Predicted Protein Interactions of the SARS-CoV-2 B.1.1.529 Variant with Neutralizing Antibodies

Each time a new SARS-CoV-2 variant pops up in the news, there are two reactions that I have: “Here we go again…🙄” and “I wonder if this actually means anything? 🤔”.

On Friday, I published a post on Medium that outlines how I used AlphaFold2 to predict the structure of the receptor binding domain from the Spike protein of the B.1.1.529 variant and posited my guesses as to its effect on vaccine efficacy.

Today, I’m taking this research a step further. I’ve now run multiple analyses that dock a couple neutralizing antibodies to my predicted receptor binding domain (RBD) structure. Using HADDOCK, a modeling tool from Utrecht University that allows you to dock two or more protein structures together, I predicted the binding affinity between the RBD of B.1.1.529 and two actual Spike reference structures (PDBs: 6XC2 and 6XC7).

About HADDOCK

HADDOCK is a tool I’ve used a ton in my malaria research when looking at interactions with human immunoproteins. You simply provide two .PDB files, specifying residues that are “active” (the assumed binding loci), and the system will return multiple PDBs with both proteins docked together. The system also quantifies the interactions by providing biophysical metrics:

Computational tools like HADDOCK allow us to quickly create in silico predictions of protein-protein interactions. Since lab-based/in vitro protein-protein interaction experiments take time and money, predictive tools can give us a good guess for cheap.

What’s Up, Dock?

In my analyses, two SARS-CoV-2 neutralizing antibodies where used: CC12.1 and CC12.3. I then docked each of these antibodies with a reference RBD and the B.1.1.529 variant RBD to see what changes.

CC12.1 Antibody

First, to get a baseline interaction, I docked CC12.1 with the RBD in PDB 6XC2. Then, I docked CC12.1 with the predicted RBD of B.1.1.529. This provided the following metrics:

You can also see the docked structures here:

Predicted docking of RBDs (PDB: 6XC2 shown in green; predicted B.1.1.529 structure shown in blue) with the antibody Fab region of CC12.1 (from PDB: 6XC2) shown in pink/magenta.

Note how the overall HADDOCK score increases (worsens) with the B.1.1.529 variant RBD. The CC12.1 antibody does not bind as closely with the variant RBD (see the “Buried Surface Area” metric) and the electrostatic interaction is slightly weaker (~18% less). However, the position of the antibody to the RBDs is still quite similar to the reference binding.

CC12.3 Antibody

Repeating the above analysis, but swapping in CC12.3 and the RDB from PDB 6XC7, this yielded the following results:

…and the docked structures:

Predicted docking of RBDs (PDB: 6XC7 shown in green; predicted B.1.1.529 structure shown in blue) with the antibody Fab region of CC12.3 (from PDB: 6XC7) shown in pink/magenta.

As you can see, the results are fairly similar: the CC12.3 antibody doesn’t bind as well to the B.1.1.529 variant RBD, but it still binds. The electrostatic interaction is weaker (~22% less), but the interaction distance remained about the same.

Interestingly, the binding of the CC12.3 antibody was weaker with the 6XC7 RBD than the CC12.1 antibody was with the 6XC2 RBD above. So, the scores between these analyses shouldn’t be compared directly.

Dock it to me! (Key Takeaways)

So, what does this all mean from a virology/immunology/public health standpoint? Here are my thoughts:

Resources

All code, sequences, and results are available on GitHub here: https://github.com/colbyford/SARS-CoV-2_B.1.1.529_Spike-RBD_Predictions

If you use any of this work, code, insights, or results, please cite:

Colby T. Ford. (2021). Predictions of the SARS-CoV-2 B.1.1.529 Variant Spike Protein Receptor Binding Domain Structure and Neutralizing Antibody Interactions (v1.0.0) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.5733161

Stay curious.

Cloud AI and genomics guy and aspiring polymath. I am a researcher in machine learning and bioinformatics and I sometimes write things here.