Introduction
Retinylidene proteins are photochemically reactive proteins that are bound to or can bind to retinal (vitamin A aldehyde) as their chromophore (Spudich et al., Reference Spudich, Yang, Jung and Spudich2000). For convenience, I will use the term ‘opsin’ interchangeably with ‘retinylidene protein’, despite the fact that ‘opsin’ sometimes refers specifically to the chromophore-free apoprotein. Retinylidene proteins are divided into two groups, animal opsins and microbial opsins, which are evolutionarily distinct but share common characteristics such as having seven transmembrane (TM) domains and a retinal-binding lysine residue (Spudich et al., Reference Spudich, Yang, Jung and Spudich2000; Yee et al., Reference Yee, Shlykov, Vastermark, Reddy, Arora, Sun and Saier2013). In this study, I will consider nine animal opsins and three microbial opsins.
Animal opsins are a type of class A GPCR (G-protein-coupled receptor), which are characterized by seven transmembrane domains, the NPxxY motif, and activation of G proteins through the outward movement of TM6 (Zhou et al., Reference Zhou, Yang, Wu, Guo, Guo, Zhong, Cai, Dai, Jang, Shakhnovich, Liu, Stevens, Lambert, Babu, Wang and Zhao2019). Almost all animal opsins include a lysine residue that forms a Schiff base link with the retinal (Gühmann et al., Reference Gühmann, Porter and Bok2022). When retinal absorbs a photon, it isomerizes, usually changing from 11-cis to all-trans. The subsequent conformational changes of the protein are well studied using bovine rhodopsin, which changes from dark state to BATHO state, then LUMI, META I, and META II (Okada et al., Reference Okada, Ernst, Palczewski and Hofmann2001). Proton transfer plays an important role in this process (Mahalingam et al., Reference Mahalingam, Martínez-Mayorga, Brown and Vogel2008). Afterward, the protein bleaches and returns to the dark state, completing what is called the photocycle. As the protein conformation changes to META, TM6 moves characteristically outward, activating the G protein. In photoreceptor cells, the G protein, transducin, activates a phosphodiesterase, which hydrolyzes cGMP into GMP, decreasing the activity of cGMP-gated cation channels and hyperpolarizing the cell (Chabre and Deterre, Reference Chabre and Deterre1989).
In this study, I select nine opsins expressed in the human nervous system: OPN1MW (UniProt ID: P04001), OPN1LW (UniProt ID: P04001), OPN1SW (UniProt ID: P03999), OPN2 (UniProt ID: P04001), OPN3 (UniProt ID: Q9H1Y3), OPN4 (UniProt ID: Q9UHM6), OPN5 (UniProt ID: Q6U736), RGR (UniProt ID: P47804), and RRH (UniProt ID: O14718). They belong to several evolutionarily distinct families of animal opsins (Terakita, Reference Terakita2005).
OPN1MW (medium-wave-sensitive opsin 1), OPN1LW (long-wave-sensitive opsin 1), and OPN1SW (short-wave-sensitive opsin 1) are expressed in retinal cone photoreceptors and are responsible for color vision (Bowmaker and Dartnall, Reference Bowmaker and Dartnall1980). Certain variants of OPN1MW, OPN1LW, and OPN1SW, respectively, cause deuteranopia, protanopia, and tritanopia, which are different types of color blindness (Baraas et al., Reference Baraas, Hagen, Dees and Neitz2012; Ueyama et al., Reference Ueyama, Kuwayama, Imai, Tanabe, Oda, Nishida, Wada, Shichida and Yamade2002). The absence of both functional OPN1MW and OPN1LW causes blue cone monochromacy, an X-linked congenital cone dysfunction syndrome (Wissinger et al., Reference Wissinger, Baumann, Buena-Atienza, Ravesh, Cideciyan, Stingl, Audo, Meunier, Bocquet, Traboulsi, Hardcastle, Gardner, Michaelides, Branham, Rosenberg, Andreasson, Dollfus, Birch, Vincent, Martorell, Català Mora, Kellner, Rüther, Lorenz, Preising, Manfredini, Zarate, Vijzelaar, Zrenner, Jacobson and Kohl2022), and cone dystrophy 5, an X-linked cone dystrophy (Gardner et al., Reference Gardner, Webb, Kanuga, Robson, Holder, Stockman, Ripamonti, Ebenezer, Ogun, Devery, Wright, Maher, Cheetham, Moore, Michaelides and Hardcastle2010).
OPN2 (opsin 2), also known as rhodopsin, is expressed in retinal rod photoreceptors and is responsible for vision at low light intensity (Hubbard and Kropf, Reference Hubbard and Kropf1958). Certain variants lead to autosomal recessive or autosomal dominant retinitis pigmentosa and congenital stationary night blindness (Fanelli et al., Reference Fanelli, Felline and Marigo2021). OPN2 is a representative animal opsin, one of the first studied. In fact, bovine OPN2 is the first opsin to be sequenced (Nathans and Hogness, Reference Nathans and Hogness1984), as well as the first GPCR whose crystal structure was resolved experimentally (Palczeski et al., Reference Palczeski, Behnke, Motoshima and Fox2000). Many studies on the functional mechanisms of animal opsins also focus on OPN2. Consequently, I chose OPN2 to conduct a functional analysis in this study in order to further explore the effectiveness of the QTY code in redesigning retinylidene proteins.
OPN3 (opsin 3), also known as encephalopsin or panopsin, is activated by blue and ultraviolet A light. It was discovered in the brain (Blackshaw and Snyder, Reference Blackshaw and Snyder1999). It is also expressed in melanocytes and keratinocytes in the skin and regulates functions such as melanogenesis, cell differentiation, and glucose uptake. Its expression is also found in the liver, pancreas, kidney, lung, heart, and skeletal muscles. When expressed in neurons, the release of neurotransmitters is inhibited by light, making OPN3 a useful inhibitory optogenetic tool (Copits et al., Reference Copits, Gowrishankar, O’Neill, Li, Girven, Yoo, Meshik, Parker, Spangler, Elerding, Brown, Shirley, Ma, Vasquez, Stander, Kalyanaraman, Vogt, Samineni, Patriarchi, Tian, Gautam, Sunahara, Gereau and Bruchas2021).
OPN4 (opsin 4), also known as melanopsin, is expressed in ipRGC (intrinsically photosensitive retinal ganglion cells) in the ganglion cell layer in the retina (Provencio et al., Reference Provencio, Jiang, De Grip, Hayes and Rollag1998). It is essential for non-image-forming responses to light, including the pupillary reflex, optokinetic visual tracking response, and photoentrainment and regulation of circadian rhythm. Certain variants of OPN4 lead to seasonal affective disorder and other circadian rhythm disorders (Berson et al., Reference Berson, Dunn and Takao2002). Rendering OPN4-containing ipRGCs capable of image formation is also a potential pathway for the treatment of various eye diseases such as retinitis pigmentosa and diabetic retinopathy.
OPN5 (opsin 5), also known as neuropsin, is activated by blue and ultraviolet A light (Tarttelin et al., Reference Tarttelin, Bellingham, Hankins, Foster and Lucas2003). It is expressed in the retina and contributes to the regulation of light-dependent vascular development and photoentrainment in the cornea and retina (Buhr et al., Reference Buhr, Yue, Ren, Jiang, Liao, Mei, Vemaraju, Nguyen, Reed, Lang, Yau and Van Gelder2015). Certain variants of OPN5 may lead to cycloplegia, paralysis of the ciliary muscle in the eye.
RGR (RPE-retinal GPCR) is expressed in RPE (retinal pigmented epithelium) and Müller cells in the retina (Shen et al., Reference Shen, Jiang, Hao, Tao, Salazar and Fong1994). Unlike the aforementioned human opsins, RGR preferentially binds all-trans-retinal and may catalyze its isomerization into 11-cis-retinal via a retinochrome-like mechanism. It is expressed only in tissue surrounding photoreceptors and plays a role in the light-dependent synthesis of visual chromophore (Radu et al., Reference Radu, Hu, Peng, Bok, Mata and Travis2008).
RRH (RPE-derived rhodopsin homolog), also known as peropsin, is localized in the microvilli of RPE cells that surround photoreceptor outer segments (Sun et al., Reference Sun, Gilbert, Copeland, Jenkins and Nathans1997). It is another protein that preferentially binds to all-trans-retinal (Cook et al., Reference Cook, Ng, Lloyd, Eddington, Sun, Nathans, Bok, Radu and Travis2017). Although not much information is known about RRH, it can reasonably be inferred that it photoisomerizes all-trans-retinal to 11-cis-retinal and may play a role in the upkeep of photoreceptor functions.
Microbial opsins are transmembrane ion pumps or channels (Findlay and Pappin, Reference Findlay and Pappin1986). They are also 7TM, though this is due to convergent evolution rather than homology (Yee et al., Reference Yee, Shlykov, Vastermark, Reddy, Arora, Sun and Saier2013). The chromophore, retinal, usually isomerizes from all-trans to 13-cis, different from the case of animal opsins (Findlay and Pappin, Reference Findlay and Pappin1986). In addition, microbial opsins often form oligomers to carry out their functions (Gmelin et al., Reference Gmelin, Zeth, Efremov, Heberle, Tittor and Oesterhelt2007).
In this study, I select three microbial opsins: BACR (UniProt ID: P02945), BACH (UniProt ID: B0R2U4), and ChR2 (UniProt ID: Q8RUT8).
BACR (bacteriorhodopsin) is a light-driven proton pump. It is one of the first microbial opsins discovered (Oesterhelt and Stoeckenius, Reference Oesterhelt and Stoeckenius1971). BACH (halorhodopsin) is a light-driven chloride pump activated by yellow light (Schobert and Lanyi, Reference Schobert and Lanyi1982). ChR2 (channelrhodopsin 2) is a light-activated sodium channel activated by blue light (Nagel et al., Reference Nagel, Szellas, Huhn, Kateriya, Adeishvili, Berthold, Ollig, Hegemann and Bamberg2003). BACH and ChR2 are among the first optogenetic tools (Han and Boyden, Reference Han and Boyden2007; Zhang et al., Reference Zhang, Wang, Brauner, Liewald, Kay, Watzke, Wood, Bamberg, Nagel, Gottschalk and Deisseroth2007). BACH is used for inhibition, while ChR2 is used for excitation.
I have asked whether retinylidene proteins could be redesigned to be more soluble. Retinylidene proteins are all integral membrane proteins with seven transmembrane alpha helices embedded in a lipid bilayer. Because of the hydrophobic properties of the transmembrane domains, they are not water-soluble without the aid of detergents.
Bacteriorhodopsin has been the subject of several protein-solubilizing studies, though with limited success (Qing et al., Reference Qing, Hao, Smorodina, Jin, Zalevsky and Zhang2022). Recently, researchers leveraged a neural network, SolubleMPNN, which was built upon ProteinMPNN, to engineer soluble variants of bacteriorhodopsin while maintaining its ligand-binding ability and light-sensing function (Nikolaev et al., Reference Nikolaev, Orlov, Tsybrov, Kuznetsova, Shishkin, Kuzmin, Mikhailov, Nikolaeva, Anuchina, Chizhov, Semenov, Kapranov, Borshchevskiy, Remeeva and Gushchin2025).
Instead of taking a computational approach, I apply the QTY (glutamine, threonine, tyrosine) code to systematically engineer water-soluble analogues with reduced hydrophobicity in membrane proteins. There are structural similarities between hydrophobic and polar amino acids: leucine (L) and glutamine (Q); isoleucine (I)/valine (V) and threonine (T); and phenylalanine (F) and tyrosine (Y), as can be observed on high-resolution electron density maps (Tegler et al., Reference Tegler, Corin, Pick, Brookes, Skuhersky, Vogel and Zhang2020; Zhang and Egli, Reference Zhang and Egli2022; Zhang et al., Reference Zhang, Tao, Qing, Tang, Skuhersky, Corin, Tegler, Wassie, Wassie, Kwon, Suter, Entzian, Schubert, Yang, Labahn, Kubicek and Maertens2018). This justifies the replacement of hydrophobic amino acids with polar ones. Zhang et al. initially applied the QTY code to design several detergent-free chemokine receptors, all of which retained structural thermal stability and native ligand-binding and enzymatic activities despite substantial changes to the transmembrane domain (Tegler et al., Reference Tegler, Corin, Pick, Brookes, Skuhersky, Vogel and Zhang2020; Zhang et al., Reference Zhang, Tao, Qing, Tang, Skuhersky, Corin, Tegler, Wassie, Wassie, Kwon, Suter, Entzian, Schubert, Yang, Labahn, Kubicek and Maertens2018). Zhang’s team also applied the QTY code to design water-soluble GPCRs, including chemokine receptors (Qing et al., Reference Qing, Han, Skuhersky, Chung, Badr, Schubert and Zhang2019; Skuhersky et al., Reference Skuhersky, Tao, Qing, Smorodina, Jin and Zhang2021; Tegler et al., Reference Tegler, Corin, Pick, Brookes, Skuhersky, Vogel and Zhang2020; Zhang et al., Reference Zhang, Tao, Qing, Tang, Skuhersky, Corin, Tegler, Wassie, Wassie, Kwon, Suter, Entzian, Schubert, Yang, Labahn, Kubicek and Maertens2018), cytokine receptors (Hao et al., Reference Hao, Jin, Zhang and Qing2020), and olfactory receptors (Johnsson et al., Reference Johnsson, Karagöl, Karagöl and Zhang2025; Skuhersky et al., Reference Skuhersky, Tao, Qing, Smorodina, Jin and Zhang2021). In addition to GPCR, they used the QTY code to design water-soluble glucose transporters (Smorodina et al., Reference Smorodina, Tao, Qing, Jin, Yang and Zhang2022), ABC transporters (Pan et al., Reference Pan, Tao, Smorodina and Zhang2024), monoamine neurotransmitter transporters (Karagöl et al., Reference Karagöl, Karagöl and Zhang2024c), glutamate transporters (Karagöl et al., Reference Karagöl, Karagöl, Smorodina and Zhang2024a, Reference Karagöl, Karagöl and Zhang2024b), mitochondrial megacomplex (Chen and Zhang, Reference Chen and Zhang2025), antibodies (Li et al., Reference Li, Wang, Tao, Xu and Zhang2024b), potassium channels (Smorodina et al., Reference Smorodina, Tao, Qing, Yang and Zhang2024), receptor kinases (Li et al., Reference Li, Tang, Qing, Wang, Liu, Wang, Lyu, Ma, Xu, Zhang and Tao2024a), transmembrane enzymes (Chen et al., Reference Chen, Pan and Zhang2025), as well as bacterial enzymes with beta barrel structures (Sajeev-Sheeja and Zhang, Reference Sajeev-Sheeja and Zhang2024; Sajeev-Sheeja et al., Reference Sajeev-Sheeja, Smorodina and Zhang2023). The QTY-designed water-soluble CXCR4 chemokine receptor was successfully used in biomimetic sensors (Qing et al., Reference Qing, Xue, Zhao, Wu, Breitwieser, Smorodina, Schubert, Azzellino, Jin, Kong, Palacios, Sleytr and Zhang2023).
AlphaFold2 was released by Google DeepMind in July 2021 (Jumper et al., Reference Jumper, Evans, Pritzel, Green, Figurnov, Ronneberger, Tunyasuvunakool, Bates, Žídek, Potapenko, Bridgland, Meyer, Kohl, Ballard, Cowie, Romera-Paredes, Nikolov, Jain, Adler, Back, Petersen, Reiman, Clancy, Zielinski, Steinegger, Pacholska, Berghammer, Bodenstein, Silver, Vinyals, Senior, Kavukcuoglu, Kohli and Hassabis2021). It has greatly facilitated the study of QTY-designed protein variants. Subsequently, AlphaFold3 was released in May 2024, featuring improved architecture and improved efficiency (Abramson et al., Reference Abramson, Adler, Dunger, Evans, Green, Pritzel, Ronneberger, Willmore, Ballard, Bambrick, Bodenstein, Evans, Hung, O’Neill, Reiman, Tunyasuvunakool, Wu, Žemgulytė, Arvaniti, Beattie, Bertolli, Bridgland, Cherepanov, Congreve, Cowen-Rivers, Cowie, Figurnov, Fuchs, Gladman, Jain, Khan, Low, Perlin, Potapenko, Savy, Singh, Stecula, Thillaisundaram, Tong, Yakneen, Zhong, Zielinski, Žídek, Bapst, Kohli, Jaderberg, Hassabis and Jumper2024). Furthermore, AlphaFold3 enabled the accurate prediction of complexes of multiple proteins, as well as complexes with modified residues, nucleic acids, ions, and certain ligands. QTY studies have made use of these new features (Chen and Zhang, Reference Chen and Zhang2025; Johnsson et al., Reference Johnsson, Karagöl, Karagöl and Zhang2025).
GROMACS is a molecular dynamics (MD) simulation program released in 1995 (Berendsen et al., Reference Berendsen, Van Der Spoel and Van Drunen1995). Its 5.0 version was released in 2015 (Abraham et al., Reference Abraham, Murtola, Schulz, Páll, Smith, Hess and Lindahl2015). GROMACS enables an efficient and realistic simulation of biomolecular systems and has been used to investigate the structural and functional properties of QTY-designed protein variants in various studies (Johnsson et al., Reference Johnsson, Karagöl, Karagöl and Zhang2025; Karagöl et al., Reference Karagöl, Karagöl and Zhang2024b; Li et al., Reference Li, Tang, Qing, Wang, Liu, Wang, Lyu, Ma, Xu, Zhang and Tao2024a, Reference Li, Wang, Tao, Xu and Zhang2024b; Smorodina et al., Reference Smorodina, Tao, Qing, Yang and Zhang2024).
In this article, I apply the QTY code to redesign nine human opsins and three microbial opsins. I provide the superpositions of the AlphaFold3-predicted hydrophobic native proteins and their water-soluble QTY variants, and experimentally determined structures when available. I also provide a comparison of the surface hydrophobicity of the variants. Furthermore, I run MD simulations of native and QTY-designed OPN2 and analyze their response to the 11-cis to all-trans isomerization of the retinal chromophore.
Methods
QTY design, protein sequence alignment, and other characteristics
The native protein sequences for OPN1MW, OPN1LW, OPN1SW, OPN2, OPN3, OPN4, OPN5, RGR, RRH, BACR, BACH, and ChR2 were obtained from UniProt (https://www.uniprot.org). The sequence for ChR2 was truncated according to the experimentally determined structure in the RCSB PDB (PDB ID: 8ZAN). The QTY designs were performed through the protein solubilizing server (https://pss.sjtu.edu.cn/) (Tao et al., Reference Tao, Tang, Zhang, Li and Xu2022). For ChR2, the secondary structure was provided to the server in SS3 format according to the RCSB PDB structure.
AlphaFold3 predictions
I predicted the structures of native proteins and their QTY variants using the AlphaFold3 website (https://alphafoldserver.com) (Abramson et al., Reference Abramson, Adler, Dunger, Evans, Green, Pritzel, Ronneberger, Willmore, Ballard, Bambrick, Bodenstein, Evans, Hung, O’Neill, Reiman, Tunyasuvunakool, Wu, Žemgulytė, Arvaniti, Beattie, Bertolli, Bridgland, Cherepanov, Congreve, Cowen-Rivers, Cowie, Figurnov, Fuchs, Gladman, Jain, Khan, Low, Perlin, Potapenko, Savy, Singh, Stecula, Thillaisundaram, Tong, Yakneen, Zhong, Zielinski, Žídek, Bapst, Kohli, Jaderberg, Hassabis and Jumper2024). For microbial opsins, the dimers and trimers have identical subunits, that is, are homodimers and homotrimers. In AlphaFold3, this was achieved by altering the number of protein copies.
Structure superpositions
PDB files for native protein structures determined experimentally by X-ray diffraction or electron microscopy were taken from RCSB PDB: OPN2 (PDB ID: 5W0P), BACR (PDB ID: 7XJC), BACH (PDB ID: 2JAF), ChR2 (PDB ID: 8ZAN). The AlphaFold3-predicted native and QTY variants were taken directly from the most probable predicted structure. Superpositions were performed via the command ‘super’ in PyMOL (https://pymol.org). I removed unstructured loops at the N and C terminals for the sake of clarity.
Structure visualization
I used PyMOL (https://pymol.org) to superpose the native predicted protein structures, their QTY variants, and the experimentally determined structures for the proteins where these existed. I then used UCSF ChimeraX (https://www.cgl.ucsf.edu/chimerax/) to render each protein model with hydrophobicity patches.
Molecular dynamics simulations
All MD simulations and analyses were executed on a desktop computer with Intel Xeon Platinum 8352 V Processor, 256 GB RAM, and 2 NVIDIA GPUs (GeForce RTX 4090) with 24 GB VRAM each. All MD simulations were conducted using GROMACS 2024.5 (Abraham et al., Reference Abraham, Murtola, Schulz, Páll, Smith, Hess and Lindahl2015) with the CHARMM36m all-atom force field (Huang et al., Reference Huang, Rauscher, Nawrocki, Ran, Feig, de Groot, Grubmüller and MacKerell2017). Data for retinal were obtained from NAMD Wiki (https://www.ks.uiuc.edu/Research/namd/wiki/index.cgi?RetinalTop for topology and https://www.ks.uiuc.edu/Research/namd/wiki/index.cgi?RetinalPar for parameters) and manually integrated into the protein file. Initial structures, configuration files, and command-line codes for the simulations are publicly available on Zenodo (https://doi.org/10.5281/zenodo.15377505).
Regarding native OPN2, the protein membrane system was constructed using the web-based membrane builder CHARMM-GUI and downloaded in GROMACS format (Jo et al., Reference Jo, Kim, Iyer and Im2008; Lee et al., Reference Lee, Cheng, Swails, Yeom, Eastman, Lemkul, Wei, Buckner, Jeong, Qi, Jo, Pande, Case, Brooks, MacKerell, Klauda and Im2016; Wu et al., Reference Wu, Cheng, Jo, Rui, Song, Dávila-Contreras, Qi, Lee, Monje-Galvan, Venable, Klauda and Im2014). The protein was centered in a rectangular box. The generated membrane models consisted of 40% POPC, 40% POPE, 10% POPS, and 10% cholesterol, which simulated a rod photoreceptor disk membrane where native OPN2 is usually expressed (Albert and Boesze-Battaglia, Reference Albert and Boesze-Battaglia2005). The system was solvated in TIP3P water with 150 mM NaCl. NaCl was used instead of KCl to simulate the environment of rod photoreceptor outer segments (Govardovskii, Reference Govardovskii1971). 11-cis-Retinal was manually added to the protein files. LINCS constraints were used for constraints, and the Verlet integrator was used. Electrostatics was handled with particle mesh Ewald (PME), with both Coulomb and van der Waals interaction cutoffs set at 1.2 nm. The energy of the system was minimized using the steepest descent until the maximum forces converged below 1000 kJ/mol/nm. The standard six-step CHARMM-GUI NP
$ \gamma $
T equilibration protocol (Jo et al., Reference Jo, Kim, Iyer and Im2008) was used, with two 125-ps NVT equilibration simulations, and one 125-ps and three 500-ps NP
$ \gamma $
T equilibration simulations. Temperature and pressure were maintained at 303.15 K and 1.0 bar, respectively, using the V-rescale thermostat and C-rescale barostat with surface tension coupling. Following equilibration, a 10-ns production MD simulation was run. The retinal molecule was then manually switched to the all-trans state by rotating C13 to C15 and associated hydrogen and methyl groups by 180 degrees around the C11–C12 bond. This mimicked the effect of the absorption of a photon by 11-cis-retinal. Energy minimization and equilibration were repeated exactly as before, and then a 120-ns production MD simulation was run.
Regarding QTY-designed OPN2, I found that helix 9 of the QTY analogue flung up and adhered to the surface that was originally the transmembrane domain, which disrupted the normal conformation. Therefore, this helix was truncated. The system was constructed directly using GROMACS. The protein was centered in a rectangular box with dimensions equal to those of native OPN2. The equilibration involved a sequence of simulations: one 250-ps NVT equilibration, followed by one 125-ps and one 1500-ps NPT equilibration, totaling an equilibration duration that matches that of the natural OPN2. All other configurations and parameters were identical to those of the original OPN2.
After the simulations, I extracted frames from the trajectory to inspect the structures before and after retinal isomerization. I also calculated the RMSD (‘gmx rms’ command), number of hydrogen bonds (‘gmx hbond’ command) and water molecules (‘gmx select’ command) inside the binding pocket, interaction energy (including short-range Coulombic interaction energy and short-range Lennard-Jones energy) (‘gmx energy’ command), the RMSF (‘gmx rmsf’ command), and radius of gyration (‘gmx gyrate’ command). The detailed commands are publicly available in the online database for this study.
Results and discussion
The rationale of the QTY code
The transmembrane segments of membrane proteins are hydrophobic in nature. This leads to a challenge to study their structure and function. Traditionally, detergent must be used to purify them. Zhang et al. considered a different approach, that is, via systematic soluble protein design. There exist structural similarities, as can be observed on high-resolution electron density maps, between certain hydrophobic and hydrophilic amino acids: leucine (L) and glutamine (Q), isoleucine (I) or valine (V) and threonine (T), and phenylalanine (F) with tyrosine (Y). This fact enables systematic replacement of L with Q, I/V with T, and F with Y in all transmembrane segments of the membrane protein, which Zhang et al. named the QTY code. QTY analogues are less hydrophobic than native membrane proteins. Although their amino acid sequences are significantly changed, they still exhibit relatively preserved structure, isoelectric points (pI), and molecular weights (MW) compared to the native membrane proteins (Table 1).
Table 1. Protein characteristics of 12 retinylidene proteins and their QTY analogs

MW, molecular weight; pI, isoelectric focusing; RMSD, root-mean-square distance; TM, transmembrane.
Protein sequence alignments and other characteristics of retinylidene proteins
The protein sequences of nine human opsins and three microbial opsins were aligned with their QTY analogues (Figure 1). The QTY substitution resulted in an overall variation of the sequence of 15.48% to 30.04% and a variation of the transmembrane domain of 35.53% to 50.24%. Despite changes in amino acid composition and sequence, the pI only experienced a slight change, between 0.00 and 0.16. This is because the amino acids Q, T, and Y have neutral charges and do not introduce additional charges to the system. The MW increased slightly by a value between 0.04 and 0.60 kDa. This is because the hydrophilic amino acids usually have nitrogen or oxygen atoms in place of carbon atoms in hydrophobic acids and thus may have higher mass: L (131.17 Da) versus Q (146.14 Da); I (131.17 Da) and V (117.15 Da) versus T (119.12 Da); F (165.19 Da) versus Y (181.19 Da).

Figure 1. The protein sequence alignments of 12 retinylidene proteins and their QTY analogs. Blue tables: human opsins; purple tables: microbial opsins. The symbols
$ \mid $
and
$ \ast $
indicate that amino acids are identical or different, respectively. Amino acids L, I/V, and F in TM (transmembrane) alpha helices (shown in blue above the sequences) are replaced with Q, T, and Y, respectively. The variation in the TM domain ranges from 35.53% to 50.24%, while the overall variation rate ranges from 15.48% to 30.04%. The characteristics of the proteins are shown above the sequences. Despite large variation rates, the pI only experiences a slight change between 0.00 and 0.16 and the MW increases slightly by a value between 0.04 and 0.60 kDa. The enlarged individual sequence alignments are available in supplementary information. The alignments are (a) OPN1MW versus OPN1MW
$ {}^{\mathrm{QTY}} $
, (b) OPN1LW versus OPN1LW
$ {}^{\mathrm{QTY}} $
, (c) OPN1SW versus OPN1SW
$ {}^{\mathrm{QTY}} $
, (d) OPN2 versus OPN2
$ {}^{\mathrm{QTY}} $
, (e) OPN3 versus OPN3
$ {}^{\mathrm{QTY}} $
, (f) OPN4 versus OPN4
$ {}^{\mathrm{QTY}} $
, (g) OPN5 versus OPN5
$ {}^{\mathrm{QTY}} $
, (h) RGR versus RGR
$ {}^{\mathrm{QTY}} $
, (i) RRH versus RRH
$ {}^{\mathrm{QTY}} $
, (j) BACR versus BACR
$ {}^{\mathrm{QTY}} $
, (k) BACH versus BACH
$ {}^{\mathrm{QTY}} $
, and (l) ChR2 versus ChR2
$ {}^{\mathrm{QTY}} $
.
Superpositions of AlphaFold3-predicted native human opsins, their water-soluble QTY variants, and experimentally determined structures
Although the chromophore, retinal, is bound to retinylidene proteins in an internal pocket, it is only incorporated into the protein after protein folding is complete. In other words, protein folding does not depend on the chromophore. Thus, it is justified to predict the ligand-free forms of the proteins. I used AlphaFold3 to predict the structure of the nine native human opsins and their water-soluble QTY analogues. The structures superposed very well (Figure 2a–i). For OPN2, which has an available X-ray diffraction structure, I also superposed the experimentally determined structure with the AlphaFold3-predicted structures. Overall, the root mean square distances (RMSD) were small, from 0.307 to 0.611 Å, with only one exception of OPN2
$ {}^{\mathrm{QTY}} $
versus OPN2
$ {}^{\mathrm{EXP}} $
(RMSD = 0.999 Å). This shows that the water-soluble QTY analogues are quite similar to native proteins in terms of structure.

Figure 2. Superposition of AlphaFold3-predicted native human retinylidene proteins, their QTY analogs, and experimentally determined structures. For clarity, unstructured N- and C-terminal ends are deleted. For (a) to (i), despite significant changes in the protein sequence, the structures superpose very well. The root-mean-square distance (RMSD) values are quite small, from 0.307 to 0.611 Å, with only one exception (OPN2
$ {}^{\mathrm{QTY}} $
vs. OPN2
$ {}^{\mathrm{EXP}} $
, RMSD = 0.999 Å). Green: AlphaFold3-predicted native structure; cyan: AlphaFold3-predicted QTY analog structure; magenta: experimentally determined structure. The superpositions are (a) OPN1MW
$ {}^{\mathrm{AF}3} $
versus OPN1MW
$ {}^{\mathrm{QTY}} $
, (b) OPN1LW
$ {}^{\mathrm{AF}3} $
versus OPN1LW
$ {}^{\mathrm{QTY}} $
, (c) OPN1SW
$ {}^{\mathrm{AF}3} $
versus OPN1SW
$ {}^{\mathrm{QTY}} $
, (d) OPN2
$ {}^{\mathrm{AF}3} $
versus OPN2
$ {}^{\mathrm{QTY}} $
versus OPN2
$ {}^{\mathrm{EXP}} $
, (e) OPN3
$ {}^{\mathrm{AF}3} $
versus OPN3
$ {}^{\mathrm{QTY}} $
, (f) OPN4
$ {}^{\mathrm{AF}3} $
versus OPN4
$ {}^{\mathrm{QTY}} $
, (g) OPN5
$ {}^{\mathrm{AF}3} $
versus OPN5
$ {}^{\mathrm{QTY}} $
, (h) RGR
$ {}^{\mathrm{AF}3} $
versus RGR
$ {}^{\mathrm{QTY}} $
, and (i) RRH
$ {}^{\mathrm{AF}3} $
versus RRH
$ {}^{\mathrm{QTY}} $
. For (j) and (k), there is a large degree of similarity between the RMSD between a pair of native proteins and that between the corresponding pair of QTY analogs. Green: OPN1MW; red: OPN1LW; blue: OPN1SW; purple: OPN2; cyan: OPN3; gray: OPN4; olive: OPN5; orange: RGR; pink: RRH. The superpositions are (j) OPN1MW
$ {}^{\mathrm{AF}3} $
versus OPN1LW
$ {}^{\mathrm{AF}3} $
versus OPN1SW
$ {}^{\mathrm{AF}3} $
versus OPN2
$ {}^{\mathrm{AF}3} $
versus OPN3
$ {}^{\mathrm{AF}3} $
versus OPN4
$ {}^{\mathrm{AF}3} $
versus OPN5
$ {}^{\mathrm{AF}3} $
versus RGR
$ {}^{\mathrm{AF}3} $
versus RRH
$ {}^{\mathrm{AF}3} $
and (k) OPN1MW
$ {}^{\mathrm{QTY}} $
versus OPN1LW
$ {}^{\mathrm{QTY}} $
versus OPN1SW
$ {}^{\mathrm{QTY}} $
versus OPN2
$ {}^{\mathrm{QTY}} $
versus OPN3
$ {}^{\mathrm{QTY}} $
versus OPN4
$ {}^{\mathrm{QTY}} $
versus OPN5
$ {}^{\mathrm{QTY}} $
versus RGR
$ {}^{\mathrm{QTY}} $
versus RRH
$ {}^{\mathrm{QTY}} $
.
All nine of these human opsins belong to the same large family, that is, animal opsins. However, they belong to different subfamilies and have structural differences from each other. In order to investigate the extent to which the water-soluble QTY analogues preserve these differences, I also conducted pairwise superpositions within all nine native human opsins and within all nine QTY analogues of human opsins (Figure 2j and k). There was a large degree of similarity between the RMSD between a pair of native proteins and that between the corresponding pair of QTY analogues. Therefore, I conclude that the structural differences are relatively well preserved before and after the QTY substitution. However, whether the functional differences are preserved remains a problem to study.
Superpositions of AlphaFold3-predicted native microbial opsins, their water-soluble QTY variants, and experimentally determined structures
I used AlphaFold3 to predict the structure of monomers of the three native microbial opsins, their water-soluble QTY analogues. Since microbial opsins often have to form oligomers to be functional, I also predicted the structure of native and QTY variants of BACR trimers, BACH trimers, and ChR2 dimers. These predicted structures were also superposed with experimentally determined structures (X-ray diffraction or electron microscopy). The structures superposed very well (Figure 3). Overall, the root-mean-square distances (RMSD) were small, with the highest value being 0.685 Å. This shows that the water-soluble QTY analogues are likely capable of forming oligomers with structures similar to those of native proteins.

Figure 3. Superposition of AlphaFold3-predicted native microbial retinylidene proteins, their QTY analogs, and experimentally determined structures. Despite significant changes in the protein sequence, the structures superpose very well. The root-mean-square distance (RMSD) values are quite small, with the highest being 0.685 Å. For clarity, unstructured N- and C-terminal ends are deleted. Green: AlphaFold3-predicted native structure; cyan: AlphaFold3-predicted QTY analog structure; magenta: experimentally determined structure. The superpositions are (a) BACR
$ {}^{\mathrm{AF}3} $
versus BACR
$ {}^{\mathrm{QTY}} $
versus BACR
$ {}^{\mathrm{EXP}} $
monomer, (b) BACR
$ {}^{\mathrm{AF}3} $
versus BACR
$ {}^{\mathrm{QTY}} $
versus BACR
$ {}^{\mathrm{EXP}} $
trimer, (c) BACH
$ {}^{\mathrm{AF}3} $
versus BACH
$ {}^{\mathrm{QTY}} $
versus BACH
$ {}^{\mathrm{EXP}} $
monomer, (d) BACH
$ {}^{\mathrm{AF}3} $
versus BACH
$ {}^{\mathrm{QTY}} $
versus BACH
$ {}^{\mathrm{EXP}} $
trimer, (e) ChR2
$ {}^{\mathrm{AF}3} $
versus ChR2
$ {}^{\mathrm{QTY}} $
versus ChR2
$ {}^{\mathrm{EXP}} $
monomer, and (f) ChR2
$ {}^{\mathrm{AF}3} $
versus ChR2
$ {}^{\mathrm{QTY}} $
versus ChR2
$ {}^{\mathrm{EXP}} $
dimer.
Analysis of the hydrophobic surface of native retinylidene proteins and their water-soluble QTY variants
As transmembrane proteins, retinylidene proteins require detergents to be separated from the lipid bilayer. The detergents act as an interface between the hydrophobic transmembrane domain of the protein and the surrounding water, thus solubilizing the protein. If these proteins are exposed in water without detergents, their hydrophobicity will induce them to aggregate and precipitate, with their structures and functions disrupted.
The hydrophobic portions of the protein surface are represented in yellow (Figure 4). In the native proteins, there are large hydrophobic patches in the transmembrane domains, which are embedded within the lipid bilayer. The nonpolar and hydrophobic amino acids, including leucine (L), isoleucine (I), valine (V), phenylalanine (F), alanine (A), methionine (M), and tryptophan (W), interact with the hydrophobic hydrocarbon chains and expel water.

Figure 4. Hydrophobic surface of 12 retinylidene proteins and their water-soluble QTY analogs. Hydrophobic patches are shown in yellow, while hydrophilic patches are shown in cyan. The native proteins have many hydrophobic patches due to the presence of hydrophobic amino acids, including L, I, V, and F. After QTY substitution, hydrophilic Q, T, and Y have respectively replaced hydrophobic L, I/V, and F, and the hydrophobic patches in the surface of transmembrane helices have become more hydrophilic. In addition, the surface shape of the native and QTY analogs are very similar. For clarity, unstructured N- and C-terminal ends are deleted. The comparisons are (a) OPN1MW versus OPN1MW
$ {}^{\mathrm{QTY}} $
, (b) OPN1LW versus OPN1LW
$ {}^{\mathrm{QTY}} $
, (c) OPN1SW versus OPN1SW
$ {}^{\mathrm{QTY}} $
, (d) OPN2 versus OPN2
$ {}^{\mathrm{QTY}} $
, (e) OPN3 versus OPN3
$ {}^{\mathrm{QTY}} $
, (f) OPN4 versus OPN4
$ {}^{\mathrm{QTY}} $
, (g) OPN5 versus OPN5
$ {}^{\mathrm{QTY}} $
, (h) RGR versus RGR
$ {}^{\mathrm{QTY}} $
, (i) RRH versus RRH
$ {}^{\mathrm{QTY}} $
, (j) BACR versus BACR
$ {}^{\mathrm{QTY}} $
monomer, (k) BACH versus BACH
$ {}^{\mathrm{QTY}} $
monomer, (l) ChR2 versus ChR2
$ {}^{\mathrm{QTY}} $
monomer, (m) BACR versus BACR
$ {}^{\mathrm{QTY}} $
trimer, (n) BACH versus BACH
$ {}^{\mathrm{QTY}} $
trimer, and (o) ChR2 versus ChR2
$ {}^{\mathrm{QTY}} $
dimer.
I investigated the surface hydrophobicity of water-soluble QTY analogues in comparison with the native proteins. After applying the QTY code to, respectively, replace the hydrophobic amino acids L, I/V, and F with hydrophilic amino acids glutamine (Q), threonine (T), and tyrosine (Y), the hydrophobic surface areas were significantly reduced. More importantly, since the electron density maps of the corresponding amino acids are similar, the alpha helices in QTY analogues had a similar contour to those in the native proteins. The QTY analogues successfully retained their structural integrity and stability.
AlphaFold3 predictions
DeepMind released AlphaFold3 in May 2024, marking a significant leap in accuracy for modeling across biomolecular space. This latest iteration outperforms state-of-the-art docking tools and its predecessor, AlphaFold-Multimer v.2.3, in protein structure and protein–protein interaction predictions (Abramson et al., Reference Abramson, Adler, Dunger, Evans, Green, Pritzel, Ronneberger, Willmore, Ballard, Bambrick, Bodenstein, Evans, Hung, O’Neill, Reiman, Tunyasuvunakool, Wu, Žemgulytė, Arvaniti, Beattie, Bertolli, Bridgland, Cherepanov, Congreve, Cowen-Rivers, Cowie, Figurnov, Fuchs, Gladman, Jain, Khan, Low, Perlin, Potapenko, Savy, Singh, Stecula, Thillaisundaram, Tong, Yakneen, Zhong, Zielinski, Žídek, Bapst, Kohli, Jaderberg, Hassabis and Jumper2024). AlphaFold3 reduces the reliance on multiple sequence alignment by integrating a diffusion-based model, enabling it to predict a broader spectrum of biomolecules, including ligands, ions, nucleic acids, modified residues, and large protein megacomplexes. On October 9, 2024, DeepMind’s founders, Demis Hassabis and John Jumper, received the Nobel Prize in Chemistry for revolutionizing protein structure prediction through leveraging machine learning.
AlphaFold3 is easily accessible online (https://alphafoldserver.com), allowing users to make 30 predictions a day at present. The structures of the QTY analogues were predicted using the AlphaFold3 server, which was free of charge, and the results were produced within a few minutes.
DeepMind also collaborated with the EBI to make over 214 million predicted protein structures available through the AlphaFold Protein Structure Database (https://alphafold.ebi.ac.uk). This number is continuously expanding, with the quality of predictions further improving with the advent of AlphaFold 3.
However, AlphaFold3 still has limitations that have been encountered in this study. Although AlphaFold3 allows the structural prediction of complexes that include small-molecule ligands, it only supports a small range of ligands, which does not include retinal. I was therefore unable to predict the structure of chromophore-bound states of retinylidene proteins with AlphaFold3. Consequently, I chose another approach, namely molecular dynamics simulation, in order to investigate the behavior of QTY opsin analogues when they are bound to the chromophore.
Simulation of 11-cis to all-trans isomerization of retinal in native OPN2 and its QTY analogue
QTY substitutions do not introduce changes to the essential NPxxY motif or the retinal-binding lysine of human retinylidene proteins. Therefore, I proposed that QTY analogues of human opsins may have conserved functions and carried out a molecular dynamics (MD) simulation. I recorded and analyzed the behavior of native OPN2 and its QTY analogue before and after the 11-cis to all-trans isomerization. Note that my aim is not to simulate the whole photocycle of OPN2, which takes longer than 1 ms (Fanelli et al., Reference Fanelli, Felline and Marigo2021), but rather to find proof that the QTY analogue conserves the function of being activated by isomerization of retinal.
The MD simulation provided evidence of conformational change in both native and QTY analogues (Figure 5). The native protein experienced a slower change from its original conformation, while the QTY analogue showed more fluctuations and sudden changes. Under further inspection, this phenomenon might be due to the entrance of water molecules into the QTY analogue. In addition, the retinal molecule had a greater degree of change in conformation in the native protein than in the QTY analogue. Nevertheless, the retinal in the QTY analogue was still able to cause changes to surrounding residues, sometimes doing so via intervening water molecules. It may be noted that the RMSD was still fluctuating, which means that the protein had not yet stabilized. This was expected, since the OPN2 was still in the process of transition from BATH to LUMI state, and this provided evidence for the activation of the protein in response to isomerization of retinal.

Figure 5. The conformational changes of native OPN2 and its QTY analog before and after 11-cis to all-trans isomerization of the chromophore, retinal. (a, b) 1 ns running averages of the root-mean-square distances (RMSD) of the protein–retinal complex, transmembrane helix 6 (TM6), the retinal-binding pocket, and retinal. By convention, the isomerization is set at time 0 ns, which is indicated by a brown, vertical dashed line. (c) Superpositions between cis-state OPN2, trans-state OPN2, cis-state OPN2
$ {}^{\mathrm{QTY}} $
, and trans-state OPN2
$ {}^{\mathrm{QTY}} $
. Both OPN2 and OPN2
$ {}^{\mathrm{QTY}} $
exhibit conformational changes, with RMSDs greater than 2 Å. Blue: cis-state protein, orange: 11-cis-retinal; yellow: trans-state protein; greenish cyan: all-trans retinal.
Zooming in on the retinal-binding pocket, I observed that retinal is held within its binding pocket by several forces in both the native and QTY analogue: the counterions of the retinal Schiff base (RSB) (113E and 181E), the hydrophobic interactions with neighboring residues (e.g., 265W, 268Y, and the 187–189 beta strand), and, sometimes, steric collisions (Figure 6). Since retinal is a ligand that binds in an internal pocket, I was unable to calculate the Gibbs free energy of its binding. Nevertheless, the distributions of interaction energies were found to be similar in native OPN2 and the water-soluble QTY analogue, except for small differences: native OPN2 experienced a change in interaction energy at about
$ t=50 $
ns due to the entrance of water near 113E; the QTY analogue had more negative interaction energy at residues 208 and 212 due to phenylalanine (F) to tyrosine (Y) substitutions. Aside from these, the similarity was a strong suggestion of functional conservation. Finally, I observed very different hydrogen bonding and water molecule patterns inside the binding pocket, which was mainly due to several F to Y substitutions in that region. This might cause retinal to interact with adjacent residues in a subtly different way, and might lead to a different absorption peak since the environment of the RSB, that is, the binding pocket, determines the spectral characteristic of opsins (Fenno et al., Reference Fenno, Yizhar and Deisseroth2011).

Figure 6. Changes in the retinal-binding pocket and protein–ligand interaction in native OPN2 and its QTY analog before and after 11-cis to all-trans isomerization of the chromophore. (a, b) Close-ups of the binding pocket in cis-state. Protein–ligand interactions with lengths shorter or equal to 3.5 Å are shown in the figure. Blue: protein residues, orange: 11-cis-retinal; green dashed lines: ion bridge; yellow dashed lines: van der Waals and/or hydrophobic interactions. (c, d) Close-ups of the binding pocket in trans-state. Protein–ligand interactions with lengths shorter or equal to 3.5 Å are shown in the figure. Yellow: protein residues, greenish cyan: all-trans-retinal; green dashed lines: ion bridge; yellow dashed lines: van der Waals and/or hydrophobic interactions. (e, f) The interaction energies (IEs) between the protein, the binding pocket, individual residues, and retinal. IE is the sum of the short-range Coulombic interaction energy and short-range Lennard–Jones energy. Note that IE is a product of MD simulation and is not necessarily a ‘real’ physical quantity. For clarity, IE is rescaled using the signed pseudo logarithm (
$ y=\operatorname{sign}(x)\cdot \ln \left(|x|+1\right) $
). By convention, the isomerization is set at time 0 ns, which is indicated by a brown, vertical dashed line. The large changes in IE in OPN2 around
$ t=50 $
ns are due to the entrance of water molecules into the binding pocket, near 113E. The QTY analog has more negative IE at residues 208 and 212 due to F to Y substitutions. Besides from these, the similarity between the IE of OPN2 and OPN2
$ {}^{\mathrm{QTY}} $
is a strong suggestion of functional conservation. (g, h) The number of hydrogen bonds formed between residues in the binding pocket, and the number of water molecules within the binding pocket. By convention, the isomerization is set at time 0 ns, which is indicated by a brown, vertical dashed line.
Overall, I observed that QTY-designed OPN2 had relatively similar behavior to the native variant despite certain small differences caused by decreased hydrophobicity. The function of OPN2
$ {}^{\mathrm{QTY}} $
awaits to be verified by experiments.
Future scopes and potential applications
The findings of this study indicate that the QTY code could be used as a robust tool to design water-soluble retinylidene proteins.
This could, in the first place, facilitate the study of these proteins. The water solubility of QTY analogues makes it easier to purify and investigate proteins, especially those such as RGR and RRH, which have not been studied much but may have considerable clinical relevance. The water solubility also reduces the difficulty in recording the different phases of the photocycle, in comparison with native membrane opsins. Another application may be the rapid design of new optogenetic tools.
Furthermore, water-soluble QTY opsins could be useful clinically, since they could be delivered and/or expressed in the eye without forming aggregates and precipitating. It is not unreasonable to hypothesize that QTY opsins can still pass through the photocycle and interact with downstream signaling proteins. In light of this, QTY opsins may facilitate the optogenetic therapy of ophthalmological diseases (Sakai et al., Reference Sakai, Tomita and Maeda2022), providing a potential pathway in restoring basic vision for those who have lost it.
Finally, water-soluble opsins have the potential to be harvested in mass and may be used to design new biomimetic light-sensing systems, which may have applications in bioengineering.
Conclusion
In this study, I selected 12 retinylidene proteins, including 9 human opsins (OPN1MW, OPN1LW, OPN1SW, OPN2, OPN3, OPN4, OPN5, RGR, and RRH) and 3 microbial opsins (BACR, BACH, and ChR2), that had clinical implications and various potential applications. I applied the QTY code to convert the hydrophobic transmembrane alpha helices into hydrophilic ones and thus create water-soluble QTY analogues of the proteins. Then, I used AlphaFold3 to predict the structures of the native proteins as well as the QTY analogues, superposed them, and found that the structure of the proteins was well conserved despite substantial residue substitutions. I also inspected the surface hydrophobicity of the proteins and found a great decrease in hydrophobic patches on the transmembrane surface. Next, I chose a representative retinylidene protein, OPN2, and employed molecular dynamics simulations to investigate the behavior of the native and QTY analogue of OPN2. I found that the QTY analogue was capable of conformational change and effective protein–ligand interaction. These findings revealed that QTY-designed OPN2 and retinylidene proteins in general might have conserved structure and function. I believe that these water-soluble QTY variants may have potential in research, therapeutic treatments, and bioengineering.
Open peer review
To view the open peer review materials for this article, please visit http://doi.org/10.1017/qrd.2025.10009.
Supplementary material
The supplementary material for this article can be found at http://doi.org/10.1017/qrd.2025.10009.
Data availability statement
The data for the AlphaFold3-predicted structures and MD simulation setting files and commands can be found on Zenodo (https://doi.org/10.5281/zenodo.15377505).
Acknowledgements
S.P. thanks Prof. Shuguang Zhang of the Lab of Molecular Architecture, Media Lab, Massachusetts Institute of Technology, for inspiring the topic of this study as well as providing valuable advice on research methods and manuscript preparation.
Author contribution
Conceptualization, formal analysis, investigation, methodology, validation, data curation, writing—original draft preparation, review and editing: S.P.
Financial support
There is no financial support for this digital bioinformatics study. I only used free tools that are publicly available.
Competing interests
Massachusetts Institute of Technology (MIT) filed several patent applications for the QTY code for GPCRs, which do not include the human opsins. There are no competing interests.
Ethics statement
All methods were carried out in accordance with relevant guidelines and regulations. Neither human biological samples nor human subjects were used in this study. This is a completely digital structural bioinformatic study using the publicly available AlphaFold3 machine learning program and GROMACS molecular dynamics simulation program.
Comments
Dear Editor,
This study is part of a series of studies on protein design with the QTY code by Prof. Shuguang Zhang and his team. I extended QTY-design to a new family of proteins, the retinylidene proteins.
I completed this manuscript individually while maintaining close communication with Prof. Zhang. He has recommended this manuscript to Prof. Bengt Nordén, who has encouraged me to submit it to QRB Discovery.
Thank you very much for considering my work. I look forward to your response.
Sincerely,
Siqi Pan
Email: siqipan2008@outlook.com