pocketz.doi.bio/2026-01-11_181946_nature_s41557-025-01929-2
De novo design of light-responsive protein–protein interactions enables reversible formation of protein assemblies | Nature Chemistry
URL: https://www.nature.com/articles/s41557-025-01929-2
Saved: 2026-01-11T02:19:51.672Z
Article Published: 28 August 2025 De novo design of light-responsive protein–protein interactions enables reversible formation of protein assemblies Bowen Yu, Jiao Liu, Zhanyuan Cui, Chu Wang, Peipei Chen, Chentong Wang, Yanzhe Zhang, Xingxing Zhu, Ze Zhang, Shichao Li, Jinheng Pan, Mingqi Xie, Huaizong Shen & Longxing Cao
Nature Chemistry volume 17, pages 1910–1919 (2025)Cite this article
11k Accesses
2 Citations
32 Altmetric
Metrics details
Abstract
Light-responsive proteins play an essential role in all domains of life by sensing and responding to environmental light signals. However, the de novo design of light-responsive proteins with precisely defined structures and reversible responsive behaviours is an unmet challenge. Here we describe a computational approach to design protein–protein interactions regulated by non-canonical amino acids, focusing on the light-responsive phenylalanine-4′-azobenzene (AzoF). Using this approach, we designed light-responsive cyclic homo-oligomers and heterodimers, which only assemble in AzoF’s trans configuration and disassemble when AzoF photoisomerizes to the cis configuration. Biophysical characterization confirms the light-responsive assembly and disassembly of these complexes, and the crystal structures match the design models with atomic accuracy. We demonstrate the applicability of these light-responsive proteins in constructing light-responsive hydrogels and engineering synthetic ligand receptors to optocontrol cell signalling in mammalian cells. Our approach opens avenues for designing environmentally responsive protein structures and broadens the toolkit for optogenetics and optochemistry.
You have full access to this article via your institution.
Download PDF Similar content being viewed by others Using azobenzene photocontrol to set proteins in motion Article 07 December 2021 Light-switchable transcription factors obtained by direct screening in mammalian cells Article Open access 02 June 2023 Spatiotemporal functional assembly of split protein pairs through a light-activated SpyLigation Article 17 April 2023 Main
Light-responsive proteins play an essential role in all domains of life by sensing and responding to environmental light signals1,2,3,4,5,6,7,8,9. Engineering these proteins to manipulate biological processes, using light as a precise, reversible and non-invasive control, has greatly advanced optogenetics and optobiochemistry10,11,12,13. These proteins typically sense light through bound chromophores, which absorb specific wavelengths and trigger protein conformational changes, initiating downstream signalling pathways. While natural photoreceptors such as LOV domains14 and phytochromes15 have been successfully applied in many contexts, their architectures and mechanisms are inherently limited by their native biological roles. Designing entirely new types of light-responsive systems offers the potential to overcome these limitations and expand the scope of dynamic and controllable protein interactions.
Protein design has become a powerful tool for creating proteins with novel structures and functions not found in nature16,17,18. Despite notable progress, programming de novo designed proteins to transition between states in response to environmental stimuli remains challenging19,20. Current efforts in engineering light-responsive proteins mostly focus on optimizing naturally occurring proteins for new biological processes21. There is a continuous need to design light-responsive switches with entirely new architectures that cannot be achieved using native proteins. Traditional protein design primarily relies on the standard 20 amino acids, which are generally inert to light. Consequently, designing de novo light-responsive proteins necessitates the integration of light-sensitive chromophores. However, achieving effective chromophore binding and efficiently translating the chromophore’s photoisomerization into protein structural changes remains an unmet challenge in de novo protein design.
Genetic code expansion offers a new route to expand the chemical capabilities of proteins beyond the canonical 20 amino acids22,23. This process uses a specialized tRNA/aminoacyl-tRNA synthetase pair, orthogonal to the host, to incorporate non-canonical amino acids at reassigned codons, introducing new functionalities such as light sensing. Azobenzene, a classic light-switching motif, undergoes reversible cis/trans isomerization with high quantum efficiency24. It has been widely applied in biological settings by chemically linking adjacent residues to control protein motion25,26,27,28. It has also been genetically engineered as a non-canonical amino acid, notably phenylalanine-4′-azobenzene (AzoF), to control DNA binding, fluorescent protein functions, peptide folding and enzyme activity29,30,31,32,33,34,35. However, previous incorporations of AzoF were case specific and lacked precise structural consideration, leaving its full potential in protein engineering largely unexplored.
Results
We sought to design de novo protein complexes whose assembly and disassembly can be controlled by the light-responsive non-canonical amino acid AzoF (Fig. 1a). This involves three main challenges: first, creating protein interactions with high shape and chemical complementarity to ensure specificity and prevent undesired states; second, accurately positioning AzoF at the protein interface core to enable light-triggered disassembly via isomerization-induced steric clashes; and third, designing unbound interfaces with minimal exposed hydrophobic residues to maintain solubility and prevent aggregation. Given the hydrophilic nature of cis-configured AzoF, we focused on complexes that assemble only in its trans configuration.
Fig. 1: Strategy for the design of light-regulated protein assemblies.
a, Schematic representation of integrating genetic code expansion with de novo protein design to precisely incorporate the non-canonical amino acid (ncAA) AzoF at the protein interface to control protein–protein interactions. The azobenzene group undergoes a reversible cis–trans photochemical isomerization. Irradiation at ∼340 nm converts the thermodynamically more stable trans isomer to the cis isomer; the cis form reverts thermally or upon irradiation at ∼420 nm. By embedding AzoF at the interface core, we can exert optical control over protein–protein interactions. b, Overview of the computational design pipeline of construction of light-regulated protein assemblies. The surface residues of the scaffolds were redesigned with hydrophobic aliphatic amino acids, and disembodied AzoF amino acids are docked onto the redesigned scaffold protein surface to generate the relative AzoF interaction field. These scaffolds were then assembled into various architectures guided by the relative AzoF interaction field and residue pair transform (RPX) score. The relative AzoF interaction field uses the scaffold as a reference, allowing it to move along with the scaffold. The interface sequences were designed to optimize the protein–protein interactions and the interactions with the AzoF. Designs with favourable interface interactions when AzoF is in its trans configuration but exhibiting severe steric clashes when AzoF is in its cis configuration were selected for experimental characterization. Some parts of a were created with BioRender.com.
Full size image Design method
We first generated a large set of helical bundle protein scaffolds consisting of three or four helices, with a total length of 65 amino acids (Fig. 1b). These scaffolds served as building blocks for construction of the light-responsive protein complexes. To identify AzoF incorporation sites, we begun by enumerating potential AzoF positions that could make favourable interactions with these scaffolds. Considering the hydrophobic nature of trans-configured AzoF, we redesigned scaffold surfaces using hydrophobic amino acids such as valine, leucine and isoleucine. We then docked AzoF residues against the modified scaffolds using the RifGen method, quickly assessing the binding energy at each position17,36. The AzoF residues making favourable non-polar interactions were identified, and their backbone coordinates relative to the targeted scaffold were stored in a six-dimensional spatial hash table for rapid look-up (Supplementary Fig. 1). This relative AzoF interaction field uses the scaffold as a reference, allowing it to move along with the scaffold. This differs from the traditional RifGen–RifDock procedure, which only permits scaffold docking to a fixed target. By adopting the relative AzoF interaction field, we can rapidly assess whether a protein subunit can incorporate an AzoF residue for favourable interactions with neighbouring subunits, even when both chains are moved.
We developed a Monte Carlo docking method to assemble protein scaffolds into larger complexes (Fig. 1b and Extended Data Fig. 1). An initial complex structure was generated based on the predefined symmetry configuration, with the initial position of each subunit randomly assigned. During each move, the backbone position of the first chain was perturbed, with changes propagated to other chains based on the symmetry configuration. For heterodimers, the second chain’s position remained fixed. The scoring function guiding the Monte Carlo optimization included: (1) the probability of incorporating an AzoF residue for favourable interactions with the neighbouring chain—rapidly evaluated by looking up the relative AzoF interaction field; and (2) the protein–protein interaction energies, assessed using the residue pair transform scoring method (RPX score), which estimates the energy of residue–residue interactions before explicit sequence design37,38. This sampling procedure, which relies solely on the backbone positions, is extremely efficient, enabling the evaluation of a million moves per minute in our protocol. For each final docked complex structure, the interface backbone position with a matching hit in the AzoF interaction field was used to install the corresponding AzoF rotamer. Then, docked configurations featuring favourable protein–protein interactions and properly installed AzoF amino acids were subjected to sequence optimization using Rosetta and ProteinMPNN to optimize the shape and chemical complementarity of interface39,40. The configuration of the installed AzoF amino acid remained fixed and was only allowed to undergo minimization during the sequence optimization process. A bias favouring polar amino acids was incorporated during the sequence design stages to increase the hydrophilicity of the unbound interface. Structures with favourable interactions in the AzoF trans configuration and steric clashes in its cis configuration were selected for experimental characterization.
Light-regulated cyclic homo-oligomers
We first applied our method to design cyclic homo-oligomers (C2–C5), which play critical roles in biological processes such as catalysis and signalling, and can also serve as building blocks for engineering biomaterials41,42,43. We selected 45 dimers, 67 trimers, 43 tetramers and 45 pentamers for experimental characterization, and named using the nomenclature: LRO (for light-regulated homo-oligomer), followed by the symmetry configuration Cn and a differentiating integer. Genes encoding these designs were synthesized with AzoF encoded by an amber stop codon. These genes were cloned into an expression vector with a C-terminal (His)6 tag. The proteins were then expressed in Escherichia coli along with an orthogonal aaRS/tRNA pair to incorporate AzoF44. The proteins were purified using immobilized nickel-affinity chromatography (Ni2+ IMAC), and 116 of the designs were recovered as yellowish eluents, indicating the incorporation of the AzoF amino acid. Size-exclusion chromatography (SEC) indicated that 22 designs exhibited an SEC elution profile that matched the expected oligomerization state. Mass spectrometry analysis indicated that the AzoF amino acid was incorporated into these proteins with high efficiency (Supplementary Fig. 2).
To test whether isomerization of AzoF can regulate protein assembly, we used SEC to evaluate protein oligomerization states before and after irradiation with 340/420-nm light. The results showed that 13 out of the 22 designs were light responsive, with seven exhibiting monomeric elution peaks after 340-nm irradiation, including two dimers, three trimers, one tetramer and one pentamer (Fig. 2 and Extended Data Fig. 2). LRO-C2-5, LRO-C3-7 and LRO-C4-13 showed almost complete disassembly into monomers when irradiated with 340-nm light (Fig. 2). Reirradiation of their disassembled monomeric fractions with 420-nm light readily restored their original oligomerization states. In contrast, LRO-C5-1 only partially disassembled into monomers and displayed alternative oligomerization states upon irradiation with 340-nm light (Fig. 2). This behaviour may be attributed to the high hydrophobicity of LRO-C5-1’s unbound surface (Supplementary Fig. 12), which increases the likelihood of non-specific associations. Analytical ultracentrifugation (AUC) results further confirmed the reversible assembly and disassembly of the complexes upon exposure to light of different wavelengths (Fig. 2). The other three designs either exhibited very weak light responsiveness, displaying a more concentration-dependent association behaviour, or tended to form large aggregates during the light-triggered disassembly–assembly process (Extended Data Fig. 2). Ultraviolet–visible (UV–vis) absorbance analysis confirmed that the azobenzene group readily isomerizes between the two configuration states within the proteins (Extended Data Fig. 3). Under 340-nm light irradiation conditions, the estimated cis-to-trans transition exceeded 90% for LRO-C2-5, LRO-C3-7 and LRO-C4-13, with a slightly lower transition ratio observed for LRO-C5-1 (Supplementary Fig. 3a). These results align with the apparent partial disassembly of LRO-C5-1 upon 340-nm irradiation. Additionally, the trans-to-cis transition was nearly complete for all tested designs. We further investigated the thermal relaxation rate of LRO-C2-5 after 340-nm irradiation and observed a thermal relaxation half-life of 82.7 h, which is comparable to previously reported values for AzoF thermal relaxation31 (Supplementary Fig. 3b). Moreover, circular dichroism (CD) experiments showed that these proteins were helical, well folded and stable up to 95 °C (Extended Data Fig. 4).
Fig. 2: De novo design and characterization of light-regulated homo-oligomers.
a, Superimposition of the computational design models (silver) on the experimentally determined crystal structures (colour). The AzoF amino acids are shown as sticks. The crystal structures are very similar to the computational design models. b, Zoom-in view of AzoF interaction environments, with key residues interacting with AzoF shown as sticks. The 2Fo − Fc omit electron densities of AzoF in the complex crystal structures are contoured at 1.5σ. c,d, Normalized SEC traces (c) and the AUC profiles (d) of the designs under different light irradiation conditions. Proteins assemble into the designed oligomeric complexes as shown by SEC and AUC (grey). Upon irradiation with 340-nm light, the azobenzene group undergoes isomerization to the cis configuration and the proteins disassemble into monomeric structures (cyan). Reversion to the trans configuration from the cis form is achieved through exposure to 420-nm light and the proteins also reassemble back to the oligomeric structures (blue dashed line). Source data for c,d are provided in the Source Data file.
Source data
Full size image
To facilitate high-resolution structure determination, we generated several sequence variants for the designs, with two to eight solvent-exposed charged residues mutated to alanine to promote crystallization (Supplementary Table 1). The selected positions for alanine mutations are distal from the protein interface and have minimal impact on their light-responsive behaviours (Supplementary Figs. 4 and 5). We successfully obtained high-quality crystals for the light-regulated homo-oligomers (Supplementary Fig. 6 and Supplementary Table 2). The crystal structure of LRO-C2-5 was resolved to a 2.4-Å resolution, closely matching the design model with a Cα r.m.s.d. of 1.45 Å (Fig. 2a). The interface features two AzoF amino acids, one from each protomer, engaging in compact hydrophobic interactions. The electron density clearly shows the AzoF configuration, with an average side chain atom r.m.s.d. of 1.0 Å, using the entire complex backbone as a reference (Fig. 2b). We obtained two crystal forms for LRO-C2-35 under two difference conditions (Extended Data Fig. 5a). Crystal form 1 had a resolution of 2.15 Å and matched the design model with a backbone Cα r.m.s.d. of 1.78 Å. However, its interface showed loosely packed hydrophobic interactions, with one AzoF amino acid appearing twisted in the electron density map (Extended Data Fig. 5a, left). Crystal form 2, with a resolution of 2.0 Å, deviated remarkably from the design model. This indicated there was energetic strain at the interface, consistent with its poor light-responsive behaviour observed in SEC analysis (Extended Data Figs. 2a and 5a, right). For the trimeric, tetrameric and pentameric complexes, crystal structures were resolved to 2.34 Å for LRO-C3-7, 2.20 Å for LRO-C4-13 and 3.00 Å for LRO-C5-1 (Fig. 2a,b). These structures closely matched their design models, with backbone Cα r.m.s.d. values of 1.01 Å, 1.74 Å and 1.13 Å, respectively. Each interface contained an AzoF amino acid establishing extensive hydrophobic interactions. The AzoF configurations in LRO-C3-7 and LRO-C5-1 were accurately represented, with average side-chain atom r.m.s.d. values of 1.20 Å and 1.07 Å. In LRO-C4-13, three AzoF residues matched the design model, but one showed a different rotamer configuration, causing minor deviations from the perfect C4 symmetry (Extended Data Fig. 5b).
Design of light-regulated heterodimer
Developing light-regulated heterodimers offers a promising tool for manipulating the proximity of two different proteins in cells, potentially influencing various cell behaviours21,45. To demonstrate the versatility of our design method, we next sought to design light-regulated heterodimers. These designs were named LRD-X, where LRD stands for light-regulated heterodimer and X is a unique identifier. The two chains were referred to as the AzoF chain (LRD-XA) and the vanilla chain (LRD-XV). We selected 36 designed heterodimers and synthesized the corresponding genes. Each heterodimeric pair was cloned into a bicistronic expression vector, with the AzoF chain tagged with a C-terminal (His)6 and the vanilla-chain tagged with a C-terminal green fluorescent protein mBaoJin46. After Ni2+ IMAC purification, the eluents of 23 designs exhibited clear green fluorescence, indicating successful heterodimer formation. We analysed these designs using SEC after exposure to 340-nm light. Three designs, LRD-2, LRD-7 and LRD-10, showed notable changes in their elution profiles, with peaks indicating disassembly into monomers (Extended Data Fig. 6a). The elution profile could be fully restored by irradiation with 420-nm light. We separately expressed the individual protomer chains and they all behaved as monomers in SEC (Extended Data Fig. 6b). Mass spectrometry analysis indicated high-efficiency incorporation of the AzoF (Supplementary Fig. 2). UV–vis analysis revealed that the azobenzene group readily undergoes isomerization between its two configurational states with high efficiency (Extended Data Fig. 3 and Supplementary Fig. 3a). Mixing AzoF and vanilla protomers led to heterodimer assembly, which disassembled under 340-nm light and reassembled with 420-nm light (Fig. 3c).
Fig. 3: De novo design and characterization of light-regulated heterodimers.
a, The crystal structures of LRD-2 and LRD-7 match the design model with atomic accuracy. The computational design models of the LRD-2 and LRD-7 (grey) are superimposed with the crystal structures (colour). For LRD-10, only the design model is shown. AzoF residues are highlighted as stick representations and coloured in pink. b, Zoom-in view of AzoF interaction environments, with key residues interacting with AzoF shown as sticks. The 2Fo − Fc omit electron densities of AzoF in the complex crystal structures are contoured at 1.5σ for LRD-2 and at 2.0σ for LRD-7. c, Normalized SEC traces of the heterodimers under different light irradiation conditions (native state, grey; 340-nm irradiation, cyan; 420-nm irradiation following treatment with 340-nm light, blue dashed line). d, BLI characterization of the binding and light responsiveness of the designed interface. The solid curves represent experimental data, and the dotted curves represent the fit curves. The fold changes in Kd were computed as the mean fold change observed between consecutive irradiation events within the reversible cycles shown in e. e, The binding affinities of the heterodimers assessed by BLI, after subjecting the AzoF protomers to sequential cycles of 340-nm and 420-nm irradiation. The process of irradiation with 340-nm light is indicated by cyan blocks, while the irradiation with 420-nm light is represented by blue blocks (see Supplementary Fig. 8). f, The interface specificity of the designed heterodimers. To assess the cross-reactivity of each AzoF protomer with each vanilla protomer, vanilla protomers were loaded onto BLI sensors, the BLI tips were then placed into solutions of AzoF protomers and their respective mutants (substituted with alanine or phenylalanine) through association, balance and dissociation processes. The heat map represents the maximum response signals obtained for each pair tested (Supplementary Fig. 7). Source data for c–f are provided in the Source Data file.
Source data
Full size image
We then investigated their binding kinetics and light responsiveness using biolayer interferometry (BLI). By immobilizing the biotinylated vanilla protomers onto streptavidin-coated sensors and placing the AzoF chain into the ligand well, we observed that these heterodimers exhibited rapid binding kinetics in native states, with affinities in the nanomolar range (Fig. 3d and Supplementary Fig. 7a). Exposure of the AzoF protomers to 340-nm light dramatically decreased their binding affinities, with a maximum fold change of 167× for LRD-7. Reirradiating the AzoF protomers with 420-nm light almost completely restored the biolayer interferometry profile. The transitions between the low- and high-affinity states were fully reversible after multiple light cycles (Fig. 3e and Supplementary Fig. 8). Moreover, mutating AzoF to alanine or phenylalanine completely disrupted the interaction for LRD-2 and LRD-7, and reduced the binding for LRD-10 (Fig. 3f and Supplementary Fig. 7b). These underscore the crucial role of AzoF at the interface and explain the substantial affinity change induced by light illumination. We also investigated the specificity of these designs. The designs showed complete orthogonality up to a concentration of 900 nM, the highest concentration tested (Fig. 3f and Supplementary Fig. 7a). Each AzoF protomer only bound its corresponding vanilla protomer, demonstrating specificity akin to natural proteins. This specificity arises from the precise AzoF incorporation site and its unique configuration designed, illustrating the potential of our approach to mimic natural specificity using non-canonical amino acids.
We successfully determined the crystal structures of the alanine mutation variants of LRD-2 and LRD-7, with resolutions of 2.18 Å and 1.40 Å, respectively (Fig. 3a and Supplementary Figs. 4–6). These structures match the design models with atomic accuracy, exhibiting a Cα r.m.s.d of 0.61 Å for LRD-2 and 0.98 Å for LRD-7. The electron density for the AzoF residues was distinctly resolved in both structures. Remarkably, in LRD-7, the two phenyl rings of AzoF were clearly identifiable, characterized by a hexagonal shape with a central hole (Fig. 3b). The configurations of the AzoF also closely matched the design model, with the side-chain atom r.m.s.d. in LRD-2 and LRD-7 being 1.26 Å and 0.81 Å, respectively, using the entire complex backbone as a reference.
Determinants of design success
Effective light responsiveness requires that the monomeric building blocks assemble into the intended oligomeric states when AzoF is in the trans configuration and remain well-behaved monomers when AzoF transitions to the cis configuration. These stringent requirements pose challenges for computational protein design, resulting in a relatively low initial design success rate (Extended Data Table 1). However, leveraging our large set of experimentally tested designs, we next sought to identify the key determinants of efficient light-responsive properties. By comparing the computational metrics of successful designs with those of failed ones, we identified several critical factors contributing to design success across various symmetric configurations (Extended Data Fig. 7 and Supplementary Figs. 9–13). Successful designs exhibit enhanced interface shape complementarity, which is crucial for stabilizing the correct oligomeric structures. Additionally, these designs feature deeply buried AzoF residues within the interface that are tightly packed by neighbouring residues. Such deep burial probably maximizes the impact of AzoF photoisomerization, inducing substantial global structural perturbations in the complex. In contrast, less buried AzoF residues may only cause minor backbone rearrangements during cis–trans isomerization, which fail to fully disassemble the complex into monomers. Furthermore, successful designs tend to have interfaces enriched with hydrogen bonds and exhibit lower surface hydrophobicity in their unbound states. A higher number of hydrogen bonds not only enhances interface specificity but also contributes to a more hydrophilic unbound surface. This could reduce the likelihood of unintended self-association or alternative oligomerization states when AzoF is in the cis configuration. Notably, LRD-7, which exhibited a large affinity fold change in response to different light irradiations, exhibited the lowest surface hydrophobicity among all tested designs and superior burial of AzoF within the interface (Supplementary Fig. 13). As with previous efforts in protein stability engineering and binder design17,47, iterative cycles of design–build–test, in which the design methodology and selection criteria are refined based on feedback from previous rounds, are expected to systematically improve the design strategy and enhance the success rate for developing light-responsive protein switches.
Design of light-regulated hydrogel
The diverse architectures and high engineering potential of our de novo designed light-responsive proteins offer building blocks for the development of light-responsive materials. Protein-based hydrogels are valuable in cellular engineering and medicine due to their biocompatibility, biodegradability and adaptability to chemical and biological modifications48,49,50. We next sought to utilize our de novo designed light-regulated homo-oligomers to create a unique protein hydrogel system. We constructed the light-regulated gel system using the light-regulated pentamer LRO-C5-1 and a previously designed obligate homodimer as the multifunctional network branch points, and cross-linked them covalently with the SpyTag/SpyCatcher system into a biopolymer network50,51 (Fig. 4a,b). We fused LRO-C5-1 to a mannose-binding protein (MBP) to increase the distance between the termini for cross-linking, which has been shown to be critical for the formation of stiff hydrogels50. Notably, the fusion of the highly soluble and monomeric MBP enhanced the light-triggered switching efficiency of LRO-C5-1 on SEC (Extended Data Fig. 8a). Both the SpyCatcher-fused homodimer and SpyTag-fused pentamer were efficiently expressed in E. coli. Bulk hydrogels were formed after mixing them with equivalent stoichiometry at a 10% w/v concentration (Fig. 4c). The hydrogel was subject to SDS–PAGE analysis, and the results indicated nearly complete cross-linking (Extended Data Fig. 8b). Upon 340-nm irradiation, the hydrogel transformed into a droplet-like fluid due to the disassembly of LRO-C5-1 (Fig. 4c). Reirradiation with 420-nm light restored the hydrogel, demonstrating a reversible sol–gel process that withstood multiple cycles without apparent property changes (Supplementary Movie 1).
Fig. 4: Design and characterization of light-regulated protein hydrogel.
a, Recombinant protein constructs for constructing the protein hydrogel. LRO-C5-1 is sequentially fused to the MBP protein and the SpyTag by flexible (GGS)n linkers to generate the pentameric cores, and the de novo designed obligate C2-symmetric homodimer is conjugated to the SpyCatcher to form the dimeric cross-linking module. b, Schematic representation of the light-regulated protein hydrogel constructed using LRO-C5-1. By mixing the SpyTag-MBP-LRO-C5-1 and the Dimer-SpyCatcher proteins at equimolar ratios, the constructs are covalently bonded through the SpyTag/SpyCatcher cross-linking chemistry, resulting in the formation of stiff gel networks (left). Irradiation with 340-nm light triggers the disassembly of MBP-LRO-C5-1, which reduces the cross-linking degree of the network and causes the gel to transform to a more fluid, droplet-like state (right). The original stiff gel network can be recovered by exposing it to 420-nm light. c, Photographs of the protein hydrogel after different light irradiations. d, Storage modulus (G′, closed circles) and loss modulus (G″, open circles) of the light-regulated protein hydrogel in the native state (black) and after irradiation with 340-nm light (cyan) derived from an angular frequency sweep. The hydrogel exhibited a higher G′ than G″ over a wide range of angular frequencies in the native state and the G′ notably dropped with 340-nm irradiation. e, Storage modulus of light-regulated protein hydrogel as a function of time for the corresponding states shown in c. Graphs in d,e show mean ± s.d. (n = 3 biologically independent samples), with individual data points overlaid. After irradiation with 420-nm light, the stiffness of the sol state gel rapidly increased, with the storage modulus swiftly reaching its plateau. f, The storage modulus and loss modulus exhibited reversible changes after alternating 340/420-nm light treatments for 12 complete cycles. Source data for d–f are provided in the Source Data file. Some parts of b,e were created with BioRender.com.
Source data
Full size image
We then assessed the viscoelastic properties of the hydrogel using oscillatory shear rheology. Initially, the hydrogel showed a higher storage modulus (G′) than loss modulus (G″) over a wide range of angular frequencies and oscillation strains, indicating the formation of a soft structure (Fig. 4d and Extended Data Fig. 8c). Irradiation with 340-nm light reduced the storage modulus to the levels of the loss modulus, indicating hydrogel dissolution. At 20% oscillation strain and 0.7 rad s−1 angular frequency, the average storage modulus dropped from ~91.0 to ~9.0 Pa, an approximate 10-fold change. Exposure to 420-nm light increased the stiffness, indicative of component cross-linking and hydrogel reformation. The storage modulus reached a plateau within 300 s, showing rapid response and association kinetics (Fig. 4e). Additionally, the hydrogel maintained reversible changes in the modulus for at least 12 complete cycles—the maximum tested—of alternating light treatments, substantiating the high reversibility of our design (Fig. 4f). Protein-based hydrogels have previously been engineered from de novo design proteins50. Our approach further advances previous work by enabling precise, spatiotemporal control over fluid properties, opening possibilities for constructing light-regulated biomaterials.
Light-regulated ligand–receptor pairs
The small size and genetically encodable nature of our de novo designed light-responsive proteins make them well suited for integration into cellular signalling pathways, enabling precise optogenetic control of cell behaviours with high spatiotemporal resolution. Leveraging the highly responsive heterodimer LRD-7, we next sought to engineer a light-regulated ligand–receptor system based on the Generalized Extracellular Molecule Sensor (GEMS) platform52. GEMS facilitates the engineering of modular receptor designs that incorporate user-defined, soluble ligand sensing through the introduction of a range of engineered ligand-binding extracellular domains, and transgene expression via the rewiring of distinct endogenous signalling pathways. We functionalized the extracellular domain of the EpoR transmembrane scaffold in the GEMS receptor by fusing it with LRD-7V using a Gly-Gly-Ser (GGS) linker. The dimeric ligand was created by coupling LRD-7A to a de novo designed obligate homodimer using either a GGSGG (5GS) or GGSGGSGG (8GS) linker (Fig. 5a and Extended Data Fig. 9). When LRD-7A associates with LRD-7V, the dimeric ligand induces a conformational change in the GEMS receptor, triggering downstream signalling. We expressed the dimeric ligand in E. coli and monitored receptor activation in HEK293T cells transiently transfected with the LRD-7V-GEMS receptor, utilizing the JAK/STAT, MAPK and PLCG pathways. Secreted embryonic alkaline phosphatase (SEAP), controlled by pathway-specific promoters, was quantified using a colorimetric assay. The synthetic ligand induced clear reporter gene expression for all pathways, with an approximate 5-fold increase for JAK/STAT and PLCG and an 18-fold increase for MAPK, using both linkers (Fig. 5b). Without further optimization, the fold increase was overall comparable to the fold change induced by small molecules in previous studies52. The 5GS linker was chosen for further assays. As expected, no activation occurred with the monomeric LRD-7 AzoF protomer. Consistent with expectations, bell-shaped dose–response curves were observed for all three signalling pathways when titrating the dimeric ligand (Fig. 5c).
Fig. 5: Design of light-regulated synthetic ligand–receptor pairs to control cell signalling in mammalian cells.
a, Schematic overview of the light-regulated ligand–receptor pair construction. LRD-7A was dimerized by fusing to the N-terminal of a designed homodimer with a GGSGG or GGSGGSGG linker, referred to as D7A-5GS and D7A-8GS hereafter. LRD-7V is fused to the extracellular part of the EpoR transmembrane scaffold of the GEMS receptor. The specific association between dimeric ligand and receptor would induce a conformational change of the GEMS receptor and trigger the activation of the target genes. b, Downstream JAK/STAT, PLCG and MAPK signal pathway activation and reporter gene expression by the dimer ligand D7A. HET293T cells were transiently transfected with LRD-7V-GEMS receptor utilizing the SATA3, NFTA and MAKA pathway, and SEAP was used as the reporter gene. SEAP activity (U l−1) was measured after incubating with D7A either non-irradiated or preirradiated with 340-nm light. The monomeric LRD-7A was used as a control. c, Titration of dimeric ligand D7A-5GS on HEK293T cells transiently transfected to express the LRD-7V-GEMS receptor utilizing the SATA3, NFTA and MAKA pathway. Cells were incubated with a range of concentrations of D7A-5GS. d, Schematic representation of the temporal control over gene expression. D7A-5GS was initially deactivated through preirradiation with 340-nm light before its introduction to HEK293T cells transiently expressing the LRD-7V-GEMS receptor. Then the mixture of deactivated ligands and transfected cells was irradiated with 420-nm light at different time points to observe a gradual restoration of full receptor activation. e, The results of the process shown in d. Graphs show mean ± s.d. from biologically independent samples (n = 3 for c; n = 4 for b,e), with individual data points overlaid. Relative SEAP activity was calculated as the fold change compared with the basal SEAP levels detected from the blank group (0 nM ligand) (see Supplementary Table 3 for significance analysis statistics). Source data for b,c,e are provided in the Source Data file. Some parts of a,d were created with BioRender.com.
Source data
Full size image
To test light responsiveness, we irradiated the homodimer-LRD-7A ligand with 340-nm light before incubation with receptor-transfected cells. This resulted in a notable decrease in signalling activation across all pathways, due to reduced binding affinities at the LRD-7 interface (Fig. 5b). Leveraging the reversibility of the interaction, we exposed the mixture of deactivated ligands and transfected cells with 420-nm light at different time points and observed a gradual restoration of full receptor activation (Fig. 5d,e). These findings demonstrate that signalling strength can be precisely regulated by light in our system. This modular framework for engineering light-regulated ligand–receptor pairs holds potential to expand applications of optogenetics in therapeutic and diagnostic contexts.
Discussion
De novo protein interface design, using either traditional physical docking or the latest deep learning methods, has enabled the creation of high-affinity binding reagents and protein complexes with diverse architectures for various applications17,37,53,54,55. However, these interfaces have typically been static, lacking the ability to control their interactions through external stimuli. Our method advances the field by enabling the precise positioning of non-canonical amino acids at protein interfaces. This expands the designable protein sequence and structural space beyond the canonical 20 amino acids. By utilizing existing non-canonical amino acids or developing new ones that can bind specific metal ions or undergo conformational changes in response to pH variations, and precisely incorporating them at protein–protein interfaces, our approach could enable external control of protein assembly through metal ions or pH. Our method offers a new route of designing dynamic and controllable protein interactions.
Although non-canonical amino acids have previously been incorporated into de novo designs, their use has been mostly limited to peptide construction56,57,58. Additionally, current methods for protein interface generation—whether through physics-based docking37,38 or deep learning-based generative approaches55—explicitly or implicitly rely on residue interaction information derived from canonical amino acids in native protein structures. As a result, the spacing and configurations of the generated protein interfaces are optimized for the packing and interactions of these canonical amino acids. Non-canonical amino acids, on the other hand, often feature larger side chains and more complex chemical properties, making their direct incorporation into such interfaces less efficient. In contrast, our approach pre-identifies optimal configurations of non-canonical amino acids that can form favourable interactions with neighbouring subunits and integrate them during the protein-docking process. This strategy enables the simultaneous optimization of canonical protein–protein interactions while creating the necessary space to effectively pack non-canonical amino acids. Despite these advances, there remains huge room for improvement because only a small fraction of the designs performed as intended under light irradiation conditions. However, the identification of several key factors from our large set of experimentally validated designs has yielded valuable insights. These findings provide a strong foundation for developing computational methods to design more sensitive and reliable light-responsive switches with higher success rates.
Compared to naturally occurring light-responsive proteins, our AzoF-based light-regulated protein assemblies offer several advantages: they are smaller, more stable and have a simplified light-switching mechanism. They also exhibit a greater diversity of architectures, including both cyclic homo-oligomers and heterodimers. The light-regulated homopentamer LRO-C5-1 appears to have no natural counterpart. We demonstrated the broad applicability of these proteins in engineering light-responsive hydrogels and in optogenetic control of cell behaviours. In our hydrogel system, the MBP protein can be easily replaced with other functional proteins to construct versatile light-responsive biomaterials for biomedical and biotechnological applications. Furthermore, the optogenetic control of cell behaviours using these de novo designed light-responsive proteins highlights their potential to confer light-sensing capabilities to cells or organisms that originally lack such perception. This is particularly advantageous for genomically recoded organisms, which offer greater compatibility with non-canonical amino acids59,60,61. Moreover, the helical topology of the scaffolds makes them ideal as building blocks for creating more complex light-regulated structures through rigid helical fusion or the latest deep-learning-based generative methods55,62,63,64. Additionally, the light-sensing spectrum of AzoF can be fine-tuned by altering the substituents on its phenyl rings, enabling the development of proteins with red-shifted sensing spectra that are more compatible with biological applications due to lower phototoxicity and better tissue penetration24,34,35,65. This spectrum tunability will also allow for multiplexed control of different processes using proteins with various AzoF derivatives.
As genetically encodable non-canonical amino acids with enhanced functionalities continue to revolutionize synthetic biology, our method can seamlessly integrate these advancements, thereby facilitating their broad application in protein engineering and de novo design. Our approach will extend de novo protein structure design into broader chemical spaces, paving the way for generating responsive protein complexes. These complexes hold promise for a wide range of applications, including intracellular signalling control in synthetic biology, the creation of responsive protein materials and the de novo design of light-powered molecular machines.
Methods AzoF param file preparation
The three-dimensional structure of AzoF was initially generated using ChemDraw 3D. Subsequent optimization was carried out via quantum mechanical calculations, utilizing the B3LYP functional alongside the DEF2-SVP basis set within the ORCA 5.0 software framework66. Conversion of the optimized structure into a non-canonical amino acid representation, AzoF, was accomplished using the molfiletoparams.py script from the Rosetta software suite. For AzoF, the backbone torsion parameters and rotamer libraries of the canonical amino acid phenylalanine were directly adopted. Because the benzene ring in phenylalanine exhibits two-fold symmetry relative to the Cβ–Cγ bond axis, in Rosetta, only half of the torsion space for the side-chain χ2 angle is utilized when modelling its side-chain configurations. In contrast, flipping the azobenzene group around the Cβ–Cγ bond axis in AzoF leads to entirely different configurations. To account for all possible configurations of AzoF, the χ2 torsion space of phenylalanine was extended to the other half during the generation of the AzoF rotamer library. Furthermore, considering the asymmetric nature of the azobenzene group, the rotation of the first phenyl ring was explicitly taken into account, resulting in a doubling of the total number of rotamers relative to those of phenylalanine. Lastly, the sequence design and backbone optimization features of Rosetta were used to refine the structures containing the AzoF non-canonical amino acid56.
Protein scaffold generation
We used helical bundle scaffolds consisting of three or four helices, with a total length of 65 amino acids. These scaffolds were constructed by the fragment assembly approach, reimplemented within our ProBuilder protein design package. Briefly, for each scaffold length, a random secondary structure configuration was sampled for each sampling trajectory. Corresponding secondary structure fragments were then integrated into the proteins through a Monte Carlo-simulated annealing protocol. The folding trajectory was guided by the rotamer pair transform score, which facilitated the protein-folding process. Initially, the well-folded backbone scaffolds were subjected to protein sequence design using the Rosetta FastDesign protocol. Subsequent to this, the amino acid sequences were further optimized through three iterative rounds of the ProteinMPNN-FastRelax method40,67. The candidate scaffolds were first screened based on the Rosetta energy score, selecting those with a scoreperres below 0.3 Rosetta Energy Units. These scaffold sequences were then subjected to the AlphaFold2 structure prediction pipeline to generate their respective predicted models68. The final set of scaffolds was chosen according to a stringent criterion: the backbone r.m.s.d. between the design model and the predicted structure had to be less than 0.6 Å across all five AlphaFold2 models. A final set of 1,149 monomeric scaffolds was selected for the design of light-responsive protein assemblies.
Relative AzoF interaction field generation
The surface residues of the final selected scaffolds underwent a redesign via the Rosetta sequence design protocol, with the allowed amino acids being constrained to hydrophobic aliphatic types, specifically valine, leucine and isoleucine. Subsequently, the disassembled AzoF amino acids were docked onto these modified scaffolds using the branch-and-bound search algorithm implemented by RifGen17,36. This step made it possible to rapidly evaluate potential favourable hydrophobic interactions with the target. The AzoF amino acids that showed favourable hydrophobic interactions with the scaffolds were then chosen, with approximately 250,000 AzoF residues selected for each scaffold (Supplementary Fig. 1). The coordinates of their backbones, relative to the reference frame of the first residue’s backbone, were stored in a six-dimensional spatial hash table, with the rotamer χ angles recorded as associated values. This hashing mechanism affords a rapid retrieval capability; given the relative positions of two protein chains, it can assess whether one chain can accommodate an AzoF residue that is capable of making favourable interactions with the other chain, based solely on their relative positioning. Our method also differs the original RifGen–RifDock pipeline by using the AzoF interaction field relative to the scaffolds, which allows both chains to move during the docking process.
Protein complex generation
The scaffolds were assembled into complex structures by utilizing the NCAA docking protocol within our ProBuilder protein design package. This protocol started by constructing an initial symmetric complex in accordance with predefined symmetry. During each Monte Carlo step, the backbone positions of the first chain were perturbed, with these perturbations being propagated to the other chains to maintain the symmetric configuration of the complex. Notably, for the design of heterodimeric complexes, only the position of the first chain was allowed to move and the position of the second chain remained fixed. We adopted the symmetry-centric framework adeptly employed by Rosetta, which confines the conformational search space by examining only symmetric degrees of freedom and explicitly simulating only a subset of the interacting monomers69. With the backbone coordinates updated, protein interactions are rapidly assessed by calculating the RPX score for each interacting residue pair37,38. This was achieved by looking up a precompiled library of full-atom Rosetta score interactions of all the residue pairs in the native protein structures. Additionally, the backbone coordinates of the second chain, relative to the first residue’s backbone frame of the first chain, are computed. Their spatial hash values are then calculated to facilitate access to the relative AzoF interaction field table. This step enables a rapid evaluation of whether an AzoF residue could potentially be incorporated into the second chain that can then make favourable interactions with the first chain. The summed RPX score and the relative RIF score was used to guide the Monte Carlo-simulated annealing docking trajectory. For each final docked structure, the interface positions were evaluated against the AzoF interaction field. For the positions with matching hits, the corresponding AzoF rotamer was installed into the respective backbone locations. For docking configurations with multiple hits in the AzoF interaction field, each hit was treated independently in the subsequent sequence design stages. The AzoF amino acid is incorporated, and its interface burial and potential clashes are rapidly evaluated. Designed structures with favourable interface interactions, optimal AzoF burial were selected for the following sequence design processes.
Protein sequence design
The structures exhibiting optimal AzoF burial without backbone clashes are then subjected to sequence design using the Rosetta FastDesign protocol. During this process, only design of the interface residues was allowed; residues that were distant from the interface or deeply buried within the scaffold core were kept fixed. Subsequently, the deep neural network protein sequence design method, ProteinMPNN, was used to further optimize the interface amino acid sequences distal from the AzoF residue40,67. Once a new ProteinMPNN sequence was sampled, it was threaded onto the scaffolds within the complex, and the entire complex was subjected to relaxation using the Rosetta FastRelax protocol. The resulting relaxed complex structure then served as the starting point for another iteration of ProteinMPNN, perpetuating the design–refinement cycle. This iterative process was repeated three times. Throughout both stages of sequence design, a bias was introduced to favour polar amino acids, thereby enhancing the hydrophilicity of the unbound interface. Moreover, the configuration of the installed AzoF amino acid was fixed to prevent repacking and was only allowed to undergo minimization during the structure relaxation process for both sequence design stages. The final selection of structures was based on their ability to maintain favourable interface interactions when the AzoF was in its trans configuration, while simultaneously making severe steric clashes when AzoF adopts its cis configuration.
Computational metrics for in silico design selection
For the sequence-optimized designs, we applied several computational metrics to facilitate in silico design selection. These metrics include:
scoreperres: the total Rosetta score for the entire complex, normalized by the protein length.
ddg_chA: the ∆∆G value of chain A relative to the rest of the complex.
contact_chA: the contact molecular surface area of chain A interacting with the rest of the complex.
sasa_chA: the buried solvent-accessible surface area upon the association between chain A and the rest of the complex.
scchA: the shape complementarity score of chain A with the rest of the complex. A higher scchA score indicates better shape complementarity between chain A and the complex.
AzoF_sasa: the solvent-accessible surface area of the AzoF residue, with lower values indicating that AzoF is deeply buried at the interface.
AzoF_contact: the contact molecular surface between the side chain of AzoF and its neighbouring residues17.
AzoF_neighbors: The number of neighbouring residues surrounding AzoF, weighted by the distance between AzoF and the Cβ atom.
ddgwithAzoF: the ∆∆G value of chain A relative to the rest of the complex without repacking the side chains.
ddgwithoutAzoF: the ∆∆G value of chain A relative to the rest of the complex after mutating AzoF to alanine, without repacking the side chains.
interface_hbonds: the number of hydrogen bonds formed at the interface between chain A and chain B.
sap_score: the hydrophobicity score of the unbound monomer, with lower values indicating reduced hydrophobicity17.
These metrics were used to identify designs with optimal interface characteristics, stability and responsiveness.
Vanilla homodimer design
The vanilla homodimer (obligate homodimer without AzoF at the interface) used in the cell signalling assay was generated using the same docking approach as previously described, with the RIF scoring disabled. This allowed us to focus on achieving stable homodimeric interactions without introducing AzoF at the interface. After docking, the resulting structures were subjected to rigorous sequence optimization using Rosetta and ProteinMPNN, ensuring the designed sequences were energetically favourable and compatible with the desired homodimeric configuration. To confirm the oligomerization state of the design, the homodimer was SEC coupled with multiangle light scattering (MALS). This analysis revealed a molecular weight consistent with that of a homodimer, confirming the successful formation of the intended structure.
Protein expression and purification
Genes were codon optimized for expression in E. coli, with the AzoF encoded as amber stop codon TAG. DNA fragments encoding the designed proteins were synthesized by Universe Gene Technology and cloned into the pET28a vector at the NcoI and XhoI restriction sites using Exnase (Vazyme). For coexpression constructs of light-regulated dimers (LRDs), the AzoF chain (with a C-terminal 6×His tag) and the vanilla chain (with a C-terminal mBaojin tag46) were synthesized in tandem, separated by a ribosome-binding site sequence (CTAGAAGGAGATATC), and inserted into pET28a at the NcoI and XhoI sites. Site-directed mutagenesis was carried out using primers tailored to introduce the desired mutations. PCR amplification was performed with KOD PCR master mix (Toyobo) to both mutate and linearize the plasmids; the products were then purified and ligated using a Blunting Kination Ligation Kit (Takara). AzoF was chemically synthesized by GL Biochem using a previously reported method29. The E. coli strain BL21(DE3) (NEB) was transformed with a chloramphenicol-resistant plasmid encoding the AzoRS4 aminoacyl-tRNA synthetase under the control of a T7 promoter, along with its corresponding tRNA, which were previously engineered to specific incorporation of the AzoF amino acid at the in-frame amber codon position44. Then its single colony was picked and further prepared into competent cells using the Inoue method. These competent cells were then transformed with the constructed plasmids. Resulting colonies were cultured in Luria–Bertani (LB) medium at 37 °C. Protein expression was induced with 0.2 mM isopropylthiogalactoside and 0.1 mg ml−1 AzoF when the OD600 reached 0.8–1.2, with subsequent incubation at 22 °C and 220 r.p.m. Screening of light-regulated oligomers was performed in 100-ml cultures, with further assays for light-responsive protein crystallization screening, and hydrogel preparation conducted in volumes of 1–6 litres. After 15 h of expression, the cells were pelleted at 10,000g for 10 min, resuspended in 15 ml TBS (150 mM NaCl, 20 mM Tris, pH 7.4), lysed by sonication, and the debris was removed by centrifugation at 17,000g for 15 min. For affinity purification, the supernatant was applied to 1 ml of Ni-NTA metal affinity chromatography resin (Qiagen), washed twice with 8× bead volume of wash buffer (150 mM NaCl, 20 mM Tris, 10/20 mM imidazole, pH 7.4), and the protein was eluted using a buffer containing 150 mM NaCl, 20 mM Tris and 300 mM imidazole, pH 7.4. Eluted proteins were analysed on a 4–20% SDS–PAGE (GenScript) and visualized with Coomassie blue staining. Control proteins, lacking AzoF, were expressed using standard BL21(DE3) cells following similar expression and purification procedures. The protein yield was measured using the BCA method with a standard protocol (Beyotime Biotechnology) (Supplementary Fig. 3a).
Light irradiation
The light source, 1 cm in diameter, emitted at power densities of 240 mW cm−2 and 1,500 mW cm−2 for wavelengths of 340 nm and 420 nm, respectively. In a typical set-up, 1 ml of a 50 μM protein solution in a 1.5-ml microcentrifuge tube was exposed to light for 120 s at 340 nm and 60 s at 420 nm. To mitigate sample heating, a consistent 3-cm gap was maintained between the light source and the samples. Following 420-nm irradiation, the protein samples were allowed to equilibrate at room temperature for 5 min prior to analysis by SEC and analytical ultracentrifugation. The duration of irradiation was adjusted in accordance with the sample volume and concentration; larger volumes were subjected to extended periods of exposure.
SEC
The chromatography system utilized was a fast protein liquid chromatography system (model UEV25D) from Union-biotech, equipped to detect at both 280- and 230-nm UV wavelengths with a flow rate set at 0.8 ml min−1. Data acquisition and processing were conducted using the system’s integrated software. The MBP-LRO-C5-1 protein was subjected to analysis and purification using a Superdex 200 Increase 10/300GL column (GE Healthcare), whereas other proteins were processed using a Superdex 75 Increase 10/300GL column. Tris-buffered saline (TBS) consisting of 150 mM NaCl and 20 mM Tris, pH 7.4, served as the running buffer. For the initial SEC screening of the LROs, 1.5 ml of protein eluate, obtained from a 5-ml Ni-NTA purification, was directly subjected to SEC analysis or irradiated at 340 nm before SEC analysis. Candidates with light responsiveness were then produced in larger volumes (1 litre), concentrated down to 1 ml, and further purified by SEC. Post-SEC, the purified samples were divided into three groups—native, 340-nm irradiation and 420- nm irradiation after 340-nm light exposure—each with a concentration of 0.3 mg ml−1 and a volume of 1 ml. These groups were then treated with the corresponding light and subjected to SEC analysis. The initial SEC screenings for LRDs were performed on coexpressed proteins, specifically the mBaojin-tagged vanilla chain and (His)6-tagged AzoF chain (Extended Data Fig. 6a). Candidate proteins, including LRD-2/7/10A and LRD-2/7/10V, were expressed individually, mixed in equimolar concentrations and incubated at room temperature for 20 min before SEC analysis (Fig. 3b). The SEC analysis procedures for LRDs were similar to LROs.
SEC-MALS
The analysis was carried out with an Arc HPLC system (Waters), which included a quaternary solvent manager and a sample manager. A Superdex 75 Increase 10/300GL column with a flow rate of 0.5 ml min−1 was used for separation. Samples with a concentration of 7.5 mg ml−1 and a volume of 50 μl were loaded for analysis. Detection was performed using an Arc 2489 UV detector (Waters), a DAWN MALS detector and an Optilab differential refractive index detector. ASTRA and HPLC CONNECT software were used for data collection and analysis.
UV–vis assay
After SEC purification, protein samples were diluted to 50 μM using TBS (150 mM NaCl, 20 mM Tris, pH 7.4) and divided into three groups—native, 340-nm irradiation and 420-nm irradiation after 340-nm light exposure. Following light treatment, 200 μl of each sample was placed in a cuvette and analysed using a Shimadzu UV-2700 spectrophotometer with a step size of 1 nm and an absorbance detection range of 200-650 nm at room temperature. The estimated cis-to-trans ratio was calculated based on a previously reported method70. The cis-to-trans ratio was calculated as [A334(native) − A334(420 nm)]/A334(native). These ratios were calculated using the absorbance at 334 nm.
Analytical ultracentrifugation
Samples were analysed using a Beckman Optima AUC equipped with an An-50 Ti rotor. Following SEC purification, the LROs were diluted with TBS (150 mM NaCl, 20 mM Tris, pH 7.4) to achieve an OD of 0.6–0.8 at 230 nm, as determined by a NanoDrop spectrophotometer (Thermo Fisher Scientific). A volume of 2 ml of the prepared sample was exposed to the designated light treatments prior to loading 380 μl into the AUC centrepiece. The AUC cell housing was properly sealed in accordance with the established experimental protocol. A blank control consisting of 400 μl TBS (150 mM NaCl, 20 mM Tris, pH 7.4) was similarly prepared. The samples were then subjected to centrifugation at 145,000g and 20 °C for 18 h. Detection was conducted at a wavelength of 230 nm with a sampling frequency of every 20 s. The resulting data were analysed using Sedimentation Analysis software, which used a continuous sedimentation coefficient (c[M]) distribution model. The partial specific volume, and the buffer density and viscosity, were calculated using SEDNTERP software.
CD
CD characterization was conducted using a Chirascan V100 instrument (Applied Photophysics). For the analysis, 130 μl of 0.04 mM LROs and 0.02 mM LRDs dissolved in TBS (150 mM NaCl, 20 mM Tris, pH 7.4) were placed into a quartz cuvette. Sequential wavelength scans were performed at 20 °C, followed by heating to 95 °C, and then a return to 20 °C to observe recovery after heating. Far-UV CD spectra were recorded within the 180–280 nm range. The average protein spectrum was obtained by subtracting the buffer baseline spectrum and this measurement was replicated three times for accuracy. The CD signal was converted to mean residue ellipticity ([θ]), dividing the corrected spectra by the product of the number of amino acid residues (N), the protein concentration (C) and the cuvette path length (L, which is 1 mm).
Mass spectrometry
Chromatographic separation was conducted on an ultraperformance liquid chromatography system (Waters), interfaced with a quadrupole time-of-flight mass spectrometer (Synapt XS, Waters) featuring an electrospray ionization source. A BEH C4 column (50 mm × 2.1 mm, 1.7 μm) at 40 °C was used for the separation process. The mobile phase A consisted of ultrapure water with 0.1% formic acid, and mobile phase B was composed of acetonitrile with 0.1% formic acid. The flow rate was maintained at 0.3 ml min−1. The gradient profile for mobile phase B started at 5% for 1 min, increased to 95% over 4 min, and was then held at 95% for an additional 2 min. Samples were injected at a volume of 1 μl (1 μM concentration). Detection was carried out using a Waters quadrupole time-of-flight mass spectrometer in electrospray ionization positive mode. The instrumental parameters were set as follows: capillary voltage, 3 kV; sample cone voltage, 50 V; source offset, 30 V; source temperature, 140 °C; desolvation temperature, 600 °C; nebulizer gas pressure, 35 p.s.i.; cone gas flow rate, 50 l h−1; desolvation gas flow rate, 800 l h−1. A full scan mass range from 200 to 2,000 m/z was executed. Data acquisition and processing were facilitated by Masslynx software.
Crystallization and structural determination
Attempts at crystallization were conducted using kits purchased from Hampton Research and Rigaku, which included Crystal Screen, PEGRx and Wizard Classic 1–4. These screenings, along with subsequent optimizations, were carried out with the aid of a two-position deck mosquito LCP (SPT Labtech) within 96-well crystallization plates using the sitting-drop method. Initial attempts to screen native protein crystals yielded either no crystals or crystals with poor diffraction quality. To improve the crystal quality, surface residue mutations were introduced into each light-regulated protein based on computational design models. For LROs, 2–4 mutation sites were chosen, while for LRD-2/7, 6–8 mutation sites were selected across two proteins, as detailed in Supplementary Table 1. To reduce the linker region, the (His)6 tag and design sequences were directly fused. For LRD crystallization screening, the mBaojin tag from the coexpression constructs was removed. This approach led to the successful crystallization of eight crystals from seven proteins. The crystallization conditions are summarized in Supplementary Fig. 6. The crystals were subjected to diffraction and indexing using an in-house X-ray diffraction system (Rigaku) utilizing Cu Kα radiation with a wavelength of 1.54 Å. The structure solution was approached using the molecular replacement method, utilizing the designed models as search templates. For LRO-C5-1 and LRD-2/7, oligomeric models served as the search models, while the remaining LROs were approached with monomeric models in a multistep search process. Molecular replacement and structure refinement were executed using the PHENIX suite71 and COOT72. Visualization of the structures was performed using ChimeraX73 and PyMOL (PyMOL Molecular Graphics System, 2002). The refinement statistics are comprehensively presented in Supplementary Table 2.
Hydrogel preparation and rheometry
Proteins SpyTag-MBP-C5-1 and Dimer-SpyCatcher were expressed in five 1-litre batches. After Ni-NTA affinity purification, the proteins were concentrated to approximately 200 mg ml−1 using ultrafiltration. Subsequently, the buffer was exchanged with TBS (150 mM NaCl, 20 mM Tris, pH 7.4). Hydrogels were prepared by mixing SpyTag-MBP-C5-1 and Dimer-SpyCatcher at a molecular ratio of 1:2.5, with a final concentration of 10% w/v. After mixing for 1 h, the cross-linking ratio between SpyTag-MBP-LRO-C5-1 and Dimer-SpyCatcher was analysed by SDS–PAGE. loading about 5 μg of sample for each lane. The rheological properties of the hydrogels were assessed using a rotational rheometer (TA Instruments-Waters, ARES-G2). The hydrogel was prepared in a 1-ml syringe and stored overnight at 16 °C. The rheological measurements were conducted using an 8-mm parallel plate with an 8-mm gap, at a temperature of 25 °C. Each test used approximately 45 μl of the hydrogel sample, and silicone oil was used to seal the edges to prevent evaporation. The linear viscoelastic regime was established at a strain of 20% across a frequency range of 0.01 to 1,000 rad s−1. Oscillation strain assays were performed at a constant frequency of 0.7 rad s−1, with strain values varying from 1% to 100%. For the dynamic modulus and repeated irradiation tests, the oscillation strain was fixed at 20% and the angular frequency at 0.7 rad s−1. Real-time monitoring of the changes in the modulus of the hydrogel samples was achieved using a detection plate under irradiation from one external 420-nm light source, spanning from 200 to 550 s, as illustrated in Fig. 4e.
BLI
BLI experiments were performed using the separately expressed LRD-2V-(His)6-AVI, 7V-(His)6-AVI, 10V-(His)6-AVI, 2A-(His)6, 7A-(His)6 and 10A-(His)6, and the AzoF mutant proteins. The experiments were conducted using an OCTET RED96E system (ForteBio) and the results were processed with the integrated software provided with the instrument. AVI-tagged LRD-2V, LRD-7V and LRD-10V proteins were first purified by SEC and the buffer changed to 80 mM NaCl, 20 mM Tris, pH 7.4. Biotinylation was carried out using a BirA-RT Kit (Avidity) at a protein concentration of 100 μM and incubation at 25 °C for 2 h. Following the reaction, the biotinylated proteins were subjected to desalting via SEC. The biotinylated LRD-2V, LRD-7V and LRD-10V proteins were diluted to a final concentration of 10 μg ml−1 prior to their loading onto the streptavidin-coated biosensors (SA, ForteBio). LRD-2A, LRD-7A and LRD-10A proteins were gradient diluted to achieve a range of concentrations from 0 to 900 nM. The buffer used for sensor activation, baseline stabilization and protein dilution was TBST (150 mM NaCl, 20 mM Tris, pH 7.4, with 0.05% Tween-20). In experiments involving 340- and 420-nm irradiation, the concentration of LRD-2A, LRD-7A and LRD-10A was maintained at 225 nM, while for the AzoF mutant proteins, a fixed concentration of 900 nM was used. For the assay involving repeated light irradiation, LRD-2A, LRD-7A and LRD-10A were prepared in a 4-ml volume within 5-ml tubes. Light sources of 340 nm and 420 nm were alternately used for irradiation, and 200-μl samples were pipetted out for analysis at each interval.
Synthetic ligand–receptor assay
The construction of receptor proteins and intracellular SEAP secretion pathways was based on previously reported methods52. The extracellular domains of the erythropoietin receptor (EpoR) plasmids for the three studied signalling pathways were identical, as outlined in Supplementary Table 4. Their intracellular domains were uniquely fused with IL6RBint, VEGFR2int and FGFR1int, corresponding to each pathway. HEK293T cells (ATCC, CRL-11268) were cultured in Dulbecco’s Modified Eagle Medium (DMEM, Thermo Fisher Scientific) supplemented with 10% fetal bovine serum (FBS, WISENT) and 1% streptomycin/penicillin (Thermo Fisher Scientific), at 37 °C in a humidified atmosphere containing 5% CO2. Cells were seeded into a 24-well plate at a density of 5 × 104 cells per well. After 12 h of culture, transfection was carried out per well as follows: for the JAK/STAT pathway, 200 ng receptor plasmid pSL889 (Psv40-7V-EpoR-IL6BRint-pA) and 100 ng STAT3 reporter plasmid pYW123 (OStat3-PhCMVmin-SEAP-pA) were utilized per well. For the PLCG pathway, 200 ng receptor plasmid pSL893 (Psv40-7V-EpoR-VEGFR2int-pA) and 100 ng NFAT reporter plasmid pSYQ034 (ONFAT-Pmin-SEAP-pA) were used. For the MAPK pathway, each well received 200 ng receptor plasmid pSLM112 (Psv40-7V-EpoR-FGFRint-pA), 100 ng TetR-dependent reporter plasmid pSLM112 (OTet-PhCMVmin-SEAP-pA) and 10 ng TetR-Elk1 fusion protein expression plasmid (pTetR-Elk1, PhCMVmin-TetR-ELK1-pA). Transfection was conducted using a DNA:PEI ratio of 1:3 (w/w). At 12 h post-transfection, the culture medium was replaced. For light irradiation studies, D7A-5/8GS was filter sterilized and then gradient diluted in PBS to the desired concentrations before being added to the cell culture supernatant. After a 15-h incubation, the supernatants were harvested for SEAP activity measurement using an Abcam SEAP Reporter Gene Assay Kit (AB133077). Briefly, 50 μl of supernatant from each well was heat inactivated for 30 min in a PCR machine. Subsequently, 10 μl of culture medium or serially diluted standard samples were transferred to the corresponding white 96-well plate wells, followed by the addition of 50 μl of the substrate to each well, with brief mixing. After a 10-min incubation, luminescence was recorded using a microplate reader (Thermo Fisher Scientific). The comparative experiments utilizing D7A-5/8GS were conducted at a protein concentration of 25 nM. For the 420-nm light recovery experiments, protein concentrations for the JAK/STAT, PLCG and MAPK pathways were set at 3 nM, 25 nM and 3 nM, respectively. Following a 3- or 10-h incubation, wells corresponding to each pathway were exposed to 420-nm light for 60 s. Statistical analyses were performed using one-way analysis of variance with a non-parametric test for multiple-group comparisons and an unpaired, two-sided t-test for comparisons between two groups. Statistically significant P values are summarized in Supplementary Table 3. All analyses were carried out using GraphPad Prism 8 for Windows (GraphPad Software, www.graphpad.com).
Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
Data availability
All data supporting the findings of this study are available in the Article and its Supplementary Information. Source data are provided with this paper. Coordinates and structure files have been deposited to the Protein Data Bank with accession codes 9JM1 (LRO-C2-5), 9JM2 (LRO-C2-35 form 2), 9JM3 (LRO-C2-35 form 1), 9JM4 (LRO-C3-7), 9JM5 (LRO-C4-13), 9JM6 (LRO-C5-1), 9JM7 (LRD-2) and 9JM8 (LRD-7). Source data are provided with this paper.
Code availability
The Rosetta macromolecular modelling suite (https://www.rosettacommons.org) is freely available to academic and commercial users. Design protocols, analysis scripts and design models used in this paper are available in the Supplementary Information and via https://github.com/LongxingLab/NCAALightAssembly. The source code for the docking protocol implemented in our ProBuilder package is freely available via https://github.com/LongxingLab/ProBuilder.
References
Herrou, J. & Crosson, S. Function, structure and mechanism of bacterial photosensory LOV proteins. Nat. Rev. Microbiol. 9, 713–723 (2011).
CAS PubMed PubMed Central Google Scholar
Filipek, S., Stenkamp, R. E., Teller, D. C. & Palczewski, K. G protein-coupled receptor rhodopsin: a prospectus. Annu. Rev. Physiol. 65, 851–879 (2003).
CAS PubMed Google Scholar
Legris, M. et al. Phytochrome B integrates light and temperature signals in Arabidopsis. Science 354, 897–900 (2016).
CAS PubMed Google Scholar
Ernst, O. P. et al. Microbial and animal rhodopsins: structures, functions, and molecular mechanisms. Chem. Rev. 114, 126–163 (2014).
CAS PubMed Google Scholar
Inoue, K. et al. A light-driven sodium ion pump in marine bacteria. Nat. Commun. 4, 1678 (2013).
PubMed Google Scholar
Maestre-Reyna, M. et al. Visualizing the DNA repair process by a photolyase at atomic resolution. Science 382, eadd7795 (2023).
CAS PubMed Google Scholar
Chen, X. et al. An extraordinary stringent and sensitive light-switchable gene expression system for bacterial cells. Cell Res. 26, 854–857 (2016).
PubMed PubMed Central Google Scholar
Zhou, Y. et al. A small and highly sensitive red/far-red optogenetic switch for applications in mammals. Nat. Biotechnol. 40, 262–272 (2022).
CAS PubMed Google Scholar
Romano, E. et al. Engineering AraC to make it responsive to light instead of arabinose. Nat. Chem. Biol. 17, 817–827 (2021).
CAS PubMed Google Scholar
Deisseroth, K. et al. Next-generation optical technologies for illuminating genetically targeted brain circuits. J. Neurosci. 26, 10380–10386 (2006).
CAS PubMed PubMed Central Google Scholar
Tischer, D. & Weiner, O. D. Illuminating cell signalling with optogenetic tools. Nat. Rev. Mol. Cell Biol. 15, 551–558 (2014).
CAS PubMed PubMed Central Google Scholar
Goglia, A. G. & Toettcher, J. E. A bright future: optogenetics to dissect the spatiotemporal control of cell behavior. Curr. Opin. Chem. Biol. 48, 106–113 (2019).
CAS PubMed Google Scholar
Kolar, K., Knobloch, C., Stork, H., Žnidarič, M. & Weber, W. OptoBase: a web platform for molecular optogenetics. ACS Synthetic Biology 7, 1825–1828 (2018).
CAS PubMed Google Scholar
Flores-Ibarra, A. et al. Light–oxygen–voltage (LOV)-sensing domains: activation mechanism and optogenetic stimulation. J. Mol. Biol. 436, 168356 (2024).
CAS PubMed Google Scholar
Hughes, J. & Winkler, A. New insight into phytochromes: connecting structure to function. Annu. Rev. Plant Biol. 75, 153–183 (2024).
CAS PubMed Google Scholar
Huang, P. S., Boyken, S. E. & Baker, D. The coming of age of de novo protein design. Nature 537, 320–327 (2016).
CAS PubMed Google Scholar
Cao, L. et al. Design of protein-binding proteins from the target structure alone. Nature 605, 551–560 (2022).
CAS PubMed PubMed Central Google Scholar
Kortemme, T. De novo protein design—from new structures to programmable functions. Cell 187, 526–544 (2024).
CAS PubMed PubMed Central Google Scholar
Boyken, S. E. et al. De novo design of tunable, pH-driven conformational changes. Science 364, 658–664 (2019).
CAS PubMed PubMed Central Google Scholar
Shen, H. et al. De novo design of pH-responsive self-assembling helical protein filaments. Nat. Nanotechnol. 19, 1016–1021 (2024).
CAS PubMed PubMed Central Google Scholar
McCue, A. C. & Kuhlman, B. Design and engineering of light-sensitive protein switches. Curr. Opin. Struct. Biol. 74, 102377 (2022).
CAS PubMed PubMed Central Google Scholar
Liu, C. C. & Schultz, P. G. Adding new chemistries to the genetic code. Annu. Rev. Biochem. 79, 413–444 (2010).
CAS PubMed Google Scholar
Chin, J. W. Expanding and reprogramming the genetic code. Nature 550, 53–60 (2017).
CAS PubMed Google Scholar
Jerca, F. A., Jerca, V. V. & Hoogenboom, R. Advances and opportunities in the exciting world of azobenzenes. Nat. Rev. Chem. 6, 51–69 (2022).
PubMed Google Scholar
Bozovic, O., Jankovic, B. & Hamm, P. Using azobenzene photocontrol to set proteins in motion. Nat. Rev. Chem. 6, 112–124 (2022).
CAS PubMed Google Scholar
Mart, R. J. & Allemann, R. K. Azobenzene photocontrol of peptides and proteins. Chem. Commun. 52, 12262–12277 (2016).
CAS Google Scholar
Hoersch, D., Roh, S. H., Chiu, W. & Kortemme, T. Reprogramming an ATP-driven protein machine into a light-gated nanocage. Nat. Nanotechnol. 8, 928–932 (2013).
CAS PubMed PubMed Central Google Scholar
Hoersch, D. & Kortemme, T. A model for the molecular mechanism of an engineered light-driven protein machine. Structure 24, 576–584 (2016).
CAS PubMed Google Scholar
Bose, M., Groff, D., Xie, J., Brustad, E. & Schultz, P. G. The incorporation of a photoisomerizable amino acid into proteins in E. coli. J. Am. Chem. Soc. 128, 388–389 (2006).
CAS PubMed Google Scholar
Padilla, M. S. & Young, D. D. Photosensitive GFP mutants containing an azobenzene unnatural amino acid. Bioorg. Med. Chem. Lett. 25, 470–473 (2015).
CAS PubMed Google Scholar
Crone, N. S. A. et al. Azobenzene-based amino acids for the photocontrol of coiled-coil peptides. Bioconjugate Chem. 34, 345–357 (2023).
CAS Google Scholar
Kneuttinger, A. C. et al. Light regulation of enzyme allostery through photo-responsive unnatural amino acids. Cell Chem. Biol. 26, 1501–1514.e1509 (2019).
CAS PubMed PubMed Central Google Scholar
Nakayama, K., Endo, M. & Majima, T. A hydrophilic azobenzene-bearing amino acid for photochemical control of a restriction enzyme BamHI. Bioconjugate Chem. 16, 1360–1366 (2005).
CAS Google Scholar
John, A. A., Ramil, C. P., Tian, Y., Cheng, G. & Lin, Q. Synthesis and site-specific incorporation of red-shifted azobenzene amino acids into proteins. Org. Lett. 17, 6258–6261 (2015).
CAS PubMed PubMed Central Google Scholar
Luo, J., Samanta, S., Convertino, M., Dokholyan, N. V. & Deiters, A. Reversible and tunable photoswitching of protein function through genetic encoding of azobenzene amino acids in mammalian cells. Chembiochem 19, 2178–2185 (2018).
CAS PubMed PubMed Central Google Scholar
Dou, J. et al. De novo design of a fluorescence-activating β-barrel. Nature 561, 485–491 (2018).
CAS PubMed PubMed Central Google Scholar
Fallas, J. A. et al. Computational design of self-assembling cyclic protein homo-oligomers. Nat. Chem. 9, 353–360 (2017).
CAS PubMed Google Scholar
Sheffler, W. et al. Fast and versatile sequence-independent protein docking for nanomaterials design using RPXDock. PLoS Comput. Biol. 19, e1010680 (2023).
CAS PubMed PubMed Central Google Scholar
Leaver-Fay, A. et al. ROSETTA3: an object-oriented software suite for the simulation and design of macromolecules. Methods Enzymol. 487, 545–574 (2011).
CAS PubMed PubMed Central Google Scholar
Dauparas, J. et al. Robust deep learning-based protein sequence design using ProteinMPNN. Science 378, 49–56 (2022).
CAS PubMed PubMed Central Google Scholar
Goodsell, D. S. & Olson, A. J. Structural symmetry and protein function. Annu. Rev. Biophys. Biomol. Struct. 29, 105–153 (2000).
CAS PubMed Google Scholar
Ali, M. H. & Imperiali, B. Protein oligomerization: how and why. Bioorg. Med. Chem. 13, 5013–5020 (2005).
CAS PubMed Google Scholar
Nishi, H., Hashimoto, K., Madej, T. & Panchenko, A. R. Evolutionary, physicochemical, and functional mechanisms of protein homooligomerization. Prog. Mol. Biol. Transl. Sci. 117, 3–24 (2013).
CAS PubMed PubMed Central Google Scholar
Israeli, B. et al. Genetically encoding light-responsive protein-polymers using translation machinery for the multi-site incorporation of photo-switchable unnatural amino acids. Adv. Funct. Mater. 31, 2011276 (2021).
CAS Google Scholar
Klewer, L. & Wu, Y. W. Light-induced dimerization approaches to control cellular processes. Chemistry 25, 12452–12463 (2019).
CAS PubMed PubMed Central Google Scholar
Zhang, H. et al. Bright and stable monomeric green fluorescent protein derived from StayGold. Nat. Methods 21, 657–665 (2024).
CAS PubMed PubMed Central Google Scholar
Rocklin, G. J. et al. Global analysis of protein folding using massively parallel design, synthesis, and testing. Science 357, 168–175 (2017).
CAS PubMed PubMed Central Google Scholar
Petka, W. A., Harden, J. L., McGrath, K. P., Wirtz, D. & Tirrell, D. A. Reversible hydrogels from self-assembling artificial proteins. Science 281, 389–392 (1998).
CAS PubMed Google Scholar
Banta, S., Wheeldon, I. R. & Blenner, M. Protein engineering in the development of functional hydrogels. Annu. Rev. Biomed. Eng. 12, 167–186 (2010).
CAS PubMed Google Scholar
Mout, R. et al. De novo design of modular protein hydrogels with programmable intra- and extracellular viscoelasticity. Proc. Natl Acad. Sci. USA 121, e2309457121 (2024).
CAS PubMed PubMed Central Google Scholar
Zakeri, B. et al. Peptide tag forming a rapid covalent bond to a protein, through engineering a bacterial adhesin. Proc. Natl Acad. Sci. USA 109, E690–E697 (2012).
CAS PubMed PubMed Central Google Scholar
Scheller, L., Strittmatter, T., Fuchs, D., Bojar, D. & Fussenegger, M. Generalized extracellular molecule sensor platform for programming cellular behavior. Nat. Chem. Biol. 14, 723–729 (2018).
CAS PubMed Google Scholar
Cao, L. et al. De novo design of picomolar SARS-CoV-2 miniprotein inhibitors. Science 370, 426–431 (2020).
CAS PubMed PubMed Central Google Scholar
Sahtoe, D. D. et al. Reconfigurable asymmetric protein assemblies through implicit negative design. Science 375, eabj7662 (2022).
CAS PubMed PubMed Central Google Scholar
Watson, J. L. et al. De novo design of protein structure and function with RFdiffusion. Nature 620, 1089–1100 (2023).
CAS PubMed PubMed Central Google Scholar
Renfrew, P. D., Choi, E. J., Bonneau, R. & Kuhlman, B. Incorporation of noncanonical amino acids into Rosetta and use in computational protein–peptide interface design. PLoS One 7, e32637 (2012).
CAS PubMed PubMed Central Google Scholar
Hosseinzadeh, P. et al. Comprehensive computational design of ordered peptide macrocycles. Science 358, 1461–1466 (2017).
CAS PubMed PubMed Central Google Scholar
Bhardwaj, G. et al. Accurate de novo design of membrane-traversing macrocycles. Cell 185, 3520–3532.e3526 (2022).
CAS PubMed PubMed Central Google Scholar
Lajoie, M. J. et al. Genomically recoded organisms expand biological functions. Science 342, 357–360 (2013).
CAS PubMed PubMed Central Google Scholar
Fredens, J. et al. Total synthesis of Escherichia coli with a recoded genome. Nature 569, 514–518 (2019).
CAS PubMed PubMed Central Google Scholar
Grome, M. W. et al. Engineering a genomically recoded organism with one stop codon. Nature 639, 512–521 (2025).
CAS PubMed PubMed Central Google Scholar
Vulovic, I. et al. Generation of ordered protein assemblies using rigid three-body fusion. Proc. Natl Acad. Sci. USA 118, e2015037118 (2021).
CAS PubMed PubMed Central Google Scholar
Hsia, Y. et al. Design of multi-scale protein complexes by hierarchical building block fusion. Nat. Commun. 12, 2294 (2021).
CAS PubMed PubMed Central Google Scholar
Divine, R. et al. Designed proteins assemble antibodies into modular nanocages. Science 372, eabd9994 (2021).
CAS PubMed PubMed Central Google Scholar
Hoppmann, C. et al. Genetically encoding photoswitchable click amino acids in Escherichia coli and mammalian cells. Angew. Chem. Int. Ed. 53, 3932–3936 (2014).
CAS Google Scholar
Neese, F. Software update: the ORCA program system—version 5.0. Wiley Interdiscip. Rev. Comput. Mol. Sci. 12, e1606 (2022).
Google Scholar
Bennett, N. R. et al. Improving de novo protein binder design with deep learning. Nat. Commun. 14, 2625 (2023).
CAS PubMed PubMed Central Google Scholar
Jumper, J. et al. Highly accurate protein structure prediction with AlphaFold. Nature 596, 583–589 (2021).
CAS PubMed PubMed Central Google Scholar
DiMaio, F., Leaver-Fay, A., Bradley, P., Baker, D. & André, I. Modeling symmetric macromolecular structures in Rosetta3. PLoS One 6, e20450 (2011).
CAS PubMed PubMed Central Google Scholar
Fischer, E. Calculation of photostationary states in systems A ⇌ B when only A is known. J. Phys. Chem. 71, 3704–3706 (1967).
CAS Google Scholar
Adams, P. D. et al. PHENIX: a comprehensive Python-based system for macromolecular structure solution. Acta Crystallogr. D 66, 213–221 (2010).
CAS PubMed PubMed Central Google Scholar
Emsley, P. & Cowtan, K. Coot: model-building tools for molecular graphics. Acta Crystallogr. D 60, 2126–2132 (2004).
PubMed Google Scholar
Goddard, T. D. et al. UCSF ChimeraX: meeting modern challenges in visualization and analysis. Protein Sci. 27, 14–25 (2018).
CAS PubMed Google Scholar
Download references
Acknowledgements
This work was supported by grants from the National Key R&D Program of China (2022YFA1303700, 2021YFC2301401), the National Natural Science Foundation of China (32370989) and the Westlake Education Foundation to L.C. This work was supported by State Key Laboratory of Gene Expression. We thank the Protein Characterization and Crystallography Facility of Westlake University for help in sample analysis; the Mass Spectrometry & Metabolomics Core Facility of Westlake University for sample analysis; the Westlake University HPC Center for computation assistance; and the Instrumentation and Service Center for Physical Sciences for sample analysis. This work was carried out with the support of Shanghai Synchrotron Radiation Facility, beamline BL19U1 (proposal 2024-NFPS-PT-501172).
Author information Authors and Affiliations
School of Life Sciences, Westlake University, Hangzhou, China
Bowen Yu, Jiao Liu, Zhanyuan Cui, Chu Wang, Chentong Wang, Yanzhe Zhang, Ze Zhang, Shichao Li, Mingqi Xie & Longxing Cao
School of Basic Medical Sciences, Shandong Second Medical University, Weifang, China
Bowen Yu
Artificial Intelligence Drug Design Core Laboratory, Westlake Laboratory of Life Sciences and Biomedicine, Hangzhou, China
Bowen Yu, Zhanyuan Cui, Chu Wang, Chentong Wang, Yanzhe Zhang & Longxing Cao
Biomedical Research Core Facilities, Westlake University, Hangzhou, China
Jiao Liu & Jinheng Pan
School of Communication Engineering, Hangzhou Dianzi University, Hangzhou, China
Peipei Chen
State Key Laboratory of Natural and Biomimetic Drugs, School of Pharmaceutical Sciences, Peking University, Beijing, China
Xingxing Zhu
Key Laboratory of Structural Biology of Zhejiang Province, School of Life Sciences, Westlake University, Hangzhou, China
Huaizong Shen
State Key Laboratory of Gene Expression, School of Life Sciences, Westlake University, Hangzhou, China
Longxing Cao
Contributions
B.Y. and L.C. designed the research. B.Y., P.C. and L.C. developed the design pipeline and made the designs. B.Y. performed most of the experiments with the help of J.L., Z.C., Chu Wang, Chentong Wang, Y.Z., X.Z., Z.Z, S.L. and J.P. All authors analysed the data. M.X., H.S. and L.C. supervised the research. B.Y. and L.C. wrote the manuscript with input from the other authors. All authors revised the manuscript.
Corresponding author
Correspondence to Longxing Cao.
Ethics declarations Competing interests
B.Y. and L.C. are coinventors on a provisional patent application that incorporates discoveries described in this manuscript. The remaining authors declare no competing interests.
Peer review Peer review information
Nature Chemistry thanks Jessica Cross, Sarel Fleishman, G. Woolley and the other, anonymous, reviewer(s) for their contribution to the peer review of this work.
Additional information
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Extended data Extended Data Table 1 Summary of LRDs and LROs during the screening process Full size table Extended Data Fig. 1 Scaffold docking and AzoF installation in complex generation processes.
The protein complex structure was generated using a Monte Carlo docking procedure, guided by a scoring function that optimizes protein-protein interactions and AzoF incorporation. For the final docked structure, the neighbouring backbone positions were queried against the AzoF interaction field, and the residue position with a matching hit was used to install the corresponding AzoF rotamer. For symmetric structures, AzoF installation was propagated to the other chains at their respective symmetric positions.
Extended Data Fig. 2 Design and characterization of light-regulated homo-oligomers.
Left, design models of LRO-C2-35 (a), LRO-C3-37 (b) and LRO-C3-64 (c). The AzoF residues are highlighted as stick representations. Right, the corresponding SEC traces of the designs under different light irradiation conditions. LRO-C2-35 showed a weak response to light, and its dimerization appeared to be more dependent on concentration, with a tendency to exist as a dimer at higher concentrations and primarily as a monomer at lower concentrations, as indicated by SEC (a, right). LRO-C3-37 predominantly dissociated into monomers following exposure to 340 nm light irradiation and formed non-specific aggregation complexes upon subsequent irradiation of the proteins with 420 nm light (b, right). Similarly, LRO-C3-64 also tended to form nonspecific aggregates when subjected to light treatment (c, right). A protein quantity of 0.3 mg was loaded onto the SEC column for LRO-C3-37 and LRO-C3-64, while two distinct loading quantities, 0.3 mg and 1.0 mg, were evaluated for LRO-C2-35.
Source data
Extended Data Fig. 3 Ultraviolet–visible absorption spectra analysis of designed LROs and LRDs.
Absorbance spectra indicated the photo stationary state of AzoF in proteins (grey) and the configuration transition of AzoF between trans and cis after 340 nm (cyan) or 420 nm (blue dash) light irradiation.
Source data
Extended Data Fig. 4 Circular dichroism analysis of designed LROs and LRDs.
The circular dichroism spectra confirmed the helical structure of the designed proteins, and most (6 out 7) were stable at 95 °C (grey, 20 °C; red, 95 °C; blue dash, 20 °C followed by 95 °C).
Source data
Extended Data Fig. 5 Crystal structures of the light-regulated assemblies.
a, Two different crystal forms were solved for LRO-C2-35. Left, Crystal form 1, crystallized under conditions comprising 200 mM lithium sulfate, 40% v/v PEG 400, 100 mM Tris hydrochloride at pH 8.5, aligned closely with the theoretical design model. In contrast, Right, Crystal form 2, grown in a solution containing 30% w/v polyethylene glycol 1500 and 0.1 M sodium acetate trihydrate at pH 4.5, displayed a marked deviation from the design model, featuring a chain that had rotated approximately 90 degrees. The design model was coloured in silver and the 2Fo − Fc omit electron densities of AzoF in the complex crystal structures are contoured at 1.5σ. b, The 2Fo − Fc omit electron densities of the four AzoF residues in the complex crystal structure of LRO-C4-13. The electron density contour is set at 1.5σ. Among the four AzoF residues, one (lower right) exhibited a configuration that diverged from the designed model. The designs models are shown as silver, and the crystal structures are shown with colours.
Extended Data Fig. 6 Size exclusion chromatography traces of the de novo designed heterodimers.
a, The vanilla protomers, fused with the monomeric green fluorescent protein mBaojin, were coexpressed with the AzoF protomers. Subsequent to various light irradiation treatments, the proteins were subjected to SEC analysis. The small peak, corresponding to the AzoF protomers, emerged post-340 nm light irradiation, which indicated the disassociation between the AzoF protomers and vanilla protomers. b, Size exclusion chromatography traces of vanilla and AzoF protomers expressed separately.
Source data
Extended Data Fig. 7 Comparison of the computational metrics distribution between all the designs with the successful designs.
The successful designs feature AzoF residues that are deeply buried within the interface (small AzoFsasa and large AzoFneighbors), well-packed by neighbouring residues (AzoFcontact), interfaces with more hydrogen bonds (interfacehbonds), lower hydrophobicity scores in their unbound states (sapscore) and improved interface shape complementarity (scchA). See Methods for the detailed description for each computational metrics.
Source data
Extended Data Fig. 8 Experimental characterization of the light-regulated hydrogel.
a, Schematic representation of the MBP-LRO-C5-1 construct (left) and the normalized size exclusion chromatography traces (right) of this protein under different light irradiation conditions. The MBP fusion notably enhanced the light responsiveness of LRO-C5-1 and exhibited almost complete switching between the pentameric state and the monomeric state. b, SDS–PAGE gel analysis the covalent cross-linking between SpyTag-MBP-LRO-C5-1 and Dimer-SpyCatcher. After hydrogel formation between the pentameric block and the dimeric block, the sample were dissolved in denaturing (SDS) running buffer and boiled before running on SDS–PAGE. Results show that upon covalent network formation through SpyLigation, the conjugated oligomers shift to higher position on the gel. c, Modulus changes of the gel during oscillation strain sweep from 1 to 100%, mean ± SD (n = 3 biologically independent samples), with individual data points overlaid.
Source data
Extended Data Fig. 9 Characterization of the designed vanilla homodimer and the dimeric ligand D7A.
a, Design model of the designed homodimer. b, Normalized size exclusion chromatography traces of the designed homodimer and D7A. c, Size exclusion chromatography-multi-angle light scattering data of the designed homodimer and D7A revealing both proteins correctly form dimer.
Source data
Supplementary information Supplementary Information
Supplementary Figs 1–13 and Tables 1–4.
Reporting Summary Supplementary Video 1
Reversible sol–gel transformation process of hydrogel.
Supplementary Data 1
Source data for supplementary information.
Source data Source Data Fig. 2
The source data file for Fig. 2c and d.
Source Data Fig. 3
The source data file for Figs. 3c, 3d, 3e and 3f.
Source Data Fig. 4
The source data file for Figs. 4d, 4e and 4f.
Source Data Fig. 5
The source data file for Figs. 5b, 5c and 5e.
Source Data Extended Data Fig. 2
The source data file for Extended Data Fig. 2.
Source Data Extended Data Fig. 3
The source data file for Extended Data Fig. 3.
Source Data Extended Data Fig. 4
The source data file for Extended Data Fig. 4.
Source Data Extended Data Fig. 6
The source data file for Extended Data Fig. 6.
Source Data Extended Data Fig. 7
The source data file for Extended Data Fig. 7.
Source Data Extended Data Fig. 8
The source data file for Extended Data Fig. 8.
Source Data Extended Data Fig. 8b
The uncropped gel image of Extended Data Fig. 8b.
Source Data Extended Data Fig. 9
The source data file for Extended Data Fig. 9.
Experimentally tested models
The PDB files for all tested models.
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
Reprints and permissions
About this article Cite this article
Yu, B., Liu, J., Cui, Z. et al. De novo design of light-responsive protein–protein interactions enables reversible formation of protein assemblies. Nat. Chem. 17, 1910–1919 (2025). https://doi.org/10.1038/s41557-025-01929-2
Download citation
Received 27 January 2025
Accepted 28 July 2025
Published 28 August 2025
Version of record 28 August 2025
Issue date December 2025
DOI https://doi.org/10.1038/s41557-025-01929-2
Share this article
Anyone you share the following link with will be able to read this content:
Get shareable link
Provided by the Springer Nature SharedIt content-sharing initiative
Subjects Protein design