Title: Sieve analysis of the Step trial: Evidence for vaccineinduced antigenic pressure on HIV
1Sieve analysis of the Step trial Evidence for
vaccine-induced antigenic pressure on HIV
- Allan deCamp and Peter Gilbert
- Statistical Center for HIV/AIDS Research and
Prevention - Fred Hutchinson Cancer Research Center
2Conclusions of the Sieve Analysis
- The analysis of HIV sequences shows significant
differences (vaccine vs placebo) in T cell
epitope regions, suggesting that the vaccine
induced immune responses that in turn put
antigenic pressure on the virus - The analysis of post-infection T cell responses
shows significant anamnestic responses to MRKAd5
proteins deriving from vaccination and subsequent
infection - The analysis of acute VL data suggests that the
vaccine transiently and modestly suppressed acute
VL, which may have been caused by these
anamnestic responses
3Two Types of Potential Selective Effects for a
T-Cell Based Vaccine
- Acquisition Sieve Effect
- The vaccine selectively blocks (or enhances)
acquisition with specific HIV variants - Post-Infection Selective Effect
- The vaccine drives HIV sequence evolution
- Longitudinal HIV sequences are needed to
distinguish these two types of effects - But at the moment we only have one time-point per
subject
4Assess the genetics of the HIVs that infected the
trial participantsAre the viruses different
depending on whether a subject got vaccine or
placebo?
What is Sieve Analysis?
5Sieve Analysis Plan
- Compare a subjects sequences to the MRKAd5
insert sequence in 2 ways - Local Evaluate each site and sets of sites
separately (i.e., antigen
scanning) - Global Summarize overall protein distance with a
single - number
- Results shown by Morgane Rolland were Global
6Local Sieve Analysis Methods and Results
7Antigen Scanning
- Test each amino acid (AA) site as a signature
site - Signature site is a site where the frequency of
AA mismatch to the MRKAd5 AA differs vaccine vs
placebo - Several statistical methods applied
MRKAd5 sequence
Vaccinee observed sequences
Placebo observed sequences
Including Gilbert, Wu, Jobes (2008, Biometrics)
8AA Site Scanning Departures From 0 Indicates
Signature
Site where the frequency of AA mismatches to the
MRKAd5 AA differs in vaccine vs placebo
sequences (q-value lt 0.20)
8
9 Describe the strongest signature
Gag 84
10In several A-list epitopes including position 8
in SLYNTVATL A0201 A0202 A0205
LANL B (N324) 65 T 34 V
11Gag 84 by A-List Epitope-Restricting Alleles
P-value
44
56
17 V 79 V
Other
Other
A-List
A-List
Placebo
Vaccine
12Additional Signature Sites
135 of 6 Vacc w/ D potentially have an
allele restricting an epitope with 211-E
Position 9 of ETINEEAAEW A2501
LANL B (N324) 93 E 6 D
14Elite-controller B57 protective epitope (Walker
and colleagues)
Position 1 of HTQGYFPDW B57
Only 1 B57 vaccinee (H at site 116)
LANL B (N824) 84 H 14 N
15LANL B (N210) 72 T 24 I 1 V 0 -
169-Mer Scanning
- For each 9-mer in Gag, Nef, and Pol, we tested
for a difference (vaccine vs placebo) in protein
distances to MRKAd5 - Found 4 regions with a q-valuelt0.2 with 3 of the
4 regions showing a greater distance to the
vaccine in among the vaccinees.
in this region vaccinee sequences are closer to
the vaccine than placebo sequences this region
overlaps the B-57 restricted HW9 epitope
17Global Sieve Analysis Methods and Results
18Summary Measure Sieve Analysis
- Compute distance v from a subjects set of
sequences to the MRKAd5 sequence - For simple and valid statistical tests, use one
number per infected subject - Wilcoxon tests of whether the distributions of
summary measures differ between infected vaccine
vs infected placebo
19Complementary Distance Measures
1) Previous results presented by Morgane Rolland
Distance Average of protein distances across
all epitopes that are predicted in both the
MRKAd5 sequence and in a subjects set of founder
sequences
2) New results presented next Percent Epitope
Mismatch Distance Estimated percentage of
predicted epitopes in the MRKAd5 sequence that
are mismatched in at least one of a subjects
observed sequences
20Percent Epitope Mismatch MRKAd5Gag/Pol/Nef
Epipred
NetMHC
The corresponding p-values based on the distances
shown previously by Morgane Rolland were both
significant (0.02 and 0.007 respectively)
21Percent Epitope Mismatch Gag, Pol, Nef
Epipred
Nef
Pol
Gag
NetMHC
Nef
Gag 0.005
Pol
The corresponding p-values based on the distances
shown previously by Morgane Rolland were
significant for Epipred/Nef (0.03) and NetMHC/Gag
(lt0.0001)
22Percent Epitope Mismatch HXB2non-insert proteins
Epipred
NetMHC
23Summary of Results
- Local sieve analysis of signature sites
- Statistical evidence for 10 AA signature sites in
Gag, Nef, Pol none in Env - One particularly strong signature (Gag 84)
- Interpretation There was greatest statistical
power to detect site 84 as a signature, because
of the large sample size (n36 subjects with a
restricting allele). Vaccine-induced selection
pressure may have operated on many other sites,
but there is low statistical power for sites in
epitopes restricted by rare alleles. - Global sieve analysis
- Statistical evidence that vaccinee sequences had
greater epitope-based distances to MRKAd5 than
placebo sequences for Gag and Nef (not Pol)
24Challenges to Interpretation of Global Sieve
Analysis
- While the results are statistically valid, what
do they mean? - The extent to which the global sequence
differences are driven by a small number of
epitope regions is not yet clear - The interpretation of the sequence differences
depends on the epitope prediction method (Epipred
or NetMHC), which do different things - Epipred predicts CTL epitopes based on all known
epitope sequence motifs found in Branders A-list
and IEDB. It uses 2-digit, 4-digit, and
supertype HLA information - NetMHC predicts CTL epitopes based on
experimental binding affinity of peptides using
4-digit HLA information - Not surprising that the results differ by
algorithm
25What are the Functional Consequences of the
Observed Sequence Selective Effect?
26Functional Consequences T-Cell Response
- Were there anamnestic responses to MRKAd5
proteins deriving from vaccination and subsequent
infection?
27Post-infection CD8 T-cell Responses to Proteins
Contained in the Vaccine are Stronger in
Vaccinees
Nicole Frahm presented these data at AIDS
Vaccine 2009
- CD8 T-cell responses were measured by ICS in 87
participants (33 placebo and 54 vaccine
recipients) - Samples were obtained 1 week (8 participants) and
2 weeks (79 participants) post HIV diagnosis
28Functional Consequences Viral Load
- Did the boosted vaccine-induced T-cells suppress
viral load? - At set-point No (except possibly for some HLA
alleles) - During acute infection Possibly
29Acute Viral Load
Acute Viral Loads
N 29 infected subjects have acute-phase VL
(out of 87 cases) Acute sample that is HIV
RNA and HIV Ab Negative (ELISA Neg and WB Neg
or Indeterm)
(n 15) (n 14)
Estimated mean difference 0.27 (95 CI -0.28
to 0.83)
Analysis by Holly Janes
30Combined Viral Load and Signature Analysis
- The antigenic selection pressure may have caused
a transient suppression of viral load, with the
suppressive effect lost within weeks or months
after HIV acquisition - Approach
- At the identified signature sites, do subjects
with matched signature sequences have lower acute
VL vaccine vs placebo?
31VL Insert Matched Residue at Gag 211
31
32Viral Load Vaccine vs Placebo for Subjects with
Insert Matched Residue at Nef 116
33VL Insert Matched Residue at Pol 541
33
34 Follow-Up Studies
- Additional sequence data
- Mullins lab measuring HIV sequences at 2-3
time-points over the first 12 months of infection
- Will allow direct assessment of whether and how
vaccination alters HIV evolution and in
particular the pattern or rate of escape
mutations - Additional T cell response data
- McElrath lab is measuring post-infection T cell
responses to an array of peptide targets, which
will allow evaluation of whether vaccination
accelerated the development of T cell responses - Step ancillary studies
35Conclusions (Sequence Data)
- The analysis of HIV sequences shows significant
differences in breakthrough viruses for vaccine
vs placebo recipients - The nature of the differences supports that the
vaccine selected against viruses with certain
amino acids in T cell epitopes, suggesting that
the vaccine induced immune responses that put
antigenic pressure on the virus - While the MRKAd5 vaccine is not clinically
useful, this result may be a milestone in T-cell
based vaccine research, providing guidance for
the development of improved T-cell based vaccines
36Conclusions (Acute Viral Load Data)
- The analysis of acute VL data suggests
(nonsignificant trend) that the vaccine
transiently and modestly suppressed acute VL - The sequence analysis suggests the hypothesis
that this suppression was due to a
vaccine-induced acceleration of T cell evolution
37Conclusions (T Cell Response Data)
- The analysis of post-infection T cell responses
shows significant anamnestic responses to MRKAd5
proteins deriving from vaccination and subsequent
infection, which is consistent with a transient
vaccine-induced suppression of VL - However, few vaccinees had measurable
pre-infection T cell responses to the protein
regions or signature sites that contributed most
to the sequence differences, raising open
questions - The forthcoming additional sequence data and T
cell response data are expected to provide
additional insights into the vaccine effects
38Acknowledgments
Mullins lab Dana Raugi Stefanie Sorensen Jill
Stoddard Kim Wong Hong Zhao Laura Heath
Morgane Rolland Jim Mullins SCHARP Craig
Magaret Holly Janes Tomer Hertz Fusheng Li Steve
Self
McCutchan lab Francine McCutchan Sodsai
Tovanabutra Eric Sanders-Buell Meera Bose
Andrea Bradfield Annemarie OSullivan
Jacqueline Crossler Teresa Jones Marty Nau
Jerome Kim
Merck Danilo Casimiro Michael Robertson HVTN
Susan Buchbinder Ann Duerr John Hural David
Chambliss Patricia Dodd Nicole Frahm David
Friedrich Dan Geraghty Julie McElrath Larry
Corey
Plus thanks to David Nickle David Heckerman
Now at the Gates Foundation
39Extra Slides
40(No Transcript)
41Matched Epitope Distances
- For each founder sequence, compute all matched
epitopes, i.e., K-mers that satisfy - A known or highly likely epitope in the founder
sequence - A known or highly likely epitope in the MRKAd5
sequence - Same positions in the two sequences
- For each qualifying K-mer, compute the protein
distance founder vs MRKAd5 sequence - Matched epitope distance average of these
distances - This distance only measures amino acid
mismatches that preserve potential epitopes - Two approaches to predicting potential
epitopes Epipred and NetMHC - - NetMHC predicts binding affinity of an
MHC-peptide pair using - 4- digit HLA
42VL Insert Matched Residue at Gag 84
42
43Vaccine Selection Pressure May Operate on Acute
Founder Virus
Test for interaction p 0.08
Acute
Post-acute
Acute
Post-acute
44Percent Epitope Mismatch Gag and Nefby
Acute/Early
Epipred
Nef
Gag
NetMHC
Nef
Gag
45Complementary Distance Measures
- Previous results presented by Jim Mullins
- Distance Average of protein distances across
all epitopes that are predicted in
both the MRKAd5 sequence and in a subjects
set of founder sequences - New results presented next
- Distance Estimated percentage of predicted
epitopes in the MRKAd5
sequence that are mismatched in at least one
of a subjects founder sequences - The first distances only count amino acid
mismatches-with-insert that preserve potential
epitopes the second distances count all
mismatches - For both distances, epitopes predicted using two
different methods Epipred and NetMHC -
Heckerman D, Kadie C, Listgarten J (2007).
Leveraging information across HLA alleles/
supertypes improves epitope prediction. J
Computational Biology 14736-746. NetMHC 3.0.
Buus S, Lauemoller SL, Worning P, Kesmir C,
Frimurer T, Corbet S, Fomsgaard A, Hilden J,Holm
A, Brunak S. Tissue Antigens., 62378-84