Transcript Darren Flower - UK-QSAR

QSAR and the Prediction
of T cell Epitopes
Darren R Flower
http://www.jenner.ac.uk/res-bio
[email protected]
VACCINE MARKET
VACCINES
1%
$5 Billion
Growth in sales
Vaccines: 12% yr-1
Drugs: 5% yr-1
DRUGS
15
$ 15 B
$5B
$ 1.75 B
5
1.75
1990
2000
2010
99%
$350 Billion
HUMAN VACCINES
ARE MOVING FROM A MARGINAL
TO A MAJOR R & D DRIVEN SECTOR
AIDS
ANTIBIOTIC
RESISTANCE
BIOTERRORISM
A limited number
A few innovative
of vaccines targeting
vaccines with
major diseases
blockbuster potential
1980
# of Biotech
companies
R &D pipeline:
100s of new vaccines
2000
10
150
IMMUNOVACCINOLOGY
Vaccines induce protective immunity.
Protective immunity is an enhanced
adaptive immune response to re-infection.
WHOLE
ORGANISM
attenuated
SUBUNIT
VACCINE
EPITOPE
VACCINE
Delivered as recombinant protein / vector
or as naked DNA
+ adjuvants &/or “danger signals”
Types of Peptide Epitope
Conformational
Antibody
or “B cell”
Epitope
Linear
B cell Epitope
NonConformational
T cell Epitope
Class I MHCs
Class II MHCs
all cells
Professional Antigen
Presenting cells
Foreign and self proteins
Foreign proteins
8-10 amino acids
8-20 amino acids
T cell response
CD8
Class I
TCR
PEPTIDE
Class I
AFFINITY
MEASURE
NARROWEST
POINT
“PEPTIDE”
FUNNEL
PROTEIN
PROTEA
SOME
TAP
MHC
TCR
IC50
IC50
KD
KD
CLEAVAGE
PATTERNS
BL50
C50
SC50
t1/2
etc.
T cell
response
Half maximal
lysis or
qualitative
measure
(thymidine
incoporation,
cell killing,
etc
RESPONSE
MEASURE
PREDICTING EPITOPES
Traditional Motifs:
X{Y/F}XXPXXWS
Frequency Matrices, Profiles
“AI” Solutions:
Neural Networks, HMMs, etc
Epitope prediction is a chemical problem.
We have taken a quantitative approach.
Molecular Dynamics
Quantitative Structure
Activity Relationships
3D-QSAR:
ComFA / CoMSIA
2D-QSAR:
Free Wilson Analysis
QUANTITITATIVE MEASURES
OF PEPTIDE-MHC AFFINITY
FROM THE LITERATURE
Compile data
RELATIONAL
DATABASE
JenPep
Extract data for
particular Allele
QSAR
CoMSIA / CoMFA
ADDITIVE METHOD
TESTABLE PREDICTIONS
EDWARD JENNER INSTITUTE
FOR VACCINE RESEARCH
JENPEP
Helen McSparron
Martin Blythe
Christianna Zygouri
JENPEP
VERSION 1.0
2061 T-Cell Epitopes
5848 MHC Binding data (IC50, BL50, t1/2, etc)
432 TAP Binding data
ACCESS Relational Database
GUI using HTML and ASP
on our website: www.jenner.ac.uk/JenPep
MJ Blythe, IA Doytchinova, and DR Flower.
JenPep: a database of quantitative functional peptide data for immunology.
Bionformatics 2002 18 434-439
JENPEP
“VERSION 2.0”
3018 T cell epitopes
12210 MHC Binding data (IC50, BL50, t1/2, Kd, etc)
441 TAP Binding data
1656 B cell epitopes
300 pMHC-TCR Binding data
bespoke postgreSQL relational database
GUI using perl and HTML
on our website: www.jenner.ac.uk/JenPep
H McSparron, C Zygouri, D Taylor, MJ Blythe, IA Doytchinova, and DR Flower.
JenPep+: Novel developments in quantitative immunological databases
Nucleic Acids Research, commissioned
Develop database system further:
extend existing databases
(T cell, MHC, TAP, B cell, pMHC-TCR)
with new data and further retrospective analysis
add new database sections:
non-natural peptides and non-natural MHC mutants
antibody binding
whole protein antigens
Host - Superantigen / Virulence Factor Binding Data
Co-receptor Binding Data
etc.
Binding Affinity of peptides
vs. Host Immunogenicity
MHCs: hundreds of alleles.
Each with a different peptide binding selectivity.
T cell epitopes bind well to MHCs.
95% of all known T cell epitopes bind to
MHC with an IC50 < 500nM.
Exact T cell response is dependent on the T cell repertoire.
Therefore, prediction of MHC binding
is “best” option for predicting T cell epitopes.
EDWARD JENNER INSTITUTE
FOR VACCINE RESEARCH
PREDICTING
T cell
EPITOPES
Irini Doytchinova
Christianna Zygouri
PingPing Guan
T - Cell Epitope Search
CAVEAT
our peptide sets are larger than
is typical in the pharmaceutical literature
the peptides themselves are physically large
physical properties of peptides are extreme:
multiple charges, zwitterions, huge range in hydrophobicity, etc.
Sequence & thus properties
are heavily biased in our peptide sets
Affinity data is “poor”:
multiple measurements of same peptide with orders
of magnitude differences, some values are clearly wrong, mix of
different standard peptides in radioligand competition assays, etc.
performing a “meta-analysis”:
probably many different binding modes
forced into one QSAR model
Predicting T cell Epitopes Using QSAR
CoMFA / CoMSIA
Towards the quantitative prediction of T-cell epitopes:
CoMFA and CoMSIA studies of peptides with
affinity to class I MHC molecule HLA-A*0201.
Doytchinova, I.A and Flower, D. R.
J. Med. Chem. 2001, 44, 3572-3581.
Physicochemical Explanation of Peptide Binding to HLA-A*0201
Major Histocompatibility Complex. A Three – Dimensional
Quantitative
Structure – Activity Relationship Study.
Doytchinova, I.A and Flower, D. R.
Proteins, in press.
FREE WILSON ANALYSIS
An Additive Method for the Prediction of
Protein-Peptide Binding Affinity.
Application to the MHC Class I Molecule HLA-A*0201
Irini A. Doytchinova*, Martin J. Blythe and Darren R. Flower
J. Proteome Research 2002, 1, 263-272.
HLA-A*0201
most common
allele in Caucasian
population: 40%
~5x more binding
data than for any
other allele
152 peptides
with affinity to
the HLA-A2.1
Comparison of
CoMFA & CoMSIA
for HLA-A*0201
Training set
102
peptides
CoMSIA model
9
9
8
8
predicted pIC50
predicted pIC50
CoMFA model
7
7
6
6
5
5
5
6
7
experimental pIC50
2 r = 0.694
r
pred
Test set
50
peptides
8
< 0.5
NC = 6 q2 = 0.480 r2 = 0.911
9
5
6
7
8
9
experimental pIC50
r = 0.783
r2pred = 0.679
NC = 5 q2 = 0.542 r2 = 0.870
Full CoMSIA Analysis of HLA-A*0201
Electrostatic Map
Steric Map
Hydrogen Bond Map
Hydrophobic Map
NC = 7 q2 = 0.683 r2 = 0.891 n = 236
ADDITIVE METHOD FOR
BINDING AFFINITY PREDICTION
P2
H
H
P4
O
N
H
O
N
N
H
H
H
H
P5
9
8
7
i 1
i 1
i 1
H
O
N
N
O
P3
P8
O
N
N
O
P1
P6
H
N
N
O
H
O
O
P7
pIC 50  const   Pi   Pi Pi 1   Pi Pi  2
HLA-A*0201:
OH
NC = 5 q2 = 0.337 r2 = 0.898 n = 340
P9
Amino acids contributions
1-2 Interactions
1-3 interactions
How does
the additive method work?
YLSPGPVTV with pIC50 exp = 7.642
pIC50 = const + 1Y + 2L + 3S + 4P + 5G + 6P +7V + 8T + 9V
+ 1Y2L + 2L3S + 3S4P + 4P5G + 5G6P + 6P7V + 7V8T + 8T9V
+ 1Y3S + 2L4P + 3S5G + 4P6P + 5G7V + 6P8T + 7V9V
pIC50 = 6.213 +0.304 + 0.219 – 0.164 + 0.135 + 0.013 - 0.008 + 0.096 + 0.035 + 0.263
+ 0.240 – 0.015 + 0 + 0 + 0.101 + 0.075 + 0.059 + 0.102
+ 0.031 + 0.044 – 0.107 + 0.046 + 0.011 + 0.008 - 0.001
= 7.700
CoMSIA & ADDITIVE METHOD
ARE COMPLEMENTARY
CoMSIA is “slow” but is “better” at extrapolating.
ADDITIVE is very fast
(analyze whole microbial genome in a few minutes)
but is worse at extrapolating to peptide sequences
very different to training data (missing values)
Our models are not perfect but
our results are at least as good as anyone else
working in predicting MHC binding
Trying to develop a range of “universal” models
each covering a different allele
A2-Supertype models
CoMSIA models.
parameter
A*0201
A*0202
A*0203
A*0206
A*6802
n
236
63
60
54
45
q2
0.683
0.534
0.621
0.523
0.385
NC
7
8
6
12
4
SEP
0.443
0.509
0.595
0.505
0.652
SEP/affinity range %
9.9
13.6
13.0
13.3
14.8
r2
0.891
0.935
0.966
0.991
0.944
parameter
A*0201
A*0202
A*0203
A*0206
A*6802
n
335
69
62
57
46
q2
0.377
0.317
0.327
0.475
0.500
NC
6
9
6
6
7
SEP
0.694
0.606
0.841
0.576
0.647
SEP/affinity range %
15.5
16.2
18.3
15.2
14.7
r2
0.731
0.943
0.963
0.989
0.983
Additive models.
MHCPred: an on-line server for
peptide MHC binding prediction
Models:
A*0101, A*0201,
A*0202, A*0203,
A*0206, A*0301,
A*1101, A*3301,
A*6801, A*6802,
B*3501
www.jenner.ac.uk/MHCPred
P Guan, IA Doytchinova, C Zygouri, and DR Flower.
MHCPred: bringing a quantitative dimension to online prediction of MHC Binding.
To be submitted
FUTURE DEVELOPMENTS
OF THIS WORK
Make “true” predictions - design new peptides
and test them experimentally
Develop models for uncharacterized MHC alleles
using peptides generated with Experimental Design
In
Progress
Develop Additive Method to be descriptor based
Develop “better” QSAR models using “clean”
thermodynamic data from ITC
and designed peptides
Planned