Standardized comparison of the relative impacts of HIV-1 reverse transcriptase (RT) mutations on nucleoside RT inhibitor susceptibility.
ABSTRACT Determining the phenotypic impacts of reverse transcriptase (RT) mutations on individual nucleoside RT inhibitors (NRTIs) has remained a statistical challenge because clinical NRTI-resistant HIV-1 isolates usually contain multiple mutations, often in complex patterns, complicating the task of determining the relative contribution of each mutation to HIV drug resistance. Furthermore, the NRTIs have highly variable dynamic susceptibility ranges, making it difficult to determine the relative effect of an RT mutation on susceptibility to different NRTIs. In this study, we analyzed 1,273 genotyped HIV-1 isolates for which phenotypic results were obtained using the PhenoSense assay (Monogram, South San Francisco, CA). We used a parsimonious feature selection algorithm, LASSO, to assess the possible contributions of 177 mutations that occurred in 10 or more isolates in our data set. We then used least-squares regression to quantify the impact of each LASSO-selected mutation on each NRTI. Our study provides a comprehensive view of the most common NRTI resistance mutations. Because our results were standardized, the study provides the first analysis that quantifies the relative phenotypic effects of NRTI resistance mutations on each of the NRTIs. In addition, the study contains new findings on the relative impacts of thymidine analog mutations (TAMs) on susceptibility to abacavir and tenofovir; the impacts of several known but incompletely characterized mutations, including E40F, V75T, Y115F, and K219R; and a tentative role in reduced NRTI susceptibility for K64H, a novel NRTI resistance mutation.
- SourceAvailable from: Charles Boucher[show abstract] [hide abstract]
ABSTRACT: HIV-1 nucleoside reverse transcriptase inhibitors (NRTIs) have been used in the clinic for over twenty years. Interestingly, the complete resistance pattern to this class has not been fully elucidated. Novel mutations in RT appearing during treatment failure are still being identified. To unravel the role of two of these newly identified changes, E40F and K43E, we investigated their effect on viral drug susceptibility and replicative capacity. A large database (Quest Diagnostics database) was analysed to determine the associations of the E40F and K43E changes with known resistance mutations. Both amino acid changes are strongly associated with the well known NRTI-resistance mutations M41L, L210W and T215Y. In addition, a strong positive association between these changes themselves was observed. A panel of recombinant viruses was generated by site-directed mutagenesis and phenotypically analysed. To determine the effect on replication capacity, competition and in vitro evolution experiments were performed. Introduction of E40F results in an increase in Zidovudine resistance ranging from nine to fourteen fold depending on the RT background and at the same time confers a decrease in viral replication capacity. The K43E change does not decrease the susceptibility to Zidovudine but increases viral replication capacity, when combined with E40F, demonstrating a compensatory role for this codon change. In conclusion, we have identified a novel resistance (E40F) and compensatory (K43E) change in HIV-1 RT. Further research is indicated to analyse the clinical importance of these changes.Retrovirology 02/2008; 5:20. · 5.66 Impact Factor
- [show abstract] [hide abstract]
ABSTRACT: Although 2 widely used susceptibility assays have been developed, their precision and sensitivity have not been assessed. To assess the precision of the Antivirogram and PhenoSense assays, we examined susceptibility results of HIV-1 isolates lacking drug resistance mutations and containing matching patterns of drug resistance mutations. To assess sensitivity, we determined for each assay the proportion of isolates with common patterns of matching drug resistance mutations having reductions in susceptibility greater than those in isolates without drug resistance mutations. We analyzed protease inhibitor (PI) susceptibility results obtained by the Antivirogram assay for 293 isolates and by the PhenoSense assay for 300 isolates. We analyzed reverse transcriptase (RT) inhibitor susceptibility results obtained by the Antivirogram assay for 202 isolates and by the PhenoSense assay for 126 isolates. For wild-type and mutant isolates, the median absolute deviance of the fold resistance of nucleoside RT inhibitor susceptibility results was significantly lower for the PhenoSense assay than for the Antivirogram assay. The PhenoSense assay was also significantly more likely than the Antivirogram assay to detect resistance to abacavir, didanosine, and stavudine in isolates with the common drug resistance mutations M41L, M184V, and T215Y (+/-L210W). We found no significant differences between the 2 assays for detecting PI and nonnucleoside RT inhibitor resistance. The PhenoSense assay is more precise than the Antivirogram assay and superior at detecting resistance to abacavir, didanosine, and stavudine.JAIDS Journal of Acquired Immune Deficiency Syndromes 05/2005; 38(4):439-44. · 4.65 Impact Factor
- [show abstract] [hide abstract]
ABSTRACT: To evaluate HIV-1 reverse transcriptase genotypic and phenotypic indicators of resistance to abacavir (ABC) as predictors of ABC antiviral efficacy. The study was a retrospective, combined analysis of five multicentre trials in which ABC was added as a single agent to background antiretroviral therapy in experienced adults. Baseline HIV-1 genotype and phenotypic susceptibility to ABC were determined and the association of genotype and phenotype with virological response after addition of ABC was analysed. Overall, 68% of these therapy-experienced subjects had a virological response (>0.5 log10 or <400 copies/ml; 42% <400 copies/ml) 4 weeks after addition of ABC. Multivariable analyses revealed no significant difference in the response rate between subjects with wild-type virus and those carrying virus with 1-2 nucleoside reverse transcriptase inhibitor (NRTI)-associated mutations. At the 4-week time-point subjects harbouring virus with > or = 3 mutations associated with NRTI resistance were significantly less likely to respond to ABC than were subjects harbouring wild-type virus (P=0.015). However, at the last viral RNA measurement after addition of ABC (12-28 weeks), > or = 4 mutations were required to diminish virological response significantly (P=0.012). Phenotypic resistance was also predictive of antiviral response. Significant breakpoints were identified for virological responses for the PhenoSense HIV assay and the Antivirogram assay. CD4 responses generally paralleled the antiviral responses with a median increase of 55 cells/microl by weeks 12-28. Virological response to ABC may be diminished significantly by multiple NRTI-associated mutations and/or by reductions in phenotypic susceptibility to ABC. However, many subjects with baseline samples showing evidence of resistance to NRTIs respond to ABC.Antiviral therapy 03/2004; 9(1):37-45. · 3.07 Impact Factor
Standardized Comparison of the Relative Impacts of HIV-1 Reverse
Transcriptase (RT) Mutations on Nucleoside RT Inhibitor
George L. Melikian,aSoo-Yon Rhee,aJonathan Taylor,bW. Jeffrey Fessel,cDavid Kaufman,dWilliam Towner,ePaolo V. Troia-Cancio,f
Andrew Zolopa,aGregory K. Robbins,gRon Kagan,hDennis Israelski,aand Robert W. Shafera
Division of Infectious Diseases, Department of Medicine, Stanford University, Stanford, California, USAa; Department of Statistics, Stanford University, Stanford, California,
USAb; Kaiser Permanente Medical Care Program—Northern California, San Francisco, California, USAc; Mount Sinai Medical Center, New York, New York, USAd;
Department of Infectious Diseases, Kaiser Permanente Los Angeles, Los Angeles, California, USAe; Division of Infectious and Immunologic Diseases, UC Davis Medical
Center, Davis, California, USAf; Division of Infectious Diseases, Harvard University, Boston, Massachusetts, USAg; and Quest Diagnostics Incorporated, San Juan Capistrano,
Each of the initial treatment regimens recommended by the De-
partment of Health and Human Services (34) and the World
an ARV belonging to a second drug class.
In a previous study, we applied several data-mining ap-
proaches to quantify associations between NRTI-associated
HIV-1 drug resistance mutations and in vitro susceptibility
data (24). About 630 susceptibility test results were available
for abacavir (ABC), didanosine (ddI), lamivudine (3TC), sta-
vudine (d4T), and zidovudine (AZT), and 350 were available
for tenofovir (TDF). In that study, we used a predefined list of
nonpolymorphic NRTI-selected mutations to reduce the number
analyze a data set that is about twice as large and uses two regres-
sion methods in tandem: one to identify genotypic predictors of
NRTI susceptibility from the many RT mutations present in the
data set (rather than relying on a predefined list of mutations, as
we did previously) and one to quantify the impact of RT muta-
tions on NRTI susceptibility. In addition, we used several ap-
proaches to determine whether models that included statistical
diction of reductions in NRTI susceptibility.
ucleoside/nucleotide reverse transcriptase (RT) inhibitors
(NRTIs) are the backbone of antiretroviral (ARV) therapy.
MATERIALS AND METHODS
HIV-1 isolates. We analyzed HIV-1 isolates in the HIV Drug Resistance
been performed by the PhenoSense (Monogram, South San Francisco,
previously by other laboratories; 65% were from studies by our research
group or from data recently contributed by one of several collaborating
the published literature previously (for a copy of the data set, see the
supplemental material). The Stanford University Human Subjects Com-
mittee approved this study.
Drug susceptibility results were expressed as the fold change in sus-
ceptibility, defined as the ratio of the 50% effective concentration (EC50)
for a tested isolate to that for a standard wild-type control isolate. EC50
results for 3TC and emtricitabine (FTC) with a fold change in suscepti-
bility of ?200 were censored (i.e., reported as ?200) by the PhenoSense
as well as for AZT, for samples which had fold change results of ?200.
The subtype of each isolate either was determined by using the REGA
subtyping algorithm (5) and the NCBI viral genotyping resource (26) or
was identified directly from the phenotype report. Mutations were de-
Received 9 August 2011 Returned for modification 16 September 2011
Accepted 3 February 2012
Published ahead of print 13 February 2012
Address correspondence to George L. Melikian, firstname.lastname@example.org.
Supplemental material for this article may be found at http://aac.asm.org/.
Copyright © 2012, American Society for Microbiology. All Rights Reserved.
The authors have paid a fee to allow immediate free access to this article.
0066-4804/12/$12.00Antimicrobial Agents and Chemotherapyp. 2305–2313 aac.asm.org
(available at http://hivdb.stanford.edu/pages/documentPage/consensus
_amino_acid_sequences.html). Nonpolymorphic mutations were de-
fined as mutations that occur at a prevalence of ?0.5% in the absence of
ARV selective pressure (1).
To minimize bias, we excluded susceptibility results obtained when
more than one virus from the same individual contained the same muta-
tions at the following influential NRTI resistance positions: 65, 74, 115,
151, 184, and 215. Because the presence of mixtures may confound gen-
otype-phenotype correlations, we also excluded viruses with sequences
containing electrophoretic mixtures at these positions.
tibility. To identify mutations that decrease susceptibility to one or more
NRTIs, we used the LASSO (least absolute shrinkage and selection oper-
ator) procedure to examine all mutations occurring in 10 or more virus
samples. LASSO is particularly useful for selecting a subset of predictors
by fitting a least-squares solution with the added constraint that ? |?i|1
(the L1norm of the parameter vector) be ?s, where s is a regularization
parameter determined by cross-validation. During cross-validation,
LASSO used four-fifths of the data for selecting a model and one-fifth for
5-fold cross-validation was repeated 10 times to estimate the variance in
to decide when to stop adding variables to the model. The regularization
parameter—the LASSO penalty used to identify the optimal number of
explanatory features—was chosen as the smallest parameter whose mean
cross-validation error was less than or equal to the minimum cross-vali-
dation error plus 1 standard deviation of the cross-validation error at the
minimum. The dependent variable was the log10fold change in HIV sus-
ceptibility. Each of the regression coefficients represented an HIV-1
amino acid mutation. LASSO coefficient means that were more than 3
standard deviations above or below zero after 10 repeated runs of 5-fold
ceptibility to NRTIs.
To quantify the effect of the LASSO-selected mutations on NRTI sus-
ceptibility, we used least-squares regression (LSR). For this regression
analysis, we also used 5-fold cross validation and 10-fold repetition to
estimate the variance among the fitted coefficients. Seven LSR models—
one for each NRTI—were created. In these models, each of the selected
mutations was an explanatory variable and the log of the fold change in
susceptibility was the response variable. For each 5-fold cross-validation,
80% of data was used for learning regression coefficients and 20% was
zero in the 10 repeated runs of 5-fold cross-validation were considered
statistically significant predictors of susceptibility to NRTIs.
Regression analyses (for both the LASSO and LSR models) were stan-
dardized by scaling the log fold distributions for each of seven NRTIs to a
distribution with a standard deviation of 1. Standardizing the regression
tibility ranges among the NRTIs. Consequently, the regression coeffi-
cients reflect the standard deviation change in log fold associated with
each specific mutation (rather than the actual log fold difference).
Contribution of NRTI mutations to decreased susceptibility. Pre-
diction accuracy was evaluated using continuous and categorical ap-
proaches. The continuous approach involved calculating the mean
squared error (MSE) between the actual and predicted standardized log
fold change in susceptibility. The categorical approach involved deter-
mining how often the predicted phenotype correlated with one of three
predefined susceptibility categories: susceptibility, low/intermediate re-
sistance, and high-level resistance. The predefined susceptibility catego-
(24). They were chosen to approximate the geometric mean of the pub-
lished estimated clinical cutoffs provided with the PhenoSense reports.
For AZT, 3TC, and FTC, an isolate with ?3-fold-decreased susceptibility
was considered susceptible; an isolate with 3- to 25-fold-decreased sus-
ceptibility was considered to exhibit low/intermediate resistance; and an
isolate with ?25-fold-decreased susceptibility was considered highly re-
indicate susceptibility; 1.5- to 3.0-fold resistance was considered low/in-
termediate resistance; and ?3.0-fold resistance was considered a high
susceptibility; 2- to 6-fold resistance was considered low/intermediate re-
sistance; and ?6-fold resistance was considered a high level of resistance.
Mutational interactions. We used four approaches to investigate
whether models with interactions improved the prediction of in vitro
plored interactions among the mutations identified by LASSO (31). (ii)
Multivariate adaptive regression splines (MARS) progressively tune the
maximum allowed interaction constraint parameter mi from 1 to 3 (8).
(iii) We extended our LSR by including the stepwise addition of interac-
tified as significantly covarying in a previous study (23). (iv) We con-
ducted an exhaustive search of all potential two-way interactions among
the LASSO-identified mutations by constructing a variable interaction
matrix that included all possible two-way interactions in addition to each
specific regression model using this larger interaction matrix. Cross-vali-
dation was used in both stages to minimize overfitting.
Summary of NRTI susceptibility analysis results. Phenotypic
susceptibility results were available for 1,739 HIV-1 isolates from
1,478 individuals. These included 1,687 clinical isolates and 52
laboratory clones or site-directed mutants. To reduce bias result-
ing from individuals who had more than one virus tested, we
excluded from our analysis 228 viruses from individuals with
more than one virus having the same mutations at each of the
following NRTI resistance positions: 65, 74, 115, 151, 184, and
215. To reduce the confounding effect of virus populations con-
taining mixtures of two or more residues at the same position, we
excluded 256 isolates with electrophoretic mixtures at the same
Among the 1,273 isolates included in our analysis, more than
1,100 susceptibility results were available for 3TC, ABC, AZT,
d4T, and ddI, 952 for TDF, and 577 for FTC. Overall, 45% of
results met the predefined criteria for susceptibility; 28% met
those for low/intermediate resistance; and 26% met those for
each susceptibility category for each of the seven NRTIs. Of the
1,273 isolates, 98.2% belonged to subtype B. Isolates were ob-
tained between 1995 and 2011 (median year: 2003; interquartile
range, 2000 to 2007).
of NRTIs by showing the correlation of the standardized log fold
change in susceptibility for each pair of NRTIs. The two cytidine
second and third highest correlations were those between the
two thymidine analogs AZT and d4T (r ? 0.83) and between
AZT and TDF (r ? 0.83). Extremely low correlations were
present between the standardized log fold susceptibilities to
TDF and 3TC (0.02), TDF and FTC (0.04), AZT and 3TC
(0.11), and AZT and FTC (0.22).
NRTI resistance mutations and their effects on specific
Melikian et al.
aac.asm.orgAntimicrobial Agents and Chemotherapy
positions as significant predictors of decreased susceptibility to
K43E, K64H, K65R, D67N, T69ins, K70R, L74V, V75T, F77L,
R83K, A98G, K102Q, Y115F, V118I, I135T, Q151M, M184V/I,
E203D, H208Y, L210W, T215F/Y/D, D218E, and K219R. To
quantify the contribution of the LASSO-identified mutations to
each of the seven NRTIs. M184I, which was present in 16 patient
samples, was combined with M184V in our analysis. T69ins in-
cludes a variety of different double amino acid insertions at this
nificantly associated with reduced susceptibility to at least one
NRTI in the LSR model. The complete list of regression coeffi-
in Table S1 in the supplemental material.
The median number of LASSO-identified mutations per sam-
highly correlated with the prevalence of these mutations in se-
in the Stanford HIV Drug Resistance Database (Pearson’s r, 0.99;
P, ?0.001) (22) (Fig. 3).
LSR models) were those for K65R, T69ins, Y115F, Q151M, and
K70R, L74V, V75T, F77L, and T215F/Y (between 0.5 and 1.0).
K64H, which was present in only 16 and 13 isolates undergoing
d4T and TDF susceptibility testing, had standardized regression
coefficients for these two drugs of 0.63 (95% confidence interval
[95% CI], 0.629 to 0.631) and 1.17 (95% CI, 1.164 to 1.176),
M184V/I, and T215F/Y each had coefficients of ?0.5 for four
susceptibility to TDF and AZT; and K65R was associated with
increased susceptibility to AZT.
Four of the 28 mutations associated with decreased NRTI sus-
ceptibility were polymorphic in one or more group M subtypes,
including K43E, V118I, I135T, and E203D.
Least-squares regression prediction performance. Table 2
summarizes the categorical and continuous prediction perfor-
The categorical performance, or classification accuracy, was the
proportion of isolates for which the regression model correctly
the three predefined susceptibility categories: susceptible, exhib-
iting low/intermediate resistance, or highly resistant. The classifi-
cation accuracies ranged from 0.77 for ddI, 0.78 to 0.82 for ABC,
AZT, TDF, and d4T, and 0.92 to 0.94 for 3TC and FTC. The
predictions and actual results were completely discordant (i.e.,
susceptible versus highly resistant) for about 0.5% of tests (range,
0.26% for ABC to 0.96% for TDF) and partially discordant (i.e.,
intermediate versus susceptible or intermediate versus highly re-
sistant), on average, for 13% of tests (range, 5.3% for FTC to
22.5% for ddI) (see Table S2 in the supplemental material).
The standardized log fold MSE of 50 trials (5-fold cross-vali-
dation performed 10 times) per NRTI ranged from 0.08 (FTC) to
(TDF) (Table 2).
NRTI mutation interactions. None of the four approaches
that incorporated mutational interactions (that is, evaluation for
nonlinear effects, such as synergy or antagonism, in NRTI resis-
(DSA) partitioning algorithm, multivariate adaptive regression
splines (MARS), extension of LASSO to include subsets of previ-
ously identified covarying mutations, and extension of LASSO to
include all pairwise interactions—improved the accuracy of pre-
diction of reductions in NRTI susceptibility over that with their
respective noninteraction versions. Although several models
identified pairs of mutations (e.g., T69ins plus T215Y, F77L plus
Q151M, and K65R plus Q151M) that interacted synergistically to
reduce NRTI susceptibility, these isolated effects did not result in
an overall improvement in prediction accuracy and therefore did
not justify the use of a complex interaction model.
poration into the HIV-1 primer DNA strand and those that pro-
mote the excision of chain-terminating NRTIs via ATP-mediated
pyrophosphorolysis. K65R, K70E, L74V, F115Y, M184V/I, and
Q151M plus the Q151M-associated mutations (A62V, V75I,
F77L, and F116Y) inhibit NRTI incorporation; whereas M41L,
D67N, K70R, L210W, T215Y/F, K219Q/E, and the amino acid
T69ins promote NRTI excision. M41L, D67N, K70R, L210W,
T215Y/F, and K219Q/E are called thymidine analog mutations
(TAMs) because they are selected primarily by the thymidine an-
alogs AZT and d4T. The TAMs have been subclassified into two
overlapping clusters: type I (M41L, L210W, and T215Y) and type
II (D67N, K70R, T215F, and K219Q/E) TAMs. The mechanisms
of action of two additional mutations, T69D and V75T, which
were reported in the 1990s to reduce susceptibility to ddC and
d4T, respectively (6, 14, 29), have been less well characterized.
With the analysis of increasingly large databases, many addi-
TABLE 1 Numbers of HIV-1 isolates with genotype-phenotype
correlations for each of the seven NRTIs by predefined resistance
No. (%) of isolatesb:
Total no. of
3,460 (45.5) 2,155 (28.3)
aNRTI, nucleoside reverse transcriptase inhibitor; 3TC, lamivudine; ABC, abacavir;
AZT, zidovudine; d4T, stavudine; ddI, didanosine; TDF, tenofovir; FTC, emtricitabine.
HIV-1 Reverse Transcriptase Mutations
May 2012 Volume 56 Number 5aac.asm.org 2307
FIG 1 Phenotypic correlation matrix showing standardized HIV-1 log fold cross-resistance between each pair of the seven NRTIs. The Pearson correlation
coefficients (r) for each of the 21 NRTI pairs are shown. ***, P ? 0.3; in all other cases, P ? 0.0001. Drug abbreviations: 3TC, lamivudine; ABC, abacavir; AZT,
zidovudine; D4T, stavudine; DDI, didanosine; TDF, tenofovir; FTC, emtricitabine.
Melikian et al.
aac.asm.org Antimicrobial Agents and Chemotherapy
FIG 2 Regression coefficients of the RT mutations found to be significantly associated with decreased susceptibility to at least one NRTI in the least-squares
regression models. The mutations shown occurred at least 10 times in the data set. Positive coefficients represent mutations that decrease drug susceptibility;
in standard deviation units) for the log fold distribution of the respective NRTI. The error bars indicate the standard deviation of the mean generalized error,
deviations from zero are blue; other coefficient bars are gray, indicating a lack of statistical significance after cross-validation. Drug abbreviations: 3TC,
lamivudine; ABC, abacavir; AZT, zidovudine; D4T, stavudine; DDI, didanosine; TDF, tenofovir; FTC, emtricitabine.
HIV-1 Reverse Transcriptase Mutations
May 2012 Volume 56 Number 5aac.asm.org 2309