Fig 1.
(A) Native or stressed conformation (PDB: 1AZX-I). (B) Cleaved state (PDB: 1EZX). (C) Latent or relaxed conformation (PDB: 1AZX-L). (D) Dimer structure, exchanging two central β-strands (PDB: 2ZNH). White, red and dark blue: serpin (antithrombin or α-1-antitrypsin). Cyan: central wild-type β-strands. Magenta: reactive center loop. Light blue: trypsin.
Fig 2.
Selection of variants for this study.
Variants were selected from a cohort of 350 patients with antithrombin deficiency and were previously described in different works.
Table 1.
Functional antigenic and biochemical data of SERPINC1 mutations identified in patients with antithrombin deficiency and selected for predictions using AlphaFold.
Fig 3.
Results for reduced (green) and full (blue) versions of AlphaFold.
(A) Wild-type antithrombin predictions. (B) M5 predictions. No major differences are perceived for antithrombin between reduced and full AlphaFold versions.
Table 2.
Summary of AlphaFold metrics between predictions of reduced and full (reference) versions.
We observe no differences between full and reduced AlphaFold models.
Fig 4.
Comparisons between wild-type antithrombin prediction and different structures.
(A) 1AZX. (B) M1. (C) M2. (D) M3. (E) M4. (F) M5. White: 1AZX. Blue: AlphaFold prediction for wild-type antithrombin. Green: corresponding mutation. Red: mutated residues. Strong correlation between wild-type, native structure and mutant predictions is observed, even for bigger changes, such as M4 and M5. M4 is included as a loop in the main structure, whilst M5 is modeled as an α-helix, as this configuration has higher thermodynamical stability than a 10-residue loop.
Table 3.
Summary of metrics for AlphaFold and molecular dynamics.
AlphaFold metrics are referred to its predictions, except for PDB:1AZX and PDB:4EB1. Molecular Dynamics metrics refer to differences between initial and final state of the proteins.
Table 4.
Summary of RMSD metrics of the difference of energy in REU (Rosetta energy units) between the wild-type and each mutant.
Differences between ROSIE 2’s predictions and AlphaFold’s are presented as RMSD and lDDT.
Fig 5.
Cross-section of M5 crystal (4EB1) compared to its AlphaFold prediction.
White: M5 crystal (4EB1). Green: M5 prediction. Cyan: natural serpins β-strands. Orange: 10 amino acid insertion on crystal. Red: 10 amino acid insertion on M5 prediction. 4EB1 shows how M5 insertion leads to an extra β-strand in patients, resembling a latent structure, whilst AlphaFold predicts M5 insertion to be fold as a new α-helix, with no change of the native, wild-type-like structure.
Table 5.
Dihedral angles (Phi/Psi) summary for the wild-type and M5 mutant.
A comparison of crystal structures and AlphaFold predictions.
Table 6.
Secondary structure comparison for the wild-type and M5 mutant.
Analysis of crystal structures and AlphaFold predictions.
Table 7.
Analysis of the hydrogen bond network, a comparative summary.
Comparison of pairs of residues linked by at least one hydrogen bond in each structure. Structure A is the reference on each analysis. Structure B is the corresponding AlphaFold prediction.
Table 8.
Summary of RMSD metrics between initial and most energetically stable MD states for all studied proteins.
Local minimums were calculated using FEL calculations.
Table 9.
Summary of lDDT metrics between initial and state with less energy for all considered proteins.