University of Southern California Dissertations and Theses

Pharmacogenetic association studies and the impact of population substructure in the women's interagency HIV study

(USC Thesis Other)

Pharmacogenetic association studies and the impact of population substructure in the women's interagency HIV study

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

PHARMACOGENETIC ASSOCIATION STUDIES AND
THE IMPACT OF POPULATION SUBSTRUCTURE
IN THE WOMEN’S INTERAGENCY HIV STUDY

by

Melissa Ann Frasco

A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(EPIDEMIOLOGY)

August 2012

Copyright 2012 Melissa Ann Frasco

ii
Table of Contents

List of Tables iv

List of Figures vii

Abstract viii

Chapter 1: Introduction 1

Chapter 2: Pharmacogenomics Background 9
Introduction 9
Genes Involved in HIV Therapeutics 12

Chapter 3: Review of Pharmacogenetic Studies of HAART 19
Introduction 19
Review of Study Designs 24
Results of Pharmacogenetic Studies 30
Pharmacogenetic Issues Specific to the WIHS 49

Chapter 4: Population Substructure in the WIHS 53
Abstract 53
Introduction 54
Methods 57
Results 61
Discussion 70
Chapter 4 References 74

Chapter 5: Genetic Variation in CYP2B6 Predicts Virologic 80
Response to NNRTIs
Abstract 80
Introduction 81
Methods 84
Results 89
Discussion 96
Chapter 5 References 99

Chapter 6: Validation of Diabetes Genetic Associations 105
Abstract 105
Introduction 106
Methods 107
Results 111
iii
Discussion 119
Chapter 6 References 122

Chapter 7: Summary and Future Directions 130

Comprehensive References 134

Appendix: Pharmacogenetic Associations in Antiretroviral Therapy 156
Introduction 156
Methods 158
Results 166

iv
List of Tables

Table 1.1 Specific drug names and abbreviations in three 3
Drug classes (NRTIs, NNRTIs, PIs) that are prescribed
in HAART regimens.

Table 2.1 Genes involved in non-nucleoside reverse transcriptase 14
inhibitor (NNRTI) and protease inhibitor (PI) metabolism
and absorption.

Table 3.1 CYP3A4 and CYP3A5 pharmacogenetic studies 31

Table 3.2 CYP2C19 pharmacogenetic studies 37

Table 3.3 CYP2B6 pharmacogenetic studies 41

Table 3.4 ABCB1 pharmacogenetic studies 44

Table 4.1 Description of 2,318 women in the Women’s 62
Interagency HIV Study (WIHS) with ancestry
informative marker data.

Table 4.2 Differences in allele frequencies across relevant 67
HapMap3 populations for 105 ancestry informative
markers.

Table 4.3 Average proportions of ancestry for self-identified 69
African American, Hispanic, and non-Hispanic White
women by WIHS site for k=4 on 105 loci (first row)
and k=4 on 168 loci (second row).

Table 5.1 Frequency counts for genotypes by virologic responder 92
status

and minor allele frequencies by self-reported race/
ethnic group in 91 HIV-infected women treated with
NNRTIs (efavirenz and nevirapine).

Table 5.2 Associations between CYP2B6 single nucleotide 94
polymorphisms (SNP) and with virologic response to
NNRTI-based regimens in women who were naïve to
antiretroviral drugs (N=91).

v

Table 5.3 Estimates for the association between the CYP2B6 95
metabolizer phenotype (using rs3745274 and rs28399499)
with virologic response to NNRTI-based HAART
regimens in women who were naïve to antiretroviral
drugs (N=91).

Table 6.1 Description of HIV-infected women in the Women’s 112
Interagency HIV Study (WIHS) by type 2 diabetes
status from 2000-2011.

Table 6.2 Associations between single nucleotide polymorphisms 114
(SNP) and type 2 diabetes in Women’s Interagency
HIV Study (WIHS) self-reported White, Hispanic, Asian,
and other women (n=517) and published results from the
European genome wide association study (GWAS) meta-
analysis and the Swiss HIV Cohort Study (SHC) with p
values for heterogeneity (p-het).

Table 6.3 Associations between single nucleotide polymorphisms 115
(SNP) and type 2 diabetes in Women’s Interagency HIV
Study (WIHS) self-reported African-American (AA)
women (n=773) and a meta-analysis of published results
from five African-American (HIV negative) replication
studies with p values for heterogeneity (p-het).

Table 6.4 Associations between type 2 diabetes and single 116
nucleotide polymorphisms (SNP) by number of NRTIs at
visit after adjustment for genetic ancestry covariates in
White, Hispanic, Asian, and other HIV+ women (n=517).

Table 6.5 Associations between incident type 2 diabetes and 117
single nucleotide polymorphisms (SNP) by number of
NRTIs after adjustment for genetic ancestry covariates
in African American HIV+ women (n=773).

Table 6.6 Associations between type 2 diabetes and single 118
nucleotide polymorphisms (SNP) by cumulative number
of nucleoside reverse transcriptase inhibitors (NRTI)-years
(dose and duration) after adjustment for genetic ancestry
covariates in White, Hispanic, Asian, and other HIV+
women (n=517).

vi

Table 6.7 Associations between type 2 diabetes and single 119
nucleotide polymorphisms (SNP) by cumulative
number of nucleoside reverse transcriptase inhibitors
(NRTI)-years (dose and duration) after adjustment for
genetic ancestry covariates in African-American HIV+
women (n=773).

Table A1 Nominally significant associations between genotypes 165
in NR1I2 and response to HAART in all HAART
initiators (N=424) and in the subset of antiretroviral
naïve HAART initiators (N=204).

Table A2 Association between genetic variants (log-additive) 166
and response to HAART in all HAART initiators (N=424)
and in the subset of antiretroviral naïve HAART initiators
(N=204) by type of HAART regimen.

vii
List of Figures

Figure 4.1 Individual ancestry proportions for self-reported WIHS 64
ethnic groups and HapMap populations.

Figure 4.2 Individual proportions of Yoruban ancestry (k=3) vs 65
individual proportions of Yoruban plus non-Yoruban
ancestry (k=4) for self-identified WIHS African American
women (n=1,416).

Figure 5.1 Principal component (PC) 1 vs. PC 2 for WIHS women 91
(n=2,318) from analysis on 168 ancestry informative
markers. Self-identified race/ethnicity groups are color-
coded.

viii
Abstract

Population stratification can result in spurious associations in genetic studies when
the outcome and genotype are associated with genetic ancestry. The objective of this
dissertation was to characterize population substructure in the Women’s Interagency HIV
Study (WIHS), a cohort of HIV-infected and uninfected participants in the United States
(U.S), and assess the impact of population substructure on genetic association studies.
First, population substructure was characterized using genotype data on 168 ancestry
informative markers (AIMs) by performing a Bayesian clustering algorithm to infer
genetic membership in three and four assumed source populations. Principal components
(PC) analysis was conducted to generate continuous genetic ancestry covariates to use as
covariates in genetic association statistical models. Population substructure was
identified within self-identified ethnic groups and across geographical regions in the U.S.,
exemplifying the importance of estimating individual genetic ancestry to control for
residual ethnic confounding in genetic association studies. Cautious interpretation of
ancestry admixture estimates should be exercised, as the labeling of source populations is
sensitive to the panel of markers and referent populations used in analyses.
The effect of variation in CYP2B6 on virologic response to its substrates, the
nonnucleoside reverse transcriptase inhibitors (NNRTIs), was explored with
consideration for self-reported race/ethnicity and underlying genetic structure. Logistic
regression was used to test the joint effect of two single nucleotide polymorphisms (SNP)
rs3745274 and rs28399499, which comprise the CYP2B6 metabolizer phenotype
ix
indicating slow, intermediate and extensive metabolizers. Substantial evidence of
confounding was present with the metabolizer phenotype when comparing the crude,
self-reported ethnicity-adjusted and genetic ancestry PCs-adjusted estimates. Women
classified as intermediate and slow metabolizers were 2.90 (95% CI 0.79-12.28) and
13.44 (95% CI 1.66-infinity) times as likely to achieve virologic suppression compared to
extensive metabolizers after adjustment for PCs (p trend =0.005). The CYP2B6
metabolizer phenotype was significantly associated with response to NNRTIs, a relation
that would have been masked by simply adjusting for self-reported race/ethnicity (p for
trend=0.04 after adjustment for self-reported race/ethnicity).
Type 2 diabetes (T2D) in the setting of HIV infection is a concern given the
prevalence of HIV-related conditions that contribute to the etiology T2D. The incidence
of T2D varies substantially between racial/ethnic groups and thus control for genetic
substructure is imperative in genetic association studies of T2D risk. Eighteen previously
confirmed T2D-associated SNPs were tested using Cox proportional hazard models with
adjustment for genetic ancestry principal components using age for the time-scale.
Exposure to nucleoside reverse transcriptase inhibitors (NRTI), a T2D risk factor, was
explored as an effect modifier. Overall, the T2D risk conferred by these SNPs was similar
in White/Hispanic HIV-infected women compared to HIV-uninfected European
individuals, as evaluated by the effect measures and p values for heterogeneity. The
magnitude of each SNP effect was smaller in African-American women (HRs~1.1).
Significant interactions were revealed between these T2D-associated SNPs and NRTI
exposure (p<0.03) in White/Hispanic women.
x
This dissertation confirmed prior report of the role of CYP2B6 SNPs in NNRTI
metabolism and efficacy. Additionally, the role of T2D-associated SNPs in the context
of HIV was validated and the synergistic effect of T2D-associated SNPs and NRTIs on
the risk of T2D was suggested. Substantial evidence of residual confounding was
revealed in both studies, which highlights the importance of controlling for population
substructure using genetic ancestry estimates in genetic association studies.

1
Chapter 1: Introduction

This dissertation will address three areas of genetic research using data from the
Women’s Interagency HIV Study (WIHS). First, the genetic ancestry of the WIHS
cohort will be characterized using a panel of ancestry-informative genetic markers. Use
of these genetic markers to construct estimates of each participant’s genetic ancestry is
crucial for control of population stratification in the subsequent genetic association
studies. Second, the association between genetic variants involved in drug biosynthesis
and metabolism and response to highly active anti-retroviral therapy (HAART) will be
evaluated. Genetic influences on virologic response to HAART may explain some of the
variance in patients’ viral load levels after initiation of treatment. Lastly, known
diabetes-associated genetic variants will be assessed in WIHS HIV-infected women to
determine if similar genetic associations are observed in HIV-related diabetes. Validating
these diabetes risk alleles in HIV-infected individuals is especially important in assessing
whether the documented excess diabetes risk related to particular HAART regimens can
be attributed to genetics, treatment, or the joint effects of the two factors.
WIHS Study Design. The WIHS is a prospective multi-ethnic cohort study of
HIV-infected and HIV-uninfected women (Barkan 1998). Enrollment into the WIHS
was completed in two phases with the first from October 1994 to November 1995 (phase
I) and the second from October 2001 to September 2002 (phase II). A total of 2,058 HIV
seropositive and 568 HIV seronegative women were recruited in phase I and 737
seropositive women and 406 seronegative women in phase II. Enrollment for both phase

2
I and II was conducted at 6 sites across the United States: Bronx NY, Brooklyn NY,
Washington DC, Los Angeles, Chicago, and the San Francisco Bay area. WIHS
recruitment focused on African-American and Latina women in order to obtain a sample
similar to the U.S. population of HIV-infected women (Barkan 1998).
HIV-infected women were recruited from HIV primary care clinics, hospital-
based programs, community outreach sites, women’s support groups, drug rehabilitation
programs, HIV testing sites, and referrals from enrolled participants (Barkan 1998).
Eligible participants were at least 13 years of age, gave written informed consent, were
tested for HIV, completed the interview in English or Spanish, were able to travel to the
regional WIHS clinic site and able to have blood drawn for laboratory testing (Barkan
1998).
The HIV-uninfected women were recruited from the same sources and enrollment
was targeted among women who engage in high-risk behaviors for HIV infection. High-
risk behavior was defined as reporting one or more of the following characteristics within
the past year: injection drug use, having a sexually transmitted disease, having
unprotected sex with three or more men or protected sex with five or more men, or
having exchanged sex for drugs, money, or shelter (Barkan 1998). HIV-negative women
were comparable to infected participants in age (median age HIV+ 36, HIV- 34),
race/ethnicity (56% African American HIV+, 54% African American HIV-), level of
education, injection drug use since 1978, and total number of sexual partners since 1980
(Barkan 1998).

3
History of clinical treatment for HIV and baseline treatment regimens in the
WIHS. HIV treatment has evolved dramatically since 1987, when the first drug was
approved to treat HIV (FDA website). Highly active antiretroviral therapy (HAART),
which became available in 1996, is currently the gold standard in treatment for HIV;
HAART is prescribed as a multi-drug regimen containing at least three different drugs
from two different drug classes (Dybul 2002). The three drug classes that make up the
HAART regimen are shown below in Table 1.1.

Table 1.1 Specific drug names and abbreviations in three drug classes (NRTIs,
NNRTIs, PIs) that are prescribed in HAART regimens.

Nucleoside reverse transcriptase inhibitors (NRTI)
Zidovudine AZT
Zalcitabine DDC
Didnosine DDI
Stavudine D4T
Lamivudine 3TC
Emtricitabine FTC
Abacavir ABC
Tenofovir TDF
Non-nucleoside reverse transcriptase inhibitors (NNRTI)
Nevirapine NVP
Delavirdine DLV
Efavirenz EFV
Protease inhibitors (PI)
Saquinavir SQV
Ritonavir RTV
Indinavir IDV
Nelfinavir NFV
Lopinavir LPV
Atazanavir ATZ
Fosamprenavir FPV

The nucleoside reverse transcriptase inhibitor (NRTI), zidovudine, was approved
in 1987 (FDA website). After 1991, several other NRTIs were approved. At this time,

4
the initial treatment of HIV consisted of one or more NRTIs, and patients failing to
respond to an existing NRTI were prescribed different or additional NRTIs (Dybul 2002).
Treatment effectiveness was highly variable, where virologic failure (increased HIV-1
replication) was often the result of acquiring a mutated, treatment-resistant HIV-1 virus
(Dybul 2002).
Additional drug classes were approved for antiretroviral therapy in the mid
1990’s, including the protease inhibitor (PI) saquinavir in 1996, and a non-nucleoside
analog reverse transcriptase inhibitor (NNRTI), delavirdine in 1997 (FDA website). The
licensing of additional drug classes in the mid 1990’s initiated the HAART era, in which
HIV-infected individuals were widely prescribed a HAART-defining regimen.
In WIHS, phase I (1994-1995) recruitment occurred prior to the advent of
HAART. Thus, on entry into the early cohort, WIHS participants were typically treated
with one drug class or they were naïve to antiretroviral treatment (Barkan 1998). In order
to address new research questions pertaining to HAART, women with varied treatment
regimen exposure were recruited in the phase II (2001-2002) cohort (Bacon 2005). Of
the women recruited in the phase II cohort, 381 (51.7%) reported HAART use at
baseline. The proportion of women who were not taking antiretroviral medications of
any type at baseline was 33.8% and 43.2% in phase I and phase II cohorts, respectively.
The cross-sectional prevalence of HAART-treated women in either cohort was 95% in
2009.
WIHS follow-up. Participants have been followed every six months since
entering the cohort. At each biannual visit, detailed information is collected by a trained

5
medical interviewer on the participants’ current treatment regimens. The women answer
yes or no to a series of questions regarding use of specific antiretroviral drugs and are
given pictures of the medication bottles to prompt their memory. In addition, they are
questioned on the antiretroviral medications they have taken in the six months prior to the
current follow-up visit. Comprehensive information on all medication use for other
ailments is also obtained and recorded at each follow-up visit. Additional data regarding
sexual behaviors, drug, alcohol, and cigarette use, medical and OB/GYN history, health
services utilization, psychological status, and demographics is obtained at each follow-up
visit. (Barkan 1998, Bacon 2005)
Physical examinations are performed at each follow-up visit; specifically, skin
and oral examinations, anthropometric measures, gynecological and lipodystrophy
examinations are conducted. Participants provide biologic samples (i.e. blood, saliva,
hair, gynecological, and urine specimens) at every six-month visit. Comprehensive
laboratory tests are completed on the biospecimens, including plasma HIV-1 RNA levels,
CD4+ and CD8+ cell counts, basic chemistry, complete blood counts, HPV detection on
cervical specimens, and pregnancy tests. Additional samples are stored at a central
repository. (Barkan 1998, Bacon 2005)
Clinical outcomes are obtained through surveillance of varied sources. WIHS
participants are followed for AIDS-related diagnoses by self-report followed by medical
record abstractions. Subject personal identifiers are also routinely linked to local and
state tuberculosis registries, state cancer registries (every two years), and the National
Death Index (annually). (Barkan 1998, Bacon 2005)

6
Additional WIHS data collection. Over time additional variables have been
collected based on the emergence of new areas of research. For example, data related to
adherence to anti-HIV medications has been collected since 1998 (visit 9) when it
became clear that these data would be critical for evaluating issues related to efficacy of
treatment regimens. Along similar lines, the WIHS investigators have kept current with
changing technology and have incorporated more sensitive HIV-1 RNA assays when they
become available (Bacon 2005). At the initiation of WIHS, the lower limit of detection
of HIV RNA was 4000 copies/ml; currently, the detection limit is 50 copies/ml.
Overall, the WIHS is a rich dataset with more than 15 years of follow-up and
includes a wide range of clinical and laboratory outcomes to investigate. The WIHS
cohort encompasses many subjects with varied treatment histories and HIV disease
phenotypes. In addition, the observational study design closely reflects the real-world
experience of HIV disease and treatment allowing research questions to be addressed in a
population that can be generalizable to many HIV-infected women in the U.S.
Mortality in the HAART era. Morbidity and mortality has declined markedly
since the advent and widespread use of HAART to treat patients with HIV (Palella 1998,
Ledergerber 1999, Gange 2001, Palella 2006). The rates of AIDS diagnoses and
mortality in WIHS women declined at the nascence of the HAART era from April 1996-
March 1999 (Gange 2001). Among women who did not have clinical AIDS at baseline,
the estimated decline in the incidence of AIDS was 23% (95% CI –29% to –16%) per
each six-month interval of follow-up (Gange 2001). Additionally, the mortality rate of
women who reported clinical AIDS at baseline decreased 21% (95% CI –27% to –14%)

7
per six-month interval over the same time period (Gange 2001). The AIDS-related
mortality rate has plateaued during 2005-2007; there were an estimated 1.5 AIDS-related
deaths per 100 person-years in 2007. This rate was comparable to the all-cause mortality
rate in HIV-uninfected women in the WIHS.
While HIV-infected individuals are living longer in the HAART era, comorbidity
rates have increased among the HIV-infected population as a function of longer survival
with HAART treatment (Palella 2006). In a prospective study evaluating mortality rates
of HIV-infected individuals from 1996-2004, the proportion of deaths due to non-AIDS
defining illnesses increased from 13.1% in 1996 to 42.5% in 2004 (p<0.001) (Palella
2006). The proportion of deaths attributable to non-AIDS diseases primarily included
hepatic, cardiovascular, and pulmonary diseases, as well as non-AIDS malignancies
(Palella 2006). This trend is also prevalent in the WIHS, where non-AIDS related deaths
accounted for nearly 40% of all deaths in HIV-infected women from 2003-2004,
compared to 10% of total deaths in 1996 (pre-HAART era, French 2009). The
interpretation of these data is that co-morbid conditions are increasing in the HAART-
era.
Part I of this dissertation includes: a) a background on pharmacogenomics and
genes involved in the metabolism of HAART drugs, b) a critical review of the literature
on the association of polymorphisms with response to HIV therapy and discussion on
pharmacogenomic-specific issues using the WIHS dataset.
Part II of this dissertation encompasses the three manuscripts. The first
manuscript characterized the population genetics in the WIHS using a panel of ancestry

8
informative markers (AIMs). This critical descriptive effort provided the basis for future
genetic association studies conducted in the WIHS by generating variables to properly
adjust for ancestry (versus self-reported race/ethnicity). The second manuscript focused
on the association between CYP2B6 genetic variation and virologic response to NNRTIs.
Additional analyses for the remaining genetic variants of interest are provided in the
Appendix. The third manuscript explored whether the European-derived genetic variants
that have been confirmed as diabetes susceptibility alleles in HIV-negative populations
confer risk of diabetes among HIV-infected individuals.

9
Chapter 2: Pharmacogenetics Background

INTRODUCTION

Pharmacology. Orally administered drugs travel through the digestive system
and are absorbed in the small intestine to reach systemic circulation. A drug becomes
inert by conversion to a metabolite or becomes activated through metabolism in the
intestine and liver where it can be absorbed or excreted, respectively (Piscitelli 2001).
First-pass drug absorption occurs in the intestine and is regulated by P-glycoprotein,
which actively transports drugs out of the intestinal epithelial cell, thereby regulating the
bioavailability and distribution of drugs to cellular targets via the bloodstream (Marzolini
2004). Excretion of metabolized drugs occurs through the renal or gastrointestinal
systems (Piscitelli 2001).
Drug response, defined by plasma drug levels or clinical efficacy, can vary
according to age, sex, drug interactions, and disease severity (Evans and McLeod 2003,
Cressey 2007). Genetic factors also influence the efficacy of a drug as well as the
likelihood of a drug-related adverse reaction (Evans and McLeod 2003, Cressey 2007).
The study of variability in response to xenobiotics attributed to hereditary factors is
termed pharmacogenetics.
Pharmacogenetics. The emergence of the pharmacogenetics field dates to the
late 1970’s, spawning from clinical observations of varied plasma or urinary drug
concentrations in patients prescribed the same drug dose (Mahgoub 1977, Eichelbaum

10
1979). The inter-individual differences in drug clearance and toxicity were thought to be
inherited metabolic phenotypes (Eichelbaum 1979). For example, family studies
demonstrated that drug toxicity, classified by high plasma drug levels of debrisoquine, a
drug used to control hypertension, and sparteine, an antiarrhymic agent, occurred in
patients who had inherited two copies of a gene that encoded an enzyme with decreased
or deficient activity (Mahoub 1977, Eichelbaum 1979). It was subsequently established
that phenotypes characterized by poor metabolism of debrisoquine and sparteine co-
segregated in Whites, suggesting interethnic variability in drug metabolism (Evans 1980).
As molecular techniques became more sophisticated, in vitro studies mapped the null
allele attributed to poor metabolism to a cytochrome p450 enzyme, CYP2D6, on
chromosome 22 (Eichelbaum 1987, Gonzalez 1988).
Since these early studies, pharmacogenetic research has expanded to include
genes responsible for the metabolism of several other drugs, as well as genes responsible
for each of the many processes a drug undergoes after it is administered. Polymorphisms
in genes involved in drug absorption, distribution and excretion may also influence drug
response phenotypes. The ultimate objective of pharmacogenetic research is to identify
genetic variation in specific pharmacologic pathways that may be utilized to individualize
dosage requirements and optimize clinical response.
Example of a pharmacogenetic association. The cytochrome p450 family of
enzymes are accountable for the oxidative metabolism of drugs, environmental toxins,
and other endogenous substrates, such as steroids (Piscitelli 2001). CYP2C9, a p450
family member, is mainly expressed in the liver and constitutes about 20% of all hepatic

11
enzymes (Ingelman-Sundenberg 2005). Two common CYP2C9 single-base substitutions
that result in amino acid changes and reduced enzyme activity have the most clinical
relevance (rs1799853 R144C and rs1057910 I359L) (King 2004). These variant alleles
are primarily present in Whites at a frequency of 20% and 12%, whereas only 5% of
Africans, Chinese, and Japanese are carriers of either variant (Lee 2002).
Numerous studies have reported associations between CYP2C9 variants and
reduced dosage requirements of the anticoagulant, warfarin (Furuya 1995, Steward 1997,
Aithas 1999, Scordo 2002, King 2004). Carriers of rs1799853 and rs1057910 variants
experience serious bleeding from standard doses of warfarin, due to reduced clearance of
the active drug (Ogg 1990, Aithas 1999, Higashi 2002). In a 2005 meta-analysis of eight
warfarin pharmacogenetic studies among 2775 subjects, carriers of at least one copy of
the rs1799853 variant allele required 0.85 mg less of warfarin daily (95% CI –1.11 to –
0.60 mg), and those who carried at least one copy of the rs1057910 variant allele required
1.92 mg less of warfarin daily (95% CI –2.47 to –1.37 mg) compared to wildtype to
achieve adequate plasma drug levels (Sanderson 2005).
The vitamin K epoxide reductase (VKOR) gene, which encodes the drug target
for warfarin, further contributes to variability in warfarin sensitivity (Rieder 2005, Yuan
2005, Lee 2006). VKORC1 and CYP2C9 genotypes explain over 50% of the variability
in warfarin dosing requirements, after accounting for age and body size (Rieder 2005,
Sconce 2005, Takahashi 2006).

12
GENES INVOLVED IN HIV THERAPEUTICS

Specific drugs in HIV therapy. Highly active antiretroviral therapy (HAART)
regimens are defined as the combination of at least three different drugs from at least two
different drug classes (i.e., nucleoside reverse transcriptase inhibitors (NRTI), non-
nucleoside reverse transcriptase inhibitors (NNRTI), protease inhibitors (PI)). The
typical HAART regimen contains an NRTI-backbone of two NRTIs, and either a PI or a
NNRTI (Dybul 2002). A newer drug class, entry inhibitors, has recently been
implemented in HAART regimens. Table 1.1 in the previous chapter lists the specific
drugs in NRTI, NNRTI, and PI drug classes that WIHS women have reported as part of
an initial HAART regimen from 1997 to 2009 (entry inhibitors were not reported by
women who initiated HAART during WIHS follow-up in this period and are therefore
not discussed further).
HAART Drug Metabolism. Nucleoside reverse transcriptase inhibitors (NRTIs)
do not undergo hepatic metabolism through the CYP450 system. Instead, NRTIs are
metabolized by hepatic 5’-glucuronidation and are intracellularly phosphorylated to their
active triphosphate form (Dooley 2008).
Non-nucleoside reverse transcriptase inhibitors (NNRTIs) and protease inhibitors
(PIs) are extensively metabolized by the CYP450 system (Piscitelli 2001, Dooley 2008).
Drugs are classified as substrates, inducers, or inhibitors of cytochromes. Substrates are
drugs that are metabolized by the particular enzyme. Drugs that induce cyotchromes
increase the transcription of a CYP gene, thereby leading to increased rates of substrate

13
metabolism and reduced plasma concentration of the substrate drug (Dooley 2008).
Drugs that inhibit P450 cytochromes compete with substrate drugs or inactivate the
enzyme, leading to decreased clearance and increased plasma concentration of the
substrate drug (Dooley 2008). A substrate of an enzyme that is also an inducer
accelerates its own metabolism affecting plasma concentration and bioavailability.
Multiple drugs in the HAART regimen are metabolized by the same CYP
isoenzyme (Piscitelli 2001). HIV-infected individuals are often prescribed concomitant
drugs for other ailments or infections that also undergo CYP family metabolism. The
competing and promoting effects of various NNRTIs, PIs, and concomitant drugs that
inhibit or induce the same CYP isoforms alter plasma drug concentrations and may result
in antiretroviral-related toxicity or reduced antivirologic response (Dooley 2008). The
results of experimental studies that have measured the individual plasma drug
concentrations of specific antiretroviral drug combinations have been implemented in the
clinical dosing requirements.
This dissertation will evaluate genetic influences on HAART virologic and
immunologic response in the following genes: CYP3A4, CYP3A5, NR1I2 (ligand
activated nuclear receptor), ABCB1 (P-glycoprotein), CYP2C9, CYP2C19, and
CYP2B6. Table 2.1 shows the relationship of various antiretroviral drugs with CYP
enzymes and the efflux transporter, P-glycoprotein (ABCB1).

14
Table 2.1 Genes involved in non-nucleoside reverse transcriptase inhibitor (NNRTI) and
protease inhibitor (PI) metabolism and absorption.

CYP3A Family. The CYP3A locus at 7q22.1 includes all genes in the subfamily
3A, in which four isoforms have been identified: CYP3A4, CYP3A5, CYP3A7, and
CYP3A43 (Lamba 2002). Two main isoforms, CYP3A4 and CYP3A5, are abundant in
adult liver and intestine, comprising nearly 50% of all CYP hepatic enzymes (Williams
2002). CYP3A5 is differentially expressed in the liver, in which individuals express high
or low levels of CYP3A5 (Lamba 2002). The expression of CYP3A5 also varies among
ethnicities, where 10-40% of Whites, 33% of Japanese and 55% of African-Americans
show considerable levels of CYP3A5 (Tateishi 1999, Kuehl 2001). CYP3A7 expression
is rarely found in adults; it is predominantly expressed in the fetal liver (Schuetz 1994).
The highest expression of CYP3A43 is found in the prostate and is rarely expressed in
the liver or intestines (Gellner 2001).
The CYP3A complex is generally regulated by two pathways -- at the
transcriptional level via a ligand-activated nuclear receptor called NR1I2 and via
inflammatory signaling pathways (Burk 2004 and Jover 2006). Briefly, CYP3A is
induced by structurally diverse drugs that bind to the nuclear membrane receptor NR1I2,
which dimerizes with another nuclear receptor, RXR, and binds to a response element in
the promoter region located 8kb upstream of the CYP3A start site (Burk 2004). A very
CYP3A4 ABCB1 CYP2C9 CYP2C19 CYP2B6
NNRTI: efavirenz,
nevirapine

PI: all protease
inhibitors
PI: all protease
inhibitors
PI: ritonavir,
nelfinavir,
atazanavir
PI: nelfinavir,
ritonavir
NNRTI:
efavirenz,
nevirapine,
ritonavir

15
important transcriptional difference exists between genders, in that women express
approximately 1.5- to 2-fold more CYP3A4 than men (Wolbold 2003), leading to greater
clearance of CYP3A4 substrates in women due to higher enzyme activity. The
mechanism leading to this gender difference is not clear, but probably involves gender-
specific hormone receptors induced transcription (Waxman 2006).
CYP3A4 and CYP3A5 substrates overlap significantly, however, experiments
have shown that CYP3A5 is inferior in catalytic activity to CYP3A4-mediated
metabolism (Williams 2002). All NNRTIs and PIs are substrates of CYP3A4 (Table
2.1); NNRTIs are generally inducers, and most PIs are inhibitors of CYP3A4. Inhibition
of CYP3A by ritonavir or nelfinavir (PIs) offsets the inductive effects of efavirenz or
nevirapine (NNRTIs) when administered concomitantly (Fellay 2005). This exemplifies
a drug-drug interaction, where CYP3A-mediated drug metabolism is altered by a CYP3A
inhibitor. As a result, the co-administration of efavirenz (NNRTI) and nevirapine (PI) is
generally avoided due to increased rates of adverse events (Van Leth 2004).
ATP-binding Cassette Family. The multidrug resistance sub-family contains
two genes, ABCB1 and MDR3, that are important transmembrane efflux pumps, which
function to transport various molecules from the intracellular to the extracellular domain
(Marzolini 2004). The product of the ABCB1 gene, p-glycoprotein (P-gp), is a drug
efflux transporter, whereas MDR3 encodes a phospholipid translocator protein. P-gp
functions as a protective barrier to keep toxins out of critical cells, such as the liver, small
intestine, kidney, capillaries in the brain-blood barrier, testes, ovaries, macrophages,
peripheral blood mononuclear cells, macrophages, natural killer cells, and T and B

16
lymphocytes (Thiebaut 1987). P-gp regulates drug absorption and bioavailability by
transporting drug compounds into the bile, urine, or the intestinal lumen for elimination
(Marzolini 2004).
P-gp has been implicated in the cytotoxicity process and with the induction of
immune responses in T cells during HIV infection (Marzolini 2004). All PIs are
substrates of P-gp with affinities in the order ritonavir > nelfinavir > indinavir >
saquinavir. Since ritonavir also strongly inhibits the transport of weaker P-gp substrates,
it is often utilized to boost intracellular concentrations of co-administered PIs (Marzolini
2004).
Most compounds that are substrates for P-gp are also substrates for CYP3A4.
Because of the similarity in substrate specificity for both CYP3A4 and P-gp, these
proteins appear to act synergistically in drug metabolism and excretion (Chiou 2000).
Both P-gp and CYP3A are located on chromosome 7 (7q21.1 and 7q.22.1, respectively)
and regulated by ligand-activated NR1I2 in the liver and small intestine tissues
(Marzolini 2004). P-gp is known to influence the intracellular concentration of NR1I2
ligands, thereby controlling the induction of itself and the bioavailability of drugs that
regulate CYP3A expression (Marzolini 2004, Lamba 2006). Consistent with this model,
Lamba et al. found a significant interaction between NR1I2 and ABCB1 genetic
polymorphisms in affecting hepatic CYP3A4 expression (Lamba 2006).
CYP2C Family. There are four genes at the CYP2C locus (10q24), of which
three are expressed in the liver (CYP2C8, CYP2C9, CYP2C19). The CYP2C family
accounts for 20% of the total p450 cytochrome expression in the liver; CYP2C9 and

17
CYP2C19 have higher levels of hepatic basal expression over CYP2C8 (Ingelman-
Sundberg 2007). The CYP2C family is inducible by NR1I2 ligands, such as the
prototypical inducer rifampicin, with varying fold increases according to the particular
substrate (Raucy 2002, Chen 2004). The induction of CYP2C family transcription with
the specific NR1I2-ligands, efavirenz (NNRTI), ritonavir (PI), and nelfinavir (PI), has not
been investigated. Given that these antiretroviral drugs are NR1I2 activators, CYP2C
enzymes may be induced through NR1I2 regulation (Raucy 2002).
The CYP2C19 isozyme is particularly important in HAART regimens since it
contributes to the secondary metabolism of ritonavir and nelfinavir (PIs). The main
metabolite of nelfinavir, M8, is exclusively metabolized by CYP2C19 before it is
eliminated (Lillibridge 1998). Additional CYP2C9 and CYP2C19 substrates are shown
in Table 2.1.
CYP2B6. Another CYP2 family member, CYP2B6, is expressed in hepatocytes
to metabolize various endogenous and xenobiotic substrates. Although CYP2B6
represents the smallest percentage of total hepatic P450 enzyme content (2-10%), its
implication in drug metabolism is substantial (Lang 2001, Wang 2008). Extrahepatic
expression of CYP2B6 includes specific regions in the brain (Milksys 2003) along with
several other tissues (Ding 2003).
There are a growing number of CYP2B6 substrates, totaling about 8% of clinical
drugs; 30% of known CYP3A4 substrates are common to CYP2B6 (Xie 2001).
Efavirenz (NNRTI) undergoes primary and secondary metabolism by CYP2B6 (Ward
2003), but it is also a substrate of CYP3A4 (Table 2.1). This reflects the apparent

18
substrate selectivity among the CYP enzymes. The majority of CYP3A4 inducers also
activate CYP2B6 expression, suggesting a common regulation mechanism through
NR1I2 (Faucette 2006).
Additional genes involved in the metabolism and excretion of antiretroviral drugs
include the nuclear receptors, RXR and CAR, and kinases that act to transform nucleotide
reverse transcriptase inhibitors into their metabolites. This dissertation research does not
encompass the aforementioned genes due to financial restrictions.

19
Chapter 3: Review of Pharmacogenetic Studies of HAART

INTRODUCTION

The chapter discusses the potential biases in pharamacogenetic studies of HAART
treatment efficacy and critically evaluates studies in the HIV literature that have tested
for associations between genetic variation and antiretroviral drug virologic and
immunologic response. The chapter concludes with pharmacogenetic issues I have
addressed in designing a study to evaluate the influence of single nucleotide
polymorphisms (SNPs) in drug metabolism (CYP), regulation (NR1I2), and
transport/excretion (ABCB1) genes.
Issues Related to Genetic Association Studies. False positives in genetic
association studies may arise for several reasons; including failure to account for multiple
comparisons, genotyping error, and population stratification. There are an estimated 21
million single nucleotide polymorphisms (SNPs), which can be tested for associations
(www.1000genomes.org). In addition, each SNP can be modeled in many ways,
assuming dominant, additive, or recessive inheritance models. Failure to account for the
number of hypotheses tested leads to false positive associations at the declared alpha
level. A number of methods to account for multiple comparisons have been developed,
such as permutation testing and false discovery rates (Benjamini 1995, Doerge 1996).
Generally, a p value less than 1.0 x 10
-8
is considered as genome-wide significance.

20
Systematic genotyping error may produce false positive or negative associations if
case and control samples are not genotyped in the same batches. To prevent the
preferential assignment of an erroneous genotype to either cases or controls, all samples
should be genotyped blinded to case-control status.
As a positive control for absence of genotyping error, the genotype frequencies
should not deviate significantly from Hardy-Weinberg equilibrium (HWE). Generally,
the power is poor to detect deviations from HWE in SNP data with minor allele
frequencies <10% (Cox 2006), and as such, SNPs with HWE deviations p<0.20 should
be investigated for assay failure and poor duplicate concordance. If the assay depicts
distinct clusters, the assay should not be eliminated from the dataset. Additionally,
standard Hapmap mother-father-child trios should be genotyped within batches.
Erroneous assays can be deduced from Mendelian errors among trio data and from cross-
referencing published genotypes in the Hapmap database.
Population stratification, which refers to confounding by ethnicity in genetic
association studies, is the primary source of systematic bias in such studies that use
unrelated subjects (Pritchard 2000a). Population stratification is present when the
sample consists of several subpopulations, or if samples are those from the recent
admixture of different ethnic groups, and these subpopulations are differentially
represented among cases and controls. Ethnic groups vary in marker allele frequencies
and disease frequencies, and if the differences in ethnic representations among case and
control groups lead to case-control differences in allele frequencies. Spurious

21
associations can result from systematic ancestry differences rather than a real genetic
association with disease.
Many genetic association studies have attempted to adjust for population
stratification by adjusting for self-reported ethnicity or matching on country of origin.
Adjusting for self-reported ethnicity would eliminate confounding if all subjects within a
self-reported ethnic category had similar genetic ancestry. However, certain culturally-
defined ethnic categories (e.g., Caucasian, Hispanic, African American, and Asian) do
not completely and uniquely reflect the heterogeneity of the underlying genetic
population structure (Pritchard 2000a). As a result, residual confounding on ethnic
background may exist in genetic association studies even after adjusting for self-reported
ethnicity.
Determining population substructure in an ethnically diverse population is
required to adequately control for underlying genetic ancestry and reduce the probability
of false positive genetic associations due to population stratification. Ancestry
informative genetic markers have been identified for several continental populations and
can be genotyped and evaluated in a study population to determine population
substructure with clustering algorithms. The resulting summary ancestry component
variables are then included as covariates in genetic association models to more
completely control for population stratification. Self-reported ethnicity may also capture
additional covariates (e.g., SES and other lifestyle factors) in addition to ethnic ancestry
and therefore may be included in association models as well.

22
False negative genetic associations are primarily attributed to the lack of power to
detect small genetic associations (RR~1.1). Many polymorphisms are not present in the
population at sufficiently high frequencies to allow detection of genetic associations in
moderate sample sizes (n=100-1,000); therefore very large studies are needed to detect
small to modest genetic effects.
Early genetic association studies focused on testing polymorphisms located in
gene promoter and coding regions for associations with clinical phenotypes. An
observed genetic effect represents an association not only with the specific variant tested,
but also with all other polymorphisms that are in linkage disequilibrium (LD) with the
SNP of interest. As such, more recent studies have utilized LD to test the minimum
number of polymorphisms that capture the large majority of genetic variation across a
region; these polymorphisms are called SNP tagging SNPs. The tag-SNPs are genotyped
and tested for associations with phenotypes. This method maximizes the coverage of a
region for less cost and increases power for detecting an association over a region by
controlling for multiple-hypothesis testing (Eskin 2008).
Phenotypes to Evaluate Genetic Modifiers of HAART Efficacy. The general
hypothesis in HIV-related pharmacogenetic studies is that variant drug metabolism alleles
(e.g. CYP3A polymorphisms) are associated with altered plasma levels or target tissue
exposures of antiretroviral drugs (e.g. efavirenz, nelfinavir) leading to a differential
clinical response.
Pharmacogenetic studies of HIV treatments have explored two phenotypes:
pharmacokinetic (PK) outcomes (e.g. drug concentrations) and clinical phenotypes, such

23
as HIV-1 viral load suppression and CD4+ T cell recovery. PK properties measure drug
absorption, distribution, metabolism, and excretion. Plasma drug levels, drug
bioavailability, intracellular drug concentrations and rates of drug biotransformation are
examples of PK parameters that have been investigated as outcomes in genetic
association studies of HIV-related medications (Singh 2006, Cressey 2006).
Pharmacodynamic phenotypes are related to the effect of a drug, including both adverse
and therapeutic effects (Singh 2006). In general, anti-HIV drugs suppress HIV-1 viral
replication and efficacy is measured by quantifying HIV-1 RNA in the plasma. The
long-term effect of anti-HIV drugs occurs immunologically, with an increase in CD4+
cell count as a function of viral suppression. Adverse effects of antiretroviral drugs
include central nervous system and liver toxicity (Haas 2004, Neff 2006).
Pharmacokinetic parameters are easily measured and often modeled as continuous
variables to increase the power of detecting a genetic association. Most studies in the
HIV pharmacogenetic literature have investigated the relationship between PK
parameters and genetic variants that have experimental evidence of affecting enzyme
activity. The major limitation of PK outcomes is the weak correlation with the intended
clinical phenotype.
The focus of this dissertation is to investigate genetic associations with
pharmacodynamic outcomes in HIV treatment. The following sections describe the study
designs and critically evaluate the results from pharmacogenetic studies in the literature
that have investigated pharmacodynamic outcomes of viral load and CD4+ counts.

24
REVIEW OF STUDY DESIGNS

Observational Pharmacogenetic Studies of HAART. Observational study
designs have the advantage of investigating genetic associations with diverse treatment
regimens since the sample population is not treated uniformly as in clinical trials.
However, stratification of cohorts into similar therapeutic regimens reduces power
considerably since there are several combinations of antiretroviral drugs that are defined
as HAART regimens. Further, the ‘real life’ setting of a cohort study complicates the
evaluation of treatment effects, potentially producing a type of bias called confounding
by indication. Prognostic factors may influence treatment decisions leading to
differential attributes among responders and non-responders. These differences must be
evaluated for confounding in observational studies since patients are seen by personal
physicians for HIV treatment in contrast to randomization in clinical trials.
The Swiss HIV Cohort Study is the largest cohort of HIV-infected Europeans,
which has enrolled 16,271 subjects (31% female) from 1988-2010 (> 60,000 person-
years of follow-up, www.shcs.ch). The Swiss HIV Cohort was one of the first studies to
conduct pharmacogenetic research in relation to HAART (Fellay 2002). Other cohort
studies of HIV-infected persons in Asia, Canada, Germany, France, Spain, South
America and the UK have contributed to HIV pharmacogenetic research; however the
number of patients who have contributed genetic data in any of these individual cohorts is
minimal (n<150). While collaborative efforts to pool prospective data, such as the
Antiretroviral Therapy Cohort Collaboration and NA-ACCORD, have had a significant

25
impact on research related to HIV prognosis, these efforts have yet to contribute to
pharmacogenetic research (Sterne 2009).
Randomized Clinical Trial-nested Pharmacogenetic Studies of HAART. The
majority of HIV pharmacogenetic studies have been nested within randomized clinical
trials (RCTs) of HAART regimens. There are considerable exclusion criteria for RCTs;
often subjects who have had experience with antiretroviral drugs and/or are treated with
concomitant medications are excluded from the sample population. These restrictions are
especially important when conducting pharmacogenetic studies of HAART regimens, as
antiretroviral drug experience and concomitant drugs may modify optimal treatment
response (Mocroft 2004, Piscitelli 2001, Marzolini 2004). The randomization of
treatment arms provides internal control for potential unknown confounders of
pharmacogenetic associations. If RCT subjects are chosen randomly for a genetic sub-
study, the advantage of internal confounding control is preserved. Issues in the
retrospective assembly of RCT subjects for a genetic sub-study include sampling subjects
from different RCT protocols, low power to detect relatively marginal genetic
associations with HAART response due to small sample size and limited generalizability
to those excluded from the trial.
The prominent clinical trials group in the U.S. is the AIDS Clinical Trials Group
(ACTG), established in 1987 (www.AACTG.org). In 1991, the ACTG split to create the
pediatric AIDS clinical trials group (PACTG) and the Adult ACTG (www.AACTG.org).
Data from various ACTG and PACTG trials have been used in post-hoc pharmacogenetic
sub-studies to explore genetic modifiers of specific antiretroviral medications on clinical

26
measures of drug efficacy and pharmacokinetic properties. The following sections will
briefly detail the design of these ACTG/PACTG trials and their pharmacogenetic sub-
studies. Pharmacogenetic study results are summarized later in this chapter.
ACTG 384 protocol. The ACTG 384 protocol was a randomized, double-blind,
2x3 factorial clinical trial with the first factor including 2 NRTI levels (d4T + ddI or AZT
+ 3TC) and the second factor including 3 levels (the NNRTI efavirenz, the PI nelfinavir
or both) (Robbins 2003, Shafer 2003). Thus two of the six groups received a 4-drug
regimen, and four groups received a 3-drug regimen. Inclusion criteria for the 72-week
trial were: age>16 years, HIV-1 viral load >500 copies/ml, and antiretroviral treatment
naïve (clinicaltrials.gov). The two 4-drug groups were compared with the four 3-drug
groups with respect to the primary endpoint, time until virologic failure (defined as a
decrease of < 1.0 log10 copies/mL in the plasma HIV-1 RNA level at week 8 of
treatment, an increase of >1.0 log10 copies/mL above a nadir measurement, an increase
to 12,000 copies/mL within the first 24 weeks of treatment, or a plasma HIV-1 RNA
level of >200 copies/mL occurring either in a subject who had 2 previous measurements
of <200 HIV-1 RNA copies/mL or at any time after week 24 of treatment) (Robbins
2003, Shafer 2003, Haas 2005).
A total of 980 subjects were enrolled in the trial and DNA was available on 504
(51%) of those subjects. Sampling bias is not a concern in this sub-study as the
characteristics of those who did not give consent or were enrolled in ACTG 384 prior to
the genetic sub-study implementation in 2002 were generally similar to those who did
provide blood specimens for genetic analyses (Haas 2005, Motsinger 2006). Subjects

27
who participated in the genetic sub-study were slightly older (mean age 37.7 vs 36.2
years, p=0.008) and had somewhat lower median CD4+ count at baseline (281 cells/mm
vs. 320 cells/mm, p=0.024, Haas 2005). Lower median CD4+ count is related to poorer
prognosis (Hulgan 2007) and if these subjects did not provide biospecimens due to their
illness, sampling bias is a concern. However, the difference in CD4+ counts is not
clinically significant, as treatment guidelines recommend the initiation of HAART when
CD4+ counts drop below 350/mm. There were no significant differences with regard to
ethnicity or median plasma HIV-1 RNA level at baseline (Haas 2005). Sampling bias is
not a concern in the ACTG 384 genetic sub-study.
PACTG 382 protocol. The PACTG 382 protocol was a Phase 1, open-label trial
designed to evaluate the pharmacokinetics, tolerance and efficacy of HAART treatments
containing efavirenz (NNRTI) and nelfinavir (PI) in children less than age 16
(www.clinicaltrials.gov). Additional inclusion criteria were baseline HIV-1 RNA >400
copies/ml, no previous treatment with NNRTIs or PIs, no history of other chronic
diseases, and current treatment with at least one NRTI (www.clinicaltrials.gov). All
subjects were given efavirenz (NNRTI) with nelfinavir (PI) for 48 weeks (Saitoh 2005).
At week 8, patients who had subtherapeutic plasma efavirenz or nelfinavir levels had
their dosages increased to achieve target plasma drug levels (www.clinicaltrials.gov).
Treatment efficacy was quantified by plasma HIV-1 RNA loads, where subjects were
followed for the event of ‘virologic response’ (<400 copies/ml, assay detection limit=400
copies/ml) (Saitoh 2005).

28
Subjects who received HAART for at least 8 weeks, had pharmacokinetic data at
2 weeks and had virologic data available at 2, 4, 8, and 20 weeks, were selected for a
pharmacogenetic sub-study (n=71 of 105 enrolled subjects) (Saitoh 2005). The baseline
distributions of covariates for subjects who were not selected for the genetic sub-study
were not available (Saitoh 2005). Plasma drug levels were optimized at week 2, 6, and
10 for therapeutic response in subjects who did not achieve target plasma concentration
with initial dose (Starr 1999). Standardizing plasma drug levels to a target level in all
subjects might mask the marginal effect of the variant being tested on clinical outcomes
measured after standardization.
PACTG 366 protocol. The PACTG protocol 366 was a partially randomized,
open label study among children (<21 years old). Eligible subjects were naïve to at least
2 drugs, of which one was nevirapine (NNRTI), ritonavir (PI), or nelfinavir (PI) (Kovacs
2005). Antiretroviral drug-experienced subjects were eligible if they had failed virologic
and immunologic response criteria, while treating for ≥8 weeks (Kovacs 2005). Enrolled
subjects were assigned to one of five treatment groups for 100 weeks; each subject’s
specific treatment assignment was based on their previous antiretroviral drug experience.
Subjects with NRTI experience were stratified by age and CD4+ percentage and then
randomized to receive one of the PIs, nelfinavir or ritonavir, plus nevirapine (NNRTI)
and two NRTIs. The study population was comprised of subjects with various
antiretroviral backgrounds, ranging from PI- and NNRTI-experienced (16%) to PI- and
NNRTI-naïve (35%). Virologic response was defined as plasma HIV RNA <400
copies/ml. (Kovacs 2005)

29
PACTG 377 protocol. The PACTG 377 protocol was a phase II, multicenter,
randomized, open-label clinical trial comparing pharmacokinetic parameters of HIV-
infected children aged 6 months to 17 years who were naïve to d4T and 3TC (NRTIs),
PIs, and NNRTIs. Subjects (n=193) were randomly assigned to one of four regimens: 1)
d4T plus nevirapine (NNRTI) and ritonavir (PI), 2) d4T, 3TC (NRTIs) plus nelfinavir
(PI), 3) d4T plus nevirapine and nelfinavir, or 4) d4T plus nevirapine and nelfinavir.
Subjects who were randomized to both nelfinavir and nevirapine (NNRTI) experienced
nelfinavir plasma concentrations that were twice as great compared to subjects who were
randomized to nelfinavir in the absence of nevirapine (Floren 2003). This is an example
of a drug-drug interaction that may disguise true genetic associations with nelfinavir
pharmacokinetics or response.
Combined PACTG 366 and 377 protocol sub-study. A retrospective genetic
study among 152 PACTG 366/377 subjects who received a nelfinavir-containing
HAART regimen was conducted to test the association of ABCB1 and CYP2C19 variants
with both virologic and immunologic response. Virologic response was defined as
reaching undetectable viral loads (<400 copies/ml) at 12 and 24 weeks. Immunologic
response was determined by change in CD4+ T cell percentage from baseline to 12 and
24 weeks. PACTG 366 children were more likely than PACTG 377 children to be male
(61% vs. 43%), and had lower baseline CD4+ percentages (19% vs. 27%) (Saitoh 2009).
Since lower baseline CD4+ percentage at HAART initiation is associated with HIV
disease progression to an AIDS-defining event or death (Hulgan 2007), the subjects of
PACTG 366 may be less likely to reach undetectable viral loads, or attain an

30
immunologic response to HAART. However, selection into the genetic study was based
upon randomized HAART treatment containing nelfinavir and thus, not likely to bias the
results.

RESULTS OF PHARMACOGENETIC STUDIES

This section summarizes the evaluations of modification of HAART response by
variants in CYP3A4/CYP3A5, CYP2C19, CYP2B6, and ABCB1 (Tables 3.1-3.4), using
virologic and immunologic outcomes.
CYP3A variants. CYP3A4 is the primary enzyme in antiretroviral drug
metabolism. All NNRTIs and PIs are substrates, inducers, or inhibitors of CYP3A4 and
there is substantial substrate overlap with CYP3A5 (Piscitelli 2001). One genetic variant
in CYP3A4 (rs2740574) and two in CYP3A5 (rs776746, rs10264272) were investigated
for associations with HAART virologic and immunologic response in ACTG and
PACTG genetic sub-studies (Table 3.1).

31
Table 3.1 CYP3A4 and CYP3A5 pharmacogenetic studies

Author RCT N Drugs* Outcome Statistical Test rs # Results†
Haas
(2005)
ACTG
384

340
EFV ±
NFV
with
NRTIs
a) time to
virologic
failure b)
median |∆ |
CD4+ at 48
wks
a) Tarone’s
exact
b) rank based
trend test

rs2740574

a) p > 0.1
b) p > 0.1
Motsinger
(2006)
ACTG
384

155 EFV
with
NRTIs
virologic
failure
prediction
model
Multifactor
dimensionality
reduction model

rs2740574
~ 50%
accuracy
Saitoh
(2005)
PACTG
382
61 EFV,
NFV,
with
NRTIs
< 400
copies/ml of
HIV RNA at
8 wks

Fisher’s exact
test
rs2740574 p = 0.92
Haas
(2005)
ACTG
384
340
EFV ±
NFV
with
NRTIs
a) time to
virologic
failure b)
median |∆ |
CD4+ at 48
wks

a) Tarone’s
exact
b) rank based
trend test

rs776746‡
a) p > 0.1
b) p > 0.1
Saitoh
(2005)
PACTG
382
71 EFV,
NFV,
with
NRTIs
<400
copies/ml of
HIV RNA at
8 wks
Fisher’s exact
test
rs776746‡
rs10264272‡

p= 0.74,
p= 0.92
* EFV= efavirenz (non-nucleotide reverse transcriptase inhibitor)
NFV=nelfinavir (protease inhibitor)
NRTI= nucleotide reverse transcriptase inhibitor
† data not shown by genotype
‡ CYP3A5 polymorphism

Associations of CYP3A variants and time to virologic failure were explored
among subjects randomized to efavirenz-containing HAART regimens in the ACTG 384
genetic sub-study (n=185 efavirenz (NNRTI) and nelfinavir (PI), n=155 efavirenz
monotherapy). Consistent with the lack of association with efavirenz and nelfinavir

32
pharmacokinetics (Fellay 2002, Singh 2003, Haas 2005, Saitoh 2005), CYP3A4
rs2740574 and CYP3A5 rs776746 genotypes were not associated with time to virologic
failure (p>0.1) using Tarone’s exact test for k-sample censored outcomes, nor did they
differ in median absolute change in CD4+ count from baseline to week 48 using the exact
Jonckheere-Terpstra (rank-based trend) test (p>0.1, Haas 2005).
It is possible that including subjects who were randomized to efavirenz with
nelfinavir (n=185/340) biased the association of CYP3A variants and treatment response
(i.e virologic failure). The co-administration of nelfinavir with efavirenz reduced the
efavirenz-associated induction of CYP3A activity in experimental results (Fellay 2005).
Thus, lack of CYP3A activity may occur regardless of CYP3A variant status, due to the
co-administration of nelfinavir. This drug-drug interaction might have hidden a true
association of the CYP3A genetic variants with response to efavirenz-containing
regimens. Unfortunately, investigators did not report results in the 155 subjects treated
with efavirenz alone.
The minor allele frequency (MAF) of CYP3A4 –392A>G (rs2740574) for
Whites, African Americans, and Hispanics in the ACTG 384 protocol were 3.7%, 63.5%,
and 12.4%, respectively (Haas 2005). The distribution is similar for CYP3A5 6986A>G
(rs776746), as these variants are in linkage disequilibrium in all populations (r
2
= 0.90).
The strong association of rs2740574 and rs776746 with self-reported ethnicity (p<0.001),
implies the potential for a significant confounding relationship of self-reported ethnicity
in the association between CYP3A4/5 variants and HAART response. Although the
investigators of this particular pharmacogenomic study did not report the association of

33
ethnicity with HAART virologic response, it has been previously observed that Whites
may respond better than African-Americans. For example, White women were more
likely than African American women to attain a virologic response, defined as ≤80 copies
HIV-1 RNA/ml, in the WIHS cohort (HR=1.34, p=0.005, Anastos 2005). With
adjustment for self-reported ethnicity in this ACTG 384 genetic study, the null
association between CYP3A4/5 variants and virologic response to efavirenz remained.
The ACTG 384 genetic model may suffer from residual confounding due to
genetic ancestry, despite adjustment for self-reported ethnicity (White, African American,
and Hispanic) (Haas 2005). Self-reported ethnicity likely does not completely account
for each subject’s underlying genetic ancestry (Pritchard 2000, 2001), and thus bias could
go in either direction since a confounder was not adequately captured (Rothman 2009).
In addition to the issue with analyses on a heterogeneously treated sample
population (efavirenz with nelfinavir, and efavirenz alone) and potential for residual
confounding due to population stratification, this ACTG 384 genetic sub-study was
under-powered to observe an effect with rs2740574 in Whites (n=177), African
Americans (n=123), and Hispanics (n=61). The authors did not report frequencies of
those with the event (virologic failure), however another study conducted with the 155
subjects treated with efavirenz alone reported 32 failures (results discussed below).
Extrapolating this information to calculate power in a homogeneously treated sample of
unknown ethnicities and assuming a MAF of 0.04 (that of Whites in this study), the
power to detect an OR of 1.1-1.3 is from 5-6% with a log-additive model. Though the

34
ACTG 384 study is severely limited by power, the interest in this SNP is not to be
overlooked, despite the potential for false negative associations in Whites (MAF=0.04).
Interactions between drug-metabolism (CYP) and transporter genes (ABCB1)
were also explored in the genetic data from the ACTG 384 trial (Motsinger 2006).
Multifactor dimensionality reduction (MDR) methods were used to detect multilocus
genotype combinations in CYP and ABCB1 genes associated with risk of virologic or
toxicity failure. Multifactor dimensionality reduction is a non-parametric method that
does not assume a particular genetic inheritance mode, such as additive, dominant or
recessive. Each possible n-loci model is tested on a training set to classify each genotype
combination as a high-risk or low-risk pattern depending on the number of cases and
controls that present the n-loci combination. The model that misclassified the fewest
individuals in the testing set was chosen to predict treatment failure. MDR methods were
employed on the 155 subjects who were randomized to efavirenz plus NRTIs to
investigate the genetic relationship in a homogenous treatment group. CYP3A variants
rs2740574 and rs776746 did not predict virologic response with statistical accuracy and
had poor cross-validation consistency (~50% accuracy) in White or African American
subjects. This study was severely limited by the small sample size, the small number of
virologic failures (n=32), and the lack of details regarding the MDR procedures.
Saitoh et al. (2005) investigated the effect of SNPs in CYP and ABCB1 genes on
virologic response at week 8 in children treated with concomitant efavirenz and nelfinavir
in PACTG 382. Reaching undetectable HIV-1 RNA levels (<400 copies/ml) was not
statistically significantly associated with CYP3A4/5 polymorphisms rs2740574,

35
rs776746, and rs10264272 (Fisher’s exact p= 0.74 - 0.92, Saitoh 2005). Unfortunately,
the frequency of children who reached < 400 HIV-1 RNA copies/ml were not reported
for each genotype. In this same PACTG 382 genetic study, investigators tested
pharmacokinetic associations with the rs2740574 genotypes (Wilcoxon rank sum test).
Specifically, there was a modestly higher median efavirenz area under the curve (AUC)
in children with the A/A genotype compared to children with the G/A and G/G genotypes
(67.5, 57.8, and 50.2 mg*h/ l, respectively p=0.04, Saitoh 2005). If the A/A genotype is
truly associated with higher plasma efavirenz concentration, this may lead to earlier
achievement of undetectable viral loads compared to the heterozygotes or wild type
homozygotes. Misclassification of the outcome (undetectable viral load) may explain the
null association of rs2740574 with virologic response summarized above. The HIV RNA
assay technology has become much more sensitive (with lower detection limits) over
time. PACTG subjects with “undetectable” viral loads between 50-400 copies would
have been classified as detectable with the current assay limit (50 copies/ml). If the
misclassified ‘undetectable’ children were mostly G/A or G/G genotypes, this association
would be negatively biased.
The most significant limitation of this PACTG 382 genetic sub-study is the
potential for selection bias, as discussed previously. The individual optimization of drug
dosages to achieve target plasma efavirenz and nelfinavir levels at weeks 2 and 6 before
entry to the sub-study (week 8) probably attenuated this association, since the outcome
(undetectable viral load) was measured after plasma concentration standardization. The
ad-hoc analyses of genetic associations in this PACTG clinical trial is not demonstrative

36
of a proper genetic epidemiology study design, as the optimization of drug levels severely
limits the potential for observing a genetic effect on clinical drug response.
Overall, the studies investigating CYP3A4/5 variants were underpowered to
detect an association with HAART virologic and immunologic response and suffered
from potential biases as discussed previously. It would have been helpful to know the
frequencies of subjects who reached undetectable viral loads by genotype in each study.
Crude pooling of the data would lend insight into whether a trend was present and the
probable mode of inheritance (i.e. dominant, recessive, or log-additive). These
CYP3A4/5 polymorphisms should be further studied, as conclusions about their
association with HAART clinical response cannot be drawn.
CYP2C19 variants. CYP2C19 exclusively metabolizes nelfinavir to its
metabolite, M8, and thus non-functional CYP2C19 alleles are hypothesized to negatively
affect the ratio of M8:nelfinavir in the plasma. Nelfinavir is additionally metabolized by
CYP3A4, but not through the M8 metabolite (Piscitelli 2001). The rate of metabolism of
nelfinavir to M8 was reduced by 50% in carriers of CYP2C19 681G>A (rs4244285)
variant allele (Hirt 2007). This polymorphism was investigated for associations with
HIV-1 viral load levels (virologic response) and CD4+ counts (immunologic response) in
the ACTG 384, PACTG 382 and 366/377 populations (Table 3.2). Virologic failure in
ACTG 384 was defined as either a decrease of < 1.0 log10 copies/mL in the plasma HIV-
1 RNA level at week 8 of treatment, an increase of >1.0 log10 copies/mL above a nadir
measurement, an increase to 12,000 copies/mL within the first 24 weeks of treatment, or
a plasma HIV-1 RNA level of >200 copies/mL occurring either in a subject who had 2

37
previous measurements of <200 HIV-1 RNA copies/mL or at any time after week 24 of
treatment. The PACTG definition of virologic response was reaching undetectable viral
load (<400 copies/ml).

Table 3.2 CYP2C19 pharmacogenetic studies
Author RCT N Drugs* Outcome Statistical
Test
rs# Results
Haas
(2005)
ACTG
384
348
NFV ±
EFV
with
NRTIs
a) time to
virologic
failure
b)
median
|∆ |
CD4+ at
48 wk

a) Tarone’s
exact
b) rank
based trend
test

rs4244285
a) probability
of failure at 48
wks
A/A 0.18
G/A 0.12
G/G 0.22
p = 0.07
b) p > 0.1†
Saitoh
(2005)
PACTG
382
61 EFV,
NFV
with
NRTIs
< 400
copies/ml
of HIV
RNA at 8
wk

Fisher’s
exact

rs4244285

p = 0.2†
Saitoh
(2009)
PACTG
366/377
152
NFV ±
NVP,
RTV
with
NRTIs
a) < 400
copies/ml
of HIV
RNA at
24 wk
b) |∆ |
CD4+ at
12wk, 24
wk

a) Cochran-
Armitage
trend test
b) rank-
based trend
test

rs4244285
a) A/A 63%
G/A 69%
G/G 46%
p = 0.01
b) p = 0.5† 12
wk, p= 0.44†
24 wk
* EFV= efavirenz (non-nucleotide reverse transcriptase inhibitor), NFV=nelfinavir (PI)
NRTI= nucleotide reverse transcriptase inhibitor
† data not shown by genotype

There was a trend toward lower risk of virologic failure with CYP2C19
rs4244285 A allele (variant) over 48 weeks of a regimen containing nelfinavir (n=348,

38
n=164 NFV only) in the ACTG 384 sub-study (p=0.07, Haas 2005). Subjects with the
A/A genotype or G/A genotype had an 18% and 12% probability of virologic failure,
compared to a 22% probability of failure in subjects with the G/G genotype (Haas 2005).
However, these analyses are likely biased by including subjects treated with concomitant
efavirenz in over half of the sample population (n=184). The modification of nelfinavir
pharmacokinetics by efavirenz exposure was demonstrated in a meta-analysis of
pharmacokinetic studies in the ACTG group. Patients who received nelfinavir plus
efavirenz had a 45% reduction in the mean nelfinavir concentration in serum (p=0.01),
and a 31% more rapid half-life (p=0.01) compared to patients who received nelfinavir
monotherapy (Smith 2005). Conflicting results have also been reported; the package
inserts for efavirenz (Sustiva
®
) and nelfinavir (Viracept
®
) claim that coadministration of
these antiretroviral drugs result in a 20% increase of nelfinavir plasma concentration.
Thus, we cannot conclude whether CYP2C19 681G>A genotypes modified nelfinavir
clinical response in the absence of efavirenz co-treatment.
Despite evidence for an association between variant CYP2C19-mediated
metabolism of nelfinavir and undetectable viral load, there was no evidence to suggest an
effect of the A allele on immunologic response over 48 weeks (Jonckheere-Terpstra rank
based p trend >0.1, Haas 2005). Unfortunately median increases in CD4+ cell count
were not reported for each genotype. The lack of statistically significant trend test is not
informative to conclude whether the magnitude of CD4+ cell increase trended or if the
major limitation of this study was low power. Long-term immunologic response occurs
after suppressed viremia (i.e. consecutive measures of <50 HIV-1 RNA copies/ml) is
attained. In the Swiss HIV cohort, median increases of CD4+ cells after suppression

39
were 87, 52, and 19 cells/ml per year in the first year, years 2-3, and years 4-5,
respectively (Wolbers 2007). It is possible that the length of follow-up (48 weeks) was
not sufficient to attain suppressed viremia and subsequently observe a significant
difference in CD4+ cell recovery between CYP2C19 681G>A (rs4244285) genotypes.
The association between an undetectable viral load at week 8 (<400 HIV-1 RNA
copies/ml) and rs4244285 genotypes was not striking in the 61 children treated with
nelfinavir and efavirenz in PACTG 382 (p=0.20, frequencies of response phenotypes not
given by genotype, Saitoh 2005). However, differences in virologic response were
observed at week 24 among a larger sample of children treated with various nelfinavir-
containing HAART regimens (n=152 PACTG 366/377, Saitoh 2009). In total, 69% and
63% of subjects with rs4244285 G/A and A/A genotypes achieved undetectable viral load
levels (<400 copies/ml) compared with 46% of those with wildtype (G/G) (p=0.01).
When stratified by self-reported ethnicity, these differences were observed both in
African American (n= 92, p=0.02) and White subjects (n=24, p=0.03). Among Hispanics
(n=32), no A/A genotypes were observed (n=8 G/A genotypes). Median absolute change
in CD4+ percentages from baseline to 12 and 24 weeks were not significantly different
among the three genotypes (CD4+ counts not shown) using the exact Jonckheere-
Terpstra (rank-based trend) test (p >0.50, 0.44 respectively, Saitoh 2009).
Further investigation of the association between rs4244285 genotype and
nelfinavir clinical response is warranted. The ACTG and PACTG nested studies suffer
from small cell numbers when stratified by uniform HAART regimens (e.g. nelfinavir
alone). Perhaps a stronger effect would have been observed in the PACTG 366/377

40
study if the subjects randomized to nelfinavir with ritonavir (PI) were excluded, since the
bioavailability of nelfinavir was significantly reduced with concomitant administration of
ritonavir, but not with nevirapine co-treatment (NNRTI) in these data (p=0.002, 0.70,
respectively Saitoh 2009).
CYP2B6 variants. Hepatic expression of CYP2B6 varies according to gender
and ethnicity (Lamba 2003). Females had higher amounts of CYP2B6 mRNA and
protein (p<0.001, p<0.009), and 1.6-fold greater activity (p< 0.05) than did male subjects
in an experimental study of human liver samples (Lamba 2003). CYP2B6 activity was
3.6 and 5.0-fold higher in Hispanic females than in White and African American women,
respectively (p<0.04, Lamba 2003).
Experimental results have shown that the rs3745274 (516G>T) polymorphism is
associated with decreased CYP2B6 expression in the human liver (Hoffman 2008). The
common polymorphism results in diminished activity through an aberrant splicing
mechanism (Hoffman 2008). CYP2B6 516G>T is most common in Hispanics
(MAF=0.35, compared with 0.24 in Whites, and 0.31 in African Americans, Haas 2005).
The less common missense SNP rs3211371 (1459C>T) has minor allele frequencies of
0.13, 0.04, and 0.07 in Whites, African Americans and Hispanics, respectively (Haas
2005). These two CYP2B6 polymorphisms have been investigated as predictors of
virologic and immunologic response with the CYP2B6 substrates, efavirenz and
nevirapine (NNRTIs) (Table 3.3).

41
Table 3.3 CYP2B6 pharmacogenetic studies
Author Study N Drugs* Outcome Statistical test rs# Results
Saitoh
(2005)
PACTG
382
71 EFV,
NFV
with
NRTIs
< 400
copies/ml
of HIV
RNA at 8
wks

Fisher’s exact rs3211371 p = 0.32†
Haas
(2005)
ACTG
384
340
EFV ±
NFV
with
NRTIs
a) time to
virologic
failure
b) median
|∆ | CD4+
at 48 wks

a) Tarone’s
exact
b) rank based
trend test

rs3745274
a) probability of
failure at 48 wk
T/T 20%
G/T 10%
G/G 10%
p = 0.35
b) p > 0.1†
Motsinger
(2006)
ACTG
384
155 EFV
with
NRTIs
virologic
failure
prediction
model
Multifactor
dimensionality
reduction
model

rs3745274
< 50% accuracy
Saitoh
(2007)
PACTG
366/377
126
NVP ±
RTV,
NFV
with
NRTIs
a) < 400
copies/ml
HIV RNA
at 24 wks
b) CD4+∆
at 12 wks

a) Fisher’s
exact
b) Multivariate
linear
regression

rs3745274
a) p = 0.24†
b) β= 4.7 p=0.03
* EFV= efavirenz (non-nucleotide reverse transcriptase inhibitor), NFV=nelfinavir,
NRTI= nucleotide reverse transcriptase inhibitor
† data not shown

In children treated with concomitant efavirenz and nelfinavir in PACTG 382,
reaching undetectable HIV-1 RNA levels (<400 copies/ml) was not significantly different
among CYP2B6 1459C>T genotypes (p=0.32, frequencies of response phenotypes not
given by genotype, Saitoh 2005). The limitations of the PACTG 382 genetic sub-study
include small sample size, lack of study results, and the potential for selection bias. The
optimization of plasma drug levels at week 2 and 6 is a concern (Starr 1999), as virologic
response was measured at week 8 (Saitoh 2005). This association was likely attenuated

42
due to optimizing drug exposure before the outcome was measured.
Despite the association with plasma and intracellular efavirenz exposure (Rotger
2005, 2007), CYP2B6 516G>T (rs3745274) was not associated with time to virologic
failure in the ACTG 384 genetic sub-study of subjects randomized to efavirenz-
containing HAART regimens (p=0.35, Haas 2005). Subjects with the T/T genotype had a
20% probability of virologic failure, compared to a 10% virologic failure probability in
G/T and T/T genotypes (Haas 2005). These results suggest a recessive mode of
inheritance and modeling the G/T and T/T genotypes together may have increased power
to observe an association with virologic failure. Additionally, there is a concern for a
drug-drug interaction that may have modified the association among the subjects
concomitantly treated with nelfinavir and efavirenz (n=185/384).
A single SNP model of CYP2B6 516G>T was not predictive of virologic
response to efavirenz in the n=155 ACTG 384 subjects randomized to efavirenz without
nelfinavir (p>0.5, Motsinger 2006). Though the multilocus dimensionality reduction
model is a limitation in the ACTG 384 dataset (n=155, 32 events), excluding concomitant
nelfinavir recipients avoided confounding by possible genetic effects mediated through
nelfinavir rather than efavirenz.
In the PACTG 366/377 sub-study of children treated with nevirapine (NNRTI)
and either of the PIs, ritonavir and nelfinavir, or both PIs, there were no differences in the
number of children achieving an undetectable viral load at 24 weeks among the three
genotypes of rs3745274 (516G>T, p=0.24, Saitoh 2007). However, the binary outcome
(< 400 copies/ml or ≥ 400 copies/ml) does not have the specificity of current viral RNA

43
detection assays (50 copies/ml). Thus, outcome misclassification (undetectable by 400
copies/ml limit, but detectable by 50 copies/ml limit) may have negatively biased the true
association of the T allele and an undetectable viral load.
In contrast to virologic response, children with the T/T genotype had larger CD4+
percentage increases (+10.5%) compared with children with G/G (+4.7%, p=0.01) and
G/T (+8.2%, p=0.06) genotypes after treatment with a nevirapine (NNRTI)-containing
regimen for 24 weeks (Saitoh 2007). After adjusting for self-reported ethnicity, gender,
concomitant PIs, and T/T genotype for ABCB1 2677G>T, the T/T genotype for
rs3745274 was statistically significantly associated with an increase of 4.7% CD4+ cells
over 12 weeks (Saitoh 2007, p=0.025). The genetic association with immunologic
response to HAART was not replicated in recipients of efavirenz and/or nelfinavir-
containing regimens over 48 weeks (p>0.1, Haas 2005).
The early immune response observed in PACTG 366/377 subjects may be unique
to nevirapine containing regimens. However, the null results in the efavirenz and/or
nelfinavir recipients of the ACTG 384 sub-study may have produced a false negative
association due to the potential for drug-drug interaction, as discussed previously.
Moreover, utilizing a non-parametric method (Jonckheere-Terpstra rank-based trend test)
to test for an association with change in CD4+ over time may be too conservative. Since
change in CD4+ counts is modeled as a continuous outcome, log transformation is
probably sufficient enough to satisfy model assumptions.
ABCB1 variants. Two SNPs, ABCB1 3435C>T (rs1045642) and 2677G>T
(rs2032581), appear to be associated with altered hepatic P-gp expression (Marzolini

44
2004). Since P-gp is crucial to the distribution and excretion of all protease inhibitors
(Table 2.2), there has been wide interest in the potential genetic influence on HAART
response. The studies shown in Table 3.4 have investigated the two missense SNPs as
predictors of virologic and immunologic response with protease inhibitors.

Table 3.4 ABCB1 pharmacogenetic studies

Author Study§ N Drugs* Outcome Statistical test rs# Results
Fellay
(2002)
Swiss
HIV
Cohort
Study
134 EFV or
NFV
with
NRTIs
>200 increase
CD4+ count
at 24 wks
Logistic
regression

rs1045642
T/T
OR=3.0
(1.3-7.1)
Nasi
(2003)
cohort of
ART
naïve,
Italy
106 PIs
with
NRTIs

a) ∆ log
HIV-1 RNA
copies/ml at
12 wks
b) CD4+
count
increase at 12
wks

a) Kruskal-
Wallis
b) Kruskal-
Wallis

rs1045642
a) p=0.45†
b) p=0.13†
Brumme
(2003)
cohort of
ART
naïve,
Canada
461 PI with
NRTIs
a) time to
virologic
failure
b) time to
immunologic
failure

a) Cox
regression
b) Cox
regression

rs1045642
T/T
a) HR=0.67
(0.45-0.98)
b) HR=1.02
(0.68-1.54)

Haas
(2005)
ACTG
384
340
EFV ±
NFV
with
NRTIs
a) time to
virologic
failure at 48
wks
b) median |∆ |
CD4+ at 48
wks

a) Tarone’s
exact
b) rank based
trend test

rs1045642
a)probability
of failure at
48 wk
T/T 0.05
C/T 0.15
C/C 0.15
p=0.08
b) p>0.1†

45
Table 3.4, continued

Author Study§ N Drugs* Outcome Statistical test rs# Results
Saitoh
(2005)
PACTG
382
71 EFV,
NFV
with
NRTIs
a) < 400
copies/ml of
HIV RNA at
8 wks

b) CD4+
count at 4, 8,
20 wks
a)Fisher’s exact
b)Kruskal-
Wallis

rs1045642
a) C/C 59%
(17/29)
C/T 91%
(29/32)
T/T 57%
(4/7)
C/C vs C/T
p=0.004,

b) p=0.87º†
p=0.26º†
p=0.30†

Anderson
(2006)
RCT
ART
naïve,
USA
33 IDV
with
NRTIs
a) mean ∆
log HIV-1
copies/ml at
52 wks
b) mean
CD4+
increase at 52
wks
a) Student’s t
test
b) Student’s t
test

rs1045642
a) p>0.1†
b) p>0.1†
Saitoh
(2007)
PACTG
366/367
126
NVP ±
RTV,
NFV
with
NRTIs
CD4+∆ at 12
wks

Multivariate
linear
regression

rs1045642
T/T
β= –0.728,
p=0.73

Saitoh
(2009)
PACTG
366/377
152
NFV ±
NVP,
RTV
with
NRTIs
a) < 400
copies/ml of
HIV RNA at
24 wk
b) |∆ | CD4+
at 12wk, 24
wk

a) Cochran-
Armitage trend
test
b) rank-based
trend test

rs1045642
a) p=0.06†
b) p=0.08†,
p=0.21†
Winzer
(2005)
cohort of
ART
naïve,
Germany
72 PI or
NNRTI
with
NRTIs
a) ∆ log
HIV-1 RNA
copies/ml at
4,12 wks
b) CD4+
increase at 4,
12 wks

a) Kruskal-
Wallis
b) Kruskal-
Wallis

rs1045642/
rs2032582
haplotype
a) p>0.8†,
p>0.08†
b) p>0.1†,
p>0.8†

46
Table 3.4, continued

Author Study§ N Drugs* Outcome Statistical test rs# Results
Haas
(2005)
ACTG
384
340
EFV ±
NFV
with
NRTIs
a) time to
virologic
failure at 48
wks
b) median |∆ |
CD4+ at 48
wks

a) Tarone’s
exact
b) rank based
trend test

rs2032582
a) p>0.1†
b) p>0.1†
Anderson
(2006)
RCT
ART
naïve,
USA
33 IDV
with
NRTIs
a) mean ∆
log HIV-1
RNA
copies/ml at
52 wks
b) mean
CD4+
increase at 52
wks

a) Student’s t
test
b) Student’s t
test

rs2032582
a) G/G vs
G/T
-2.2 vs –3.2
p=0.008
G/G vs T/T
-2.2 vs –3.7
p=0.04

b) G/G vs
G/T p>0.1†,
G/G vs T/T
p>0.1†
Motsinger
(2006)
ACTG
384
155 EFV
with
NRTIs
virologic
failure
prediction
model
Multifactor
dimensionality
reduction
model

rs2032582

54%
accuracy
§ RCT= randomized clinical trial, ART=antiretroviral therapy
* NRTI= nucleotide reverse transcriptase inhibitor, NNRTI= non-nucleotide reverse transcriptase
inhibitor, EFV= efavirenz (NNRTI), NVP (NNRTI), PI= protease inhibitor, NFV=nelfinavir (PI),
RTV= ritonavir (PI), IDV=indinavir (PI)
†data not shown by genotype

The relationship between immunologic response and the ABCB1 3435C>T
(rs1045642) variant was initially explored in the Swiss HIV Cohort Study, a retrospective
study of patients treated with efavirenz (NNRTI) or nelfinavir (n=134). After adjusting
for other factors, Fellay et al. (2002) observed that subjects with the T/T genotype were 3
times (95% CI 1.3-7.1) as likely to achieve an increase of at least 200 CD4+ cells after 24
weeks of HAART treatment compared to C/T and C/C genotypes. Half of the study

47
population (n=67) was chosen based on demonstration of long-term viral suppression
(<400 copies/ml); the remaining patients initiated treatment with nelfinavir or efavirenz
plus two NRTIs. Selection bias is a key issue in this finding, where half of eligible
subjects were selected based on viral suppression, which has been attributed to the loss of
productively infected CD4+ cells (Perelson 1997). If the ABCB1 3435C>T
polymorphism is truly associated with immunologic response, then the retrospective
selection of subjects with sustained virologic suppression may over-represent the T/T
genotype in this study population.
Subsequent studies attempted to validate this association in other populations
without success (Nasi 2003, Brumme 2003, Winzer 2005, Haas 2005, Saitoh 2005,
Anderson 2006, Saitoh 2007). In a study population of 155 Italian subjects treated with a
PI-containing regimen (n=106), CD4+ counts did not differ among rs1045642 genotypes
after 12 weeks (p=0.13, Nasi 2003). Similarly, the median change in CD4+ cells were
not different among the three genotypes in ACTG 384 subjects randomized to efavirenz-
containing regimens over 48 weeks (p>0.1, Haas 2005). Specifically, the T/T genotype
was associated with a decrease of 0.728 CD4+ cells over 12 weeks in PACTG 366/377
children treated with nevirapine (NNRTI) and the PIs, ritonavir or nelfinavir, though this
association was not significant (p=0.73, n=126, Saitoh 2007).
In British Columbia, a retrospective study of 455 treatment-naïve patients
initiating a triple drug regimen of unknown class, were followed for 40 months or until
time to immunological failure, defined as the time to the second of two successive CD4+
cell counts below baseline (Brumme 2003). Evaluation of adherence to medication was a

48
strength in this observational study; adherence was based on patients’ pharmacy records,
where 100% adherence was defined as filling every prescription on time. After adjusting
for baseline CD4+ cell count, plasma viral load, type of antiretroviral therapy and 100%
adherence, the ABCB1 3435C>T T/T genotype was not associated with immunologic
failure (HR=1.02, 95% CI 0.68-1.54). However, there was a trend for a protective effect
of the T allele on virologic failure in these data. After adjusting for age, sex, baseline
viral load and CD4+ count, 100% adherence, type of antiretroviral therapy and ‘time to
virologic success’ (<500 copies/ml), rs1045642 variant homozygotes (T/T) were 0.67
times (95% CI 0.45-0.98) as likely and C/T heterozygotes were 0.71 times (95% CI 0.47-
1.06) as likely to experience virologic failure compared to wild types (C/C). The
association of rs1045642 and virologic failure did not remain statistically significant in a
sensitivity analysis of patients who were 100% adherent to their medications (n=240/461,
HR not shown, p>0.05). The lack of a statistically significant association in the model
based on 100% adherent subjects is likely due to reduced power, since a statistically
significant association was found after adjustment for 100% adherence (T/T genotype
HR=0.67, 95% CI 0.45-0.98).
An association with virologic failure was also observed in the ACTG 384 sub-
study, though 46% of subjects (n=156/340) were not randomized to a protease inhibitor
(i.e. ABCB1 substrate nelfiniavir). Consistent with Brumme et al., ABCB1 3435 T/T
genotypes had more favorable virologic responses after 48 weeks compared to C/T and
C/C combined genotypes (p=0.02, log rank test, Haas 2005). The association persisted
using Cox regression after adjusting for self-reported ethnicity (effect measure not

49
reported, p=0.02). Perhaps a stronger effect would have been observed in the subset of
nelfinavir plus efavirenz recipients, as evidenced in the PACTG 382 substudy (p=0.004,
n=61 efavirenz/nelfinavir recipients).
The discordant results of the relationship between ABCB1 polymorphisms and
clinical response to HAART (Table 3.4) mainly spawn from underpowered studies and
lack of control for concomitant antiretroviral drugs. The ACTG and PACTG studies
have investigated the influence of ABCB1 polymorphisms on virologic response to
nelfinavir with varying results; however, the genetic influence of ABCB1 on clinical
response may be modified by particular ABCB1 substrates. Protease inhibitors have
varying affinities for ABCB1 (Marzolini 2004) and thus, the conflicting results of a
genetic influence on clinical response to an array of protease inhibitors is expected.
Multivariate models should be employed to account for the varying PI-based regimens
among a heterogeneously treated sample population.

PHARMACOGENETIC ISSUES SPECIFIC TO THE WIHS

Therapy in an observational cohort. As reviewed above, studies investigating
the influence of variation in CYP and ABCB1 genes on pharmacodynamic outcomes of
HAART response have primarily been conducted in a clinical trial setting with controlled
drug exposure. In observational studies such as the WIHS, the subjects are treated
independently from the study, thus HAART regimens are heterogeneous among the study
population. We will therefore have limited power to investigate the genetic influence on

50
clinical response to specific antiretroviral drugs. Instead, we will test genetic associations
with NNRTI-containing regimens compared to PI-containing regimens. To address
confounding by indication, prognostic factors, such as baseline CD4+ count, will be
compared in NNRTI, PI, and mixed HAART treatment groups.
The study population in this genetic association study design consists of WIHS
women who have initiated HAART during follow-up (n=609); this will be the largest
observational study to assess genetic associations with virologic and immunologic
response to HAART to my knowledge. The inclusion criteria extend to the initiation of
at least 2 new drugs (NRTI, NNRTI, PI) in a ≥3-drug HAART regimen containing a
NNRTI or PI. Antiretroviral-experienced (≤ 2 drugs) women who added only one new
drug to an existing regimen were not considered HAART initiators, as initiating one new
drug was not considered a sufficient difference in regimen to define the initiation of a
stringent HAART regimen.
A limitation of the WIHS study sample is the inclusion of women who are
antiretroviral drug experienced prior to initiating at least 2 new antiretroviral drugs in a
≥3-drug HAART regimen. It has been reported that treatment-experienced patients have
a poorer response to HAART and are less likely to achieve a viral load below the limit of
detection (Ledergerber 1999, Mocroft 2004). Treatment-experienced patients may have
accumulated drug-resistance through the mutation of the virus. We will conduct a
sensitivity analysis by excluding women who had prior antiretroviral drug experience
before initiating a HAART regimen, as defined above.
Many of the pharmacogenetic studies investigating clinical response outcomes

51
have been conducted with study populations treated with antiretroviral drugs that have
evidence of drug-drug interaction. We will test for genetic associations stratified by PI-
containing, NNRTI-containing and potential drug-drug interaction (e.g. efavirenz and
nelfinavir comedication, ritonavir and another PI) HAART regimens to control for the
potential modification of genetic influence on clinical response with particular regimens.
Adherence to medication. Adherence to treatment regimens is critical to attain
optimal response to treatment. The genetic sub-studies of randomized clinical trials have
the advantage of controlled drug exposure. In contrast, observational studies must collect
adherence data through questionnaire. Adherence data was implemented into the WIHS
questionnare at visit 9 (about 2 years into follow-up). At each visit, the women were
asked, ‘How often did you take antiretrovirals as prescribed over past 6 months?’ We will
control for self-reported adherence at both the HAART initiation visit and the visit after
HAART initiation for the women who initiated HAART subsequent to or at visit 9. A
dummy category for adherence will be defined for women who initiated HAART before
visit 9 (n=250/609 eligible women). Additionally, we will run separate models for the
women without adherence data to observe if the genetic effect estimate changes.
Clinical enpoint. In previous studies, phenotypes defining virologic
failure/success have been measured at a spectrum of time intervals, from initial response
(< 8 weeks) to 52 weeks post-HAART initiation. Viral loads at week 4 and week 8 are
correlated with virologic response at 24 weeks of HAART treatment (Cozzi 2001). It has
been commonly reported that 24 weeks of HAART treatment is sufficient to observe the
initial virologic response and a gradual increase in CD4+ counts (Gullick 1997,

52
Ledergerber 1999, Cozzi 2001, Mocroft 2004). In this WIHS pharmacogenetic study,
virologic response is defined as achieving an undetectable viral load (<50 copies/ml) at
the follow-up visit immediately succeeding reported HAART initiation (within 12
months).
Population Substructure. A major limitation of the aforementioned
pharmacogenetic studies of HAART is the lack of adjustment for population substructure.
Defining population substructure is crucial to ensuring the observed association is not due
to confounding. The WIHS is a multiethnic cohort of women, where over half of the
population is African American, an admixed population. WIHS women will be
characterized according to their unique genetic ancestry in the population substructure
study, which is part of this dissertation. Cluster analyses of ancestry informative markers
in the cohort will produce a continuous variable for each woman that captures her
underlying genetic make-up from several continental populations (e.g. European, East
African, Native American, Asian, etc.). Confounding by genetic ancestry will be
adjusted for by including ancestry component covariates in the logistic regression model.

53
Chapter 4: Population Substructure in the WIHS

ABSTRACT

Background. Population stratification can result in spurious associations in genetic
studies when the outcome and genotype are associated with genetic ancestry.
Objective. To characterize population substructure in the Women’s Interagency HIV
Study (WIHS), a cohort of HIV-infected and uninfected participants in the United States
(U.S).
Methods. Genotype data on 168 ancestry informative markers (AIMs) were utilized to
assess population substructure in the WIHS. A Bayesian clustering algorithm was used
to infer genetic membership in three and four assumed source populations. Principal
components analysis (PCA) was conducted to generate continuous genetic ancestry
covariates for future genetic association studies in the cohort.
Results. Participants shared ancestry with four source populations and heterogeneity was
detected within self-identified ethnic groups. African ancestry from two distinct
populations was revealed; the average combined African ancestry in African-Americans
was 0.80 with admixture from European and Asian source populations. There was an
increasing east-to-west gradient of Asian ancestry in self-identified Hispanics and small
proportions of African components were detected in Hispanics from the sites in the East.
The average European ancestry in Non-Hispanic Whites was 0.97 across sites.

54
Conclusions. Population substructure was identified within self-identified ethnic groups
and across geographical regions in the U.S., exemplifying the importance of estimating
individual genetic ancestry to control for residual ethnic confounding in genetic
association studies. Cautious interpretation of ancestry admixture estimates should be
exercised, as the labeling of source populations is sensitive to the panel of markers and
referent populations used in analyses.

INTRODUCTION

Genetic association studies to understand HIV-1 infection, pathogenesis,
progression, and variation in treatment response are increasingly common. However, as
observed with many other phenotypes [1,2], results from HIV studies are often not
replicated [3-7]. Although the primary reason for false positive genetic associations is
chance [1], a serious concern relates to confounding arising from population substructure
in studies of unrelated participants [8,9]. Confounding due to population substructure
occurs when both allele frequencies and disease occurrence differ across ethnic groups or
subpopulations [8-12]. For example, given that height differs across populations and
most alleles also vary across populations, potential confounding by race/ethnicity must be
considered.
Gross population stratification (confounding) can be addressed by adjusting
statistical models for self-reported ethnicity, but cryptic relatedness and admixture
(described below) require more detailed genetic ancestry data to guard against false

55
positive results [10-13]. Cryptic relatedness occurs within seemingly homogenous
populations within the same self-reported race/ethnicity group [10]. In theory, affected
individuals (cases) may be more related to one another than controls and as a result the
allelic distribution among cases is skewed [10]. There are examples of genetic
heterogeneity even among suspected homogeneous populations, such as Icelanders [14-
18].
Spurious associations can also occur in a study population that contains
individuals from a single ethnic group with recent genetic admixture [19,20]. Varying
proportions of ancestry from distinct ‘parental’ populations contribute to a population’s
genome; admixture describes this mosaic of genetic ancestry. Admixture is particularly
evidenced in African-American and Hispanic-American populations [20-23]. For
example, African-Americans from different United States (U.S.) geographical regions
have European ancestry varying between 11.6% and 22.5%; the remaining proportion
includes African and Native American ancestry [21]. Native American, European, and
African ancestries also contribute to the Hispanic genome in varying proportions
depending on locality [22,23]. Hispanics in the Western U.S. tend to represent historic
migration patterns from the Americas, and thus carry greater proportions of Native
American genetic ancestry than Hispanics in the East [22,23]. Additionally, Eastern U.S.
Hispanics’ possess modest proportions of African ancestry, prominent in Cuban and
Puerto Rican populations [22,23].
Addressing population substructure in genetic association studies beyond
adjustment for self-reported race/ethnicity is critical to protect against spurious results.

56
Several methods utilizing random genetic markers throughout the genome to adjust for
differences in underlying genetic ancestry have been developed [10-12,24,25]. The
‘genomic control’ method corrects for an average effect of differences in allele
frequencies across unknown sample subpopulations by applying a chi square inflation
factor for all markers tested [10]. The STRUCTURE software utilizes a Bayesian
algorithm to assign participants to multiple subpopulations by estimating admixture
coefficients based on the assumption of k ancestral populations represented in the sample
[11,24]. More recently, the EIGENSTRAT software employs principal components
analysis (PCA) to capture the main axes of variation in the sample pairwise SNP
correlation matrix [25]. The several largest principal components (PCs) are utilized in
the statistical model to control for allele frequency differences between cases and controls
due to genetic ancestry [25,26]. PCA is favored over the other methods since it is more
computationally efficient for handling large sets of markers and does not depend on prior
knowledge of the underlying populations in the sample [26,27]. However, unlike
analyses based on STRUCTURE, individual ancestry admixture proportions cannot be
easily interpreted from PCA [27].
The Women’s Interagency HIV Study (WIHS) is a prospective observational
study of multi-ethnic women from across the U.S. Over half of the cohort is self-
identified as African-American, nearly 25% identify as Hispanic, and approximately15%
are non-Hispanic White [28,29]. WIHS participants were genotyped using a focused
panel of ancestry informative markers (AIMs), which were selected to distinguish
European, West African, Asian, and Native American ancestry, to enable investigators to

57
adjust for genetic ancestry in statistical models to test genetic associations. We describe
the results of this effort herein and discuss the implications of detecting population
substructure using a particular set of markers.

METHODS

WIHS Study Design. The WIHS is a prospective study of HIV-infected women
and a comparison group of HIV-uninfected women [28,29]. Participants were recruited
from six sites across the U.S., located in Bronx/Manhattan, New York; Brooklyn, New
York; Washington, D.C.; Los Angeles, CA; San Francisco/Bay Area, CA; and Chicago,
IL. Enrollment was conducted in two phases; 2,054 HIV seropositive and 569
seronegative women were enrolled in 1994-1995 and a second phase from 2001-2002
resulted in the addition of 737 HIV seropositive and 406 seronegative women to the
cohort. Participants are seen for visits every six months during which time blood and
other biological samples are obtained, an extensive questionnaire is administered and a
physical examination is performed. The characteristics of this cohort are comparable to
reported characteristics of HIV cases in the U.S. at the time of enrollment. A more
detailed description of the WIHS cohort is published elsewhere [28,29].
Race and ethnicity were collected at entry into the cohort. Participants indicated
which ethnic group they identify with from the following choices: non-Hispanic White,
Hispanic White, ‘other Hispanic’, non-Hispanic African-American, Hispanic African-
American, Asian/Pacific Islander, Native American/Alaskan and ‘other’. For purposes of

58
this study, we collapsed Hispanic White and ‘other Hispanic’ into one group and non-
Hispanic and Hispanic African-Americans into one group as a means to graphically
display the substructure results. The substructure analyses do not utilize any data other
than genomic DNA, and thus collapsing the self-reported ethnicities did not affect the
results.
Selection of Ancestry Informative Markers. A subset of markers that were
originally identified by Smith and colleagues was utilized to distinguish between West
African, European, East Asian, and Native American populations [30]. Briefly, Smith et
al. used a greedy algorithm to select 3,011 markers dispersed throughout the genome that
were maximally informative (Shannon information content > 0.035 out of a maximum of
0.709) for high-density admixture mapping [30,31]. These markers conformed to
Hardy-Weinberg equilibrium in West African and European founder populations, were
spaced 50 kb from each other, were not in linkage disequilibrium with one another or any
known human disease loci, and were similar in frequency for intracontinental populations
[30]. These criteria are widely accepted for the selection of markers for admixture
mapping [32,33]. Smith and colleagues [30] then identified four lists of 100 markers
(n=345 unique markers) that were optimal for distinguishing between West African,
European, East Asian, and Native American populations. The average difference in
allele frequencies (Fδ) for these 345 unique markers were 78% between West
African/European, 85% between West African/Native American, 56% between
European/Native American, and 57% for European/East Asian [30].

59
This panel of 345 markers was reduced to 185 markers that had a high probability
for successful genotyping on the Illumina Goldengate
®

platform and that were also
genotyped as part of The International HapMap Project [34]). In a study by Shtir et al.,
genotype data on the 185 markers were successfully used to estimate population
substructure for Hispanics in the Los Angeles Latino Eye Study (LALES) and a multi-
ethnic panel of participants from the Multiethnic Cohort (MEC), in which results
supported the separation of four unique clusters (European, West African, East Asian,
and Native American) [35].
Genotyping and Quality Control. From a total of 3,766 women enrolled in the
WIHS cohort, 2,437 women who consented to genetic studies for which DNA samples
were available were genotyped (65% of the cohort). Out of the 1,329 remaining
participants in the cohort, 356 did not consent to genetic studies and 973 consented, but
did not have an adequate amount of DNA available for genotyping.
Data for the 185 AIMs in the WIHS cohort were generated as a subset of a larger
panel (n= 384 single nucleotide polymorphisms [SNPs]). The raw data were edited using
Illumina BeadStudio

software to cluster the three genotypes (AA, AB, BB) at each locus.
Quality control was a multi-step process, in which the first round eliminated 24 of 384
SNPs (6%) for which three distinct genotypes were not observed. Next, 76 samples (3%)
were eliminated that had missing data for more than 20% of the remaining 360 SNPs.
Assay performance for the remaining 360 SNPs was then assessed and seven additional
SNPs (2%) were eliminated because they had a call rate < 90%. Lastly, 110 samples (5%)
were excluded because they had more than 5% missing data after the elimination of the

60
31 SNPs. Hardy-Weinberg equilibrium was tested within each self-identified ethnic
group; SNPs with p-values < 0.05 in any one population were visually reviewed for
presence of three distinct clusters for accurate genotype determination. Three SNPs did
not meet this criterion and were therefore excluded from the analysis. Duplicates for 53
participants were included and concordance was 99.9%.
In summary, 34 SNPs and 119 participants did not meet quality control criteria
and were excluded from the analysis. Of the remaining 350 SNPs, 168 were AIMs and
were included in the analysis (91% of the AIMs). The SNP assay failure rate was similar
for the non-AIMs genotyped (8.5% of the non-AIMs).
Characterization of population substructure and admixture
Structured association. Individual genetic ancestry proportions for 2,318 WIHS
women were inferred using the software package STRUCTURE 2.3.1 [11,24], which
employs a Bayesian Markov chain Monte Carlo (MCMC) clustering algorithm. A 30,000
repetition burn-in period and 10,000 subsequent iterations for different values of k
(number of assumed subpopulations, k=3-6) were initiated under the admixture model
with independent allele frequencies. Four independent simulations for each value of k
were performed to ensure that estimates were consistent across runs. The admixture
model with the greatest log likelihood for each value of k was selected.
Reference population genotype data was obtained from 11 known populations
included in HapMap2 and HapMap3 [34]; European ancestry individuals from Utah
(CEU), Yorubans from Nigeria (YRI), Chinese Han from Beijing (CHB), Japanese from
Tokyo (JPT), Chinese Han from Denver, Colorado (CHD), Mexicans from Los Angeles

61
(MEX), African Americans from the southwest, U.S. (ASW), Gujarati Indians from
Houston, Texas (GIH), Toscani from Tuscany, Italy (TSI), Maasai people from southern
Kenya (MKK), and Luhya people from Kenya (LWK). HapMap data were included in
the STRUCTURE analyses to increase the accuracy of admixture estimation [33]. The
MCMC scheme was run on two merged datasets: 1) WIHS (n=2,318) and HapMap2
(CEU, YRI, CHB, JPT, n=270) participants for 168 AIMs, and 2) WIHS (n=2,318) and
HapMap3 (all populations, n=1002) participants for the 105 AIMs included in the
HapMap3 dataset. Results were formatted and graphically displayed using the Distruct
1.1 software package [36].
Principal components analysis. Population substructure was also assessed with
PCA on the WIHS genotype data for 168 AIMS (n=2,318). Following the method used
with the EIGENSTRAT software [25], principal components were estimated using the
prcomp function of the R software package to detect the major axes of genetic variation
across hidden population structure in the WIHS.

RESULTS

A total of 2,318 WIHS participants who passed all QC procedures were included
in the analysis. The number of women at each site and self-reported ethnicity by HIV
seropositive (n=1,796) and HIV seronegative (n=522) status are given in Table 4.1. The
majority of seropositive (60.4%) and seronegative women (63.4%) reported African-
American ethnicity.

62
Table 4.1 Description of 2,318 women in the Women’s Interagency HIV Study (WIHS)
with ancestry informative marker data.

HIV+ HIV-
N (%) N (%)
N 1796 522
Site
Bronx, NY 316 (17.6) 146 (28.0)
Brooklyn, NY 371 (20.7) 120 (23.0)
Washington, DC 287 (16.0) 68 (13.0)
Los Angeles, CA 390 (21.7) 70 (13.4)
San Francisco, CA 189 (10.5) 77 (14.7)
Chicago, IL 243 (13.5) 41 (7.9)
Self-identified ethnicity
African American 1,085 (60.4) 331 (63.4)
Hispanic 418 (23.3) 117 (22.4)
White 240 (13.4) 51 (9.8)
Asian/Pacific Islander 20 (1.1) 7 (1.3)
Native American/Alaskan 7 (0.4) 5 (1.0)
Other 26 (1.4) 11 (2.1)

Since the AIMs selected for this study were part of a panel that was designed to
optimally distinguish four populations [30], we focus on the modeling results assuming
three and four source populations. Individual ancestry proportions by WIHS self-
reported ethnic groups for k=3 and k=4 cluster models are illustrated in Figure 4.1 (results
for self-reported Asian, Native American and ‘other’ groups are not illustrated due to
small numbers). Each model was run with HapMap2 (Figure 4.1A & B) and HapMap3
(Figure 4.1C & D) reference populations.

63
Using WIHS genotype data for 168 AIMs and HapMap2 ancestral population data
(CEU, YRI, CHB, JPT) to assess the population substructure of the WIHS, the estimated
log likelihood for the k=3 model (-449462.3) and the k=4 model (-449517.3) was
substantially greater than the likelihood for k=2 (-450089.7), thus indicating that
participants in the WIHS cohort likely descended from at least three source populations.
The k=3 model distinguishes European (green), Yoruban (pink) and Asian (purple)
ancestry components, color-coded for ease of comparison with corresponding HapMap2
populations (Figure 4.1A). Assuming k=4 subpopulations, the fourth subpopulation
(yellow) was represented in WIHS African-American and Hispanic women, but was not
substantially present in any of the HapMap2 populations (Figure 4.1B). The k=4 model
suggests that there is structure in the WIHS that is not represented in the HapMap2
populations.
The availability of 105 of the AIMS also genotyped in HapMap3 permitted the
comparison of population substructure found in the WIHS with additional reference
populations. The estimated log likelihood for k=4 (-358970.2) was greater than the
likelihood for k=3 (-365860.4), indicating that the assumption of four subpopulations was
more consistent with these data. The fourth subpopulation (orange) was present in WIHS
self-identified African-Americans (Figure 4.1C) in similar proportions to the fourth
subpopulation (yellow) in the k=4 model using HapMap2 populations (Figure 4.1B).
With the inclusion of additional African populations (Luhya and Maasai from Kenya),
the fourth subpopulation (orange) was over-represented in the Maasai, in which the
average proportion of ancestry attributed to the fourth subpopulation was 79% (Figure

64
4.1 D). Since the Maasai population show admixture with Yoruban ancestry (12%), the
fourth component was labeled as non-Yoruban ancestry.

Figure 4.1 Individual ancestry proportions for self-reported WIHS ethnic groups and
HapMap populations.
A. k=3, HapMap2 B. k=4, HapMap2 C. k=3, HapMap3 D. k=4, HapMap3
A B C D

65
The AIM panel [30] utilized for this study was characterized using one African
source population (HapMap2 Yorubans); however, our results partition two African
components as indicated after the inclusion of additional African source populations from
HapMap3 (Figure 4.1C & D). To investigate whether the fourth subpopulation (yellow)
represents African ancestry separate from Yoruban ancestry, the Yoruban component
from the k=3 model (Figure 4.1A) was plotted against the Yoruban plus non-Yoruban
ancestry components from the k=4 model (Figure 4.1B) on 168 AIMs in WIHS women
(Figure 4.2). There was a strong correlation (r
2
= 0.94) between the single African
component (k=3) and the combination of two separate African components (k=4). The
correlation is indicative of a broader clustering of African ancestry, which splits into two
distinct clusters when the number of assumed populations increases. Additionally, the
log likelihood was greater in the k=4 model, thus suggesting that we can detect two
unique African source populations in WIHS women using these markers.

Figure 4.2 Individual proportions of Yoruban ancestry (k=3) vs individual proportions of
Yoruban plus non-Yoruban ancestry (k=4) for self-identified WIHS African American
women (n=1,416).

R
2
= 0.9369
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
proportion of Yoruban ancestry (k =3)
proportion of Yoruban+ non-Yoruban ancestry (k =4)

66
To understand the partition of subpopulations utilizing this particular set of
markers, the difference in allele frequencies (Fδ) was computed between any two
populations in HapMap2 data over 168 AIMs and in HapMap3 data over 105 AIMs. The
average Fδ

and frequency of SNPs with differences of 0-0.1, 0.11-0.3, 0.31-0.50, and
greater than 0.50 are given for several HapMap3 population comparisons in Table 4.2.
As expected, the majority of SNPs had Fδ

greater than 0.50 between CEPH Europeans
and the three African populations. The Yoruban and Luhya populations share similar
allele frequencies across the 105 SNPs, which is depicted in relatedness (average Fδ

=
0.06). However, only 19 (18.1%) SNPs had Fδ between 0-0.1 in the Yoruban and Maasai
population comparison (average Fδ=0.17). The separation of two African components is
also supported by the dissimilarity in allele frequencies between the Maasai and Yoruban
populations across this particular panel of AIMs. These findings suggest that women in
the WIHS cohort likely share ancestry with Yoruban, non-Yoruban African, Asian, and
European populations, as supported by the k=4 models run with HapMap2 (Figure 4.1B)
and HapMap3 data (Figure 4.1D).

67
Table 4.2 Differences in allele frequencies across relevant HapMap3 populations for
105 ancestry informative markers.

No. (%) of SNPs with allele frequency difference (Fδ)
Population comparison* 0 - 0.1 0.11 - 0.3 0.31 - 0.50 0.51+ Avg.Fδ

YRI-LWK 92 (88) 12 (11) 0 1 (1) 0.06
YRI-MKK 19 (18) 79 (75) 6 (6) 1 (1) 0.17
YRI-CEU§ 0 0 7 (7) 98 (93) 0.68
YRI-MEX 0 11 (11) 24 (23) 70 (67) 0.58
YRI-Asian§ 9 (9) 14 (13) 29 (28) 53 (51) 0.50
LWK-MKK 34 (32) 69 (66) 2 (2) 0 0.13
LWK-Asian 10 (10) 15 (14) 33 (31) 47 (45) 0.47
LWK-MEX 4 (4) 12 (11) 23 (22) 66 (63) 0.54
LWK-CEU 0 0 11 (11) 94 (90) 0.64
MKK-Asian 12 (11) 33 (31) 31(30) 29 (28) 0.37
MKK-MEX 11 (11) 19 (18) 30 (29) 45 (43) 0.41
MKK-CEU 0 3 (3) 42 (40) 60 (57) 0.51
CEU-MEX 46 (44) 48 (46) 11 (11) 0 0.14
CEU-Asian§ 23 (22) 49(47) 23 (22) 10 (10) 0.24
MEX-Asian 50 (48) 38 (37) 13 (13) 3 (3) 0.15

*HapMap3 population abbreviations: CEU= CEPH European, Asian=Chinese Han Beijing and
Japanese Tokyo, MEX= Mexican Los Angeles, LWK=Luhya tribe Western Kenya, YRI=
Yoruban, Nigeria, MKK= Maasai tribe Kenya

§ Proportions of SNPs with Fδ
of 0-0.1, 0.11-0.3, 0.31-0.5, and 0.5+ in HapMap2 population
comparisons across 168 markers were not different than HapMap3 population comparisons (χ
2
, p
value > 0.1).

68
Admixture proportions in African-American, Hispanic, and non-Hispanic White
women were evaluated by the geographic location of the WIHS sites. The mean (± SD)
proportions of European, Yoruban, non-Yoruban, and Asian ancestry for the k=4
admixture model using 168 AIMs (Figure 4.1B) and 105 AIMs (Figure 4.1D) across the
six WIHS site locations are shown in Table 4.3. The proportions of Yoruban (0.29-0.41)
and non-Yoruban (0.48-0.56) ancestry in African-Americans using HapMap2 reference
populations were similar to proportions using HapMap3 reference populations (Yoruban
0.39-0.49, non-Yoruban 0.41-0.51). In contrast to African-Americans, ancestry
proportions in Hispanics varied across sites. There was a gradient of decreasing
European ancestry and increasing Asian ancestry in Hispanics from East to West.
Additionally, African admixture (Yoruban and non-Yoruban) was substantial in
Hispanics in New York (0.13-0.19) but minimal among Los Angeles Hispanics (0.03-
0.04).

69
Table 4.3 Average proportions of ancestry for self-identified African American,
Hispanic, and non-Hispanic White women by WIHS site for k=4 on 105 loci (first row)
and k=4 on 168 loci (second row).*

Estimated average ancestral proportions (± SD)
Self-identified ethnicity (N) European Yoruban Non-Yoruban Asian
African-American
Bronx, NY (282) 0.07 (0.13) 0.42 (0.29) 0.46 (0.28) 0.04 (0.07)
0.09 (0.13) 0.37 (0.25) 0.49 (0.26) 0.05 (0.07)
Brooklyn, NY (395) 0.05 (0.10) 0.49 (0.30) 0.41 (0.27) 0.04 (0.06)
0.07 (0.11) 0.41 (0.27) 0.48 (0.26) 0.04 (0.05)
Washington, DC (276) 0.07 (0.11) 0.39 (0.28) 0.50 (0.27) 0.05 (0.07)
0.10 (0.12) 0.32 (0.24) 0.54 (0.26) 0.04 (0.06)
Los Angeles, CA (120) 0.08 (0.13) 0.39 (0.26) 0.47 (0.26) 0.05 (0.07)
0.10 (0.13) 0.29 (0.21) 0.56 (0.25) 0.05 (0.08)
San Francisco, CA (131) 0.06 (0.09) 0.40 (0.28) 0.50 (0.28) 0.04 (0.05)
0.09 (0.10) 0.34 (0.24) 0.52 (0.25) 0.04 (0.05)
Chicago, IL (208) 0.05 (0.09) 0.43 (0.28) 0.48 (0.27) 0.04 (0.04)
0.07 (0.11) 0.35 (0.23) 0.55 (0.24) 0.04 (0.04)
Hispanic
Bronx, NY (146) 0.53 (0.23) 0.14 (0.12) 0.18 (0.14) 0.16 (0.16)
0.52 (0.21) 0.13 (0.10) 0.19 (0.16) 0.16 (0.15)
Brooklyn, NY (66) 0.53 (0.19) 0.14 (0.12) 0.17 (0.10) 0.16 (0.18)
0.51 (0.18) 0.13 (0.08) 0.19 (0.16) 0.17 (0.19)
Washington, DC (16) 0.41 (0.28) 0.08 (0.12) 0.13 (0.20) 0.39 (0.33)
0.37 (0.27) 0.07 (0.09) 0.12 (0.21) 0.43 (0.33)
Los Angeles, CA (234) 0.36 (0.24) 0.03 (0.07) 0.04 (0.08) 0.57 (0.27)
0.31 (0.22) 0.03 (0.05) 0.04 (0.08) 0.62 (0.25)
San Francisco, CA (40) 0.46 (0.30) 0.03 (0.04) 0.06 (0.18) 0.46 (0.31)
0.44 (0.29) 0.03 (0.05) 0.06 (0.15) 0.48 (0.31)

70
Table 4.3, continued

Estimated average ancestral proportions (± SD)
Self-identified ethnicity (N) European Yoruban Non-Yoruban Asian
Hispanic
Chicago, IL (33) 0.46 (0.27) 0.08 (0.14) 0.07 (0.08) 0.38 (0.33)
0.40 (0.28) 0.08 (0.12) 0.08 (0.11) 0.44 (0.35)
Non-Hispanic White
Bronx, NY (27) 0.92 (0.08) 0.02 (0.03) 0.02 (0.03) 0.04 (0.04)
0.91 (0.07) 0.02 (0.03) 0.03 (0.03) 0.04 (0.03)
Brooklyn, NY (18) 0.89 (0.12) 0.03 (0.03) 0.03 (0.04) 0.05 (0.10)
0.89 (0.13) 0.03 (0.03) 0.03 (0.03) 0.06 (0.11)
Washington, D.C. (49) 0.91 (0.19) 0.01 (0.02) 0.04 (0.14) 0.04 (0.08)
0.91 (0.17) 0.03 (0.09) 0.02 (0.07) 0.04 (0.06)
Los Angeles, CA (91) 0.94 (0.12) 0.01 (0.02) 0.01 (0.03) 0.04 (0.11)
0.93 (0.12) 0.01 (0.02) 0.01 (0.03) 004 (0.11)
San Francisco, CA (66) 0.95 (0.08) 0.01 (0.01) 0.01 (0.01) 0.03 (0.07)
0.94 (0.08) 0.01 (0.01) 0.01 (0.01) 0.03 (0.08)
Chicago, IL (40) 0.95 (0.06) 0.01 (0.02) 0.01 (0.03) 0.02 (0.03)
0.95 (0.06) 0.01 (0.02) 0.01 (0.03) 0.01 (0.03)

DISCUSSION

The AIMs genotyped in the WIHS participants identified heterogeneity within
each self-reported ethnic group and across WIHS sites, demonstrating the importance of
* Average proportion of ancestry and standard deviations in labeled subgroups
from k=4 results (first row) using 105 markers with HapMap3 reference populations and k=4
results (second row) using 168 markers and HapMap2 reference populations.

71
using AIMs to control for population stratification when disease rates and genetic
markers are both associated with subpopulations; such as they do for many phenotypes
including susceptibility to HIV [37], drug toxicity [38], and co-morbid conditions
common in HIV infection, such as type-2 diabetes [39]. This set of markers was
adequate to estimate individual proportions of ancestry from broad continental groups,
such as European, African, and Asian. In contrast, the panel was not able to distinguish
Native American and East Asian ancestry. However, the modest set of markers was able
to detect two African subpopulations in a predominately African-American cohort.
Obtaining a finer resolution of ancestry to control for population stratification may be
necessary among those closely related subpopulations in which disease rates vary.
Ancestry from African source populations constituted 77%-83% of African-
Americans’ genetic makeup with admixture from European (5%-14%) and Asian (4%-
7%) source populations in the WIHS participants sampled. There was an increasing
gradient of Asian ancestry in Hispanics from the East to the West WIHS sites, which
ranged from 16%-63%. More subtle diversity in the genetic ancestry of non-Hispanic
Whites, where European ancestry ranged from 89%-95% across sites, was observed.
Population substructure in the WIHS is similar to reports of genetic ancestry in
admixed populations in the U.S. For example, several studies of African-American
populations across the U.S. [21,23,30,41] have estimated African ancestry between 75%-
83%, which is consistent with our findings. Similar to WIHS Hispanics, Halder and
colleagues [23] observed an east/west gradient of increasing Native American and
decreasing African heritage in a sample of self-identified Hispanics from the east coast

72
(Farmington, CT) and west coast (Southern California). Shared genetic membership
from the cluster representing Asian ancestry in WIHS Hispanics and HapMap3 Mexicans
(Fig. 4.2B) was observed. This can be attributed to evidence of genetic similarity
between Native Americans and Siberians, as identified by Wang et al. [42]. Given the
modest panel of AIMs utilized in this study and the relatively small difference in allele
frequencies between Asian and Mexican HapMap3 reference populations (average Fδ

=0.15, Table 4.2), Native American ancestry is represented within the Asian ancestry
cluster.
An ancestral component in African-American and Hispanic women in the k=4
model that did not correspond to any HapMap2 reference population was observed. It
appears that the unidentified ancestral component from the k=4 model using HapMap2
source populations emerged from the Yoruban component in the k=3 model (Fig. 2).
Subsequent analysis incorporating HapMap3 reference population data using 105
markers indicated an African ancestry component represented in the Maasai that was not
appreciably represented in Yorubans. These results may be driven by a selection bias in
the 105 markers used for analysis as indicated by the dissimilarity in allele frequencies
for these markers between the Yoruban and Maasai reference populations (average Fδ.=
0.17). The remaining 63 SNPs genotyped in WIHS women were not available in the
HapMap3 database. Thus, it is unclear whether the 105 markers included in the analysis
consisted of a disproportionate number of SNPs with allele frequency differences greater
than 0.10 between the Yoruban and Maasai populations.

73
However, the separation of two African source populations using this panel of
markers, although unintended, revealed additional population substructure in a cohort of
women who are predominately African-American. The finding of two African
components is consistent with the extreme ancestral diversity across Africa, as reported
by Tishkoff and colleagues [40]. Tishkoff et al. genotyped 1,327 markers on 2,432
Africans from 113 geographic locations, revealing population structure consisting of 14
ancestral components and thus our identification of two African components is not
surprising. While we believe that this structure exists in our sample, the suggested
selection bias of the SNPs potentially resulted in an over-estimation of the non-Yoruban
component for all individuals.
The interpretation of an individual’s global ancestry using a highly selective,
modest number of markers and without a diverse panel of reference populations, is
difficult. The naming of the fourth component as “non-Yoruban” is preliminary, as the
reference populations included in the analysis can influence the clustering algorithm. It
merits emphasis that the primary goal of estimating individual ancestry is to control for
population stratification in genetic association studies. Considering this, the issue of
whether to include three or four ancestral components from a STRUCTURE analysis in
statistical models to address population stratification is dependent on the ability to control
for resulting confounding bias, if disease rates and disease risk factors differ substantially
between Yorubans and other African populations.
The observation of substantial diversity based on AIMs in the WIHS underscores
the importance of providing a measure of genetic ancestry in all genetic association

74
studies. Even within our non-Hispanic White population, we observed up to 11% non-
European ancestry and thus even studies of only Whites are potentially subject to residual
confounding when genetic ancestry is not taken into account when testing genetic
associations. Future WIHS reports will incorporate PCs to ensure that results generated
from this cohort are not subject to population stratification.

CHAPTER REFERENCES

1. Lohmueller KE, Pearce CL, Pike M, Lander ES, Hirschhorn JN. Meta-analysis of
genetic association studies supports a contribution of common variants to
susceptibility to common disease. Nat Genet 2003;33(2):177-82.
2. Ioannidis JP. Non-replication and inconsistency in the genome-wide association
setting. Hum Hered 2007;64:203-213.
3. An P, Winkler CA. Host genes associated with HIV/AIDS: advances in gene
discovery. Trends Genet 2010;26(3):119-31.
4. O’Brien SJ, Nelson GW. Human genes that limit AIDS. Nat Genet
2006;36(6):565-574.
5. Lama J, Planelles V. Host factors influencing susceptibility to HIV infection and
AIDS progression. Retrovirology 2007;4:52.

75
6. Bushman FD, Malani N, Fernandes J, D’orso I, Cagney G, Diamond TL, Zhou H,
Hazuda DJ, Espeseth AS, Konig R, Banyopadhyay S et al. Host cell factors in
HIV replication: meta-analysis of genome-wide studies. PloS Pathogens
2009;5:e1000437.
7. Vidal F, Gutierrez F, Gutierrez M, Olana M, Sanchez V, Mateo G, Peraire J,
Vilades C, Veloso S, Lopez-Dupla M, Domingo P. Pharmacogenetics of adverse
effects due to antiretroviral drugs. AIDS Rev 2010 12:15-30.
8. Thomas DC, Witte JS. Point: population stratification: a problem for case-control
studies of candidate-gene associations? Cancer Epidemiol Biomarkers Prev
2002;11(6):505-512.
9. Gorroochurn P, Hodge SE, Heiman G, Greenberg DA. Effect of population
stratification on case-control association studies. II False-positive rates and their
limiting behavior as number of subpopulations increases. Hum Hered
2004;58(1):40-48.
10. Devlin B, Roeder K, Wasserman L. Genomic control, a new approach to genetic-
based association studies. Theor Popul Biol 2001;60:155-166.
11. Pritchard JK, Stephens M, Donnelly P. Inference of population structure using
multilocus genotype data. Genetics 2000;155:945-959.
12. Pritchard JK, Donnelly P. Case-control studies of association in structured or
admixed populations. Theor Popul Biol 2001;60:227-237.
13. Voight BF, Pritchard JK. Confounding from cryptic relatedness in case-control
associations. PloS Genet 2005;1(3):e32.

76
14. Helgason A, Yngvafottir B, Hrafnkelsson B, Gulcher J, Stefansson K. An
Icelandic example of the impact of population structure on association studies.
Nat Genet 2005;37(1):90-95.
15. Price AL, Helgasson A, Palsson S, Stefansson H, St Clair D, Andreassen OA,
Reich D, Kong A, Stefansson K. The impact of divergence time on the nature of
population structure: an example from Iceland. PloS Genet 2009;5(6):e1000505.
16. Campbell DC, Ogburn EL, Lunetta KL, Lyon HN, Freedman ML, Groop LC, et
al. Demonstrating stratification in a European American population. Nat Genet
2005;37(8):868-872.
17. Hirschhorn JN, Daly MJ. Genome-wide association studies for common disease
and complex traits. Nat Rev Genet 2005;6(2):95-108.
18. McCarthy MI, Abecasis GR, Cardon LR, Goldstein DB, Little J, Ioannidis JP,
Hirschhorn JN. Genome-wide association studies for complex traits: consensus,
uncertainty and challenges. Nat Rev Genet 2008;9(5):356-369.
19. Freedman ML, Reich D, Penney KL, McDonald GJ, Mignault AA, Patterson N, et
al. Assessing the impact of population stratification on genetic association studies.
Nat Genet 2004;36(4):388-393.
20. Hoggart CJ, Parra EJ, Shriver MD, Bonilla C, Kittles RA, Clayton DG,
McKeigue PM. Control of confounding of genetic associations in stratified
populations. Am J Hum Genet 2003;72:1492-1504.

77
21. Parra EJ, Marcini A, Akey J, Martinson J, Batzer MA, Cooper R et al. Estimating
African American admixture proportions by use of population-specific alleles.
Am J Hum Genet 1998;63:1839-1851.
22. Salari K, Choudhry S, Tang H, Naqvi M, Lind D, Avila PC, et al. Genetic
admixture and asthma related phenotypes in Mexican American and Puerto Rican
asthmatics. Genet Epidemiol 2005;29:76-86.
23. Halder I, Yang BZ, Kranzler HR, Stein MB, Shriver MD, Gelernter J.
Measurement of admixture proportions and description of admixture structure in
different U. S. populations. Hum Mutat 2009;30(9):1299-1309.
24. Falush D, Stephens M, Pritchard JK. Inference of population structure using
multilocus genotype data: linked loci and correlated allele frequencies. Genetics
2003;164:1567-1587.
25. Price AL, Patterson NJ, Plenge RM, Weinblatt ME, Shadick NA, Reich D.
Principal components analysis corrects for stratification in genome-wide
association studies. Nat Genet 2006;38(8):904-909.
26. Price AL, Zaitlen NA, Reich D, Patterson N. New approaches to population
stratification in genome-wide association studies. Nat Rev Genet 2010;11:459-
463.
27. Engelhardt BE and Stephens M. Analysis of population structure: a unifying
framework and novel methods based on sparse factor analysis. PloS Genet
2010;6(9) e1001117.

78
28. Barkan SE, Melnick SL, Preston-Martin S, Weber K, Kalish LA, Miotti P, et al.
The Women’s Interagency HIV Study. Epidemiology 1998;9:117-25.
29. Bacon MC, Von Wyl V, Alden C, Sharp G, Robison E, Hessol N, et al. The
Women's Interagency HIV Study: an observational cohort brings clinical sciences
to the bench. Clin Diagn Lab Immunol 2005;12(9): 1013-19.
30. Smith MW, Patterson N, Lautenberger JA, Truelove AL, McDonald GJ,
Waliszewska A, et al. A high-density admixture map for disease gene discovery
in African Americans. Am J Hum Genet 2004;74(5):1001-1003.
31. Patterson N, Hattangadi N, Lane B, Lohmueller KE, Hafler DA, Okenberger JR,
et al. Methods for high-density admixture mapping of disease genes. Am J Hum
Genet 2004;74(5):979-1000.
32. Rosenberg N, Li LM, Ward R, Pritchard JK. Informativeness of genetic markers
for inference of ancestry. Am J Hum Genet 2003;73(6):1402-1422.
33. Kosoy R, Nassir R, Tian C, White PA, Butler LM, Silva G, et al. Ancestry
informative marker sets for determining continental origin and admixture
proportions in common populations in America. Hum Mutat 2009; 30(1):69-
78.33.
34. The International HapMap Consortium. The International HapMap Project.
Nature 2003;426:789-796.
35. Shtir CJ, Marjoram P, Azen S, Conti DV, Le Marchand L, Haiman CA, et al.
Variation in genetic admixture and population structure among Latinos; the Los
Angeles Latino eye study (LALES). BMC Genetics 2009; 10:71-84.35.

79
36. Rosenberg, NA. Distruct: a program for the graphical display of population
structure. Molecular Ecology Notes 2004;4:137-138.
37. Telenti A, McLaren P. Genomic approaches to the study of HIV-1 acquisition. J
Infect Dis. 2010;202 Suppl 3:S382-386.
38. Tedaldi EM, Absalon J, Thomas AJ, Shlay JC, van den Berg-Wolf M. Ethnicity,
race and gender. Differences in serious adverse events among participants of an
antiretroviral initiation trial CPCRA058. J Acquir Immune Defic Syndr.
2008;47:441-448.
39. Shai I, Jiang R, Manson JE et al. Ethnicity, obesity, and risk of type 2 diabetes in
women: a 20 year follow-up study. Diabetes Care 2006;29(7):1585-1590.
40. Tishkoff SA, Reed FA, Friedlander FR, Ehret C, Ranciaro A, Froment A et al.
The genetic structure and history of Africans and African Americans. Science
2009;324:1035-1044.
41. Reiner AP, Carlson CS, Ziv E, Iribarren C, Jaquish CE, Nickerson DA.
Population structure, admixture, and disease-related traits in African American
participants in the CARDIA study. Hum Genet 2007;121:565-575.
42. Wang S, Lewis CM, Jakobsson M, et al. Genetic Variation and Population
Structure in Native Americans. PloS Genet 2007;3(11):e185.

80
Chapter 5: Genetic Variation in CYP2B6 Predicts Virologic Response to NNRTIs

ABSTRACT

Background. CYP2B6 variation predicts the pharmacokinetic characteristics of its
substrates. Consideration for underlying genetic structure is critical to protect against
spurious associations with the highly polymorphic CYP2B6 gene. The effect of variation
in CYP2B6 on response to its substrates, the nonnucleoside reverse transcriptase
inhibitors (NNRTIs), was explored in the Women’s Interagency HIV Study (WIHS).
Methods. Five CYP2B6 putative functional polymorphisms were tested for associations
with virologic suppression within one year after NNRTI initiation in women naïve to
antiretroviral agents (n=91). Simultaneously, principal components (PCs) were generated
to explain population substructure. Logistic regression was used to test the joint effect of
rs3745274 and rs28399499, which comprise the CYP2B6 metabolizer phenotype
indicating slow, intermediate and extensive metabolizers.
Results. Rs3745274 was significantly associated with virologic suppression (OR=3.61,
95% CI 1.16-11.22, p trend =0.03); the remaining polymorphisms tested were not
significantly associated with response. Women classified as intermediate and slow
metabolizers were 2.90 (95% CI 0.79-12.28) and 13.44 (95% CI 1.66-infinity) times as
likely to achieve virologic suppression compared to extensive metabolizers after
adjustment for PCs (p trend =0.005). Failure to control for genetic ancestry resulted in

81
substantial confounding of the relationship between the metabolizer phenotype and
treatment response.
Discussion. The CYP2B6 metabolizer phenotype was significantly associated with
response to NNRTIs, a relation that would have been masked by simply adjusting for
self-reported race/ethnicity. Given the appreciable genetic heterogeneity that exists
within self-reported ethnicity, these results exemplify the importance of characterizing
underlying genetic structure in pharmacogenetic studies. Further follow-up of the
CYP2B6 metabolizer phenotype is warranted given the potential clinical importance of
this finding.

INTRODUCTION

Genetic variation in pathways of drug absorption, delivery, metabolism and
excretion (ADME) contributes to the variability in pharmacokinetic parameters of drug
levels and clearance in plasma or cells over time, such as half life or area under the curve
(AUC) [1-2]. Pharmacokinetic characteristics are often associated with clinical
endpoints, such as fasting glucose levels after exposure to anti-diabetic medications or
viral suppression after exposure to antiretroviral agents [1,2]. One of the most robust
examples of a pharmacokinetic association is observed with genetic variation in the
cytochrome p450 2B6 gene (CYP2B6) [1-2], whose protein product is an enzyme that
metabolizes an array of medications [3], including bupropion, tamoxifen, propofol, and
the non-nucleoside reverse transcriptase inhibitors (NNRTI) efavirenz and nevirapine that

82
are used to treat human immunodeficiency virus (HIV) infection. Several CYP2B6 loss-
of-function alleles are associated with pharmacokinetic characteristics of NNRTIs [4,5].
Particularly, consistent associations of two single nucleotide polymorphisms (SNP),
rs3475274 and rs28399499, have been observed with plasma and intracellular NNRTI
levels [4-12] and to a lesser degree with clinical response [9-12]. The CYP2B6
metabolizer phenotype, derived from the composite genotype for rs3745274 and
rs28399499, categorizes individuals as extensive, intermediate, and slow metabolizers of
CYP2B6 substrates [13,14]. An association between the metabolizer phenotype and
efavirenz pharmacokinetics was reported [13] and an association with virologic response
in African-Americans has been suggested [14].
Confounding by race/ethnicity is a concern in all genetic association studies and
the two variants comprising the CYP2B6 metabolizer phenotype may be particularly
susceptible to this type of confounding. Substantial differences in minor allele
frequencies for these two SNPs occur across racial/ethnic groups [5-8] and race/ethnicity
has often been associated with pharmacokinetic parameters and treatment responses [15].
Gross confounding by race/ethnicity (population stratification) can be addressed by
adjusting statistical models for self-reported ethnicity, but cryptic relatedness and
admixture require more detailed genetic ancestry data to guard against false positive or
false negative results [16-20]. Such cryptic relatedness occurs in seemingly homogenous
populations within the same self-reported race/ethnicity group [20]. In theory, affected
individuals (cases) may be more related to one another than controls and as a result the
allelic distribution among cases may be skewed [20].

83
Spurious associations can also occur in a study population that contains
individuals from a single ethnic group that has experienced recent genetic admixture
[21,22]. Varying proportions of ancestry from distinct ‘parental’ populations contribute
to a population’s genome; admixture describes this mosaic of genetic ancestry.
Admixture is particularly evident in African-American and Hispanic-American
populations [23-25]. For example, among African-Americans from different
geographical regions in the United States (U.S.), European ancestry varies between
11.6% and 22.5%; the remaining proportion includes African and Native American
ancestry [23]. Thus, addressing population substructure in genetic association studies
beyond adjustment for self-reported race/ethnicity is critical to protect against potentially
spurious results.
The Women’s Interagency HIV Study (WIHS) is a multi-ethnic cohort study of
HIV-infected and uninfected women who have been followed every six months since
1995. We report here the association between genetic variation in CYP2B6 and virologic
response to initial antiretroviral therapy regimens containing a NNRTI. Simultaneously,
population substructure was characterized using a panel of ancestry informative markers
(AIM) selected to distinguish between European, West African, Asian, and Native
American ancestry [26]. CYP2B6 genetic associations were used to demonstrate that
confounding can be introduced when underlying genetic structure is ignored.

84
METHODS

WIHS Study Design. The WIHS is a prospective cohort study of HIV-infected
women and a comparison group of HIV-uninfected women. Participants were recruited
from six sites across the U.S., located in Bronx/Manhattan, New York; Brooklyn, New
York; Washington, D.C.; Los Angeles, California; San Francisco/Bay Area, California;
and Chicago, Illinois. Enrollment was conducted in two phases; 2,054 HIV seropositive
and 569 seronegative women were enrolled in 1994-1995 and a second enrollment from
2001-2002 resulted in the addition of 737 HIV seropositive and 406 seronegative women
to the cohort. A more detailed description of the WIHS cohort has been published
elsewhere [27,28].
Inclusion and Exclusion Criteria. HIV-infected women who consented to
genetic studies, initiated antiretroviral therapy containing a NNRTI during study follow-
up, maintained a ≥ 3 antiretroviral drug regimen at the subsequent visit, had HIV RNA
measurements at the visit immediately prior to, the visit of, and the visit immediately
after initiating an NNRTI-containing regimen were eligible for this study. Women who
were treated with antiretroviral drugs prior to reporting treatment with a NNRTI were
excluded to limit the inclusion of women infected with a virus that acquired resistance to
antiretroviral agents.
Virologic Response. HIV viral loads were quantified with an assay with a lower
limit of detection of 80 HIV RNA copies/ml. A positive virologic response
(“responders”) was defined as achievement of an undetectable viral load at the visit of

85
during which the NNRTI regimen was first reported or at the visit subsequent to it, which
corresponds to a maximum of 54 weeks of treatment. Participants not achieving an
undetectable viral load at the visit of first reported NNRTI regimen use (in the previous
six months) or subsequent visit were classified as “non-responders”. This definition
conforms to clinical guidelines for treatment with antiretroviral regimens [29].
SNP Selection. Polymorphisms in CYP2B6 were selected based on putative
function (located in exons, 3’ UTR, 5’ UTR and splice sites) and a minor allele frequency
(MAF) of 5% or higher, published pharmacogenetic associations, and assay performance
on Illumina’s GoldenGate
®
genotyping platform (San Diego, CA). A total of 11 CYP2B6
SNPs were selected: rs3211370, rs3211397, rs36079186, rs34128717, rs3211374,
rs3211371, rs8192709, rs34097093, rs1042389, rs3745274, and rs28399499.
Genotyping and Quality Control. Data for the eleven CYP2B6 exonic SNPs
were generated as a subset of a larger genotyping panel (n=384 SNPs). Four of the eleven
CYP2B6 SNP assays (rs3211370, rs3211397, rs36079186, rs34128717) failed to generate
three distinct genotypes and were excluded from the analysis. Of the remaining seven
SNPs that were successfully genotyped, rs3211374 was monomorphic in all race/ethnic
groups and thus, excluded from the analysis. In summary, six CYP2B6 polymorphisms
(rs3211371, rs8192709, rs34097093, rs1042389, rs3745274, and rs28399499) passed the
QC criterion of successful genotyping on 89.5% of the samples.

86
Characterization of Genetic Ancestry in the WIHS
Selection of AIMs, Genotyping and Quality Control. A subset of markers that
were originally identified by Smith and colleagues was utilized to distinguish between
West African, European, East Asian, and Native American populations [26]. A total of
185 AIMs were selected for genotyping. All subjects for the WIHS cohort who
consented to genetic studies and had DNA available were characterized for genetic
ancestry as these markers are being used for all genetic association studies being
conducted in the WIHS. A genotyping panel of 384 SNPs was developed for the
Illumina GoldenGate platform (San Diego, CA), which included 185 AIMs along with
199 SNPs in genes of broad interest in the WIHS. Overall, 34 of the 384 SNPs did not
meet quality control criteria, which included 17 AIMs. This left 168 of the 185 SNPs
available for the genetic ancestry analysis. A total of 119 participants did not meet QC
criteria and were excluded from the genetic ancestry analysis.
Estimation of Genetic Ancestry Components. Individual genetic ancestry
proportions for 2,318 WIHS women were inferred using the software package
STRUCTURE 2.3.1 [18,30], which employs a Bayesian Markov chain Monte Carlo
clustering algorithm. A 30,000 repetition burn-in period and 10,000 subsequent iterations
for different values of k (number of assumed subpopulations, k=3-6) were initiated under
the admixture model with independent allele frequencies. Four independent simulations
for each value of k were performed to ensure that estimates were consistent across runs.
The admixture model with the greatest log likelihood for each value of k was selected.
HapMap2 and HapMap3 [31] reference population data on 168 AIMs and 105 AIMs,

87
respectively, were included in the STRUCTURE analyses to increase the accuracy of
admixture estimation [32]. Results were formatted and graphically displayed using the
Distruct 1.1 software package [33].
Genetic ancestry components were also evaluated with principal components
analysis on the WIHS genotype data for 168 AIMs (n=2,318) following the method used
with the EIGENSTRAT software [34,35]. Adjusting for PCs is the preferred method to
control for population substructure, as the model does not depend on an assumption of
the number of source populations [34,35]. Genetic ancestry PCs were used in the models
examining the association between CYP2B6 genotypes and virologic response to therapy.
Statistical Analysis. Logistic regression was used to test associations between
each CYP2B6 polymorphism and achievement of an undetectable viral load up to 54
weeks after first reported NNRTI-based regimen. Odds ratios (OR) per allele and 95%
CIs were estimated by modeling the genotypes as an ordinal variable, where common
allele homozygotes, heterozygotes and minor allele homozygotes were coded as 0, 1, and
2, respectively. This log-additive model provides a p-value for corresponding test of the
trend for increased probability of virologic response per allele.
CYP2B6 metabolizer phenotypes were constructed using two polymorphisms,
rs3745274 and rs28399499, to test the association between the metabolizer phenotype
and virologic response. Women who were common allele homozygotes at rs3745274
and rs28399499 (GG and TT, respectively) were coded as 0 “extensive metabolizers”.
Women with one heterozygote genotype and one common allele homozygote genotype at
either polymorphism were coded as 1 “intermediate metabolizers”. Women with a total

88
of two minor alleles (one minor allele homozygote genotype, or two heterozygote
genotypes) across both SNPs were coded as 2 “slow metabolizers”. No women carried
one minor allele at one SNP and two minor alleles at the other SNP, or four minor alleles
across the two SNPs.
Metabolizer phenotype-specific ORs and 95% CIs for intermediate metabolizers
and slow metabolizers compared with extensive metabolizers, were estimated with exact
logistic regression, since there were zero non-responders with the slow metabolizer
phenotype. Additionally, the metabolizer phenotype was treated as an ordinal variable to
obtain the exact p for trend.
To assess the potential confounding effects of population substructure, models
were fit unadjusted, adjusted for self-reported race/ethnicity (Non-Hispanic White,
African American, Hispanic, and Asian/Other), and adjusted for genetic ancestry
principal components (PCs). The three most important PCs that accounted for the
largest change in the main effect β in the individual SNP analyses were adjusted for in
the metabolizer phenotype model.
Self-reported adherence was also evaluated as a potential confounder (change in
the genotype main effect β of 10% or more was considered confounding). Adherence
data were taken at the visit at which the participant achieved the virologic response
outcome since the adherence variable at this visit reflects treatment adherence in the six
months leading up to the visit in which the outcome was achieved. For modeling
purposes, adherence was dichotomized as ≥ 95% adherent and < 95% adherent from
original categories as collected (100%, 95-99%, 75-95%, <75%).

89
RESULTS

Detection of population substructure. Figure 4.1B and 4.1D in the previous
chapter illustrate the individual ancestry proportions for 2,318 WIHS participants
(n=1,796 HIV seropositive and n=522 seronegative) by self-reported ethnic groups for
k=4 cluster STRUCTURE models (results for self-reported Asian, Native American and
‘other’ groups are not illustrated due to small numbers). Admixture was estimated by
including reference population data from HapMap2 (Figure 4.1B) and HapMap3 (Figure
4.1D). Assuming k=4 subpopulations (log likelihood= -449517.3), the fourth
subpopulation (yellow) was represented in WIHS African-American and Hispanic
women, but was not substantially present in Yoruban, European, or East Asian reference
populations (Figure 4.1B). The k=4 model suggests that there is ancestry in the WIHS
that is not represented in the HapMap2 populations. Ancestry characterization was also
performed assuming k=4 subpopulations (log likelihood= -358970.2) with HapMap3 data
which include additional reference populations (Figure 4.1D). The fourth subpopulation
(orange) present in WIHS African-American and Hispanic women was over-represented
in the Maasai Kenyans, in which the average proportion of ancestry attributed to the
fourth subpopulation was 79% (Figure 4.1D). Since the Maasai population show
admixture with Yoruban ancestry (12%), the fourth component was labeled as non-
Yoruban African ancestry. The HapMap3 Mexican-American population shows a
considerable amount of admixture represented by the Asian subpopulation, which was
inferred to represent Native American ancestry as Native American ancestry is correlated

90
with East Asian ancestry [36]. In summary, the WIHS women likely descended from
four source populations, European, Asian, Yoruban, and non-Yoruban African ancestry.
The estimated admixture proportions in self-reported African-American,
Hispanic, and non-Hispanic White women were evaluated by the geographic location of
the WIHS sites (results not shown). In contrast to African-Americans and Whites,
ancestry proportions in Hispanics varied across sites. There was a gradient of decreasing
European ancestry and increasing Asian ancestry in Hispanics from East to West.
Additionally, African admixture (Yoruban and non-Yoruban) was substantial in
Hispanics in New York (0.13-0.19) but minimal among Los Angeles Hispanics (0.03-
0.04).
While estimating ancestry proportions using STRUCTURE is beneficial to
labeling subpopulations for ease of interpretation, PCs are preferred as ancestry
covariates in statistical models. Graphical display of the PCs illustrate the separation of
broad ethnic groups and the dispersion of individuals along the axes confirms the
variation that exists within self-reported ethnic groups as seen in Figure 4.1. Figure 5.1
shows the plot of PC1 vs. PC2 with color-coded self-reported ethnic groups. PC1 (x
axis) separates African vs non-African ancestry and PC2 (y axis) separates Native-
American/Hispanic vs non-Hispanic ancestry.

91
Figure 5.1 Principal component (PC) 1 vs. PC 2 for WIHS women (n=2,318) from
analysis on 168 ancestry informative markers. Self-identified race/ethnicity groups
are color-coded.

CYP2B6 associations and impact of population substructure. A total of 91
antiretroviral naïve women initiated a NNRTI-based regimen; 21 were virologic non-
responders. Nine non-responders (43%) had a nadir CD4+ count lower than 200 cells/ml
prior to reporting a NNRTI-based regimen, a difference that was not statistically
significantly different from responders (N=24/70, 34%, p=0.47). The median viral load
measured at the visit prior to first report of a NNRTI was 28,000 copies/ml (IQR 6,300-
110,000) for responders and 45,000 copies/ml (IQR 13,000-69,000) for non-responders
(p=0.57). Nearly all of the non-responders (n=18/21) reported having African-American
race/ethnicity and the remaining three women reported having Hispanic race/ethnicity.
-9
-7
-5
-3
-1
1
3
5
-12 -10 -8 -6 -4 -2 0 2 4 6 8 10
PC2
PC1
Non-Hispanic White African-American Hispanic Asian Native American/Alaskan other

92
Among the responders (N=70), 53% self-reported as African-American, 31% as White,
and 7% as Hispanic.
The genotype frequencies by responder status and MAF by self-reported
ethnicity are given in Table 5.1 for the six SNPs that were successfully genotyped. There
were no non-responders among carriers of the minor allele for rs3211371 and therefore
effect estimates could not be calculated.
Table 5.1. Frequency counts for genotypes by virologic responder status
1
and minor
allele frequencies by self-reported race/ethnic group in 91 HIV-infected women treated
with NNRTIs (efavirenz and nevirapine).

Minor Allele Frequency
CYP2B6 SNP
2
Responders
(%)
Non-
responders
(%)
African
American
(n=55)
White
(n=22)
Hispanic
(n=8)
Other
3

(n=6)
rs3211371 0.03 0.07 0 0
CC 63 (90) 21 (100)
CT 6 (10) 0
TT 0 0
missing 1 0
rs3409703 0.22 0 0 0.08
TT 52 (76) 14 (74)
TC 16 (24) 4 (21)
CC 0 1 (5)
missing 2 2
rs1042389 0.25 0.14 0.21 0.08
TT 46 (66) 13 (62)
TC 21 (30) 7 (33)
CC 3 (4) 1 (5)
rs8192709 0.06 0 0.07 0
CC 64 (91) 20 (95)
CT 6 (9) 1 (5)
TT 0 0

93
Table 5.1, continued

Minor Allele Frequency
CYP2B6 SNP
2

Responders
(%)
Non-
responders
(%)
African
American
(n=55)
White
(n=22)
Hispanic
(n=8)
Other
3

(n=6)
rs3745274 0.28 0.24 0.21 0.33
GG 32 (46) 14 (67)
GT 34 (48) 7 (33)
TT 4 (6) 0
rs28399499 0.10 0 0 0
TT 62 (88) 19 (90)
TC 7 (10) 2 (10)
CC 1 (2) 0

1 Virologic response defined as achievement of undetectable viral load up to 54 weeks after self-
reported initiation of NNRTI-based regimen
2 SNP= single nucleotide polymorphism; all SNPs are missense with the exception of rs1042389
which is located in the 3’ untranslated region
3 Asian, Native American, Other

The associations between the remaining five SNPs (rs3409703, rs1042389,
rs8192709, rs3745274 and rs28399499) and response to a NNRTI-based regimen are
shown in Table 5.2. Overall, no statistically significant associations between these five
SNPs and virologic response were observed after adjustment for self-reported
race/ethnicity (second column). However, a high magnitude of effect on virologic
response was observed for the two SNPs which comprise the metabolizer phenotype
(rs3745274 and rs28399499) after adjustment for genetic ancestry PCs and the effect of
rs3745274 reached statistical significance (p trend =0.03).

94
Table 5.2 Associations between CYP2B6 single nucleotide polymorphisms
(SNP) and with virologic response
1
to NNRTI-based regimens in women who
were naïve to antiretroviral drugs (N=91).

Crude
Adjusted for
SR Race/Ethnicity
3

Adjusted for
top three PCs
4
CYP2B6 SNP
2
OR (95% CI) OR (95% CI) OR (95% CI)
rs34097093
per C allele 0.70 (0.25-1.99) 1.21 (0.39-3.77) 1.25 (0.37-4.22)
p trend 0.50 0.74 0.72
rs1042389
per C allele 0.88 (0.38-2.03) 1.14 (0.46-2.84) 1.13 (0.43-2.94)
p trend 0.76 0.77 0.80
rs8192709
5

CC 1.00 1.00 1.00
CT 1.87 (0.21-16.52) 3.35 (0.38-29.90) 2.96 (0.30-28.87)
rs3745274
per T allele 2.42 (0.93-6.30) 2.49 (0.90-6.88) 3.61 (1.16-11.22)
p trend 0.07 0.08 0.03
rs28399499
per C allele 1.32 (0.30-5.91) 2.23 (0.47-10.61) 3.63 (0.69-19.18)
p trend 0.71 0.31 0.13

1 Virologic response was defined as achievement of undetectable viral load up to 54 weeks after
self-reported initiation of NNRTI-based regimen
2 rs3211371 could not be analyzed due to lack of non-responders with the TC genotype
3 Adjusted for self-reported (SR) race/ethnicity (White, Hispanic, African American,
Asian/Native American/Other)
4 Adjusted for the top three genetic ancestry principal components (PCs)
5 No carriers of the TT genotype observed

When considering the joint effects of the two SNPs, intermediate and slow
metabolizers were 2.90 (95% CI 0.79-12.28) and 13.44 (95% CI 1.66-infinity) times as
likely to respond to NNRTIs compared to extensive metabolizers after adjustment for
genetic ancestry PCs (p for trend =0.005; Table 5.3). Substantial evidence of confounding

95
is present with the metabolizer phenotype when comparing the crude, self-reported
ethnicity-adjusted and genetic ancestry PCs-adjusted estimates. After adjustment for self-
reported race/ethnicity, the OR for slow metabolizers compared to extensive metabolizers
increased by 54%. After adjustment for genetic ancestry PCs, the confounding effect of
population substructure was further reduced; the OR for slow metabolizers compared to
extensive metabolizers increased by 169% over the unadjusted estimate and by 75% over
the self-reported race/ethnicity estimate. The level of confounding attributable to
population substructure is not surprising given the appreciable heterogeneity that exists
within each self-reported ethnic group (Figures 5.1 and 5.2).

Table 5.3. Estimates for the association between the CYP2B6 metabolizer
phenotype (using rs3745274 and rs28399499) with virologic response
1
to
NNRTI-based HAART regimens in women who were naïve to antiretroviral
drugs (N=91).

Metabolizer
Phenotype
2
Responders
(%)
Non-
responders
(%)
Crude
Adjusted for SR
Race/Ethnicity
3

Adjusted for
top three PCs
4

OR (95% CI) OR (95% CI) OR (95% CI)

Extensive 28 (40) 12 (57) 1.00 1.00 1.00
Intermediate 33 (47) 9 (43) 1.56 (0.52-4.88) 1.66 (0.47-6.11) 2.90 (0.79-12.28)
Slow 9 (13) 0 5.00 (0.70-inf) 7.69 (1.04-inf) 13.44 (1.66-inf)
p trend 0.09 0.04 0.005

1 Virologic response defined as achievement of undetectable viral load up to 54 weeks after
self-reported initiation of NNRTI-based regimen
2 Extensive = no minor alleles at either rs3745274 or rs28399499
Intermediate = one minor allele at either rs3745274 or rs28399499
Slow = two minor alleles at either rs3745274 or rs28399499, or one minor allele at both
polymorphisms
3 Adjusted for self-reported race/ethnicity (White, Hispanic, African American, Asian/Other)
4 Adjusted for the top three genetic ancestry principal components (PCs)

96
DISCUSSION

We observed a statistically significant association between the CYP2B6
metabolizer phenotype (joint effect of rs3745274 and rs28399499) and virologic response
to NNRTI-based antiretroviral regimens (p trend =0.005). This association was strongly
confounded by underlying population substructure for which adjustment of self-reported
race/ethnicity was not a sufficient control. Our findings confirm prior report of the role of
these SNPs in NNRTI metabolism and further highlight the importance of controlling for
population substructure using genetic ancestry estimates, such as PCs, in genetic
association studies.
It is possible that both genetic ancestry estimates and self-reported race/ethnicity
may confound the association between SNPs and outcome, such as if the additional
factors captured by self-reported ethnic group (e.g. SES) are also associated to the SNP of
interest. However, adjusting for both PCs and self-reported race/ethnicity did not change
the estimated effect of CYP2B6 metabolizer phenotype on virologic response to NNRTIs
compared to PCs alone. These findings suggest that PCs alone capture the underlying
population substructure and the additional factors related to self-identification in a
particular ethnic group are not associated with these SNPs.
To our knowledge, we are the first to report a statistically significant joint effect
of rs3745274 and rs28399499 (metabolizer phenotype) on virologic response to NNRTIs
in a multi-ethnic cohort of HIV-infected women. Previously, a study of participants
randomized to efavirenz in the AIDS Clinical Trial Group (ACTG), of which 19% were

97
female and 34% were African-American, explored the effect of the metabolizer
phenotype on virologic failure [14]. The incidence of virologic failure adjusted for the
competing event of efavirenz discontinuation was lower among self-reported African
American participants with the slow metabolizer phenotype (p=0.02) [14]. However,
this effect was not seen in Whites or Hispanics, which is likely an artifact of the lower
allele distribution in non-African Americans. The authors reported that the results were
similar after incorporation of genome-wide association study (GWAS) PCs, though it is
unclear whether this model was tested in all participants.
The two polymorphisms that contribute to the CYP2B6 metabolizer phenotype
are functionally relevant. The minor allele (T) at rs3745274 causes an aberrant splicing
mechanism that decreases expression and activity of CYP2B6 in the liver [37]. The
minor allele (C) at rs28399499 has been consistently associated with reduced CYP2B6
expression [1,7], though the mechanism is unclear [37]. Additionally, it is well
established that both SNPs predict several short-term NNRTI pharmacokinetic
parameters, such as AUC, oral clearance, and half-life [4-14,38]. In a separate WIHS
study, carriers of the TT genotype compared to GG and GT genotypes at rs3745274
experienced an over 3-fold increase in efavirenz concentration in hair specimens after
adjustment for non-genetic predictors of long-term efavirenz exposure (e.g. adherence
and consumption of orange juice) [Gandhi, in press]. The minor allele C at rs28399499
was associated with a 1.70 fold increase of efavirenz levels in hair (p=0.02) [Gandhi, in
press]. The independent associations of these SNPs with long-term exposure to efavirenz
lends support to our results of increased probability of virologic suppression in

98
intermediate and slow metabolizers through the mechanism of sustained increased drug
levels.
A limitation of this study is the absence of viral genotyping on pre-NNRTI
specimens from each participant, which would allow us to limit exclusion of treatment-
experienced NNRTI users to those who are resistant to NNRTIs. There is no reason to
believe that the exclusion of treatment-experienced women who reported initial
treatment with a NNRTI would bias the results away from the null. Another limitation of
our study is the small sample size; however, we observed a robust and statistically
significant association that survived a Bonferroni correction for the number of tests in
CYP2B6 (α =0.008). The functional role of the SNPs comprising the metabolizer
phenotype and associations within and external to the WIHS with pharmacokinetic
parameters is compelling enough to assign a high prior probability that this association is
real.
The association of the CYP2B6 metabolizer phenotype and response to NNRTIs
has important potential for clinical decision making, in which the genetic background of
an individual is taken into account prior to initiation of a given antiretroviral agent. In
fact, this paradigm has already been implemented with the antiretroviral agent, abacavir,
in which acquisition of genetic information is highly recommended prior to the institution
of this drug, in order to avoid potential hypersensitivity reactions (39). In terms of
CYP2B6, although patients with intermediate and slow metabolizer phenotypes are more
likely to achieve undetectable viral loads after treatment with NNRTIs, study of low
grade toxicities experienced in carriers of the minor allele at rs3745274 [4,9] should also

99
be addressed, as these potential toxicities may impact adherence to NNRTI’s over time.
The genetic determinants of anti-retroviral drug absorption, delivery, metabolism and
excretion will be important areas of study in the years ahead.

CHAPTER REFERENCES

1. Zanger UM, Turpeinen M, Klein K, and Schwab M. Functional
pharmacogenetics/genomics of human cytochromes P450 involved in drug
biotransformation. Anal Bioanal Chem 2008;392:1093-108.
2. Ingelman-Sundberg M, Sim SC, Gomez A, and Rodriguez-Antona C. Influence
of cytochrome P450 polymorphisms on drug therapies: pharmacogenetic,
pharmacoepigenetic, and clinical aspects. Pharmacol Ther 2007;116:496-526.
3. Wang HB, Tompkins LM. CYP2B6: new insights into a historically overlooked
cytochrome P450 isozyme. Curr Drug Metab 2008;9:598-610.
4. Rotger M, Colombo S, Furrer H, et al. Influence of CYP2B6 polymorphisms on
plasma and intracellular concentrations and toxicity of efavirenz and nevirapine in
HIV-infected patients. Pharmacogenet Genomics 2005;15:1-5.
5. Klein K, Lang T, Saussele T, et al. Genetic variability of CYP2B6 in populations
of African and Asian origin: allele frequency variants and possible implications
for anti-HIV therapy with efavirenz. Pharmacogenetic Genomics 2005;15:861-
73.

100
6. Wang J, Sonnenberg A, Rane A, et al. Identification of a novel specific CYP2B6
allele in Africans causing impaired metabolism of the HIV drug efavirenz.
Pharmacogenet Genomics 2006;16:191-198.
7. Rotger M, Tegude H, Colombo S, et al. Predictive value of known and novel
alleles of CYP2B6 for efavirenz plasma concentrations in HIV-infected
individuals. Clin Pharmacol Ther. 2007;81:557-66.
8. Wyen C, Hendra H, Vogel M, et al. Impact of CYP2B6 983T>C polymorphisms
on non-nucleoside reverse transcriptase inhibitor plasma concentrations in HIV-
infected patients. J Antimicrob Chemother 2008;61:914-8.
9. Haas DW, Ribaudo HJ, Kim RB, et al. Pharmacogenetics of efavirenz and central
nervous system side effects: An Adult AIDS Clinical Trials Group study. AIDS
2004; 18:2391-400.
10. Haas DW, Smeaton LM, Shafer RW, et al. Pharmacogenetics of long-term
responses to antiretroviral regimens containing efavirenz and/or nelfinavir: An
Adult AIDS Clinical Trials Group Study. J Infect Dis. 2005; 192:1931-1942.
11. Motsinger AA, Ritchie MD, Shafer RW, et al. Multilocus genetic interactions
and response to efavirenz-containing regimens: An adult AIDS Clinical Trials
Group study. Pharmacogenetics and Genomics 2006; 16:837-845.
12. Saitoh A, Sarles E, Capparelli E, et al. CYP2B6 genetic variants are associated
with nevirapine pharmacokinetics and clinical response in HIV-1 infected
children. AIDS 2007;21:2191-9.

101
13. Haas DW, Koletar SL, Laughlin L et al. Associations between CYP2B6
polymorphisms and pharmacokinetics after a single dose of nevirapine or
efavirenz in African Americans. J Infect Dis 2009;199:872-880.
14. Ribaudo HJ, Huan L, Schwab M, et al. Effect of CYP2B6, ABCB1, and CYP3A5
polymorphisms on efavirenz pharmacokinetic and treatment response: An AIDS
Clinical Trials Group study. J Infect Dis 2010;202:717-22.
15. Anastos K, Schneider MF, Gange SJ, et al. The association of race,
sociodemographic, and behavioral characteristics with response to highly active
antiretroviral therapy in women. J Acquir Immune Defic Syndr 2005;39:537-544.
16. Thomas DC, Witte JS. Point: population stratification: a problem for case-control
studies of candidate-gene associations? Cancer Epidemiol Biomarkers Prev
2002;11(6):505-512.
17. Gorroochurn P, Hodge SE, Heiman G, et al. Effect of population stratification on
case-control association studies. II False-positive rates and their limiting behavior
as number of subpopulations increases. Hum Hered 2004;58(1):40-48.
18. Pritchard JK, Stephens M, Donnelly P. Inference of population structure using
multilocus genotype data. Genetics 2000;155:945-959.
19. Pritchard JK, Donnelly P. Case-control studies of association in structured or
admixed populations. Theor Popul Biol 2001;60:227-237.
20. Voight BF, Pritchard JK. Confounding from cryptic relatedness in case-control
associations. PloS Genet 2005;1(3):e32.

102
21. Freedman ML, Reich D, Penney KL, et al. Assessing the impact of population
stratification on genetic association studies. Nat Genet 2004;36(4):388-393.
22. Hoggart CJ, Parra EJ, Shriver MD, et al. Control of confounding of genetic
associations in stratified populations. Am J Hum Genet 2003;72:1492-1504.
23. Parra EJ, Marcini A, Akey J, et al. Estimating African American admixture
proportions by use of population-specific alleles. Am J Hum Genet
1998;63:1839-1851.
24. Salari K, Choudhry S, Tang H, et al. Genetic admixture and asthma related
phenotypes in Mexican American and Puerto Rican asthmatics. Genet Epidemiol
2005;29:76-86.
25. Halder I, Yang BZ, Kranzler HR, et al. Measurement of admixture proportions
and description of admixture structure in different U. S. populations. Hum Mutat
2009;30(9):1299-1309.
26. Smith MW, Patterson N, Lautenberger JA, et al. A high-density admixture map
for disease gene discovery in African Americans. Am J Hum Genet
2004;74(5):1001-1003.
27. Barkan SE, Melnick SL, Preston-Martin S, et al. The Women’s Interagency HIV
Study. Epidemiology 1998;9:117-25.
28. Bacon MC, Von Wyl V, Alden C, et al. The Women's Interagency HIV Study: an
observational cohort brings clinical sciences to the bench. Clin Diagn Lab
Immunol 2005;12(9): 1013-19.

103
29. Barlett JG, Lane HC. Guidelines for the use of antiretroviral drugs in HIV-1-
infected adults and adolescents. Clinical Guidelines for the treatment and
management of HIV Infection. Edited by Infection PCPTHIV. USA, Department
of Health and Human Services; 2005:1-118.
30. Falush D, Stephens M, Pritchard JK. Inference of population structure using
multilocus genotype data: linked loci and correlated allele frequencies. Genetics
2003;164:1567-1587.
31. The International HapMap Consortium. The International HapMap Project.
Nature 2003;426:789-796.
32. Kosoy R, Nassir R, Tian C, et al. Ancestry informative marker sets for
determining continental origin and admixture proportions in common populations
in America. Hum Mutat 2009; 30(1):69-78.33.
33. Rosenberg, NA. Distruct: a program for the graphical display of population
structure. Molecular Ecology Notes 2004;4:137-138.
34. Price AL, Patterson NJ, Plenge RM, et al. Principal components analysis corrects
for stratification in genome-wide association studies. Nat Genet 2006;38(8):904-
909.
35. Price AL, Zaitlen NA, Reich D, et al. New approaches to population stratification
in genome-wide association studies. Nat Rev Genet 2010;11:459-463.
36. Wang S, Lewis CM, Jakobsson M, et al. Genetic Variation and Population
Structure in Native Americans. PloS Genet 2007;3(11):e185.

104
37. Hoffman MH, Blievernicht JK, Klein K, et al. Abberant splicing caused by single
nucleotide polymorphism c.516G>T [Q172H], a marker of CYP2B6*6, is
responsible for decreased expression and activity of CYP2B6 in liver. J
Pharmacol Exp Ther 2008;325:284-292.
38. Saitoh A, Spector SA. Effect of host genetic variation on the pharmacokinetics
and clinical response of non-nucleoside reverse transcriptase inhibitors. Futur
HIV Ther 2008;2:69-81.
39. Chaponda M and Pirmohamed M. Hypersensitivity reactions to HIV therapy. Br
J Clin Pharmacol 2011;71(5):659-671.

105
Chapter 6: Validation of Diabetes Genetic Associations

ABSTRACT

Background. Type 2 diabetes (T2D) in the setting of human immunodeficiency virus
(HIV) infection is a concern given the prevalence of HIV-related conditions that
contribute to the etiology T2D. Genome wide association studies (GWAS) have
identified several common single nucleotide polymorphisms (SNP) that have been
reproducibily associated with T2D in separate European and Asian populations. Eighteen
T2D-associated SNPs were tested in a multi-ethnic population of HIV-infected women.
Methods. Incident T2D criteria included a combination of self-reported anti-diabetic
medication, fasting glucose ≥ 126 mg/dL, and hemoglobin A1C ≥ 6.5% measurements
with confirmation at the subsequent visit. Cox proportional hazard models were fitted
between each SNP and T2D with adjustment for genetic ancestry principal components
using age for the time-scale. Exposure to nucleoside reverse transcriptase inhibitors
(NRTI) was explored as an effect modifier.
Results. Overall, the T2D risk conferred by these SNPs was similar in White/Hispanic
HIV-infected women compared to HIV-uninfected European individuals, as evaluated by
the effect measures and p values for heterogeneity. The magnitude of each SNP effect
was smaller in African-American women (HRs~1.1). Significant interactions were
revealed between these T2D-associated SNPs and NRTI exposure (p<0.03) in
White/Hispanic women.

106
Conclusions. Further study is warranted in large-scale datasets to investigate the modest
effects of these SNPs in ethnically diverse HIV-infected individuals and the suggested
synergistic effects of NRTIs on risk of T2D.

INTRODUCTION
Evidence for a genetic influence on insulin levels, insulin resistance and type 2
diabetes mellitus (T2D) has been reported since the late 1980’s [1-5]. In 2007, four
genome-wide association studies (GWAS) identified multiple alleles associated with T2D
in predominately White populations [6-10]. Replication studies in independent European
and Asian populations have confirmed the T2D risk variants [11- 23] and their
associations with quantitative traits of the diabetic phenotype (i.e. insulin resistance,
insulin response, and fasting plasma glucose)

[11,20-29]. However, there is conflicting
evidence as to whether these markers confer similar risk in the African American
population. With the exception of variants at the TCF7L2 (rs7903146) and JAZF1
(rs864745) loci, the majority of the established T2D risk variants have failed to replicate
in African American populations, though the direction of the effect measure is consistent
with European- and Asian- derived susceptibility loci [30-33]. A recent GWAS with
replication and validation phases in African Americans revealed a single genetic
determinant of T2D near the gene RND3 [34]. This association was not observed in the
primary European GWAS meta-analysis [6-10].
T2D incidence among individuals infected with human immunodeficiency virus
(HIV) is a concern given the prevalence of HIV-related conditions that contribute to the

107
etiology and pathogenesis of T2D, such as premature cardiovascular disease [35,36],
dyslipidemia [37], and advanced immunosupression [38]. In addition, exposure to
nucleoside reverse transcriptase inhibitors (NRTI), particularly thymidine analogues, is
linked to T2D risk [39,40]. Although debated, there is substantial evidence in favor of a
direct association between HIV infection and T2D [39-42].
The Women’s Interagency HIV Study (WIHS) is a prospective study of HIV-
infected and uninfected women from six United States (U.S.) sites who have been
followed every six months since 1995. The aim of the current study was to examine the
individual associations of the European- and Asian- derived T2D risk variants in a multi-
ethnic cohort of HIV-infected women with consideration for genetic ancestry estimates
and HIV-related predictors of T2D.

METHODS
WIHS Study Design. The WIHS is a prospective study of HIV-infected women
and a comparison group of HIV-uninfected women. Participants were recruited from six
sites across the U.S., located in Bronx/Manhattan, New York; Brooklyn, New York;
Washington, D.C.; Los Angeles, CA; San Francisco/Bay Area, CA; and Chicago, IL.
Enrollment was conducted in two phases; 2,054 HIV seropositive and 569 seronegative
women were enrolled in 1994-1995 and a second enrollment from 2001-2002 resulted in
the addition of 737 HIV seropositive and 406 seronegative women to the cohort. A more
detailed description of the WIHS cohort has been published elsewhere [43,44].

108
Participants are seen for in-person visits every six months during which time
trained medical interviewers administer an extensive questionnaire, a clinical
examination is performed, and biological samples are collected for various laboratory
tests, including HIV RNA viral load and white blood cell counts. Beginning in October
2000, fasting glucose (FG) and hemogloblin A1C (A1C) were measured at each follow-
up visit. Detailed information on participants’ antiretroviral medications and medications
used for other conditions, such as anti-diabetic medications, is obtained and recorded at
each follow-up visit.
T2D case definition. Participants were followed after the index visit for
incidence of T2D. T2D was defined by (i) first report of anti-diabetic medication, (ii)
first visit with A1C ≥ 6.5% confirmed by either a report of anti-diabetic medication at the
subsequent visit or FG ≥ 126 mg/dL at the same visit or subsequent visit, (iii) first visit
with FG ≥ 126 mg/dL confirmed by report of anti-diabetic medication or FG ≥ 126
mg/dL at the subsequent visit or A1C ≥ 6.5% at the same visit or subsequent visit, (iv)
first report of T2D confirmed by report of anti-diabetic medication at the subsequent visit
or A1C ≥ 6.5% plus FG ≥ 126 mg/dL at the subsequent visit or two measurements of FG
≥ 126 mg/dL at the following two subsequent visits. These definitions conform to
recommendations made by the American Diabetes Association [45].
Inclusion criteria. The first visit for which FG and A1C measurements were
available was considered the index visit. HIV-infected women who consented to genetic
studies, did not have a positive or missing report of pregnancy, and had at least one
follow-up visit with FG and A1C measurements after the index visit were eligible for the

109
study. Women satisfying any of the above T2D definitions at the index visit (with
confirmation at the subsequent visit) were flagged as prevalent T2D cases and excluded
from the study. The final study population consisted of 1,561 women.
T2D-associated SNP selection, Genotyping and Quality Control. A total of 18
single nucleotide polymorphisms (SNPs) across 14 genes identified from European-
derived GWAS were selected for genotyping. Nine of these SNPs were genotyped on
Illumina’s Golden-Gate platform (San Diego, CA): rs10811661 CDKN2A/B, rs1111875
HHEX, rs13266634 SLC30A8, rs1470579 and rs4402960 IGFBP2, rs5219 KCNJ11,
rs7754840 CDKAL1, rs7903146 TCF7L2, and rs6698181 PKN2. The remaining nine
were genotyped using TaqMan allelic discrimination assays (Applied Biosystems, Foster
City, CA): rs1801282 PPARG, rs12779790 CAMK1D, rs564398 CDKN2A/B, rs8050136
FTO, rs7923837 HHEX, rs864745 JAZF1, rs2237895 KCNQ1, rs2934381 NOTCH2, and
rs7901695 TCF7L2.
Sufficient DNA material for genotyping was available for 88% of eligible
participants (n=1,366/1,561). Data for nine of the T2D-associated SNPs were generated
as a subset of a larger genotyping panel (n=384 SNPs) conducted on all consented WIHS
women with available DNA in the entire cohort (n=2,533). The panel included 185
ancestry informative markers (AIM) along with 199 variants in genes of broad interest in
the WIHS. The systematic quality control (QC) process eliminated 24 of 384 SNPs (6%)
for which three distinct genotypes were not observed. Next, 76 samples (3%) were
eliminated that had missing data for more than 20% of the remaining 360 SNPs. Assay
performance for the remaining 360 SNPs was then assessed and seven additional SNPs

110
(2%) were eliminated because they had a call rate < 90%. Lastly, 110 samples (5%) were
excluded because they had more than 10% missing data after the elimination of the 31
SNPs. Three SNPs were eliminated due to deviation from Hardy Weinberg Equilibrium
(HWE). Duplicates for 53 participants were included and concordance was 99.9%.
Duplicate concordance rates were 99% across all assays for the additional nine SNPs
genotyped with Taqman and none deviated from HWE at p<0.05.
In summary, 76 (6%) of the 1,366 women genotyped in the T2D study were
excluded for failure to meet QC criteria. The nine SNPs genotyped with GoldenGate,
and the nine SNPs genotyped with TaqMan passed QC criteria.
Genetic ancestry estimation to control for population substructure. Genetic
ancestry proportions were estimated using data for 168 ancestry informative markers that
were included in the large genotyping panel. In a similar method to EIGENSTRAT [46],
principal components [PC] analysis was performed to generate genetic ancestry
covariates. A more detailed description on genetic ancestry determination in the WIHS
has been addressed in a separate study.
Statistical analysis. The final analytic dataset included 1,290 women consisting
of 150 T2D cases and 1,140 controls. Cox proportional hazard models were fitted to test
associations between each SNP and T2D with adjustment for the top 10 genetic ancestry
PCs using age for the time-scale. A log-additive mode of inheritance was modeled for
each SNP coded on the established risk allele [47]. Hazard ratios (HR) per allele and
95% confidence intervals (CI) were estimated by modeling the genotypes as an ordinal
variable, where non-risk allele homozygotes, heterozygotes and risk allele homozygotes

111
were coded as 0, 1, 2, respectively. To investigate the suggested heterogeneity of effect
between African and non-African populations, the analyses were stratified by self-
reported ethnicity (African American vs. White, Hispanic, Asian, Other).
Body mass index (BMI) at study entry and time-dependent antiretroviral therapy
were evaluated as potential confounders, defined as a change in the SNP main effect β by
≥ 10%. The dose of NRTIs (number of NRTIs at each visit) reported at each visit and the
combined dose-duration of NRTIs (NRTI-years defined as the number of NRTIs reported
at each visit multiplied by 0.5, since each visit is approximately 6 months apart) were
explored as potential effect modifiers based on a priori evidence of NRTI exposure as a
risk factor for T2D in the WIHS [40].
A meta-analysis was performed on African-American replication studies [30-34]
using published odds ratios and 95% CIs as input for the metan procedure in STATA
(College Station, TX). The Q statistic with 1 df was used to obtain the p for
heterogeneity between associations in the WIHS and previously published associations in
HIV negative populations [10, 30-34] and the Swiss HIV Cohort Study (SHC) [48].

RESULTS

Characteristics of the 1,290 women with successful genotyping are presented in
Table 6.1. T2D cases (n=150) had a median follow-up time of three years, compared to
eight and a half years in controls; the total follow-up time was 60,004 person-years. The
median age (40 years) and the distribution of race/ethnicity were similar between cases

112
and controls. As expected, cases were more likely to have a BMI over 30 kg/m
2
compared to controls (41% vs 28%). Cases (15%) were more likely to remain naïve to
antiretroviral drugs compared to controls (9%), which is likely a reflection of the shorter
follow-up among cases.

Table 6.1 Description of HIV-infected women in the Women’s Interagency HIV Study
(WIHS) by type 2 diabetes (T2D) status from 2000-2011.

T2D T2D-free
N (%) N (%)

N 150 1,140
Median age (IQR) 46.4 (40.3-50.8) 46.1 (40.4-52.6)
Median follow-up years (IQR) 3 (1.5-5.5) 8.5 (6.5-9.5)
Site
Bronx, NY 32 (21) 195 (17)
Brooklyn, NY 22 (15) 240 (21)
Washington, DC 19 (13) 174 (15)
Los Angeles, CA 22 (15) 235 (21)
San Francisco, CA 17 (11) 141 (12)
Chicago, IL 38 (25) 155 (14)
Self-identified ethnicity
African American 86 (57) 687 (60)
Hispanic 37 (25) 272 (24)
White 22 (15) 142 (13)
Asian/other 5 (3) 39 (3)
Phase of Cohort Enrollment
1995-1996 110 (73) 723 (63)
2000-2001 40 (27) 417 (37)
BMI at index visit
1

< 25 44 (29) 449 (39)
26-30 44 (29) 342 (30)
> 30 61 (41) 316 (28)
missing 1 (1) 33 (3)
Antiretroviral treatment at censored visit
2

Naïve to antiretroviral drugs 23 (15) 98 (9)

1 Body mass index (BMI) was taken at the first visit where measurements for fasting
glucose and hemoglobin A1C were available (index visit)
2 Self-reported use of antiretroviral drugs at visit of incident T2D or end of follow up

113
The associations between individual SNPs and T2D after adjustment for genetic
ancestry PCs among White, Hispanic, Asian, and other women (n=517) are presented in
Table 6.2. The results were similar after adjustment for BMI and current NRTI exposure.
Overall, the T2D risk conferred by these SNPs was similar in HIV-infected women
compared to HIV-uninfected European individuals [10], as evaluated by the effect
measures and p values for heterogeneity. However, between study heterogeneity was
observed for two SNPs, rs6698181 and rs7903146 (p-het=0.01). A subset of these SNPs
were tested for associations with T2D in the Swiss HIV Cohort Study (SHC) [48] which
allowed a direct comparison between two HIV-infected populations. With the exception
of rs7903146 (p-het=0.03), the associations with T2D-susceptibility alleles were similar
between the WIHS and the SHC. Of note, modest heterogeneity was detected between
the estimate for rs5219 compared to the SHC (p=0.14) but not in comparison to the
GWAS-derived estimate (p=0.63).

114
Table 6.2 Associations between single nucleotide polymorphisms (SNP) and type 2
diabetes in Women’s Interagency HIV Study (WIHS) self-reported White, Hispanic,
Asian, and other women (n=517) and published results from the European genome wide
association study (GWAS) meta-analysis and the Swiss HIV Cohort Study (SHC) with p
values for heterogeneity (p-het).

* Adjusted for genetic ancestry principal components 1-10
1 Established risk allele in European populations
2 RAF= risk allele frequency
3 The G allele is the established protective allele
4 Zeggini 2007
5 Rotger 2010

The estimates for the T2D-susceptibility alleles in African-Americans (n=773) are
presented in Table 6.3. Overall, the magnitude of effect was smaller in African-
American women compared to their non-African counterparts. The meta-analysis of five
validation studies in African-American populations [30-34] revealed odds ratios < 1.1
except for rs864745 (OR=1.16, 95% CI 1.09-1.23) and rs7903146 (OR=1.35, 95% CI
1.28-1.43). The HRs for rs864745 and rs7903146 in WIHS African-American women

115
were not significantly different than the meta-ORs (p-het=0.09 and 0.29, respectively),
though neither association reached statistical significance in the WIHS.

Table 6.3 Associations between single nucleotide polymorphisms (SNP) and type 2
diabetes in Women’s Interagency HIV Study (WIHS) self-reported African-American
(AA) women (n=773) and a meta-analysis of published results from five African-
American (HIV negative) replication studies with p values for heterogeneity (p-het).

* Adjusted for genetic ancestry principal components 1-10 with age as time scale
1 Established risk allele in European populations
2 RAF= risk allele frequency
3 Q statistic with 1 df
4 The G allele is the established protective allele
[1] Saxena 2012 [2] Cooke 2012 [3] Waters 2010 [4] Lewis 2008 [5] Palmer 2012

The associations of the T2D-risk alleles were substantially modified by the
number of NRTIs reported at each visit (NRTI dose) in White, Hispanic and other
women (Table 6.4). Statistically significant associations between these SNPs and T2D in
women who reported zero, one or two NRTIs were not observed. In contrast, the risk
conferred by these T2D-associated alleles ranged from 1.53-3.35 (p<0.01) in women who

116
reported three or more NRTIs. The estimates for rs6698181 and rs7903146 were in the
opposite direction in the pooled results (Table 6.2), but a nominal risk of T2D with these
loci (HRs~1.53) was observed in women who reported three or more NRTIs at the
current visit. The strongest modifying effect of NRTI dose was observed at the two risk
alleles in the IGFBP2 gene, where no association was observed in women who reported
zero, one or two NRTIs, but the associations between these two SNPs and T2D were over
three-fold in women who reported three or more NRTIs.

Table 6.4 Associations between type 2 diabetes and single nucleotide polymorphisms
(SNP) by number of NRTIs at visit after adjustment for genetic ancestry covariates in
White, Hispanic, Asian, and other HIV+ women (n=517).

SNP
0 NRTIs at visit
12 cases / 7348 PY
HR (95% CI)
1-2 NRTIs at visit
31 cases / 13,271 PY
HR (95% CI)
3+ NRTIs at visit
21 cases / 3,466 PY
HR (95% CI)
rs12779790 1.43 (0.72-2.84) 0.50 (0.23-1.07) 1.82 (0.80-4.16)
rs7754840 1.09 (0.58-2.06) 1.01 (0.64-1.60) 2.92 (1.82-4.69)
rs564398 0.81 (0.46-1.45) 0.93 (0.59-1.48) 1.80 (1.11-2.94)
rs10811661 1.23 (0.66-2.30) 1.28 (0.72-2.28) 2.55 (1.44-4.54)
rs8050136 1.12 (0.59-2.13) 0.97 (0.62-5.17) 3.35 (2.17-5.17)
rs7923837 1.11 (0.66-1.88) 1.06 (0.68-1.65) 2.50 (1.60-3.93)
rs1111875 0.95 (0.58-1.58) 0.94 (0.63-1.42) 1.95 (1.29-2.95)
rs1470579 0.70 (0.32-1.55) 1.14 (0.74-1.76) 3.28 (2.06-5.20)
rs4402960 0.70 (0.32-1.55) 1.05 (0.67-1.65) 3.39 (2.12-5.42)
rs864745 1.05 (0.62-1.78) 0.96 (0.62-1.50) 2.22 (1.41-3.48)
rs5219 0.84 (0.41-1.75) 0.93 (0.58-1.51) 1.85 (1.01-3.39)
rs2237895 0.88 (0.47-1.62) 0.79 (0.50-1.24) 2.11 (1.34-3.33)
rs2934381 0.39 (0.06-2.69) 0.77 (0.35-1.70) 2.30 (1.17-4.51)
rs6698181 0.64 (0.34-1.20) 0.55 (0.34-0.88) 1.53 (0.88-2.64)
rs1801282 1.34 (0.43-4.17) 0.83 (0.41-1.70) --
rs13266634 1.42 (0.83-2.43) 1.31 (0.80-2.13) 2.64 (1.60-4.35)
rs7901695 1.17 (0.64-2.11) 1.26 (0.81-1.95) 2.81 (1.74-4.55)
rs7903146 0.65 (0.30-1.39) 0.48 (0.24-0.93) 1.54 (0.84-2.84)

117
In contrast to the effect modification observed in White, Hispanic, and other
women, the associations of T2D-risk alleles were similar between African-American
women who reported use of zero, one to two, or three or more NRTIs (Table 6.5).

Table 6.5 Associations between incident type 2 diabetes and single nucleotide
polymorphisms (SNP) by number of NRTIs after adjustment for genetic ancestry
covariates in African American HIV+ women (n=773).

Table 6.6 reports the modifying effect of NRTIs after considering the combination
of dose and duration (NRTI-years). There was a significant interaction between NRTI-
years and two SNPs in the IGFBP2 gene (p<0.04). Specifically, women who were
carriers of the risk alleles at either rs1470579 or rs4402960 and had three or more NRTI-
years were twice as likely to develop T2D (Table 6.6). Overall, the genetic effect on
SNP
0 NRTIs at visit
25 cases / 14,276 PY
HR (95% CI)
1-2 NRTIs at visit
52 cases / 18,234 PY
HR (95% CI)
3+ NRTIs at visit
9 cases / 3,396 PY
HR (95% CI)
rs12779790 0.61 (0.23-1.65) 1.42 (0.88-2.30) 1.49 (0.56-3.93)
rs7754840 0.81 (0.54-1.22) 0.98 (0.71-1.36) 1.01 (0.63-1.63)
rs564398 0.90 (0.48-1.67) 1.03 (0.56-1.88) 0.98 (0.50-1.91)
rs10811661 0.58 (0.33-1.01) 0.65 (0.38-1.11) 0.60 (0.32-1.13)
rs8050136 0.88 (0.56-1.39) 1.12 (0.79-1.60) 1.21 (0.68-2.16)
rs7923837 1.07 (0.56-2.06) 1.26 (0.67-2.38) 1.08 (0.53-2.23)
rs1111875 1.00 (0.64-1.56) 1.21 (0.82-1.80) 1.16 (0.70-1.94)
rs1470579 0.99 (0.63-1.55) 1.12 (0.76-1.67) 1.09 (0.65-1.83)
rs4402960 1.04 (0.69-1.57) 1.09 (0.79-1.52) 1.11 (0.63-1.95)
rs864745 0.82 (0.54-1.23) 0.89 (0.64-1.25) 0.89 (0.55-1.41)
rs5219 1.76 (0.77-4.05) 1.07 (0.52-2.21) --
rs2237895 0.78 (0.35-1.75) 0.97 (0.60-1.57) 1.57 (0.68-3.62)
rs2934381 0.94 (0.58-1.52) 0.98 (0.67-1.44) 1.10 (0.54-2.26)
rs6698181 0.76 (0.31-1.83) 1.16 (0.72-1.88) 0.37 (0.05-2.54)
rs1801282 0.71 (0.10-5.13) -- --
rs13266634 0.94 (0.52-1.67) 1.00 (0.57-1.78) 0.95 (0.50-1.83)
rs7901695 1.01 (0.67-1.52) 1.18 (0.85-1.64) 0.92 (0.50-1.68)
rs7903146 1.05 (0.62-1.78) 1.17 (0.79-1.74) 1.08 (0.52-2.22)

118
T2D risk across these loci was elevated among those women who reported exposure to
three or more NRTI-years.

Table 6.6 Associations between type 2 diabetes (T2D) and single nucleotide
polymorphisms (SNP) by cumulative number of nucleoside reverse transcriptase
inhibitors (NRTI)-years (dose and duration) after adjustment for genetic ancestry
covariates in White, Hispanic, Asian, and other HIV+ women (n=517).

Effect modification by the cumulative number of NRTI-years was also explored
in African-American women (Table 6.7). Among women who were exposed to three or
more NRTI-years, most T2D-risk alleles were significantly associated with T2D
incidence, including the TCF7L2 SNPs which are highly reproducible in the HIV
uninfected African-American population. These effects were not observed in African-
SNP
0 NRTI-years
9 cases / 6,819 PY
HR* (95% CI)
0.5-2.9 NRTI-years
9 cases / 14,372 PY
HR* (95% CI)
≥ 3 NRTI-years
46 cases / 2,834 PY
HR* (95% CI)
p-
interaction
rs12779790 1.69 (0.78-3.68) 0.25 (0.06-0.99) 1.15 (0.63-2.09) 0.01
rs7754840 1.08 (0.53-2.19) 0.59 (0.28-1.23) 1.85 (1.25-2.72) 0.81
rs564398 0.77 (0.41-1.43) 0.55 (0.30-1.00) 1.40 (0.91-2.16) 0.55
rs10811661 1.15 (0.61-2.20) 0.83 (0.43-1.60) 1.93 (1.12-3.32) 0.53
rs8050136 1.13 (0.54-2.33) 0.49 (0.23-1.08) 1.99 (1.37-2.89) 0.35
rs7923837 1.11 (0.65-1.90) 0.55 (0.28-1.05) 1.72 (1.15-2.56) 0.49
rs1111875 0.86 (0.49-1.49) 0.50 (0.26-0.93) 1.54 (1.07-2.21) 0.71
rs1470579 0.45 (0.15-1.35) 0.59 (0.28-1.22) 2.12 (1.45-3.11) < 0.01
rs4402960 0.44 (0.14-1.35) 0.68 (0.24-1.34) 1.99 (1.36-2.91) 0.03
rs864745 0.97 (0.54-1.74) 0.61 (0.33-1.13) 1.61 (1.09-2.38) 0.66
rs5219 0.69 (0.27-1.76) 0.72 (0.37-1.41) 1.38 (0.88-2.17) 0.56
rs2237895 0.80 (0.39-1.66) 0.45 (0.21-0.94) 1.43 (0.98-2.09) 0.97
rs2934381 0.56 (0.08-3.76) 0.34 (0.08-1.38) 1.81 (1.00-3.27) 0.29
rs6698181 0.48 (0.21-1.10) 0.50 (0.26-0.95) 0.95 (0.63-1.46) 0.52
rs1801282 1.23 (0.31-4.86) 0.36 (0.09-1.43) 1.18 (0.56-2.49) 0.47
rs13266634 1.30 (0.73-2.29) 0.94 (0.53-1.67) 2.04 (1.28-3.24) 0.05
rs7903146 0.80 (0.38-1.71) 0.30 (0.11-0.84) 0.99 (0.59-1.68) 0.04
rs7901695 0.99 (0.51-1.93) 0.68 (0.37-1.25) 1.96 (1.30-2.97) 0.09

119
American women who did not accumulate three or more NRTI-years, suggesting effect
modication by NRTI exposure.

Table 6.7 Associations between type 2 diabetes (T2D) and single nucleotide
polymorphisms (SNP) by cumulative number of nucleoside reverse transcriptase
inhibitors (NRTI)-years (dose and duration) after adjustment for genetic ancestry
covariates in African-American HIV+ women (n=773).

DISCUSSION

We tested 18 European-derived T2D-associated SNPs for risk of T2D in a
longitudinal study of multi-ethnic women infected with HIV, using fasting glucose and
hemoglobin A1C to define incident T2D. The effect of these SNPs in White, Hispanic,
and other women trended toward the same direction as seen in HIV-uninfected European
SNP
0 NRTI-years
14 cases / 13,368 PY
HR* (95% CI)
0.5-2.9 NRTI-years
12 cases / 19,006 PY
HR* (95% CI)
≥ 3 NRTI-years
60 cases / 3,470 PY
HR* (95% CI)
p-
interaction
rs12779790 -- 0.82 (0.39-1.69) 2.33 (1.43-3.79) 0.05
rs7754840 0.59 (0.35-0.98) 0.51 (0.31-0.85) 1.33 (0.98-1.81) 0.97
rs564398 0.72 (0.37-1.37) 0.55 (0.28-1.07) 1.37 (0.75-2.47) 0.20
rs10811661 0.48 (0.27-0.86) 0.35 (0.19-0.65) 0.91 (0.54-1.56) 0.59
rs8050136 0.67 (0.38-1.18) 0.45 (0.23-0.86) 1.65 (1.19-2.28) 0.97
rs7923837 0.79 (0.40-1.46) 0.63 (0.31-1.28) 1.61 (0.86-3.00) 0.47
rs1111875 0.79 (0.48-1.30) 0.64 (0.39-1.05) 1.66 (1.13-2.44) 0.38
rs1470579 0.64 (0.37-1.10) 0.54 (0.32-0.91) 1.49 (1.02-2.17) 0.05
rs4402960 0.66 (0.37-1.16) 0.52 (0.30-0.89) 1.64 (1.20-2.22) 0.33
rs864745 0.56 (0.34-0.93) 0.48 (0.30-0.76) 1.23 (0.89-2.40) 0.68
rs5219 1.43 (0.52-3.89) 1.25 (0.54-2.91) 0.90 (0.34-2.17) <0.01
rs2237895 0.68 (0.25-1.87) 0.23 (0.06-0.86) 1.60 (1.05-2.44) 0.78
rs2934381 0.69 (0.37-1.28) 0.46 (0.24-0.89) 1.50 (1.06-2.13) 0.64
rs6698181 0.43 (0.11-1.72) 0.33 (0.11-1.03) 1.76 (1.10-2.82) 0.24
rs1801282 0.98 (0.13-7.13) -- -- --
rs13266634 0.71 (0.38-1.30) 0.49 (0.25-0.95) 1.32 (0.75-2.30) 0.29
rs7903146 1.04 (0.59-1.86) 0.47 (0.22-0.98) 1.68 (1.16-2.44) 0.07
rs7901695 0.57 (0.32-1.02) 0.52 (0.30-0.90) 1.61 (1.19-2.19) 0.38

120
populations, although none of the associations reached statistical significance. The
nominal risk of T2D conferred by these SNPs in European and Asian populations was not
observed in HIV-infected African-American women, similar to reports in HIV-uninfected
African-American populations [30-34]. We also show for the first time, that NRTI
treatment significantly modified the associations between four SNPs and T2D in White,
Hispanic, and other women (p<0.04). Effect modification by NRTI treatment was
suggested in the remaining genetic associations in non-African women and across all
genetic associations in African-American women.
The results of this study exemplify that the genetic contribution to the
development of T2D in the context of HIV infection is similar to that in the European
general population. Disconcertingly, there was an absence of support for one of the two
TCF7L2 variants tested in this study, despite this gene’s strong influence on T2D in the
European population. In White and Hispanic WIHS women, the association between
rs7903146 and T2D was not reproduced in the pooled analysis or among those with the
highest category of NRTI exposure. The majority of the non-African subset was
Hispanic and thus, the association was driven by this group. Interestingly, the allele
frequency of rs7903146 was identical in WIHS Whites and Hispanics, but the odds ratio
was smaller in Hispanics. The lack of association of rs7903146 is consistent with a
GWAS performed in Mexican and Mexican-Americans [49].
Associations between these SNPs and T2D in Whites and Hispanics were mostly
consistent with the associations observed in the SHC [48]. The SNP with the largest
effect in the SHC study (rs5219, RR=1.63, 95% CI 1.05-2.55) was associated with T2D

121
with a comparable magnitude exclusively in WIHS women who were treated with three
or more NRTIs (HR=1.85 95% CI 1.01-3.39). Effect modification by dose or duration of
NRTIs was not explored in the SHC, however, 84% of the 644 individuals in the study
population reported treatment with NRTIs [48]. Thus, the association in the SHC may
have been driven by the effect of NRTIs, given that the effect of rs5219 in the general
European population was 1.15 (95% CI 1.09-1.21) [10]. Alternatively, the definition of
T2D in the SHC was based on fasting and nonfasting gluose levels, which may have lead
to misclassification of some borderline T2D cases ultimately creating a bias away from
the null.
To date, associations with several European-derrived T2D-risk loci have not been
replicated in African-American populations [30-34]. A meta-analysis including the
associations in the WIHS did not favor much overlap with the established associations in
European and Asian populations. The highly reproducible association between TCF7L2
variants (rs7903146 and rs7901695) and T2D was observed in African-American women,
but only after stratification of NRTI treatment categories. The European-derived alleles
associated with T2D are unlikely to be the causal alleles and it is possible that these
markers are not close proxies for the causal alleles in the disparate linkage disequilibrium
structures found in African-Americans.
Strengths of this study include the longitudinal assessment of the influence of
NRTI exposure on T2D-associated alleles, the stringent definition of diabetes that
included A1C measurements and required confirmation at the subsequent visit, and
adjustment for genetic ancestry PCs to account for the potential confounding effects of

122
population stratification [46]. The main limitation of this study is the modest number of
cases, which did not provide adequate power to detect the nominal effects of these T2D-
associated loci. Further study is warranted in large-scale datasets to investigate the
modest effects of these SNPs in ethnically diverse HIV-infected individuals and the
suggested synergistic effects of NRTIs on risk of T2D.

CHAPTER REFERENCES

1. Haffner SM, Stern MO, Hazuda HP et al. Increased insulin concentrations in
nondiabetic offspring of diabetic parents. NEJM 1988; 319:1297-1301.
2. Schumacher MC, Hasstedt SJ, Hunt SC et al. Major gene effect for insulin levels
in familial NIDDM pedigrees. Diabetes 1992; 41:416-423.
3. Mayer EJ, Newman B, Austin MA, et al. Genetic and environmental influences
on insulin levels and the insulin resistance syndrome: and analysis of women
twins. Am J Epidemiol 1996; 143:323-332.
4. Beck-Nielson, H, Vaag A, Poulsen P, GasterM. Metabolic and genetic influence
on glucose metabolism in type 2 diabetic subjects-experience from relative and
twin studies. Best Pract Res Clin Endocrinol Metab 2003; 17:445-467.
5. Altshuler D, Hirschhorn JN, Klannemark M, et al. The common PPARgamma
Pro12Ala polymorphism is associated with decreased risk of type 2 diabetes. Nat
Genet 2000; 26(1):76-80.

123
6. Sladek R, Rocheleau G, Rung J et al. A genome-wide association study identifies
novel risk loci for type 2 diabetes. Nature 2007; 445:881-885.
7. The Wellcome Trust Case Control Consortium: genome-wide association study of
14,000 cases of seven common diseases and 3,000 shared controls. Nature 2007;
447:661-678.
8. Saxena R , Diabetes Genetics Initiative of Broad Institute of Harvard and MIT,
Lund University, and Novartis Institutes of Biomedical Research, et al. Genome-
wide association analysis identifies loci for type 2 diabetes and triglyceride levels.
Science 2007; 316:1331-1336.
9. Scott LJ, Mohlke KL, Bonnycastle LL, Willer CJ, Li Y, et al. A genome-wide
association study of type 2 diabetes in Finns detects multiple susceptibility
variants. Science 2007; 316:1341-1345.
10. Zeggini E, Weedon MN, Lindgren CM, Frayling TM, et al. Wellcome Trust Case
Control Consortium (WTCCC). Replication of genome-wide association signals
in UK samples reveals risk loci for type 2 diabetes. Science 2007; 316:1336-1341.
11. Grarup N, Rose CS, Andersson EA, et al. Studies of Association of variants near
the HHEX, CDKN2A/B, and IGF2BP2 genes with type 2 diabetes and impaired
insulin release in 10,705 Danish subjects: validation and extension of genome-
wide association studies. Diabetes 2007; 56:3105-3111.

124
12. Zeggini E, Scott LJ, Saxena R and Voight BF for the Diabetes Genetics
Replication and Meta-Analysis (DIAGRAM) Consortium. Meta-analysis of
genome-wide association data and large-scale replication identifies additional
susceptibility loci for type 2 diabetes. Nature Genetics 2008; 40:638-645.
13. Lyssenko V, Jonsson A, Almgren P, Pulizzi N, et al. Clinical risk factors,
variants, and the development of type 2 diabetes. N Engl J Med 2008; 359: 2220-
2232.
14. Yasuda K, Miyake K, Horikawa Y, et al. Variants in KCNQ1 are associated with
susceptibility to type 2 diabetes mellitus. Nat Genet 2008; 40:1092–1097.
15. Unoki H, Takahashi A, Kawaguchi T, et al. SNPs in KCNQ1 are associated with
susceptibility to type 2 diabetes in East Asian and European populations. Nat
Genet. 2008; 40(9):1098-102.
16. Lango H, Palmer C, Morris A, et al. Assessing the combined impact of 18
common genetic variants of modest effect sizes on type 2 diabetes risk. Diabetes
2008; 57:3129–3135.
17. Cauchi S, Proença C, Choquet H, Gaget S, De Graeve F, Marre M, Balkau B,
Tichet J, Meyre D, Vaxillaire M, Froguel P. Analysis of novel risk loci for type 2
diabetes in a general French population: the D.E.S.I.R. study. J Mol Med. 2008;
86(3):341-8.
18. Van Hoek M, Dehgan A, Witteman JC, et al. Predicting type 2 diabetes based on
polymorphisms from genome wide association studies: a population-based study.
Diabetes 2008; 57:3122–3128.

125
19. Tabara Y, Osawa H, Kawamoto R, et al. Replication study of candidate genes
associated with type 2 diabetes based on genome-wide screening. Diabetes 2009;
58:493-498.
20. Wu Y, Li H, Loos RJF, et al. Common variants in CDKAL1, CDKN2A/B,
IGF2BP2, SLC30A8, and HHEX/IDE genes are associated with type 2 diabetes
and impaired fasting glucose in a Chinese Han population. Diabetes 2008;
57:2834-2842.
21. Lyssenko V, Nagorny CL, Erdos MR, et al. Common variant in MTNR1B
associated with increased risk of type 2 diabetes and impaired early insulin
secretion. Nat Genet. 2009; 41(1):82-8.
22. Jonsson A, Isomaa B, Tuomi T, et al. A variant in the KCNQ1 predicts future
types 2 diabetes and mediates impaired insulin secretion. Diabetes. 2009; Jul 7.
[Epub ahead of print]
23. Sparsø T, Bonnefond A, Andersson E, et al. G-allele of intronic rs10830963 in
MTNR1B confers increased risk of impaired fasting glycemia and type 2 diabetes
through an impaired glucose-stimulated insulin release: studies involving19,605
Europeans. Diabetes. 2009; 58(6):1450-6.
24. Palmer ND, Goodarzi MO, Langefeld CD, et al. Quantitative trait analysis of
type 2 diabetes susceptibility loci identified from whole-genome association
studies in the Insulin Resistance Atherosclerosis Family study. Diabetes 2008;
57:1093-1100.

126
25. Ruchat SM, Elks CE, Loos RJF, et al. Association between insulin secretion,
insulin sensitivity and type 2 diabetes susceptibility variants identified in genome-
wide association studies. Acta Diabetol. 2008; epub.
26. Stancˇa´kova´ A, Pihlajama¨J, Kuusisto J, et al. Single Nucleotide Polymorphism
rs7754840 of CDKAL1 is associated with impaired insulin secretion in
nondiabetic offspring of type 2 diabetic subjects and in a large sample of men
with normal glucose tolerance. J Clin Endocrinol Metab 2008; 93(5):1924-1930.
27. Grarup N, Andersen G, Krarup NT, et al. Association testing of novel type 2
diabtes risk alleles in the JAZF1, DCD123/CAMK1D, TSPAN8, THADA,
ADAMTS9, and NOTCH2 loci with insulin release, insulin sensitivity, and
obesity in a population-based samples of 4516 glucose-tolerant middle-aged
Danes. Diabetes. 2008; 57(9): 2534-40.
28. Holmkvist J, Banasik K, Andersen G, et al. Type 2 diabetes associated with
minor allele of rs2237895 KCNQ1 associates with reduced insulin release
following oral glucose load. PLoS One. 2009 Jun 11; 4(6):e5872.
29. Prokopenko I, Langenberg C, Florez JC, et al. Variants in MTRNB1 influence
fasting glucose levels. Nat Genet. 2009; 41(1):77-81.
30. Lewis JP, Palmer ND, Hicks PJ, Sale MM, et al. Association analysis in African
Americans of European-derived type 2 diabetes single nucleotide polymorphisms
from whole-genome association studies. Diabetes 2008; 57:2220-2225.

127
31. Waters KM, Stram DO, Hassanein MT, et al. Consistent association of type 2
diabetes risk varaints found in Europeans in diverse racial and ethnic groups.
PLoS Genet 2010;6(8):e1001078.
32. Cooke JN, Hester JM, Ng MCY, et al. Genetic risk assessment of type 2
diabetes-associated polymorphisms in African Americans. Diabetes Care
2012;35:287-292.
33. Saxena R, Elbers EE, Guo Y, et al. Large-scale gene-centric meta-analysis across
39 studies identifies type 2 diabetes loci. Am J Hum Gen 2012; epub ahead of
print.
34. Palmer ND, McDonough CW, Hiscks PJ, et al. A genome-wide association
search for type 2 diabetes genes in African Americans. PLoS One
2012;7(1):e29202.
35. Kaplan RC, Kingsley LA, Gange SJ, et al. Low CD4+ T-cell count as a major
athersclerosis risk factor in HIV-infected women and men. AIDS 2008;22:1615-
1624.
36. Triant VA, Lee H, Hadigan C, Grinspoon SK. Increased acute myocardial
infarction rates and cardiovascular risk factors among patients with human
immunodeficiency virus disease. J Clin Endocrinol Metab 2007;92:2506-2512.
37. Bucher HC, Richter W, Glass TR, et al. Small dense lipoproteins, apolipoprotein
B and risk of coronary events in HIV-infected patients on antiretroviral therapy:
The Swiss HIV Cohort Study. J Acquir Immune Defic Syndr 2011; e-pub ahead
of print.

128
38. Salter ML, Lau B, Go VF, Mehta SH, Kirk GD. HIV infection, immune
suppression, and uncontrolled viremia are associted with increased
multimorbidity among aging injection users. Clin Infect Dis 2011;53:1256-1264.
39. Brown TT, Cole SR, Li X, et al. Antiretroviral therapy and the prevalence and
incidence of diabetes mellitus in the Multicenter AIDS Cohort Study. Arch Intern
Med 2005;165:1179-1184.
40. Tien PC, Schneider MF, Cole SR, et al. Antiretroviral therapy exposure and
incidence of diabetes mellitus in the Women’s Interagency HIV Study. AIDS
2007;21:1739-1745.
41. Brar I, Shuter J, Thomas A, Daniels E, Absalin J. A comparison of factors
associated with prevalent diabetes mellitus among HIV-infected antiretroviral-
naïve individuals versus individuals in the National Health and Nutritional
Examination Survey cohort. J Acquir Immune Defic Syndr 2007;45:66-71.
42. Ledergerber B, Furrer H, Rickenbauch M, et al. Fractors associated with the
incidence of tupe 2 diabetes mellitus in HIV-infected particpants in the Swiss HIV
Cohort Study. Clin Infect Dis 2007;45:111-119.
43. Barkan SE, Melnick SL, Preston-Martin S, et al. The Women’s Interagency HIV
Study. Epidemiology 1998;9:117-25.
44. Bacon MC, Von Wyl V, Alden C, et al. The Women's Interagency HIV Study: an
observational cohort brings clinical sciences to the bench. Clin Diagn Lab
Immunol 2005;12(9): 1013-19.

129
45. American Diabetes Association. Diagnosis and classification of diabetes
mellitus. Diabetes Care 2010;33(1S):S62-69S.
46. Price AL, Patterson NJ, Plenge RM, et al. Principal components analysis corrects
for stratification in genome-wide association studies. Nat Genet 2006;38(8):904-
909.
47. Salanti G, Southam L, Altshuler D, et al. Underlying genetic models of
inheritance in established type 2 diabetes associations. Am J Epidemiol 2009;
170(5): 537-45.
48. Rotger M, Gsponer T, Martinez R, et al. Impact of single nucleotide
polymorphisms and of clinical risk factors on new-onset diabetes mellitus in HIV-
infected individuals. Clin Infect Dis 2010;51:1090-1098.
49. Below JE, Gamazon ER, Morrison JV, et al. Genome-wide association and meta-
analysis in populations from Starr County, Texas and Mexico City identify type 2
diabetes susceptibility lovi and enrichment for expression quantitative trait loci in
the top signals. Diabetologia 2011.; 54:2047-2055.

130
Chapter 7: Summary and Future Directions

The focus of this dissertation was to characterize individual genetic ancestry in a
multi-ethnic cohort of HIV-infected and uninfected women from six geographical regions
across the U.S. and to incorporate genetic ancestry components to control for population
stratification in genetic association studies. Genetic ancestry was evaluated two ways
using 168 markers. A hierarchical clustering algorithm called STRUCTURE was used to
classify ancestral proportions according to known reference populations in HapMap. The
second approach used principal components analysis (PCA) to reveal hidden population
substructure without assumptions of ancestral populations. The genetic ancestry
components derived from PCA were utilized in the subsequent statistical models
evaluating genetic associations with HAART response and incident diabetes.
The second analysis was a candidate gene study of variants in the xenobiotic
metabolism and distribution pathway. Specific genes in this pathway that are implicated
in the metabolism of HIV antiretroviral drugs were selected; it was hypothesized that
genetically-modified drug metabolism would influence virologic supression. Population
substructure had a major impact on the association between the CYP2B6 metabolizer
phenotype and virologic suppression after treatment with non-nucleoside reverse
transcriptase inhibitors (NNRTIs). After adjusting for individual genetic ancestry,
intermediate and slow metabolizers of CYP2B6 substrates, NNRTIs, were many times
more likely to achieve virologic suppression compared to extensive metabolizers

131
(OR
intermediate metabolizers
= 2.90, 95% CI 0.79-12.28 and OR
slow metabolizers
=13.44, 95% CI 1.66-
infinity).
The third analysis was a validation study of established European-derived type
two diabetes (T2D)-risk alleles in respect to T2D incidence in HIV-infected individuals.
The hypothesis was that the alleles that contribute risk of T2D in the HIV-uninfected
population would also confer risk in the HIV-infected population. The impact of
population substructure was not as evident in these association models testing T2D-risk
loci due to a priori evidence of heterogeneous effects between European and African-
American populations; the association models were therefore stratified by race
accordingly. Despite stratification, incorporation of genetic ancestry components as
model covariates was critical since the majority of the non-African subset were self-
reported Hispanics. The second aim was to explore effect modification by nucleoside
reverse transcriptase inhibitors (NRTI), as NRTI exposure is a T2D risk factor in the
HIV-infected population. The interaction of two SNPs in the IGFBP2 gene with NRTI
dose and NRTI dose-duration was statistically significant in the Whites, Hispanics, and
“other” group. There was also evidence of effect modification by NRTI exposure in the
associations of the remaining 16 T2D-risk alleles among the Whites, Hispanics, and other
group; however, the study had limited power to detect interactions. The nominal effects
of these T2D-risk loci were not observed in African-American women (except for the
highly replicated association of a variant in TCF7L2), which was consistent with
expectations.

132
Independent genetic association models evaluated in the WIHS cohort
incorporated the genetic ancestry components to safeguard from potentially spurious
results due to confounding by ancestry, as demonstrated in the CYP2B6 metabolizer
phenotype association. Recently, WIHS women were genotyped using a genome-wide
SNP chip, which will allow estimation of individual genetic ancestry at a higher
resolution. Further investigation of the non-Yoruban African cluster that was revealed in
WIHS women in Chapter 4 will be possible with the genome-wide data.
While this dissertation addressed estimation of global genetic ancestry, local
genetic ancestry is also important to characterize in admixed populations, such as the
WIHS. Local genetic ancestry describes ancestry at a particular chromosomal locus. The
nuance of genetic ancestry is that while two individuals may share similar proportions of
ancestry from distinct populations across the genome, they can differ in ancestry at
particular loci. This is particularly important in reference to T2D-risk loci since this
dissertation and others (Saxena 2012, Palmer 2012) did not observe associations with
these loci in African-Americans. In theory, risk of T2D in African-Americans with
European ancestry at these T2D-risk loci would be similar to that of Whites. As a future
direction, genome-wide variation can be utilized to measure local genetic ancestry in the
WIHS using established algorithms, such as HAPMIX (Price 2009).
Replication of the association between the CYP2B6 metabolizer phenotype and
virologic response to NNRTIs in an independent population would lend support to the
results presented in this dissertation. The genome-wide data in the WIHS can be utilized

133
to acquire additional women who initiated a NNRTI-containing regimen after genotyping
for our study began.
In reference to the potential effect modification of T2D genetic associations with
exposure to NRTIs, future work should focus on teasing out the effect of NRTI dose
versus duration. The WIHS dataset can be utilized to test the main effect of NRTIs over
the entire follow-up by modeling cumulative NRTI exposure (duration) along with dose
as separate time-dependent variables. If the effect of dose and duration are independently
or differentially associated with T2D incidence, interaction terms with T2D-risk loci
should be fitted separately. However, the modifying effect of dose and duration
separately requires a very large sample size. Pooling data across cohorts, for example
between the WIHS and Swiss HIV Cohort studies, is warranted to properly address this
question.
In conclusion, the pharmacogenetic associations investigated in this dissertation
have potential for impacting clinical practice. Medical decisions may be influenced by
knowledge of the patient’s genetic profile. Further work to establish the synergistic
effect of HIV therapy and genetic variation on virological and immunological response
outcomes is critical to advancing the field of pharmacogenetics in the context of HIV-
infection. Additionally, further work to establish whether NRTI dose or duration or a
combination of the two, significantly modifies the genetic effect of T2D is worthwhile
since NRTIs are the backbone of all antiretroviral therapy regimens.

134
Comprehensive References

An P, Winkler CA. Host genes associated with HIV/AIDS: advances in gene discovery.
Trends Genet 2010;26(3):119-31.

Aithal GP, Day CP, Kesteven PJ, Daly AK. Association of polymorphisms
in the cytochrome P450 CYP2C9 with warfarin dose requirement and risk of bleeding
complications. Lancet. 1999; 353:717–719.

Altshuler D, Hirschhorn JN, Klannemark M, et al. The common PPARgamma Pro12Ala
polymorphism is associated with decreased risk of type 2 diabetes. Nat Genet 2000;
26(1):76-80.

American Diabetes Association. Diagnosis and classification of diabetes mellitus.
Diabetes Care 2010;33(1S):S62-69S.

Anastos K, Schneider MF, Gange SJ, et al. The association of race, sociodemographic,
and behavioral characteristics with response to highly active antiretroviral therapy in
women. J Acquir Immune Defic Syndr. 2005; 39:537-544.

Anderson PL, Lamba J, Aquilante CL, Schuetz E, Fletcher CV. Pharmacogenetic
characteristics of indinavir, zidovudine, and lamivudine therapy in HIV-infected adults: a
pilot study. J Acquir Immune Defic Syndr 2006;42:441-449.

Anastos K, Kalish LA, Palacio H, Benson CA, Delapenha R, Chirgwin K, Stonis L,
Telzak EE. Prevalence of and risk factors for tuberculin positivity and skin test anergy in
HIV-1-infected and un-infected at-risk women. Women’s Interagency HIV Study
(WIHS). J Acquir Immune Defic Syndr 1999; 21(2):141-147.

Aquilante CL, Langaee TY, Lopez LM, Yarandi HN, Tromberg JS, Mohuczy D, et al.
Influence of coagulation factor, vitamin K epoxide reductase complex subunit 1, and
cytochrome P450 2C9 gene polymorphisms on warfarin dose requirements. Clin
Pharmacol Ther. 2006; 79:291–302.

Arab-Alameddine M, Di Iulio J, Buclin T, Rotger M, Lubomirov R, Cavassini M, et al;
the Swiss HIV Cohort Study. Pharmacogenetics-based population pharmacokinetic
analysis of efavirenz in HIV-1-infected individuals. Clin Pharmacol Ther. 2009; 85:485–
94.

135
Bacon MC, Von Wyl V, Alden C, Sharp G, Robison E, Hessol N, et al. The Women's
Interagency HIV Study: an observational cohort brings clinical sciences to the bench.
Clin Diagn Lab Immunol 2005;12(9): 1013-19.

Barban M, Padrini R, Ieiri I, Otsubo K, Kashima T, Kimura S, Kijima S, Echizen H.
Different contributions of polymorphisms in VKORC1 and CYP2C9 to intra- and inter-
population differences in maintenance dose of warfarin in Japanese, Caucasians and
African-Americans. Pharmacogenet Genomics. 2006; 16:101–110.

Barkan SE, Melnick SL, Preston-Martin S, Weber K, Kalish LA, Miotti P, oung M,
Greenblatt R, Sacks H, Feldman J. The Women’s Interagency HIV Study.
Epidemiology 1998;9:117-25.

Barnholtz-Sloan JS, Chakraborty R, Seller TA, Schwartz AG. Examining population
stratification via individual ancestry estimates versus self-reported race. Cancer
Epidemiol Biomarkers Prev (2005) 14:1545-1551.

Barlett JG, Lane HC. Guidelines for the use of antiretroviral drugs in HIV-1-infected
adults and adolescents. Clinical Guidelines for the treatment and management of HIV
Infection. Edited by Infection PCPTHIV. USA, Department of Health and Human
Services; 2005:1-118.

Beck-Nielson, H, Vaag A, Poulsen P, GasterM. Metabolic and genetic influence on
glucose metabolism in type 2 diabetic subjects-experience from relative and twin studies.
Best Pract Res Clin Endocrinol Metab 2003; 17:445-467.

Below JE, Gamazon ER, Morrison JV, et al. Genome-wide association and meta-
analysis in populations from Starr County, Texas and Mexico City identify type 2
diabetes susceptibility lovi and enrichment for expression quantitative trait loci in the top
signals. Diabetologia 2011.; 54:2047-2055.

Benjamini Y, Hochberg Y. Controlling the false discovery rate: a practical and powerful
approach to multiple testing. JR Stat Soc 1995; 57:289-300.

Bosma PJ, Chowdhury JR, Bakker C, Gantla S, de Boer A, Oostra BA,Londhout D,
Tytgat GN, Jansen PL, Oude Elferink RP. The genetic bases of the reduced expression of
bilirubin UDP-glucuronosyltransferase 1 in Gilbert’s syndrome. 1996 N Engl J Med
334(12):802-803.

Brar I, Shuter J, Thomas A, Daniels E, Absalin J. A comparison of factors associated
with prevalent diabetes mellitus among HIV-infected antiretroviral-naïve individuals
versus individuals in the National Health and Nutritional Examination Survey cohort. J
Acquir Immune Defic Syndr 2007;45:66-71.

136
Bresolin T, de Freitas Rebelo M, Celso Dias Bainy A. Expression of PXR, CYP3A and
MDR1 genes in the liver of zebrafish. Comp Biochem Physiol C Toxicol Pharmacol.
2005; 140(3-4):403-407.

Brown TT, Cole SR, Li X, et al. Antiretroviral therapy and the prevalence and incidence
of diabetes mellitus in the Multicenter AIDS Cohort Study. Arch Intern Med
2005;165:1179-1184.

Brumme ZL, Dong WWY, Chan KJ, Hogg RS, Montaner JSG, O’Shaughnessy MV,
Harrigan PR. Influence of polymorphisms within the CX
3
CR1 and MDR-1 genes on
intial antiretroviral therapy response. AIDS 2003;17:201-208.

Bucher HC, Richter W, Glass TR, et al. Small dense lipoproteins, apolipoprotein B and
risk of coronary events in HIV-infected patients on antiretroviral therapy: The Swiss HIV
Cohort Study. J Acquir Immune Defic Syndr 2011; e-pub ahead of print.

Burchell B. Genetic variation of human UDP-gluconosyltransferase:implications in
disease and drug glucronidation. Am J Pharmacogenomics 2003; 3:37-52.

Burk O, Koch I, Raucy J, Hustert E, Eicherlbaum M, Brockmoller N, et al. The
induction of cytochrome P450 3A5 (CYP3A5) in the human liver and intestine is
mediated by the xenobiotic sensors pregane X receptor (PXR) and constitutively
activated receptor (CAR). J Biol Chem. 2004; 279(37):3879-3885.

Bushman FD, Malani N, Fernandes J, D’orso I, Cagney G, Diamond TL, et al. Host cell
factors in HIV replication: meta-analysis of genome-wide studies. PloS Pathogens
2009;5:e1000437.

Caldwell MD, Berg RL, Zhang KQ, Glurich I, Schmelzer JR, Yale SH, Vidaillet HJ,
Burmester JK. Evaluation of genetic factors for warfarin dose prediction. Clin Med Res.
2007; 5:8 –16.

Campbell DC, Ogburn EL, Lunetta KL, Lyon HN, Freedman ML, Groop LC, et al.
Demonstrating stratification in a European American population. Nat Genet
2005;37(8):868-872.

Carr DF, la Porte CJ, Pirmohamed M, Owen A, Cortes CP. Haplotype structure of
CYP2B6 and association with plasma efavirenz concentrations in a Chilean HIV cohort.
J Antimicrob Chemother 2010; 65:1889-1893.

Cauchi S, Proença C, Choquet H, Gaget S, De Graeve F, Marre M, Balkau B, Tichet J,
Meyre D, Vaxillaire M, Froguel P. Analysis of novel risk loci for type 2 diabetes in a
general French population: the D.E.S.I.R. study. J Mol Med. 2008; 86(3):341-8.

137
Chaponda M and Pirmohamed M. Hypersensitivity reactions to HIV therapy. Br J Clin
Pharmacol 2011;71(5):659-671.

Chen Y, Ferguson S, Negishi M, Goldstein JA. Induction of human CYP2C9 by
rifampicin, hyperforin, and Phenobarbital is mediated by the pregnane X receptor. J
Pharmacol Exp Ther 2004; 308:495-501.

Clarke A, Stein CR, Townsend ML. Drug-drug interactions with HIV antiretroviral
therapy. US Pharm. 2008; 33(4):HS-3-HS-21.

Clinical Trials Information on PACTG 382, NIH website (Accessed on December 20,
2009 at http://clinicaltrials.gov/ct2/show/NCT00000893)

Cooke JN, Hester JM, Ng MCY, et al. Genetic risk assessment of type 2 diabetes-
associated polymorphisms in African Americans. Diabetes Care 2012;35:287-292.

Cox DG, and Kraft P. Quantification of the power of Hardy-Weinberg equilibrium
testing to detect genotyping error. Hum Hered 2006; 61:10-4.

Cressey TR, and Lallemant M. Pharmacogenetics of antiretroviral drugs for the
treatment of HIV-infected patients: an update. Infection, Genetics, and Evolution
2007;7:333-42.

Crum NF, Riffenburgh RH, Wegner S et al. Comparisons of causes of death and
mortality rates among HIV-infected persons; analysis of the pre-, early, and late HAART
eras. J Acquir Immune Defic Syndr 2006; 41(2):194-200.

de Maat MM, Ekhart GC, Huitema AD, et al. Drug interactions between antiretroviral
drugs and comedicated agents. Clin Pharmacokinet. 2003; 42:223-82.

Desta Z, Zhao XJ, Shin JG, Flockhart DA. Clinical significance of the cytochrome P450
2C19 genetic polymorphism. Clin Pharmacokinet. 2002;41:913-58.

Devlin B, Roeder K, Wasserman L. Genomic control, a new approach to genetic-based
association studies. Theor Popul Biol 2001;60:155-166.

Ding X, Kaminsky LS. Human extrahepatic cytochrome P450: function in xenobiotic
metabolism and tissue-selective chemical toxicity in the respiratory and intestinal tracts.
Annu Rev Pharmacol Toxicol 2003;43:149-73.

Doerge RW, Churchill GA. Permutation tests for multiple loci affecting a quantitative
characteristic. Genetics 1996; 142:285-94.

138
Dooley KE, Flexner C, Andrade AS. Drug interactions involving combination antiviral
therapy and other anti-infective agents: repercussions for resource-limited countries. J
Infect Dis 2008;198:948-61.

Dybul M, Fauci AS, Bartlett JG, Kaplan JE, Pau AK; Panel on Clinical Practices for
Treatment of HIV. Guidelines for using antiretroviral agents among HIV-infected adults
and adolescents. Ann Intern Med 2002;137:381-433.

Edlund CK, Lee WH, Li D, Van Den Berg DJ, and Conti DV. Snagger: a user-friendly
program for incorporating additional information for tagSNP selection. BMC
Bioinformatics 2008;9:174-187.

Egger M, May M, Chene G, Ledergerber B, et al.; ART Collabortation. Prognosis of
HIV-1 infected patiens starting HAART; a collaborative analysis of prospective studies.
Lancet 2002; 360:119-129.

Eichelbaum M, Spannbrucker N, Steinke B, and Dengler HJ. Defective Noxidation
of sparteine in man: a new pharmacogenetic defect. Eur J Clin Pharmacol 1979;
16:183–187.

Eichelbaum M, Baur MP, Dengler HJ, Osikowska-Evers BO, Tieves G, Zekorn C,
Rittner C. Chromosomal assignment of human cytochrome P-450
(debrisoquine/sparteine type) to chromosome 22. Br J Clin Pharmacol 1987; 23:455-458.

Eichelbaum M, Fromm MF, Schwab M. Clinical aspects of the MDR1 (ABCB1) gene
polymorphism. The Drug Monit. 2004; 26(2):180-185.

Engelhardt BE and Stephens M. Analysis of population structure: a unifying framework
and novel methods based on sparse factor analysis. PloS Genet 2010;6(9) e1001117.

Eskin E. Increasing power in association studies by using linkage disequilibrium
structure and molecular function as prior information. Genome Res 2008; 18:653-60.

Evans DA, Mahgoub A, Sloan TP, Idle JR, Smith RL. A family and population study of
the genetic polymorphism of ebrisoquine oxidation in a white British population. J Med
Genet 1980; 17:102-105.

Falush D, Stephens M, Pritchard JK. Inference of population structure using multilocus
genotype data: linked loci and correlated allele frequencies. Genetics 2003;164:1567-
1587.

Faucette SR, Sueyoshi T, Smith CM, Negishi M, Lecluyse EL, Wang H. Differential
regulation of hepatic CYP2B6 and CYP3A4 genes by the constitutive andostane receptor
but not pregnane receptor X. J Pharmacol Exp Ther 2006; 317:1200-1209.

139
FDA information on antiretroviral drugs. (Accessed on Dec 5, 2008 at
http://www.fda.gov/ForConsumers/byAudience/ForPatientAdvocates/HIVandAIDSActiv
ities/ucm118915.htm)

Fellay J, Marzolini C, Decosterd L, Golay KP, Baumann P, Buclin T, Telenti A, Eap CB.
Variations of CYP3A activity induced by antiretroviral treatment in HIV-1 infected
patients. Eur J Clin Pharmacol. 2005; 60(12):865-73.

Fellay J, Marzolini C, Meaden ER, et al. Response to antiretroviral treatment in HIV-1
infected individuals with allelic variants of the multidrug resistance transporter 1: a
pharmacogenetics study. Lancet. 2002; 359:30-6.

Floren KC, Wiznia A, Hayashi S et al. Nelfinavir pharmacokinetics in stable HIV-
positive children: Peadtric AIDS Clinical Trial Group Protocol 377. Pediatrics
2003;112:e220-27.

Freedman ML, Reich D, Penney KL, et al. Assessing the impact of population
stratification on genetic association studies. Nat Genet 2004;36(4):388-393.

French AL, Gawel SH, Hershow R, Benning L, Hessol NA, Levine AM, et al. Trends in
mortality and causes of death among women with HIV in the United States: A 10-year
study. J Aquir Immune Defic Syndr 2009;51:399-406.

Furuta T, Shirai N, Xiao F, Ohashi K, Ishizaki T. Effect of high dose lansoprazole on
tragastric pH in subjects who are homozygous extensive metabolizers of cytochrome
P4502C19. Clin. Pharmacol. Ther. 2001; 70:484-492.

Furuta T, Ohashi K, Kamata T et al. Effect of genetic differences in omeprazole
metabolism on cure rates for Helicobacter pylori infection and peptic ulcer. Ann Intern
Med. 1998; 129:1027-30.

Furuya H, Fernandez-Salguero P, Gregory W, Taber H, Steward A, Gonzalez FJ, Idle JR.
Genetic polymorphisms of CYP2C9 and its effect on warfarin maintenance dose
requirement in patients undergoing anticoagulation therapy. Pharmacogenetics 1995;
5:389-92.

Gardiner SJ, Begg EJ. Pharmacogenetics, drug metabolizing enzymes, and clinical
pratice. Pharmacol Rev 2006; 58:521-90.

Gange SJ, Barron Y, Greenblatt RM, Anastos K et al. Effectiveness of highly active
antiretroviral therapy among HIV-1 infected women. J Epidemiol Community Health
2002;56:153-159.

140
Gellner K, Eiselt R, Hustert E, Arnold H, Koch I, Haberl M, et al. Genomic organization
of the CYP3A locus: identification of the new, inducible CYP3A gene.
Pharmacogenetics 11:111-21.

Gonzalez FJ, Vilbois F, Hardwick JP, McBride OW, Nebert DW, Gelboin HV, and
Meyer UA. Human debrisoquine 4-hydroxylase (P450IID1): cDNA and deduced amino
acid sequence and assignment of the CYP2D locus to chromosome 22. Genomics 1988;
2(2): 174-9.

Gorroochurn P, Hodge SE, Heiman G, et al. Effect of population stratification on case-
control association studies. II False-positive rates and their limiting behavior as number
of subpopulations increases. Hum Hered 2004;58(1):40-48.

Grarup N, Rose CS, Andersson EA, et al. Studies of Association of variants near the
HHEX, CDKN2A/B, and IGF2BP2 genes with type 2 diabetes and impaired insulin
release in 10,705 Danish subjects: validation and extension of genome-wide association
studies. Diabetes 2007; 56:3105-3111.

Grarup N, Andersen G, Krarup NT, et al. Association testing of novel type 2 diabtes risk
alleles in the JAZF1, DCD123/CAMK1D, TSPAN8, THADA, ADAMTS9, and
NOTCH2 loci with insulin release, insulin sensitivity, and obesity in a population-based
samples of 4516 glucose-tolerant middle-aged Danes. Diabetes. 2008; 57(9): 2534-40.

Gullick RM, Mellors JM, Havlir D et al. Treatment with indinavir, zidovudine and
lamivudine in adults with human immunodeficiency virus infection and prior
antiretroviral therapy. N Eng J Med 1997; 337:734-9.

Haas DW, Wu H, Li H, Bosch R, Lederman M, Kuritzkes D, et al. MDR1 gene
polymorphisms and phase 1 viral decay during HIV-1 infection: an adult AIDS clinical
trials group study. J Acquir Immune Defic Syndr. 2003; 34:295-298.

Haas DW, Ribaudo HJ, Kim RB, et al. Pharmacogenetics of efavirenz and central
nervous system side effects: An Adult AIDS Clinical Trials Group study. AIDS 2004;
18:2391-400.

Haas DW, Smeaton LM, Shafer RW, Robbins GK, Morse GD, Labbe L, et al.
Pharmacogenetics of long-term responses to antriretroviral regimens containg efavirenz
and/or nelfinavir: An Adult AIDS Clinical Trials Group Study. J Infect Dis. 2005;
192:1931-1942.

Haas DW, Koletar SL, Laughlin L et al. Associations between CYP2B6 polymorphisms
and pharmacokinetics after a single dose of nevirapine or efavirenz in African
Americans. J Infect Dis 2009;199:872-880.

141
Haffner SM, Stern MO, Hazuda HP et al. Increased insulin concentrations in nondiabetic
offspring of diabetic parents. NEJM 1988; 319:1297-1301.

Halder I, Yang BZ, Kranzler HR, et al. Measurement of admixture proportions and
description of admixture structure in different U. S. populations. Hum Mutat
2009;30(9):1299-1309.

Halder I, Yang BZ, Kranzler HR, Stein MB, Shriver MD, Gelernter J. Measurement of
admixture proportions and description of admixture structure in different U. S.
populations. Hum Mutat 2009;30(9):1299-1309.

Helgason A, Yngvafottir B, Hrafnkelsson B, Gulcher J, Stefansson K. An Icelandic
example of the impact of population structure on association studies. Nat Genet
2005;37(1):90-95.

Higashi MK, Veenstra DL, Kondo LM, Wittkowsky AK, Srinouanprachanh SL, Farin
FM, Rettie AE. Association between CYP2C9 genetic variants and anticoagulation-
related outcomes during warfarin therapy. JAMA 2002; 287: 1690-1698.

Hiratsuka M, Takekuma Y, Endo N, et al. Allele and genotype frequencies of CYP2B6
and CYP3A5 in the Japanese population. Eur J Clin Pharmacol 2002;58:417-21.

Hirschhorn JN, Daly MJ. Genome-wide association studies for common disease and
complex traits. Nat Rev Genet 2005;6(2):95-108.

Hoffmann MH, Blievernicht JK, Klein K, Saussele T, Schaeffeler E, Schwab M, Zanger
UM. Aberrant splicing caused by single nucleotide polymorphism 516G>T [Q172H], a
marker of CP2B6*6, is responsible for decreased expression and activity of CP2B6 in the
liver. J Pharmacol Exp Ther. 2008; 325:284-92.

Hoggart CJ, Parra EJ, Shriver MD, et al. Control of confounding of genetic associations
in stratified populations. Am J Hum Genet 2003;72:1492-1504.

Holmkvist J, Banasik K, Andersen G, et al. Type 2 diabetes associated with minor allele
of rs2237895 KCNQ1 associates with reduced insulin release following oral glucose
load. PLoS One. 2009 Jun 11; 4(6):e5872.

Hooshyar D, Hanson DL, Wolfe M, et al. Trends in perimortal conditions and mortality
rates among HIV-infected patients. AIDS 2007; 21:2093-2100.

Hulgan T, Shepard BE, Raffanti SP, Fusco JS, Beckerman R, Barkanic G, Sterling TR.
Absolute count and percentage of CD4 lymphocytes are independent predictors of
disease progression in HIV-infected persons initiating highly active antiretroviral therapy.
J Infect Dis 2007;195:425-31.

142

Ibeanu GC, Goldstein JA, Meyer U et al. Identification of new human CYP2C19 alleles
(CYP2C19*6 and CYP2C19*2B) in a Caucasian poor metabolizer of mephenytoin. J
Pharmacol Exp Ther. 1998; 286:1490-5.

Ingelman-Sundberg M, Sim SC, Gomez A, and Rodriguez-Antona C. Influence of
cytochrome P450 polymorphisms on drug therapies: pharmacogenetic,
pharmacoepigenetic, and clinical aspects. Pharmacol Ther 2007;116:496-526.

Inomata S, Nagashima A, Itagaki F, Homma M, Nishimura M, Osaka Y, et al. CYP2C19
genotype affects diazepam pharmacokinetics and emergence from general anesthesia.
Clin Pharmacol Ther. 2005 Dec;78(6):647-55.

International HapMap Consortium. The International HapMap Project. Nature
2003;426:789-796.

Ioannidis JP. Non-replication and inconsistency in the genome-wide association setting.
Hum Hered 2007;64:203-213.

Jackson KA, Rosenbaum SE, Kerr BM, et al. A population pharmacokinetic analysis of
nelfinavir mesylate in human immunodeficiency virus-infected subjects enrolled in a
phase III clinical trial. Antimicrob Agents Chemother. 2000; 44:1832–1837.

Johnson JA. Ethnic differences in cardiovascular drug response: potential contribution of
pharmacogenetics. Circulation. 2008; 118:1383-1393.

Jonsson A, Isomaa B, Tuomi T, et al. A variant in the KCNQ1 predicts future types 2
diabetes and mediates impaired insulin secretion. Diabetes. 2009; Jul 7. [Epub ahead of
print]

Jover R, Bort R, Gimez-Lechon MJ, Castell JV. Down-regulation of human CYP3A4 by
the inflammatory signal interleukin-6: molecular mechanism and transcription factors
involved. FASEB J. 2006; 16(13):1799-1801.

Kaplan RC, Kingsley LA, Gange SJ, et al. Low CD4+ T-cell count as a major
athersclerosis risk factor in HIV-infected women and men. AIDS 2008;22:1615-1624.

Keuhl P, Zhang J, Lin Y, Lamba J, Assem M, Schuetz J, et al. Sequence diversity in
CYP3A promoters and chracterization of the genetic basis of polymorphic CYP3A5
expression. Nat Genet 2001 27(4):383-391.

King BP, Khan TI, Aithal GP, Kamali F, Daly AK. Upstream and coding region
CYP2C9 polymorphisms:correlation with warfarin dose and metabolism.
Pharmacogenetics 2004; 14: 813-822.

143

Kirchheiner J and Brockmoller J. Clinical consequences of cytochrome P450 2C9
polyorphisms. Clin. Pharmacol. Ther 2005; 77:1-16.

Klein K, Lang T, Saussele T, et al. Genetic variability of CYP2B6 in populations of
African and Asian origin: allele frequencies, novel functional variants and possible
implications for anti-HIV therapy with efavirenz. Pharmacogenet Genomics 2005;
15:861-73.

Kosoy R, Nassir R, Tian C, et al. Ancestry informative marker sets for determining
continental origin and admixture proportions in common populations in America. Hum
Mutat 2009; 30(1):69-78.

Kovacs A, Montepiedra G, Carey V, et al. Immune reconstitution after receipts of
hightly active antiretroviral therapy in children with advanced or progressive HIV disease
and complete or partial viral load response. J Infect Dis 2005;192:296-302.

Lankisch TO, Moebius U, Wehmeir M, Behrens G, Manns MP, Schmidt RE, Strassburg
CP. Gilbert’s disease and atazanavir: from phenotype to UDP-glucuronosyltranferase
haplotype. Hepatology. 2006; 44(5):1324-1332.

Lama J, Planelles V. Host factors influencing susceptibility to HIV infection and AIDS
progression. Retrovirology 2007;4:52.

Lamba JK, Lin YS, Schuetz EG, and Thummel KE. Genetic contribution to variable
human CYP3A-mediated metabolism. Advanced Drug Delivery Reviews 2002;54:1271-
94.

Lamba V, Lamba J, Yasuda K, Strom S, Davila J, Hancock ML, Fackentha JD, Rogan
PK, Ring B, Wrighton SA, Schuetz EG. Hepatic CYP2B6 expression: gender and ethnic
differences and relationship to CYP2B6 genotype and CAR expression. J Pharmacol Exp
Ther 2003; 307:906-922.

Lamba JK, Strom R, Venkataramanan R, Thummel K , Lin YS, Lui W, Cheng C, Lamba
V, Watkins PB, Scheutz EG. MDR1 genotype is associated with hepatic cytochrome
P450 3A4 basal and induction phenotype. Clin Pharmacol Ther. 2006; 79(4): 325-338.

Lang T, Klein K, Ficher J et al. Extensive genetic polymorphism in the human CYP2B6
gene with impact on expression and function in the human liver. Pharmacogenetics
2001; 11:339-415.

Lang T, Klein K, Richter T, et al. Multiple novel nonsynonymous CYP2B6 gene
polymorphism in Caucasians: demonstration of phenotypic null alleles. J Pharmacol Exp
Ther. 2004;311:34-43.

144
Lango H, Palmer C, Morris A, et al. Assessing the combined impact of 18 common
genetic variants of modest effect sizes on type 2 diabetes risk. Diabetes 2008; 57:3129–
3135.

Lankisch TO, Moebius U, Wehmeir M, Behrens G, Manns MP, Schmidt RE, Strassburg
CP. Gilbert’s disease and atazanavir: from phenotype to UDP-glucuronosyltranferase
haplotype. Hepatology. 2006; 44(5):1324-1332.

Lazo M, Gange SJ, Wilson TE, Anastos K, Ostrow DG, Witt MD, Jacobson LP. Patterns
and predictors of changes in adherence to highly active antiretroviral therapy:
longitudinal study of men and women. CID 2007; 45: 1377-1385.

Ledergerber B, Egger M, Opravil M, et al. Clinical progression and virological failure on
highly active antiretroviral therapy in HIV-1 patients: a prospective cohort study: Swiss
HIV Cohort Study. Lancet 1999; 353:863–886.

Ledergerber B, Furrer H, Rickenbauch M, et al. Fractors associated with the incidence of
tupe 2 diabetes mellitus in HIV-infected particpants in the Swiss HIV Cohort Study. Clin
Infect Dis 2007;45:111-119.

Lederman MM, Connick E, Landay A, et al. Immunologic responses associated with 12
weeks of combination antiretroviral therapy consiting of zidovudine, lamivudine, and
ritonavir: results of the AIDS Clinical Trials Group Protocol 315. J Infect Dis. 1998;
178:70-79.

Lee CR, Goldstein JA and Pieper JA. Cytochrome p450 2C9 polymorphisms: a
comprehensive review of the in-vitro and human data. Pharmacogenetics 2002; 12:251-
263.

Lee SC, Ng SS, Oldenburg J, Chong PY, Rost S, Guo JY, Yap HL, Rankin SC, Khor HB,
Yeo TC, Ng KS, Soong R, Goh BC. Interethnic variability of warfarin maintenance
requirement is explained by VKORC1 genotype in an Asian population. Clin Pharmacol
Ther. 2006; 79:197–205.

Lewis JP, Palmer ND, Hicks PJ, Sale MM, et al. Association analysis in African
Americans of European-derived type 2 diabetes single nucleotide polymorphisms from
whole-genome association studies. Diabetes 2008; 57:2220-2225.

Li X, and Chan WK. Transport, metabolism and elimination mechanisms of anti-HIV
agents. Advanced Drug Delivery Reviews 1999;39:81-103.

Lohmueller KE, Pearce CL, Pike M, Lander ES, Hirschhorn JN. Meta-analysis of
genetic association studies supports a contribution of common variants to susceptibility to
common disease. Nat Genet 2003;33(2):177-82.

145
Lucas GM, Chaisson RE, Moore RD. Survival in an urban HIV-1 clinic in the era of
highly active antiretroviral: a 5-year cohort study. J Acquir Immune Defic Syndr. 2003;
33:321-328.

Lyssenko V, Jonsson A, Almgren P, Pulizzi N, et al. Clinical risk factors, variants, and
the development of type 2 diabetes. N Engl J Med 2008; 359: 2220-2232.

Lyssenko V, Nagorny CL, Erdos MR, et al. Common variant in MTNR1B associated
with increased risk of type 2 diabetes and impaired early insulin secretion. Nat Genet.
2009; 41(1):82-8.

Mahgoub A, Idle JR, Dring DG, Lancaster R, Smith RL. Polymorphic
hydroxylation of debrisoquine in man. Lancet 1977; 2:584–586.

Marzolini C, Paus E, Buclin T, Kim RB. Polymorphisms in human MDR1 (P-
glycoprotein): recent advances and clinical relevance. Clin Pharmacol Ther 2004; 75:13-
33.

Mayer EJ, Newman B, Austin MA, et al. Genetic and environmental influences on
insulin levels and the insulin resistance syndrome: and analysis of women twins. Am J
Epidemiol 1996; 143:323-332.

McCarthy MI, Abecasis GR, Cardon LR, Goldstein DB, Little J, Ioannidis JP, Hirschhorn
JN. Genome-wide association studies for complex traits: consensus, uncertainty and
challenges. Nat Rev Genet 2008;9(5):356-369.

Miksys S, lerman C, Shields PG, Mash DC, Tndale RF. Smoking, alcoholism and
genetic polymorphisms alter CYP2B6 levels in human brain. Nueropharmaology 2003;
45:122-132.

Mocroft A and Lundgren JD. Starting highly active antiretroviral therapy: why, when
and response to HAART. J Antimicrobial Chemotherapy 2004; 54:10-13.

Motsinger AA, Ritchie MD, Shafer RW, et al. Multilocus genetic interactions and
response to efavirenz-containing regimens: An adult AIDS Clinical Trials Group study.
Pharmacogenetics and Genomics 2006; 16:837-845.

Nebert DW and Menon AG. Pharmacogenomic ethnicity and susceptibility genes.
Pharmacogenomics J. 2001;1:19-22.

Neff GW, Jayaweera D, and Sherman KE. Drug-induced Liver Injury in HIV Patients.
Gastroenterology & Hepatology 2006; 2:430-7.

O’Brien SJ, Nelson GW. Human genes that limit AIDS. Nat Genet 2006;36(6):565-574.

146
Ogg MS, Brennan P, Meade T, Humphries SE. CYP2C9*3 allelic variant and bleeding
complications. Lancet 1990; 354:1124.

Owen A, Khoo SH. Pharmacogenetics of antiretroviral agents. Curr Opin HIV AIDS
2008;3(3)288-295.

Padol S, Yuan Y, Thabane M, Padol It, Hunt RH. The effect of CYP2C19
polymorphisms on H. pylori eradication rate in dual and triple first-line PPI therapies; a
meta-analysis. Am J Gastroenterol. 2006; 101(7):1476-1478.

Palella FJ, Delaney KM, Moorman AC, et al. Declining morbidity and mortality among
patients with advanced human immunodeficiency virus infection. N Engl J Med 1998;
338:853–60.

Palella FJ, Baker RK, Moorman AC, Chmiel JS, Wood KC, Brooks JT, Holmberg SD,
and HIV Outpatient Study Investigators. Mortality in the highly active antiretroviral
therapy era: changing causes of death and disease in the HIV outpatient study. J Acquir
Immune Defic. Syndr 2006; 43(1):27-34.

Palmer ND, Goodarzi MO, Langefeld CD, et al. Quantitative trait analysis of type 2
diabetes susceptibility loci identified from whole-genome association studies in the
Insulin Resistance Atherosclerosis Family study. Diabetes 2008; 57:1093-1100.
Perelson AS, Essunger P, Cao YK et al. Decay characteristics of HIV-1 infected
compartments during combination therapy. Nature 1997; 387:188-191.

Palmer ND, McDonough CW, Hiscks PJ, et al. A genome-wide association search for
type 2 diabetes genes in African Americans. PLoS One 2012;7(1):e29202.

Parra EJ, Marcini A, Akey J, et al. Estimating African American admixture proportions
by use of population-specific alleles. Am J Hum Genet 1998;63:1839-1851.

Patterson N, Hattangadi N, Lane B, Lohmueller KE, Hafler DA, Okenberger JR, et al.
Methods for high-density admixture mapping of disease genes. Am J Hum Genet
2004;74(5):979-1000.

Phillips AN, Staszewski S, Weber R, Kirk O, Francioli P, Miller V, et al. Swiss HIV
Cohort Study; Frankfurt HIV Clinic Cohort; EuroSIDA Study Group. HIV viral load
response to antiviral therapy according to the baseline CD4 cell count and viral load after
initiation triple-drug therapy. JAMA 2001; 286: 2560-2567.

Pikuleva IA. Cytochrome P450s and cholesterol homeostasis, Pharmacol Ther. 2006;
112(3):761-773.

147
Piscitelli SC, Gallicano KD. Interactions among drugs for HIV and opportunistic
infections. N Engl J Med 2001; 344:984 –96.

Price AL, Patterson NJ, Plenge RM, Weinblatt ME, Shadick NA, Reich D. Principal
components analysis corrects for stratification in genome-wide association studies. Nat
Genet 2006;38(8):904-909.

Price AL, Tandon A, Patterson N, Barnes KC, et al. Sensitive detection of chromosomal
segments of distinct ancestry in admixed populations. PLoS Genet 2009; 5:e1000519.

Price AL, Helgasson A, Palsson S, Stefansson H, St Clair D, Andreassen OA, Reich D,
Kong A, Stefansson K. The impact of divergence time on the nature of population
structure: an example from Iceland. PloS Genet 2009;5(6):e1000505.

Price AL, Zaitlen NA, Reich D, et al. New approaches to population stratification in
genome-wide association studies. Nat Rev Genet 2010;11:459-463.

Pritchard JK, Stephens M, Donnelly P. Inference of population structure using
multilocus genotype data. Genetics 2000;155:945-959.

Pritchard JK, Donnelly P. Case-control studies of association in structured or admixed
populations. Theor Popul Biol 2001;60:227-237.

Prokopenko I, Langenberg C, Florez JC, et al. Variants in MTRNB1 influence fasting
glucose levels. Nat Genet. 2009; 41(1):77-81.

Qui XY, Jiao Z, Zhang M, Zhong LJ, Liang HQ, Ma CL, Zhang L, Zhong MK.
Association of MDR1, CYP3A4*1B, and CYP3A5*3 polymorphisms with cyclosporine
pharmakinetics in Chinese renal transplant recipients. Eur J Clin Pharmacol. 2008;
64(22)1069-1084.

Raucy JL, Mueller L, Duan K, Allen SW, Strom S, Lasker JM. Expression andinduction
of CYP2C P450 enzymes in primary cultures of human hepatocytes. J Molecular Med
2002; 302:475-482.

Reiner AP, Carlson CS, Ziv E, Iribarren C, Jaquish CE, Nickerson DA. Population
structure, admixture, and disease-related traits in African American participants in the
CARDIA study. Hum Genet 2007;121:565-575.

Ribaudo HJ, Huan L, Schwab M, et al. Effect of CYP2B6, ABCB1, and CYP3A5
polymorphisms on efavirenz pharmacokinetic and treatment response: An AIDS Clinical
Trials Group study. J Infect Dis 2010;202:717-22.

148
Rieder MJ, Reiner AP, Gage BF, Nickerson DA, Eby CS, McLeod HL, Blough DK,
Thummel KE, Veenstra DL, Rettie AE. Effect of VKORC1 haplotypes on transcriptional
regulation and warfarin dose. N Engl J Med. 2005; 352:2285–93.

Robbins GK, De Guttola V, Shafer RW, Smeaton LM, Snyder SW, Petinelli C, AIDS
Clinical Trials Group 384 Team, et al. Comparison of sequential three-drug regimens as
initial therapy for HIV-1 infection. N Engl J Med. 2003; 349:2293–303.

Rosenberg NA, Li LM, Ward R, Pritchard JK. Informativeness of genetic markers for
inference of ancestry. Am J Hum Genet 2003;73(6):1402-1422.

Rosenberg, NA. Distruct: a program for the graphical display of population structure.
Molecular Ecology Notes 2004;4:137-138.

Rotger M, Colombo S, Furrer H, Bleiber G, Buclin T, Lee BL, Keiser O, Biollaz J,
Decosterd L, Telenti A; Swiss HIV Cohort Study. Influence of CYP2B6 polymorphsim
on plasma and intracellular concentrations and toxicity of efavirenz and nevirapine in
HIV-infected patients. Pharmacogenet Genomics 2005;15:1-5.

Rotger M, Taffe P, Bleiber G, et al. Gilbert syndrome and the development of
antiretroviral therapy-associated hyperbilirubinemia. J Infect Dis 2005;192:1381-1386.

Rotger M, Tegude H, Colombo S, Cavassini M, Furrer H, Decosterd L et al. Predictive
value of known and novel alleles of CYP2B6 for efavirenz plasma concentrations in
HIV-infected individuals. Clin Pharmacol Ther. 2007;81:557-66.

Rotger M, Gsponer T, Martinez R, et al. Impact of single nucleotide polymorphisms and
of clinical risk factors on new-onset diabetes mellitus in HIV-infected individuals. Clin
Infect Dis 2010;51:1090-1098.

Ruchat SM, Elks CE, Loos RJF, et al. Association between insulin secretion, insulin
sensitivity and type 2 diabetes susceptibility variants identified in genome-wide
association studies. Acta Diabetol. 2008; epub.

Saitoh A, Singh KK, Powell CA, Fenton T, Fletcher CV, Brundge R, Starr S, Spector SA.
An MDR1-3435 variant is associated with higher plasma nelfinavir levels and more rapid
virologic response in HIV-1 infected children. AIDS 2005; 19:371-380.

Saitoh A, Sarles E, Capparelli E, Aweeka F, Kovacs A, Burchett SK, Wiznia A,
Nachman S, Fenton T and Spector SA. CYP2B6 genetic variants are associated with
nevirapine pharmacokinetics and clinical response in HIV-1 infected children. AIDS
2007;21:2191-9.

149
Saitoh A, Spector SA. Effect of host genetic variation on the pharmacokinetics and
clinical response of non-nucleoside reverse transcriptase inhibitors. Futur HIV Ther
2008;2:69-81.

Saitoh A, Capparelli E, Aweeka F, Sarles E, Singh KK, Kovacs A, et al. CYP2C19
genetic variants affect nelfinavir pharmacokinetics and virologic response in HIV-1
infected children receiving highly active antiretroviral therap. J Acquir Immune Defic
Syndr. 2009; Nov 3 [epub ahead of print].

Salanti G, Southam L, Altshuler D, et al. Underlying genetic models of inheritance in
established type 2 diabetes associations. Am J Epidemiol 2009; 170(5): 537-45.

Salari K, Choudhry S, Tang H, et al. Genetic admixture and asthma related phenotypes
in Mexican American and Puerto Rican asthmatics. Genet Epidemiol 2005;29:76-86.

Salter ML, Lau B, Go VF, Mehta SH, Kirk GD. HIV infection, immune suppression, and
uncontrolled viremia are associted with increased multimorbidity among aging injection
users. Clin Infect Dis 2011;53:1256-1264.

Sandanaraj E, Lal S, Selvarajan V, Ooi LL, Womg ZW, Wong NS, et al. PXR
pharmacogenetics: association of haplotypes with hepatic CYP3A4 and ABCB1
messenger RNA expression and doxorubicin clearance in Asian breast cancer patients.
Clin Cancer Res. 2008; 24(21):7116-7126.

Sanderson S, Emery J, Higgins J. CYP2C9 gene variants, drug dose, and bleeding risk in
warfarin-treated patients: a HuGEnet systematic review and meta-analysis. Genet Med.
2005; 7:97-104.

Saxena R , Diabetes Genetics Initiative of Broad Institute of Harvard and MIT, Lund
University, and Novartis Institutes of Biomedical Research, et al. Genome-wide
association analysis identifies loci for type 2 diabetes and triglyceride levels. Science
2007; 316:1331-1336.

Saxena R, Elbers EE, Guo Y, et al. Large-scale gene-centric meta-analysis across 39
studies identifies type 2 diabetes loci. Am J Hum Gen 2012; epub ahead of print.

Schalekamp T, Brasse BP, Roijers JF, Chahid Y, van Geest-Daalderop JH, de Vries-
Goldschmeding H, et al. VKORC1 and CYP2C9 genotypes and acenocoumarol
anticoagulation status: interaction between both genotypes affects overanticoagulation.
Clin Pharmacol Ther. 2006; 80:13–22.

Schelleman H, Chen Z, Kealey C, Whitehead AS, Christie J, Price M, et al. Warfarin
response and vitamin K epoxide reductase complex 1 in African Americans and
Caucasians. Clin Pharmacol Ther. 2007;81:742–7.

150
Schwarz UI. Clinical relevance of genetic polymorphisms n the human CYP2C9 gene.
Eur J Clin Investi 2003;33:23-30.

Schuetz JD, Beach DL, and Guzelian PS. Selective expression in cytochrome P450
CYP3A mRNAs in embryonic and adult human liver. Pharmacogenetics 1994;4:11-20.

Schumacher MC, Hasstedt SJ, Hunt SC et al. Major gene effect for insulin levels in
familial NIDDM pedigrees. Diabetes 1992; 41:416-423.

Sconce EA, Khan TI, Wynne HA, Avery P, Monkhouse Lm, King BP, et al. The impact
of CYP2C9 and VKORC1 genetic polymorphism and patient characteristics upon
warfarin dose requirements: proposal for a new dosing regimen. 2005; Blood 106: 2329-
2333.

Scordo MG, Pengo V, Spina E, Dahl ML, Gusella M, Padrini R. Influence of CP2C9 and
CYP2C19 genetic polymorphisms on warfarin maintenance does and metabolic
clearance. Clin Pharmacol Ther 2002; 72:702-710.

Scott LJ, Mohlke KL, Bonnycastle LL, Willer CJ, Li Y, et al. A genome-wide
association study of type 2 diabetes in Finns detects multiple susceptibility variants.
Science 2007; 316:1341-1345.

Shackman BR, Ribaudo HJ, Krambrink A, Hughes V, Kuritzkes DR, Gulick RM. Racial
differences in virologic failure associated with adherence and quality of life on efavirenz-
containing regimens for initial HIV therapy. J Acquir Immune Defic Syndr. 2007;
46:547-554.

Shai I, Jiang R, Manson JE et al. Ethnicity, obesity, and risk of type 2 diabetes in
women: a 20 year follow-up study. Diabetes Care 2006;29(7):1585-1590.

Shafer RW, Smeaton LM, Robbins GK, De Gruttola V, Snyder SW, D’Aquilla RT, et al.
AIDS Clinical Trials Group 384 Team. Comparison of four-drug regimens and pairs of
sequential three-drug regimenas as initial therapy for HIV-1 infection. N Engl J Med
2003;349:2304-15.

Shirai N, Furata T, Xiao F, Kajimura M, Hania H, Ohashi K et al. Comparison of
lansoprazole and famotidine for gastric acid inhibition during the daytime and night time
in different CYP2C19 genotype groups. Aliment. Pharmacol. Ther. 2002;16:837-846.

Shtir CJ, Marjoram P, Azen S, Conti DV, Le Marchand L, Haiman CA, et al. Variation
in genetic admixture and population structure among Latinos; the Los Angeles Latino eye
study (LALES). BMC Genetics 2009; 10:71-84.35.

151
Singh SS. Preclinical pharmacokinetics: an approach towards safer and efficacious
drugs. Curr Drug Metab 2006;7:165-82.

Sladek R, Rocheleau G, Rung J et al. A genome-wide association study identifies novel
risk loci for type 2 diabetes. Nature 2007; 445:881-885.

Smith MW, Patterson N, Lautenberger JA, et al. A high-density admixture map for
disease gene discovery in African Americans. Am J Hum Genet 2004;74(5):1001-1003.

Smith PF, Robbins GK, Shafer RW, Wu H, Yu S, Merigan TC, et al. Pharmacokinetics
of nelfinavir and efavirenz in antiretroviral-naıve, human immunodeficiency virus-
infected subjects when administered alone or in combination with nucleoside analog
reverse transcriptase inhibitors. Antimicro. Agents Chemother. 2005; 49: 3558-3561.

Song P, Lamba JK, Zhang L, Schuetz E, Shukla N, Meibohm B, Yates CR. G2677T and
C3435T genotype and haplotype are associated with pehatic ABCB1 (MDR1)
expression. J Clin Pharmacol. 2006; 46(3):379-373.

Sparsø T, Bonnefond A, Andersson E, et al. G-allele of intronic rs10830963 in
MTNR1B confers increased risk of impaired fasting glycemia and type 2 diabetes
through an impaired glucose-stimulated insulin release: studies involving19,605
Europeans. Diabetes. 2009; 58(6):1450-6.

Stancˇa´kova´ A, Pihlajama¨J, Kuusisto J, et al. Single Nucleotide Polymorphism
rs7754840 of CDKAL1 is associated with impaired insulin secretion in nondiabetic
offspring of type 2 diabetic subjects and in a large sample of men with normal glucose
tolerance. J Clin Endocrinol Metab 2008; 93(5):1924-1930.

Starr SE, Fletcher CV, Spector SA, Young FH, Fenton T, et al. Combination therapy
with efavirenz, nelfiniavir, and nucleoside reverse-transcriptase inhibitors in children
infected with human immunodeficiency virus type 1. Pediatric AIDS Clinical Trials
Group 382. N Engl J Med 1999; 341(25):1925-6.

Sterne JA, May M, Costagliola D, de Wolfe F, Phillips AN, Harris R, Funk MJ, Geskus
RB et al. Timing and initiation of antiretroviral therapy in AIA-free HIV-1-infected
patients: a collaborative analysis of 18 HIV cohort studies. Lancet 2009; 373:1352-63.

Steward DJ, Haining RL, Henne KR, Davis G, Rushmore TH, Trager WF, Rettie AE.
Genetic association between sensitivity to warfarin and expression of CYP2C9*3.
Pharmacogenetics 1997; 7:361-367.

Storch CH, Theile D, Lindenmaier H, Haefeli WE, Weiss J. Comparison of the
inhibitory activity of anti-HIV drugs on P-glycoprotein. Biochem Pharmacol
2007;73:1573-81.

152
Strassburg CP, Lankisch TO, Manns MP, Ehmer U. Family 1 uridine-5’-diphosphate
glucuronosyltransferases (UGT1A): from Gilbert’s syndrome to genetic organization and
variability. Arch Toxicol 2008;82:415-433.

Sugimoto M, Furuta T, Shirai N, Ikuma M, Hishida A, Ishizaki T. Initial 48-hour
inhibition by intravenous infusion of omepravole, famotidine, or both in relation to
cytochrome P450 2C19 genotype status. Clin Pharmacol Ther. 2006; 80(5):539-548.

Sustiva (efavirenz) [package insert]. Wilmongton, DE: DuPont Pharmaceuticals Co.;
2003.

Tabara Y, Osawa H, Kawamoto R, et al. Replication study of candidate genes associated
with type 2 diabetes based on genome-wide screening. Diabetes 2009; 58:493-498.

Takahashi H, Wilkinson GR, Nutescu EA, Morita T, Ritchie MD, Scordo MG, et al.
Different contributions of polymorphisms in VKORC1 and CYP2C9 to intra- and inter-
popumation difference in maintenace dose of warfarin in Japanese, Caucasians, and
African-Americans. Pharmacogenetic Genomics 2006; 16:101-110.

Tateishi T, Watanabe M, Moriya H, Yamaguchi S, Sato T, Kobayashi S. No ethnic
difference between Caucasian and Japanese hepatic samples in the expression frequency
of CYP3A5 and CYP3A7 proteins. Biochem Pharmacol 1999; 57:935-9.

Tedaldi EM, Absalon J, Thomas AJ, Shlay JC, van den Berg-Wolf M. Ethnicity, race
and gender. Differences in serious adverse events among participants of an antiretroviral
initiation trial CPCRA058. J Acquir Immune Defic Syndr. 2008;47:441-448.

Telenti A, McLaren P. Genomic approaches to the study of HIV-1 acquisition. J Infect
Dis. 2010;202 Suppl 3:S382-386.

Thiebaut F, Tsuruo T, Hamada H, Gottesman MM, Pastan I, Willingham MC. Cellular
localization of the multidrug-resistance gene product P-glycoprotein in normal human
tissues. Proc Natl Acad Sci 1987; 84:7735-7738.

Thomas DC, Witte JS. Point: population stratification: a problem for case-control studies
of candidate-gene associations? Cancer Epidemiol Biomarkers Prev 2002;11(6):505-512.

Tien PC, Schneider MF, Cole SR, et al. Antiretroviral therapy exposure and incidence of
diabetes mellitus in the Women’s Interagency HIV Study. AIDS 2007;21:1739-1745.

Tishkoff SA, Reed FA, Friedlander FR, Ehret C, Ranciaro A, Froment A et al. The
genetic structure and history of Africans and African Americans. Science
2009;324:1035-1044.

153
Tozzi V. Pharmacogenetics of antiretrovirals. Antiviral Res 2010;85(1):190-200.

Triant VA, Lee H, Hadigan C, Grinspoon SK. Increased acute myocardial infarction
rates and cardiovascular risk factors among patients with human immunodeficiency virus
disease. J Clin Endocrinol Metab 2007;92:2506-2512.

Unoki H, Takahashi A, Kawaguchi T, et al. SNPs in KCNQ1 are associated with
susceptibility to type 2 diabetes in East Asian and European populations. Nat Genet.
2008; 40(9):1098-102.

Van Hoek M, Dehgan A, Witteman JC, et al. Predicting type 2 diabetes based on
polymorphisms from genome wide association studies: a population-based study.
Diabetes 2008; 57:3122–3128.

Van Leth F, Phanuphak P, Ruxrungtham K, et al. for the 2NN Study team. Comparison
of the first-line antiretroviral therapy with regimens including nevirapine, efavirenz, or
both drugs, plus stavudine and lamivudine: a randomized open-label trial, the 2NN Study.
Lancet 2004;363:1253-1263.

Vidal F, Gutierrez F, Gutierrez M, Olana M, Sanchez V, Mateo G, et al.
Pharmacogenetics of adverse effects due to antiretroviral drugs. AIDS Rev 2010 12:15-
30.

Viracept (nelfinavir) [package insert]. La Jolla, CA: Agouron Pharmaceuticals, Inc.; Sept.
2004.

Voight BF, Pritchard JK. Confounding from cryptic relatedness in case-control
associations. PloS Genet 2005;1(3):e32.

Von Moitke LL, Greenblatt DJ, Granda BW, Giancarlo GM, Duan SX, Daily JP,
Harmatz JS, Shader RI. Inhibition of human cytochrome P450 isoforms by
nonnucleoside reverse transcriptase inhibitors. J Clin Oharmacol 2001;41(1):85-91.

Wag HB, Tompkins LM. CYP2B6: New insights into a historically overlooked
cytochrome P450 isozyme. Curr Drug Metab 2008; 9:598-610.

Wang J, Sonnenborg A, Rane A et al. Identification of a novel specific CYP2B6 allele in
Africans causing impaired metabolism of the HIV drug efavirenz. Pharmacogenet
Genomics 2006;16:191-8.

Wang S, Lewis CM, Jakobsson M, et al. Genetic Variation and Population Structure in
Native Americans. PloS Genet 2007;3(11):e185.

154
Ward BA, Gorski JC, Jones DR, Hall SD, Flockhart DA, Desta Z. The cytochrome p450
2B6 is the main catalyst of evafirenz primary and secondary metabolism: implication for
HIV/AIDS therapy and utility of efavirenz as a sybstrate marker of CYP2B6 catalytic
activity. J Pharmacol Exp Ther 2003; 306:287-300.

Waters KM, Stram DO, Hassanein MT, et al. Consistent association of type 2 diabetes
risk varaints found in Europeans in diverse racial and ethnic groups. PLoS Genet
2010;6(8):e1001078.

Waxman DJ and O’Connor DJ. Growth hormone regulation of sex-dependent liver gene
expression. Mol Endocrinol. 2006; 20(11):2613-2629.

Weinshilbourn R. Inheritance and Drug Response. N Eng Med J 2003; 348: 529-537.

Weiss J, Weiss N, Ketabi-Kiyanvash N, Storch CH, Haefeli WE. Comparison of the
induction of P-glycoprotein activity by nucleotide, nucleoside, and non-nucleoside
reverse transcriptase inhibitors. Eur J Pharmacol 2007; 21:411-8.

The Wellcome Trust Case Control Consortium: genome-wide association study of 14,000
cases of seven common diseases and 3,000 shared controls. Nature 2007; 447:661-678.

Williams JA, Ring BA, Cantrell VE, Jones DR, Eckstein J, Ruterbories K, Hamman MA,
Hall SD, Wrighton SA. Comparative metabolic capabilities of CYP3A4, CYP3A5,
CYP3A7. Drug Metabolism and Disposition 2002;30:883-891.

Wolbers M, Battegay M, Hirschel B, Furrer H, Cavassini M, Hasse B, et al. Swiss HIV
Cohort Study. Antivir Ther 2007;12(6):889-897.

Wolbold R, Klein K, Burk O, Nussler AK, Neuhas P, Eichlbaum M, Schwab M, Zanger
UM. Sex is a major determinant of CYP3A4 expression in the human liver. Hepatology.
2003; 38(4): 978-988.

Wu H, Kurtizkes DR, McClernon DR, et al. Characterization of viral dynamics in HIV-1
infected patients treated with combination antiretroviral therapy:relationships to host
factors, cellular restoration, and virologic endpoints. J Infect Dis. 1999; 179:799-807.

Wu Y, Li H, Loos RJF, et al. Common variants in CDKAL1, CDKN2A/B, IGF2BP2,
SLC30A8, and HHEX/IDE genes are associated with type 2 diabetes and impaired
fasting glucose in a Chinese Han population. Diabetes 2008; 57:2834-2842.

Wyen C, Hendra H, Vogel M, et al. Impact of CYP2B6 983T>C polymorphisms on non-
nucleoside reverse transcriptase inhibitor plasma concentrations in HIV-infected patients.
J Antimicrob Chemother 2008;61:914-8.

155
Xie W, Evans RM. Orphan nuclear receptors: the exotics of xenobiotics. J Biol Chem
2001; 276:37739-37742.

Yasuda K, Miyake K, Horikawa Y, et al. Variants in KCNQ1 are associated with
susceptibility to type 2 diabetes mellitus. Nat Genet 2008; 40:1092–1097.

Yuan HY, Chen JJ, Lee MT, Wung JC, Chen YF, Charng MJ, et al. A novel functional
VKORC1 promoter polymorphism is associated with inter-individual and inter-ethnic
differences in warfarin sensitivity. Hum Mol Genet. 2005; 14:1745–1751.

Zanger UM, Turpeinen M, Klein K, Schwab M. Functional pharmacogenetics/genomics
of human cyctochromes P450 involved in drug biotransformation. Anal Bioanal Chem.
2008; 392:1093-1108.

Zeggini E, Weedon MN, Lindgren CM, Frayling TM, et al. Wellcome Trust Case Control
Consortium (WTCCC). Replication of genome-wide association signals in UK samples
reveals risk loci for type 2 diabetes. Science 2007; 316:1336-1341.

Zeggini E, Scott LJ, Saxena R and Voight BF for the Diabetes Genetics Replication and
Meta-Analysis (DIAGRAM) Consortium. Meta-analysis of genome-wide association
data and large-scale replication identifies additional susceptibility loci for type 2 diabetes.
Nature Genetics 2008; 40:638-645.

156
Appendix : Pharmacogenetic Associations in Antiretroviral Therapy

INTRODUCTION

Morbidity and mortality has declined markedly with the advent and widespread
use of highly active antiretroviral therapy (HAART) for individuals with HIV-1 infection
(Palella 1998, Gange 2002, Palella 2006). However, clinical response to HAART varies
in populations; variation in genes that affect the absorption, distribution, metabolism, and
excretion (ADME) of drugs is thought to account for a proportion of the variability in
HAART response (Owen 2008, Tozzi 2010).
Several pharmacogenetic studies, i.e., studies that examine the effect of genetic
variation on drug therapy response outcomes, have been conducted to evaluate the
association between genetic variants and antiretroviral drug response. Drug responses
evaluated as outcomes include pharmacokinetic properties, hypersensitivity reaction
syndrome, neurotoxicity, hepatotoxicity, hyperbilirubinemia, and virologic and immune
responses (Owen 2008, Tozzi 2010). The genes which have received the most attention
in pharmacogenetic association studies in HIV disease include those in the cytochrome
p450 family (CYP), ABCB1, and UGT1A1 genes. The protein products of these genes
are involved in the metabolism and transport of two antiretroviral drug classes, non-
nucleoside reverse transcriptase inhibitors (NNRTIs) and protease inhibitors (PIs) (Weiss
2007, Storch 2007, Von Moitke 2001).

157
Numerous studies have evaluated CYP genetic determinants of antiretroviral drug
pharmacokinetics; whether these polymorphisms translate to drug efficacy is
controversial (Rotger 2005, Haas 2005, Fellay 2005, Anderson 2006, Motsinger 2006,
Rotger 2007, Saitoh 2007, Haas 2009, Saitoh 2009, Carr 2010). There has also been
wide interest in the potential genetic influence of ABCB1 on HAART response owing to
the crucial role of P-glycoprotein (encoded by ABCB1) in the distribution and excretion
of all protease inhibitors (Storch 2007, Fellay 2002, Brumme 2003, Haas 2003, Saitoh
2005). However, modification of response to PI-containing regimens in HIV-infected
patients who carry variant alleles of two different ABCB1 polymorphisms have not been
consistently observed (Fellay 2002, Brumme 2003, Haas 2003, Saitoh 2005). Loss of
function alleles in the UGT1A gene are associated with hyperbilirubinemia (Strassberg
2008); carriers of the variant with the PI atazanavir develop more severe
hyperbilirubinemia, as atazanavir, acts as an inhibitor to UGT1A (Rotger 2005, Lankisch
2006). Despite the potential for personalized HAART regimens based on genetic
predictors of response, pharmacogenetic studies of HAART have not provided conclusive
evidence of associations.
The Women’s Interagency HIV Study (WIHS) is a prospective multi-site
observational study of multi-ethnic women infected with HIV who are treated with
various HAART regimens as part of their routine care. Over half of the cohort is self-
identified as African-American, nearly 25% identify as Hispanic, and nearly 15% are
non-Hispanic White (Barkan 1998, Bacon 2005). WIHS participants initiating a HAART
regimen were genotyped for 68 polymorphisms suggested to affect protein function in

158
CYP1A2, CYP2B6, CYP2C9, CYP2C19, CYP2D6, CYP2E1, CYP3A4, CYP3A5,
ABCB1, ADH1, and UGT1A1 genes. We also explored genetic associations with
clinical treatment response in the NR1I2 gene (37 polymorphisms), which is a common
nuclear transcription receptor for enzymes that metabolize both NNRTIs and PIs
(Faucette 2006). We report our results herein.

METHODS

WIHS Study Design. The WIHS is a prospective study of HIV-infected women
and a comparison group of HIV-uninfected women (Barkan 1998, Bacon 2005).
Participants were recruited in two phases from six sites across the U.S.
(Bronx/Manhattan, New York; Brooklyn, New York; Washington, D.C.; Los Angeles,
CA; San Francisco/Bay Area, CA; and Chicago, IL). The original recruitment phase was
conducted in 1994-1995, in which 2,054 HIV seropositive and 569 seronegative women
were enrolled. The second phase of recruitment (2001-2002) resulted in the addition of
737 HIV seropositive and 406 seronegative women to the cohort. The characteristics of
WIHS HIV-infected women at entry to the cohort are comparable to reported
characteristics of HIV cases in the U.S. A more detailed description of the WIHS cohort
is published elsewhere (Barkan 1998, Bacon 2005).
Participants are seen for in-person visits every six months during which time
trained medical interviewers administer an extensive questionnaire, a clinical exam is
performed, and biological samples are collected. Laboratory tests are conducted on the

159
women’s blood samples to measure measure HIV-1 viral load (copies/ml) and white
blood cell counts, including CD4+ cells. At each biannual visit, detailed information on
participants’ current treatment regimen and the antiretroviral medications taken in the six
months prior to the current follow-up visit is obtained. The women answer yes or no to a
series of questions regarding use of specific antiretroviral drugs and are given pictures of
the medication bottles to prompt their memory. Comprehensive information on
medication use for other conditions is also obtained and recorded at each follow-up visit.
Commencing at visit 9 (1998), data were collected on self-reported adherence to
antiretroviral medication (100%, 95-99%, 75-95%, <75% over the past six months) and
whether any side effects were experienced.
Inclusion Criteria for Pharmacogenetic Study. HIV-infected participants who
initiated HAART during follow-up and had measures of HIV-1 RNA viral load at the
visit immediately prior to, the visit of, and the visit immediately after reporting an initial
HAART regimen were eligible for the this pharmacogenetic study. HAART was defined
according to the DHHS/Kaiser Panel 2008 guidelines as the reported use of “three or
more antiretroviral medications, one of which has to be a protease inhibitor (PI), a non-
nucleoside reverse transcriptase inhibitor (NNRTI), one of the nucleoside reverse
transcriptase inhibitors (NRTIs) abacavir or tenofovir, an integrase inhibitor (e.g.
raltegravir), or an entry inhibitor (e.g. Maraviroc or enfuvitide)”. Participants who
received an antiretroviral drug prior to initiating HAART were included in the study only
if two new antiretroviral drugs were implemented as part of a ≥ 3 drug HAART regimen.

160
Participants who did not consent to genetic analyses or did not remain on HAART
the visit after the initial HAART visit were excluded from the study.
Selection of genetic variants. Single nucleotide polymorphisms (SNP)
implicated in drug absorption, distribution, metabolism and excretion (ADME) pathways
were selected based on putative function, pharmacogenomic association studies, minor
allele frequency (MAF), and assay performance on Illumina’s GoldenGate
®
genotyping
platform. The UCSC Genome Browser was referenced to compile a list of putative
functional SNPs (located in exons, 3’ UTR, 5’ UTR and splice sites), across cytochrome
p450, ABCB1, ADH, NR1I2, and UGT1A genes. Putative functional SNPs that have
been investigated for associations with antiretroviral response or antiretroviral
pharmacokinetic properties were queried on www.hiv-pharmacogenomics.org and added
to the bin. NCBI’s dbSNP was used to acquire minor allele frequencies (MAF) for
African, European, Han Chinese, and Japanese populations. The list was filtered for
SNPs with a MAF ≥ 0.10 in any one population and SNPs that scored favorably for
successful assay design with a high probability for successful genotyping on the Illumina
platform . SNPs that passed all criteria were cross-checked with each other for linkage
disequilibrium (LD) r
2
≥ 0.80 to confirm that each locus was independent at this level.
Genetic variants with MAFs ≥ 0.10 within the coding region of NR1I2 were
tagged using a pairwise linkage disequilibrium approach at r
2
≥ 0.80 in European,
Yoruban, and combined East Asian (Chinese Han and Japanese) populations from
HapMap 2 (28). Putative functional SNPs identified in the original list (above) were
force included into the tag SNP selection software, Snagger (Edlund 2008), and a

161
minimum physical distance of 60 base pairs was enforced. The minimal set of tagSNPs
that captured the underlying genetic structure of NR1I2 in all three populations was
selected.
Outcome. HIV viral load levels were quantified with an assay that has a lowest
detectable limit of 80 copies/ml. A positive HAART response was defined as
achievement of an undetectable viral load at the visit they first reported HAART or at the
visit subsequent to HAART initiation (“responders”), which corresponds to a maximum
of 54 weeks. Participants not achieving an undetectable viral load at the visit first
reporting HAART initiation or at the post-HAART initiation visit were classified as
“non-responders”.
Covariates. In a separate WIHS study, genetic ancestry was quantified using a
panel of ancestry informative markers. Principal components analysis was performed to
generate genetic ancestry covariates to control for population stratification in WIHS
genetic association studies. HAART adherence data was available for participants who
initiated HAART at visit 9 (1998) or later (data collected at visit 9 and thereafter). The
adherence data for this analysis was taken at the visit the participant achieved an
undetectable viral load (visit of HAART initiation or post-HAART initiation) since the
adherence variable at this visit reflects HAART adherence in the six months leading up to
the visit in which the outcome was achieved. Adherence data for non-responders was
taken at the study endpoint, which is defined as the first visit post-HAART initiation. For
modeling purposes, adherence was dichotomized as ≥ 95% adherent and < 95% adherent.

162
Type of HAART regimen was categorized as NRTI/PI, NRTI/NNRTI, NRTI/PI/NNRTI,
and 3 or more NRTIs.
Genotyping and Quality Control. From a total of 582 eligible HAART
initiators, 524 (90%) women who initiated at least two new drugs and for which DNA
samples were available were genotyped. The 58 remaining eligible participants did not
have an adequate amount of DNA available for genotyping.
Data for the 102 ADME SNPs in the study were generated as a subset of a larger
panel (n= 384 SNPs) and eligible participants were genotyped as a subset of a larger
study population (N=1,252). The raw data were edited using Illumina BeadStudio


software to cluster the three genotypes (AA, AB, BB) at each locus. Quality control was
a multi-step process, in which the first round eliminated 11 of 384 SNPs (3%) for which
three distinct genotypes were not observed. Next, 51 samples (4%) were eliminated that
had missing data for more than 20% of the remaining 373 SNPs. Assay performance for
the remaining 373 SNPs was then assessed and six additional SNPs (2%) were eliminated
because they had a call rate < 90%. Lastly, 13 samples (5%) were excluded because they
had more than 10% missing data after the elimination of the 17 SNPs. Hardy-Weinberg
equilibrium was tested within each self-identified ethnic group; SNPs with p-values <
0.05 in any one population were visually reviewed for presence of three distinct clusters
for accurate genotype determination. Duplicates for 78 participants were included and
concordance was 99.9%.
In summary, 10 out of 102 ADME SNPs (10%) and 48 (9%) out of the 524
participants who were genotyped did not meet quality control (QC) criteria and were

163
excluded from the analysis. Six out of 92 ADME SNPs that passed QC were
monomorphic in this study population and thus excluded from the analysis.
Statistical Analysis. Analysis was restricted to women with ancestry informative
marker data (N=460/476 with successful genotyping) and those who reported African-
American, Hispanic, or Non-Hispanic White ethnicity (n=445). Additionally, participants
who initiated a HAART regimen that contained both a NNRTI and a PI were excluded
from the analysis, due to small numbers (N=21) and the potential for substrate-competing
effects for enzymatic activity. The final analytic dataset included 424 women who
initiated HAART during the period 1996-2008.
Logistic regression was used to test the associations between genotypes and
response to HAART; the heterozygotes and rare homozygotes were categorized
separately and relative risks were estimated using the major allele homozygotes as the
referent group. Analyses were performed with adjustment for population stratification
incorporating the top 5 principal components (PCs) that accounted for 44% of the
population genetic variance. Antiretroviral drug classes vary in their affinity for
particular enzymes, and as such the analyses were stratified based on type of HAART
regimen to evaluate heterogeneity between subsets of initiators by antiretroviral drug
class. Separate models were fitted among self-reported race/ethnicity groups (African-
American, Hispanic, and Non-Hispanic White) to evaluate whether there was
heterogeneity between ethnicities after adjustment for PCs. All analyses were conducted
in the subset of HAART initiators who had never been treated with antiretroviral drugs
(N= 204/424) as a sensitivity analysis.

164
CYP2B6 metabolizer phenotypes were constructed using two polymorphisms,
rs3745274 and rs28399499, to test the association of metabolizer phenotype and virologic
response (16). Women who were common homozygotes at both polymorphisms were
coded as 0 ‘extensive metabolizers’. Women with one heterozygote genotype and one
common homozygote genotype at either polymorphism were coded as 1 ‘intermediate
metabolizers’. Women with a total of two variants (one variant homozygote genotype, or
two heterozygote genotypes) across both SNPs were coded as 2 ‘slow metabolizers’.
Adherence, pre-HAART viral load and nadir CD4+ count prior to initiation were
evaluated as confounders by testing for an association between adherence and genotypes
in women who were naïve to antiretroviral treatment and reponded to their HAART
regimen.

165
RESULTS

Table A1. Nominally significant associations between genotypes in NR1I2 and
response to HAART* in all HAART initiators (N=424) and in the subset of
antiretroviral naïve HAART initiators (N=204).

ALL NAIVE
SNP & MAF Genotype N
R
†
N
NR
‡
RR
§
95% CI p
trend
RR
§
95% CI p
trend

rs2472661 GG 229 (127) 149 (52) 1.00 ref
1.00 ref
MAF 0.06 GA 33 (19) 12 (6) 2.16 1.05-4.43 1.77 0.63-4.94
AA 1 (0) 0 (0) -- -- -- -- -- --
rs4687883 GG 208 (116) 137 (47) 1.00 ref
1.00 ref
MAF 0.10 GA 54 (30) 21 (11) 2.04 1.14-3.65 1.41 0.62-3.21
AA 1 (1) 2 (0) 0.35 0.03-3.99 -- -- -- --
rs2461817 AA 75 (39) 60 (19) 1.00 ref
1.00 ref
MAF 0.45 AC 129 (77) 63 (25) 1.61 1.02-2.55 1.43 0.69-2.96
CC 59 (30) 37 (13) 1.23 0.71-2.11 0.33 1.03 0.43-2.47 0.84
rs16830505 AA 163 (95) 87 (26) 1.00 ref
1.00 ref
MAF 0.24 AG 84 (43) 61 (26) 0.79 0.50-1.23 0.43 0.21-0.89
GG 16 (8) 12 (5) 0.77 0.34-1.73 0.30 0.43 0.12-1.55 0.04
rs4687887 AA 179 (103) 100 (32) 1.00 ref
1.00 ref
MAF 0.19 AG 75 (37) 53 (22) 0.87 0.54-1.40 0.52 0.24-1.13
GG 7 (4) 8 (4) 0.54 0.18-1.58 0.29 0.31 0.07-1.40 0.05

* HAART responders defined as achievement of undetectable viral load up to 54 weeks after
self-reported initiation of HAART
† N
R
: N for all HAART responders, N for antiretroviral naïve HAART responders in
parentheses
‡ N
NR
: N for all HAART non-responders, N for antiretroviral naïve HAART non-responders in
parentheses
§ Adjusted for genetic ancestry principal components

166
Table A2. Association between genetic variants (log-additive) and response to HAART
in all HAART initiators (N=424) and in the subset of antiretroviral naïve HAART
initiators (N=204) by type of HAART regimen.

Gene
(ARV substrate
†
)
SNP
(Risk Allele)
Regimen
†

ALL
NAIVE
RR
‡
95% CI p
trend
RR
‡
95% CI p
trend

ABCB1 rs1128503 (A) NRTI alone 0.84 0.20-3.53 0.81 0.01 0-1.50 0.07
(PI&NNRTI) NRTI/NNRTI 2.68 1.35-5.33 0.005 3.36 1.11-10.19 0.03
NRTI/PI 0.91 0.61-1.38 0.66 1.64 0.77-3.47 0.20
ABCB1 rs2235046 (T) NRTI alone 0.80 0.24-2.67 0.71 0.02 0.0-5.35 0.17
(PI&NNRTI) NRTI/NNRTI 2.37 1.24-4.55 0.01 2.09 0.77-5.69 0.15
NRTI/PI 0.88 0.59-1.32 0.55 1.74 0.85-3.55 0.13
ABCB1
(PI&NNRTI)
rs34800935 (C) NRTI alone 0.84 0.14-5.00 0.85 --
NRTI/NNRTI 3.96 1.18-13.27 0.03 7.17 0.79-64.7 0.08
NRTI/PI 0.96 0.57-1.63 0.89 0.97 0.43-2.16 0.93
NR1I2 rs2472661 (A) NRTI alone 0.87 0.06-13.35 0.92 --
(PI&NNRTI) NRTI/NNRTI 1.70 0.57-5.04 0.34 1.78 0.39-8.19 0.46
NRTI/PI 3.25 1.16-9.16 0.03 3.08 0.64-14.86 0.16
NR1I2 rs1403526 (G) NRTI alone 0.78 0.15-3.98 0.76 1.01 0.05-19.28 0.99
(PI&NNRTI) NRTI/NNRTI 0.57 0.34-0.95 0.03 0.53 0.22-1.25 0.14
NRTI/PI 0.89 0.62-1.26 0.50 1.06 0.58-1.93 0.86
NR1I2 rs16830505 (G) NRTI alone 0.33 0.06-1.74 0.19 0.87 0.03-25.19 0.94
(PI&NNRTI) NRTI/NNRTI 0.44 0.22-0.87 0.02 0.25 0.08-0.79 0.02
NRTI/PI 1.12 0.73-1.71 0.60 0.73 0.35-1.51 0.40
NR1I2 rs4687887 (G) NRTI alone 0.27 0.03-2.15 0.22 0.36 0-45.11 0.68
(PI&NNRTI) NRTI/NNRTI 0.42 0.20-0.85 0.02 0.37 0.12-1.10 0.07
NRTI/PI 1.09 0.65-1.82 0.75 0.59 0.25-1.40 0.23
NR1I2 rs2472681 (T) NRTI alone 10.36 1.26-85.54 0.03 --
(PI&NNRTI) NRTI/NNRTI 0.74 0.40-1.37 0.34 0.87 0.35-2.19 0.77
NRTI/PI 0.80 0.53-1.19 0.26 0.84 0.42-1.67 0.61
ADH rs1693482 (T) NRTI alone 0.10 0.01-0.80 0.03 --
(NRTI) NRTI/NNRTI 1.11 0.54-2.27 0.78 0.94 0.34-2.61 0.91
NRTI/PI 1.10 0.69-1.74 0.69 0.98 0.47-2.05 0.97
ADH rs698 (C) NRTI alone 0.03 0-0.78 0.03 --
(NRTI) NRTI/NNRTI 1.26 0.60-2.65 0.54 1.12 0.36-3.52 0.84
NRTI/PI 1.17 0.75-1.83 0.50 1.12 0.55-2.28 0.75

167
Table A2, continued

† ARV= antiretroviral drug, NRTI= nucleoside reverse transcriptase inhibitor (n=39, NRTI alone
regimen), NNRTI= non-nucleoside reverse transcriptase inhibitor containing regimen (n=138),
PI=protease inhibitor containing regimen (n=247), NFV=nelfinavir (n=75)
‡ Adjusted for genetic ancestry principal components

Abstract (if available)

Abstract Population stratification can result in spurious associations in genetic studies when the outcome and genotype are associated with genetic ancestry. The objective of this dissertation was to characterize population substructure in the Women’s Interagency HIV Study (WIHS), a cohort of HIV-infected and uninfected participants in the United States (U.S), and assess the impact of population substructure on genetic association studies. First, population substructure was characterized using genotype data on 168 ancestry informative markers (AIMs) by performing a Bayesian clustering algorithm to infer genetic membership in three and four assumed source populations. Principal components (PC) analysis was conducted to generate continuous genetic ancestry covariates to use as covariates in genetic association statistical models. Population substructure was identified within self-identified ethnic groups and across geographical regions in the U.S., exemplifying the importance of estimating individual genetic ancestry to control for residual ethnic confounding in genetic association studies. Cautious interpretation of ancestry admixture estimates should be exercised, as the labeling of source populations is sensitive to the panel of markers and referent populations used in analyses. ❧ The effect of variation in CYP2B6 on virologic response to its substrates, the nonnucleoside reverse transcriptase inhibitors (NNRTIs), was explored with consideration for self-reported race/ethnicity and underlying genetic structure. Logistic regression was used to test the joint effect of two single nucleotide polymorphisms (SNP) rs3745274 and rs28399499, which comprise the CYP2B6 metabolizer phenotype indicating slow, intermediate and extensive metabolizers. Substantial evidence of confounding was present with the metabolizer phenotype when comparing the crude, self-reported ethnicity-adjusted and genetic ancestry PCs-adjusted estimates. Women classified as intermediate and slow metabolizers were 2.90 (95% CI 0.79-12.28) and 13.44 (95% CI 1.66-infinity) times as likely to achieve virologic suppression compared to extensive metabolizers after adjustment for PCs (p trend =0.005). The CYP2B6 metabolizer phenotype was significantly associated with response to NNRTIs, a relation that would have been masked by simply adjusting for self-reported race/ethnicity (p for trend=0.04 after adjustment for self-reported race/ethnicity). ❧ Type 2 diabetes (T2D) in the setting of HIV infection is a concern given the prevalence of HIV-related conditions that contribute to the etiology T2D. The incidence of T2D varies substantially between racial/ethnic groups and thus control for genetic substructure is imperative in genetic association studies of T2D risk. Eighteen previously confirmed T2D-associated SNPs were tested using Cox proportional hazard models with adjustment for genetic ancestry principal components using age for the time-scale. Exposure to nucleoside reverse transcriptase inhibitors (NRTI), a T2D risk factor, was explored as an effect modifier. Overall, the T2D risk conferred by these SNPs was similar in White/Hispanic HIV-infected women compared to HIV-uninfected European individuals, as evaluated by the effect measures and p values for heterogeneity. The magnitude of each SNP effect was smaller in African-American women (HRs~1.1). Significant interactions were revealed between these T2D-associated SNPs and NRTI exposure (p<0.03) in White/Hispanic women. ❧ This dissertation confirmed prior report of the role of CYP2B6 SNPs in NNRTI metabolism and efficacy. Additionally, the role of T2D-associated SNPs in the context of HIV was validated and the synergistic effect of T2D-associated SNPs and NRTIs on the risk of T2D was suggested. Substantial evidence of residual confounding was revealed in both studies, which highlights the importance of controlling for population substructure using genetic ancestry estimates in genetic association studies.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Population substructure and its impact on genome-wide association studies with admixed populations

PDF

Genetic studies of cancer in populations of African ancestry and Latinos

PDF

Identifying genetic, environmental, and lifestyle determinants of ethnic variation in risk of pancreatic cancer

PDF

Using genetic ancestry to improve between-population transferability of a prostate cancer polygenic risk score

PDF

Genomic risk factors associated with Ewing Sarcoma susceptibility

PDF

Associations between physical activities with bone mineral density in postmenopausal women

PDF

Cross-sectional association of blood pressure, antihypertensive medications, MRI volumetric measures and cognitive function scores in an aging population

PDF

Observed and underlying associations in nicotine dependence

PDF

Shortcomings of the genetic risk score in the analysis of disease-related quantitative traits

PDF

HIF-1α gene polymorphisms and risk of severe-spectrum hypertensive disorders of pregnancy: a pilot triad-based case-control study

PDF

Genetic epidemiological approaches in the study of risk factors for hematologic malignancies

PDF

Genetic association studies of age-related macular degeneration from candidate gene to whole genome

PDF

The role of genetic ancestry in estimation of the risk of age-related degeneration (AMD) in the Los Angeles Latino population

PDF

Associations between isoflavone soy protein (ISP) supplementation and breast cancer in postmenopausal women in the Women’s Isoflavone Soy Health (WISH) clinical trial

PDF

Utility of polygenic risk score with biomarkers and lifestyle factors in the multiethnic cohort study

PDF

Native American ancestry among Hispanic Whites is associated with higher risk of childhood obesity: a longitudinal analysis of Children’s Health Study data

PDF

Insulin’s effect on lactate levels in extremely low birth weight neonates. a multi-center, observational study

PDF

The associations between ultra-processed food consumption and type 2 diabetes and obesity among young adults

PDF

Diet quality and pancreatic cancer incidence in the multiethnic cohort

PDF

Adaptive set-based tests for pathway analysis

Asset Metadata

Creator Frasco, Melissa Ann (author)

Core Title Pharmacogenetic association studies and the impact of population substructure in the women's interagency HIV study

School Keck School of Medicine

Degree Doctor of Philosophy

Degree Program Epidemiology

Publication Date 07/25/2012

Defense Date 03/16/2012

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag confounding,HIV,OAI-PMH Harvest,pharmacogenetic association studies,population substructure,Women

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Mack, Wendy Jean (committee chair), Pearce, Celeste Leigh (committee chair), Conti, David V. (committee member), Shibata, Darryl K. (committee member), Van Den Berg, David (committee member)

Creator Email frasco.melissa@gmail.com,mfrasco@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c3-66253

Unique identifier UC11290008

Identifier usctheses-c3-66253 (legacy record id)

Legacy Identifier etd-FrascoMeli-1003.pdf

Dmrecord 66253

Document Type Dissertation

Rights Frasco, Melissa Ann

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA