examined the light-responsive behavior of co-assembled ^βCDRhuA-^azoRhuA nanotubes. Exposure to visible or low-intensity UV irradiation had negligible effect on the nanotube structures (Figures S19 and S20). However, a 60-fold increase in the UV light intensity resulted in rapid (τ_1/2 ~ 4 min) disassembly of the ^βCDRhuA-^azoRhuA nanotubes as monitored by light scattering (Figures 4D and S19). The disassembly of the nanotubes was confirmed by ns-TEM images of the irradiated samples, which were largely devoid of self-assembled RhuA structures except small 2D fragments (Figures 4D, inset and S20). The irradiation of these samples with visible light led to the recovery of the ^βCDRhuA-^azoRhuA nanotubes over the course of 30 min followed by dark incubation for 11 h. The reversible UV- and visible-light mediated assembly/disassembly cycle of the nanotubes could be repeated over at least three cycles (Figures 4E, S21 and S22).

Because HG interactions are inherently responsive to chemical agents that can compete for these interactions⁶¹, we surmised that the assembly and disassembly of ^βCDRhuA-^azoRhuA nanotubes could also be controlled specifically by such competing ligands. As a test case for a competing host molecule, we chose α-cyclodextrin (α-CD), which binds E-azobenzene with two-fold higher affinity than βCD (Figure 4F)^56,62. As anticipated, the addition of increasing amounts of α-CD to a suspension of pre-assembled ^βCDRhuA-^azoRhuA nanotubes led to increasingly faster disassembly kinetics (>3-fold faster at 20 molar equivalents of α-cyclodextrin), eventually resulting in fragmented nanotubes and individual protomers (Figure 4G). Conversely, the inclusion of α-cyclodextrin in solutions of ^βCDRhuA and ^azoRhuA prior to self-assembly slowed down the HG-mediated nanotube formation by >4-fold at 5 molar equivalents of α-cyclodextrin (Figure 4H). Aside from α-cyclodextrin, any molecule that can form complexes with either azobenzene or β-cyclodextrin can in principle be used as a modulator of ^βCDRhuA-^azoRhuA assembly/disassembly. Indeed, the addition of increasing concentrations (1-20 equiv.) of free β-CD, azobenzene or adamantane suspensions of ^βCDRhuA-^azoRhuA nanotubes all led to greater extents of nanotube disassembly as determined by ns-TEM measurements (Figures S23-S25).

Mechanism of the self-assembly of RhuA nanotubes

Interestingly, even when the rate of formation of ^βCDRhuA-^azoRhuA nanotubes was considerably slowed down by the addition of competing ligands or by lowering the overall protein concentration, we did not observe a lag phase in the assembly kinetics (Figure 5A). Helical assemblies such as cytoskeletal filaments (and helical polymers in general) overwhelmingly display cooperative polymerization behavior with a characteristic lag phase arising from the thermodynamically unfavorable formation of a critical nucleus⁶³. Instead, the kinetics of ^βCDRhuA-^azoRhuA nanotube assembly appeared to be more consistent with an isodesmic process, which is typical of the formation of 1D polymer chains^64,65. Indeed, the kinetics of ^βCDRhuA-^azoRhuA nanotube formation obtained at different protein concentrations can be reasonably well described by an isodesmic polymerization model (Scheme S1). In this model, all ^βCDRhuA-^azoRhuA association/ dissociation steps occur with a single set of rate constants: k_assoc. = 36 M⁻¹s⁻¹ and k_dissoc. = 2.3 x 10⁻⁴ s⁻¹ (Figure 5B).

Figure 5. Formation kinetics and structural evolution of ^βCDRhuA-^azoRhuA nanotubes.

(A) Kinetics of formation of ^βCDRhuA-^azoRhuA nanotubes at different total protein concentrations, monitored by changes in turbidity at 425 nm, show no evidence of a lag phase.

(B) Global fits to the formation kinetics of ^βCDRhuA-^azoRhuA nanotubes by an isodesmic polymerization model described in Methods.

(C) Proposed mechanism for the isodesmic, row-by-row growth model of ^βCDRhuA-^azoRhuA assemblies.

(D) Cryo-EM images showing the early stages of ^βCDRhuA-^azoRhuA co-assembly. The samples contained 50 μM each of ^βCDRhuA and ^azoRhuA, and were rapidly diluted 10-fold prior to deposition onto the cryoEM grids. At 15 s after initiation of co-assembly, some 1D structures are evident, followed by the formation of multi-row, 2D assemblies at 150 s and the formation of short, tubular structures at 300 s.

Besides a lag phase, an important characteristic of nucleated/cooperative self-assembly processes is the presence of a critical concentration (Cc), which is defined as the minimal protomer concentration below which no self-assembly/polymerization occurs and above which all protomers (aside from the Cc amount) are incorporated into the assembly/polymer. By contrast, in an isodesmic process, there is no Cc: self-assembly/polymerization occurs even at low concentrations, and the concentration of both the polymers and free protomers continuously increase with increasing total protomer concentration⁶⁵. To investigate if a Cc of protomers exists in ^βCDRhuA-^azoRhuA nanotube assembly, we carried out sedimentation analyses. In these experiments, solutions with different total protomer concentrations (10-125 μM) were centrifuged at 21000 x g (after reaching steady-state), and the amount of protomers in the pellets and the supernatants were quantified. It is important to note that at this centrifugation velocity, we estimate only oligomers/polymers larger than n~7 to sediment efficiently, whereas any smaller species (n= 1-6) should remain in the supernatant. As shown in Figure S26, we observed a continuous increase in the amounts of both the large oligomers/polymers and free protomers/small oligomers, with no plateauing in the concentration of the latter at higher total protomer concentrations, which is consistent with an isodesmic polymerization mechanism.

Given that ^βCDRhuA-^azoRhuA nanotubes are not 1D chains of protomers but can rather be considered as curved 2D crystalline arrays, the lack of a lag phase and a Cc in their assembly kinetics is surprising. For such 2D arrays, classical nucleation theory (CNT) predicts a free energy barrier that stems from the different dimensional dependencies of the decrease in chemical potential (which scales with surface area) and the increase in interfacial/line tension (which scales with surface perimeter) during crystallization/self-assembly. In contrast, for 1D chains, no free energy barrier is predicted, as both chemical potential and interfacial tension terms are proportional to the length of the assemblies^66,67. Thus, the question arises as to how ^βCDRhuA-^azoRhuA nanotubes, which are 2D arrays, display barrierless self-assembly kinetics like the isodesmic polymerization of 1D chains. Insights into this question are provided by two prior studies. In one study, De Yoreo and colleagues reported that heptapeptides specifically selected for MoS₂ binding can form 2D arrays on crystalline MoS₂ substrates without a nucleation barrier⁶⁸. A detailed in situ atomic force microscopy (AFM) analysis of the nucleation/growth process revealed that the self-assembly kinetics of the heptapeptides were dominated by strong longitudinal peptide-peptide interactions which led to their growth one row at a time without the formation of a critical nucleus. In contrast, the growth of additional rows to form the final 2D crystalline arrays were mediated by weaker, lateral peptide-peptide interactions which do not incur a nucleation barrier. In a second study, Sibley and colleagues examined the self-assembly of actin filaments from Toxoplasma gondii and observed that these assemblies uncharacteristically also displayed barrierless growth kinetics⁶⁹. In analogy to the aforementioned heptapeptide arrays, this unusual isodesmic polymerization behavior of T. gondii actin filaments was attributed to strong longitudinal protein-protein interactions to form the 1D protofilaments, but weak lateral interactions to yield the final bi-filamentous assembly. In turn, it was suggested that isodesmic polymerization was important for the assembly short filaments that power parasite motility rather than the typical long, stable actin filaments of yeast and mammalian cells formed through a nucleation-condensation mechanism.

A consideration of the HG bonding network within the ^βCDRhuA-^azoRhuA nanotubes provides a rationale for how they may also follow a similar row-by-row self-assembly mechanism to avoid a nucleation barrier⁷⁰. Each ^βCDRhuA or ^azoRhuA unit is equipped in its four corners with four β-cyclodextrin or azobenzene groups attached to the protein backbone with >10-Å linkers. These linkers are sufficiently long that they would allow the edge-to-edge association of the complementary ^βCDRhuA and ^azoRhuA protomers through two pairs of HG interactions (i.e., a bivalent association), which would provide a strong driving force to form 1D rows. (Note that such bivalent, edge-to-edge association would not be possible with the disulfide bonds between the 98 positions, which can only allow monovalent, corner-to-corner attachment of ^C98RhuA protomers due to their short length.) At the same time, the linkers are also flexible enough and the HG interactions are sufficiently labile such that they would enable the lateral attachment of protomers to the 1D rows through a HG swapping mechanism as illustrated in Figure 5C. Because the lateral attachment process incurs no net change in the number of total HG interactions (one longitudinal interaction is replaced with an equivalent lateral one), this mechanism would allow the initiation of additional rows with no net change in bond/interaction energy, thereby avoiding the emergence of a nucleation barrier. The addition of each protomer in the newly started rows would occur through formation of two HG interactions as in the case of the first row followed by HG swapping, such that each additional row also grows isodesmically, thus leading to a barrier-free formation of 2D arrays. Strong support for the feasibility of an HG bond swapping mechanism is provided by the dimer-of-dimer arrangement of the HG complexes between four neighboring protomers, which clearly shows that their corner positions are in close enough proximity to exchange HG partners.

In accordance with the row-by-row growth model, cryoEM snapshots taken immediately (~15 s) after the mixing of ^βCDRhuA and ^azoRhuA show the formation of short 1D chains. Within 2-3 min, we observe the emergence of 2D arrays rather than long, isolated 1D chains. This observation, similar to that made for the aforementioned heptapeptides on MoS₂ surfaces, suggests that the lateral attachment of RhuA protomers to 1D chains must be highly favorable. This can be understood by an increase in the likelihood for lateral attachment in a growing linear chain relative to attachment to the two termini (Figure 5D, left). By 5 min after mixing, the formation of the nanotubular architectures is evident, suggesting that the folding of the 2D arrays into nanotubes occurs once a critical size limit is reached (Figure 5D, middle, right). Indeed, rapid SAXS scans of ^βCDRhuA and ^azoRhuA mixtures reveal that a cylindrical form factor is attained within the ~3 min deadtime of the measurements and remains unchanged thereafter, which is consistent with a rapid 1D chain-to-2D array-to nanotube evolution and the continued elongation of the nanotubes by protomer attachment (Figure S27).

DISCUSSION

Owing to recent advances in deep learning, it has now become possible to design new protein folds and architectures of arbitrary complexity with relative ease. There also is an increasing number of dynamic proteins that can alter their structures or self-assembly states in response to environmental cues^71,72. Despite this progress, designing proteins that can simultaneously respond to different types of chemical and physical stimuli from first principles still remains quite challenging, but constitutes an important step toward engineering artificial proteins with controllable functions in complex environments (e.g., artificial cells) and adaptive biomaterials with properties not achievable with natural proteins. In our ongoing quest to address this challenge by rationally designing chemical interactions (e.g., metal coordination or reversible covalent bonds), we have now implemented synthetic HG chemistry to direct protein self-assembly. This strategy is distinct from earlier efforts which have used HG interactions between proteins and their natural ligands/cofactors^42-45 or free cucurbituril units and tagged proteins^49-53, and offers several advantages: (1) it requires minimal interaction/design footprints on the proteins of interest, (2) it is inherently modular, in that the host or the guest functionality can be synthetically altered–without having to redesign the proteins–to modulate the thermodynamics and the dynamics/flexibility of protein-protein interactions, (3) it allows protein-protein interactions to be controlled by multiple stimuli based on the choice of host-guest pairs, (4) it may potentially be applied to any protein pair of choice to form heteromeric complexes, although this was only demonstrated here for chemically distinct variants of the same protein. In our case, we exploited these advantages by using the canonical azobenzene-βCD HG pair to construct heteromeric RhuA assemblies. The reversibility of the azobenzene-βCD interactions allowed the formation of well-ordered architectures that were responsive to both physical (i.e., light) and chemical (i.e., external ligands) stimuli, while the flexibility of the azobenzene and βCD linkers led to the interchangeable formation of 1D nanotubes and 2D crystalline sheets. Though unplanned, the 1D ^βCDRhuA-^azoRhuA nanotubes displayed unusual self-assembly kinetics that were free of a nucleation barrier, enabled again by the reversibility and flexibility of HG interactions to mediate a bond swapping mechanism. Some of our future goals include examining the effects of linker lengths and alternative HG motifs on the structure, dynamics and formation kinetics of RhuA assemblies as well as testing the generalizability of our HG-based strategy by exploring alternative protein building blocks.

EXPERIMENTAL PROCEDURES

Full experimental procedures can be found in Supplemental Information.

RESOURCE AVAILABILITY

Lead contact

Further information and requests for resources should be directed to and will be fulfilled by the lead contact, F. Akif Tezcan (tezcan@ucsd.edu)

Materials availability

All materials generated in this study are available from the lead contact without restriction.

Data and code availability

The principal data supporting the findings of this work are available within the figures and the Supplementary Information. Additional data that support the findings of this study are available from the corresponding authors on request. Structural data obtained from cryo-EM have been deposited in Electron Microscopy Data Bank (EMDB) and Protein Data Bank (PDB) with accession code: EMD-46767 (10-start nanotubes), EMD-46768 (11-start nanotubes), EMD-46769 (12-start nanotubes), 9DGH/EMD-46826 (single particle reconstruction of focused view on nanotubes).

Supplementary Material

2

Supplemental Video S1. Detailed view of the dimer-of-dimer arrangement of the host-guest complexes within the ^βCDRhuA-^azoRhuA nanotubes

THE BIGGER PICTURE.

Most cellular functions depend on the ability of proteins to change their structures in response to environmental stimuli. Yet, despite great advances in the design of static protein structures, it remains challenging to construct artificial protein assemblies that adopt different conformations in response to external cues. This challenge arises from minute energetic differences between these conformational states each of which is stabilized by a large ensemble of interconvertible non-covalent interactions encoded within a single amino-acid sequence. To circumvent the difficulty in programming such multi-state non-covalent networks, we have implemented supramolecular host-guest interaction motifs that provide stability and specificity without requiring large design footprints, while also rendering the resulting protein assemblies responsive to diverse physical and chemical stimuli. Coupled with the synthetic modularity of host-guest complexes that provide access to diverse interaction modalities, this approach opens the path to constructing artificial protein assemblies with new dynamic and functional properties with minimal design burden.

Highlights.

• Design of protein assembly driven by host-guest interactions

• Helical protein assemblies that are responsive to multiple external stimuli

• Unusual barrierless growth of the tubular protein assembly