Predictions of the Docking of an Engineered Antibody to Anthrax Toxin

1
Predictions of the Docking of an Engineered
Antibody to Anthrax Toxin
Arvind Sivasubramanian Jeffrey J. Gray ACS 2006
San Francisco
PROGRAM IN
Molecular Biophysics
2
Antibody therapeutics

• At least 20 therapeutic antibodies have been
  approved by regulatory authorities, with another
  150 molecules in the clinic

Key Growth driver!
mAbs entering Clinical study
Reichert, Nat Biotech, 2005 Das et al, Intl. Sci.
Comm, 2006
3
Blind Prediction of Docking (CAPRI)

• Laminin Nidogen, model 2
• 53 contacts, rmsd 4.6 Å, interface rmsd 0.66 Å
• One of two best predictions out of 20 groups in
  international challenge

RedNative laminin GreenPredicted laminin
BlueNidogen
D800, N802, V804 constrained near interface
Methods in Gray et al. 2003 JMB
Xtal by T. Springer, Harvard
4
Typical requirements and assumptions of docking
predictions

• Monomer structures known
• Strong binding (Kd lt µm)
• Small proteins (1-2 domains per partner)
• Little significant backbone movement
• Some experimental data are helpful

5
Mechanism of toxin action

• Therapy Prevent PA binding to cells using mAbs

6
Structure of pre-pore

• Santelli et al, Nature 2004.

7
mAb 14B7 functions by blocking PA receptor
binding site

• Small loop (679-693) residues are critical for
  receptor binding.
• 14B7 family antibodies function by blocking
  receptor binding site (Rosovitz et al, JBC 2003)

PA Domain IV
PA Domain II
Small loop
CMG Receptor
8
Affinity maturation in 14B7 family
9
Mutations in mAbs14B7, 1H and M18
Heavy
Light
Mutations
L2 mutations
10
Errors in mAb 14B7 homology model
H1
H3

• White and Gray Crystal structure
• Green and Yellow WAM model
• CDR H3 rmsd is 2.7Å

11
Objectives

• Predict the structure of the 14B7-PA complex
• Propose structural explanation for affinity
  maturation in 14B7 family?
• Investigate effect of homology model inaccuracies
  on docking.

12
Model 14B7 using WAM
PA-14B7 (Xtal or WAM) docking using RosettaDock
1x105 modelsbest scorescluster200 structures
Filter structures using known PA and 14B7 hotspot
residues
6 preliminary models
Calculate ??G of mutations using RosettaInterface
Consistent with experiments?
No
Yes
Reject
Validate with new computational and experimental
mutagenesis
13
Docking Algorithm Overview
Random Start Position
Low-Resolution Monte Carlo Search
High-Resolution Refinement
105
Clustering
Predictions
14
Low-Resolution Decoy
High-Resolution Refinement
Small Rigid-Body Move

• Simultaneous rigid-body and side-chain
  optimization

Repack Side Chains
Rigid-Body Minimization
Monte Carlo Accept?
Filter
Reject
50x
Clustering
15
Full-Atom scoring

• Terms (combined linearly)
• Attractive Van der Waals tight packing
• Repulsive Van der Waals avoid clashes
• Solvation (pair-wise Gaussian exclusion) bury
  non-polars (Lazaridis/Karplus)
• Hydrogen bonding (Kortemme)
• Side-chain conformational energy use common
  rotamers
• Residue pair scores statistical catch-all
• Solvent-Accessible Surface Area (ASPs)
• Electrostatics

16
Experimental information is used to guide
predictions post-docking
Antigen
Antibody
Rosovitz et. al. JBC 2003. Maynard et. al.
unpublished
17
Contact filter based on known hotspots
18
Consensus between models

• Homology model docking generates low resolution
  representations of crystal structure docking
  models

19
Computational mutagenesis using known hotspots
20
Final models
Model 1 Antigen hotspots contact chain L
Model 3 Antigen hotspots contact chain H
21
Final models
Models 1 and 3 share 67 and 87 of receptor and
ligand interface residue identities BUT 0
common 14B7-PA interface contacts
22
New antigen epitope regions identified
712-720
681-688 (Known)
648-660
23
Model 3 hydrogen bonding
R99
24
Model 3 hydrophobic interactions
L685
Y688
Y52
E654
L652
L97
W33
Y100
Y50
25
Model 1 hydrogen bonding interactions
K653
D648
N691
N719
E654
D683
Y52
R53
W33
Y49
R99
N92
26
Model 1 hydrophobic interactions
I656
E654
L97
Y688
L652
L98
Y50
Y32
W33
Y100
27
Site-directed antigen mutations for model
validation
Neutral - Reduced binding
28
Implications for affinity maturation pathway

• Mutated residues do not contact the antigen
  significantly

Model 1
Model 3
29
Affinity maturation hypotheses

• Loop entropy contribution?
• VL-VH interface stabilization?

30
Affinity maturation

• Previously proposed explanations
• Increased hydrophobic surface burial (HyHEL-HEL,
  Li et. al, 2003)
• Enhanced electrostatic complementarity (TEM-BLIP,
  Joughin et. al, 2005)
• Cumulative effect of minor structural alterations
  (4M5.3-Fluorescein, Midelfort et. al., 2004)
• Potential explanations for the mAb 14B7-PA
  interaction
• Rigidification of CDR L2 (Q55L and S56P
  mutations) or CDR L3 (L94P mutation)
• Rigidification of CDR H3 contributes 2kcal/mol
  (? ? G -1.5RT ln(n), Wang et. al. 2005)
• Stabilization of VL-VH interface (L46F, Q55L and
  L94P mutations)

VL-VH interface stabilization?
Electrostatics (K64E)?
31
Homology model docking produces low-res
representations of crystal structure docking
Model 1
Model 3
32
CDR H3 conformation modeling critical for
high-res docking
WAM
WAM
H3 chimera
H3 chimera
14B7
14B7
33
WAM model errors frustrate high-resolution
docking prediction
34
Conclusions

• Two candidate structures for 14B7-PA complex
• Computational mutagenesis suggests good agreement
  with experimental data on interface.
• Affinity maturation not mediated directly by
  contacting residues
• Alternate explanations are needed
• Homology docking agrees with crystal structure
  docking at low-resolution
• Improvements in CDR H3 modeling
• Docking with backbone flexibility

35
Acknowledgments

• Georgiou Iverson Labs University of Texas
• Jennifer Maynard (UMn)
• Andrew Hayhurst
• Johns Hopkins
• Carlos Castaneda
• Sony Somarouthu
• Mike Daily
• Aroop Sircar
• David Baker Lab University of Washington
• Tanja Kortemme (UCSF)
• Brian Kuhlman (UNC)
• Carol Rohl (UCSC)
• Rees Whitelegg
• NIH/NHGRI K01-HG02316