mAb 806 binds a 16 residue domain II peptide which contains the complete epitope'

1
Computational structure prediction of therapeutic
antibody-antigen complexes Arvind
Sivasubramanian1 Jeffrey J. Gray1,2 1Chemical
Biomolecular Engineering 2Program in Molecular
Computational Biophysics, Johns Hopkins
University
Introduction
Prediction of mAb 14B7-PA toxin complex structure
Prediction of mAb 806-EGFR complex structure
mAb 14B7 blocks binding of anthrax toxin to
cellular receptor
mAb 806 binds EGFR, a popular target for tumor
therapy
Computational structure prediction is useful for
Epidermal Growth Factor Receptor Signaling
Both mAb 14b7 cellular receptor CMG2 bind PA
residues 681-688
Protective Antigen (PA) is a component of the
tripartite anthrax toxin.

• Complete paratope epitope identification.
• Rational engineering of antibody affinity based
  on structure.
• Determination of disease and antibody therapeutic
  mechanism.

• mAb 806 binds EGFR when it is over-expressed or
  truncated (as in cancer), but not EGFR found in
  endogenous skin or liver tissue.
• The EGFR conformational state recognized by mAb
  806 has not been established conclusively.

Ligand
Dimerization
Receptor domain
mAbs

PA binds cellular receptors (CMG2 or TEM8)
heptamerizes to form a prepore, which in turn
binds Lethal Factor (LF) Edema Factor (EF)
Extracellular
A pre-pore to pore conformational switch allows
LF EF to enter kill the host cell.
Membrane
PA residues important for binding to Only
CMG2 Red Only 14B7 Green Both Yellow
Intracellular
Kinase domain
TKIs
Rosetta algorithm Protein folding, design
docking
K
K
K
K
Phosphorylation
Mishima et al, Cancer. Res., 2001 Johns et al,
Int. J. Cancer, 2002 Panousis et al, British J.
Cancer, 2005.
Protein folding Rosetta Ab initio
Protein design RosettaDesign
Protein docking RosettaDock
Tumor response
Cell division
Yellow
EGFR conformational change is critical to mAb 806
binding
Ligand (e.g. EGF) induced closed-to-open
conformation change
PA binding surface
Open (Hypothetical, from dimer)
Closed

• mAb 806 binds a 16 residue domain II peptide
  which contains the complete epitope.
• mAb 806 does not bind the closed EGFR, but EGFR
  mutations that weaken auto-inhibition lead to
  stronger binding.

• mAb 14B7 affinity matured variants protect
  rats from lethal toxin challenge by binding to PA.

Ligand
Santelli et al, Nature , 2004 Lacy et al, PNAS,
2004 Maynard et al, Nat. Biotech, 2002 Rosovitz
et al, JBC, 2003, Gao et al, Biophys. J, 2006.
Dimerization arm
Ligand
?
Best docking model generated with crystal
structure of mAb 14B7 PA is consistent with 60
of known mutations
HOX-B1
TOP7
Cohesin-dockerin complex
Bradley et al, Science, 2005 Kuhlman et al,
Science, 2003 Daily et al, Proteins, 2005.
Peptide epitope
The best model (Model 2), predicts that PA
residues 648-660 712-720 are also important for
mAb 14B7 binding.
Comparison with known experimental data
RosettaDock generates models of protein complexes
Peptide
Known mutations that decrease 14b7-PA binding
14B7 PA Hotspots shown as spheres
Auto-inhibitory tether
Ferguson et al, Molecular Cell, 2003 Garrett et
al, Cell, 2002 Walker et al, JBC, 2004.
Random Start Position
Terms in the energy function include
PA

• van der Waals tight packing/avoid clashes
• Solvation bury non-polars
• Hydrogen bonding
• Backbone dependent rotamers
• Solvent accessible surface area
• Electrostatics

PA
mAb 806 homology model is docked to known peptide
epitope
Newly Predicted PA hotspots
Known PA hotspots
Low-Resolution Monte Carlo Search

• Canonical conformations for mAb 806 CDRs,
  except H3.
• H3 Length is 7. Seven residue H3's have minor
  role
• in antigen binding.

14B7
High-Resolution Refinement
Gray et al, JMB, 2003.
5 unique peptide conformations capture
flexibility.
VH
VL
? Consistent with the experimental result X
Not consistent with the experimental result
105
Predictions
Blind post-docking mutagenesis predictions
validate docking model
Clustering
Site-directed mutants were constructed based on
computational mutagenesis of initial models. The
final model, Model 3, has the best agreement with
both known blindly predicted post-docking
mutagenesis experiments.
RosettaInterface estimates ??G values for
mutations at protein interfaces
Predicted Known PA hotspots interact with known
14B7 hotspots
Known 14B7 hotspots compete with CMG2 for H-bonds
with PA hotspots
Comparison with known experimental data
Validation with new post-docking experiments
D1.3 HEL
HYHEL 10 HEL
Crystal structure or docking model
Interface mutation
New mutations are neutral towards 806-EGFR binding
Known mutations that decrease 806-EGFR binding
PA-CMG2 (Crystal structure)
PA-14B7 (Model 2)
Expt. ??G
N682
RosettaInterface
PA
PA
D658
N682
D683
E654
D658
??Gcalc
R99
G116
??Gcalc (RosettaInterface prediction)
? Consistent with the experimental result X
Not consistent with the experimental result
Sivasubramanian et al, Structure, 2006 Chao et
al, JMB, 2004.
T716
D683
N92
Predictions 0 lt ??Gcalc lt 1.0 Neutral
mutation ??Gcalc gt 1.0 Binding loss mutation
S87
E654
T716
R50
Kortemme et al, PNAS, 2002
Y100
mAb 806 binds "Open" EGFR in the model,
consistent with hypothesis
K51
T118
Computational protocol combines RosettaDock
validation using RosettaInterface
Q27
CMG2
Q88
VH
VL
The final docking model was structurally
superposed (using the peptide) with the EGFR
end-states to infer the mAb 806-full-length EGFR
global orientation.
Model predicts E654 and D658 as new hotspots
mAb 806-EGFR (Open) Minimal Ab-EGFR van der Waals
clashes
mAb 806-EGFR (Closed) Extensive Ab-EGFR van der
Waals clashes
Antibody Coordinates
Antigen Coordinates
Affinity maturation may not be directly due to
enhanced Ab-Ag interface interactions.
EGFR
EGFR
Johns et al, JBC, 2004.
Generate a docking model using RosettaDock
PA

• Variant mAb M18 with 120-fold higher affinity
  has 10 mutations (5 each in CDR framework)
  compared to wt 14B7.
• These mutations do not change antibody
  conformation.
• Mutated residue positions do not make
  energetically significant interface contacts in
  model.

Dimerization arm
Calculate ??G of interface mutations in model
using RosettaInterface
Interface mutations
VL
VH
Consistent with experiments?
VL
VL
No
Yes
VH
Validate model with new blind computational
experimental mutagenesis
Reject
VH
VH
Peptide
Harvey et al, PNAS, 2004.
VL
Spheres Affinity increasing mutations Darker
shading CDR residues
Acknowledgements Funding
Conclusions
The predicted interaction suggests that

• Best docking model agrees with 60 of known
  experimental binding loss mutations.
• PA residues D648, L652, E654, D658 T717 might
  be important for 14B7 binding.
• Affinity maturation in the 14B7 family may not be
  directly due to enhanced Ab-Ag interface
  interactions

• mAb 806 binds the Open EGFR, and sterically
  prevents dimerization.
• Computational structure prediction is useful to
  model large targets with both Ab Ag
  conformational uncertainty.

mAb 14B7 project Jennifer Maynard, George
Georgiou, Brent Iverson, Clint Leysath (U.Texas)
mAb 806 project Dane Wittrup, Ginger Chao (MIT)
Funding NIH R01 GM078221 NIH K01 HG002316 Sidney
Kimmel Cancer Center
Antibody homology modeling Web Antibody Modeling
(WAM)
Reference Sivasubramanian et al, "Structural
Model of the mAb 806-EGFR Complex Using
Computational Docking followed by Computational
and Experimental Mutagenesis". Structure 14,
401-414, 2006.
Preprint available on request
Figures PyMOL