High-throughput Immunomics

1

• High-throughput Immunomics
• Professor David Moss
• Birkbeck, U of London
• FISS07 25th August 2007

2
Plan of talk

• Important components of the adaptive immune
  system
• Immunomics in vaccine design
• Immunomics in tissue transplantation
• Grid technology in immunomics
• T-cell epitopes in bacteria

3
Cells of the Immune System
Stem cell
Myeloid progenitor
Lymphoid progenitor
Monocyte
B cell
T cell
Eosinophil
Neutrophil
Natural killer cell
Mast cell
Basophil
MANY CLONES
Plasma cell
Memory B cell
Helper T cell
Killer T cell
Dendritic cell
Macrophage
ENORMOUS REPERTOIRE
V. Brusic 2007
4
Conventional Vaccine Design

• Pathogens cultivated under laboratory conditions,
  dissected into their component proteins
• Antigens offering protective immunity are then
  identified by testing individual components
• Many of the proteins that are expressed during
  infection are not necessarily expressed in vitro
• Good candidate antigens can be overlooked

5
ImmunoVaccinology
Vaccines induce protective immunity Protective
immunity is an enhanced adaptive immune response
to re-infection
delivered as recombinant protein, via a viral
vector, as naked DNA, loaded onto APCs etc.
adjuvants
6
Vaccine Pipeline
Antigen Recognition
Antigen Delivery
Antigen Processing
MHC
B or T Memory cells
T cell Receptor
T cell
B or T effector cells
Epitope
Antigen Uptake
7
What is an Epitope?

• Epitope Any molecular structure that can be
  recognised by immune, or other, biological system
• T cell epitope A short peptide that interacts
  with a T cell receptor while bound to a Major
  Histocompatibility Complex (MHC) molecule
• B cell epitope Region of antigenic peptide
  bound by an antibody

8
T Cell Epitope Prediction
T-cell Receptor
Peptide
MHC
9
Molecular Detail
T cell Binding
Proteasomal Cleavage
Cathepsin Cleavage
TAP Transporter Binding
MHC class II Peptide Binding
MHC class I Peptide Binding
10
Intro - Proteasomal Cleavage
Proteasome
11
MHC Structure
CLASS I
CLASS II
12
Intro - MHCPeptide Binding
T-cell Receptor
Peptide
MHC
APC
APC Antigen Presenting Cell
MHC Major Histocompatibility Complex
13
T cell Epitope Prediction
250 AA protein
Computational route
Experimental route
242 8mers, 241 9mers, 240 10mers, 239 11mers etc
Peptide synthesis
2000 peptide binding assays
20 CTL assays
20 CTL assays
1-2 epitopes
14
Some of the tasks for in silico vaccinology

• Protein cleavage
• Peptide transport into the ER
• MHCpeptide binding
• MHCpeptide off-rates
• T-cell receptor binding to MHC complex
• T-cell signalling and cytokine production

15
Some available Tools and Services
This is a list (incomplete) of tools and services
that we might want to use in everyday
computational immunology

• MHCPeptide Binding Prediction
• SYFPEITHI, BIMAS, SMM, MHCPred, RANKPEP,
  TEPITOPE, NetMHC, SVMHC, EpiJen
• Molecular Dynamics NAMD, Amber, BLEEP
• Proteasomal Cleavage Prediction
• PAProC, MAPPP, NetChop
• TAP Binding - TAPPred
• T-cell Binding - EpiJen

16
A peptide in an MHC groove
17
Free Energy Calculations

• Free energy governs chemical equilibria
• Direct calculation of free energy using
  intermolecular forces involves evaluation of a
  6-dimensional integral
• The derivative of the free energy with respect to
  an appropriate co-ordinate can be evaluated using
  Molecular dynamics or Monte Carlo sampling
  methods
• r

18
ABF Calculations for MHC-peptide

• Adaptive Biasing Force (ABF) eliminates the force
  holding the peptide in the groove
• Molecular dynamics can then be used to sample the
  force at different distances between protein and
  peptide
• Integrating this force along the diffusion path
  gives the free enerfy of interaction

19
ABF simulation
20
Discovering minor histocompatibilty antigens

• These often occur through non-synonymous single
  nucleotide polymorphisms (SNPs) and are usually a
  problem in tissue transplantation
• Kidney transplants
• Bone marrow transplants
• They may occur in peptides that are recognised as
  foreign in the patient and cause graft rejection
• However, their presence on cancer cells may also
  be exploited to enable T-cells to attack them

21
Minor Histocompatibility Antigens
HA-1H
HA-1R
V L H D D L L E A
V L R D D L L E A
HLA-A0201
HLA-A0201
mHAgs are immunogenic peptides from polymorphic
proteins
22
mHAgs Leukaemia Relapse Treatment
If we select a donor/patient pair that differ at
a specific mHAg Donor Lymphocyte Infusion using
Donor cytotoxic T cells specific to that mHAg.
Leukaemia Relapse Patient
Bone Marrow Donor
mHAg specific T-cells
V L H D D L L E A
V L R D D L L E A
23
mHAg Discovery Process
Predicted mHAgs

• Peptides synthesised and refolded with MHC class
  I
• Correctly refolded MHC class I is purified by
  gel filtration using an HPLC Superdex column
• Elution is monitored by spectrophotometer

ANRI
24
Predicting mHAgs
348474
Peptide
Expression
15256
Protein
Proteasomal Cleavage
SAAP
12893538
Binding Predictions
45094
SNP
Binding Predictions
Proteasomal Cleavage
SNP
SAAP
Protein
Peptide
Expression
VLRDDLLEA
-1.46
G/A
R/H
NP_036424
5.28
Thyroid gland
VLHDDLLEA
0.768
Halling-Brown et al, 2005. SiPep A system for
the prediction of Minor Histocompatability
Antigens. European Journal of Immunology.
25
Predict difference in Binding
Scoring Methods
Peptide/Mutated
Consensus
SYFPEITHI
CombiPred
Binder
DILAGERAF
BIMAS
nHLApred
Non-binder
DILAGEYAF
MHCPred
26
Predicted mHAgs
Peptide Protein HLA Allele Prediction Score Binder ve, nonBinder -ve Prediction Score Binder ve, nonBinder -ve Prediction Score Binder ve, nonBinder -ve
SPANVSSCL C2orf24 A2 Reference (P) -1.679
SPANVSSCL C2orf24 A2 Mutated (L) 1.917
ARTWPCTLL CTLA4 A3 Reference (T) -0.078
ARTWPCTLL CTLA4 A3 Mutated (A) 1.254
QSLYSLTGL SLC26A1 A3 Reference (Q) 1.066
QSLYSLTGL SLC26A1 A3 Mutated (R) -0.307
NKDFFLRSL SLC26A1 A3 Reference (Q) -0.348
NKDFFLRSL SLC26A1 A3 Mutated (R) 1.025
SMDPLKLFD HS1BP3 A3 Reference (M) 2.788
SMDPLKLFD HS1BP3 A3 Mutated (V) -0.357
PLLDLAAYD NUP210 A3 Reference (R) -0.028
PLLDLAAYD NUP210 A3 Mutated (L) 3.557
QLLNLTLNT NALP4 A2 Reference (Q) -2.457
QLLNLTLNT NALP4 A2 Mutated (L) 0.025
27
Predicted mHAgs
Peptide HLA Allele Prediction Score Binder ve, nonBinder -ve Prediction Score Binder ve, nonBinder -ve Prediction Score Binder ve, nonBinder -ve
SPANVSSCL A2 Reference (P) -1.679
SPANVSSCL A2 Mutated (L) 1.917
ARTWPCTLL A3 Reference (T) -0.078
ARTWPCTLL A3 Mutated (A) 1.254
QSLYSLTGL A3 Reference (Q) 1.066
QSLYSLTGL A3 Mutated (R) -0.307
NKDFFLSRL A3 Reference (Q) -0.348
NKDFFLSRL A3 Mutated (R) 1.025
SMDPLKLFD A3 Reference (M) 2.788
SMDPLKLFD A3 Mutated (V) -0.357
PLLDLAAYD A3 Reference (R) -0.028
PLLDLAAYD A3 Mutated (L) 3.557
QLLNLTLNT A2 Reference (Q) -2.457
QLLNLTLNT A2 Mutated (L) 0.025
28
S(L/P)ANVSSCL bound with A2
I Aggregate II Protein/Peptide III Peptide
A2/(Control)
A2/SLANVSSCL
Optical density (mAU)
A2/SPANVSSCL
Time
29
Using mHAgs in treating leukaemia relapse

• The proteins containing mHAgs must be expressed
  preferentially on lymphocytes
• The mHAgs must occur with reasonable frequency in
  the population
• T-cells containing mHAgs can be used as an
  immunotherapy to attack leukaemia cells

30
GRID Definition
Grid computing is distributed computing
performed transparently across multiple
administrative domains Peter Coveney
31
GRID Rationale

• Distribute workflows and simulations
• Allow exploration of parameter space
• Simulate allelic variation in populations
• Access to large machines for computationally
  expensive simulations in a transparent way

32
Examples of Technology to Access
Resource Type Technology to Access
Groups individual PCs Web service/GridSAM
Groups clusters Web service/GridSAM/UNICORE
CINECA UNICORE/EnginFrame
National Grid Service Globus/GridSAM
DEISA UNICORE/DESHL/GridSAM
TeraGrid Globus
33
GRID Rationale (3)
We intend to utilise and amalgamate multiple
different GRID solutions
GridSAM
AHE
This approach will provide flexibility and
redundancy
34
GRID Implementation
Application Hosting Environment Prof. Peter
Coveney
Recent developments have made this approach
possible
RSL - Resource Specification Language
JSDL - Job Submission Description Language
NJS - Network Job Supervisor
35
Application Hosting Environment
Web Interface
Job launcher
AHE Client
UNICORE
AHE Server
JSDL
GridSAM
GridSAM
GridSAM
GATEWAY
GATEWAY
GridSAM
FORK
FORK
RSL
NJS
Web Service
Group Local
NGS
DEISA
CINECA
Group Cluster
GLOBUS
36
Application Hosting Environment

• Through GridSAM, the AHE allows access to
• Local resources
• Local clusters
• National Grid Service
• UNICORE
• TeraGrid

Easy installation AHE, GridSAM, OMII-BPEL
Tomcat can all be installed via the OMII-stack
37
Characteristics of subunit vaccines

• Examine a database of proteins that are known to
  confer protection
• Where are these proteins located on the pathogen?
• Are these proteins relatively rich in T-cell
  epitopes?

38
A Vaccine Antigen Database

• 70 proteins from pathogenic bacteria
• Each known to elicit an immune response in a
  disease model
• Either an animal model of human disease
• Or an animal disease known to infect humans
  (zoonosis)
• Most are exposed to the immune system
• 23/70 (33) exported from the cell
• 39/70 (56) outer membrane / cell wall

Mayers et al. (2003), Comp. Funct. Genomics 4, 468
39
An Unexpected Result?
Outer Membrane
Inner Membrane
Secreted
Cytosolic
Periplasmic
Combined

• Proteins in the vaccine antigen dataset had much
  lower scores than control proteins - over 5
  standard deviations from mean

40
Class I Binding A Similar Anomaly
Outer Membrane
Inner Membrane
Secreted
Cytosolic
Periplasmic
Combined

• Proteins were screened for Class I alleles using
  a similar technique Known antigens also found to
  be deficient in Class I epitopes

41
MHC Score correlates with FILMVWY
42
Co-Evolution

• HIV can generate epitope mutations that impair
  the immune response generated by Class I alleles
  it sees
• Bacteria are thought to use similar strategies
• Generating mutations to mimic host peptides
• Generating mutations to prevent MHC presentation,
  T-cell binding or response
• This may explain why proteins visible to the
  immune system lack MHC alleles

43
Acknowledgements

• Birkbeck
• Adrian Shepherd
• Mark Halling-Brown
• Clare Sansom
• Matt Davies
• Renata Kabiljo
• Collaborators
• Paul Travers, Anthony Nolan Research Institute
• Darren Flower, Jenner Institute for Vaccine
  Design
• Mark Coles, University of York

• ImmunoGrid Partners
• CINECA Elda Rossi, Andrew Emerson
• Brisbane Vladimir Brusic
• CNR Massimo Bernaschi,
• Filippo Castiglione
• CNRS Marie-Paule Lefranc
• DTU Søren Brunak
• Bologna Pierre-Luigi Lollini
• Catania Santo Motta

44
S(L/P)ANVSSCL bound with A2
A2/(Control)
Refolding Results (area under peak II) Binding Prediction
263 1.917
- -1.679
A2/SLANVSSCL
Optical density (mAU)
A2/SPANVSSCL
Time
45
Application Hosting Environment
46
Prediction of MHC class I epitopes

• Gibbs sampling
• Sequence motifs, matrices
• Sequence weighted matrices performance of the
  method (e.g. ROC analysis) depends on the number
  of training peptides (Immunological
  Bioinformatics O. Lund, 2005)
• Hidden Markov Models
• Artificial Neural Networks

ALAKAAAAM ALAKAAAAN ALAKAAAAV GMNERPILT GILGFVFTM
TLNAWVKVV KLNEPVLLL AVVPFIVSV Peptides known to
bind to the HLA-A0201 molecule.
For T-cell epitopes the most selective
requirement is the ability to bind an MHC with
high affinity.
47
T cell Epitope Prediction
Prediction Class I and II T cell
epitope prediction B cell epitopes and Antigens
Experimental Verification and Data
Discovery Test prediction and generate new
binding data
Database Antigens, B cell and T
cell epitopes Peptide binding, Protein-Protein Int
eractions
48
Prediction of MHC class I epitopes

• Gibbs sampling
• Sequence motifs, matrices
• Sequence weighted matrices performance of the
  method (e.g. ROC analysis) depends on the number
  of training peptides (Immunological
  Bioinformatics O. Lund, 2005)
• Hidden Markov Models
• Artificial Neural Networks

ALAKAAAAM ALAKAAAAN ALAKAAAAV GMNERPILT GILGFVFTM
TLNAWVKVV KLNEPVLLL AVVPFIVSV Peptides known to
bind to the HLA-A0201 molecule.
For T-cell epitopes the most selective
requirement is the ability to bind an MHC with
high affinity.
49
Application Hosting Environment
50
Adaptive Biased Force Calculations

• Computation of free energy profiles based on an
  enhanced exploration of chosen conformational
  degrees of freedom
• Adaptive biasing force (ABF) method uses a
  continuously updated estimate of a free energy
  profile to apply a bias
• Alternative methodology to umbrella sampling

51
ABF methodology

• Optimal uniform sampling of the phase space along
  chosen ordering parameter ?
• Independent set of routines that collect the
  Cartesian co-ordinates and the forces exert on
  those atoms involved in the reaction co-ordinate
  ?
• Feeds back the biasing force to the core MD
  integrator

52
Complexity in the Immune System
Systems
Organs
Cells
Intra- cellular