CROI2003

1
Role of adaptive immunity and viral evolution in
vaccine design Immune selection of viral
variants
Simon Mallal Executive Director Centre for
Clinical Immunology and Biomedical
Statistics Royal Perth Hospital and Murdoch
University Perth, Australia
2
HLA diversity1600 different alleles (Class I
II)Evolved over milleniaDiverse within and
between racial groups
HIV diversity Error-prone RTRapid replication
rateRecombination High plasticity?APOBEC
3
HIV infection- host vs virus
4
Immune Response and Compromised Replicative
Capacity (RC) Decrease Viral Load
VIRAL LOAD
HIV-SPECIFIC CTL
Emerging Escape Mutations ? ?RC
Reverting Escape Mutations ? ?RC
Acute Infection
AIDS
TIME
5
Transmission of pre-adapted vs non-adapted SIV
Escaped
Wild type
100
Adaptation
Transient Reversion
Wildtype Sequence
Tetramer Binding
Selecting HLA allele
0
Absent Present
100
Reversion
Wildtype Sequence
Tetramer Binding
0
0
52
52
Barouch D et al and M John, S Mallal. Nature
Immunology 6(3)232,2005
Weeks
Weeks
6
HIV Escape Molecular to Individual to Population
Level

• CTL selection pressure is exerted at single amino
  acids in the virus
• CTL escape mutations should be HLA
  allele-specific at population level
• Escape is determined by CTL pressure VS genetic
  hurdle AND cost to replicative capacity

7
HIV-1 RT polymorphisms are HLA allele-specific
Moore et al Science 2002, 296 1439
8
HIV-1 RT polymorphisms are HLA allele-specific
Moore et al Science 2002, 296 1439
9
(No Transcript)
10
HLA-driven adaptation genome wide
11
Dominant influence of HLA-B in mediating the
potential co-evolution of HIV and HLA
Plt0.0003
P Kiepiela et al Nature 432769,2004
12
HLA-B5701 and 5801 are strongly associated with
a Gag T242N mutation

p24 110 Gag 242
13
Gag T242N associated with HLA-B5701 has a a
replicative cost evident at population level
Viral polymorphism
Gag T242N (Escaped)
Gag T242T (WT)
Replicative cost (0.6, p0.02)
Host polymorphism
CTL on non- escaped (1.1, p0.02)
Non- B5701
4.9
4.3
CTL on escaped (pNS)
B5701
3.8
4.7
Mean viral load (log), n197
14
HLA-Adaptation and Viral Load (Plt0.0001)
126,000
64,000
35,000
1,600
15
HLA-Adaptation and Viral Load clade B only
(Plt0.00001)
126,000
64,000
35,000
1,600
M John, Symposium CROI 2003
16
Higher Replicative Capacity Favors Negative Odds
Ratios
17
Accumulation of escape mutation -KF9 A83G, LI9
I31V
A Leslie et al, J Exp Med 2005
18
Interactive selection effects of HLA and ART on
HIV
M John et al, Antiviral Therapy 10551-555 2005
19
From HLA association to epitope discovery

• HLA associations may be within or flanking the
  true epitope
• Imputed non-adapted viral sequence indicates
  the epitope sequence most likely to be
  immunogenic in-vivo
• Large numbers of associations, multiple HLA
  associations at the same residue and flanking
  escape genome-wide means that a probabilistic
  approach to epitope prediction is required
• Can use epitopes to improve sensitivity of
  testing of ex-vivo CTL responses and for vaccine
  design

20
From HLA association to epitope discovery

• Existing methods
• Machine learning (e.g., SYFPEITHI)
• Molecular binding studies (e.g., Parker et al.
  1994)
• Our approach Combine HLA driven adaptation
  studies and machine learning
• Collaboration with Microsoft Research

21
From HLA association to epitope discovery

• Assume that there is one epitope for every escape
  found in a population of viral strains.
• For every escape, find the kmer in its vicinity
  that is most likely to contain the epitope (if
  the escape had not occurred).

22
Prediction Model

• Mixture of decision tree and logistic regression
  model
• Trained with 8-, 9-, and 10-mers from LANL
• Length 6 flanking regions determined by consensus
  amino acids from WA HIV Cohort Study

23
Decision Tree
24
Predicting new epitopes Example
Pol 770 A1101, Non-escaped K
AETGQETAYFILKLAGRWPVQTIHTDNGGNFISTTVKAACW
25
Predicting new epitopes Example
Pol 770 A1101, Non-escaped K
AETGQETAYFILKLAGRWPVKTIHTDNGGNFISTTVKAACW
26
Predicting new epitopes Example
Pol 770 A1101, Non-escaped K
AETGQETAYFILKLAGRWPVKTIHTDNGGNFISTTVKAACW
Scan nearby 18mers with probabilistic classifier
27
Predicting new epitopes Example
Pol 770 A1101, Non-escaped K
AETGQETAYFILKLAGRWPVKTIHTDNGGNFISTTVKAACW
28
Predicting new epitopes Example
Pol 770 A1101, Non-escaped K
AETGQETAYFILKLAGRWPVKTIHTDNGGNFISTTVKAACW
29
Experimental testing of putative epitopesiTOPIA
30
Experimental testing of putative epitopesiTOPIA
31
Implications for Vaccine Design

• In the Western Australian population
• No two individuals have the same HIV aa sequence
• No two individuals are HLA identical
• Individuals whose virus had adapted at the
  HLA-associated viral polymorphic sites
  corresponding to their HLA type had higher viral
  loads
• This marks the in-vivo HIV residue/epitope
  targets of HLA-restricted responses
• Those same epitopes should afford protection if
  given in a vaccine before exposure

32
Vaccine immunogen design

• what versus how
• Empirical approach has not worked to date
• Genetic diversity of HIV is extreme
• Vaccine responses are narrow, type-specific
• Need to maximize breadth of reactivity but
  minimize antigens for feasibility of vaccine
  development
• How can genetics and computing help?
• Knowledge of adaptation effects genome-wide
  in-vivo, across large real human populations
• Anti-wildtype responses would not protect against
  pre-adapted viruses
• HLA diversity can be exploited to overcome HIV
  diversity
• Collaboration with Microsoft Research, Seattle
• Machine Learning
• Computing capability

33
Comparison of 3 similar full-length vaccines
with a mixture of all 3
Gag-Pol-Env DNA/MVA vaccine
34
Current approaches

• Consensus
• Isolates, strains, population consensus
• multiple consensus sequences (clustering)
• Phylogeny-ancestral state reconstructions
• M group consensus
• MRCA
• COT
• None that directly optimise for immunogenicity
  against circulating strains using knowledge of
  strain diversity in HLA context

35
A polyallelic HIV immunogen Our focus and
assumptions

• Cellular vaccines, MHC-I epitopes
• More epitopes are better BUT longer immunogens
  are worse

36
HIV immunogen design Our focus and assumptions

• Cellular vaccines, MHC-I epitopes
• More epitopes are better BUT longer immunogens
  are worse, therefore should exploit overlap of
  epitopes
• Epitope must have proper flanking region to be
  presented
• Cross reactivity should be considered
• Immunodominance must be addressed

37
Vaccine Immunogen Optimization

• Exploit epitope overlap to shorten the immunogen
• Examine various optimization criteria (scores)
  and perform sensitivity analyses
• What if we over or under estimated the number of
  new epitopes?
• What model of cross reactivity should be used?

38
Optimization Score
HIV sequence
epitope
?
immunogen
pop-average(fraction of epitopes that are covered)
Viral strains are 245 Nef sequences from WA HIV
Cohort Study .
39
Epitomes vaccine immunogens that overlap
epitopes
40
Exploiting overlap yields immunogensthat cover
more epitopes per unit length
consensus
41
Cross-reactivity Sensitivity analysis

• Construct vaccine immunogens to optimize for
  various models of cross reactivity
• Score each set of vaccines according to each of
  the cross reactivity models
• Is there a model of cross reactivity that
  produces a good vaccine regardless of which model
  is true?

42
Cross Reactivity Models

• Not much in the literature (e.g., McKinney et al.
  2004)
• One to two conservative or semi-conservative
  changes are typically well tolerated (provided
  the epitope still binds MHC-I)

43
Models of cross-reactivity
non-escaped assocations for some HLA-I allele of
the patient
epitope
K
A
A
portion of immunogen
K
A
A
No-play epitope is covered if, for some HLA of
the patient, both epitope and vaccine segment
have non-escaped associations and an exact match
on every other amino acid 2-play epitope is
covered if, for some HLA of the patient,
both epitope and vaccine segment have
non-escaped associations and at most two amino
acids differ only by conservative amino-acid
substitutions elsewhere at non-anchor
sites N-play epitope is covered if, for some
HLA of the patient, both eptiope and vaccine
segment have non-escaped associations and differ
only by conservative amino-acid substitutions
elsewhere at non-anchor sites
44
Cross-reactivity Sensitivity analysis
45
Cross-reactivity Sensitivity analysis
46
Cross-reactivity Sensitivity analysis
Vaccine optimized for no-play model does well by
all three cross-reactivity models.
47
Number of epitopes Sensitivity analysis

• Baseline LANL predicted
• Less LANL only
• More All 9mers are epitopes
• Is there a vaccine optimized for one of these
  that does well regardless of the true number?

48
What if there are fewer epitopes?
Fewer epitopes
49
What if there are more epitopes?
More epitopes
If uncertain, should err in favor of
more epitopes (overlap provides some robustness)
50
Epitomes vaccine immunogens that overlap
epitopes
51
Exploiting overlap
overlap
no overlap
52
Immunodominance

• Rodriguez et al. 2002 If dominant and
  subdominant epitopes are delivered separately,
  then responses to both are elicited
• Generalization to vaccine (covering numerous
  patients)
• Becomes an HLA-sensitive optimization problem

cocktail vaccine

• dom1
• sub2
• sub3

• sub1
• dom2
• dom3

53
Conclusions

• HLA-driven adaptation in HIV is evident at
  population level as associations between HLA
  alleles and HIV sequence
• Such HLA associations can be used to predict
  immunogenic epitopes in HIV in-vivo
• We have optimized HIV vaccine immunogen sequences
  for maximum coverage of all (non-adapted)
  epitopes by exploiting epitope overlap and
  knowledge of adaptation effects
• Assuming no cross reactivity yields near-optimal
  vaccine even if there is some cross-reactivity
• If uncertain about the identity of the epitopes,
  it is better to err in favor of more epitopes.
  The more the epitopes overlap, the less important
  it becomes to know their identity (and the less
  they overlap, the shorter the vaccine)

54
A poly-allelic vaccine

• Immunogen ensures that all amino acids identified
  as target sites of CTL selection for all HLA
  alleles in a human population are present
• Uniquely exploits HLA diversity to combat viral
  diversity (clade and intra-clade)
• Based on observations from real human populations
• Takes into account the sites at which the
  non-escaped amino acid is NOT in consensus
  sequence (negative associations)
• Can be also optimised to-
• take account of overlapping effects of different
  HLA alleles at single sites
• take account of relative contribution of escape
  sites to viral fitness
• take account of the escape substitution that
  immune system may see when exposed to virus

55
Summary

• The interaction between host HLA and HIV
  polymorphism suggests an approach to the study of
  host-pathogen population genetics which may
  generalise to other co-evolutionary systems
• other adaptable pathogens
• other host factors
• cancer cell/host cell interactions
• animal or plant hosts/pathogen interactions
• Results can inform basic immunology and virology
  (escape, reversion, processing,
  fitness,compensatory, class II etc ),
  evolutionary biology, drug resistance,
  preventative and therapeutic vaccine design

56
Acknowledgements

• Microsoft Corporation
• David Heckerman, Nebojsa Jojic, Carl Kadie,
  Chris Meeks, Jennifer Listgarten
• Mina John, Corey Moore, Ian James, David Nolan,
  Silvana Gaudieri, Larry Park, Abha Chopra, Filipa
  Carvalho, Jean Keller, Annette Patterson, David
  Sayer, Karen Sutton, Damian Goodridge, Susan
  Herrmann
• Immunomics, Beckman Coulter (iTOPIA testing)
  Roberto Rentería, Mark Thorburn, Lori Lofaro
• National Health and Medical Research Council of
  Australia
• National Institutes of Health
• Bill and Melinda Gates Foundation