Antibody Humanization

Paolo Marcatili

HAMA

HUMANIZATION

Murine monoclonal Abs

Rationale

We can engineer the antibody so that it "looks" human.

CH2  

CL  

CH1  

CH3  

VH  

VL  

Antibodies are modular

!

VH VL

ABS

Chimeric antibody

Human Constant domain Murine variable domains

Chimeric antibody

Human Constant domain Murine variable domains Still antigenic!

Chimeric vs humanised

Chimeric vs humanised

How much can we humanise without loosing affinity?

Humanization strategies

CDR grafting: Use only the murine CDRS SDR grafting: CDR + structurally important residues (Canonical Structures)

Resurfacing: Change all surface residues (apart to CDR) to human Superhumanization: Check for each segment in the Ab and add mutation

to maximize local similarity to the human

Humanization strategies

CDR grafting: Use only the murine CDRS SDR grafting: CDR + structurally important residues (Canonical Structures)

Resurfacing: Change all surface residues (apart to CDR) to human Superhumanization: Check for each segment in the Ab and add mutation

to maximize local similarity to the human

Humanization strategies

CDR grafting: Use only the murine CDRS usually the antibody loses affinity SDR grafting: CDR + structurally important residues (Canonical Structures) often binding is poor

Antibody Paratope

The binding probability of each residue depends on all the environment. Even distant residues might change the binding mode

Framework residues affect antigen binding (Mouse)

Antigen

CDR

FR residue

Framework residues affect antigen binding (Mouse)

FR residue

CDR

Antigen

Humanization process

“Trial and error cycle” Human

Framework

In vitro evaluation

Back mutations

Regions for grafting

A novel, integrated method to facilitate and automate the humanization process.

Project Goal

Known data (Antibody sequences, Antigen contacting

residues)

Machine learning techniques

Prediction method

Input data (Antibody

sequences)

Output data (Antigen contacting

residues)

Project Workflow

Known data 313 antibody-antigen

complexes Machine learning

Random Forest algorithm

proABC (prediction of

AntiBody Contacts)

Input data (Antibody sequence)

Output data (residue-based

contact probability)

Pipeline

Performance Evaluation

AUC: 84% MCC: 52%

���������������

�������������

��� ��� ��� ��� ��� ���

���

���

���

���

���

���

Predicted to be in contact but not in contact

Pre

dict

ed to

be

in c

onta

ct a

nd in

co

ntac

t

Prediction of Antibody Contacts

Olimpieri et al. Prediction of site-specific interactions in antibody-antigen complexes: the proABC method and server. Bioinformatics 29, 2285-2291

www.biocomputing.it/proABC

Antibody Fingerprint

Humanization of an Anti-Vascular Endothelial Growth Factor Monoclonal Antibody for the Therapy of Solid Tumors and Other Disorders. Leonard G. Presta, Helen Chen, Shane J. O’Connor et al. Cancer Res 1997; 57:4593-4599 Antibody against VEGF developed on mouse Drafting of the CDRs Test of affinity Back mutation (8 rounds)

proABC for humanization Humanization of an Anti-Vascular Endothelial Growth Factor Monoclonal

Antibody for the Therapy of Solid Tumors and Other Disorders.

Leonard G. Presta, Helen Chen, Shane J. O’Connor et al.

Cancer Res 1997; 57:4593-4599

Why this paper?

1 6 12 18 24 30 34 40 46 52 55 61 67 73 79 83 89 95 100C 100U 108

020

4060

80Mouse

Pro

babi

lity

(Per

cent

age)

Position

1 6 12 18 24 30 34 40 46 52 55 61 67 73 79 83 89 95 100C 100U 108

020

4060

80Grafted vs Mouse

Pro

babi

lity

(Per

cent

age)

Position

1 6 12 18 24 30 34 40 46 52 55 61 67 73 79 83 89 95 100C 100U 108

020

4060

803 back mutations

Pro

babi

lity

(Per

cent

age)

Position

1 6 12 18 24 30 34 40 46 52 55 61 67 73 79 83 89 95 100C 100U 108

020

4060

808 back mutations

Pro

babi

lity

(Per

cent

age)

Position

Humanization process

“Trial and error cycle” Human

Framework

In vitro evaluation

Back mutations

Regions for grafting

The Tabhu web server

Template Selection

www.biocomputing.it/tabhu

The Tabhu web server

Antibody humanization is needed for any Ab drug It is still a "trial and error" procedure non CDR residues can greatly affect the binding Structural analysis is extremely important

Conclusion