Download Prediction of B cell epitopes

Pernille Haste Andersen,
Ph.d. student
Immunological Bioinformatics
CBS, DTU
CENTER FOR BIOLOGICAL SEQUENCE ANALYSIS TECHNICAL UNIVERSITY OF DENMARK DTU
B cell epitopes and predictions

Antibodies are produced by B lymphocytes
(B cells)

Antibodies circulate in the blood


They are referred to as “the first line of
defense” against infection
Antibodies play a central role in immunity
by attaching to pathogens and recruiting
effector systems that kill the invader
CENTER FOR BIOLOGICAL SEQUENCE ANALYSIS TECHNICAL UNIVERSITY OF DENMARK DTU
B cells and antibodies
B cell epitopes

Accessible and
recognizable structural
feature of a pathogen
molecule (antigen)

Antibodies are developed
to bind the epitope with
high affinity by using the
complementarity
determining regions
(CDRs)
Antibody Fab
fragment
CENTER FOR BIOLOGICAL SEQUENCE ANALYSIS TECHNICAL UNIVERSITY OF DENMARK DTU
What is a B cell epitope?
B cell epitope
 Prediction of B cell epitopes can potentially
guide experimental epitope mapping
 Predictions of antigenicity in proteins can be
used for selecting subunits in rational vaccine
design
 Predictions of B cell epitopes may also be
valuable for interpretation of results from
experiments based on antibody affinity binding
such as ELISA, RIA
CENTER FOR BIOLOGICAL SEQUENCE ANALYSIS TECHNICAL UNIVERSITY OF DENMARK DTU
Motivations for prediction of B cell epitopes
>PATHOGEN PROTEIN
KVFGRCELAAAMKRHGLDNYR
GYSLGNWVCAAKFESNF
Rational Vaccine
Design
CENTER FOR BIOLOGICAL SEQUENCE ANALYSIS TECHNICAL UNIVERSITY OF DENMARK DTU
Computational Rational
Vaccine Design
 Classified into linear (~10%) and
discontinuous epitopes (~90%)
 Databases: AntiJen, IEDB, BciPep, Los
Alamos HIV database, Protein Data Bank
 Large amount of data available for linear
epitopes
 Few data available for discontinuous epitopes
 In general, B cell epitope prediction methods
have relatively low performances
CENTER FOR BIOLOGICAL SEQUENCE ANALYSIS TECHNICAL UNIVERSITY OF DENMARK DTU
B cell epitopes, linear or
discontinuous?
An example: The epitope of
the outer surface protein A
from Borrelia Burgdorferi
(1OSP)
•
SLDEKNSVSVDLPGEMK
VLVSKEKNKDGKYDLIATVD
KLELKGTSDKNNGSGVLEGV
KADKCKVKLTISDDLGQTTLE
VFKEDGKTLVSKKVTSKDKS
STEEKFNEKGEVSEKIITRADG
TRLEYTGIKSDGSGKAKEVLK
G
•
..\Discotope\1OSP_epitope\1OSP_epitope.psw
CENTER FOR BIOLOGICAL SEQUENCE ANALYSIS TECHNICAL UNIVERSITY OF DENMARK DTU
Discontinuous B cell epitopes
Binding strength




Salt bridges
Hydrogen bonds
Hydrophobic interactions
Van der Waals forces
CENTER FOR BIOLOGICAL SEQUENCE ANALYSIS TECHNICAL UNIVERSITY OF DENMARK DTU
The binding interactions
 Many of the charged groups and hydrogen
bonding partners are present on highly flexible
amino acid side chains.
 Most crystal structures of epitopes and
antibodies in free and complexed forms have
shown conformational rearrangements upon
binding.
 “Induced fit” model of interactions.
CENTER FOR BIOLOGICAL SEQUENCE ANALYSIS TECHNICAL UNIVERSITY OF DENMARK DTU
B-cell epitopes are dynamic
B-cell epitope – structural feature of a molecule or
pathogen, accessible and recognizable by B-cells
Linear epitopes
One segment of the
amino acid chain
Discontinuous epitope
(with linear determinant)
Discontinuous epitope
Several small segments brought
into proximity by the protein fold
CENTER FOR BIOLOGICAL SEQUENCE ANALYSIS TECHNICAL UNIVERSITY OF DENMARK DTU
B-cell epitope classification
• Linear epitopes:
– Chop sequence into small pieces and measure
binding to antibody
• Discontinuous epitopes:
– Measure binding of whole protein to antibody
• The best annotation method : X-ray crystal
structure of the antibody-epitope complex
CENTER FOR BIOLOGICAL SEQUENCE ANALYSIS TECHNICAL UNIVERSITY OF DENMARK DTU
B-cell epitope annotation
• Databases: AntiJen, IEDB, BciPep, Los
Alamos HIV database, Protein Data
Bank
• Large amount of data available for
linear epitopes
• Few data available for discontinuous
CENTER FOR BIOLOGICAL SEQUENCE ANALYSIS TECHNICAL UNIVERSITY OF DENMARK DTU
B-cell epitope data bases
CENTER FOR BIOLOGICAL SEQUENCE ANALYSIS TECHNICAL UNIVERSITY OF DENMARK DTU
B cell epitope prediction
 Protein hydrophobicity – hydrophilicity algorithms
Parker, Fauchere, Janin, Kyte and Doolittle, Manavalan
Sweet and Eisenberg, Goldman, Engelman and Steitz (GES), von Heijne
 Protein flexibility prediction algorithm
Karplus and Schulz
 Protein secondary structure prediction algorithms
GOR II method (Garnier and Robson), Chou and Fasman, Pellequer
 Protein “antigenicity” prediction :
Hopp and Woods, Welling
TSQDLSVFPLASCCKDNIASTSVTLGCLVTG
YLPMSTTVTWDTGSLNKNVTTFPTTFHETY
GLHSIVSQVTASGKWAKQRFTCSVAHAEST
AINKTFSACALNFIPPTVKLFHSSCNPVGDT
HTTIQLLCLISGYVPGDMEVIWLVDGQKATN
IFPYTAPGTKEGNVTSTHSELNITQGEWVSQ
KTYTCQVTYQGFTFKDEARKCSESDPRGVT
SYLSPPSPL
CENTER FOR BIOLOGICAL SEQUENCE ANALYSIS TECHNICAL UNIVERSITY OF DENMARK DTU
Sequence-based methods for
prediction of linear epitopes
• The Parker
hydrophilicity scale
• Derived from
experimental data
D
E
N
S
Q
G
K
T
R
P
H
C
A
Y
V
M
I
F
L
W
2.46
1.86
1.64
1.50
1.37
1.28
1.26
1.15
0.87
0.30
0.30
0.11
0.03
-0.78
-1.27
-1.41
-2.45
-2.78
-2.87
-3.00
Hydrophilicity
CENTER FOR BIOLOGICAL SEQUENCE ANALYSIS TECHNICAL UNIVERSITY OF DENMARK DTU
Propensity scales: The principle
….LISTFVDEKRPGSDIVEDLILKDENKTTVI….
(-2.78 + -1.27 + 2.46 +1.86 + 1.26 + 0.87 + 0.3)/7 = 0.39
Prediction scores:
0.38 0.1 0.6 0.9 1.0 1.2 2.6 1.0 0.9 0.5 -0.5
Epitope
CENTER FOR BIOLOGICAL SEQUENCE ANALYSIS TECHNICAL UNIVERSITY OF DENMARK DTU
Propensity scales: The principle
• A Receiver
Operator Curve
(ROC) is useful
for finding a good
threshold and
rank methods
CENTER FOR BIOLOGICAL SEQUENCE ANALYSIS TECHNICAL UNIVERSITY OF DENMARK DTU
Evaluation of performance
• Pellequer found that 50% of the epitopes in a
data set of 11 proteins were located in turns
•Turn propensity
scales for each
position in the turn
were used for
epitope prediction.
1
4
2
Pellequer et al.,
Immunology letters, 1993
3
CENTER FOR BIOLOGICAL SEQUENCE ANALYSIS TECHNICAL UNIVERSITY OF DENMARK DTU
Turn prediction and B-cell epitopes
• Extensive evaluation of propensity
scales for epitope prediction
• Conclusion:
– Basically all the classical scales perform
close to random!
– Other methods must be used for epitope
prediction
CENTER FOR BIOLOGICAL SEQUENCE ANALYSIS TECHNICAL UNIVERSITY OF DENMARK DTU
Blythe and Flower 2005
• Parker hydrophilicity scale
• Hidden Markov model
• Markov model based on linear epitopes
extracted from the AntiJen database
• Combination of the Parker prediction scores
and Markov model leads to prediction score
• Tested on the Pellequer dataset and
epitopes in the HIV Los Alamos database
CENTER FOR BIOLOGICAL SEQUENCE ANALYSIS TECHNICAL UNIVERSITY OF DENMARK DTU
BepiPred: CBS in-house tool
Evaluation
on HIV Los
Alamos data
set
CENTER FOR BIOLOGICAL SEQUENCE ANALYSIS TECHNICAL UNIVERSITY OF DENMARK DTU
ROC evaluation
• Pellequer data set:
– Levitt
– Parker
– BepiPred
AROC = 0.66
AROC = 0.65
AROC = 0.68
• HIV Los Alamos data set
– Levitt
– Parker
– BepiPred
AROC = 0.57
AROC = 0.59
AROC = 0.60
CENTER FOR BIOLOGICAL SEQUENCE ANALYSIS TECHNICAL UNIVERSITY OF DENMARK DTU
BepiPred performance
• BepiPred conclusion:
– On both of the evaluation data sets,
Bepipred was shown to perform better
– Still the AROC value is low compared to Tcell epitope prediction tools!
– Bepipred is available as a webserver:
www.cbs.dtu.dk/services/BepiPred
CENTER FOR BIOLOGICAL SEQUENCE ANALYSIS TECHNICAL UNIVERSITY OF DENMARK DTU
BepiPred
Structural
determination
• X-ray crystallography
• NMR spectroscopy
Both methods are time
consuming and not
easily done in a larger
scale
Structure prediction
• Homology modeling
• Fold recognition
Less time consuming,
but there is a
possibility of incorrect
predictions, specially
in loop regions
CENTER FOR BIOLOGICAL SEQUENCE ANALYSIS TECHNICAL UNIVERSITY OF DENMARK DTU
How can we get information about the
three-dimensional structure?
• Homology/comparative modeling
>25% sequence identity (seq 2 seq alignment)
• Fold-recognition/threading
<25% sequence identity (Psi-blast search/ seq 2 str
alignment)
• Ab initio structure prediction
0% sequence identity
CENTER FOR BIOLOGICAL SEQUENCE ANALYSIS TECHNICAL UNIVERSITY OF DENMARK DTU
Protein structure prediction methods
 A data set of 75 discontinuous epitopes was
compiled from structures of antibodies/protein
antigen complexes in the PDB
 The data set has been used for developing a method
for predictions of discontinuous B cell epitopes
 Since about 30 of the PDB entries represented
Lysozyme, I have used homology grouping (25
groups of non-homologous antigens) and 5 fold
cross-validation for training of the method
 Performance was measured using ROC curves on a
per antigen basis, and by weighted averaging of AUC
values
CENTER FOR BIOLOGICAL SEQUENCE ANALYSIS TECHNICAL UNIVERSITY OF DENMARK DTU
A data set of 3D
discontinuous epitopes
Table 1. The Parker hydrophilicity
scale and epitope log-odds ratiosa
 Frequencies of amino acids in epitopes
Amino acid
Parker
Log-odds
Ratios
D
2.460
0.691
compared to frequencies of non-epitopes
E
1.860
0.346
N
1.640
1.242
S
1.500
-0.145
Q
1.370
1.082
G
1.280
0.189
K
1.260
1.136
T
1.150
-0.233
R
0.870
1.180
P
0.300
1.164
H
0.300
1.098
C
0.110
-3.519
A
0.030
-1.522
Y
-0.780
0.030
V
-1.270
-1.474
M
-1.410
0.273
I
-2.450
-0.713
F
-2.780
-1.147
L
-2.870
-1.836
W
-3.000
-0.064
Several discrepancies compared to the
Parker hydrophilicity scale which is often
used for epitope prediction
Both methods are used for predictions
using a sequential average of scores
Predictive performance of B cell epitopes:
Parker
0.614 AUC
Epitope log–odds
0.634 AUC
a
Amino acids are listed with descending
hydrophilicity using the values of the Parker
scale.
CENTER FOR BIOLOGICAL SEQUENCE ANALYSIS TECHNICAL UNIVERSITY OF DENMARK DTU
Epitope log-odds ratios
• Surface exposure
and
• structural protrusion
can
• be measured by
residue
The predictive performance:
• contact numbers
Parker
0.614 AUC
Epitope log–odds
Contact numbers
0.634 AUC
0.647 AUC
CENTER FOR BIOLOGICAL SEQUENCE ANALYSIS TECHNICAL UNIVERSITY OF DENMARK DTU
3D information: Contact
numbers
• A combination of:
– Sequentially averaged epitope logodds values of residues in spatial
proximity
– Contact numbers
.LIST..FVDEKRPGSDIVED……ALILKDENKTTVI.
-0.145
+0.691+0.346+1.136+1.180+1.164
+0.346
Contact number :
K 10
+1.136
Sum of log-odds
values
DiscoTope prediction
value
CENTER FOR BIOLOGICAL SEQUENCE ANALYSIS TECHNICAL UNIVERSITY OF DENMARK DTU
DiscoTope : Prediction of Discontinuous
epitopes using 3D structures
 Improved prediction of residues in discontinuous B
cell epitopes in the data set
 The predictive performance on B cell epitopes:
Parker
0.614 AUC
Epitope log–odds
0.634 AUC
Contact numbers
0.647 AUC
DiscoTope
0.711 AUC
CENTER FOR BIOLOGICAL SEQUENCE ANALYSIS TECHNICAL UNIVERSITY OF DENMARK DTU
DiscoTope : Prediction of
Discontinuous epiTopes
• Apical membrane antigen 1 from
Plasmodium falciparum (not used
for training/testing)
• Two epitopes were identified using
phage-display, point-mutation
(black side chains) and sequence
variance analysis (side chains of
polyvalent residues in yellow)
• Most residues identified as epitopes
were successfully predicted by
DiscoTope (green backbone)
..\Discotope\1Z40_epitope\1Z40_movie.mov
DiscoTope is available as web server:
http://www.cbs.dtu.dk/services/DiscoTope/
CENTER FOR BIOLOGICAL SEQUENCE ANALYSIS TECHNICAL UNIVERSITY OF DENMARK DTU
Evaluation example AMA1
 Add epitope predictions for protein-protein complexes
 Visualization of epitopes integrated in web server
 Testing a score for sequence variability fx based on
entropy of positions in the antigens
 Combination with glycosylation site predictions
 Combination with predictions of trans-membrane
regions
 Assembling predicted residues into whole epitopes
CENTER FOR BIOLOGICAL SEQUENCE ANALYSIS TECHNICAL UNIVERSITY OF DENMARK DTU
Future improvements
CENTER FOR BIOLOGICAL SEQUENCE ANALYSIS TECHNICAL UNIVERSITY OF DENMARK DTU
Presentation of the web
server
CENTER FOR BIOLOGICAL SEQUENCE ANALYSIS TECHNICAL UNIVERSITY OF DENMARK DTU
Presentation of the web
server output
• Conformational epitope server
http://202.41.70.74:8080/cgi-bin/cep.pl
• Uses protein structure as input
• Finds stretches in sequences which are
surface exposed
CENTER FOR BIOLOGICAL SEQUENCE ANALYSIS TECHNICAL UNIVERSITY OF DENMARK DTU
Use the CEP server
• CBS server for prediction of
discontinuous epitopes
• Uses protein structure as input
• Combines propensity scale values of
amino acids in discontinuous epitopes
with surface exposure
• www.cbs.dtu.dk/services/DiscoTope
CENTER FOR BIOLOGICAL SEQUENCE ANALYSIS TECHNICAL UNIVERSITY OF DENMARK DTU
Use the DiscoTope server
>PATHOGEN PROTEIN
KVFGRCELAAAMKRHGLDNYR
GYSLGNWVCAAKFESNF
Rational Vaccine
Design
CENTER FOR BIOLOGICAL SEQUENCE ANALYSIS TECHNICAL UNIVERSITY OF DENMARK DTU
Rational vaccine design
• Protein target choice
• Structural analysis of antigen


Model


Known structure or homology model
Precise domain structure
Physical annotation (flexibility,
electrostatics, hydrophobicity)
Functional annotation (sequence
variations, active sites, binding sites,
glycosylation sites, etc.)
Known 3D structure
CENTER FOR BIOLOGICAL SEQUENCE ANALYSIS TECHNICAL UNIVERSITY OF DENMARK DTU
Rational B-cell epitope design
• Protein target choice
• Structural annotation
• Epitope prediction and ranking




Surface accessibility
Protrusion index
Conserved sequence
Glycosylation status
CENTER FOR BIOLOGICAL SEQUENCE ANALYSIS TECHNICAL UNIVERSITY OF DENMARK DTU
Rational B-cell epitope design
•
•
•
•
Protein target choice
Structural annotation
Epitope prediction and ranking
Optimal Epitope presentation




Fold minimization, or
Design of structural mimics
Choice of carrier (conjugates, DNA
plasmids, virus like particles)
Multiple chain protein engineering
CENTER FOR BIOLOGICAL SEQUENCE ANALYSIS TECHNICAL UNIVERSITY OF DENMARK DTU
Rational B-cell epitope design
B-cell
epitope
T-cell
epitope
Rational optimization of
epitope-VLP chimeric
proteins:



Design a library of possible
linkers (<10 aa)
Perform global energy
optimization in VLP (virus-like
particle) context
Rank according to estimated
energy strain
CENTER FOR BIOLOGICAL SEQUENCE ANALYSIS TECHNICAL UNIVERSITY OF DENMARK DTU
Multi-epitope protein design
• Selection of protective B-cell epitopes involves
structural, functional and immunogenic analysis of the
pathogenic proteins
• When you can: Use protein structure for prediction
• Structural modeling tools are helpful in prediction of
epitopes, design of epitope mimics and optimal epitope
presentation
CENTER FOR BIOLOGICAL SEQUENCE ANALYSIS TECHNICAL UNIVERSITY OF DENMARK DTU
Conclusions