Designing the peptide

The Design–Build–Test–Learn (DBTL) cycle is integral to any synthetic biology-based engineered solution for biological problems. As a result, to arrive at the best possible peptide, we performed multiple iterations of the engineering cycle. Each iteration involved designing our peptide, building it and finally testing it in-silico. The changes involved in each iteration and the information learnt have been documented.

The DBTL cycle
Design, Build, Test and Learn

The predicted structures of MAM7 and Heparin Binding Domain (HBD) of Fibronectin were docked with each other on ClusPro [1]-[4], followed by visualisation and observation of these interacting residues on PyMOL [5]. The sequence signature of the structural repeats in the proteins of Fibronectin and MAM7 were detected using HHrepID [6],[7]—a tool for de novo identification of repeats in protein sequences. This confirmed the existence of 5 continuous repeats of type 1 modules, forming the Heparin-Binding Domain of Fibronectin and the presence of seven Mammalian Cell Entry domains (MCEs) in MAM7. Fibronectin is known to require five tandem MCE domains to dock with MAM7; as a result, designing a peptide that would bind to one of the central MCE domains was deemed sufficient [8].

Each of the five repeats of the type I module present in Fibronectin was considered an individual peptide, and the structure for each of these peptides was predicted. Amber, an energy mininmization tool, was employed to stabilize the predicted peptide structures. The peptides were then docked with MAM7 to test its binding using ClusPro [1]-[4]. It was observed that the peptides numbered three and five derived from modules three and five of Fibronectin, respectively, docked with the MCE3 of MAM7, gave the highest docking scores.

It was seen that the fifth peptide had the highest docking score with MCE3. Thus, this peptide was taken as the basis for the next set of modifications.

Schrodinger Maestro[9] was used to visualise and study the specific amino acids involved in the interaction between the peptide and MAM7. The residues within 5 Å were assumed to be close enough for the interaction and thus were considered during the evaluation. Further, this was verified by subjecting the peptide to an alanine scan on BUDE. An alanine scan replaces all the amino acid residues in a peptide with alanine one at a time and then compares the values of the respective Gibbs free energies to determine the extent of contribution of every amino acid in the docking process [10]. Five such mutatable positions were identified in the wild type. This was followed by improving the interactions by substituting aforementioned amino acids. These replacements were made based on the study [11], but each of them had a plethora of possibilities. For example, Methionine (present at the 29th position) could be replaced by Glutamic acid, Aspartic acid, Phenylalanine, Tyrosine, Tryptophan, Histidine or Leucine.

Therefore, the docked structure was subjected to mutation Cutoff Scanning Matrix (mCSM) [12], which helped us analyse which mutations were truly beneficial. With newly learnt information, the peptide was subjected to new changes. Hence, it was necessary to predict the structure of the peptide and dock it with MAM7 at every stage. The docked complex was subjected to an alanine scan to account for conformational changes by the substitutions, that mCSM does not account for, and to verify the interactions. This gave us twelve variants of the peptide with different combinations of the various substitutions that could be made. These were then visualised on Schrodinger Maestro and further analysed using Prodigy to identify the strongly interacting peptides.

Four of the twelve peptides showed an increased binding affinity, and the best among these was selected for the next round. The mutations performed in the previous iteration introduced new amino acids in the docked structure, opening up an avenue for a new set of substitutions and hence the process was repeated. Subsequently, a new set of variants were obtained (16 variants). These were further subjected to docking and analysis to select the next best peptide. At this stage, to prevent the peptides from mutually interacting and consequently decreasing their interaction efficiency with MAM7, we performed AGGRESCAN [13] on the four peptides obtained from the last iteration, to highlight the hotspots of aggregation. One such hotspot involving eight amino acids was observed and was rectified by substituting Cytosine at the 34^th position with Asparagine in every peptide. After this process, our very own software—GRASP, helped us increase the scope of possible mutations.

Two rounds of running the peptide through the software gave us the final one. The structure of the peptide was then predicted and docked with MAM7, and run through Prodigy [14],[15] to compare its binding affinity and K_d value with the previous variants. These values are tabulated below:

Table- Peptide variants with gibbs free energy and dissociation constant References:

1. I. T. Desta, K. A. Porter, B. Xia, D. Kozakov, and S. Vajda, “Performance and Its Limits in Rigid Body Protein-Protein Docking,” Structure, vol. 28, no. 9, pp. 1071-1081.e3, Sep. 2020, doi: 10.1016/j.str.2020.06.006.
2. S. Vajda et al., “New Additions to the ClusPro Server Motivated by CAPRI”, doi: 10.1002/prot.25219.
3. D. Kozakov et al., “The ClusPro web server for protein-protein docking”, doi: 10.1038/nprot.2016.169.
4. D. Kozakov et al., “How Good is Automated Protein Docking?”, doi: 10.1002/prot.24403.
5. PyMOL The PyMOL Molecular Graphics System, Version 2.5.2 Schrödinger, LLC.
6. L. Zimmermann et al., “A Completely Reimplemented MPI Bioinformatics Toolkit with a New HHpred Server at its Core,” J. Mol. Biol., vol. 430, no. 15, pp. 2237–2243, Jul. 2018, doi: 10.1016/j.jmb.2017.12.007.
7. F. Gabler et al., “Protein Sequence Analysis Using the MPI Bioinformatics Toolkit,” Curr. Protoc. Bioinforma., vol. 72, no. 1, Dec. 2020, doi: 10.1002/cpbi.108.
8. A. M. Krachler and K. Orth, “Functional characterization of the interaction between bacterial adhesin Multivalent Adhesion Molecule 7 (MAM7) protein and its host cell ligands,” J. Biol. Chem., vol. 286, no. 45, pp. 38939–38947, Nov. 2011, doi: 10.1074/JBC.M111.291377.
9. Maestro Schrödinger Release 2022-3:Maestro, Schrödinger, LLC, New York, NY, 2021.
10. A. A. Ibarra et al., “Predicting and Experimentally Validating Hot-Spot Residues at Protein-Protein Interfaces,” ACS Chem. Biol., vol. 14, no. 10, pp. 2252–2263, Oct. 2019, doi: 10.1021/ACSCHEMBIO.9B00560/SUPPL_FILE/CB9B00560_SI_002.XLSX.
11. N. Z. Xie, Q. S. Du, J. X. Li, and R. B. Huang, “Exploring Strong Interactions in Proteins with Quantum Chemistry and Examples of Their Applications in Drug Design,” PLoS One, vol. 10, no. 9, p. 137113, Sep. 2015, doi: 10.1371/JOURNAL.PONE.0137113.
12. D. E. V. Pires, D. B. Ascher, and T. L. Blundell, “mCSM: predicting the effects of mutations in proteins using graph-based signatures,” Bioinformatics, vol. 30, no. 3, pp. 335–342, Feb. 2014, doi: 10.1093/BIOINFORMATICS/BTT691.
13. O. Conchillo-Solé, N. S. de Groot, F. X. Avilés, J. Vendrell, X. Daura, and S. Ventura, “AGGRESCAN: A server for the prediction and evaluation of ‘hot spots’ of aggregation in polypeptides,” BMC Bioinformatics, vol. 8, no. 1, pp. 1–17, Feb. 2007, doi: 10.1186/1471-2105-8-65/FIGURES/1.
14. A. Vangone and A. M. Bonvin, “Contacts-based prediction of binding affinity in protein–protein complexes,” Elife, vol. 4, Jul. 2015, doi: 10.7554/eLife.07454.
15. L. C. Xue, J. P. Rodrigues, P. L. Kastritis, A. M. Bonvin, and A. Vangone, “PRODIGY: a web server for predicting the binding affinity of protein–protein complexes,” Bioinformatics, p. btw514, Aug. 2016, doi: 10.1093/bioinformatics/btw514.