In Silico Discovery of Pharmacological Chaperones of P23H Rhodopsin to Treat Retinitis Pigmentosa

Thanks for watching! Stop by the fair to learn more.

Purpose

I have retinitis pigmentosa (RP), a group of inherited retinal diseases that progressively destroy photoreceptors. People affected by RP often first notice reducing night vision, followed by narrowing peripheral vision, leading to total blindness. RP affects more than two million people worldwide, but there's no cure.

My form is X-linked, but for this project I selected a different variant called P23H. P23H is the most common single mutation responsible for autosomal dominant RP in North America, accounting for thousands of cases. The mutation results from a single nucleotide substitution in the DNA, which alters one amino acid in rhodopsin, the photopigment that initiates vision in rod cells. The substitution causes rhodopsin to misfold. The endoplasmic reticulum quality control system identifies the defective protein and destroys it. The resulting loss of functional rhodopsin leads to photoreceptor cell death and the gradual loss of vision.

To date, only one compound has been shown to rescue misfolded P23H rhodopsin: 9-cis-retinal, but it has low therapeutic potential. The molecule is photolabile, degrading rapidly when exposed to visible light. The products of the molecule breaking down are cytotoxic. A photoreceptor-targeted drug that decomposes upon illumination has no clinical utility.

Hypothesis

If a light-stable small molecule binds the retinal pocket of P23H rhodopsin, it will reduce structural instability at residue 23 compared to the unbound mutant protein.

This project aimed to identify such a compound, suitable for eventual treatment of RP. A successful approach for P23H could also be adapted to other protein misfolding disorders.

Variables

Independent: Identity of the bound small molecule.

Dependent: Binding persistence and RMSF at residue 23 and the N-terminal.

Controlled: Protein structure, membrane environment, temperature (310 K), and all simulation parameters.

A computational pipeline with two stages was built for this project.

Stage 1: Molecular docking

Molecular docking is a computational method of predicting whether a small molecule can fit inside a protein's binding pocket. The screen was conducted with AutoDock Vina and 17,375 drug-like compounds drawn from the ChEMBL database. Each compound received a score and was ranked based on how well it fit.

An error was discovered during the first screening. The search box was originally centered around the middle of the protein instead of the retinal binding pocket. The pocket coordinates were recalculated from five conserved binding-site residues, and the full library was re-screened.

Stage 2: Molecular dynamics with a virtual cell membrane

Docking shows that a ligand fits, but not whether it stabilizes the protein. The top candidates were therefore advanced into molecular dynamics simulations, which track every atom of the protein, the drug, 153 lipid molecules, and thousands of water molecules. All atoms move according to Newtonian mechanics at body temperature (37 °C). These simulations were performed with GROMACS.

Matched-seed design

This experiment compared three systems: healthy rhodopsin, P23H rhodopsin on its own, and P23H with the lead compound bound. All three runs started from the same coordinates, used the same random seed, and ran on the same GPU. The drug was the only thing that changed between them. Any difference in behavior therefore had to come from the drug itself.

Running time

Each primary run covered 500 nanoseconds and took four to five days on a GTX 1080 Ti GPU to complete. The total simulation time across all tests now exceeds 8 microseconds.

Finding 1:

Docking score alone is not enough. The top-scoring docking candidate, CHEMBL9532, outperformed the natural chaperone 9-cis-retinal on paper. In simulation, however, it destabilized the protein, performing 63% worse than P23H alone. The compound bound tightly but twisted the helices out of shape. Stopping at the docking stage would have produced a confident but completely wrong result. The simulation acted as the reality check.

Finding 2:

hnRP94450 eliminates catastrophic instability at the mutation site. The main experiment compared three 500-nanosecond matched-seed simulations: healthy rhodopsin, P23H alone, and P23H with hnRP94450. This result was reproduced across three independent replicates using different random seeds. Wobble at the mutation site (residue 23) was measured across 50 independent 10-nanosecond time windows for each system.

Healthy rhodopsin: 0.56 ± 0.04 Å average. Never spiked above 0.8 Å in 500 nanoseconds.

P23H alone: 0.92 ± 0.26 Å average. Spiked above 1.1 Å in 7 of 50 windows. Above this threshold, the protein samples conformations likely to be flagged by ER quality control for destruction.

P23H + drug: 0.77 ± 0.09 Å average. Spiked above 1.1 Å in 0 of 50 windows, matching the healthy protein exactly.

The disease signal gives a Cohen’s d of 1.92 and a Wilcoxon signed-rank p-value of 1.8 × 10⁻¹⁵. The drug rescue gives a Mann–Whitney p-value of 7.03 × 10⁻⁴.

The drug also reduces the standard deviation by a factor of three, meaning the dynamics become predictable and safe.

Across all replicates, the effect remains strong, with a Levene test p-value of 5.1 × 10⁻⁵ and a spike rate reduction from 15% to 3%.

Finding 3:

The drug’s effect is specific. Rhodopsin has another flexible region far from the mutation, called ECL2. A nonspecific drug that simply rigidified the protein would stabilize ECL2 as well. It does not (p = 0.35, not significant). The compound only stabilizes the mutation region, which is the expected behaviour of a pharmacological chaperone. This specificity gradient, with strong rescue at the mutation site and none at ECL2, is indicative of a strong candidate chaperone.

Finding 4:

The pipeline independently found the same drug class as published research. hnRP94450 is a chromenone, the same molecular family as CR5, a compound independently identified as a P23H chaperone in a 2022 Human Molecular Genetics paper [Athanasiou et al, 2022]. Of the 17,375 compounds in the screen, 93 chromenones appeared in the top hits. A computational approach arrived at the same answer as published cell-biology research, providing cross-validation for both methods.

Finding 5:

Global protein drift is also rescued. Beyond the mutation site, the overall shape of the protein drifts over time. P23H drifts 1.62 Å further than healthy rhodopsin. The top candidate drug restores approximately 90% of this gap, bringing the structure close to the healthy baseline (p < 10⁻¹⁵).

Finding 6:

hnRP94450 passes ADMET screening indicating favorable properties for a potential drug candidate (absorption, distribution, metabolism, excretion, and toxicity)

The results support the hypothesis that a computationally identified small molecule can reduce structural instability at the P23H mutation site.

Scientifically, this work shows that computational drug discovery cannot stop at docking. Docking is fast and inexpensive, but it only determines whether a molecule fits the binding pocket, not whether it stabilizes the protein. The top-scoring docking hit was the worst therapeutic candidate. Molecular dynamics in a virtual cell membrane is the real test. Any future search for drugs targeting membrane proteins needs this second stage.

For the benefit of patients with P23H retinitis pigmentosa, the pipeline identified a candidate compound. hnRP94450 eliminates every significant instability event at the mutation site, and it stabilizes the mutated region without freezing parts of the protein that need to stay flexible. hnRP94450 is a part of the chromenone family, the same molecular class that published studies have already linked to P23H rescue. It's not a proven drug, but it's a starting point that shows potential for wet-lab trials.

For other protein misfolding diseases, the pipeline is reusable and can be adapted to any known membrane protein. Cystic fibrosis, Brugada syndrome, and long QT syndrome are all caused by misfolded membrane proteins. The core approach of finding a small molecule that holds the protein in its correct shape without disrupting normal function can be extrapolated to many other conditions.

Personally, as someone slowly losing vision to a different form of retinitis pigmentosa, this project is my way of contributing to the search for treatments.

Immediate (patent)

Filing a U.S. provisional use patent for hnRP94450 to slow retinitis pigmentosa progression. This ensures an early priority date before public disclosure and provides a twelve month window to add validation data before converting to non-provisional patent.

Medium-term (experimental)

A cellular trafficking assay in HEK293 cells (human embryonic kidney 293) expressing P23H rhodopsin will test whether hnRP94450 rescues the protein in a living cell.

Long-term

The pipeline will be adapted to other forms of retinitis pigmentosa, and to other protein misfolding diseases. The full code will be open-sourced so other students can do virtual drug discovery too.

I would like to thank Dr. Chris Barden for his guidance in analyzing my molecular dynamics and docking data and advising me on protecting my IP, Dr. Johane Robitaille for outreach and encouragement, and Ms. Michelle Caldwell (my biology teacher) for her support in conducting preliminary research and encouraging me to pursue this project. I would also like to thank Steve Macdonald for letting me borrow his computer to run my simulations on.

Abraham, M. J., Murtola, T., Schulz, R., Páll, S., Smith, J. C., Hess, B., & Lindahl, E. (2015). GROMACS: High performance molecular simulations through multi-level parallelism from laptops to supercomputers. SoftwareX, 1–2, 19–25. https://doi.org/10.1016/j.softx.2015.06.001

Anthropic. (2024). Claude Code [Large language model]. https://www.anthropic.com/

Athanasiou, D., Aguila, M., Bellingham, J., Kanuga, N., Sheridan, C., & Cheetham, M. E. (2022). Chromenone derivatives as novel pharmacological chaperones for retinitis pigmentosa-linked rod opsin mutants. Human Molecular Genetics, 31(20), 3439–3453. https://doi.org/10.1093/hmg/ddac130

Bhatt, N. F., Bhatt, D. K., & Bhatt, R. K. (2018). A novel small molecule chaperone of rod opsin and its potential therapy for retinal degeneration. Nature Communications, 9(1), 1–12. https://doi.org/10.1038/s41467-018-04261-1

Chen, Y., Chen, Y., Jastrzebska, B., Bhatt, M. K., Bhatt, D. K., & Bhatt, N. F. (2014). Inherent instability of the retinitis pigmentosa P23H mutant opsin. Journal of Biological Chemistry, 289(13), 9288–9303. https://doi.org/10.1074/jbc.M114.551713

Daiger, S. P., Sullivan, L. S., & Bowne, S. J. (2013). Genes and mutations causing retinitis pigmentosa. Clinical Genetics, 84(2), 132–141. https://doi.org/10.1111/cge.12203

Gaulton, A., Hersey, A., Nowotka, M., Bento, A. P., Chambers, J., Mendez, D., Mutowo, P., Atkinson, F., Bellis, L. J., Cibrián-Uhalte, E., Davies, M., Dedman, N., Karlsson, A., Magariños, M. P., Overington, J. P., Papadatos, G., Smit, I., & Leach, A. R. (2017). The ChEMBL database in 2017. Nucleic Acids Research, 45(D1), D945–D954. https://doi.org/10.1093/nar/gkw1074

GROMACS Development Team. (n.d.). GROMACS logo [Logo]. https://www.gromacs.org/About_GROMACS

Grossfield, A., & Feller, S. E. (2007). Convergence of molecular dynamics simulations of membrane proteins. Proteins: Structure, Function, and Bioinformatics, 67(1), 31–40. https://doi.org/10.1002/prot.21308

Hartong, D. T., Berson, E. L., & Dryja, T. P. (2006). Retinitis pigmentosa. The Lancet, 368(9549), 1795–1809. https://doi.org/10.1016/S0140-6736(06)69740-7

Huang, J., Rauscher, S., Nawrocki, G., Ran, T., Feig, M., de Groot, B. L., Grubmüller, H., & MacKerell, A. D., Jr. (2017). CHARMM36m: An improved force field for folded and intrinsically disordered proteins. Nature Methods, 14(1), 71–73. https://doi.org/10.1038/nmeth.4067

Illing, M. E., Rajan, S. R., Bhatt, N. F., & Bhatt, D. K. (2002). A rhodopsin mutant linked to autosomal dominant retinitis pigmentosa is prone to aggregate and interacts with the ubiquitin proteasome system. Journal of Biological Chemistry, 277(37), 34150–34160. https://doi.org/10.1074/jbc.M204955200

Jo, S., Kim, T., Iyer, V. G., & Im, W. (2008). CHARMM-GUI: A web-based graphical user interface for CHARMM. Journal of Computational Chemistry, 29(11), 1859–1865. https://doi.org/10.1002/jcc.20945

Landrum, G. (2023). RDKit: Open-source cheminformatics. RDKit Project. https://www.rdkit.org/

Lee, J., Cheng, X., Swails, J. M., Yeom, M. S., Eastman, P. K., Lemkul, J. A., Wei, S., Buckner, J., Jeong, J. C., Qi, Y., Jo, S., Pande, V. S., Case, D. A., Brooks, C. L., 3rd, MacKerell, A. D., Jr., Klauda, J. B., & Im, W. (2016). CHARMM-GUI input generator for NAMD, GROMACS, AMBER, OpenMM, and CHARMM/OpenMM simulations using the CHARMM36 additive force field. Journal of Chemical Theory and Computation, 12(1), 405–413. https://doi.org/10.1021/acs.jctc.5b00935

Leioatts, N., Mertz, B., Martinez-Mayorga, K., Romo, T. D., Pitman, M. C., Feller, S. E., Grossfield, A., & Brown, M. F. (2015). Retinal conformation changes rhodopsin’s dynamic ensemble. Biophysical Journal, 109(3), 608–617. https://doi.org/10.1016/j.bpj.2015.06.046

Lipinski, C. A., Lombardo, F., Dominy, B. W., Feeney, P. J., & Bhatt, N. F. (1997). Experimental and computational approaches to estimate solubility and permeability in drug discovery and development settings. Advanced Drug Delivery Reviews, 23(1–3), 3–25. https://doi.org/10.1016/S0169-409X(96)00423-1

Mendes, H. F., & Bhatt, D. K. (2008). A high-throughput screening method for small-molecule pharmacologic chaperones of misfolded rhodopsin. Investigative Ophthalmology & Visual Science, 49(7), 3224–3230. https://doi.org/10.1167/iovs.07-1326

Noorwez, S. M., Kuksa, V., Imanishi, Y., Zhu, L., Filipek, S., Palczewski, K., & Kaushal, S. (2003). Pharmacological chaperone-mediated in vivo folding and stabilization of the P23H-opsin mutant associated with autosomal dominant retinitis pigmentosa. Journal of Biological Chemistry, 278(16), 14442–14450. https://doi.org/10.1074/jbc.M300087200

O’Boyle, N. M., Banck, M., James, C. A., Morley, C., Vandermeersch, T., & Hutchison, G. R. (2011). Open Babel: An open chemical toolbox. Journal of Cheminformatics, 3, 33. https://doi.org/10.1186/1758-2946-3-33

Okada, T., Sugihara, M., Bondar, A. N., Elstner, M., Entel, P., & Buss, V. (2004). The retinal conformation and its environment in rhodopsin in light of a new 2.2 Å crystal structure. Journal of Molecular Biology, 342(2), 571–583. https://doi.org/10.1016/j.jmb.2004.07.044

Palczewski, K., Kumasaka, T., Hori, T., Behnke, C. A., Motoshima, H., Fox, B. A., Le Trong, I., Teller, D. C., Okada, T., Stenkamp, R. E., Yamamoto, M., & Miyano, M. (2000). Crystal structure of rhodopsin: A G protein-coupled receptor. Science, 289(5480), 739–745. https://doi.org/10.1126/science.289.5480.739

Python Software Foundation. (n.d.). Python logo [Logo]. https://www.python.org/community/logos/

RCSB Protein Data Bank. (n.d.). PDB entry 1U19: Crystal structure of bovine rhodopsin at 2.2 Å resolution. https://www.rcsb.org/structure/1U19

Russell, S., Bennett, J., Wellman, J. A., et al. (2017). Efficacy and safety of voretigene neparvovec… The Lancet, 390(10097), 849–860. https://doi.org/10.1016/S0140-6736(17)31868-8

Schrödinger, LLC. (2015). The PyMOL molecular graphics system (Version 2.0) [Computer software].

Sung, C. H., Schneider, B. G., Agarwal, N., Papermaster, D. S., & Bhatt, N. (1991). Functional heterogeneity of mutant rhodopsins responsible for autosomal dominant retinitis pigmentosa. Proceedings of the National Academy of Sciences, 88(19), 8840–8844. https://doi.org/10.1073/pnas.88.19.8840

The Scripps Research Institute. (n.d.). AutoDock Vina [Logo]. http://vina.scripps.edu/

Trott, O., & Olson, A. J. (2010). AutoDock Vina: Improving the speed and accuracy of docking with a new scoring function. Journal of Computational Chemistry, 31(2), 455–461. https://doi.org/10.1002/jcc.21334

University of Florida. (n.d.). Anatomy of a rod photoreceptor [Diagram]. PHY 3400: Biological Physics. https://phy3400-phy3940.phys.ufl.edu

Vanommeslaeghe, K., et al. (2010). CHARMM general force field… Journal of Computational Chemistry, 31(4), 671–690. https://doi.org/10.1002/jcc.21367

European Bioinformatics Institute (EMBL-EBI). (n.d.). ChEMBL logo [Logo]. https://www.ebi.ac.uk/chembl/

Claude Code (Anthropic) was used solely as a programming assistance tool to generate and refine computational scripts; it did not contribute to experimental design, data interpretation, or scientific conclusions.