1 sieve analysis of hiv sequences in the step hiv vaccine trial peter gilbert vaccine infectious...
TRANSCRIPT
1
Sieve Analysis of HIV Sequences in the Step HIV Vaccine Trial
Peter Gilbert
Vaccine Infectious Disease Institute
Fred Hutchinson Cancer Research Center
27 May 2009
2
Coworkers
Mullins lab Dana Raugi Stefanie Sorensen Jill Stoddard Kim Wong Hong Zhao Laura Heath Morgane Rolland Jim Mullins
SCHARP Peter Gilbert Allan deCamp Fusheng Li Craig Magaret Steve Self
McCutchan lab Francine McCutchan* Sodsai Tovanabutra Eric Sanders-Buell Meera Bose Andrea Bradfield Annemarie O’Sullivan Jacqueline Crossler Teresa Jones Marty Nau Jerome Kim
Plus thanks to David Nickle & David Heckerman *Now at the Gates Foundation
Merck Danilo Casimiro Michael Robertson
HVTN John Hural David Chambliss Patricia Dodd Nicole Frahm David Friedrich Julie McElrath
3
Merck’s Ad5 Trivalent Vaccine
ITRL ITRRgaghCMV
pA
MRKAd5 HIV-1 gag
ITRL ITRR
polhCMV
pAMRKAd5 HIV-1 pol
ITRL ITRRnefhCMV
pA
E1
MRKAd5 HIV-1 nef
• Vaccine: 1:1:1 admixture of 3 Ad5 vectors– Encoded transgenes: codon-optimized, near-consensus clade B
HIV-1 sequences
• Placebo: vaccine dilution buffer without Ad5
4
Step Study sites
Study conducted December 2004 to present
5
Cumulative Number of HIV Infections
Cases accrued as of Oct 17, 2007
Time to event (weeks)
Cum
ulat
ive
num
ber
of H
IV in
fect
ions
(ev
ents
)
0 10 20 30 40 50 60 70 80 90 100
0
5
10
15
20
25
30
35
40
45
50
55
60
49 Vaccine
33 Placebo
2-tailed p-value = 0.077
Primary study results
reported in
Buchbinder et al. (2008,
Lancet)
Surprising result: The
vaccine may have increased
the risk of HIV
6
No Vaccine Effect on Viral Load
• No difference between vaccine and placebo groups (p = 0.441)
7
Assess the genetics of the HIVs that infected the trial participants
Are the viruses different depending on whether a subject got vaccine or placebo?
8
Potential effects of CTL-based vaccines
X Vaccine blocks
infection
XVaccine blocks
specific variants
CTL-driven evolution
9
Given the failure of the vaccine to block infection:
• Our Overriding Questions Become: 1. Can we detect a “sieve” effect on the virus founder, in which some strains are blocked,
presumably due to strain-specific immunity?
2. Can we detect selection on the evolving viral population, presumably due to anamnestic responses deriving from vaccine immunization and subsequent infection?
• Sequence viral genomes from infected vaccine and placebo recipients– Compare overall viral protein sequences in infected volunteers– Restrict analysis to predicted viral epitopes and compare sequences to vaccine– Use placebo recipients as control for these comparisons
• Methods:– Amplify and directly sequence 5-10 individual, near-full-length (9.1kb) viral genomes– Assess phylogenetic tree structure, diversity, divergence from vaccine, selective pressure– Assess conservation of predicted epitopes shared between vaccine and infection founder
10
Mullins Laboratory Overview
• 93 volunteers infected until Dec. 2007
• Plasma samples available from 88• 51 Vaccine, 37 Placebo
• Samples from 68 individuals were PCR positive:• 39 Vaccine, 29 Placebo
• WG sequences derived from 66 volunteers:
• Near-full-length (9.1kb) genomes
• Single half-genomes from 2 volunteers
• First target was 5 genome sequences:
• Assess sequence variation by counting the number of phylogenetically-informative sites:
• Little variation: 5 WG
• Detectable variation in first 5 WG: obtain 10 WG
• 459.5 individual, PCR-amplified viral genomes were directly sequenced
11
Time between the last immunization and HIV-1 infection
• The 68 plasma samples from which WG were obtained corresponded to the :
• First HIV-1 positive samples for 66 volunteers: 39 Vaccine, 27 Placebo
• Second HIV-1 positive samples for 2 volunteers*: 1 Vaccine (d.364), 1 Placebo (d.247)
3
2
1Days since last immunization
Vaccine recipient who ended up completing 2 immunizations
Vaccine recipient who ended up completing 3 immunizations
Placebo recipient who ended up completing 2 immunizations
Placebo recipient who ended up completing 3 immunizations
*
*
• Samples were collected at the same time after the last immunization for Vaccine (156 days) and Placebo (163 days) recipients
Number of immunizations at the time of infection
12
Number of Founder Viruses Detected
49/65 = 75% of subjects replicating a single variant 16/65 = 25% of subjects replicating multiple variants
• Among vaccinees: 10/40 = 25% of subjects replicating multiple variants• Among placebos: 6/25 = 24% of subjects replicating multiple variants• Insufficient data from 3 individuals
Number of founder variants
Phy
log
enet
ical
ly-in
form
ativ
e si
tes
Vaccine
Placebo
1 2* 4
3
74
49 subjects 16 subjects
13nef
CRF02-AG
Vaccine
Placebo
Yellow highlighting indicates multiple variants from one subject
HXB2
STEP vaccine
Linked transmission pair
NYC 502-0309 26 October 06NYC 502-0879 22 March 07
Are there phylogenetic clusters consistent with transmission between trial participants?
14nef
LimaIquitos
Atlanta
Atlanta
Atlanta
Birmingham
Boston
Denver
Denver
Los Angeles
Miami
Toronto
NYC
NYC
NYC
NYC
NYC
NYC
San Francisco
San Francisco
Seattle
St. Louis
St. Louis
Atlanta
Atlanta
BirminghamDenver
Denver
Denver
Houston
Houston
Los Angeles
MiamiMiami
NYC
NYC
NYC
NYC
NYC
NYC
NYC
NYC
Rochester
San Francisco
San Francisco
San Francisco
Seattle
Seattle
Seattle
St. Louis
NYC
CRF02-AGHXB2
STEP vaccineLima
Lima
LimaLima Lima
Lima
Lima
Lima
IquitosIquitos
Toronto
Iquitos
Toronto
Vaccine
Placebo
Yellow highlighting indicates multiple variants from one subject
Linked transmission pair
NYC 502-0309 26 October 06NYC 502-0879 22 March 07
15
p=0.3331 p=0.3766p=0.3275
Gag Pol Nef
Overall, are breakthrough/founder viruses unusually divergent from the vaccine? (No)
*Nickle D, Heath L, Jensen M, Gilbert P, Mullins J, Pond S. 2007. HIV-specific probabilistic models of protein evolution. PLoS ONE, June 6; 2:e503.
Distances from breakthrough sequences to STEP vaccine sequence were calculated using an HIV-specific model of protein evolution*
16
Summary - 1
• Phylogenetic analysis of breakthrough viruses from 66 trial volunteers Single HIV-1 variants established infection in 75% of volunteers One cluster with 2 HIV-1 infected individuals
2 vaccine recipients from NYC All subtype B infections except one CRF02-AG
• No difference between Placebo and Vaccine in the genetic distances from the breakthrough to the STEP vaccine sequences
17
Methods
• Known and potential CTL epitopes were predicted using Epipred* (with a posterior probability > 0.80).
• Epitopes were predicted based on each volunteer’s HLA type in:o Breakthrough viral sequences (WG)o STEP vaccine sequence (Gag, Pol, Nef)
• Viral sequences from 3 individuals were excluded:o One female individual: 502.1115 (Placebo)o CRF02_AG isolate: 502.2696 (Vaccine)o No HLA genotype data was available from 1 placebo recipient: 502.1504
• Epitope prediction on:o 64 WG (39 vaccine recipients; 25 placebo recipients)o 2 partial sequences from 2 placebo recipients
CTL-mediated selection for breakthrough viruses?
*Epipred. Listgartner, Cadie, Heckerman, Journal of Computational Biology 2007.available at: http://atom.research.microsoft.com/bio/epipred.aspx
18
• Gag + Pol + Nef: ~1700 AA• Predicted epitopes: ~120 AA
Analysis on predicted epitopes is restricted to ~7% of the Gag-Pol-Nef sequence data
Num
ber
of e
pito
pes
Placebo Vaccine n = 26 n = 39
12 13
Breakthrough vs. Vaccine: Predicted Epitopes Only
19
Viruses infecting vaccinees were more likely to have epitopes that differed from those in the vaccine
Dis
tanc
e
Placebo Vaccine n = 26 n = 39
0.011
p =0.0232
0.030
Breakthrough vs. Vaccine: Predicted Epitopes Only
• Protein distances between the breakthrough and the STEP vaccine epitopes were calculated using an empirical HIV-specific model of protein evolution
• Epitope-specific distances were summarized to obtain one ‘breakthrough to STEPvax’ distance value per subject
20
Breakthrough vs. Vaccine: Predicted Epitopes OnlyD
ista
nce
Placebo Vaccine n = 25 n = 37
0.011
p =0.1465
0.028
Placebo Vaccine n = 26 n = 36
0.008
p =0.8299
0.008
Placebo Vaccine n = 26 n = 38
0.021
p =0.0298
0.064
NefGag Pol
•Viruses infecting vaccinees were more likely to have epitopes that differed from those in the vaccine
•The effect is primarily driven by mutations seen in Nef epitopes
21
Analysis of predicted epitopes
•Viruses infecting vaccinees were more likely to have epitopes that differed from those in the vaccine
•These data indicate that the vaccine may have blocked establishment of infection by those variants sharing more epitopes with the vaccine
•Whether vaccine-induced CTL-mediated pressure drives subsequent viral evolution requires sequences from later time-points
Summary - 2
22
Context of Sieve Analysis: Challenged Statistical Power
• Achieving high statistical power requires:– Large n of infected subjects with sequence data– A vaccine that induces immune responses that ‘react strongly’
with the infecting viruses
• For Step, the sieve analysis has relatively low power– Small number of infections (n=66)
• Phase 2b, not Phase 3 (VaxGen: n=336)
– At an epitope level, the vaccine appeared to induce limited potential selective pressure
• On average, a vaccinee recognizes < 1 reactive epitope in an average exposing HIV
• Can only detect relatively large sieve effects
23
Structure of Sieve Analysis
• Assess Gag, Nef, Pol, Env separately
• Assess either 1 sequence per subject (majority consensus) or use all individual sequences
• Compare a subject’s sequences to the StepVx sequence in 2 ways:– Global: Summarize overall ‘similarity’ or ‘distance’ with a single
number – Local: Evaluate each site and sets of sites separately (i.e.,
‘antigen scanning’, machine learning classification)
24
Global Sieve Analysis: Methods and Results
25
Summary Measure Sieve Analysis
• Compute similarity or distance measures v between the StepVx sequence and a subject’s set of sequences– 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
26
Summary Measure Sieve Analysis
• Epitope-based summary measures: Compare known and predicted T cell epitope sequences (8-mers through 11-mers) in StepVx sequence to a subject’s corresponding sequences
E.g., StepVx-sequence 9-mer Gag 77-85: A subject’s sequences:
• These results focus on simplest measure that scores 0 or 1 for match or mismatch
GAG SLYNTVATL
Con . . F . . . . V .
Seq . . . . . . . V .
Seq . . . . . . . V .
Seq . . F . . . . V .
Seq . . F . . . . V .
27
Summary Measures Used for These Analyses (Based on Shared Epitopes)
• ‘Absolute’ Similarity Score: Number epitopes in both the StepVx sequence and in all of a subject’s sequences
• ‘Percent’ Similarity Score: Percent of epitopes in the StepVx sequence that are also in all of a subject’s sequences
Estimate in 2 ways, based on all of a subject’s HLA alleles:– Known & Highly Likely Epitopes: Restrict to all 8-mers through 11-
mers in the StepVx sequence that are known epitopes or predicted epitopes with probability > 0.80 of being an epitope (from Epipred*)
– All Possible Epitopes: Consider all 8-mers through 11-mers in the StepVx sequence with positive probability of being an epitope
*Heckerman D, Kadie C, Listgarten J (2007). Leveraging information across HLA alleles/
supertypes improves epitope prediction. J Computational Biology 14:736-746.
28
Translation of Percent Similarity Score to Percent Mismatch Distance
• Percent Similarity Score: Percent of epitopes in the StepVx sequence that are also in all breakthrough sequences
• We report results using the equivalent
Percent Mismatch Distance = 1 - Percent Similarity Score
Percent Mismatch Distance = Estimated percent of epitopes in the
StepVx sequence that are not in any of the subject’s sequences
29
Estimated Number Shared Epitopes (Known & Highly Likely)
p=.46 p=.91 p=.24
30
Estimated Number Shared Epitopes (Account for all 8-11 Mers)
p=.07 p=.19 p=.04
31
Percent Mismatched Epitopes (Known & Highly Likely)
p=.16 p=.32 p=.09
32
Percent Mismatched Epitopes (Account for all 8-11-mers)
p=.09 p=.46 p=.06
33
More Sophisticated Epitope-Based Summary Measures (Ongoing Analyses)
• Similar to the above except account for biological knowledge of HIV evolution and MHC-peptide interactions– Weight AA positions by
• Entropy• Whether a primary or secondary anchor site
– Weight distances between K-mer peptides by• Predicted change in binding energy• Evolutionary cost of AA mismatches*
*Nickle D, Heath L, Jensen M, Gilbert P, Mullins J, Pond S (2007). HIV-specific probabilistic models of protein evolution. PLoS ONE, June 6; 2:e503.
Thanks to Tomer Hertz for discussions about defining peptide-distances
GAG SLYNTVATL
Con . . F . . . . V .
Seq . . . . . . . V .
Seq . . . . . . . V .
Seq . . F . . . . V .
Seq . . F . . . . V .
(The results Morgane reported use this weighting)
34
Evolutionary-Cost Weighted Epitope Distances (Shown Earlier)
Dis
tanc
e
Placebo Vaccine n = 25 n = 37
0.011
p =0.1465
0.028
Placebo Vaccine n = 26 n = 36
0.008
p =0.8299
0.008
Placebo Vaccine n = 26 n = 38
0.021
p =0.0298
0.064
Gag Pol Nef
35
The analyses shown did not account for the timing of sequencing relative to the development of immune responses
Based on knowledge of early HIV infection dynamics, a vaccine selective effect may be expected to be restricted to (or stronger on)
early viruses
Break down results by whether sequences were measured pre-seroconversion (n=27
infected subjects)
36
Estimated Number Shared Epitopes (Known & Highly Likely)
Interaction p-values: p=.09 p=.43 p=.002
P-values for Ab-: p=.05 p=.64 p=.002
37
Percent Mismatched Epitopes (Known & Highly Likely)
Interaction p-values: p=.81 p=.90 p=.01
P-values for Ab-: p=.20 p=.43 p=.0009
38
• Pol: No statistical evidence of sieving
• Gag: Weak/borderline statistical evidence of sieving (overall p-values .07, .09, .15, .16, .46)
– Timing analysis tentatively supports sieve effect may be concentrated on pre-seroconversion viruses
• Nef: Fairly strong statistical evidence of sieving (overall p-values .03, .04, .06, .09, .24)
– Timing analysis supports sieve effect concentrated on pre-seroconversion viruses
• Interaction p-values .002, .01, .08, .13
• Pre-seroconversion subgroup p-values .0009, .002, .01, .01
Summary of Global Sieve Analysis of Epitope-Based Summary Measures
39
Local Sieve Analysis: Methods and Results
40
Antigen Scanning of AA Sites
• Test each AA site as a signature site: – Signature site = a site where the frequency of AA mismatches to the StepVx
AA differs in vaccine vs placebo sequences
• 2 analyses: – 1 sequence per subject (majority consensus variant)*– All individual sequences**
– Use adjusted p-values, q-values to guard
against false positives
*Method of Gilbert, Wu, Jobes (2008, Biometrics)**Nonparametric bootstrap pairwise mismatch method
Vx reference sequence
Vaccinee breakthrough sequences
Placebo breakthrough sequences
41
Machine Learning Methods to Classify Sequences by Vaccine/Placebo
• Classify vaccine/placebo status from AA characters at sets of AA sites
• Use all individual sequences
• Cross-validation (at subject level) to estimate classification accuracy on hold-out data
• Inductive learning methods:– Divide and conquer algorithms (decision trees)– Induction rules– Ensemble models (boosting, bagging, bumping)
42
Results: AA Site Scanning
Signature Sites with a q-value < .20AA Site (HXB2
Numbering)Majority Cons
Variant ScanningAll Sequences
ScanningAll Sequences
Machine Learning
Unadj p (q) Unadj p (q) In Selected Model?
Gag 84* (in several
A-list epitopes)
<.0001 (<.0001) <.0001 (.02) Yes
211* .002 (.09) Yes
Pol 541 <.0001 (.02) Yes
721 .0009 (.11)
Nef 64a .003 (.16)
82 .006 (.16)
116* (in HW9) .002 (.16) Yes
173 .004 (.16) Yes*Known CTL epitope escape site
Bonferroni adjustment:
Gag 84: adjusted p < .01 (majority consensus scanning) and adjusted p = .015 (all sequences scanning)
Pol 541: adjusted p = .024 (all sequences scanning)
43
Machine Learning Results: Certain Sets of AAs Classify Vaccine/Placebo Status Better Than Chance
• Correct classification rate of vaccine/placebo status on hold-out data:– Gag: ~78%– Nef: ~ 68% – Pol: ~ 64% Benchmark: random guessing gives rate of
60%
Best classifying sets of AA sites
– Gag 84V, 124N, 406R• 77% vaccinee sequences; 0% placebo sequences
• 84 (SLYNTVATL) A*0201 A*0202 A*0205 + several other epitopes• 124 (NSSKVSQNY) B*3501• 406 (CRAPRKKGC) B14
– Nef 116N, 120Y• 56% vaccinee sequences; 2% placebo sequences
• 116 (HTQGYFPDW) B57• 120 (YFPDWQNYT) A29 B*3701 B*5701 Cw6
44
LANL B
(N=324):
65% T
34% V
In several A-list epitopes including position 8 in SLYNTVATL A*0201 A*0202 A*0205
45
LANL B
(N=324):
93% E
6% D
Position 9 of ETINEEAAEW A*2501
46
LANL B
(N=824):
84% H
14% N
Position 1 of HTQGYFPDW B57
Elite-controller
protective epitope
(Walker and colleagues)
47
Prior to infection, did the breakthrough vaccinees react with epitopes containing
these signature sites?
48
Week 8 ELISpot Reactions with StepVx Sequence 15-mers (N=37 Vaccinees)
49
Vaccinee T Cell Reactions to Vaccine 15-Mers Containing Signature Sites (N=37)
• Of N=37 infected vaccinees evaluated, 4 had a positive ELISpot response to an epitope including a signature site
– Gag 84 Signature: 1 vaccinee (A*0211) had a positive response to Gag SLYNTVATLYCVHQK
2 vaccinees (A*1101) had a positive response to Gag SLYNTVATLYCVHQK
– Pol 721 Signature: 1 vaccinee had a positive response to Pol GIRKVLFLDGIDKAQ and to
Pol DGIDKAQDEHEKYHS
• All Other Signatures: No Vaccinee Reactions
84
721
721
84
50
Breakthrough Sequences for 3 Vaccinees With a Reaction to SLYNTVATLYCVHQK
GAG SLYNTVATLYCVHQK
Con . . F . . . . V . . . . . . .
Seq . . F . . . . V . . . . . . .
Seq . . F . . . . V . . . . . . .
Seq . . F . . . . V . . . . . . .
Seq . . F . . . . V . . . . . . .
Seq . . F . . . . V . . . . . . .
Reacting
Vaccinee 1A0101 A1101
B0801 B35G1
C04G1 G07G1
GAG SLYNTVATLYCVHQK
Con . . F . . . . V . . . . . . .
Seq . . F . . . . V . . . . . . .
Seq . . F . . . . V . . . . . . .
Seq . . F . . . . V . . . . . . .
Seq . . F . . . . V . . . . . . .
Seq . . F . . . . V . . . . . . .
Seq . . F . . . . V . . . . . . .
Seq . . F . . . . V . . . . . . .
Seq . . F . . . . V . . . . . . .
Seq . . F . . . . V . . . . . . .
Seq . . F . . . . V . . . . . . .
Seq . . F . . . . V . . . . . . .
Reacting
Vaccinee 2A1101 A3101
B3503 B51G1
C04G1 C1502
GAG SLYNTVATLYCVHQK
Con . . . . . . . V . . . . . . R
Seq . . . . . . . V . . . . . . R
Seq . . . . . . . V . . . . . . R
Seq . . . . . . . V . . . . . . R
Seq . . . . . . . V . . . . . . R
Seq . . . . . . . V . . . . . . R
Seq . . . . . . . V . . . . . . R
Seq . . . . . . . V . . . . . . R
Seq . . . . . . . V . . . . . . R
Seq . . . . . . . V . . . . . . R
Seq . . . . . . . V . . . . . . R
Seq . . . . . . . V . . . . . . R
Reacting
Vaccinee 3A0211 A02G1
B1504 B1504
C0101 C0102
51
Drill down on the ‘strongest hit’: Gag 84 signature
52
Other OtherA-List A-List
Placebo Vaccine
A-List Alleles N Placebo Vaccine
C14 1 0:1 0:0A0205 2 1:1 0:0A2902 2 1:0 1:0
B58 2 0:1 0:1A1101 5 1:0 0:4B4403 5 1:0 1:3A02G1 26 8:1 5:12
A02*28 9:2 5:12 0.02
A-List 36 10:2 5:19 0.0008
Other 28 7:6 3:12 0.11
T:V
* A02G1, A0202 or A0205
Gag 84 by A-List Epitope-Restricting Alleles
P-value
17% V 79% V
Placebo A*02 (n=11) Vaccine A*02 (n=17)
0.089 0.273Gag 77-85: Mean distance to StepVx SLYNTVATL:
44%
56%
53
Vaccine Selection Pressure May Operate Early
54
Does the T to V difference impact viral load?
Iversen et al. showed that A*02+ patients with efficient CTL selection in SYNTVATL (at sites 3,
6, 8) had low plasma viral loads
55
56
Conclusions
57
Summary and Conclusions
• Global sieve analysis– Borderline significant evidence that vaccinee sequences have
greater epitope-based distances to StepVx than placebo sequences for Gag and especially Nef (not Pol)
• Local sieve analysis of ‘signature’ sites – Statistical evidence for ~10 AA signature sites in Gag, Nef, Pol;
none in Env– Greatest evidence for site 84 in 7 A-list CTL epitopes
• One interpretation: The vaccine-induced selection pressure is specific to the set of HLA-restricted epitopes containing Gag 84
• Another interpretation: The vaccine-induced selection pressure operates on many sites, but for sites in epitopes restricted by rare alleles, there is low statistical power
58
Summary and Conclusions
• Taken together these results support that the vaccine selected against viruses with certain amino acids in T cell epitopes
• This selection pressure did not appear to lead to a vaccine effect on early post-infection markers of disease progression (Janes et al., 2008)
• However, the demonstration that a T cell-based vaccine imposed constraints on the viruses establishing infection may provide guidance for the development of improved T-cell based vaccines
59
Future Analyses and Research
• Ongoing analyses of available data– Evaluate ‘biologically weighted’ epitope-based distances– Use alternative epitope prediction methods– 9-mer scanning analyses– Expand classification analyses to include physical/chemical
properties of AAs– Additional analyses of vaccine-induced selection pressure (e.g.,
compare intra-subject diversity between infected vaccine group and infected placebo group)
– Additional analyses accounting for the timing of sequence-sampling (relative to the timing of development of immune responses)
60
Future Analyses and Research
• Possible follow-up experimentation– Evaluate validity of epitope-based distances via fine-epitope
mapping, especially for those with rare alleles– Compare other phenotypes of breakthrough viruses vaccine vs
placebo (e.g., fitness, infectivity)– Evaluate post-infection T cell responses to peptides (and
variants) containing the signature sites– Deep sequencing of targeted regions at earliest time-point– Measure sequences at a later time-point, especially in those with
a pre-seroconversion sample
61
Acknowledgements
McCutchan labFrancine McCutchan
Sodsai Tovanabutra
Eric Sanders-Buell
Marty Nau
Meera Bose
Andrea Bradfield
Annemarie O' Sullivan
Jacqueline Crossler
Teresa Jones
VIDI/SCHARP Craig Magaret
Allan deCamp
Fusheng Li
Steve Self
Step Study team, including Mike Robertson
Susan Buchbinder
Mullins labDana RaugiStefanie SorensenJill StoddardKim WongHong Zhao
Laura HeathMorgane RollandJim Mullins
VIDI/HVTN labNicole FrahmDavid FriedrichJulie McElrath
Acknowledgments for Helpful AdviceTomer HertzDavid Heckerman David Nickle
62
Extra Slides
63
LANL B
(N=210):
72% T
24% I
1% V
0% -
64
Gag 84 by A*02+/A*02-
SLYNTVATL a well-known
A*02+ immunodominant
Epitope
Edwards et al. (2005, J
Virol) showed positive
selection at site 84 for
A*02+ but not for A*02-
A*02+ =
A*0201 or
A*0202 or
A*0205 for
at least 1 allele
65
Positions 3, 6, 8 in SLYNTVATL (Gag 77-85)
• Iversen et al. (2006, Nat Immun, 7:179-189) found that, for A*02 individuals, SYLNTVATL often acquires CTL escape mutations at positions 3, 6, and 8
• For all 29 A*02 infected subjects, Gag 77-85 in their majority consensus sequence is a known or predicted epitope (w/ prob >.8)
• Gag 77-85: Mean distance to StepVx:
Pos 3 Pos 6 Pos 8
Y F V I T V
Placebo 9 3 (25%)
9 3 (25%)
11 1 (8%)
Vaccine 10 7 (41%)
14 3 (18%)
5 12 (73%)
Numbers of A*02 Subjects with StepVx AA or Mismatch (% Mismatch)
Placebo Vaccine
0.089 0.273
66
67
Estimated RR of Infection by Percent Epitope Mismatch Distance (Account for all 8-11-mers)
68
Estimated RR of Infection by Percent Epitope Mismatch Distance (Account for all 8-11-mers)
69
Estimated RR of Infection by Percent Epitope Mismatch Distance (Account for all 8-11-mers)
70
Estimated RR of Infection by Percent Epitope Mismatch Distance (Known & Likely Epitopes)
71
Estimated Number Shared Epitopes (Known & Likely Epitopes)
72
Estimated Number Shared Epitopes (Account for all 8-11-mers)
73
74
Antigen Scanning of 9-Mers (Ongoing Analyses, Not Reported Here)
• Test each 9-mer as a signature peptide: – Signature 9-mer = a 9-mer where the distribution of peptide-distances to the
StepVx peptide differs in vaccine vs placebo sequences
Vx reference sequence
Vaccinee breakthrough sequences
Placebo breakthrough sequences
H H
H
H
H
H
H
H
H
H
H
H
H
75
Estimated Number Shared Epitopes (Account for all 8-11 Mers)
Interaction p-values: p=.85 p=.64 p=.13
P-values for Ab-: p=.12 p=.18 p=.01
76
Percent Mismatched Epitopes (Account for all 8-11-mers)
Interaction p-values: p=.77 p=.71 p=.08
P-values for Ab-: p=.19 p=1.0 p=.01
77
Structure of Sieve Analysis
• Consider 2 sets of AA sites for the analyses: – Include all sites or linear peptides of length 8, 9, 10, 11
– Restrict to ‘Immunogenic’ sites/linear peptides: • Contained in a StepVx-sequence15-mer recognized by ‘many’ vaccinees (Week 8
ELISpot)
Vaccinee Week 8
ELISpot reactions
with StepVx-sequence
15-mers (N=37)
Data generated by
Nicole Frahm,
David Friedrich,
Julie McElrath
Keep 5% Keep 42%
Keep 8%
78
Sieve Analysis of Step Sequences
VIDI / SCHARP
Craig Magaret, Allan deCamp, Peter Gilbert
in collaboration with
Morgane Rolland, Laura Heath, Jim Mullins
May, 2009
79
80
*
*Major evolving sites [Iversen et al., Nature Immunology, 7:179-189]
81
Sequence Data
• 65 HIV infected male subjects (39 vaccine, 26 placebo)– 62 known to be infected prior to October 17, 2007 (unblinding)– 3 with later 2007 dates of first evidence of HIV infection
• Oct 23, Nov 11, Dec 6
82
Detecting a Sieve Effect
Vaccine
Strains
Placebo group
Infecting HIVs
83
Detecting a Sieve Effect
Vaccine
Strains
Placebo group
Infecting HIVs
Vaccine group
Infecting HIVs
84
Power of Sieve Analysis (n=39 Vaccine; n=25 Placebo)
• Example: Number of shared epitopes in Gag (S1)
85
AA Site Scanning: Departures From 0 Indicates Signature
86
Percent Epitope Mismatch (Known & Likely Epitopes)
Distance
87
Percent Epitope Mismatch (Account for all 8-11-mers)
88
Context of Sieve Analysis
• Type 1 sieve analysis: Conceived as evaluation of ‘selective protection’ against infection– For Step, evaluation of ‘selective enhancement’ is more germane
• Type 1 or 2 sieve analysis: Achieving high statistical power requires:– A large number of infected subjects with sequence data
– A vaccine that induces immune responses that ‘react strongly’ with the infecting viruses (in order to apply selective pressure)
• For Step, the sieve analysis has relatively low power– Small number of infections (Phase 2b, not Phase 3)
– At an epitope level, the vaccine appeared to induce limited potential selective pressure
• On average, a vaccinee recognizes < 1 reactive epitope in an average exposing HIV
89
Number of Amino Acid Sites
Gag Pol Nef Env
All sites 542 886 236 957
‘Immunogenic’ sites
26
(5%)
70
(8%)
98
(42%)
N/A
90
LANL B
(N=324):
52% T
15% A
18% N
10% S
.3% H
91
LANL B
(N=324):
77% I
17% V
3% M
1% A
92
LANL B
(N=324):
95% K
2% R
.3% G
93
LANL B
(N=824):
200% -
94
LANL B
(N=824):
7% I
84% M
2% V
1% A
2% T
.1% L
95
LANL B
(N=210):
92% D
4% E
0% -
.5% N