Transcript Document

80
60
20
Percent (%)
40
0
* G314W
I318V
N305D
A243T
* R259H
* Q147L
E151A
I135L
* T130V
I135I
T130T
* N105S
* M115L
* E125E
* V72L
* K77G
* V69L
M54L
* I62V
* L44M
* E49D
* V28A
* N42S
L7L
* M24M
* G3A
* G3GT
I4I
* V2^
Overall
Average LFCIC50 corrected by overall average
94%
96%
Effects
produce clusters that associated with ENF susceptibility.
Further elucidation of the clusters is ongoing.
 Cluster analysis on B subtype gp41 sequences did not
produce clusters that associated with ENF susceptibility.
B Cluster
(n=365)
Non-B Cluster
(n=12)
Diff.
95% CI
P-value
0.208
0.236
-0.109
-0.109
-0.317
-0.346
(-0.558, -0.076)
(-0.602, -0.089)
0.010
0.009
 Analyses were also conducted using only the data from B
subtype samples versus the complete data. Though there
were some differences in results, the polymorphisms identified
in multi-factor ANOVA and tree regression analysis were
similar.
Regression Tree Analysis Results
-2.315
-2.327
10
13.7
16.67
16.67
27.78
30.96
40.56
45.45
-0.759
-0.771
50
50
66.67
66.67
(-1.409, -0.108)
(-1.402, -0.140)
0.023
0.017
Table 3: Summary of Regression Tree Analysis
0.362
0.675
0.112
0.206
0.128
0.237
[1] Least square means were calculated from ANCOVA model with covariates: Number of ARV, BL CD4, Prior Lopinavir/r
use, GSS, PSS, and BL viral load.
[2] Least square means were calculated from ANCOVA model with covariates: BL CD4, Prior Lopinavir/r use, GSS, PSS,
Adherence to all treatments, and BL viral load.
17
*: This analysis differs from the previous results by Greenberg et
al.6 in using only the R5 tropic data and a novel method for
subtype assignments.
8
* K492X
5
* L486I
5
* L475X
6
* N470X
* L444M
13 10
* M426I
5
* S361T
5
* K340T
* I330V
92 22
D322N
6
* P297L
7
* N287T
5
* Q256H
* L124I
D61
* T50I
* M25I
* G22W
Overall
Average LFCIC50 corrected by overall
5
6.31
3.98
2.51
1.58
1
0.63
0.40
0.25
0.16
H331N
37
H331N
50
100%
N
H331N
60
96%
-1.556
-1.557
A306T
N305D
V301I
Q294Q
S293G
L289L
W286G
W285L
I270I
V267A
L264L
L263F
I245A
R229Q
0.80
0.60
0.40
0.20
0.00
-0.20
-0.40
-0.60
-0.80
7
A306T
N305D
V301I
Q294Q
S293G
L289L
W286G
W285L
I270I
V267A
L264L
L263F
I245A
R229Q
1.9
1.9
Predictors
(ENV regions)
gp41 only
gp41 only
gp120 only
gp120 only
gp120+gp41
gp120+gp41
Forms of
Predictors
a.a. position
indicator
a.a. position
indicator
a.a. position
indicator
# of Tree Mean
Splits
Residual
Deviance *
4
0.144
4
0.147
4
0.143
6
0.154
4
0.139
5
0.141
*: the mean deviance at the root note is 0.176
377 15 13
A306T
N305D*
V301I
Q294Q
S293G
L289L
W286G
W285L
I270I
V267A
L264L
L263F
I245A
R229Q
Figure 2b: Effects of significant gp120
polymorphisms (Dunnett's p-value <0.05).
1.7
Table 2: Summary of BL ENF susceptibility and the
main efficacy endpoints *
Change from BL HIV-1 RNA
(log10 copies/ml)
TORO 2 only (LSM) [1]
TORO 1 & 2 (LSM) [2]
HIV-1 RNA <50 Copies/ml
% responder
TORO 2 only
TORO 1 & 2
HIV-1 RNA <400 Copies/ml
% responder
TORO 2 only
TORO 1 & 2
HIV-1 RNA  log10 decrease %
responder
TORO 2 only
TORO 1 & 2
G221G
The ‘*’ indicates the effect is greater than 0.176, which is
equivalent to greater than 1.5 fold difference from the overall
geometric mean (GM) of fold change IC50. V2^ indicates
deletion at a.a. position 2. G3GT indicates insertion T at a.a.
position 3. The error bars indicate the standard error of the
mean. The numbers on the top indicates the number of
recombinants for each polymorphism. The numbers on the
right y-axis indicate the fold change from the overall GM.
G221G
-0.10
G221G
-1.00
N42S (75% vs. 16%), E151A (92% vs. 23%), N305D (75% vs.
11%) and T130T (42% vs. 9%), which were all associated with
increased ENF susceptibility (Figure 2a).
 Preliminary cluster analysis on gp120 sequences did not
1.6
S [1] D H X
I LTVX
M
24
I LVX
M
mean= 0.235
n=217
dev=29.4
mean= -0.036
n=36
dev=4.0
mean= 0.573
n=43
dev=8.4
mean= 0.149
n=14
dev=2.0
The a.a. positions in the green ovals were used for tree splitting. The
blue and red color indicate an amino acid as wild type or
polymorphism respectively. Rectangles indicate terminal nodes.
Numbers in each terminal node are the LFCIC50 average, the number
of cases in the node and the deviance of the node. [1]: 62 cases of
‘S’ at postition 42, [2]: 297 cases of no insertion at position 3.
 This tree had an ~20% reduction in deviance.
G3GT was
associated with decreased BL ENF susceptibility, N42S and
M24M was associated with increased BL ENF susceptibility,
which was consistent with the results in Figure 2a.
 Non-B Cluster had higher frequencies of polymorphisms
[1]: Differences between clusters were statistically significant (p-value<0.05) from a chi-square test.
[2]: Genotypic and phenotypic sensitivity scores (GSS and PSS) represent the number of drugs in the
regimen to which virus was sensitive, excluding ENF.
Parameter
BL ENF susceptibility (LFCIC50)
TORO 2 only (mean)
TORO 1 & 2 (mean)
E220E
0.16
E220E
-0.80
E220E
0.25
G215G
-0.60
G215G
0.40
G215G
-0.40
A212A
0.63
A212A
-0.20
A212A
1
89.5
4
L210L
0.00
L210L
1.58
L210L
0.20
94
3.8
L209L
2.51
L209L
0.40
L209L
3.98
5 (42%)
8 (75%)
40
7 (58%)
5.43
E151A
0.60
E151A
6.31
E151A*
0.80
336 (91%)
335 (92%)
41
3 (0.82%)
5.20
0
E148D
10
E148D
1.00
E148D
25 27 39 17 225 152 32 94 37 30 48 11 167
12 (100%)
20
T130L
5
T130L
15 280 13 11 33 5
T130L
377 13 12 57 114 105 62 11 68 6
180 (49%)
40
T130T
Figure 2a: Effects of significant gp41
polymorphisms (Dunnett's p-value <0.05).
T130T
TORO 2 (N, %)
[1]
Male (N, %)
White (N, %)
Age (median, years)
[1]
Region (N, % in Beglium)
BL HIV-1 RNA (median, log10 copies/ ml)
3
BL CD4+ T cells (median, cell/mm )
Number of ARV in OB (mean)
[2]
GSS at entry (mean)
[2]
PSS at entry (mean)
Prior Lopinavir/r use (%)
Adherence to ENF (mean)
Adherence to all regimen (mean)
Non-B Cluster
(N=12)
60
E125D
[1]
B Cluster
(N=365)
80
Polymorphisms displayed (with the exception of T130T) are those
whose frequency in clusters differed by >50% or >25 fold. The ‘*’
indicates a significant effect on BL ENF susceptibility (Figure 2a).
1.5
ANOVA Results
E125D
B Cluster (n=365)
T101S
1.0
0
T101S
Regression Tree Analysis
Trees predicting ENF susceptibility were built using gp41,
gp120 or gp41 + gp120 amino acid sequences as
predictors. Initially, amino acid positions were used in the
tree model as predictors. Such an approach may favor
the selection of more variable amino acid positions. To
study this possibility, trees were also built using indicator
variables created for all amino acids at each position.
0.0
0.5
LFCIC50
20
100
Cluster analysis identified two clusters of 365 (Cluster
1) and 12 (Cluster 2) recombinants. Based on the
novel methods of Fenger el at.5, recombinants in
Cluster 1 and Cluster 2 were identified as B subtype
and non-B subtype, respectively. They can be also
referred to as B Cluster and non-B Cluster.
24
42
40
N42S
Cluster Analysis
The agglomerative hierarchical clustering algorithm was
applied using the number of differing amino acids
between two sequences as the distance metric. The
optimal number of clusters was selected to achieve the
maximum Silhouette value (Rousseeuw 1987).
The frequencies of polymorphisms were compared
among the clusters; the genotypic difference among the
clusters were represented by those polymorphisms which
were distributed most differently across the clusters. BL
ENF susceptibility was compared among the clusters
using a Student’s t-Test. Clinical efficacy outcomes were
compared among the clusters using analysis of
covariance (ANCOVA) with baseline characteristics as
covariates.
-0.5
60
N42S
ANOVA
One-way ANOVA was used as a filter for detecting
potentially “interesting” polymorphisms rather than as a
tool for testing a hypothesis. At each gp160 position,
including insertions and deletions, ANOVA was performed
to compare the polymorphisms to JRCSF reference in
terms of the LFCIC50 averages. Dunnett’s adjustment3
was applied for multiple comparisons within each gp160
position. For those polymorphisms with Dunnett’s pvalues < 0.05, individual effects on ENF susceptibility
were expressed as LFCIC50 averages corrected by the
overall LFCIC50 average and the joint effects were
studied through multi-factor ANOVA with backward model
selection using a significance level to stay of 0.05.
-1.0
80
Cluster Analysis Results
Table 1: Demographics, baseline characteristics
and adherence
-1.5
100
T
ALMV
X None[2]
mean= -0.027
n=66,
dev=11.0
Non-B Cluster (n=12)
M24I
Min. -1.398
Median 0.185
Mean 0.198
Max. 1.576
SD
0.421
Geometric
Mean 1.578
E125D
Figure 1: Distribution of BL ENF susceptibility and
summary statistics
T101S
Summary of BL ENF susceptibility
final multi-factor ANOVA model were 3, 3Ins, 24, 42,
54, 69, 77, 130, 135, 243 for gp41 and 297, 340, 361,
471 for gp120. This model had a R2 of 57.6% with 85
degrees of freedom (df). If only the gp41 positions
were used, the ANOVA model had a R2 of 48.0% (83%
of 57.6%) with 59 df, demonstrating a dominating
contribution from gp41.
T130T*
Results
1.59
1.26
1
0.79
0.63
0.50
0.40
0.32
0.25
0.2
0.1
0
-0.1
-0.2
-0.3
-0.4
-0.5
-0.6
The following results would be best confirmed by analysis
with a larger number of non-B samples:
 Non-B Cluster (n=12) has higher BL ENF susceptibility
 Trees built from different forms of gp41 predictors
were similar. Trees built from different forms of gp120
predictors were different, indicating that the tree
algorithm favored the more variable gp120 a.a.
positions.
(p-value =0.01) than B Cluster (n=365).
 The inclusion of gp120 predictors in tree building
 Patients with non-B subtype viruses experienced a
did not significantly improve the tree performance in
terms of reducing mean residual deviance.
larger drop in viral load (p-value=0.023) than patients with
B subtype viruses.
Conclusions
 Data mining of genotypic and phenotypic data have
revealed an association between polymorphic sites in HIV
Env and BL ENF susceptibility in HIV-1 R5 tropic
recombinants. This association was predominantly caused
by gp41 genotype and explained a fraction of the baseline
variability of ENF susceptibility.
 Gp41 polymorphisms G3GT, N42S and M24M were
identified by both ANOVA and regression tree analysis as
associated with ENF susceptibility. In addition, regression
tree models provided insights into how they interacted
together to affect BL ENF susceptibility.
 Cluster analysis on R5 tropic gp41 viral sequences
resulted in divisions of two clusters, non-B subtype (n=12)
and B subtype (n=365) recombinants. Multiple virological
response metrics were compared between clusters. The
only significant differences were a higher BL ENF
susceptibility and larger drop of viral load from baseline for
the non-B Cluster. (See * under Table 2). However, the
subtype B group clearly also demonstrated clinically
meaningful reductions in plasma HIV-1 RNA. The differences
in ENF susceptibility were associated with the higher
frequencies of a combination of polymorphisms in the non-B
Cluster.
 Clusters derived from a preliminary analysis on gp120
sequences, while not associated with BL ENF susceptibility,
will be investigated in future studies.
 Our studies demonstrated the utility of combining multiple
statistical methods to study a large number of potential
factors in exploring the relationship between Env phenotype
and genotype.
References
1. A. Sevin, V. Degruttola, et al. The journal of infectious disease, 2000
2. SB Needleman, CD Wunsch, J Mol Biol 1970; 48(3):443-53.
3. CW Dunnett, J American Statistical Association, 1995, 50: 1096 -1121.
4. TM Therneau, EJ, Atkinson, Rochester, MN: Mayo Clinic, 1997.
5. D. Fenger, C. Su, et al. HIV-1 Subtype Analysis Using gp41 Sequences from
Patients in the Enfuvirtide Phase III Clinical Trials, in preparation.
6. ML Greenberg, T Melby, et al. Baseline and On-Treatment Susceptibility to
Enfuvirtide Seen in TORO 1 and TORO 2 Through 24 Weeks, presented at 10th CROI,
Boston, Feburary 2003.
Poster 54 - XII International HIV Drug Resistance Workshop
 The gp160 amino acid positions included in the
3Ins
M24I
HIV-1 Envelope Genotypic and Phenotypic Data
Using the GeneSeqTM and PhenoSenseTM HIV Entry
assays, BL HIV-1 envelope (complete gp160) amino acid
sequences and ENF susceptibility were generated for
377 R5 tropic recombinant Env pools from patients prior
to their receiving ENF plus an optimized background
(OB) regimen. Needleman-Wunsch algorithm2 was used
to align the gp160 sequences to JRCSF reference.
Insertions occurred at 84 positions. Ambiguous amino
acids were annotated as “X”. ENF susceptibility was
expressed as log10-scaled fold change IC50 (LFCIC50)
relative to JRCSF reference.
listed in Figure 2a and 2b, the best descriptors for
baseline variability of ENF susceptibility were positions
42 (R2=9.3%) and 3Ins (insertion at position gp41 a.a.
3) (R2=9.2%).
Figure 4: Pruned tree with 4 splits for predicting BL
ENF susceptibility based on BL gp41 a.a. sequences.
Figure 3: Polymorphism frequency in clusters and their
effects on BL ENF susceptibility.
N42S*
Methods
 Among the gp41 and gp120 amino acids positions
M24I
Here we present the results of a study of the relationship
between susceptibility to enfuvirtide (ENF) of baseline
(BL) viral recombinants and polymorphisms in the env
region of R5-tropic HIV-1. We explored such
relationships by mining the TORO 1 and TORO 2 clinical
trial databases using both univariate and multivariate
statistical methods. Analysis of variance (ANOVA) was
used to explore this relationship at each individual gp160
position. Cluster analysis and tree regression were used
to explore the joint relationship between the entire gp160
sequence and ENF susceptibility1.
The S-Plus RPART routines of Therneau et al.4 were used to
build trees. The final or optimal trees were determined by
pruning. The within-node sum of square, or deviance, was
used as the measure of variability in trees.
* W471X
Introduction
C. Su1, G. Heilek-Snyder1, D. Fenger1, P. Ravindran1, K. Tsai1, N. Cammack1, P. Sista2, S. Chiu1
1Roche, Palo Alto, CA, USA; 2Trimeris, Inc., Durham, NC, USA
Percent
XII International HIV Drug
Resistance Workshop
Cabo Del Sol, Los Cabos, Mexico
June 10-14, 2003
The Relationship between Susceptibility to Enfuvirtide of Baseline Viral Recombinants and Polymorphisms in
the Env Region of R5-tropic HIV-1
Percent
Poster 54
Cheng Su
Roche Palo Alto
3431 Hillview Avenue
Palo Alto, CA 94304
Tel: 650-855-6510
Fax: 650-855-5627
E-mail: [email protected]
- XII International HIV Drug Resistance Workshop, Cabo Del Sol, Los Cabos, Mexico, June 10-14, 2003
Poster 54
XII International HIV Drug Resistance Workshop
Cabo Del Sol, Los Cabos, Mexico
June 10-14, 2003
The Relationship between
Susceptibility to Enfuvirtide of
Baseline Viral Recombinants and
Polymorphisms in the Env
Region of R5-tropic HIV-1
C. Su1, G. Heilek-Snyder1, D. Fenger1,
P. Ravindran1, K. Tsai1, N. Cammack1,
P. Sista2, S. Chiu1
1Roche,
Palo Alto, CA, USA
2Trimeris, Inc., Durham, NC, USA