CAGI5 - Frataxin Challenge

Alexey Strokach, Carles Corbi-Verge, Philip M. Kim

2018-07-06

Overview

• Discuss the motivation for developing ELASPIC.

• Present the ELASPIC webserver.

• Discuss predictions made by ELASPIC and other methods for 8 mutations in Frataxin.

• ΔΔG datasets are hard to come by,

Introduction

We need tools for improving protein stability and increasing the affinity of proteins to their targets. Design proteins that are easier to crystallize and that show higher yield and less aggregation when expressed in heterologous systems. Design biocatalysts which remain active in inhospitable conditions, including high temperatures and pressures, acidic or basic pH, different salt types and concentrations, and admixture of organic solvents Optimize protein and peptide “biologics” to improve their shelf life and increase their affinity and specificity to de- sired targets. We need tools which provide a mechanistic explanation as to why a variant may be deleterious or benign. ELASPIC uses sequential and structural information to predict the effect of mutations on protein-folding and protein-protein interaction on a genome-wide scale.

• Research laboratories
• Industrial processes engineering
• Pharmaceutical industry
• CAR-T: Chimeric antigen receptor therapy
• Optimize the affinity between recombinant proteins and their targets (for example, to be used in biological drugs, CAR-T, etc.).

Methods

Berliner et. al. (2014) PLoS ONE 9(9): e107353.

FoldX energy function

$$ \begin{align} ΔG =& w_1 ⋅ ΔG_\text{vdw} + w_2 ⋅ ΔG_\text{solvH} + w_3 ⋅ ΔG_\text{solvP} + w_4 ⋅ ΔG_\text{hbond} + w_5 ⋅ ΔG_\text{wb} + \\ & w_6 ⋅ ΔG_\text{el}+ w_7 ⋅ ΔG_\text{clash} + w_8 ⋅ TΔS_\text{mc} + w_9 ⋅ TΔS_\text{sc} + w_{10} ⋅ ΔG_\text{kon} \end{align} $$

• $\{w_1 ... w_{10}\}$ - tuned parameters.
• $ΔG_\text{vdw}$ - experimental transfer energies from water to vapour.
• $ΔG_\text{solvH}$ - hydrophobic desolvation score, scaled with the burial of the residue.
• $ΔG_\text{solvP}$ - hydrophilic desolvation score, scaled with the burial of the residue.
• $ΔG_\text{hbond}$ - hydrogen bonds, inferred through geometric contrains and empirical data.
• $ΔG_\text{wb}$ - water molecules that have persistent interactions with the protein.
• $ΔG_\text{el}$ - electrostatics, calculated using Coulomb's law.
• $ΔG_\text{clash}$ - steric overlaps between atoms in the structure.
• $TΔS_\text{mc}$ - the entropic penalty for fixing the backbone in a given conformation.
• $TΔS_\text{sc}$ - the entropy cost of fixing a side chain in a particular conformation.
• $ΔG_\text{kon}$ - electrostatic interactions between different chains.

For energy terms, we relied primarily on the FoldX energy function.

Most of these terms are loosely based on small molecule experimental data or theoretic experiments.

Note that while FoldX can predict the ∆∆G of mutation, we only used the FoldX energy terms and not its prediction.

The energy function includes terms that have been found to be important for protein stability, where the energy of unfolding (∆G) of a target protein is calculated using the equation:

∆G = ∆Gvdw + ∆GsolvH + ∆GsolvP + ∆Ghbond + ∆Gwb + ∆Gel + ∆Smc + ∆Ssc

• ∆Gvdw is the sum of the Van der Waals contributions of all atoms with respect to the same interactions with the solvent.
• ∆GsolvH and ∆GsolvP is the difference in solvation energy for apolar and polar groups, respectively, when going from the unfolded to the folded state.-
• ∆Ghbond is the free energy difference between the formation of an intra-molecular hydrogen-bond compared to inter-molecular hydrogen-bond formation (with solvent).
• ∆Gwb is the extra stabilizing free energy provided by a water molecule making more than one hydrogen-bond to the protein (water bridges) that cannot be taken into account with non-explicit solvent approximations.
• ∆Gel is the electrostatic contribution of charged groups, including the helix dipole.
• ∆Smc is the entropy cost for fixing the backbone in the folded state. This term is dependent on the intrinsic tendency of a particular amino acid to adopt certain dihedral angles.
• ∆Ssc is the entropic cost of fixing a side chain in a particular conformation.

The energy values of ∆Gvdw, ∆GsolvH, ∆GsolvP and ∆Ghbond attributed to each atom type have been derived from a set of experimental data, and ∆Smc and ∆Ssc have been taken from theoretical estimates. The Van der Waals contributions are derived from vapor to water energy transfer, while in the protein we are going from solvent to protein.

For protein-protein interactions, or protein-DNA interactions FoldX calculates ∆∆G of interaction :

∆∆Gab = ∆Gab- (∆Ga + ∆Gb) + ∆Gkon + ∆Ssc

∆Gkon reflects the effect of electrostatic interactions on the kon. ∆Ssc is the loss of translational and rotational entropy upon making the complex.

Provean methodology

Source: http://provean.jcvi.org/about.php.

Provean score

Source: http://provean.jcvi.org/about.php.

ELASPIC webserver

http://elaspic.kimlab.org/

Witvliet^†, Strokach^†, et. al. (2016) Bioinformatics 32(10): 1589-1591.

• SIFT -> Provean
• NACCESS -> Stide

Also

• Homology models of all human proteins and PPIs
• Provean supporting sets for all human proteins
• Run jobs on our local Sun Grid Engine cluster

Spearman correlation coefficient because our primary goal is to rank mutations for experimental validation

Performance on core mutations

	ΔΔG datasets	Phenotype datasets
Training / Validation
Test

Performance on interface mutations

	ΔΔG datasets	Phenotype datasets
Training / Validation
Test

Performance on oncogenes / tumor suppresor genes in DoCM

• Predictions for passenger mutations (green), mutations in tumour suppressors genes (purple), and mutations in oncogenes (cyan).

• In molecular dynamics, we aim to model the movement of atoms in a molecule using laws of physics.
• We use a simplifying assumption where atoms are treated as balls connected to each other using springs.

Results

ELASPIC results

FoldX	Provean	ELASPIC

Note: ΔΔG values are changes in the Gibbs free energy of unfolding in kcal / mol.

ELASPIC results

http://elaspic.kimlab.org/result/a5393e/

Rosetta results

Protocol: ddg_monomer Weights: soft_rep_design	Protocol: cartesian_ddg Weights: talaris2013	Protocol: cartesian_ddg Weights: beta_nov16

Amber (thermodynamic integration) results

Conclusions

• Evolutionary information can be very useful for protein structure prediction and design.

• The ELASPIC webserver provides a user-friendly interface for evaluating the structural impact of mutations.

• Rosetta cartesian_ddg protocol with the talaris2013 or beta_nov15 weights produces the most accurate results.

• Relatively cheap to calculate and very useful (if a mutation is predicted to be very deleterious, you probably don't want to do it, even if you think it will stabilize the protein).

• Can fold proteins using evolutionary information.

• ...

  • The inverse is not neccessarily true.
  • Only a fraction of missense mutations.

Additional resources

• http://kimlab.org/

• http://elaspic.kimlab.org/

• https://gitlab.com/kimlab/elaspic/

• https://gitlab.com/ostrokach/msc_thesis/

• https://ostrokach-presentations.gitlab.io/cagi5/

Acknowledgements

Supervisor

• Philip M. Kim

Members of kimlab

• David Becerra
• Recep Colak
• Carles Corbi
• Michael Garton
• Clare Juhyun Jeon
• Mark Sun
• Joan Teyra

Project students

• Daniel Witvliet