Clinical validation of saquinavir/ritonavir genotypic resistance score in protease-inhibitor-experienced patients.

Abstract

To identify a genotypic score for resistance to saquinavir boosted with ritonavir (SQV/r; 1,000/100 mg twice daily)-based regimens in protease inhibitor (PI)-experienced patients.
One-hundred and fifty-one PI-experienced patients receiving a SOV/r-containing regimen were enrolled retrospectively. The virological response (VR) was defined as the decrease in HIV RNA at months 3-5. The effect of each mutation in the protease gene on the VR to SQV/r regimen was assessed using non-parametric univariate analyses and then a step-by-step analysis was carried out using a Jonckheere-Tepstra (JT) nonparametric test to retain the group of mutations most strongly associated with VR.
Among the 138 patients with detectable plasma SQV, the median VR was -1.48 [range: -4 to +1.2] log10 copies/ml. Changes at 12 codons were associated with a reduced VR to SQV/r: codons 10, 15, 20, 24, 46, 54, 62, 71, 73, 82, 84 and 90. The JT procedure led to selection of the following genotypic score, 10+15+20+ 24+62+73+82+84+90, as providing the strongest association with VR. In the 35 patients with none of the mutations in this score, the median decrease in HIV RNA was -2.24 log10 copies/ml and it was -1.88 (n=29), -1.43 (n=24), -0.52 (n=30), -0.18 (n=9), -0.11 (n=6) and -0.30 (n=5) log10 copies/ml in those with 1, 2, 3, 4, 5 and 6 mutations, respectively.
With this resistance score to SQV/r, the isolates were classified as having no evidence of resistance (0-2), possible resistance (3) or resistance (> or =4) by grouping the number of mutations in samples for which the viral load reduction was similar.

Full text

Page 1

Objective: To identify a genotypic score for resistance to
saquinavir boosted with ritonavir (SQV/r; 1,000/100 mg
twice daily)-based regimens in protease inhibitor
(PI)-experienced patients.
Methods: One-hundred and fifty-one PI-experienced
patients receiving a SQV/r-containing regimen were
enrolled retrospectively. The virological response (VR) was
defined as the decrease in HIV RNA at months 3–5. The
effect of each mutation in the protease gene on the VR
to SQV/r regimen was assessed using non-parametric
univariate analyses and then a step-by-step analysis was
carried out using a Jonckheere–Tepstra (JT) non-
parametric test to retain the group of mutations most
strongly associated with VR.
Results: Among the 138 patients with detectable plasma
SQV, the median VR was -1.48 [range: -4 to +1.2]
log10copies/ml. Changes at 12 codons were associated
with a reduced VR to SQV/r: codons 10, 15, 20, 24, 46,
54, 62, 71, 73, 82, 84 and 90. The JT procedure led to
selection of the following genotypic score, 10+15+20+
24+62+73+82+84+90, as providing the strongest asso-
ciation with VR. In the 35 patients with none of the
mutations in this score, the median decrease in HIV RNA
was -2.24 log10copies/ml and it was -1.88 (n=29), -1.43
(n=24), -0.52 (n=30), -0.18 (n=9), -0.11 (n=6) and -0.30
(n=5) log10 copies/ml in those with 1, 2, 3, 4, 5 and 6
mutations, respectively.
Conclusion: With this resistance score to SQV/r, the
isolates were classified as having no evidence of resis-
tance (0–2), possible resistance (3) or resistance (≥4) by
grouping the number of mutations in samples for which
the viral load reduction was similar.
Clinical validation of saquinavir/ritonavir genotypic
resistance score in protease-inhibitor-experienced
patients
Anne-Genevieve Marcelin1*, Philippe Flandre2, Carmen de Mendoza3, Benedicte Roquebert1,
Gilles Peytavin4,5, Luisa Valer3, Marc Wirden1, Sarah Abbas6, Christine Katlama4, Vincent Soriano3
and Vincent Calvez1
1Department of Virology, Pitié-Salpêtrière Hospital, AP-HP, Université Pierre et Marie Curie, Paris, France
2INSERM U720,Pitié-Salpêtrière Hospital, Paris, France
3Department of Infectious Diseases, Carlos III Hospital, Madrid, Spain
4Department of Pharmacy,Bichat-Claude Bernard Hospital, AP-HP, Université Paris 7,Paris, France
5Department of Infectious Diseases,Pitié-Salpêtrière Hospital, AP-HP, Université Pierre et Marie Curie, Paris, France
6ROCHE France Laboratory, Neuilly, France
Corresponding author: Tel: +33 142177401; Fax: +33 142177411; E-mail: anne-genevieve.marcelin@psl.aphp.fr
This work was presented at the 4th European HIV Drug Resistance Workshop, 29–31 March 2006, Monte Carlo, Monaco. Abstract 54.
Antiviral Therapy 12:247–252
Saquinavir (SQV) is a potent protease inhibitor (PI)
in vitro, but its poor oral bioavailability limits its clin-
ical activity when used as a single PI [1,2]. Ritonavir
(RTV), a potent inhibitor of cytochrome P450, leads to
an approximately 20-fold increase in SQV plasma
exposure when the drugs are coadministered [3].
Consequently, the coadministration of SQV with a low
dose of RTV (100 mg twice daily) increases the expo-
sure to SQV. Numerous clinical trials have evaluated
SQV plus RTV (SQV/r) treatment regimens
(1000 mg/100 mg twice daily and 1,600 mg/100 mg
once daily) in PI-naive and -experienced patients,
showing a potent and sustained viral suppression [4].
Reduced susceptibility to SQV is most often associ-
ated with acquisition of the G48V and L90M muta-
tions in the HIV-1 protease when SQV is used as first PI
[5,6]. Individually these mutations result in <10-fold
changes in SQV susceptibility, and the less frequent
double mutations (G48V and L90M) lead to up to 100-
fold resistance [7,8]. In the Stanford HIV drug resis-
tance database (http://hivdb.stanford.edu/index.html),
the median fold resistance for G48V as the only major
Introduction
© 2007 International Medical Press 1359-6535
247

Page 2

mutation is 8 and for L90M the corresponding value is
2.2 [9]. Additional mutations have also been reported to
arise during SQV therapy at protease codons 10, 36, 63,
71, 73, 82 and 84 [5,6,10,11]. L90M, the most
common mutation selected by SQV, confers cross-resis-
tance to other PIs, especially when associated with
minor protease mutations. The specific mutation G48V
is usually selected later than L90M and at higher SQV
concentrations [12].
PI-resistance mutations detected before initiation of
SQV/r therapy in PI-experienced patients are the
strongest predictors of virological response [13].
However, most of the current resistance algorithms
consider only the mutations affecting the virological
response to unboosted SQV therapy or, in some cases,
only the response to ritonavir-boosted SQV
(800/100 mg twice daily) [14]. A previous study
assessed the effect of protease mutations on the
response to salvage therapy with SQV/r (1,000/100 mg
twice daily), but considered only mutations included in
the International AIDS Society (IAS) list of mutations
(2000) [15]. With the availability of the new SQV
formulation (500 mg in SQV mesylate tablet), there is
a need to re-evaluate the genotypic interpretation of
resistance to SQV/r (1,000/100 mg twice daily). It is
now recognized that correlation studies analysing the
virological response in treatment-experienced patients
according to the genotypic profile at baseline provide
relevant information for establishing resistance algo-
rithms. To this end, we developed a clinically relevant
genotypic interpretation score for resistance to SQV/r.
Patients and methods
Patients and antiretroviral regimens
One-hundred and fifty-one PI-experienced patients were
retrospectively selected. All were treated with nucleo-
side reverse transcriptase inhibitors (NRTI) and RTV
(100 mg twice daily) plus SQV (1,000 mg twice daily).
No PIs other than SQV and RTV were used in the anti-
retroviral combinations. The main characteristics of the
study population are reported in Table 1.
HIV-1 RNA quantification
Quantification of plasma HIV-1 RNA was performed
at baseline and after 3–5 months of treatment using the
Amplicor Monitor assay (Cobas 1.5, Roche
Diagnostics, Basel, Switzerland), which has a lower
detection limit of 200 copies/ml.
Genotypic resistance testing
Plasma samples were collected at baseline. The reverse
transcriptase (RT) and protease gene sequences were
determined using population sequencing according to
the Agence Nationale de Recherches sur le SIDA
consensus method with an ABI 3100 Genetic Analyzer
(PE Applied Biosystems, Foster City, CA, USA). The
sequences were analysed using Sequence Navigator
software (PE Applied Biosystems, as before) and
reported as amino acid changes from the reference
sequence of the wild-type virus HXB2.
PI plasma concentration determinations
Plasma samples were collected at months 3–5 approx-
imately 12 h after the last intake at steady state. Last
intake and time of sampling were recorded. SQV
plasma concentrations were measured by a specific
and validated high-performance liquid chromato-
graphic assay coupled with a UV detection at 240 nm.
Briefly, SQV and its internal standard were isolated
from plasma after alkalinization by a double-step-back
liquid–liquid extraction by methyl tert-butylether. The
separation was carried out on an octadecylsilyl analyt-
ical column with a mobile phase containing 0.1%
trifluoroacetic acid, acetonitrile and methanol (55:45,
v/v). The assay was found to be linear and has been
validated over the SQV concentration range
9–1,000 μg/l from 500 μl of plasma. The coefficients
of variation within day and between days were 4.6%
and 4.5%, respectively. The lower limit of quantifica-
tion was 9 ng/ml and was defined as the concentration
for which the relative standard deviation and the
percent deviation from the nominal concentration
were lower than 20%. Patients taking SQV/r
(1,000/100 mg twice daily) should have SQV plasma
concentrations >9 ng/ml at all times, so only patients
with SQV plasma concentrations >9 ng/ml were
retained for the determination of the resistance score
[16]. The rest were excluded and considered as non-
compliant with the medication.
Statistical methods
The endpoint for the analysis was the change in
plasma HIV-1 RNA (log10copies/ml) between day 0
and months 3–5. First the effect of the presence of
A-G Marcelin et al.
© 2007 International Medical Press
248
Plasma HIV-1 RNA*, log10copies/ml
CD4+T-cell cell count*, cells/mm3
NRTIs and NNRTIs associated with SQV/r, n (%)
1 NRTI
1 NRTI and 1 NNRTI
2 NRTIs
2 NRTIs and 1 NNRTI
3 NRTIs
4.24 (3.7 to 4.9)
292 (185 to 425)
2 (1)
7 (5)
120 (87)
5 (4)
4 (3)
*Values given as median (range). NRTI, nucleoside reverse transcriptase
inhibitor; NNRTI, non-nucleoside reverse transcriptase inhibitor; SQV/r,
saquinavir boosted with ritonavir.
Table 1. Baseline characteristics of the study population
(n=138)

Page 3

each mutation along the protease gene (codons 1 to
99) on virological response was analysed, by
comparing the change in plasma HIV-1 RNA in
patients with and without the mutation using a non-
parametric Wilcoxon–Mann–Whitney test. Mutations
that were present in at least 5% of patients and for
which the P-value was lower than 0.10 in the
univariate analysis were retained for the removal
procedure. The removal procedure, using a non-para-
metric test, allowed selection of a final set of muta-
tions associated with virological response [17–19].
The removal procedure begins with all k mutations
retained from the univariate analysis. We used the
non-parametric Jonckheere–Terpstra (JT) test for
ordered alternatives, which allows testing of a specific
a priori sequence [18–20]. From the initial set of k
mutations, the first step is to remove one mutation.
One by one all combinations of k-1 mutations are
investigated and the combination providing the lowest
P-value with the JT test is retained. In the second step,
mutations are again removed one by one to compare
all combinations of k-2 mutations, the combination
providing the lowest P-value is again retained, and the
process is repeated [17]. The procedure ends when
removing a mutation does not provide a lower P-value
than the previous one.
Finally, we used the bootstrap resampling method
to assess the robustness of the score obtained [21,22].
Multivariate analyses of the score were performed on
500 samples drawn from the 138 patients by
sampling with replacement. We report the number of
times the score had a P-value <0.05, <0.001 and
<0.0001 and the median reductions in viral load in
the resistance groups.
The statistical program used for analyses was SAS
(version 9.0, SAS Institute Inc., Cary, NC, USA).
Results
Baseline patient characteristics
All patients included were followed in two medical
centres (Pitié-Salpêtrière Hospital, Paris, France and
Carlos III Hospital, Madrid, Spain) and all had available
baseline genotype and viral load measurements at base-
line and at months 3–5. Among the 151 patients enrolled
in the study, 13 patients were excluded from the analysis
because they had undetectable SQV plasma levels,
suggesting that they were completely non-adherent. For
the 138 patients retained in the analysis, the details of
the antiretroviral drugs associated with SQV/r are
summarized in Table 1 and the numbers of major and
minor PI-resistance mutations using the IAS-USA panel
list are described in Table 2 [23].
Effect of PI-resistance mutations on virological
response
The median decrease in plasma HIV RNA between
baseline and months 3–5 in the patients exposed to
SQV/r was -1.48 log10copies/ml (range: -4 to +1.2).
Mutations at 12 codons were found to be associated
with a reduced virological response to SQV/r (P<0.1):
L10F/I/M/R/V, I15A/V, K20I/M/R/T, L24I, M46I/L,
I54A/V, I62V, A71I/T/V, G73S/T, V82A/F/S/T, I84V
and L90M. Table 3 shows the univariate analysis of the
virological response according to the presence of
mutated or wild-type codons at specific sites of the
protease gene. Three mutations were significantly asso-
ciated with a better virological response to SQV/r:
T12A/D/N/P/Q/S, I13V and K14N/R.
Boosted SQV/r genotypic score
Removal procedure. From the univariate analysis we
retained 12 codon changes (L10F/I/M/R/V, I15A/V,
K20I/M/R/T, L24I, M46I/L, I54A/V, I62V, A71I/T/V,
G73S/T, V82A/F/S/T, I84V and L90M). In the
removal procedure, only the mutations associated
with poorer virological response were included. The
strongest association between the decrease in viral load
and the ordered number of mutations was observed
with the following set L10F/I/M/R/V+I15A/V+
K20I/M/R/T+L24I+I62V+G73S/T+V82A/F/S/T+I84V+
L90M (P=2.23×10-11). Figure 1 shows the median
decrease in viral load according to the number of muta-
tions from the best combination present. The viral load
decrease was significantly smaller when the number of
mutations increased. In the 35 patients with 0 muta-
tions within this score, the median decrease in HIV
RNA was -2.24 log10 copies/ml and it was -1.88
(n=29), -1.43 (n=24), -0.52 (n=30), -0.18 (n=9), -0.11
Antiviral Therapy 12:2
249
Boosted saquinavir resistance score
Major IAS mutations, n
0
1
2
3
4
5
Minor IAS mutations, n
0
1–3
4–6
7–11
Patients with no IAS major mutations, n=46
0 minor mutation
1–3 minor mutations
4–6 minor mutations
Patients, n (%)
46 (33)
27 (20)
45 (33)
10 (7)
9 (7)
1 (1)
2 (1)
50 (36)
65 (47)
21 (15)
2 (4)
34 (74)
10 (22)
IAS, International AIDS Society.
Table 2. Baseline genotypic characteristics of the study
population (n=138)

Page 4

(n=6) and -0.30 (n=5) log10copies/ml in those with 1,
2, 3, 4, 5 and 6 mutations, respectively. A second
analysis was carried out, in which all mutations
described in Table 3, the 12 mutations associated with
poorer virological response and the three mutations
providing better virological response, were included at
the beginning of the removal procedure, but the inclu-
sion of these mutations in the model did not signifi-
cantly improve the P-value of the resistance score.
Bootstrap analyses. Bootstrap analysis with the geno-
typic score showed that 485/500 (97%) samples,
463/500 (93%) samples and 439/500 (88%) of
samples had a P-value of <0.05, <0.001 and <0.0001,
respectively (JT test). The estimated median decreases
in log10copies/ml of HIV-1 RNA were 2.24, 1.82,
1.40, 0.76, 0.18, 0.01 and 0.20 in patients with a
genotypic score of 0, 1, 2, 3, 4, 5 and 6, respectively.
A-G Marcelin et al.
© 2007 International Medical Press
250
Figure 1. Median decrease in plasma HIV-1 RNA according to
the number of mutations in the resistance score
Mutations are L10F/I/M/R/V, I15A/V, K20I/M/R/T, L24I, I62V, G73S/T, V82A/F/S/T,
I84V and L90M. JT, Jonckheere–Terpstra.
Codon (wild-type
virus HXB2)
Median
change in VL
P-value
(Mann–Whitney)Amino acid
n
IQR
L10L
F,I,M,R,V
I
A,V
K
I,M,R,T
L
I
M
I,L
I
A,V
I
V
A
I,T,V
G
S, T
V
A,F,S,T
I
V
L
M
T
A,D,K,N,P,Q,S
I
S,V
K
N,R
84
54
113
25
113
25
127
11
94
44
110
28
97
41
81
57
128
10
87
51
125
13
105
33
122
16
105
33
116
22
-1.91
-0.41
-1.59
-1.07
-1.59
-1.05
-1.60
-0.20
-1.75
-1.10
-1.62
-0.58
-1.60
-0.97
-1.73
-1.05
-1.56
-0.37
-1.88
-0.88
-1.61
-0.12
-1.75
-0.30
-1.39
-2.09
-1.31
-1.87
-1.34
-1.96
-2.49 to -1.26
-1.73 to -0.08
-2.35 to -0.39
-1.88 to -0.10
-2.45 to -0.28
-1.85 to -0.15
-2.35 to -0.30
-0.78 to -0.00
-2.45 to -0.28
-1.83 to -0.19
-2.35 to -0.44
-1.72 to -0.08
-2.39 to -0.39
-1.95 to -0.05
-2.39 to -0.44
-1.95 to -0.18
-2.35 to -0.28
-1.23 to -0.12
-2.45 to -0.77
-1.73 to -0.08
-2.36 to -0.63
-0.22 to -0.05
-2.45 to -0.78
-1.59 to -0.10
-2.19 to -0.22
-2.89 to -1.18
-2.19 to -0.23
-2.39 to -0.94
-2.20 to -0.22
-2.53 to -1.07
<0.0001
I150.058
K200.076
L24 0.006
M46 0.022
I540.020
I62 0.020
A710.028
G730.034
V820.0007
I84<0.0001
L900.0002
T120.030
I130.086
K140.083
IQR, interquartile range; VL, viral load.
Table 3. Univariate analysis of the virological response according to the presence of mutated or wild-type codons at specific
sites of the protease gene

Page 5

Discussion
The clinical validation of genotypic resistance scores is
based on the assessment of the effect of genotypic
profiles at baseline on the subsequent virological
response. This provides objective information for guid-
ance of drug selection, particularly for treatment-expe-
rienced patients [17,24]. We have determined a
genotypic resistance score for ritonavir-boosted SQV in
PI-experienced patients using stepwise methodologies,
as previously proposed [17–19].
The most widely used combination of SQV/r is at
doses of 1,000 mg and 100 mg, respectively, twice a day.
This results in adequate plasma SQV concentrations in
almost all cases [25]. In our study, the median decrease
in plasma HIV RNA between baseline and months 3–5
in PI-experienced patients receiving SQV/r-based regi-
mens was -1.48 log10copies/ml (range: -4 to +1.2). HIV
RNA changes at months 3–5 seem to be a reasonable
and convenient endpoint for determining success. Other,
similar studies have used HIV RNA changes from base-
line to earlier and later timepoints as their variables. Use
of an earlier endpoint could result in underestimation of
the response and a later endpoint could be compromised
by treatment changes. However, it might be useful to
define a standard endpoint for this type of study, to
allow clinicians to compare the scores of different PIs.
Changes at nine codons were found to constitute a
resistance score significantly associated with viro-
logical response to SQV/r: L10F/I/M/R/V+I15A/V+
K20I/M/R/T+L24I+I62V+G73S/T+V82A/F/S/T+I84V
+L90M. In the 35 patients with 0 mutations within
this score, the median decrease in HIV RNA was -2.24
log10copies/ml and it was -1.88 (n=29), -1.43 (n=24),
-0.52 (n=30), -0.18 (n=9), -0.11 (n=6) and -0.30 (n=5)
log10copies/ml in those with 1, 2, 3, 4, 5 and 6 muta-
tions, respectively. To determine a resistance score we
grouped the number of mutations for which the viral
load reduction was similar. Indeed, in these experi-
enced patients there was a high probability of a
>1 log10reduction of viral load with 0–2 mutations,
around 0.5 log10 reduction with 3 mutations and
around 0 log10reduction with more than 4 mutations.
This suggests that isolates can be classified, according
to the number of mutations in the genotypic score, as
no evidence of resistance (0–2), possible resistance (3)
or resistance (≥4) (P=4.2×10-9). One of the limitations
of this study is that all mutations are considered as
having equal importance. We did not try to weight the
different mutations in the score, because there is no
standardized method at present. However, such an
approach should be developed and would improve the
clinical value of resistance-score interpretation.
Some of these nine mutations (at codons 24, 62, 82,
84 and 90) have been previously described in a resistance
score for SQV/r (800/100 mg twice daily) [14] and some
(at codons 10, 82, 84 and 90) have been previously asso-
ciated with reduced virological response to SQV/r [15].
The mutation at position 73 has already been described
as associated with SQV resistance and is now included in
the IAS list of mutations for SQV.
Two mutations (at positions 15 and 20) not previously
described as involved in SQV resistance were identified in
our resistance score for SQV/r. Mutations at position 20
are known to be associated with resistance to several PIs,
such as atazanavir, indinavir, lopinavir/r and tipranavir.
This mutation appears to result in cross resistance to PIs
and therefore it is not surprising to find an association
between its presence and a decreased virological response
to SQV/r. One of the nine mutations (at position 15) has
not been previously associated with the reduction of PI
efficacy. However, mutation I15A has never been
described in PI-naive patients with any HIV subtype, is
very rare, even in PI-experienced patients, and is always
associated with other major PI resistance mutations. In
PI-naive patients, I15V exists in 25% of patients carrying
subtype B and in up to 80% of patients with C, D or F
subtypes. The frequency of mutation I15V is higher in PI-
experienced patients, especially in CRF_02 variants, and
appears from the Stanford database to be linked to the
use of amprenavir or indinavir [26,27]. Previous rules for
HIV-1 genotypic resistance testing for SQV advised
consideration of resistance only when mutations G48V,
I84V or L90M were present. However, these rules were
mainly based on ritonavir-unboosted SQV-containing
regimens. Moreover, the G48V mutation was underrep-
resented in the population we studied because our
patients were SQV-naive and the G48V mutation is very
specific for this drug, and so it could not be accounted for
in the genotypic score. However, it could have an effect
on the response to SQV/r. Indeed, the most common
mutations selected by SQV at failure are G48V and
L90M. The G48V mutation is specifically selected by
SQV and occurs later than the L90M mutation [12]. This
suggests that isolates harbouring this mutation should be
considered at least partially resistant to SQV/r.
In this study, three mutations at positions 12, 13
and 14 were significantly associated with a better viro-
logical response to SQV/r in univariate analysis, but,
when added to the model, did not significantly
improve the accuracy of the resistance score. This
suggests that these mutations could have a minor
effect on virological response to SQV/r, but less than
other mutations remaining in the final model.
It is now recognized that validation, or measurement
of the ability of a rule to predict clinical response, is
important for establishing clinical utility. Ideally, this
new SQV/r resistance score should be cross-validated in
a data set independent of the discovery set from which
the model was derived. Moreover, it has been shown
Antiviral Therapy 12:2
251
Boosted saquinavir resistance score