A Biometric Approach to Vaccine Insert Design

1
A Biometric Approach to Vaccine Insert Design

• Steve Self, Fusheng Li, Larry Corey
• Vaccine and Infectious Disease Institute (VIDI)
• HIV Vaccine Trials Network (HVTN)

2
Two Major Questions

• How to increase the breadth of HIV-specific
  responses to T-cell based vaccines?
• How to improve the magnitude and quality of the
  HIV-specific immune responses to T-cell based
  vaccines?
• We will consider biometric approaches to
  answering the first of these two questions.

3
Biometric Approach Design/Evaluation Cycle
Model-based Insert Design

Predictive Models
Empirical Evaluation
Model Validation and Refinement
4
Biometric Approach Design/Evaluation Cycle
Model-based Insert Design

Start here!
Predictive Models
Empirical Evaluation
Model Validation and Refinement
5
HVTN 054 Phase I Dose Escalation Trial
Vaccine 1 dose of Ad5 (NIH Vaccine Research
Center) encoding Gag/Pol (clade
B), Env (clades A,B,C) Study Participants n
48, low Ad5 neutralizing Ab titers Study
Design Group 1 20/4 1010 PU
vaccine/placebo Group 2 20/4 1011 PU
vaccine/placebo PBMC Cryopreservation within 8
hours after venipuncture on site Immunogenicity
Assessment ELISpot, ICS at day 28 with Global
PTE peptides (160 peptides per pool, 8 pools)
6
HVTN 054 Overall Response Rates by ELISpot
Day Treatment Responses
Placebo 0/7 (0)
0 Ad5 1010 0/18 (0)
Ad5 1011 2/19 (10.5)

Placebo 0/7 (0)
28 Ad5 1010 14/16 (87.5)
Ad5 1011 17/20 (85)

Placebo 0/7 (0)
364 Ad5 1010 15/19 (78.9)
Ad5 1011 13/16 (81.2)
7
Epitope Mapping

• Among those who mounted vaccine-induced T cell
  responses, we identified the specific 15 a.a. of
  HIV-1 recognized (n34).
• We then further fine-mapped the specific HLA
  class I restricted epitopes to the specific 8-11
  a.a. level.
• In total, 95 responses within 46 different
  epitopes were identified

8
NE Number of Reactive Epitopes per Vaccinee
(positive responders only)
9
Examples of individual responses, recognizing
more than one PTE variant

• PTID PROTEIN PTE SEQ 9MER HLA
• ENV TRVLAIERYLKDQQL ERYLKDQQL B14
• VLAVERYLKDQQFLG
• TRVLAVERYLRDQQL
• POL VQKITTESIVIWGKI ITTESIVIW B58
• VQKIATESIVIWGKT
• SNFTSTTVKAACWWA STTVKAACWW B58
• STAVKAACWWANVTQ
• POL LTEVIPLTEEAELEL IPLTEEAEL B35
• KALTEVVPLTEEAEL
• LTDIVTLTEEAELEL
• ENV VPTDPNPQEVVLGNV DPNPQEVVL B35
• CIPTDPNPQEIVLEN
• VPTDPNPQEMVLENV

10
Defining Breadth of Immune Response

• For HVTN 054, the average value of NE is 2.64
  reactive epitopes / vaccinee
• Distribution of NE (or its expected value) is an
  measure of breadth of immune response.
• How many of these are present in a virus sampled
  randomly from a target viral population?
• What is the coverage of response?

11
Example Vaccinee with NE 4 Reactive Epitopes
Viral Population Reactive Epitopes Reactive Epitopes Reactive Epitopes Reactive Epitopes Total Epitopes in Virus
Viral Population E1 E2 E3 E4 Total Epitopes in Virus
1 1 0 0 0 1
2 1 1 1 1 4
3 0 1 1 0 2
. . . . .
.N 0 0 0 0 0
p1 (freq of E1) p2 (freq of E2) p3 (freq of E3) p4 (freq of E4) EC (mean)
12
Defining Coverage

• EC expected number of a vaccinees reactive
  epitopes represented in a virus that is randomly
  sampled from target viral population
• Because of the lack of linkage disequilibrium in
  the viral genome, Ec is well approximated by the
  sum of epitope population frequencies (p)
• Two estimable parameters of immune response

(NE, EC)
Breadth
Coverage
13
HVTN 054 Distribution of EC for Subtype B Viral
Population
Mean EC 1.5
14
HVTN 054 Distribution of EC for Subtype A and C
Viral Populations
Mean EC 0.8
Mean EC 0.6
15
Aside Is worse for multiple independent exposures

• Expected number of independent exposures to virus
  with zero coverage by vaccinees reactive epitopes

Subtype A (mean 3.4)
Subtype B (mean 4.0)
Subtype C (mean 3.0)
16
Summary of Breadth/Coverage Analysis of 054

• Breadth
• Expected value of NE is 2.65 reactive epitopes /
  vaccinee
• Coverage
• For Subtype B viral populations,
• The expected coverage (EC) is 1.5 epitopes /
  vaccinee
• There is a loss of more than 1 epitope / vaccinee
  (over 40) due to imperfect coverage
• For Subtypes A and C viral populations,
• The expected coverage (EC) is 0.8 and 0.6
  epitopes / vaccinee
• There is a loss of nearly 2 epitopes / vaccinee
  (70 -80) due to imperfect coverage

17
Empirical criteria for evaluating insert designs

• Rank different insert designs based on
• Estimated absolute coverage values of EC
• Estimated efficiency of coverage given breadth of
  response (1 EC / NE)
• Measurement issues
• Peptide reagents other than autologous to insert
  will not give complete assessment of NE or EC
• If reagents cover all but least frequent peptides
  (eg, PTE, toggle) then
• Estimated EC will likely have only small bias
• Estimated NE may still have large bias

18
Biometric Approach Design/Evaluation Cycle
Model-based Insert Design

Predictive Models
Empirical Evaluation

• Vaccine immunogenicity
• Epitope mapping
• PTE freq for different
• viral populations
• Estimation of EC
• Rank insert designs
• based on estimated EC

Model Validation and Refinement
19
Biometric Approach Design/Evaluation Cycle
Model-based Insert Design

Predictive Models
Empirical Evaluation

• Vaccine immunogenicity
• Epitope mapping
• PTE freq for different
• viral populations
• Estimation of EC
• Rank insert designs
• based on estimated EC

Model Validation and Refinement
Correlate predicted ranking of insert designs
with ranking based on estimated EC
20
Biometric Approach Design/Evaluation Cycle
Model-based Insert Design

Predictive Models
Empirical Evaluation

• Vaccine immunogenicity
• Epitope mapping
• PTE freq for different
• viral populations
• Estimation of EC
• Rank insert designs
• based on estimated EC

Model Validation and Refinement
Correlate predicted ranking of insert designs
with ranking based on estimated EC
21
Insert Design Strategies

• Increase breadth (NE) Overcome immunodominance
• Protein fragments
• Heterotopic administration
• Increase coverage (EC) Direct responses to
  conserved PTEs
• Central sequences
• CoT and CoT Rosetta, UW and Microsoft
• Consensus and Mosaics LANL
• Heterologous prime-boost

Nickle et al. (2007) Coping with viral
diversity in HIV vaccine design. PLoS Comp
Biol 3(4) 754. Fischer et al (2007).
Polyvalent vaccines for optimal coverage of
potential T-cell epitopes in global HIV-1
vaccines. Nat Med 13 100-106
22
Status of Predictive Models
More than 90 of immunodominance can be
explained by the finding that only 1 of
peptides bind with sufficient affinity to a given
class I allomorph to form a complex of sufficient
stability to be presented in adequate numbers to
activate naive CD8 T cells. Although class I
binding selectivity narrows the peptide
repertoire considerably, it still leaves
thousands of viral peptides for the immune system
to potentially choose among.

From Yewdell (2006). Confronting
complexity Real-world immunodominance in
antiviral CD8 T-cell immune responses. Immunity
25533-543.
23
Predictive Models for HLA Peptide Binding

• Can predictive models be used to predict binding
  affinity of peptides rather than classify into
  binders and non-binders?
• Yes but only with moderate
  individual-level precision.
• Can predictions be used to identify T-cell
  epitopes amongst HLA ligands?
• No precision neither sufficient to rank
  peptides nor
• to predict number of
  reactive epitopes (NE)

BMC Immunology 98, 2008.
24
Binding Prediction Scores HLA-A A0201
Peters et al (2006). A community resources
benchmarking predictions of peptide binding to
MHC-1 molecules. PLoS Comp Biol 2(6) 574-584.
25
Model-based Design Strategy

• Cant predict breadth (NE) or which specific
  epitopes will emerge as reactive from exposure to
  a complex antigen but can potentially influence
  process
• Insert design provides a bin of PTEs from which
  the biological black box of immunodominance
  samples NE epitopes we determine what goes in
  the bin
• Assume uniform sampling of NE reactive epitopes
  from insert
• Evaluate insert by expected value of coverage per
  reactive epitope (EC / NE) as breadth-independen
  t index of coverage
• Is assumption of uniform sampling reasonable?

26
A Simple HLA Binding Prediction Score

Known CTL epitopes
Simulated PTE
gag PTE.
27
Joint Distribution PTE Frequency and HLA
Binding Score

28
Application to Heterologous Prime Boost Designs

29
Selective Boosting By HetPB Vaccination

• Hypotheses
• When heterologous antigens are administered
  sequentially the shared epitopes are
  preferentially recognized.
• Prime-specific responses are canceled at
  boosting stage.
• Boost-specific responses are actively suppressed
  in the existence of cross-reactive responses
  (original antigenic sin).

30
Selective Boosting Immunogen Design

• Design pairs of immunogens such that
• The set of PTEs that are common to the prime and
  boost are targeted as reactive epitopes (eg, high
  frequency, HIV-specific)
• The set of PTEs that are unique to the prime and
  the boost are not targeted (eg low frequency
  HIV-specific, non-HIV)
• Design can be achieved by
• Selecting pairs from a population of field
  isolates.
• Design pairs of synthetic HIV sequences
• Combination of natural and synthetic sequences
• Pairs of heterologous vectors
• Optimization
• Compute expected value of per-epitope coverage
  (EC / NE ) for each design
• Model-based optimal design has highest value of
  E EC / NE

31
HIV-1 Epidemic Pattern in China

• Three major viral forms co-circulating in China
• CRF07_BC.
• CRF01_AE.
• Subtype B.
• Different regions have distinct epidemic
  patterns
• Xinjiang Autonomous Region is dominated by
  CRF07_BC.
• Yunnan province are co-circulated by all the
  three viral forms.
• A specific CRF07_BC strain, CN54, has been
  selected as insert for candidate HIV vaccine in
  homologous PB strategy.
• What is potential advantage of a HetPB strategy?

CRF01_AE
CRF07_BC
From LANL
32
Per-Epitope Coverage of HomPB and HetPB Designs
Gag Env
EEC/NE
HomPB HetPB1 HetPB2 HomPB
HetPB1 HetPB2
HomPB prime boost single CRF07_BC isolate
sequence HetPB1 prime ? boost CN54 plus
distinct CRF07_BC isolate sequence HetPB2 prime
? boost, two distinct CRF07_BC isolate sequences
33
Expected Coverage and Number of Shared PTEs
HetPB1 Designs Env
34
Biometric Approach Design/Evaluation Cycle
Model-based Insert Design

• Reactive epitopes
• Uniform sampling
• HetPB hypotheses

Predictive Models
Empirical Evaluation

• Vaccine immunogenicity
• Epitope mapping
• PTE freq for different
• viral populations
• Estimation of EC
• Rank insert designs
• based on estimated EC

Model Validation and Refinement
Correlate predicted ranking of insert designs
with ranking based on estimated EC
35
Biometric Approach Design/Evaluation Cycle
Model-based Insert Design

Optimize by E EC/NE

• Reactive epitopes
• Uniform sampling
• HetPB hypotheses

Predictive Models
Empirical Evaluation

• Vaccine immunogenicity
• Epitope mapping
• PTE freq for different
• viral populations
• Estimation of EC
• Rank insert designs
• based on estimated EC

Model Validation and Refinement
Correlate predicted ranking of insert designs
with ranking based on estimated EC
36
Potential immunologic outcomes after
heterologous prime-boost vaccination

HomPB
Additive
?
HetPB hypotheses
HetPB
37
Biometric Approach Design/Evaluation Cycle
Model-based Insert Design

Optimize by E EC/NE

• Reactive epitopes
• Uniform sampling
• HetPB hypotheses

Predictive Models
Empirical Evaluation

• Vaccine immunogenicity
• Epitope mapping
• PTE freq for different
• viral populations
• Estimation of EC
• Rank insert designs
• based on estimated EC

Model Validation and Refinement
Correlate predicted ranking of insert designs
with ranking based on estimated EC Evaluate
HetPB hypotheses
38
Discussion Points

• Biometric approach reinforces a principled,
  iterative cycle of vaccine design and evaluation
• Breadth and coverage are related but different
  concepts that each need to be addressed in HIV
  vaccine design
• Approaches to increase breadth are still in
  infancy
• Promising approaches to increase coverage are
  nearing clinical evaluation (Con, mosaic)
• HetPB designs may provide more control of
  coverage but have not yet been fully explored
  experimentally

39
Thanks

• HVTN Lab
• Julie McElrath
• John Hural
• Steve Derosa
• Helen Horton
• SCHARP
• Peter Gilbert

• HVTN Core
• Ann Duerr
• Jim Kublin