Goal
Nanobodies are single-domain antigen-binding antibody fragments around 12-15kDa, one-tenth the size of conventional antibodies (Liu and Yang, 20022). With nanobodies exhibiting greater stability, better tissue penetration, higher binding efficiency, improved safety and easier production, it can address many issues that current monoclonal antibodies and antibody fragments have, and there is no doubt that nanobodies are the next step forward for protein binders in cell therapy and beyond (Maali et al., 2023). With our success in using nanobodies to replace the single-chain variable fragment (scFv) in our CAR constructs, our team believes more research teams around the world should have the opportunity to utilize nanobodies.
However, public information in nanobodies is extremely limited. Since nanobodies are primarily produced by Camelids (camels, alpacas, sharks etc.), immune or naive nanobody libraries are time-consuming and expensive, therefore off-limits to most laboratories (Liu and Yang, 20022). Given the importance of nanobody libraries to identify antigen-specific nanobodies, we must turn to synthetic nanobody libraries. Given the promising success and prospects in AlphaFold and diffusion models like RFdiffusion, we turned to deep learning models to help.
Instead of generating an entire nanobody library, we wanted to create a model that can produce a short list of antigen-specific nanobody designs in silico which wet lab researchers can test, simplifying researchers’ use of nanobodies and greatly lowering the barrier of entry to this exciting new technology. In fact, other researchers are already working on generating antigen-specific antibodies and nanobodies with diffusion models (Bennett et al., 2024; Luo et al., 2022). Here, we are working towards our own model, which we dub the “Nanobody Initiative”.
Nevertheless, this is no small feat. Since CAR_Ma remains the main focus of our project, this year we will work on a proof of concept to generate a nanobody that targets CD14, to functionalize our nanoparticles with. With a lack of consensus on defining features of synthetic nanobodies, however, we have instead chosen to generate protein binders, with future sights set on identifying common nanobody characteristics to implement into the protein binder generation process.
This year, we have collaborated with the Po Leung Kuk Ngan Po Ling College iGEM team on tackling this modeling project.
Generation of de novo Protein Binders
In our project, we have decided to target CD14 for nanoparticle transfection since it is a prominent and crucial antigen expressed on macrophages to identify bacteria, through binding to a bacteria’s lipopolysaccharide (LPS) region (Zamani et al., 2013). Given that it has been extensively studied, it was chosen to target the CD14’s hydrophobic LPS binding pocket, with the extra rim residues found in human CD14 being noted (Kelley et al., 2013).
Now armed with a target binding site, we turned to the generation of the protein binder. The protein backbone was generated through RFDiffusion, a potent deep-learning network for protein design, most notable de novo binder designs (Watson et al., 2023). Using the publicly available Google Collab notebook of RFDiffusion, 16 designs using 50 iterations of denoising were run with the hotspot set to the rim residues of human CD14’s LPS-binding pocket. Then, ProteinMPNN was used to generate sequences that potentially fold back to the diffused backbone. 64 ProteinMPNN sequences were produced for each RFDiffusion-designed backbone, creating 1024 unique binder designs. From here, AlphaFold2-multimer was used to predict the folding of the protein binders, and their protein-protein interaction with CD14’s LPS binding pocket. A filter for iPAE score under 10 was applied to the results to identify the most promising sequences. While iPAE was chosen as a global confidence indicator that better reflects a bigger picture, other confidence scores like pLDDT and iPTM were also considered (EMBL-EBI, n.d.b). All filtered sequences had a pLDDT score larger than 0.767 and an iPTM score larger than 0.619, suggesting high local confidence and a well-predicted protein structure (EMBL-EBI, n.d.a; EMBL-EBI, n.d.c). This all suggests that Alphafold is confident regarding the interaction between the binder and CD14’s LPS binding pocket.
Molecular Dynamics Evaluation of Protein Binders
With our protein binders now generated, they have to be evaluated for their binding affinities, through evaluating their free energy terms. Molecular dynamics (MD) simulations governed by Newton’s laws of motion to reflect physio-chemical properties have emerged as a promising tool to investigate protein structure and binding (Sinha et al., 2022). Specifically, GROMACS was used to generate the MD simulations due to its versatility, speed and open access (Van Der Spoel et al., 2005). To investigate binding affinity, the binding free energies are evaluated with the Molecular Mechanics Generalized Born and Surface Area Continuum Solvation (MM/GBSA) method. Since MD simulations are extremely computationally intensive to run, we were fortunate enough to be able to use HKU’s HPC2021, and used GROMACS version 2021.3 loaded on HPC2021. If iGEM teams wishing to run MD simulations without high performance computing access, CUDA-based GPU acceleration can also provide significant performance improvements (GROMACS Development Team, n.d.).
Firstly, we’ve chosen to utilize the Amber99SB forcefield, which is a set of energy functions and empirical parameters used to evaluate potential energies in MD simulations (Hornak et al., 2006). Amber99SB was chosen since it is an improvement over the wildly successful and popular ff94 and ff99 Amber forcefields, focused on improving protein backbone parameterization, notably regarding the balance of secondary structures (Hornak et al., 2006). This provides a refined simulation that would be more applicable to the physical world and thus wet lab work.
Next, the MD simulation is built to reflect water at room temperature (293K, 1atm), with Na+ and Cl- ions to balance the system’s charge. Then, energy minimisation is done to relax the structure, removing any steric clashes or inappropriate geometry, and creating a reasonable starting structure (Lemkul, n.d.). Afterwards, NVT (constant Number of particles, Volume and Temperature) and NPT (pressure instead of volume) equilibration is conducted for temperature and pressure to equilibrate solvents and ions around the protein (Lemkul, n.d.). With the system now ready, a 10ns simulation was generated for all 12 filtered binder sequences and their interaction with CD14’s LPS-binding pocket.
From here, MM/GBSA is implemented via gmx_MMPBSA (Valdés-Tresanco et al., 2021). MM/GBSA is an end-point method that evaluates the free energy difference between the bound complex and the unbound components to calculate free energy through estimating a solution to the Generalized Born equation (Walker et al., 2006). Given that this is simply a proof of concept with no therapeutic implementation yet, MM/GBSA was chosen over other methods due to its balance of precision and low computational demands, and because the results are more stable and reproducible by others (Genheden and Ryde, 2015). By inputting the MB structure and trajectory, the average and SD of the free energy value for our binder designs are outputted, with one binder design encountering an error during the MD simulation. One binder backbone design experienced errors during this analysis and was discarded, leaving us with 11 binder designs. The 11 binders and their free energy value are graphed below, with error bars of 1 SD.
Figure 1. Delta G Values with error bars of 1 standard deviation for 11 tested binder designs, calculated with MM/PBSA.
Here, smaller Delta G values suggest a higher binding affinity, with binder designs 28, 17 and 2 performing the best. Below contains a visualization of these three binder designs binding to CD14.
Figure 2. 10ns simulation of binder design 28 binding to CD14’s LPS binding pocket.
Figure 3. 10ns simulation of binder design 17 binding to CD14’s LPS binding pocket.
Figure 4. 10ns simulation of binder design 2 binding to CD14’s LPS binding pocket.
Future Prospects
With a workflow for generating and evaluating protein binders now created and validated, the next step is to further specify this workflow to generate nanobodies. A possible idea to investigate is to generate nanobodies from a universal humanized nanobody scaffold, as identified by Vincke et al. (2009). With the field of de novo protein binder modeling developing, we hope to be able to communicate and learn from other research groups and realize the dream of the nanobody initiative.
References
Cafferty, S. M., De Temmerman, J., Kitada, T., Becraft, J. R., Weiss, R., Irvine, D. J., Devreese, M., De Baere, S., Combes, F., & Sanders, N. N. (2021). In vivo validation of a reversible small Molecule-Based switch for synthetic Self-Amplifying mRNA regulation. Molecular Therapy, 29(3), 1164–1173. https://doi.org/10.1016/j.ymthe.2020.11.010
Cancer Research UK. (2023, December 12). Other ways of having chemotherapy. https://www.cancerresearchuk.org/about-cancer/treatment/chemotherapy/how-you-have/ways-of-having
Cleaveland Clinic. (2024, August 23). CAR T-Cell Therapy. Cleveland Clinic. https://my.clevelandclinic.org/health/treatments/17726-car-t-cell-therapy
Cigna Worldwide Life Insurance Company Limited. (2021, July 30). Radiotherapy: side effects, procedures, treatment costs & diet tips. https://www.cigna.com.hk/en/smarthealth/medical/radiotherapy-side-effects-procedures-treatments-costs-and-diet
Cliff, E. R. S., Kelkar, A. H., Russler-Germain, D. A., Tessema, F. A., Raymakers, A. J., Feldman, W. B., & Kesselheim, A. S. (2023). High cost of chimeric antigen receptor T-Cells: challenges and solutions. American Society of Clinical Oncology Educational Book, 43. https://doi.org/10.1200/edbk_397912
Food and Drug Administration. (2024, April 18). FDA requires boxed warning for T cell malignancies following treatment with BCMA-Directed or CD19-Directed autologous chimeric antigen receptor (CAR) T cell immunotherapies. U.S. Food And Drug Administration. https://www.fda.gov/vaccines-blood-biologics/safety-availability-biologics/fda-requires-boxed-warning-t-cell-malignancies-following-treatment-bcma-directed-or-cd19-directed
Hadiloo, K., Taremi, S., Heidari, M., & Esmaeilzadeh, A. (2023). The CAR macrophage cells, a novel generation of chimeric antigen-based approach against solid tumors. Biomarker Research, 11(1). https://doi.org/10.1186/s40364-023-00537-x
HKU iGEM. (2023). CAR_Ma. https://2023.igem.wiki/hku-hongkong/
Laskar, S. G., Sinha, S., Krishnatry, R., Grau, C., Mehta, M., & Agarwal, J. P. (2022). Access to radiation therapy: from local to global and equality to equity. JCO Global Oncology, 8. https://doi.org/10.1200/go.21.00358
HKU iGEM. (2023). CAR_Ma. https://2023.igem.wiki/hku-hongkong/
Light, D. W., & Lexchin, J. (2021). The costs of coronavirus vaccines and their pricing. Journal of the Royal Society of Medicine, 114(11), 502–504. https://doi.org/10.1177/01410768211053006
Maeda, H., & Khatami, M. (2018). Analyses of repeated failures in cancer therapy for solid tumors: poor tumor‐selective drug delivery, low therapeutic efficacy and unsustainable costs. Clinical and Translational Medicine, 7(1). https://doi.org/10.1186/s40169-018-0185-6
Nadeem, H., Jayakrishnan, T. T., Rajeev, R., Johnston, F. M., Gamblin, T. C., Turaga, K. K., Nadeem, H., Jayakrishnan, T. T., Rajeev, R., Johnston, F. M., Gamblin, T. C., & Turaga, K. K. (2016). RECAP: Cost differential of chemotherapy for solid tumors. Journal of Oncology Practice, 12(3), 251. https://doi.org/10.1200/jop.2015.006700
UpToDate lnc. (2024). Trimethoprim: Drug information. MediLib. https://medilib.ir/uptodate/show/10024