Protein assembly: Controllable design strategies and applications in biology

Protein assembly is the structural basis for the collaboration between proteins to accomplish life activities due to its realization of the domain‐limited and precise spatial arrangement of proteins. Therefore, artificial manipulation of protein self‐assembly has profound implications in areas such as exploring the mysteries of life and developing biomaterials. In this review, we not only summarize the classical assembly strategies and the structures of exquisite protein assemblies, but also aim to generalize the flexible and controllable assembly tools and the “interaction” between the assembled protein‐based materials and the external environment. On the basis of this, the application, challenges and further development of protein assembly in biology are reviewed.


INTRODUCTION
Proteins, because of their involvement in life activities such as structural support, life metabolism, chemical regulation and signal transduction of living organisms, are considered as one of the most important "building block of life". [1][2][3][4][5] In the case of proteins themselves, the amino acid sequences of their primary structures are encoded and translated from the genetic information of the organism and folded by interactions between peptide chains to obtain a superior topology. [6][7] Moreover, most natural proteins exist and exercise their functions as symmetric oligomers formed by interactions between monomers, probably because the formation of symmetric interfaces reduces their energy. This symmetry also makes proteins much more designable and editable. [8][9] S C H E M E 1 (A) Multiple construction strategies for protein assembly and their diverse applications. (B) Schematic diagram of the relationship between three driving forces for the flexible construction of protein assemblies. For a more visual illustration, we present the inclusion or juxtaposition relationships of several driving forces in pie charts with numbered regions. The forces corresponding to the numbered regions can be found in the text description below the diagram.
precise self-assembly. Hierarchical single-or multicomponent assemblies can be easily obtained through bottom-up protein assembly strategies. [17][18][19][20] Many classical protein-assembly works with ingenious constructing strategies have been designed and carried out to further explore the mysteries of nature and create proteinbased practical and advanced biomaterials (Scheme 1A). Both flexible supramolecular interactions and powerful covalent modifications provide chemical tools for hierarchical protein assembly. [21][22][23] In addition, bioengineering interfaces of proteins to modify the protein-protein interactions (PPIs) is also regarded as a proven approach to drive the assembly. [24][25] In fact, there is no absolute boundary between the scope of PPIs and the scope of the two previous construction strategies; for example, hydrogen bonds or hydrophobic interactions formed between protein structural domains belong to supramolecular interactions, whereas direct point mutation to design protein crosslinks belongs to the covalent interaction-driven assembly. Here, we give a relatively intuitive classification schematic to explain the different classical construction strategies (Scheme 1B). Significantly, almost all construction strategies are influenced by their properties (orientation, complexation constants, charge number, stimulus responsiveness, etc.) and solution environment (temperature, pH, ion concentration, etc.), which makes the assembly process more challenging but much more regulable. [26] Thus, in addition to multi-dimensional nanoscale structures, protein assemblies are expected to be dynamically tunable and responsive to the environment, which are the advancements of intelligent biomaterials. [27] Furthermore, the biocompatibility and self-functionality of proteins will be greatly preserved in their assemblies, which provides excellent support for their application in the biological field. [16] As protein-assembly-based biomaterials are gradually employed in a wide range of applications such as biocatalysis, drug delivery, and light-harvesting system bionics, [28][29][30][31] protein assembly is no longer just regarded as an approach to building high-ordered nanostructures. [32] In addition to the pursuit of precise assembly and the construction of delicate structures and morphologies, the regulation of the assembly processes and the application of the structures are gradually occupying the focus of the study on protein assembly. In this review, we refer to Classical Construction Strategies as those assembly works dedicated to the fabrication of subtle protein morphologies and structures and their corresponding driving forces, while Flexible (controllable or regulable) Construction Strategies refers to the related works and means for designing structural regulation of protein assembly that are further derived from the above-mentioned works. We believe that this change from a focus on "results" to a focus on "processes" reflects the change from "static" to "dynamic" protein assembly. This shift in focus brings new prospects for the development of protein assembly. Herein, classical protein assembly strategies and corresponding examples of regulatory tools are summarized and classified; advanced design and outstanding work in recent years based on protein assembly and some of the applications in biology are also illustrated.

VARIOUS STRUCTURES OF PROTEIN ASSEMBLIES BASED ON SYMMETRY RULES
The spatial symmetry present in natural proteins plays a dominant role in the direction of their growth in different dimensions. Based on classical protein assembly construction strategies, a variety of delicate and detailed nanostructures have been successfully obtained, such as one-dimensional (1D) linear structures, [33][34][35] two-dimensional (2D) lamellar structures and vesica, [36][37] and three-dimensional cage structures and crystals. [33,38] The n-fold symmetry axes (n = 2, 3, 4, 6) and symmetry planes of natural protein building blocks allow for spatially oriented cycling and ordered assembly of proteins by chemical, physical and biological tools, which provides a necessary structural basis for the construction of protein nanomaterials. [39][40] 1D protein nanostructures are usually obtained from centrosymmetric protein dimers or binary domains grown along a single orientation, which generally requires that each protein monomer that makes up the dimer interacts separately with a monomer from another module. [41] Straight 1D protein nanowires can be successfully gained when there is no F I G U R E 1 Illustrations of the various subtle nanostructures of protein assemblies. (A) 1D nanowires consist of SP1 variants. Reproduced with permission: Copyright 2017, American Chemical Society. [42] (B) Nanorings obtained by the growth of proteins with a curvature. Reproduced with permission: Copyright 2013, American Chemical Society. [43] (C) Nanosheets formed by the planar assembling of proteins. Reproduced with permission: Copyright 2019, Springer Nature. [44] (D) 2D protein vesicles. Reproduced with permission: Copyright 2021, Wiley-VCH. [45] (E) Complex and ordered protein nanocages. Reproduced with permission: Copyright 2016, Wiley-VCH. [50] (F) 3D defined nanostructures of protein crystals. Reproduced with permission: Copyright 2014, Springer Nature [54] deviation in the growth direction ( Figure 1A). [42] At the same time, when there is a certain angle of departure in the raw growth direction, the protein assemblies exhibit curvature and eventually assemble into nanoring structures ( Figure 1B). [43] The formation of 2D lamellar protein assemblies requires the presence of more than two symmetry axes of the building blocks to satisfy the in-plane growth direction in both dimensions. Usually, the plane in which the symmetry axes are located is the plane in which the assembly is formed ( Figure 1C). [44] In contrast, when the protein nanosheet is curved, or its growth direction is angled, protein vesicles are expected to be fabricated ( Figure 1D). [45] The formation of protein cages requires that assembly driving forces act along any two intersecting symmetry axes of the polyhedral protein domain ( Figure 1E). [46][47][48][49][50] The formation of protein crystals, on the other hand, is more complex and is often the result of simultaneous thermodynamic and kinetic equilibria reaching equilibrium ( Figure 1F). [51][52][53][54] The multifaceted structure of protein assemblies can be regarded as the external basis for their bionic functions, while the flexible assembly interactions and their stimulus responsiveness are the internal basis. The structural information of most natural proteins can be found in Protein Data Bank (PDB), which also provides great support for the rational design of protein assembly.

CLASSIC CONSTRUCTION STRATEGIES OF PROTEIN ASSEMBLY AND THEIR REGULATION
Many of the activities of life in nature are inextricably linked to the stimulus-response behavior of their protein assemblies in their physiological structures to the external environment. The responsiveness of protein assemblies to external stimuli is not only involved in cellular-level activities such as transcription, translation, and apoptosis, but also the circulation, metabolism, and regulation of biomolecules in living organisms. [55][56] Inspired by nature, regulable protein assembly has been progressively applied to develop intelligent bio-nanomaterials. [57][58][59] Formed protein assemblies can be endowed with novel structures or additional functions by responding to the unfamiliar stimulus under the artificial tune of solution environment. The adjustability makes the proteinassembly process more programmable and more predictable, which exhibits significant consequences for the development of new-generation biomaterials.
The discussion of typical examples of construction strategies and their regulation has profound implications for the study of protein assemblies and their stimulus responsiveness. Despite the intersection with covalent/non-covalent interactions, PPIs are usually dominated by biological design strategies. The introduction of PPIs can usually be achieved through direct design and modification of proteins at the gene level. In contrast, the supramolecular or covalent construction strategies described in this paper focus on those protein assemblies that are driven by interaction forces between chemically modified non-protein components. In this section, the classical protein assembly works and their derived regulatory strategies will be exemplified and illustrated.

Devisable protein-protein interactions
PPIs, which are elaborated by biotechnologies based on the symmetric structure of proteins, generally refer to direct protein-protein interactions without the involvement of linkers or cofactors. PPIs can be cross-linkages between amino acid sites or non-covalent interactions between protein interfaces (e.g., hydrogen bonds, hydrophobic interactions, etc.). In addition, specific recognition interactions (or site-specific protein reactive pairs) between protein or short peptide tags can be considered a PPI in a broad sense. This bioconstruction strategy by direct protein design does not require the introduction of foreign chemical modifications. Still, it requires more stringent requirements for the structure and symmetry of the protein building blocks. [59][60] Functional side chains of amino acids offer the possibility of point cross-linking between proteins. The observation of the protein surface and genetic engineering techniques allow point mutation of natural proteins, where the original amino acid at the selected site is mutated to the target amino acid to take advantage of its side chain covalent cross-linking. Cysteine is the most commonly targeted amino acid due to its extremely low content in natural proteins. The cysteine introduced on the protein surface can successfully drive covalent assembly between proteins by forming disulfide bonds under oxidizing conditions. [61] The modification and manipulation of interprotein interfaces are much more complex and challenging. A de novo interface design of rigid domains between proteins provides a broader opportunity. [62][63] Recently, Baker's group reported a programmable two-component 2D material that can be co-assembled on the cell surface. [64] The D 3 (A) and D 2 (B) building blocks were selected to construct the binary p6m lattice structure by screening the symmetry of the protein to avoid out-of-plane curvature. Low-energy interfaces were obtained by designing the amino acid sequences between the interfaces of the two building blocks to drive the self-assembly ( Figure 2A). By labeling with the fluorescent protein (GFPs), this protein lattice has been shown to offer a high density of binding sites for intracellular transmembrane receptors, thereby triggering their significant cluster. Artificial regulation of PPIs to obtain protein assembly structural transformations provides a new approach for building diversified protein nanostructures. By redesigning protein interfaces, Zhao's group successfully realized the conversion of natural hollow protein nanocage assemblies into nanorod or ribbonlike structures ( Figure 2B). [65] Thermotoga maritima ferritin (TmFtn) would assemble into 24-poly protein nanocages in the presence of calcium ions due to its natural dimeric structure head-to-side interaction manner. However, upon Rosetta docking and designing, the building blocks are arranged in a fully or partially side-by-side manner, generating new PPIs between adjacent dimeric TmFtn. As a result, building blocks can be transformed from cage-shaped assemblies into 1D nanorods or 2D nanoribbons in the presence of calcium ions or PEG. The design strategy for protein interface reconfiguration provides different ideas for hierarchical controlled protein self-assembly.
Fusion proteins are also considered to be an important and effective tool for designing PPIs. By fusing specific protein domains or peptide tags into building blocks, the protein is expected to generate new interactions based on the original PPIs. SpyCatcher and SpyTag are well known as sitespecific protein reactive pairs. [66] Fusing these two fragments at the termini of multiple different functional building blocks (ELPs, RGD cell-binding ligands, globular protein, etc.), respectively, Spy-X networks with genetically programmable viscoelasticity were successfully generated ( Figure 2C). [67] Altered chromophore conformation of Dronpa145N protein under ultraviolet light irradiation promotes the formation of multiple hydrogen bonds between protein monomers, resulting in their tetramerization, while disassembly occurs upon visible light irradiation. [68][69] Inspired by this, Li's group cleverly constructed a switchable hydrogel network consisting of backbones formed by small globular protein (GB1)-Spycatcher tandem structures and anchored Dronpa145N fused with SpyTag. [70] The protein hydrogel could switch between solution-hydrogel states according to the conformational change of Dronpa145N, and the fluorescence of the system could be turned off and on ( Figure 2D). Rather unfortunately, this state switching can only be maintained for 1.5 cycles, which may be due to the weakening of the interactions between the structural domains of Dronpa145N by the fusion protein.
As hydrophilic blocks, dimerized coiled coils can also be fused to the N-or C-terminal of the protein domains for the thermally triggered assembly. By biologically designing two precisely designed antiparallel coiled-coil sequences (CCE and CCK) attached to V-shaped Smac proteins, nanofibers with silver ion adsorption were successfully fabricated by Liu's group ( Figure 2E). [71] Typical coiled-coil structures, such as leucine zipper, arginine-rich motif (Z R ), and glutamic acid-rich motif (Z E ), have also been utilized for creating protein nanostructures. Interestingly, when fusing different building blocks, the multiple ways protein assemblies are formed depend on the forces between protein domains. [72] More recently, a tunable protein vesicle based on the interaction between fusion proteins m-Cherry-Z E and ELP-Z R have been reported by Champion and his co-workers. [73] Due to the lower critical solution temperature behavior of ELPs, [74] there is a thermal driving force (∆T) between operation temperature and transition temperature as their difference, which is determined by the protein concentration. With the increase of ∆T, the average size of protein vesicles increased significantly, and the vesicle membrane structure changed from monolayer to bilayer ( Figure 2F), thus enabling artificial control of the size of protein vesicles.
PPIs based on genetic engineering design enable direct editing of protein constructs fast and are considered to be a dynamic constructing strategy. The de novo design opens an avenue for the development of diverse protein nanomaterials. However, to control the directionality of the assembly process, the additional PPIs as driving forces between protein domains expect a higher symmetry to the original structure of the protein. Moreover, the interface design strategy with modifications of the protein at the genetic level adds editability and designability to the protein assemblies and increasing the complexity of their upstream experiments at the same time.

Flexible supramolecular interactions
In recent years, with the rapid development of supramolecular protein self-assembly technology, various non-covalent interactions have been used to drive the formation of F I G U R E 2 Protein nanostructures based on PPIs with the illustrations of related regulation. (A) 2D protein structures obtained by designing the interprotein interfaces of D 3 and D 2 symmetric building blocks. Reproduced with permission: Copyright 2021, Springer Nature. [64] (B) Reconstructed protein assembly manners realized by the regulation of the amino acid sequences. Reproduced with permission: Copyright 2021, Springer Nature. [65] (C) Functional protein networks constructed via spy-pairs. Reproduced with permission: Copyright 2018, American Chemical Society. [67] (D) Switchable protein hydrogels tuned by illumination. Reproduced with permission: Copyright 2017, Royal Society of Chemistry. [70] (E) Nanofibers formed via the hydrophilic interactions between coiled-coil domains. Reproduced with permission: Copyright 2018, Wiley-VCH. [71] (F) Protein concentration and temperature dependent vesicles. Reproduced with permission: Copyright 2017, Wiley-VCH [73] ordered protein nanostructures. [16] As an essential interdisciplinary field that integrates chemistry and biology, flexible supramolecular construction strategies provide energetic tools for protein assembly. [16] The adjustability of the supramolecular interactions tuned by the solution environment also offers excellent possibilities for the transformation of the delicate microstructure of protein assemblies.

Metal coordination interaction-based protein assembly
Metal ions (M n+ ) and their cofactors, as "assistants" of functional proteins, play critical roles in the life activities of living organisms. [75] The interaction between M n+ and proteins is mainly achieved by the coordination between ions and amino acids with high electron cloud density (histidine (His),

F I G U R E 3 Protein nanostructures based on supramolecular interactions with the illustrations of related regulation. (A) Micron protein sheets formed via
Cu 2+ -histidine coordination. Reproduced with permission: Copyright 2019, Wiley-VCH. [20] (B) Covalence-regulated redox stimulus-responsive protein hydrogels. Reproduced with permission: Copyright 2019, American Chemical Society. [77] (C) Protein-inorganic nanoparticles assembly mediated by electrostatic interactions. Reproduced with permission: Copyright 2014, American Chemical Society. [81] (D) Binary protein-nanoparticle crystal formed by encapsulation and co-assembly with modified proteins. Reproduced with permission: Copyright 2016, American Chemical Society. [84] (E) Supramolecular nanofibers driven by hydrogen bond between FUS LC domains. Reproduced with permission: Copyright 2017, American Chemical Society. [92] (F) Biofilms formed based on hydrogen bond interactions that can be used as adhesives. Reproduced with permission: Copyright 2018, Royal Society of Chemistry [94] cysteine (Cys), glutamic acid (Glu), and aspartic acid (Asp), etc.). In particular, due to the defined ratio and the directional coordination between histidine and M n+ , His6-tag can be introduced to the surface of proteins in a targeted manner to drive protein self-assembly. Additionally, the absence of Cys in most of the natural proteins makes it possible for protein assembly based on Cys-M n+ coordination.
By accurately designing the surface of the natural dimeric protein, glutathione S-transferase (GST), with V-shaped angular His-metal chelating sites, Liu and his co-workers successfully employed directional coordination to construct protein nanoring structures (diameter ∼370 nm) grown in a fixed bending manner. [43] Mimicking the delicate metal-organic frameworks (MOFs) structure, Tezcan's group introduced three histidine sites on the surface of the highly symmetric globular ferritin, making it a tripodal coordination motif for zinc ion (Zn 2+ ) with tetrahedral geometry. Through the precise metal-ligand interaction, a porous 3Dcrystalline protein framework was obtained. [76] Employing Cu 2+ -His interactions, Wang and his group achieved an accurate assembly of tobacco mosaic virus coat protein (TMV) disks and obtained hierarchical, ordered, single-layer protein nanostructures with large sizes up to tens of micrometers ( Figure 3A). [20] The highly structured protein arrangements can provide periodic binding sites for nanoparticles to con-struct arrays. In addition to the natural histidine, unnatural chelating amino acids with imidazole moieties, such as bidentate bipyridyl-alanine (bpy-Ala), can also satisfy the binding of metal ligands as sufficient and dominant driving forces for protein assembly. [44] The valence state of M n+ is one of the influencing factors in regulating the coordination assembly strategy. The thermodynamic stability and ligand exchange kinetics of cobalt ions depend on their redox states, making the Co-liganddriven protein assemblies redox-stimulated responsive. [77] Driven by SpyTag/SpyCatcher chemistry and Co/His6-tag coordination, protein hydrogel networks formed. Notably, the mechanical properties of the hydrogel could switch with the change of oxidation states of the cross-linking metal ions ( Figure 3B), which provides a new strategy for the biochemical regulating of protein materials. In addition to the above ions, more transition metal ions, such as Ni 2+ , Fe 2+ , Mn 2+ , etc., can be involved in the chelation as M n+ to drive the assembly of proteins. [78] For further regulation, mild acidic condition or ethylenediaminetetraacetic acid (EDTA) complexation triggered protein disassembly may provide us with more inspiration. [79] Metal coordination strategy provides a directional means of supramolecular assembly for proteins, where the difference in affinity of metal ions for different ligands endows it a certain degree of selectivity. It is worth noting that this interaction is sensitive to ion concentrations and ligand molecule concentrations, so the ion density of the relevant components must be strictly controlled during the assembly process to avoid unwanted protein uncontrolled aggregation.

Electrostatic interaction-based protein assembly
The distribution of amino acids on the surface of proteins determines the charge enriched on their surface, which makes them promising for ordered aggregation guided by electrostatic interactions and regulated by controlling the ionic concentration and pH of the solution.
Stable protein one (SP1), of which the isoelectric point (pI) is about 4.3, has a lot of negative surficial charges in a neutral solution. [80] Utilizing three kinds of mercaptoethylamine (positive charged)-capped CdTe quantum dots (QDs) with different diameters (QD1: ∼3 nm, QD2: ∼5 nm, and QD3: ∼10 nm), SP1 were driven into straight nanowires (with QD1 and QD2) and branching nanowires (with QD3), respectively ( Figure 3C). [81] This ordered arrangement of quantum dots provides a structural basis for constructing light-harvesting antennas. TMV rod, which could be regulated by the pH and ionic concentration of the solution, is also enriched with negatively charged amino acids superficially, which allows it to spontaneously assembled driven by electrostatic interactions. [82] In order to avoid the random aggregation between TMV rods and oppositely charged Au nanoparticles (AuNPs), the initial assembly conditions were chosen as high ionic strength (c NaCl = 500 mM) to screen the charges of proteins and AuNPs. The NaCl was slowly removed by gradient dialysis, resulting in a highly ordered superlattice structure. [83] Furthermore, the superstructure exhibits structure-dependent chiral plasmonic function due to the right-handed helical twisting caused by the right-handed form of the virus.
Except for using the protein's own charge, modification of the protein surface charge by genetic engineering techniques is also an effective electrostatic design strategy. By introducing positively charged (Arg, Lys) and negatively charged (Asp, Glu) amino acids onto the building block proteins, human heavy chain ferritin with encapsulation function, respectively, reengineered protein surfaces were achieved to obtain variant Ftn (pos) and variant Ftn (neg) , respectively, with oppositely charged surfaces. [84] The electrostatically driven crystallization process of the protein is jointly controlled by a fixed amount of NaCl (2 mM for Ftn (pos) and 0.6 mM for Ftn (neg) , respectively) and a gradient concentration of magnesium salt (100 mM ∼ 200 mM) in the buffer, and results in a tetragonal unit cell ( Figure 3D). Prior to crystal formation, two inorganic nanoparticles, cerium oxide and cobalt oxide nanoparticles, can be encapsulated in the cage-like protein building blocks, which can then co-crystallize with the protein and co-dope into the highly ordered crystal structure. On the other hand, the natural capsid-forming enzyme lumazine synthase of the hyperthermophilic bacterium Aquifex aeolicus (AaLS) is able to spontaneously assemble into icosahedrally symmetric particles and achieve encapsulation of riboflavin synthase. [85] When four negatively charged amino acids were obtained by genetic engineering mutations in the lumen of this cage, its encapsulation properties for positively charged proteins were increased to 10 times the original. [86] It is easy to see that the charge of protein building blocks (determined by the environmental pH) and the ion concentration of the solution regulate the electrostatic interaction-based protein assembly process together. In addition, sufficiently high ion strengths are able to cause the disassembly of constructed protein nanostructures. [87] Compared to the other non-covalent interactions, non-directional electrostatic interactions are more perturbed by the ionic concentration of the solution environment and are prone to agglomeration, which makes their controllable precise assembly more challenging.

3.2.3
Hydrogen bond-based protein assembly Protein assembly driven by the formation of hydrogen bonds formation includes two types: direct exploitation of natural hydrogen bonding between proteins, such as the dense hydrogen-bonding network between amyloids, [88] and DNA-driven protein assembly, in which single strands of DNA (ssDNA)/RNA are covalently modified on the protein surface to achieve recognition between building blocks based on the principle of complementary base pairing.
There are two standard tools to fabricate protein nanostructures based on DNA-pairing interactions: (1) using DNA origami as template scaffolds, proteins modified with corresponding complementary ssDNA/RNA are anchored to the origami backbone to drive well-organized alignment [89][90] ; (2) modifying two complementary ssDNA strands on each of the protein building blocks to drive ordered protein aggregation. [91] Compared with the original stable hydrogen bonds between proteins, the DNA pair-based construction strategy may be more programmable and controllable.
With classical slow kinetics of amyloid fiber assembly, the low-complexity sequence domain fused in sarcoma protein (FUS LC) was assembled into multiblock fibers. [92] By genetically fusing functional, structural domains (e.g., fluorescent proteins) or binding tags (e.g., SpyTag) to the terminal of FUS LC, Zhong's group obtained ordered supramolecular structures consisting of the functional modules by utilizing FUS LC's thermodynamically driven tendency to self-assemble ( Figure 3E). This assembly can obtain blocked protein 1D nanostructures by incorporating building blocks fused with different fluorescent proteins. Besides, the curli nanofibers (CNFs) of Escherichia coli (E.coli) biofilms have been proven to be formed by the spontaneous assembly of the protein subunit, CsgA, in an Amyloid-liked manner and can be genetically engineered to fuse co-secreted protein domains. [93] Glutamine and asparagine in CsgA proteins are responsible for forming the hydrogen bonding network that stabilizes the assembly structure ( Figure 3F), [94] while hydrophobic amino acids such as alanine exhibit strong adhesion to hydrophobic substrates. In addition, polar amino acids such as serine and tyrosine in CsgA proteins are rich in -COOH, -NH 2 , and other groups, which are required for the nucleation of MOFs. Controlled MOF growth on the biofilm-forming coating results in a surface coverage that is nearly 100 times larger than that of the original substrate.
In the DNA pair-driven protein assembly strategy, DNA origami first completes the construction of its framework F I G U R E 4 Protein nanostructures based on supramolecular interactions with the illustrations of related regulation. (A) Protein assembly on DNA origami through recognition and anchoring. Reproduced with permission: Copyright 2020, American Chemical Society. [97] (B) Divisible protein nanowires based on incomplete DNA pairing. Reproduced with permission: Copyright 2018, American Chemical Society. [99] (C) 1D GPX mimics formed via host-guest interactions between FGG and CB [8]. Reproduced with permission: Copyright 2013, Wiley-VCH. [102] (D) Controllable 2D protein nanosheets regulated by the guest competition mechanism. Reproduced with permission: Copyright 2021, Royal Society of Chemistry. [57] (E) Protein crystals based on π-stacking interactions and sugar-receptor recognitions. Reproduced with permission: Copyright 2014, Springer Nature. [54] (F) Protein assembly driven by π-stacking interactions and regulated by the pH and charges. Reproduced with permission: Copyright 2018, Royal Society of Chemistry [105] through the principle of complementary base pairing and then is employed to anchor the protein. [95] Due to its natural recognition of RNA strands, TMV has been shown to be in situ assembled on hybrid strands of DNA origami and RNA strands. [96] By programming the hybridized backbone, the programmable and controllable dynamic assembly of TMV can be further achieved ( Figure 4A). [97] This is done by designing the binding region and release-toehold regions of the RNA on the DNA backbone and exposing the RNA-protein binding sites through strand exchange. Editable DNA motifs make it possible to achieve control over the protein assembly process. Mirkin and his co-workers modified two sets of complementary ssDNA at the Cys site of the green fluorescent protein (mGFP) and used the DNA conformation to program the energy potential of the entire assembly process, thus enabling the controlled growth of protein nanostructures. [98] In addition to this, targeted cleavage of protein assemblies based on DNA-Pair interactions can be achieved by exploiting the difference in the degree of complementarity between ssDNA. [99] Aida's group modified partial complementary (66%) ssDNA on the upper and lower surfaces of the building block, chaperone protein (GroEL), to construct 1D nanostructures ( Figure 4B). When an oligomeric strand of DNA is added to the system that is completely paired (100%) with one of the modified strands, the fabricated assembly is "cleaved", and the protein tube is disassembled.
The directionality of H-bond interactions provides the stability of the protein assembly structures. On the other hand, the introduction of designable motifs, such as DNA strands, makes the assembly process more dynamic and controllable. However, direct introduction of hydrogen bond interactions between proteins should take into account of the inherent hydrogen networks to avoid disrupting the original conformation of proteins. In addition, the modification of protein surfaces with ssDNA is usually accompanied the synthesis of functional groups at the end of the base chains, which often requires more complex procedures and higher costs.

3.2.4
Host-guest interaction-based protein assembly As classical and effective supramolecular forces, the hostguest interactions have received much attention for their selectivity, high affinity, and stimulus responsiveness, and are therefore applied in the assembly and preparation of protein materials. [100][101] Among all of the host molecules, pumpkin-shaped cucurbit[n]urils exhibit clear advantages in driving protein assembly due to their self-identifying behavior in aqueous solutions. [32] Notably, the association constant K a between cucurbit [8]uril (CB [8]) and the guest molecule, Phe-Gly-Gly peptides (FGG), is as high as 1.5 × 10 11 M −2 , which makes it a cogent driving force for protein self-assembly. Liu's group genetically engineered FGG to be fused to the symmetrical N terminus of dimeric GST to obtain FGG-GST. [102] The fusion protein was then assembled with CB [8] along the two FGG linkage directions, and 1D protein nanowires were finally constructed ( Figure 4C). A Se-containing active center was successfully introduced into FGG-GST by cysteine auxotrophic expression. The selenocysteine-containing nanowires exhibit antioxidative glutathione peroxidase (GPX) activities which are 262 folds higher than other selenium-containing GPX mimics (2-SeCD). However, since FGG can only be fused to the ends of protein building blocks, higher dimensional assembly structures are difficult to obtain by this strategy for dimeric proteins like GST. More recently, a dynamically reversible 2D protein nanosheet based on host-guest interactions was reported by Liu and his co-workers. [57] The protein nanoarrays are formed by interaction-driven formation between CB [8] and an analogue of methyl viologen (MMV + , guest molecule, K a = 2 × 10 7 M −2 ) and can be disassembled based on the competition mechanism between the host-guest pairs ( Figure 4D). When the more affinity guest molecule, FGG, is introduced, CB[8] exhibits selectivity for binding to this guest molecule, thus releasing MMV + covalent-modified on the protein motif and triggering the disassembly of the protein nanostructures. Experimental results after adequate dialysis show that the process is fully reversible. Host-guest interactions based on the competition mechanism of guest molecules bring new inspirations for regulable protein self-assembly.

3.2.5
π-stacking interaction-based protein assembly Molecules with highly conjugated structures, such as one or more benzene rings, are able to form stable aggregated structures in aqueous solutions via π-stacking interactions. However, π-conjugated molecules are often limited in their applications due to their solubility in the physiological conditions of aqueous solutions. Rhodamine B (RhB), which possesses a conjugated structure, has been demonstrated to form dimeric structures in poor solvents at high concentrations, and the process is regulated by its own concentration and ionic concentration. [103] Based on this, Tezcan and his group modified RhB on the surface of two-dimensional protein nanosheets and achieved to promote the reassembly of protein structures. [104] Inspired by this, Chen and her co-workers realized the construction of two-dimensional protein crystals based on sugar-receptor recognition and π-stacking interactions. [54] Modification of a-D-mannopyranoside (Man) at the end of RhB molecule yields the "inducing ligand", Rh3Man, which can bind specifically to the sugar-receptor terminus of lectin ConA and drive the self-assembly of ConA into square protein crystal nanosheets together with π-stacking interactions ( Figure 4E). The dimerization process of RhB is thermodynamically controlled, which provides for the orderly structure of the protein assemblies. Via the dual non-covalent strategy, natural microtubule-liked protein nanotubes can be successfully constructed using native lectin soybean agglutinin (SBA) as the building block and R3GN as the inducing ligand to drive protein assembly. [105] Notably, the alternate passage of CO 2 and N 2 was able to dissolve and recover the assembled structure of the nanotubes, respectively. Chen's group explained the responsiveness of the system to CO 2 stimulation: the acidic gas influx changed the pH of the solution, which caused the SBA surface to change its charge from negative to positive, and the distance between the SBA and R3GN increased due to repulsion, thus inhibiting the formation of the assemblies ( Figure 4F). In addition, Liu's group anchored maleimide-functionalized RhB molecule (RhG2M) on the surface of enhanced green fluorescent proteins (EGFPs) by a covalent modification to construct oversized protein nanosheets up to 10 μm in size via π-stacking interactions between RhG2M. [106] In addition, this extra-large fluorescent protein sheets enable Förster resonance energy transfer (FRET) between the proteins and the RhB linkers.
In the design of the introduced π-stacking interactions, it is worth considering of the water solubility of the conjugated molecules. In addition, molecules with π-stacking properties may have quenching effects on fluorescent chromophores (e.g., green fluorescent proteins), which limits their application in the construction of fluorescent systems. Although the π-stacking strategy is still to be further developed, it has also brought new life to the construction of protein assembly.
The stimulus responsiveness of classical supramolecular interactions provides a dynamic and flexible approach to constructing protein assemblies and also allows this strategy to manifest great potential for the construction of innovative biomaterials.

Forceful covalent interactions
As previously mentioned, functional side chains of amino acids at protein surface sites offer the possibility of the covalent assembly of proteins. Covalent bonding through direct formation between amino acids or through covalent co-assembly with designed linkers is a practical construction method to achieve stable protein nanostructures. The covalent strategy is considered to be a powerful tool for protein assembly, especially disulfide bonds that are responsive to redox stimuli and provide a structural basis for reversible assembly. Ruthenium(II) trisbipyridyl dication (Ru(bpy) 3 2+ ) has been proven to catalyze tyrosine (Tyr) residues rapidly and efficiently coupling under visible light. [107] On the basis of this, protein nanosheets that were based on surface Tyr subjected to photo-induced cross-linking were successfully fabricated. [108] The growth process of the protein assembly could be fully photo-controlled and exhibits a satisfactory thermal stability ( Figure 5A). Furthermore, direct cross-linking between Cys is a common chemical construction strategy for protein assembly. Tezcan's group successfully constructed 2D protein crystal arrays using computer-aided simulations. [109] Using the regular tetrameric protein l-rhamnulose-1-phosphate aldolase (RhA) as the building block, the original amino acid at the four vertices of the square was mutated to Cys, and under the  [108] (B) Protein assembly driven by disulfide bond with redox responsiveness. Reproduced with permission: Copyright 2018, American Chemical Society. [110] (C) Protein light-harvesting systems based on the stable covalent coupling between linkers and cysteine. Reproduced with permission: Copyright 2019, American Chemical Society. [29] (D) pH-regulated protein active biomaterials used for repairing bone damages. Reproduced with permission: Copyright 2022, Wiley-VCH [113] oxidation of a small amount of β-mercaptoethanol, the Cys of the RhA mutants was covalently cross-linked to drive the formation of protein crystals. Under this construction strategy, the fine structures of the assemblies can be changed by artificially designing the cross-linking sites. Taking advantage of the cross-coupling of the sulfhydryl groups on the Cys side chain between the discoidal TMVs, Wang and his coworkers obtained monolayer protein nanosheets with a filtering function. [110] Cys has been introduced uniformly distributed on the side surface of the TMV discs to drive the assembly by copper ion-catalyzed oxidized disulfide bond formation. Due to the electrostatic repulsion between the upper and lower surfaces of the TMV disk, single-layer protein nanosheets with sizes up to tens of microns could finally be obtained ( Figure 5B). The homogeneous pore size structure of 4 nm of the nanosheet allows it to be utilized to separate particles of nanoscale size.
On the other hand, covalent cross-linking reactions between proteins and specific linkers are also confirmed to be an effective way to obtain ordered protein array structures. As a designed linker, a PEG chain with two maleimides "heads" was synthesized to couple with the Cys on the building block proteins (EGFPs) by "click" reactions to form ordered nanosheets ( Figure 5C). [29] This stratrgy enables a template-free method for constructing artificial light-harvesting systems. Notably, different lengths of PEG chains lead to differences in the size of the protein nanosheets, thus also reflecting the influence of the features of the linker on the structure of the assemblies.
Using the redox stimulus responsiveness of disulfide bonds, dynamically reversible protein assembly-disassembly processes can be achieved. [111][112] Very recently, Chen's group has developed a protein-based active biomaterial capable of participating in bone regeneration. [113] By using pH-sensitive Tris(2-carboxyethyl)phosphine (TCEP) as a reducing agent to break the original intramolecular disulfide bonds of salmon calcitonin (sCT, the building block protein) with the aid of pH regulation, and by promoting the formation of intermolecular disulfide bonds, protein assemblies with different structures in solutions of different acidity were obtained ( Figure 5D). The active material based on sCT is both a drug and a carrier for bone repair. And this regulable covalent interaction, which seeks a precise balance between oxidation conditions and pH, further advances the dynamic regulation of the protein assemblies. Besides, template-based strategies for protein self-assembly can also be used to control the density of protein assemblies by adjusting protein concentration to achieve multiple functions. [114] The covalent strategy brings a powerful approach to the construction of protein assembly; however, rapid and stable covalent cross-linking may also trigger defects in the structure of the assemblies. Tunable covalent assembly strategies with low reaction temperature can effectively improve this class of problems. In addition, there are relatively few amino acids in natural proteins that can be used for covalent crosslinking at precise locations. The introduction of non-natural amino acids may be an effective way to solve this problem.
In order to more clearly highlight the respective characteristics of the driving forces, we have compared the construction strategies mentioned in this review and their illustrations in Table 1  and the corresponding figures and references have been appended, respectively. We believe that flexible construction strategies can provide more possibilities for the smarter protein assembly and its interactions with the environment.

APPLICATIONS OF PROTEIN ASSEMBLY IN CONSTRUCTING BIOMATERIALS
The hierarchical structure and biocompatibility of protein assemblies exhibit great promise for the construction of biomaterials. The efficient specific recognition of protein building blocks, the complex structure of protein nanostructures, and the stimulus responsiveness of flexible construction strategies have together promoted the further functionalization and application of protein assemblies in biology. In this section, the application of protein assembly involved in biocatalytic material construction, encapsulation and release, artificial light-harvesting systems fabrication, and other functional biomaterials development will be highlighted.

Biocatalytic materials
The self-assembly of natural enzymes to construct multienzyme complexes has been a hot topic of research in the field of biocatalysis because it promises to improve enzyme activity. [115][116] The immobilized multiple enzymes have shorter traveling distances compared to free enzymes and are able to accelerate the catalysis of intermediates. [117] Based on the extremely robust isopeptide bonding between Spy Pairs, Xia and his research team have prepared stable protein scaffolds and realized the immobilization and complexation of multiple enzymes. [118] By using ELPs as linkers, the SpyCatcher domains, the binding domains of various enzymes, and the SpyTag peptides are linked in sequence, and the resulting tandem structures can be physically entangled into cross-linked protein scaffolds ( Figure 6A). The protein network scaffolds with four different docking domains can anchor multiple enzyme molecules of the terminal fusion receptor, resulting in a multi-enzyme complex. Evaluating the reaction kinetics of the free enzyme and the assembled enzyme complexes during the (1R,6R)-2-Succinyl-6hydroxy-2,4-cyclohexadiene-1-carboxylate (SHCHC) generation reaction, the experimental results showed that the value of k cat /K M of the immobilized system was more than twice that of the free enzyme (0.10 vs 0.04 min −1 ⋅M −1 ). In addition, the artificial simulation of natural enzymes with efficient catalytic properties is one of the highly anticipated research directions in biocatalytic materials science. Protein assemblies with special structural relationships provide the structural basis for the artificial mimicry of natural enzyme molecules, while the specific binding sites of natural proteins offer the possibility of functional mimicry of natural enzymes. [119][120][121] Taking advantage of the pH-regulated self-assembly behavior of TMV, Liu and his group successfully constructed protein nanowires with GPX catalytic activity through the surface engineering of the protein. The GPX mimics with muti-selenoenzyme active centers exhibited remarkable cat-alytic activity (the whole of 550,400 U⋅μmol −1 ), which even reaches the activity of native GPX (5780 U⋅μmol −1 ). [31] Using a similar approach to introduce the Se active site on the inner surface of the negatively charged SP1 rings, Seleno-SP1 with GPX-liked catalytic activity was consequently obtained. [87] When assembled into nanowires with the oppositely charged poly(amidoamine) dendrimers (PD5) via electrostatic interactions, the catalytic centers will be blocked in the assembly with a 95.5% blocking efficiency ( Figure 6B) and exposed with the recovery efficiency of 98.7% after dissociation by shielding effect of ions. The construction of "smart" artificial enzymes with switchable catalytic activity witnesses the progress in the development of intelligent biomaterials. Besides, by modifying manganese porphyrin in the dendrimer PD5 to obtain superoxide dismutase-active (SOD) mimics and co-assembling them with GPX-active Se contained-SP1, it was also promising to construct protein nanowires with dual enzyme synergy. [122] Catalytic functional protein assemblies containing natural enzymes exhibit an efficient catalytic activity. However, it is still difficult to catalyze reactions in organic solvents or aqueous solutions in extreme environments rely on protein assemblies due to the possibility of protein deactivation.

Encapsulation and release
Proteins and their assemblies are considered to be very appropriate for drug encapsulation and release from organisms due to their desirable biocompatibility and low cytotoxicity. In fact, protein assemblies have been widely used for the encapsulation and release of drugs or biomolecules. [123][124] Protein nanostructures are expected to be assembled and disassembled by stimulation of the external environment; for example, ferritin shows a pH-regulated encapsulation capacity, which forms into cages to encapsulate molecules such as drugs or enzymes under neutral conditions, while disassembles to release contents at pH = 2 or pH = 13. [125][126] The ordered assembly of 1D arrays of superparamagnetic iron oxide Nanoparticles (SNPs) is expected to be achieved by protein encapsulation. Variant GroELMC was obtained by covalently modifying 14 Cys contained photochromic merocyanine (MC) on the upper and lower surfaces of the barrel-shaped chaperonin GroEL, and when being added with Mg 2+ , the mutant would assemble into a nanowire structure by MC-Mg 2+ -MC interactions. [127] Ingeniously, SNPs could be encapsulated by GroELMC and consequently coassembled into micron-length 1D arrays ( Figure 6C). [128] In addition, the nanowires can be assembled into bundles controllably by an applied 0.5 T magnetic field and disassembled after the magnetic field disappears. Furthermore, TMV nanotubes with serum albumin (SA) as a stealth coating and targeting ligand were constructed for drug encapsulation and delivery. The delivery of the chemotherapy doxorubicin (DOX) was significantly better than that of free DOX and consequently retarded tumor growth.
On the other hand, Spherical protein vesicles or nanoparticles have also shown promising results in encapsulation and release. For instance, with surface polyethylene glycol coupling and intermolecular disulfide cross-linking, the electrostatic-attracted proteins and polypeptides were assembled into nanospheres by Yan and his coworkers F I G U R E 6 Applications of protein assembly in biology. (A) Multi-enzyme systems based on protein scaffolds. Reproduced with permission: Copyright 2019, American Chemical Society. [118] (B) Switchable catalytic protein nanowires regulated by ion concentrations. Reproduced with permission: Copyright 2020, American Chemical Society. [87] (C) Protein-encapsulated nanoparticles assembled into 1D nanowires. Reproduced with permission: Copyright 2018, American Chemical Society. [128] (D) Protein nanospheres responsive to tumor environment for drug delivery and release. Reproduced with permission: Copyright 2016, Wiley-VCH. [129] (E) Tunable catalytic artificial light-harvesting systems in response to redox stimuli. Reproduced with permission: Copyright 2022, American Chemical Society. [134] (F) Thermo-responsive artificial light-harvesting systems based on protein nanoparticles. Reproduced with permission: Copyright 2020, American Chemical Society. [135] (G) Functional biofilms with anchored inorganic nanoparticles. Reproduced with permission: Copyright 2019, Oxford University Press. [138] (H) Chemical protein hydrogels, physical protein hydrogels, and hybridized gels of both by covalent cross-linking or hydrophobic interactions, respectively. Reproduced with permission: Copyright 2016, Wiley-VCH [141] ( Figure 6D). [129] Photosensitizers, such as Chlorin e6 (Ce6), can be encapsulated in the constructed nanosphere and can be released rapidly through pH, redox potential, and protease concentration control, thus showing significant advantages in photodynamic therapy. The Z R /Z E interaction-based ELPs-mCherry protein vesicles, as described in the previous section, can further be modified to be size-controlled by the addition of non-natural amino acids (para-azido phenylalanine, PAP). [45,72] Furthermore, the protein vesicle excels in the encapsulation and delivery of DOX.
Protein carriers with excellent biocompatibility and designability for encapsulation and release of small molecules and nucleic acids are promising for medical applications, but at the same time, they have to face problems such as being degraded in the organism or triggering immune reactions. For this, the use of immune escape proteins or highly stable proteins as the building blocks may bring effective improvements.

Artificial light-harvesting systems
The "bottom-up" construction of artificial light-harvesting systems (LHSs) to mimic the natural LHSs is a good guide and facilitator for a deeper understanding of photosynthetic mechanisms and better use of the solar irradiation. [130][131] Since the capture and transfer of solar energy are almost inseparable from light-trapping complexes formed by protein-pigment interactions, it is of profound significance to construct artificial LHSs with protein scaffolds for mimicking. [132][133] Inspired by nature, Liu's group successfully constructed a series of artificial LHSs through the interactions between fluorescent chromophores and highly ordered protein assemblies. [28][29]42] Finding that the ring structure of SP1 is highly similar to that of the LH2 light-harvesting complexes of photosynthetic bacteria, donor chromophores were covalently modified at the periphery of the SP1 ring, and positively charged micelle were electrostatic-assembled with the modified protein as acceptor chromophores. [28] The resulting protein supramolecular nanowires controlled the distance between the donor and acceptor molecules to be limited to 2 nm for an efficient energy transfer. However, the energy transfer with a single path in the 1D structures may be interrupted due to possible point defects. The development of multi-dimensional LHSs can provide more paths for energy transfer process and can effectively address this issue. Further, via Tyr-Tyr cross-linking, Liu and his coworkers designed artificial LHSs based on 2D protein nanosheets. CdTe QDs were employed as fluorescent chromophores arranged on single-layered protein scaffolds with a 56% energy transfer efficiency, which structurally mimicked the natural LHSs. [42] More recently, a dynamic covalent bond-based artificial LHSs with sequential multistep FRET was constructed to realize photocatalysis. [134] Interestingly, the LHSs with an 84% energy transfer efficiency could be switched between "on/off" states by redox-regulated SP1 protein assembly or disassembly ( Figure 6E). The LHSs have been proven to promote the yield of cross-coupled hydrogen evolution reactions and to control the catalytic performance of the system through the degree of assembly.
In addition, protein nanoparticle-based temperatureresponsive artificial ALHSs were also constructed, whose structures highly mimic the LHSs of natural purple bacteria (Rhodobacter sphaeroides). [135] Hexameric tyrosinecoordinated heme protein (HTHP), a thermostable hemoprotein mutant, was employed to react with maleimide-tethering poly(N-isopropylacrylamide) (PNIPAAm) for fabricating the spherical protein nanoparticles at 60 • C ( Figure 6F). When the heme in the micellar assembly was substituted by a photosensitizer, Zn protoporphyrin IX (ZnPP), the protein spheres were successfully modified to artificial LHSs. Moreover, due to the temperature stimulus responsiveness of PNIPAAm, the protein nanoparticle can achieve at least 5 cycles between assembly and disassembly.
In summary, the defined structure of protein assemblies provides the structural basis for chromophore arrangements, and the response to environmental stimuli offers the possibility to mimic the self-protection mechanism of the natural LHSs. However, the artificial LHSs based on protein assembly also face new challenges: while the protein backbone that relies on surface modification or electrostatic interactions to precisely arrange chromophores achieves FRET, the loose distance between the photosensitizer and the catalytic center makes it difficult to realize the utilization of photoelectrons. Further design of the protein assembly structure is still needed to realize the possibility of the separation and utilization of photo-induced electrons.

Other functional biomaterials
In addition to the applications described above, the orderliness, flexibility, and precision at the nanoscale of protein assemblies offer great potential for the construction of other functional biomaterials. Zhong's group has been working on the development of amyloids, especially bacterial biofilms, into functional biomaterials. [136][137] For example, by fusing His6-tag at the terminal of the CsgA subunit of E.coli through gene engineering, functional biofilms that can anchor inorganic nanoparticles (NPs) were successfully obtained. [138] Presynthesized nanoscale QDs or AuNPs are arranged in a spatially precise manner on the surface of the biofilms, thus allowing the construction of three reusable and efficient catalytic systems for nitro contaminants reduction, dye degradation, hydrogen production, and functionalization, respectively ( Figure 6G). Moreover, using genetically engineered mussel foot proteins (Mefp3 or Mefp5) as the functional domains, they designed bi-layer protein hydrogels with tunable composition and mechanical properties via leucine zipper interactions. [139] Heart patches made of the constructed protein gel through 3D printing technologies can effectively reduce cardiac fibrosis, which exhibits a promising perspective for developing biomaterials.
As we have seen, protein hydrogels are considered an excellent biomaterial due to their modular designability, biocompatibility, and excellent viscoelasticity. Tirrell and his team have carefully genetically integrated the sequences of ELPs, matrix metalloproteinase-1 cleavage sites (MMP-1), cell adhesion ligands, and mCherry-leukemia inhibitory factor (LIF) to construct a modular protein hydrogel network entirely encoded by genes via Spys interactions. [140] The protein networks could encapsulate cells while maintaining the level of cellular activity and differentiation. The research group then introduced cysteines or curly helix structures into the terminals of the constructed building block proteins and thus created chemical or physical protein hydrogels and hybridized gels of both by covalent cross-linking or hydrophobic interactions, respectively ( Figure 6H). [141] In addition, using the UV-visible light-controlled tetrameric protein, Dronpa145N, [68][69] as the building block, Cao's group successively designed controllable functional protein hydrogels that can switch between sol-gel states. [142] The construction of these protein hydrogels also points to a general strategy for developing viscoelastic materials.
For example, in the construction of medical materials, a biodegradable wound dressing sponge based on the covalent cross-linking of ELPs, water-soluble N,O-carboxymethyl chitosan (N,O-CS), and oxidized cellulose nanocrystals has been successfully prepared for the problem of wound tissue adhesion and secondary injury caused by normal dry dressings. And the prepared composite sponge has been demonstrated to have an obvious preventive effect on the above problems. [143][144]

CONCLUSION AND OUTLOOK
For the past few years, protein-based nanostructures have been widely used in developing biocatalytic materials, encapsulation & release systems, artificial light-harvesting systems, and other functional biomaterials due to their designability, defined structures, ordered arrangement, and biocompatibility. The hierarchical and subtle protein assemblies could be designed and constructed in an artificially controlled manner, which makes them considered advanced materials. The recent developments of protein assembly based on classical controllable constructing strategies reflect its flexibility, editability, and the "intelligence" to interact with the environment, which is essential for further exploration and mimicry of living organisms. In fact, the construction and further application of protein assemblies also face many challenges: (1) misfolding of fusion proteins: the fusion of proteins is often involved in the design and generation of new protein interfaces based on symmetry principles, and artificially constructed proteins may exist in the inclusion body due to their misfolding under expression, making variant proteins inaccessible and unavailable. Based on experience, pre-theoretical calculations and computer-aided simulations have shown effects in ameliorating this problem. (2) Stability of protein assembly: the complex spatial structures formed via multiple non-covalent interactions, proteins and their assemblies are extremely sensitive to the microenvironment. To improve the stability of protein assembly structures, suitable physiological assembly conditions selection of stable building blocks and immobilized scaffolds are worth considering. (3) Intersection and control of the constructing strategies: the construction of protein assemblies generally requires the synergistic cooperation of multiple driving forces, and the optimal conditions for each interaction are sometimes difficult to completely orthogonalize. Therefore, the rational design and precise control of using non-orthogonal multiple interactions are the key points in protein assembly studies. (4) To date, artificially constructed protein nanostructures are still behind natural protein-based biomolecules in terms of the complexity of higher-order topologies. [145] And the most challenging issue is that artificially prepared protein nanomaterials still find it difficult to achieve the division of labor among assemblies, and to precisely manipulate complex life activities at the molecular level as protein assemblies in natural life structures.
Nevertheless, initial promising progress has been made in the structural mimicry and functionalization of protein assemblies by constructing hierarchical and precisely arranged protein assemblies. We have faith that with the continuous optimization and application of regulable and dynamic construction strategies, protein assemblies will be endowed with the ability to stimulate responsiveness and interact with the environment, and then synergistic cooperation between Assembly-Assembly and Assembly-Environment is expected to be achieved. In conclusion, protein assembly opens promising avenues for many fields such as bio-nanotechnology, exploration of living organisms, development of biomaterials.

C O N F L I C T O F I N T E R E S T
The authors declare no conflict of interest.