University of Southern California Dissertations and Theses

Novel multi-site brain imaging approaches to map HIV-related neuropathology

(USC Thesis Other)

Novel multi-site brain imaging approaches to map HIV-related neuropathology

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

Novel Multi-Site Brain Imaging Approaches
to Map HIV-related Neuropathology

By Talia Miriam Nir

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

May 2019
i

ACKNOWLEDGEMENTS

To all of my family, friends, colleagues, mentors and professors, thank you for your patience, guidance,
and support throughout this process; I would not be where I am without you.

I would like to especially thank my graduate mentor, Paul M. Thompson, for the endless scientific
opportunities, intellectual support, and enthusiasm. I am immensely grateful for my experience working
at the Imaging Genetics Center, where the scope of expertise and available resources are truly
unparalleled. My committee chair, Neda Jahanshad, was also an incredible source of knowledge and
advice, not only academically, but personally. She worked with me through more than one meltdown. She
patiently offered both thoughtful critiques and friendly encouragement as needed (boy did I need it). Her
expertise and mentorship were essential in moving each and every project forward. I would also like to
thank my committee member Greg Ver Steeg for his guidance and encouragement.

Lastly, I would like to thank my husband, David Mannheim, Bubbies, and so many labmates, classmates,
and friends who were there for me when I needed it most (and deserved it least); Artemis Zavaliangos-
Petropulu, Chris Ching, Julio Villalon, Priya Bhatt, Sophia Thomopoulos, Dan Rinker, Melanie Sweeney,
Louise Menendez, Brenton Keller, Rorry Brenner, Katie Zyuzin, and Kirsten Lynch, you kept me
sane…ish.

Funding Agency and Co-author Acknowledgements

Funding for the ENIGMA Center for Worldwide Medicine Imaging and Genomics was provided by the
NIH Big Data to Knowledge Program (BD2K; U54 EB020403). The HIV Neuroimaging Consortium
(HIVNC) was funded by the National Institute of Neurological Disorders and Stroke (NINDS R01
NS080655). The Alzheimer's Disease Neuroimaging Initiative (ADNI) was funded by the National
Institute on Aging (NIA U01 AG024904) and the Department of Defense (DOD W81XWH-12-2-0012).

ii

Chapters 2-5 are based on the following articles, and I would like to thank all of the co-authors and
collaborators for their contributions. Additional information about funding and support for each article is
provided in the respective acknoweldgments section of each chapter.

Nir TM, Jahanshad N, Ching CRK, Cohen RA, Harezlak J, Schifitto G, Lam HY, Hua X, Zhong J,
Zhu T, Taylor MJ, Campbell TB, Daar ES, Singer E, Alger JR, Thompson PM*, Navia BA*, for the
HIVNC (2019). Progressive brain atrophy in chronically infected and treated HIV+ individuals.
Journal of NeuroVirology, Epub ahead of print.

Nir TM, Fouche JP, Ananworanich J, Ances B, Boban J, Brew BJ, Chaganti J, Ching CRK, Cysique
L, Gupta V, Harezlak J, Heaps J, Hinken C , Hoar J, Joska J, Kallianpur K, Kuhn T, Lebrun-Frenay
C, Levine A, Mondot L, Nakamoto B, Navia B, Paul RH, Pennec X, Porges ES, Pruksakaew K,
Shikuma C, Thames A, Valcour VG, Vassallo M, Woods AJ, Thompson PM, Cohen RA, Stein DJ,
Jahanshad N, for the ENIGMA-HIV Working Group (2019). Smaller limbic brain volumes are
associated with greater immunosuppression in over 1000 HIV-infected adults across five continents:
Findings from the ENIGMA-HIV Working Group. In preparation for submission to Lancet HIV, April
2019.

Nir TM, Jahanshad N, Villalon-Reina JE, Isaev D, Zavaliangos-Petropulu A, Zhan L, Leow AD, Jack
Jr CR, Weiner MW, Thompson PM, for ADNI (2017). Fractional anisotropy derived from the
diffusion Tensor Distribution Function boosts power to detect Alzheimer’s disease deficits. Magnetic
Resonance in Medicine, 78(6):2322-2333.

Zavaliangos-Petropulu A*, Nir TM*, Thomopoulos SI, Reid RI, Bernstein MA, Borowski B, Jack Jr
CR, Weiner MW, Jahanshad N, Thompson PM (2019). Diffusion MRI indices of cognitive impairment
in brain aging: The updated multi-protocol approach in ADNI3. Frontiers in NeuroInformatics, 13(2).

Nir TM, Lam HY, Ananworanich J, Boban J, Brew BJ, Cysique L, Fouche JP, Kuhn T, Porges ES,
Law M, Paul R, Thames A, Woods AJ, Valcour VG, Thompson PM, Cohen RA, Stein DJ, Jahanshad
N, for the ENIGMA-HIV Working Group (2018). Effects of diffusion MRI model and harmonization
iii

on the consistency of findings in an international multi-cohort HIV neuroimaging study. 2018 MICCAI
Workshop on Computational Diffusion MRI (CDMRI).

Nir TM, Lam HY, Jahanshad N, Ching CRK, Harezlak J, Martinez K, Schifitto G, Zhu T, Cohen RA,
Thompson PM, Navia BA, for the HIVNC (2018). Multimodal brain imaging predicts neurocognitive
impairment in chronic HIV. Conference on Retroviruses and Opportunistic Infections, Boston, MA,
March 2018.

Nir TM, Thomopoulos SI, Villalon-Reina JE, Zavaliangos-Petropulu A, Reid RI, Bernstein MA,
Borowski B, Jack Jr CR, Weiner MW, Jahanshad N, Thompson PM (2019). Multi-shell diffusion MRI
measures of brain aging: A preliminary comparison from ADNI3. 2019 IEEE International
Symposium on Biomedical Imaging (ISBI). © 2019 IEEE. Reprinted, with permission.

Nir TM, Jahanshad N, Toga AW, Jack Jr CR, Weiner MW, Thompson PM, ADNI (2015).
Connectivity network measures predict volumetric atrophy in mild cognitive impairment.
Neurobiology of Aging, 36(Suppl 1):S113-S120.

iv

TABLE OF CONTENTS

Chapter 1. Introduction…………………………………………………………………………….……..1
1.1 NeuroHIV………………………………………………………………………………….…..2
1.2 Mapping NeuroHIV and Neurological Disease…………………………………….……….…3
1.2.1 Mapping Anatomical Brain Morphometry …………………….……………………4
1.2.2 Brain Mapping with Diffusion-weighted MRI……………….………………….…..5
1.2.3 Connectomic Analysis of Brain Networks………………………………….……….6
1.2.4 ENIGMA-HIV………………………………………………………………………7
1.3 HIV and Aging…………………………………………………………………………….......8
1.4 Organization of the Dissertation……………………………………………………………..10
Chapter 2. T1-Weighted Volumetric Brain Abnormalities………………………………………..……12
2.1 Longitudinal Brain Atrophy in Treated HIV-infected Individuals…………………………..13
2.1.1 Introduction………………………………………………………………………...15
2.1.2 Methods…………………………………………………………………………….16
2.1.3 Results……………………………………………………………………………...19
2.1.4 Discussion…………………………………………………………………….……26
2.1.5 Acknowledgements…………………………………………………………..…….30
2.1.6 Supplementary Appendix…………………………………………………….…….31
2.2 Consistent Subcortical Brain Volume Associations in 1000 HIV-infected Individuals
from Five Continents……………………………………………………………….....……...37
2.2.1 Introduction………………………………………………………………………...40
2.2.2 Methods…………………………………………………………………………….43
2.2.3 Results……………………………………………………………………...………45
2.2.4 Discussion…………………………………………………………………….……51
2.2.5 Acknowledgements……………………………………………………………..….55
2.2.6 Supplementary Appendix………………………………………………………..…56
Chapter 3. Diffusion-weighted MRI Microstructural White Matter Abnormalities………………….…65
3.1 Novel Microstructural Measures Boost Power to Detect Neurodegenerative Disease…....…66
3.1.1 Introduction………………………………………………………………………...68
3.1.2 Methods……………………………………………………………………...….….70
v

3.1.3 Results…………………………………………………………………………...…76
3.1.4 Discussion……………………………………………………………………….…83
3.1.5 Acknowledgements…………………………………………………………….…..85
3.2 Pooling and Harmonizing Multi-protocol Diffusion MRI Data in ADNI.…………………...87
3.2.1 Introduction…………………………………………………………………….…..89
3.2.2 Methods……………………………………………………………………….…....91
3.2.3 Results………………………………………………………………………….…..96
3.2.4 Discussion………………………………………………………………………...107
3.2.5 Acknowledgements……………………………………………………………….112
3.2.6 Supplementary Appendix………………………………………………………....113
3.3 Pooling and Harmonizing Multi-site Diffusion MRI Data in an
International Multi-Cohort HIV Study…………………………………………...................133
3.3.1 Introduction……………………………………………………………………….135
3.3.2 Methods………………………………………………………………………..….137
3.3.3 Results………………………………………………………………………….…141
3.3.4 Discussion……………………………………………………………………...…144
3.3.5 Acknowledgements……………………………………………………………….146
Chapter 4. Multimodal Predictors of Cognitive Impairment………………………………………..…147
4.1 Multimodal Brain Imaging Predicts Neurocognitive Impairment in Chronic HIV…………148
4.1.1 Introduction…………………………………………………………………….....150
4.1.2 Methods……………………………………………………………………..…….151
4.1.3 Results……………………………………………………………………….……158
4.1.4 Discussion……………………………………………………………...…………163
4.1.5 Acknowledgements……………………………………………………………….165
4.1.6 Supplementary Appendix…………………………………………………………166
Chapter 5. Future Work: Advanced Diffusion MRI Techniques………….…….…………………….168
5.1 Multi-shell Diffusion MRI Measures of Brain Aging……………………………………….169
5.1.1 Introduction……………………………………………………………………….171
5.1.2 Methods……………………………………………….…………………………..172
5.1.3 Results…………………………………………………….………………………175
5.1.4 Discussion………………………………………………………………………...178
vi

5.1.5 Acknowledgements……………………………………………………………….179
5.2 Connectivity Network Measures Predict Brain Atrophy in Aging and AD………………….180
5.2.1 Introduction…………………………………………………………………...…..182
5.2.2 Methods…………………………………………………………………………...183
5.2.3 Results…………………………………………………………………………….189
5.2.4 Discussion………………………………………………………………………...191
5.2.5 Acknowledgements……………………………………………………………….194
References…………………………………………………………………………………...…………195

1

CHAPTER 1

Introduction

2

1.1 NeuroHIV

Although the incidence of new HIV infections each year has declined to 1.8 million, there are still an
estimated 36.9 million people living with HIV worldwide (UNAIDS 2018), and HIV remains a significant
public health concern. An estimated 59% of infected individuals have access to combination antiretroviral
therapy (cART), which has reduced transmission rates, increased survival rates, and improved quality of
life for those infected. Nevertheless, in less affluent contexts, many people eligible for treatment are not
currently receiving it. Moreover, the implications of long term survival with treatment are not yet well
understood.
After transmission, HIV rapidly replicates and crosses the blood brain barrier (BBB), either
through endothelial cells as free virions, or more likely, through infected CD4+ T-lymphocytes or infected
macrophage-monocytes (i.e., the “Trojan horse” hypothesis). It then settles and replicates in perivascular
macrophages and parenchymal microglial cells that express both the CD4 receptor and the CCR5 or
CXCR4 chemokine co-receptor (Peluso et al. 1985, Gonzalez-Scarano and Martin-Garcia 2005, Spudich
and Gonzalez-Scarano 2012). Although neurons are not directly infected by HIV, once in the central
nervous system (CNS), they are injured via indirect mechanisms, including viral proteins such as gp120,
Tat, and Vpr, neurotoxins resulting from the immune response inflammatory cascade (i.e., inflammatory
cytokines and chemokines (e.g., TNF, MCP1, IL-6), nitric oxide, and quinolinic acid), and glutamate
excitotoxicity resulting from glial cell death (Gonzalez-Scarano and Martin-Garcia 2005).
Neuronal dysfunction is often accompanied by neurocognitive deficits (McArthur et al. 2010).
About half of adults infected with HIV experience a range of neuropsychological symptoms that can
include a conjunction of motor (slowness, loss of balance), behavioral (apathy, mood disturbances), and
cognitive (mental slowing, attention/memory deficits) facets (Navia et al. 1986b, Heaton et al. 2010). The
prevalence of HIV-associated dementia has greatly diminished with cART access (Ances and Ellis 2007),
but many infected adults continue to experience a broad spectrum of neuropsychological symptoms known
as HIV-associated neurocognitive disorders or HAND (Navia et al. 1986a, Sacktor et al. 2002, Antinori
et al. 2007, Tozzi et al. 2007, Brew et al. 2009, Heaton et al. 2010, Heaton et al. 2011, Robertson et al.
2012). The incidence of HAND may, paradoxically, be increasing in the cART era and a spectrum of
abnormalities of varying extents have been observed in numerous different cognitive domains (Cysique
et al. 2004, McArthur 2004, Cysique et al. 2006, Ances and Ellis 2007, Woods et al. 2009, Sacktor et al.
2016). These persistent neurocognitive deficits may reflect distinct underlying HIV neuropathology in the
3

setting of long term survival that needs to be understood to inform treatment strategies. They may be
attributable to factors such as 1) prolonged cART exposure, which may be neurotoxic (Robertson et al.
2012); 2) a reservoir of ongoing low-grade viral replication in the CNS due to poor cART penetrance
(Cysique et al. 2004, Ellis et al. 2007, Anthony and Bell 2008, Hult et al. 2008); 3) accelerated
cerebrovascular disease from chronic immune activation and inflammation, coupled with other
comorbidities (hypercholesterolemia, diabetes, renal and hepatic dysfunction, etc.) (Valcour et al. 2005,
Becker et al. 2009); and 4) neurodegenerative processes that can occur with aging (Brew et al. 2009).
In the cART era, HIV-related comorbidities, including symptoms of brain dysfunction, remain
common among individuals on suppressive treatment. A better understanding of the neurobiological
consequences of HIV infection, both alone and in the context of cART, is essential to develop thorough
treatment guidelines, including when and how to intervene to optimize long term neuropsychological
outcomes.

1.2 Mapping NeuroHIV and Neurological Disease

In addition to neurocognitive testing, non-invasive, in vivo neuroimaging techniques will play an
important role in better understanding the spectrum of CNS impairments in HIV+ individuals. The virus
crosses the BBB soon after infection, potentially triggering brain aberrations before patients show any
signs or symptoms of disease. As in Alzheimer’s disease, where magnetic resonance imaging (MRI)
markers have been shown to be abnormally altered prior to the clinical presentation of the disease (Jack
et al. 2013), there is a need to identify reliable imaging biomarkers of HIV to understand, measure, and
predict disease neuropathology and progression.
To date, numerous neuroimaging studies have been published in an effort to identify the
neuropathological hallmarks of HIV and resulting neurocognitive impairments. Standard anatomical and
diffusion MRI (dMRI) studies frequently reveal greater ventricular expansion, cortical gray matter, basal
ganglia, and white matter (WM) atrophy and microstructural differences in HIV-infected patients
compared to seronegative controls (Tate et al. 2009, Ances and Hammoud 2014, Masters and Ances 2014,
Rahimian and He 2017, Chang and Shukla 2018), consistent with early neuropathological and
immunohistochemical studies of HIV encephalitis that show the presence of multinucleated giant cells
and microglial nodules, as well as viral antigens, with a predilection for subcortical structures and WM
(Navia et al. 1986a, Navia et al. 1986b, Neuen-Jacob et al. 1993, Brew et al. 1995, Berger and Nath 1997,
4

Morgello 2018). Many imaging studies have linked microstructural and volumetric changes to history of
disease severity (nadir CD4+ T-cell count), impaired immune response (lower current CD4+ count),
greater viral load, and degree of cognitive impairment. Furthermore, growing evidence suggests that
despite effective viral suppression on cART, chronic HIV infection can still promote ongoing brain
deficits (Cardenas et al. 2009, Cohen et al. 2010, Becker et al. 2011, Harezlak et al. 2011, Tate et al. 2011,
Ances et al. 2012, Hua et al. 2013a, Harezlak et al. 2014).
Ongoing work evaluating new and emerging brain imaging techniques with improved sensitivity
and specificity to track subtle changes in brain structure, coupled with initiatives that bring together
researchers and data from all over the world, will vastly increase the scope and power of HIV
neuroimaging studies to discover reliable neuroanatomical consequences of HIV infection and related
therapeutic targets (Thompson and Jahanshad 2015).

1.2.1 Mapping Anatomical Brain Morphometry

Standard anatomical MRI offers information about morphological features of the brain, which may change
with various neurological diseases. Numerous methods have been developed to segment brain structures
and tissue compartments from T1-weighted scans, and characterize morphological variations. Traditional
T1-weighted MRI-derived measures include cortical thickness, region of interest (ROI) volume, surface
area, and shape analysis metrics, among many others. Tensor-based morphometry (TBM) has been
validated as a powerful and unbiased technique to map disease-related regional brain volume differences,
both between groups cross-sectionally and within groups longitudinally, not only in HIV (Thompson et
al. 2005, Chiang et al. 2007, Wang et al. 2010, Hua et al. 2013a), but in large multi-site studies such as
the Alzheimer’s Disease Neuroimaging Initiative (Hua et al. 2016). Through non-linear registration of T1-
weighted scans, voxel-wise volumetric variations can be identified by examining the gradients of the
resulting deformation fields. Longitudinal neuroimaging studies are relatively rare in HIV+ populations,
but by mapping rates of brain tissue loss over time in HIV-infected individuals, we can determine which
brain systems lose tissue fastest, relate these loss patterns to cognitive impairment, and disentangle which
disease-related factors, at baseline, predict higher atrophy rates. Unlike traditional ROI-based approaches,
which may limit the scope of findings, TBM 3D Jacobian ‘expansion/contraction factor’ maps reflect local
voxel-wise changes throughout the brain, without requiring prior anatomical hypotheses. This may be
5

especially helpful in studying diseases with unknown or small local effects, or to evaluate regional WM
atrophy which, in T1-weighted images, lacks the contrast for reliable parcellation into tracts.

1.2.2 Brain Mapping with Diffusion-Weighted MRI

While anatomical MRI-based measures of HIV pathology are widely reported, diffusion MRI (dMRI)
may be more sensitive to subtle WM microstructural changes. dMRI is a variant of MRI that measures the
hindered or restricted diffusion of water molecules in brain tissue. Water diffusion in the brain is hindered
by barriers such as hydrophobic myelin sheaths which promote highly anisotropic water diffusion along
axons. By characterizing the diffusion process at the voxel level, it is therefore possible to make tentative
inferences about the underlying WM microstructure (Descoteaux and Poupon 2012).
Since the development of dMRI, along with improvements in acquisition protocols to increase
angular, spatial, and spectral resolution, multiple mathematical models have been developed to describe
the diffusion process. One of the first, and still most popular, methods developed to summarize diffusion
properties in a specific voxel is the single diffusion tensor model (DTI) (Basser et al. 1994). This model
assumes purely Gaussian diffusion, and is limited as it can only model a single fiber population at every
voxel. It cannot resolve complex features of WM architecture, such as dispersing, crossing, or ‘kissing’
fibers. At the current resolution of typical dMRI acquisitions, at least two-thirds of WM voxels contain
multiple fiber crossings (Behrens et al. 2007, Descoteaux 2008, Jeurissen et al. 2013). That said, DTI-
derived fractional anisotropy (FA) is still the scalar measure most widely used to characterize WM micro-
architecture in HIV and in most clinical populations (O’Connor et al. 2017). Furthermore, although FA
may be sensitive, it is somewhat non-specific as it is influenced by many microstructural factors such as
axonal diameter, packing density, membrane permeability, myelination, and intra-voxel orientation
coherence (Descoteaux 2008).
In recent years, a surge of new models and imaging techniques have been proposed to overcome
some of the single tensor model limitations. Signal based methods including high angular resolution
diffusion imaging (HARDI) —q-ball imaging (Tuch 2004) and spherical deconvolution (SD) (Tournier et
al. 2004)— diffusion spectrum imaging (DSI) (Wedeen et al. 2005), and multi-shell diffusion kurtosis
imaging (DKI) (Jensen et al. 2005), have helped resolve multiple dominant fiber directions within voxels.
Multi-shell dMRI acquisitions (i.e., with multiple b-values) are also better suited for fitting novel
biophysical models such as the composite hindered and restricted model of diffusion (CHARMED) (Assaf
6

and Basser 2005), and neurite orientation dispersion and density imaging (NODDI) (Zhang et al. 2012).
In addition to crossing fibers, the low spatial resolution of a single voxel typically captures partial volumes
from different tissue components (i.e., the CSF, myelin, intra and extracellular compartments) which
biophysical approaches attempt to isolate by fitting different models for each compartment; thus, derived
scalar measures may provide greater specificity, if not sensitivity, to underlying pathology.
Due to the numerous types of biological, neuropsychological, and imaging data often acquired for
clinical populations, time constraints are often placed on imaging protocols to reduce patient discomfort,
patient attrition, as well as motion, and ensure adequate sample sizes. This may prevent reliable
reconstruction of many of the aforementioned models, which can require extremely dense or multi-shell
dMRI acquisitions. However, multi-tensor models, such as the tensor distribution function (TDF), which
models crossing fibers as a probabilistic ensemble of Gaussian tensors (Leow et al. 2009), may still be
feasible, and derived measures more sensitive to disease-related microstructural differences than DTI.
Prior DTI studies of HIV+ patients have found lower FA and increased mean diffusivity (MD) in
the corpus callosum and frontal WM among other regions, indicating loss of WM “integrity” (Chang et
al. 2008, Wright et al. 2012, O’Connor et al. 2017, Oh et al. 2018). However, few published studies of
HIV take advantage of these newer models which may result in more robust measurements of disease
effects and give a richer understanding of the underlying HIV-related WM microstructural changes and
pathology.

1.2.3 Connectomic Analysis of Brain Networks

In addition to scalar diffusion metrics, dMRI tractography techniques may be used to track WM fiber
bundles connecting cortical and subcortical regions, allowing for characterization of the brain's WM
structural networks (Hagmann et al. 2008). Comparing structural neural network topology— using graph
theory metrics such as modularity, centrality, efficiency, and clustering— rather than evaluating regional
structures of the brain independently, can reveal organizational properties of brain networks, and may
therefore be more sensitive to alterations in brain systems as a whole. While these methods have been
used to characterize diseases such Alzheimer’s disease (Daianu et al. 2013a, Daianu et al. 2013b, Nir et
al. 2013, Prasad et al. 2013b), few studies have explored HIV-related changes in network topology
(Jahanshad et al. 2012, Baker et al. 2017, Bell et al. 2018).

7

1.2.4 ENIGMA-HIV

Across HIV neuroimaging studies, inconsistencies in the effect size, regional distribution, and even
direction of volumetric and microstructural brain associations reported have limited the generalizability
of the conclusions to date (O’Connor et al. 2017, O'Connor et al. 2018). Sources of heterogeneity between
findings of single cohort studies include methodological variability as well as differences in study
participants, including age, sex, and environmental, socioeconomic or lifestyle attributes of the cohorts,
and differences in study inclusion and exclusion criteria. While this is true for studies across multiple
diagnostic conditions, HIV is further complicated by differences in factors such as viral load status,
comorbidities and co-infections, drug use, age at infection, mode of transmission, duration of infection,
treatment regimen (access, type, timing, etc.), and degree of neurocognitive impairment, among others,
which can all differ drastically across studies.
Such factors may modulate how HIV impacts the brain. For example, the mode of transmission,
whether via intravenous drug use or perinatal infection, may play an important role in the profile of the
disease. The pattern of brain deficits in HIV+ children may vary from those who acquired HIV in
adulthood, for instance, due to HIV and potentially cART exposure during a time of critical brain
development and reorganization (Tardieu et al. 2000, Hoare et al. 2014). On the other end of the spectrum
are the clinical implications of infected individuals surviving into old age (Brew et al. 2009). The duration
of HIV infection (and chronic immune activation) and delay in initiation, type, and duration of treatment
may also moderate the impact of the disease on the brain (Cohen et al. 2010, Carvalhal et al. 2016).
Substance abuse is a major risk factor and cause of HIV transmission (CDC 2016). Substance use
not only increases the risk of transmission, but also affects general health, increasing comorbid conditions
which can accelerate disease progression and obstruct treatment adherence (Lucas 2011). For example,
co-infection with HIV and hepatitis C (HCV) most commonly occurs in individuals who use intravenous
drugs. Around 6.2% of people living with HIV have HCV, with co-infection estimates of 82.4% in drug
users (Platt et al. 2016). HIV, HCV, and drug use may each affect brain integrity and possibly interact
(Martin-Thormeyer and Paul 2009, Weissenborn et al. 2009, Ersche et al. 2013)
As HIV disproportionately affects less affluent parts of the world, many factors linked to
socioeconomic status, outside of access to cART and treatment compliance alone (Falagas et al. 2008),
may also play a large role in HIV prognosis (Hogg et al. 1994, Perry 1998). In addition to a higher risk of
HIV contraction, individuals are often exposed to several cumulative risk factors, including comorbid
8

illness, poor nutrition, adverse living conditions, and educational disadvantages (WHO 2003, Hackman et
al. 2010). Childhood health and social factors, often associated with poverty, can impact brain function
and psychological health later on in life (Duncan et al. 2010, Walker et al. 2011, Mani et al. 2013, Blair
and Raver 2016), and therefore may play a role in HIV prognosis.
It is not then surprising that, for example, both hypertrophy (Castelo et al. 2007) and hypotrophy
(Jernigan et al. 2005) have been reported in the basal ganglia of HIV+ individuals, or that there is
disagreement whether brain changes are always linked to the degree of immunosuppression (e.g., nadir or
current CD4+ count) and specific cognitive domain deficits (Chiang et al. 2007, Gongvatana et al. 2009,
Cohen et al. 2010, Becker et al. 2011, Jernigan et al. 2011, Ances et al. 2012). Ultimately, the inconsistent
findings in many single cohort neuroimaging studies may be explained by variations in study design
(differences in the characteristics of infected individuals selected compounded by heterogeneity in MRI
acquisition protocols and analysis techniques), insufficient power to detect differences with small sample
sizes, or an incomplete consideration of the many confounds associated with the heterogeneous HIV
population.
To address variations in methods, and boost statistical power, the HIV Working Group was
established within the Enhancing Neuro Imaging Genetics through Meta-Analysis (ENIGMA) consortium
to harmonize data analysis from neuroimaging studies around the world. By pooling datasets from
independent studies of diverse HIV cohorts across the world, well powered ENIGMA-HIV studies may
determine sources of brain differences that are otherwise difficult to disentangle, and assess whether these
factors are specific to one cohort or likely to generalize to HIV patients globally. It is important to not
only identify biomarkers that are consistently related to disease burden and functional impairment, but
also distinguish disease modulators and brain effects that may differ worldwide. Understanding common
neuropathogenic pathways of HIV-infection across international populations could ultimately help lead to
improved therapeutic targets, and surrogate markers to evaluate treatment effects in clinical trials.

1.3 HIV and Aging

An estimated 50% of HIV-infected individuals in the U.S. are over the age of 50 (the fastest growing age
group) (CDC 2017), as cART has improved survival in HIV-infected adults to near-normal longevity
(Rodger et al. 2013, Costagliola 2014). Despite viral suppression, people with chronic HIV infection have
a higher risk of multiple health conditions linked to advancing age including geriatric syndromes and
9

frailty, cardiovascular disease (hypertension and stroke), diabetes, cancers, liver, renal diseases and
chronic neurological complications (Brew et al. 2009, Wing 2016). This suggests that common age and
HIV-related pathological processes, such as immune dysregulation and senescence, and inflammation,
may accelerate aspects of the aging process (Pathai et al. 2014).
HIV-associated neurocognitive impairments and those associated with normal aging share many
similarities (Brew et al. 2009). For example, similarities in both aging and HIV include executive function
deficits and memory dysfunction as well as brain abnormalities in fronto-striatal and fronto-temporal
networks and hippocampal tissue (Cherner et al. 2004, Tucker et al. 2004, Raz and Rodrigue 2006, Brew
et al. 2009, Tate et al. 2009, Woods et al. 2009, Schouten et al. 2011, Cysique and Brew 2014, Pfefferbaum
et al. 2014, Kamkwalala and Newhouse 2017). The question remains whether HIV further facilitates
neurocognitive degeneration associated with aging, leading to early or accelerated aging (Brew et al. 2009,
Holt et al. 2012, Pathai et al. 2014). Older HIV-infected individuals show cognitive impairment twice as
often as their younger counterparts with the same duration of infection (Valcour et al. 2004a).
Furthermore, older age at the time of seroconversion increases the risk for cognitive impairment (Valcour
et al. 2004b, Bhaskaran et al. 2008). Chronic infection and increasing age may exacerbate brain injury and
neurodegeneration resulting in compounding or premature detrimental effects on cognition (Goodkin et
al. 2001, Wendelken and Valcour 2012, Cysique and Brew 2014, Cohen et al. 2015). Many studies report
independent detrimental effects of age and HIV on the brain (Gongvatana et al. 2011, Valcour et al. 2011,
Ances et al. 2012, Becker et al. 2012, Nir et al. 2014, Cohen et al. 2015), while some have successfully
detected the proposed interaction (Harezlak et al. 2011, Scott et al. 2011, Cysique et al. 2013, Seider et al.
2016, Kuhn et al. 2017).
The appearance and progression of age-related neurodegenerative diseases, such as Alzheimer's
disease (AD), may also be facilitated by HIV. Some studies suggest that the HIV inflammatory cascade
may promote an over production of beta-amyloid precursor protein (APP), as well as factors that degrade
APP into neurotoxic beta-amyloid, a hallmark of neurodegeneration in AD (Forloni et al. 1992, Stanley
et al. 1994, Adle-Biassette et al. 1999, Nebuloni et al. 2001, Liao et al. 2004). The HIV Tat protein may
further inhibit effective amyloid-beta degradation (Rempel and Pulliam 2005). In fact, levels of total-tau,
phosphorylated-tau, and beta-amyloid, hallmark pathological markers of AD, are highly correlated with
HIV as well in infected populations (Brew et al. 2005, Green et al. 2005, Anthony et al. 2006, Clifford et
al. 2009, Cohen et al. 2015). The apolipoprotein E4 (ApoE4) polymorphism is yet another link between
HIV infection and AD neurologic symptoms. The ApoE4 allele, which impacts amyloid metabolism, is
10

significantly linked to neurodegeneration in AD, and is the greatest known genetic risk factor for dementia
in aging individuals (Raber et al. 2004). In HIV+ individuals, ApoE4 genotype has been associated with
faster disease progression (Burt et al. 2008), increased neurotoxicity associated with the HIV Tat protein
(Turchan-Cholewo et al. 2006), higher brain beta-amyloid deposits (Green et al. 2005, Soontornniyomkij
et al. 2012), greater brain atrophy (Wendelken et al. 2016), and increased risk of HAND (Corder et al.
1998, Valcour et al. 2004b, Spector et al. 2010, Andres et al. 2011, Chang et al. 2011). Ultimately, amyloid
pathways —implicated in aging and AD— may also be perturbed by HIV-related inflammation, BBB
disruption, and neurotoxic proteins, and may accelerate pathogenesis (Milanini and Valcour 2017).
Given the increased life expectancy of HIV+ individuals in the cART era compared to the pre-
cART era, future neuropsychological and neuroimaging studies are needed to better understand chronic
HIV-infection in the context of aging. By using a combination of the anatomical and diffusion MRI
techniques described, as well as functional neuroimaging techniques, across a large collection of different
cohorts, it may become possible to map common brain deficits in the HIV+ population, while
disentangling factors that may modulate these effects, like age.

1.4 Organization of the Dissertation

Each of the studies described in this dissertation ultimately aims to serve ongoing and future research
efforts to map HIV-associated neurological impairments, and to better understand reliable biomarkers of
HIV neuropathology. Much of the preliminary work establishing the utility, sensitivity, and power of
novel neuroimaging methods to detect neurodegenerative disease effects in multi-site, clinical quality
data, was completed with publically available data from the Alzheimer’s Disease Neuroimaging Initiative
(ADNI). Compared to potentially small or unknown effects of HIV, AD-related brain changes have a well-
established signature and are robustly detected. ADNI data, therefore, serve to determine whether novel
neuroimaging methods identify expected associations. There are also many common mechanisms between
aging, AD, and HIV-related neurodegenerative brain processes, suggesting that methods sensitive to aging
and AD, may likely be sensitive tools for HIV studies. In Chapter 2, we analyze disease related differences
in regional brain volumes derived from T1-weighted MRI. We first use tensor-based morphometry
(TBM), a method previously validated in ADNI and shown to be one of the most sensitive markers for
detecting longitudinal brain atrophy, to evaluate longitudinal volumetric brain changes in a multi-site
study of HIV-infected individuals on stable treatment. We then assess whether common subcortical
11

volumetric signatures can be identified across 12 heterogeneous and independent ENIGMA-HIV cohorts
worldwide. In Chapter 3, we evaluate neurodegenerative disease effects on white matter microstructure
using diffusion MRI. We first use ADNI to validate and test a new dMRI model, the Tensor Distribution
Function (TDF), and show that, compared to the most frequently used single tensor model (DTI), the TDF
is a more accurate, stable, and sensitive model to detect disease effects with multi-site dMRI data of the
quality often acquired under clinical time constraints. We then test whether TDF remains more sensitive
when we pool and harmonize heterogeneous dMRI acquisition protocols from ADNI. Finally we test
multi-site, multi-protocol harmonization approaches and the utility of TDF to detect consistent HIV effects
on white matter microstructure, in 6 independent ENIGMA-HIV cohorts with dMRI data. In Chapter 4,
we combine measures from multiple neuroimaging modalities, including dMRI microstructural and T1-
weighted volumetric measures, to identify the best predictors of cognitive decline in treated individuals.
Finally, in Chapter 5, we complete preliminary studies in ADNI using advanced dMRI techniques,
including structural white matter connectivity analyses and a comparison of novel scalar measures derived
from multi-shell models, which we hope to extend to HIV in future work.

12

CHAPTER 2

T1-Weighted MRI Volumetric Brain Abnormalities

13

2.1 Longitudinal Brain Atrophy in Treated HIV-infected Individuals

This section is adapted from:

Nir TM, Jahanshad N, Ching CRK, Cohen RA, Harezlak J, Schifitto G, Lam HY, Hua X, Zhong J,
Zhu T, Taylor MJ, Campbell TB, Daar ES, Singer E, Alger JR, Thompson PM*, Navia BA*, for the
HIVNC (2019). Progressive brain atrophy in chronically infected and treated HIV+ individuals.
Journal of NeuroVirology, Epub ahead of print.

14

Progressive Brain Atrophy in Chronically Infected and Treated HIV+ Individuals

Talia M Nir
1
, Neda Jahanshad
1
, Christopher RK Ching
1,2
, Ronald A Cohen
3
, Jaroslaw Harezlak
4
,
Giovanni Schifitto
5
, Hei Y. Lam
1
, Xue Hua
1
, Jianhui Zhong
6
, Tong Zhu
7
, Michael J Taylor
8
, Thomas B Campbell
9
,
Eric S Daar
10
, Elyse J Singer
11
, Jeffrey R Alger
11
, *Paul M Thompson
1
, *Bradford A Navia
12
,
On behalf of the HIVNC
* These authors contributed equally to the manuscript

1
Imaging Genetics Center, Mark and Mary Stevens Neuroimaging and Informatics Institute, Keck School of Medicine,
University of Southern California, Marina del Rey, CA, USA
2
Graduate Interdepartmental Program in Neuroscience, UCLA School of Medicine, Los, Angeles, CA, USA
3
Department of Aging and Geriatric Research, University of Florida, Gainesville, FL, USA
4
Indiana University School of Public Health, Bloomington, IN, USA
5
Department of Neurology, University of Rochester, Rochester, NY, USA
6
Department of Imaging Sciences, University of Rochester, Rochester, NY, USA
7
Department Radiation Oncology, University of North Carolina, Chapel Hill, NC, USA
8
Department of Psychiatry, University of California, San Diego, CA, USA
9
Medicine/Infectious Diseases. University of Colorado Denver, Aurora, CO, USA
10
Los Angeles Biomedical Research Institute, Harbor-UCLA Medical Center, University of California, Los Angeles, CA, USA
11
Department of Neurology, David Geffen School of Medicine at UCLA, Los Angeles, CA, USA
12
Department of Public Health, Infection Unit, Tufts University School of Medicine, Boston, MA, USA

Abstract. Growing evidence points to persistent neurological injury in chronic HIV infection. It
remains unclear whether chronically HIV-infected individuals on combined antiretroviral therapy
(cART) develop progressive brain injury and impaired neurocognitive function despite successful
viral suppression and immunological restoration. In a longitudinal neuroimaging study for the HIV
Neuroimaging Consortium (HIVNC), we used tensor-based morphometry to map the annual rate
of change of regional brain volumes (mean time interval: 1.0 ± 0.5 yrs), in 155 chronically infected
and treated HIV+ participants (mean age 48.0 ± 8.9 yrs; 83.9% Male), using between two time
points. We tested for associations between rates of brain tissue loss and clinical measures of
infection severity (nadir or baseline CD4+ T-cell count and baseline HIV plasma RNA
concentration), HIV duration, cART CNS penetration-effectiveness scores, age, as well as change
in AIDS Dementia Complex stage. We found significant brain tissue loss across HIV+ participants,
including those neuro-asymptomatic with undetectable viral loads, largely localized to subcortical
regions. Measures of disease severity, age, and neurocognitive decline were associated with greater
atrophy. Chronically HIV-infected and treated individuals may undergo progressive brain tissue
loss despite stable and effective cART, which may contribute to neurocognitive decline.
Understanding neurological complications of chronic infection, and identifying factors associated
with atrophy may help inform strategies to maintain brain health in people living with HIV.

Keywords: HIV, ADC, MRI, Brain Volume, cART, TBM

15

2.1.1 INTRODUCTION

Combined antiretroviral therapy (cART) has dramatically improved life expectancies for HIV-infected
individuals (Nakagawa et al. 2013). Today, acute HIV encephalitis and HIV-associated dementia are far
less prevalent (Ances and Ellis 2007), but many HIV-infected adults still experience a range of
neurocognitive impairments (NCI) known as HIV-associated neurocognitive disorders (HAND), or the
AIDS dementia complex (ADC) (Navia et al. 1986a, Navia et al. 1986b, Antinori et al. 2007, Robertson
et al. 2007, Heaton et al. 2011). It is not well understood whether neurologically asymptomatic, chronically
HIV-infected adults on stable cART are at heightened risk for neurodegeneration and NCI as they age.
Brain imaging studies provide evidence of persistent brain decline in chronically HIV-infected
individuals on cART. Reduced cortical gray matter, basal ganglia, and white matter volumes, as well as
larger ventricular volumes, have been associated with duration of infection, cognitive impairment, brain
metabolite disruption, and immunological markers of disease severity –particularly nadir CD4+ count
(Cardenas et al. 2009, Cohen et al. 2010, Becker et al. 2011, Harezlak et al. 2011, Tate et al. 2011, Ances
et al. 2012, Hua et al. 2013a, Harezlak et al. 2014). These trends have even been identified in neuro-
asymptomatic individuals whose viral loads are suppressed by cART (Cohen et al. 2010, Tate et al. 2011).
To date, most studies have been cross-sectional with few longitudinal studies assessing rates of
brain atrophy. Whether viral suppression with stable treatment protects against premature decline is still
unresolved (Heaton et al. 2015, Correa et al. 2016, Sacktor et al. 2016, Sanford et al. 2018). As part of the
HIV Neuroimaging Consortium (HIVNC), we assessed chronically infected individuals on stable cART,
including virologically suppressed individuals with minimal or no NCI, to determine whether these
individuals continue to show patterns of progressive brain atrophy beyond that expected from aging. We
used a longitudinal brain mapping technique, tensor-based morphometry (TBM), to generate maps of
annual brain tissue loss. We hypothesized that 1) measures of infection severity at baseline would predict
brain atrophy rates, and 2) greater atrophy would be associated with decline in neurocognitive function.
Identifying the pattern of degeneration in treated asymptomatic individuals, along with factors associated
with rates of decline, may provide targeted neurological bases for treatments and help identify pre-
symptomatic individuals at heightened risk for NCI.

16

2.1.2 METHODS

HIVNC Participants

Between 2003 and 2009, 1.5 T T1-weighted MRI, clinical, and neuropsychological data were collected at
two time points (mean time interval: 1.0 ± 0.5 yrs) from 155 chronically HIV-infected HIVNC participants
(mean baseline age: 48.0 ± 8.9 yrs; 83.9% Male) on stable cART with a history of advanced disease (nadir
CD4+ count < 200 cells/mm
3
) across seven sites. Baseline demographic and clinical characteristics are
reported in Table 1. Inclusion criteria, clinical assessments, and MRI image acquisition parameters have
been previously described (Harezlak et al. 2011, Gongvatana et al. 2013, Hua et al. 2013a) and are
summarized in Supplementary Appendix A. Procedures were approved by local institutional review
boards. Participants gave written informed consent.
NCI was assessed using AIDS Dementia Complex (ADC) staging, as previously described (Navia
et al. 1986a, Navia et al. 1986b, Price and Brew 1988). Based on both clinical evaluations and
neuropsychological tests, participants were classified at baseline and prospectively as follows: ADC stage
0- neurologically asymptomatic (NA) with no evidence of cognitive, functional or neuropsychological
impairment; ADC stage 0.5- subclinical impairment, with evidence of neuropsychological impairment
only, as defined above; ADC stage 1- mild neurocognitive impairment with evidence of definite cognitive
and functional impairment on activities of daily living (ADL) but without loss of independent functioning;
ADC stage 2 -moderate impairment, requiring assistance on ADLs; or ADC stage 3- severe global
cognitive and functional impairment (Price and Brew 1988). Of 143 participants with available baseline
ADC stage, none were ADC stage 3, and only one was ADC stage 2. Two years after the start of this
study, a revised classification of HIV-associated cognitive impairment, referred to as the Frascati criteria,
was proposed (Antinori et al. 2007), where in general, ADC stage 0.5 approximately corresponds to
asymptomatic neurocognitive impairment (ANI), ADC stage 1 to mild neurocognitive disorder (MND),
and ADC stage 2 or greater to HIV-associated dementia or HAD.

17

Table 1. Clinical characteristics of 155 HIVNC study participants. Mean and standard deviation (SD) are listed for
continuous variables. Percent and absolute number are noted for categorical variables.
Continuous Variables Mean (SD)
Age (yrs) 48.0 (8.9)
HIV Duration (yrs) 11.6 (6.9)
cART Duration (yrs) 5.4 (4.5)
CPE Score
a
8.2 (3.1)
Nadir CD4+ (cells/mm
3
) 58.6 (58.2)
CD4+ (cells/mm
3
)
a
369.1 (203.0)
Inter-Scan Time Interval (yrs) 1.0 (0.5)
Categorical Variables % (N)
Sex (male) 83.9% (130)
Education (≤ high school) 43.9% (68)
Race/Ethnicity
Caucasian
African American
Asians/Native Alaskan or American Indian

69.7% (108)
27.8% (43)
2.6% (4)
Suppressed CD4+ (≤ 350 cells/mm
3
)
a
53.6% (82)
Detectable Plasma HIV RNA (≥400 copies/mL)
a
24.8% (38)
ADC
a

Stage 0:
Stage 0.5:
Stage 1:
Stage 2:

64.3% (92)
24.5% (35)
10.5 % (15)
0.01% (1)
a
Measures only available in subset of participants: CPE score N=152; CD4+ and Suppressed CD4+ N=153;
Detectable Plasma HIV RNA N=153; ADC Stage N=143.

Image Processing and Tensor-Based Morphometry (TBM)

Tensor-based morphometry (TBM), as described in (Hua et al. 2013a, Hua et al. 2013b, Hua et al. 2016),
is a robust and sensitive technique to create three-dimensional maps that reflect the rates of change at each
point (voxel) in the brain. Briefly, each subject's pre-processed follow-up T1-weighted scan was registered
to its baseline scan using a non-linear inverse-consistent elastic mutual information based registration
algorithm (Leow et al. 2007). The resulting Jacobian determinant map, or change map, characterizes the
local volume differences (expansion/contraction) between the two scans. For each subject, change maps
were normalized by the respective time interval (in years) between scans to reflect the annualized rate of
change in each voxel. To align maps across participants for statistical analysis, baseline T1-weighted scans
were warped to a study specific template, and the resulting warp was applied to each subject’s change
map. The cerebellum was often partly outside of the field of view (FOV; i.e., cropped during image
acquisition) and was therefore excluded from all analyses. For further details, see Supplementary
Appendix C.
18

Group Average Maps of Annual Rates of Brain Atrophy

We computed the mean annual rate of brain tissue loss (percent) in HIV+ participants at each voxel. Mean
maps were computed in HIV+ subgroups classified as either neurologically asymptomatic (NA; ADC=0;
N=92) or NA with undetectable viral load (NA-uVL; plasma HIV RNA < 400 copies/mL; N=76). A two-
tailed, one-sample T-test was used to find regions with significant change (change not equal to zero). We
corrected for multiple comparisons across voxels using the searchlight false discovery rate (sFDR) method
at q=0.05 (Langers et al. 2007).
For each group, voxel-wise annual change was averaged within 26 regions of interest (ROIs)
spanning cortical and subcortical gray matter (GM), white matter (WM), and cerebrospinal fluid (CSF)
compartments (Supplementary Table S3). To generate the ROIs, the template to which all Jacobian maps
were spatially normalized was segmented with FreeSurfer (Fischl et al. 2004). A two-tailed, one-sample
T-test was used to find ROIs with significant mean change across participants, at the multiple comparisons
FDR corrected P-value (Benjamini and Hochberg 1995).

Statistical Associations with Annual Rates of Brain Atrophy

Random-effects multiple linear regressions were used to test associations between brain atrophy rates and
variables of interest. We covaried for baseline age, sex, race/ethnicity, and inter-scan time-interval. To
ensure baseline brain volumes did not confound our analyses of brain tissue change, we also covaried for
baseline TBM maps representing volumetric differences from the template (i.e., the degree of volume
differences already present at baseline). To account for possible scanner effects, scanning parameters and
acquisition site were used as the grouping variable. All voxel-wise analyses were corrected for multiple
comparisons using sFDR at q=0.05 (Langers et al. 2007).
We tested for group differences in annual brain volume changes between HIV+ neuro-
asymptomatic (NA) and symptomatic (NS) participants, and between the subset of NA-uVL participants
and the remaining population. Across HIV+ participants, we evaluated the effects of age as well as HIV-
related clinical variables at baseline on annual volumetric change: nadir CD4+, immunologically
reconstituted current CD4+ status —defined as a CD4+ > 350 cells/mm
3
, the threshold below which cART
is generally recommended (World Health Organization 2015), and detectable HIV RNA viral load (dVL)
19

in plasma (≥ 400 copies/mL). Secondary analyses evaluated duration of HIV infection and cART CNS
penetration-effectiveness (CPE) scores (Letendre et al. 2009).
Variables that showed a statistically significant association in the entire HIV+ population were
then subsequently tested in two subsets: 1) all NA participants (the majority of the study population) and
2) a smaller NA-uVL subset. Finally, we assessed whether brain volume changes were associated with
longitudinal changes in ADC stage. Details on variable definitions are available in Supplementary
Appendix D.

Uninfected Comparison Group - Parkinson's Progression Markers Initiative (PPMI) Controls

Because healthy controls were not part of the original study design, healthy control 3 T T1-weighted MRI
scans were obtained from the publicly-available, multi-center Parkinson's Progression Markers Initiative
(PPMI; Supplementary Appendix B) (Marek et al. 2011). Longitudinal MRIs from 63 controls (mean
age: 59.3 ± 10.9 yrs; 61.9% male) were downloaded from the PPMI database (http://www.ppmi-info.org/).
A subset of 48 controls were selected to match on age, sex, race/ethnicity, and education with a subset of
65 HIV+ HIVNC participants. This HIV+ subset was representative of the full cohort for all HIV-related
measures. Demographic and clinical characteristics for the full and matched PPMI and HIVNC groups are
reported in Supplementary Tables S1-S2. As was done for HIVNC, PPMI baseline and follow-up T1-
weighted images were pre-processed, and change maps generated and registered to the HIVNC template
for an exploratory comparison with HIV+ individuals.

2.1.3 RESULTS

Clinical and Demographic Characteristics

Demographic and clinical factors were compared between groups using one-way ANOVA or chi-squared
analyses. Within the full HIV+ cohort, compared to neuro-symptomatic participants (NS), the NA group
was comprised of a marginally higher number of Caucasian participants (P=0.021) and fewer participants
with detectable viral loads (P=0.011; Table 2). Compared to NA-uVL participants, those either with dVL
or NS were significantly older (P=0.010) with longer HIV duration (P=0.004).

20

Annual Rates of Brain Volume Change

The full HIVNC cohort, NA participants, and those NA with uVL all showed significant CSF expansion
and widespread tissue atrophy (Figure 1a-c), especially within subcortical structures and WM (Pcorrected ≤
0.05). Mean regional changes, summarized in Supplementary Table S3, reveal the greatest annual
changes were consistently detected in the pallidum and thalamus (0.4-0.5% atrophy annually), and
ventricles (up to 0.7% expansion), followed by the putamen and nucleus accumbens (0.3-0.4% atrophy).
Rates of brain change were similar between HIV+ groups. There were no significant differences
between NA participants and those with an ADC stage greater than 0, or between NA-uVL and the
remainder of the population.

Table 2. Comparison of demographic and clinical characteristics between neuro-asymptomatic subgroups and
symptomatic participants using ANOVA or chi-squared tests. Key: NA: Neuro-asymptomatic (ADC=0); NS:
Neuro-symptomatic (ADC>0); dVL: Detectable Viral Load (plasma HIV RNA ≥ 400 cp/mL); uVL: Undetectable
Viral Load (plasma HIV RNA <400 cp/mL); SD: Standard Deviation.

NS
N=51
NA
N=92
P-Value
NS or dVL
a
N=66
NA-uVL
N=76
P-Value
Continuous Variables Mean (SD) ANOVA Mean (SD) ANOVA
Age (yrs) 50.0 (8.0) 47.0 (9.6) 0.066 50.2 (8.3) 46.3 (9.5) 0.011*
HIV Duration (yrs) 12.9 (6.5) 10.6 (7.2) 0.068 13.1 (6.4) 9.8 (7.1) 0.004*
cART Duration (yrs) 5.1 (4.9) 4.7 (4.0) 0.376 5.1 (4.52) 4.7 (4.1) 0.508
CPE Score
8.1 (3.3)
b
N=49
8.1 (2.7) 0.977
7.8 (3.24)
b
N=64
8.3 (2.6) 0.298
Nadir CD4+ (cells/mm
3
) 67.8 (62.5) 57.1 (56.7) 0.297 65.0 (60.3) 57.6 (58.0) 0.456
CD4+ Count (cells/mm
3
) 387.0 (205.8)
370.2 (197.3)
b
N=91
0.632 372.8 (211.5) 379.3 (190.6) 0.848
Inter-Scan Time Interval (yrs) 1.1 (0.5) 1.0 (0.6) 0.613 1.0 (0.5) 1.1 (0.5) 0.855
Categorical Variables % (N) χ
2
-Test % (N) χ
2
-Test
Sex (male) 78.4% (40) 84.8% (78) 0.338 80.3% (53) 84.2% (64) 0.542
Education
(≤ high school)
47.1% (24) 41.3% (38) 0.506 42.4% (28) 44.7% (34) 0.782
Race/Ethnicity (Caucasian) 54.9% (28) 75.0% (69) 0.014* 53.0% (35) 80.3% (61) 0.001*
Suppressed CD4+
(≤350 cells/mm
3
)
49.0% (25)
53.9% (49)
b
N=91
0.581 53.0% (35) 51.3% (39) 0.838
Detectable Plasma HIV RNA
(≥400 copies/mL)
35.3% (18)
16.5% (15)
b
N=91
0.011* 50.0% (33) 0% (0) 1.98x10
-12
*
* P ≤ 0.05
a
N = 18 were both NS with dVL
b
Total N with available data are noted

21

Age and HIV Infection Severity Predict Brain Tissue Loss Rates

We evaluated the independent effects of age on brain atrophy rates in the entire HIV+ cohort, and found
that older age at baseline was associated with greater tissue loss (Pcorrected ≤ 0.05). For each year in age,
HIV+ participants showed an average of 0.04% greater ventricular/CSF expansion and 0.04% greater
tissue atrophy (Table 3a; Figure 1d). Similar patterns and rates were detected in the subsets of NA and
NA-uVL participants (Figure 1e,f). The association between brain atrophy and age was relatively
localized and did not account for the pervasive brain atrophy detected across participants over time. Age
and duration of infection were not strongly correlated (Pearson’s r=0.30; P = 0.0001) and associations
between age and atrophy rates remained significant when covarying for HIV duration.
When covarying for age, indices of HIV infection also predicted greater brain atrophy and CSF
space expansion in the full HIV+ cohort. Participants with a lower current CD4+ cell count showed on
average a 0.78% increase in ventricular/CSF expansion rates and 0.78% increase in atrophy rates in the
WM, bilateral thalami, caudate, and putamen, and right globus pallidus and amygdala (Figure 2a; Table
3b; Pcorrected ≤ 0.05). On average, every 10 unit reduction in nadir CD4+ count was associated with 0.06%
faster sulcal CSF expansion and 0.06% increase in atrophy rates in the left thalamus and internal capsule
WM (Figure 2b; Table 3c). Compared to virologically suppressed participants, those with detectable
plasma HIV RNA (dVL) showed 1.0% greater ventricular/CSF expansion, and 0.82% greater atrophy
throughout the WM, thalamus, globus pallidus, and putamen as well as the left hippocampus and amygdala
(Figure 2c; Table 3d). Atrophy rates were not associated with duration of infection, or CPE score after
correcting for multiple comparisons. Ranking the significant baseline HIV-related predictors based on the
spatial extent of their effects across the brain revealed the following order of relative importance on brain
changes: dVL, current CD4+ status at baseline, and nadir CD4+ cell count (Table 3).
The subset of NA participants showed similar atrophy rates to the full HIV+ cohort (Table 3);
greater ventricular/CSF expansion and tissue atrophy were associated with dVL and lower current or nadir
CD4+ counts. However, findings were less widespread, perhaps due to a smaller sample size (Figure 2d-
f). While atrophy in NA-uVL participants was not associated with nadir CD4+, significant associations
were detected with current CD4+ (Figure 2g).

22

Figure 1 Atrophy rates in HIV+ individuals. Annual volumetric change (%) maps averaged across (a) the full
HIVNC cohort (mean age: 48.01 ± 8.88 yrs; N=155) revealed significant ventricular expansion (red) and tissue
atrophy (blue) throughout the brain (regions with no change are colored gray/white; one-sample T-test; P corrected ≤
0.05). Maps of the (b) subset of neuro-asymptomatic individuals (NA; ADC Stage=0; N=92) and (c) those who
were NA with undetectable viral load (NA-uVL; plasma HIV RNA < 400 copies/mL; N=76) showed a similar
distribution and magnitude of change as the full cohort. (d-f) In the full cohort and NA subsets, older age at baseline
was significantly associated with greater annual ventricular and sulcal expansion and tissue atrophy (P corrected ≤ 0.05).
However, the association between brain atrophy and age was relatively localized and did not account for the
pervasive brain atrophy detected across participants over time.
23

Figure 2. Brain tissue loss rates were associated with measures of HIV infection severity. (a-c) In the full cohort
and (d-f) the subset of neuro-asymptomatic (NA) participants (ADC=0), lower nadir CD4+ cell count, suppressed
immune status at baseline (current CD4+ ≤350 cells/mm
3
), and detectable viral load (plasma HIV RNA ≥ 400
cp/mL) at baseline were significantly associated (P corrected ≤ 0.05) with a greater annual ventricular and sulcal
expansion (red) and tissue atrophy (blue). Only (g) current CD4+ was associated with volume loss in NA
participants with undetectable viral load (NA-uVL).

Changes in ADC Stage and Tissue Loss

134 HIVNC participants had ADC staging at both time points; 78.4% remained stable, 7.5% improved
(ADC decrease), and 14.2% declined (10.4% went from ADC=0 to ADC ≥ 0.5; Figure 3a). Change in
ADC status was not significantly associated with either change in viral load or current CD4+ status
(CD4+: Spearman's r = -0.15, P = 0.096; VL: Spearman's r = 0.10, P = 0.614; Supplementary Appendix
H). Compared to those who remained stable or improved (N=115), an increase in ADC stage (N=19),
24

reflecting neurocognitive decline, was associated with on average 2.1% greater ventricular expansion and
0.91% greater atrophy sparsely distributed in temporal WM (Table 3e; Figure 3b). Greater differences
were detected when an increase in ADC stage (N=19) was compared only to participants who improved
(N=10): 3.1% greater ventricular expansion and 1.4% greater atrophy in the left frontal, temporal,
occipital, and internal capsule WM, as well as the left putamen, pallidum, and thalamus (Figure 3c). No
significant difference was detected between those who remained stable and those who improved.

Figure 3. Brain tissue loss rates were associated with change in ADC stage. (a) Total number of participants (out
of 134) for each type of change in ADC stage between baseline and follow-up assessments. Relative to participants
with (b) stable or improved neurocognitive status (no change or decreased ADC stage; N=115) or (c) just improved
neurocognitive status (decreased ADC stage; N=10) over time, a decline in neurocognitive status (an increase in
ADC stage; N=19) was significantly associated (P corrected ≤ 0.05) with greater rates of tissue atrophy (blue) and
ventricular expansion (green-red).
25

Table 3. Summary of effects from Figures 1-3 are reported as (1) the spatial extent or percent of voxels (out of
1,625,341 voxels tested) that were significant after multiple comparisons correction (P corrected ≤ 0.05) and (2) the
interquartile range (IQR) and mean percent change in those voxels with positive or negative associations with each
measure. Key: NA: Neuro-asymptomatic; uVL: Undetectable Viral Load.
Analysis
Significant
Voxels
Positive/Expansion
IQR (Mean)
Negative/Atrophy
IQR (Mean)
Baseline Clinical Predictors
A. Age
(per year older)
Full Cohort 3.62% 0.032-0.047% (0.043%) 0.030-0.045% (0.038%)
NA 6.69% 0.036-0.052% (0.046%) 0.033-0.048% (0.042%)
NA-uVL 3.88% 0.037-0.053% (0.045%) 0.036-0.055% (0.046%)
B. Suppressed current CD4+
(≤ 350 cells/mm
3
)
Full Cohort 3.67% 0.58-0.84% (0.78%) 0.57-0.94% (0.78%)
NA 1.85% 0.64-0.83% (0.74%) 0.83-0.99% (0.84%)
NA-uVL 0.85% 0.69-0.92% (0.81%) 0.71-1.01% (0.88%)
C. Nadir CD4+
(every 10 cells/mm
3
lower)
Full Cohort 1.45% 0.049-0.066% (0.057%) 0.047-0.078% (0.063%)
NA 0.15% 0.057-0.072% (0.069%) 0.045-0.054% (0.049%)
NA-uVL 0 -- --
D. Detectable plRNA
(≥ 400 copies/mL)
Full Cohort 8.95% 0.77-1.21% (1.03%) 0.66-0.94% (0.82%)
NA 1.13% 1.62-2.62% (2.14%) 0.93-1.22% (1.08%)
Neurocognitive Impairment
E. ADC Change
Decline
a
vs
Stable/Improve

0.48% 1.42-2.67% (2.07%) 0.75-1.11% (0.91%)
Decline
a
vs
Improve
2.12% 1.96-4.05% (3.05%) 1.030-1.79% (1.43%)
a
Decline: Neurocognitive decline (increase in ADC stage)

Annual Rates of Brain Volume Change in HIV+ Participants Compared to Controls

HIVNC HIV+ participants were significantly younger (P=1.16x10
-13
), less educated (P=0.0001), and were
comprised of fewer females (P=0.0004) and Caucasians (P=0.0004) than the full PPMI control cohort
(Supplementary Table S1). The subset of HIV+ participants selected to match the demographic
characteristics of PPMI were not significantly different clinically to the full HIV+ group, however they
were significantly older (P=2.47x10
-6
) and had fewer males (P=0.048) than the full group in order to
match demographic characteristics of the PPMI cohort (Supplementary Table S2).
Voxel-wise analyses (Figure 4) revealed that compared to healthy controls (N=48; mean age:
mean age: 55.6 ± 9.5 yrs), matched HIV+ participants (N=65; mean age: 54.2 ± 7.9 yrs) had significantly
26

greater tissue atrophy rates in small regions bilaterally in the temporal lobes and in the left precuneus
(Pcorrected≤0.05, Mean: 0.67%; IQR: 0.58-0.77%; Significant Voxels: 0.12%).

Figure 4. Atrophy rates in HIV+ individuals compared to healthy controls. Compared to a subset of age matched
controls (N=48; mean age: 55.6 ± 9.5 yrs), HIV+ individuals (N=65; mean age: 54.2 ± 7.9 yrs) showed significantly
(P corrected ≤ 0.05) greater rates of GM atrophy bilaterally in the temporal lobes and the left precuneus.

2.1.4 DISCUSSION

This study represents one of few published longitudinal studies to map patterns of ongoing regional brain
volume loss in a large cohort of chronically HIV-infected middle-aged participants on cART, including
neurologically asymptomatic and virologically suppressed individuals. While recent cross-sectional
studies have identified disrupted brain processes in HIV-infected individuals with chronic, stable disease
(Cohen et al. 2010, Tate et al. 2011, Harezlak et al. 2014), relatively few longitudinal studies have been
conducted, and there is little consensus regarding the pattern and degree of ongoing brain change. Our
study has two main findings: 1) atrophy (CSF expansion and tissue loss) persists in HIV-infected
individuals despite cART and 2) rates of brain atrophy in HIV are associated with plasma measures of
infection severity and neurocognitive decline.
As life expectancy of HIV-infected individuals has increased significantly in the setting of cART,
it has been postulated that HIV-associated brain injury and NCI may accelerate with aging (Brew et al.
2009, Holt et al. 2012, Brew and Cysique 2017, Ding et al. 2017). Several cross-sectional studies have
reported either independent or additive effects of age and HIV (Ances et al. 2012, Becker et al. 2012, Nir
et al. 2014, Cohen et al. 2015) or HIV-by-age interactions (Harezlak et al. 2011, Cysique et al. 2013, Kuhn
27

et al. 2017). The HIVNC study did not include seronegative individuals as part of the original study,
necessary to directly compare HIV effects with healthy aging. However, the association between brain
atrophy and age was relatively localized and did not account for a majority of the annual tissue loss
detected throughout the brains of HIV+ participants. We found that markers of HIV severity and age were
both independent, and sometimes overlapping, predictors of progressive brain atrophy in HIV+
participants, suggesting atrophy beyond that expected from healthy aging alone. To further confirm
atrophy rates beyond that expected with healthy aging, a group of seronegative participants from the multi-
site longitudinal PPMI study were used in an exploratory analysis. We found suggestive evidence that
longitudinal atrophy rates in HIV+ individuals on stable cART were significantly higher than those in a
group of age-matched healthy controls, in line with other longitudinal studies (Cardenas et al. 2009,
Clifford et al. 2017). Compared to HIVNC T1-weighted brain scans, which were acquired with 1.5 T MRI,
PPMI scans were acquired with 3 T MRI, which can result in improved tissue contrast, and affect volume
estimates (Jovicich et al. 2009, Heinen et al. 2016, Lysandropoulos et al. 2016). The magnitude and extent
of the differences between HIV+ and HIV- individuals presented in this paper should be interpreted with
caution due to differences in participant inclusion criteria and MRI acquisition between studies. These
preliminary findings motivate the need for additional longitudinal MRI studies with matched seronegative
controls.
Within HIV+, atrophy was associated with baseline measures of infection severity: lower current
CD4+ count, higher HIV RNA levels, and lower nadir CD4+. The subcortical pattern of atrophy associated
with markers of infection severity in this study is consistent with early neuropathological and
immunohistochemical studies of HIV encephalitis that show the presence of multinucleated giant cells
and microglial nodules, as well as viral antigens, with a predilection for subcortical structures (Navia et
al. 1986a, Navia et al. 1986b, Neuen-Jacob et al. 1993, Brew et al. 1995, Berger and Nath 1997, Morgello
2018), such as the putamen where decreased neuronal densities have been identified post mortem (Everall
et al. 1995). Similarly, more recent in vivo neuroimaging studies suggest that HIV prominently affects the
basal ganglia and WM (Tucker et al. 2004, Cohen et al. 2010, Jernigan et al. 2011, Ances et al. 2012,
Heaps et al. 2015). HIVNC participants all had a history of advanced disease (nadir CD4+ count < 200
cells/mm
3
); nadir CD4+ count has been associated with brain volume deficits and cognitive impairment
in several cross sectional studies, suggesting that severe immunosuppression may lead to persistent and
potentially irreversible brain injury despite immune recovery in individuals on cART (Ellis et al. 2011,
Jernigan et al. 2011, Tate et al. 2011, Hua et al. 2013a), and reinforcing the need for early intervention.
28

However, we found that among baseline measures of disease severity, detectable plasma VL was a better
predictor of progressive atrophy throughout the brain, followed by current CD4+ status, and then nadir
CD4+. Studies of post mortem brain specimens in the antiretroviral era show correlations between elevated
plasma HIV viral load and HIV brain-tissue viral load and pathology (Everall et al. 2009, Gelman et al.
2013). These results suggest that, against a background of a severe immunosuppression and aging,
ongoing HIV-related processes contribute to progressive brain change. Two smaller longitudinal studies
have similarly shown accelerated loss of WM volume and regional cortical volume in individuals with
lower current CD4+ count (Pfefferbaum et al. 2014) and detectable VL (Cardenas et al. 2009). These
results support the importance of achieving viral suppression and adequate immunological restoration with
effective cART.
Importantly, tissue loss throughout the brain was found in NA participants with uVL, and was not
significantly different than that detected in NS participants or those with dVL, consistent with cross-
sectional studies that suggest brain decline persists in these individuals (Harezlak et al. 2011, Tate et al.
2011). A prospective MRS study of the HIVNC cohort revealed significant decreases in neuronal and glial
cell function in NA participants (Gongvatana et al. 2013). Together, these findings suggest that HIV-
related brain injury may persist or worsen over time even in treated HIV-infected individuals with a history
of advanced disease, who have successful viral suppression and stable disease. In contrast to the full HIV+
cohort, other than age, only current CD4+ count (i.e., not nadir CD4+) was significantly associated with
these changes, and with modest effect sizes. We cannot, however, discount that ongoing low-grade viral
replication, undetected by the older assay thresholds used in this study, may drive these brain changes.
Sequestered HIV RNA has been found in postmortem brain tissue from cART patients with uVL (Lamers
et al. 2016). Two smaller longitudinal studies (N ≤ 48) of similarly aged HIV+ individuals with VL < 50
copies/mL report similar changes in brain volume over time between HIV+ and uninfected participants
(Correa et al. 2016, Sanford et al. 2018). Concordantly, participants in Correa et al. (2016) and Sanford et
al. (2018) showed higher median current CD4+ counts than HIVNC participants (678 and 630 cells/mm
3

respectively), and, in Sanford et al. higher median nadir CD4+ counts (190 cells/mm
3
, implying a history
of severe immunosuppression), prompting further investigation to determine viral load, current and nadir
CD4+ count thresholds that mitigate atrophy. Although the underlying mechanisms remain unclear, pro-
inflammatory factors including chemokines such as MCP-1 and sCD14, which were shown to contribute
to subcortical and WM injury in the HIVNC cohort, may also play a role (Anderson et al. 2015). Similar
29

to the mechanisms that underlie some of the systemic complications of chronic HIV infection, persistent
immune activation also likely contributes to HIV neuropathogenesis (Deeks et al. 2013).
Building on evidence suggesting that neuro-asymptomatic HIV-infected individuals can develop
progressive decline in cognitive function (Grant et al. 2014), our results show that such neurocognitive
decline is significantly associated with atrophy rates in subcortical GM and WM regions. A recent HIVNC
study found that reduced levels of the neuronal marker N-acetylaspartate in the basal ganglia was the most
significant predictor of conversion to neurocognitive impairment, relative to the same clinical predictors
evaluated here (Navia et al. In Review). Together, these results are consistent with a growing body of
evidence pointing to the pathogenic role of these subcortical structures in HAND (Navia et al. 1986a,
Navia et al. 1986b).
When viewed closely, maps of associations between greater indices of infection severity and CSF
expansion may appear to extend into neighboring cortical GM (indicating slower rates of GM loss with
increased disease severity), which could reflect compensatory hypermetabolic or inflammatory processes
that occur at early stages of neurocognitive impairment (a majority of the cohort) (Rottenberg et al. 1987,
Hinkin et al. 1995, von Giesen et al. 2000, Chang et al. 2004, Castelo et al. 2007), or ceiling effects at the
parenchyma/CSF interface. TBM has been validated as a powerful and unbiased technique to map
longitudinal brain change, especially in large multi-site studies (Hua et al. 2016). However, changes in
cortical GM are difficult to measure with registration based methods, as the cortex is thin and prone to
partial volume effects, with sulcal and morphological variability leading to subtle misalignments. TBM is
arguably better powered to detect subcortical changes (e.g., WM and subcortical GM structures) (Hua et
al. 2009b, Hua et al. 2013a, Hua et al. 2013b, Hua et al. 2013c). Even so, voxel-wise studies allow for
more regionally unbiased analyses compared to predetermined ROIs, which limit our power to map
patterns of complex effects. Further validation of findings using other independent methods and cohorts
may help evaluate cortical GM effects.
Another potential limitation of the TBM method used here is that evaluating change between two
time points assumes that changes are linear. Although estimating longitudinal brain changes in a cohort is
commonly achieved with two time points (Cardenas et al. 2009; Clifford et al. 2017; Correa et al. 2016;
Sanford et al. 2018), more informative rates and trajectories may be calculated with a greater number of
time points.
The immediate clinical utility for neuroimaging in the setting chronic HIV infection and cART is
limited given the small number of longitudinal studies published to date, but further evaluation of regional
30

brain changes along with the contributing factors will be important to form a better understanding of
disease progression in relation to cognitive decline and treatment intervention. Based on available
literature in other fields, notably Alzheimer’s disease (Reiman and Jagust 2012, Hua et al. 2016, Veitch
et al. 2018), imaging biomarkers have provided an important non-invasive, in vivo approach to identify
subgroups at higher risk for cognitive decline and a marker to monitor the effects novel drug treatments
in clinical trials that aim to slow or halt such decline (McArthur 2012, Chang and Shukla 2018).
Identifying regions of disease-specific vulnerability in chronic HIV infection using noninvasive MRI
techniques may be an important step towards determining brain regions that may be associated with
cognitive decline which in turn may suggest targets for therapeutic intervention. In addition, future studies
of associations between volumetric changes in chronically HIV-infected patients as described in this
manuscript and pro-inflammatory factors or cellular changes as detected by proton MRS, would further
our understanding of processes contributing to HIV-related brain injury.
While standards in neuropsychological and viral load evaluation may have shifted from the time
the HIVNC study was conducted, our findings add to a body of evidence that chronically HIV-infected
individuals, even on cART, are at increased risk for future brain tissue loss and cognitive decline. The
findings from this study support a critical unmet need to identify novel therapies to protect the CNS, even
in the era of effective cART treatment.

2.1.5 ACKNOWLEDGEMEMNTS

The study was funded by NIH NINDS R01 NS080655 and U54 EB020403. Data used in the preparation
of this article were obtained from the Parkinson’s Progression Markers Initiative (PPMI) database
(www.ppmi-info.org/data). For up -to-date information on the study, visit www.ppmi-info.org. PPMI– a
public-private partnership– is funded by the Michael J. Fox Foundation for Parkinson’s Research and
funding partners, including AbbVie, Allergan, Avid Radiopharmaceuticals, Biogen, BioLegend, Bristol-
Myers Squibb, Denali, GE Healthcare, Genentech, GlaxoSmithKline (GSK), Eli Lilly and Company,
Lundbeck, Merck, Meso Scale Discovery (MSD), Pfizer, Piramal Imaging, Roche, Sanofi Genzyme,
Servier, Takeda, Teva, and UCB (www.ppmi-info.org/fundingpartners).

31

2.1.6 SUPPLEMENTARY APPENDIX

A. Participants, Clinical Assessments, and MRI Acquisition

Participants were enrolled at seven sites: Colorado (N=15), UCLA (N=25), Harbor-UCLA (N=43), UCSD
(N=16), Rochester (N=37), Stanford (N=8), and Pittsburgh (N=11). Inclusion criteria included nadir
CD4+ cell count (lowest cell count based on review of the participant’s medical charts) < 200 cells/mm
3
;
stable combined antiretroviral regimen (cART) with any FDA-approved therapy (≥ 12 consecutive weeks
prior to study); hemoglobin > 9.0 g/dl; serum creatinine ≤ 3× upper limit of normal (ULN); AST (SGOT),
ALT (SGPT), and alkaline phosphatase ≤ 3× ULN. Exclusion criteria included premorbid or comorbid
psychiatric disorders, confounding focal or diffuse neurologic disorders such as chronic seizures, stroke,
head trauma resulting in loss of consciousness of more than 30 min, multiple sclerosis, brain infection
other than HIV, or brain neoplasms, including CNS lymphoma; active alcohol and drug abuse or related
medical complications within 6 months of study; and diabetes mellitus with a fasting glucose > 140 mg/dl.
Longitudinal T1-weighted anatomical MRI scans were collected using either GE or Siemens 1.5
T scanners. GE spoiled gradient-recalled echo (SPGR) T1-weighted scans (N=132) were acquired on
either Signa Genesis, Signa HDx, or Signa Excite models, with a repetition time (TR) that ranged from
20 to 23 ms; echo time (TE) ranged from 3 to 9 ms; voxel dim = 0.9375 to 1.015625 mm, slice thickness
= 1.2 to 1.5 mm; flip angle = 30°; matrix size = 256 × 128. Siemens scans were performed on Sonata or
Symphony models. Siemens T1-weighted scans (N=23) were acquired using a pulse sequence and
parameters matched to the protocol used on GE scanners (TR=20-24 ms; TE=10.1 ms; slice thickness =
1.2-1.3 mm; flip angle = 30°; matrix size = 256 × 192). Inter-scanner reliability was shown to be 95% in
a prior HIVNC cross-sectional brain imaging study

(Hua et al. 2013a). While scanning parameters differed
slightly across subjects and sites, we ensured that scanning parameters between all intra-subject baseline
and follow-up scans were the same. Subjects with different intra-subject acquisition protocols were
excluded, and between-subject differences accounted for statistically.

32

B. Parkinson's Progression Markers Initiative (PPMI) MRI Acquisition

PPMI T1-weighted scans were acquired at 10 sites on 3 T Siemens Tim Trio or Verio scanners with an
MPRAGE sequence: TR = 2300 ms, TE = 2.98 ms, flip angle = 9°, voxel dim= 1 × 1 × 1 mm
3
, matrix
size=256 x 240).

C. Image Processing and Tensor-Based Morphometry (TBM)

All HIVNC and PPMI baseline and follow-up T1-weighted anatomic scans were denoised using the
nonlocal means filter (Coupe et al. 2008), N3 corrected for intensity inhomogeneity (Sled et al. 1998), and
further normalized by scan mean intensity. FreeSurfer was used to create baseline and follow-up brain-
tissue masks (Fischl et al. 2004). The intensities of all T1-weighted scans were then normalized to one
reference scan via histogram matching (Fedorov et al. 2012).
As in Jahanshad et al. (2015b), a 3-channel, study specific minimal deformation template (MDT)
was created using ANTs (Avants et al. 2011) from 52 HIV+ baseline T1-weighted images and two
additional FreeSurfer cortical, and subcortical segmentation (Reuter et al. 2012) channels. The MDT was
then resampled to the Colin27 template (Holmes et al. 1998) (220 × 220 × 220 matrix of 1 mm isotropic
voxels). Masked baseline T1 scans were rigidly aligned to the MDT with a 6 degree of freedom (6-dof)
registration to resample and align images to the same 3D coordinate space.
Follow-up scans were first linearly aligned (9-dof) to their respective baseline scan before brain
extraction to avoid scaling of brain tissue. To ensure both baseline and follow-up scans were each only
resampled once, the longitudinal 9-dof and cross-sectional 6-dof transformation matrices were
concatenated and applied to the original follow-up scan. Transformations were also applied to the brain
masks, and a common joint mask was created from the dilated union of the two mutually aligned masks.
The joint mask was used to uniformly remove all extra-cerebral tissue from the registered scans.
Each participant's follow-up T1-weighted scan was warped to its respective baseline scan using a
non-linear inverse-consistent elastic mutual information based registration algorithm (Leow et al. 2007).
To spatially normalize resulting Jacobian determinant maps, masked baseline T1-weighted scans were
linearly (9-dof) and elastically registered to the MDT, and the resulting cross-sectional 9-dof
transformation matrix and 3D deformation field were applied to each respective longitudinal change map.
The natural logarithm of each determinant Jacobian map was subsequently normalized by the respective
33

time interval (in years) between scans, to reflect the annualized rate of change. In these voxel-wise log-
determinant Jacobian maps (logJacs), a negative value indicates a loss of tissue, 0 indicates no change,
and a positive value represents tissue expansion.

D. Categorical Variables for Statistical Associations

For statistical analyses, race/ethnicity was coded using two dummy variables for the three racio-ethnic
groups: African Americans, Caucasians, and “Other” (including those of mixed races/ethnicities, Asians,
Native Alaskan and American Indians). Baseline ADC stage was coded as two dummy variables, where
ADC > 0 (NS) was coded as 1 and ADC=0 (NA) as 0 for V1, and an ADC > 0.5 was coded as 1 for V2
and ADC ≤ 0.5 as 0. Differences between the subset of NA-uVL participants and the remainder of the
population (NS or dVL) was coded with three dummy variables: where NS or dVL was coded as 1 for
V1, dVL was coded as 1 for V2, and NS was coded as 1 for V3. Longitudinal changes in ADC stage were
coded as two dummy variables: an increase in ADC (increase in NCI) was coded as 1 for V1 relative to
those who decreased or remained stable, and an ADC decrease was coded as 1 for V2 relative to those
that increased or remained stable.

E. PPMI and HIVNC Demographic Comparisons

Supplementary Table S1. Comparison of demographic characteristics between PPMI healthy controls and HIVNC
HIV+ participants both in the full cohorts and matched subsets using analysis of variance (ANOVA) or chi-squared
tests.

Matched Subset Full Cohort
PPMI
N=48
HIVNC
N=65
P-Value
PPMI
N=63
HIVNC
N=155
P-Value
Continuous Variables Mean (SD) ANOVA Mean (SD) ANOVA
Age (yrs) 55.6 (9.5) 54.2 (7.9) 0.395 59.3 (10.85) 48.0 (8.9) 1.16x10
-13
*
Inter-Scan Time Interval (yrs) 1.1 (0.2) 1.1 (0.5) 0.852 1.1 (0.19) 1.0 (0.5) 0.449
Categorical Variables % (N) χ
2
-Test % (N) χ
2
-Test
Sex (Male) 58.3% (28) 73.9% (48) 0.107 61.9% (39) 83.9% (130) 0.0004*
Education (≤ high school) 18.8% (9) 33.9% (22) 0.084 15.9% (10) 43.9% (68) 0.0001*
Race/Ethnicity (Caucasian) 89.6% (43) 80.0% (52) 0.127 92.1% (58) 69.9% (108) 0.0004*
* P ≤ 0.05

34

F. HIVNC Full Cohort and Matched Subset Characteristics

Supplementary Table S2. Comparison of demographic and clinical characteristics between the subset of HIVNC
cohort matched to PPMI controls (N=65) and the full HIVNC cohort (N=155; Table 1) using ANOVA or chi-
squared tests.
Continuous Variables Mean (SD) ANOVA P-Value
Age (yrs) 54.2 (7.9) 2.47x10
-6
*
HIV Duration (yrs) 11.7 (7.2) 0.937
cART Duration (yrs) 5.4 (4.9) 0.488
CPE Score 8.0 (2.9) 0.634
Nadir CD4+ (cells/mm
3
) 68.7 (64.2) 0.256
CD4+ (cells/mm
3
) 390.5 (189.0) 0.468
Inter-Scan Time Interval (yrs) 1.1 (0.5) 0.447
Categorical Variables % (N) χ
2
P-Value
Sex (male) 72.3% (47) 0.048*
Education (≤ high school) 32.9% (22) 0.167
Race/Ethnicity (Caucasian) 78.7% (51) 0.184
Suppressed CD4+ (≤ 350 cells/mm
3
) 46.2% (30) 0.315
Detectable Plasma HIV RNA (≥ 400 copies/mL) 29.2% (19) 0.499
a
ADC Stage 0 65.6% (42) 0.858
* P ≤ 0.05
a
N=64

35

G. HIVNC Annual Volume Change ROI Summaries

Supplementary Table S3. In the full HIV+ cohort and neuro-asymptomatic subgroups annual volume change
(in percent) was averaged in voxels within four tissue compartments. Key: NA: Neuro-asymptomatic (ADC=0);
uVL: Undetectable Viral Load (plasma HIV RNA < 400 cp/mL); SD: Standard Deviation; GM: Gray Matter;
WM: White Matter; CSF: Cerebrospinal Fluid; N: Nucleus.
Brain Region
% Annual Volume Change
ROI Mean (SD)
Full Cohort NA NA-uVL
Cortical GM
ALL -0.21 (0.38)* -0.28 (0.35)* -0.29 (0.37)*
Frontal -0.22 (0.42)* -0.29 (0.39)* -0.29 (0.42)*
Temporal -0.24 (0.47)* -0.33 (0.47)* -0.33 (0.48)*
Parietal -0.18 (0.43)* -0.24 (0.42)* -0.26 (0.41)*
Occipital -0.20 (0.45)* -0.28 (0.43)* -0.30 (0.40)*
Insula -0.08 (0.97)* -0.09 (0.87)* -0.13 (0.86)*
Cingulate -0.26 (0.93)* -0.32 (0.74)* -0.29 (0.74)*
WM
ALL -0.27 (0.77)* -0.35 (0.76)* -0.33 (0.74)*
Frontal -0.31 (0.91)* -0.39 (0.93)* -0.37 (0.90)*
Temporal -0.31 (0.90)* -0.41 (0.86)* -0.38 (0.80)*
Parietal -0.23 (0.69)* -0.31 (0.69)* -0.31 (0.67)*
Occipital -0.18 (0.88)* -0.30 (0.8)* -0.33 (0.79)*
Corpus Callosum -0.27 (1.05)* -0.29 (0.91)* -0.28 (0.85)*
Insula -0.32 (0.63) -0.37 (0.57) -0.35 (0.52)
Cingulate -0.29 (0.90)* -0.30 (0.79)* -0.28 (0.78)*
Subcortical GM
ALL -0.27 (0.57)* -0.31 (0.42)* -0.33 (0.36)*
Pallidum -0.49 (1.39)* -0.49 (1.28)* -0.43 (1.27)*
Putamen -0.37 (0.77)* -0.41 (0.61)* -0.39 (0.63)*
Thalamus -0.40 (0.93)* -0.45 (0.86)* -0.43 (0.72)*
N. Accumbens -0.31 (0.74)* -0.40 (0.76)* -0.40 (0.78)*
Amygdala -0.19 (0.57)* -0.20 (0.58)* -0.19 (0.57)*
Hippocampus -0.14 (0.92) -0.11 (0.72) -0.15 (0.69)
Caudate -0.001 (1.10) -0.12 (0.94) -0.25 (0.88)*
CSF
Insula 0.07 (1.45) 0.08 (1.38) 0.03 (1.37)
Ventricles 0.69 (2.39)* 0.73 (2.24)* 0.39 (2.05)
Full Brain -0.24 (0.49)* -0.32 (0.48)* -0.31 (0.48)*
FDR Critical P-Value 0.010 0.003 0.012
* Significant P-value ≤ FDR critical P-value indicated in the bottom row

36

H. Change in ADC vs Change in Markers of Infection Severity

Of 134 participants with ADC staging at both time points, 131 had baseline and follow-up CD4+ counts
and viral load available. CD4 status remained stable in 86.3% of participants, improved in 10.7%, and
declined in 3.1% (Table S4). Viral load status remained stable in 85.5% of participants, improved in 7.6%,
and declined in 6.9%. A Spearman's rank-order correlation between a decline, stable, or improved ADC
stage and a decline, stable, or improved CD4+ or viral load status showed no significant correlation
(CD4+: r = -0.15, P = 0.096; VL: r = 0.10, P = 0.614).

Supplementary Table S4. Total number of participants (out of 131) for each type of change in CD4+ or viral
load status between baseline and follow-up assessments.
Change
CD4+ Count
(cells/mm
3
)
N (%)
Viral Load
(copies/mL)
N (%)
Decline >350 to ≤350 4 (3.1%) <400 to ≥400 9 (6.9%)
Stable
>350 to >350 54 (41.2%) <400 to <400 94 (71.8%)
≤350 to ≤350 59 (45.0%) ≥400 to ≥400 18 (13.7%)
Improve ≤350 to >350 14 (10.7%) ≥400 to <400 10 (7.6%)

37

2.2 Consistent Subcortical Brain Volume Associations in 1000 HIV-infected
Individuals from Five Continents

This section is adapted from:

Nir TM, Fouche JP, Ananworanich J, Ances B, Boban J, Brew BJ, Chaganti J, Ching CRK, Cysique L,
Gupta V, Harezlak J, Heaps J, Hinken C , Hoar J, Joska J, Kallianpur K, Kuhn T, Lebrun-Frenay C,
Levine A, Mondot L, Nakamoto B, Navia B, Paul RH, Pennec X, Porges ES, Pruksakaew K, Shikuma
C, Thames A, Valcour VG, Vassallo M, Woods AJ, Thompson PM, Cohen RA, Stein DJ, Jahanshad N,
for the ENIGMA-HIV Working Group (2019). Smaller limbic brain volumes are associated with greater
immunosuppression in over 1000 HIV-infected adults across five continents: Findings from the
ENIGMA-HIV Working Group. In preparation for submission to Lancet HIV, April 2019.

38

Smaller Limbic Structures are Associated with Greater Immunosuppression in over 1000 HIV-
infected Adults across Five Continents: Findings from the ENIGMA-HIV Working Group

Talia M. Nir
1
, Jean-Paul Fouche
2
, Jintanat Ananworanich
3
, Beau Ances
4
, Jasmina Boban
5
, Bruce J. Brew
6
, Joga
Chaganti
6
, Christopher R.K. Ching
1
, Lucette Cysique
6
, Vikash Gupta
1
, Jaroslaw Harezlak
7
, Jodi Heaps
8
, Charles
Hinken
9
, Jacqueline Hoar
2
, John Joska
2
, Kalpana Kallianpur
10
, Taylor Kuhn
9
, Christine Lebrun-Frenay
11
,
Andrew Levine
12
, Lydiane Mondot
11
, Beau Nakamoto
10
, Bradford A. Navia
13
, Robert H. Paul
8
, Xavier Pennec
14
,
Eric S. Porges
15
, Kanchana Pruksakaew
3
, Cecilia Shikuma
10
, April Thames
9
, Victor G. Valcour
16
, Matteo
Vassallo
17
, Adam J. Woods
15
, Paul M. Thompson
1
, Ronald A. Cohen
15,18
, Dan J. Stein
2
, Neda Jahanshad
1
for the
ENIGMA-HIV Working Group

1
Imaging Genetics Center, Mark and Mary Stevens Neuroimaging and Informatics Institute, Keck School of Medicine, University of
Southern California, Marina del Rey, CA, USA
2
Department of Psychiatry and Mental Health, University of Cape Town, and MRC Unit on Anxiety & Stress Disorders
3
HIV-NAT, Thai Red Cross AIDS Research Centre, Bangkok, Thailand
4
Washington University in St Louis, St Louis, MO, USA
5
Faculty of Medicine, Diagnostic Imaging Center, University of Novi Sad, Novi Sad, Serbia
6
Departments of Neurology and HIV Medicine, St Vincent’s Hospital and University of New South Wales, Sydney, Australia
7
Indiana University School of Public Health, Bloomington, IN, USA
8
Missouri Institute of Mental Health, University of Missouri in Saint Louis, Saint Louis, MO, USA
9
Semel Institute for Neuroscience and Human Behavior, UCLA, CA, USA
10
Department of Medicine, John A. Burns School of Medicine, University of Hawaii at Manoa, HI, USA
11
Departments of Neurology and Radiology, University of Nice-Sophia Antipolis, Nice, France
12
Department of Neurology, David Geffen School of Medicine, UCLA, CA, USA
13
Department of Public Health, Infection Unit, Tufts University School of Medicine, Boston, MA, USA
14
University Côte d’Azur, Inria Sophia-Antipolis, France
15
Institute on Aging, Department of Aging and Geriatric Research, School of Medicine, University of Florida, USA
16
Memory and Aging Center, Department of Neurology, University of California, San Francisco, San Francisco, CA, USA
17
Department of Internal Medicine, Cannes General Hospital, France
18
Department of Psychiatry and Human Behavior, Warren Alpert Medical School, Brown University, Providence, RI, USA

Abstract. Human immunodeficiency virus (HIV) infection can be controlled with combination
antiretroviral therapy (cART), but neuropsychological impairments often persist. Identifying the
neuroanatomical pathways associated with infection has potential to delineate neuropathological
processes underlying these deficits. The ENIGMA-HIV Working Group was established to
harmonize data from diverse studies to identify the common effects of HIV-infection on brain
structure. Data were pooled from 12 independent neuro-HIV studies from Africa, Asia, Australia,
Europe, and North America. Volume estimates for eight subcortical brain regions were extracted
from T1-weighted MRI from 1,044 HIV+ adults (aged 22-81 years; 72.4% on cART; 70.3% male;
41.6% detectable viral load, dVL), to identify associations with plasma markers of current
immunosuppression (CD4+ T-cell count) or dVL. Follow-up analyses examined cART status and
sex. A Bonferroni correction threshold determined significance. Lower CD4+ count was
associated with smaller hippocampal (β = 20.3 mm
3
per 100 cells/mm
3
; p = 0.0001) and thalamic
volumes (β = 29.3 mm
3
per 100 cells/mm
3
; p = 0.003). A dVL was associated with smaller
hippocampal (Cohen’s d = 0.24; p = 0.0003) and amygdala volumes (d = 0.18; p = 0.0058). In the
subset of participants not on cART, CD4+ count was associated with putamen volume (β = 65.1
39

mm
3
per 100 cells/mm
3
; p = 0.0009). In HIV+ individuals across five continents, smaller limbic
volumes were consistently associated with current plasma markers. By assessing cohorts with
different inclusion/exclusion criteria and demographic distributions, these deficits may represent
a generalizable brain-signature of HIV-infection in the cART era. Our findings support the
importance of achieving viral suppression and immune restoration for maintaining brain health.

Keywords: MRI, Subcortical Volume, HIV, CD4+, Viral Load, Multi-site, ENIGMA, cART era

40

2.2.1 INTRODUCTION

In the era of globally accessible combination antiretroviral treatments (cART), morbidity and mortality
rates have dramatically decreased for individuals who have contracted the human immunodeficiency
virus (HIV). Infected individuals can live with the virus for decades and reach near-normal lifespans
(Rodger et al. 2013, Costagliola 2014). However, HIV-related comorbidities, including symptoms of
brain dysfunction, remain common among individuals on suppressive treatment.
The reported frequency of cognitive symptoms in HIV+ populations varies across studies: some
reports suggest that cognitive difficulties are present in nearly half of HIV+ individuals, but other studies
report less than 20% are affected(Nightingale et al. 2014). This variability may reflect heterogeneous
viral-host dynamics and/or inconsistencies in cognitive testing methods. Cognitive testing requires
significant quality control and access to demographically-appropriate norms. Given the challenges
associated with standardized cognitive testing (Pedraza and Mungas 2008, Fernández and Abe 2018),
there is a need to identify objective quantitative markers of central nervous system (CNS) impairment.
Systemic CD4+ T-cell count and viral load are two biomarkers universally used to monitor
immune function and treatment response. These are also the most consistently available clinical markers
in human studies of HIV, but the degree to which they capture CNS impairment is not fully understood.
Low nadir CD4+ has been identified as a predictor of neurocognitive impairment in the era of cART
(Ellis et al. 2011), suggesting that severe immunosuppression may lead to persistent and potentially
irreversible brain injury. However, nadir CD4+ is frequently self-reported, so it may be unreliable or
unknown. The duration of immunosuppression before recovery may also play a role, but this also may
not be well documented. Immune restoration and viral suppression are tracked through routine clinical
assessments, and may be achieved through modifiable factors including timely testing, treatment
initiation and compliance. The neurological implications of maintaining or achieving healthy targets, are
therefore important to establish.
Structural neuroimaging provides a promising array of quantitative biomarkers for assessing CNS
function and decline. To date, neuroimaging biomarkers have provided an important non-invasive, in
vivo approach to understand many psychiatric or neurodegenerative diseases and monitor efficacy of
clinical interventions (Shimizu et al. 2018). Many neuroimaging studies report structural abnormalities
in HIV+ individuals; and correspondence between structural changes and neurocognitive impairment
(Ances and Hammoud 2014, Chang and Shukla 2018). While neuroimaging studies tend to show
41

abnormal volumes of subcortical structures in HIV+ individuals on average, the correlations between
these volumes and current indices of HIV activity or immune status remain unclear.
Inconsistencies in the effect sizes and regional distribution of brain abnormalities associated with
CD4+ count or viral load in HIV+ individuals have limited the generalizability of the conclusions drawn
to date. For example, several studies report a positive association between subcortical volumes and
current CD4+ T-cell counts (Chiang et al. 2007, Cardenas et al. 2009, Heaps et al. 2015), while some
studies report no effects (Becker et al. 2011, Kallianpur et al. 2013, Ortega et al. 2013), and still others
report opposite effects (Jernigan et al. 2011). Demographic characteristics such as age and sex - or
socioeconomic and lifestyle attributes of study participants could, in part, explain differences in reported
findings. Moreover, comorbidities, chronicity of infection, and cART status, timing, duration, and
regimen can differ substantially across study samples. Differences in findings may also reflect
methodological heterogeneity, including differences in statistical power in studies with variable sample
sizes, MRI scanners and image acquisition protocols, and image processing or statistical analysis
techniques. By reducing methodological variability, boosting sample size, and assessing a diverse set of
HIV+ cohorts, a generalizable pattern of HIV-related brain effects may be easier to identify.
The HIV Working Group was established within the Enhancing Neuro Imaging Genetics through
Meta Analysis (ENIGMA) consortium to pool data from neuroimaging studies using harmonized data
analysis pipelines. The working group is a growing international collaboration open to all researchers
investigating the neurological consequences of HIV infection. This current benchmark analysis included
active involvement from investigators from 12 neuro-HIV studies from six different countries: the United
States, France, Serbia, Australia, Thailand, and South Africa (Table 1). The current study aimed to
investigate structural brain volume associations with the most commonly collected clinical assessments
of HIV burden. Here, we surveyed plasma CD4+ T-cell counts and the detectability of plasma RNA viral
load, and determined their relationship with MRI-derived subcortical brain volumes in HIV+ individuals.

42

Table 1. Demographic and clinical information by study and scanning site.
Site name
Total
N
Male
% (N)
Age [yrs]
Mean (SD),
Range
on cART
% (N)
Detectable
Viral Load
% (N)
CD4+ Count
cells/mm
3
Mean
(SD)
CD4+ Count
< 200 cells/mm
3

% (N)
Viral Load
> 400 copies/mL
% (N)
HIVNC Consortium (7 Sites),
United States
229 85.2 (195)
48.6 (8.3)
24-71
100 (229)
27.4 (62)
N=226
374.0 (229.6) 22.3 (51)
17.7 (40)
N=226
Site 1:
University of California,
San Diego
25 96.0 (24)
47.3 (4.6)
37-56
100 (25)
59.1 (13)
N=22
308.8 (306.7) 48.0 (12)
31.8 (7)
N=22
Site 2:
Harbor UCLA Medical
Center
54 79.6 (43)
46.6 (8.9)
24-70
100 (54) 14.8 (8) 350.6 (174.4) 27.8 (15) 9.3 (5)
Site 3:
Stanford University
10 90.0 (9)
46.4 (10.4)
31-62
100 (10) 0 (0) 313.5 (157.5) 20.0 (2) 0 (0)
Site 4:
Colorado
37 97.3 (36)
49.5 (7.8)
31-66
100 (37) 37.8 (14) 439.4 (241.8) 10.8 (4) 16.2 (6)
Site 5:
Pittsburgh
19 94.7 (18)
49.9 (9.7)
34-71
100 (19) 21.1 (4) 438.9 (298.2) 15.8 (3) 15.8 (3)
Site 6:
Rochester University
42 64.3 (27)
48.6 (7.8)
26-62
100 (42) 26.2 (11) 379.3 (206.0) 14.3 (6) 16.7 (7)
Site 7:
University of California,
Los Angeles
42 90.5 (38)
50.7 (8.3)
28-69
100 (42) 28.6 (12) 364.9 (224.3) 21.4 (9) 28.6 (12)
University of Hawaii,
United States
53 84.9 (45)
50.9 (8.0)
40-71
100 (53) 11.3 (6) 491.1 (208.5) 7.5 (4) 3.8 (2)
University of California,
San Francisco, United States
50 98.0 (49)
63.6 (2.5)
60-69
98 (49)
28.6 (14)
N=49
529.3 (218.5) 0 (0)
8.2 (4)
N=49
Brown University,
United States
79 60.8 (48)
45.2 (9.5)
23-65
83.5 (66) 27.8 (22) 476.2 (228.8) 8.9 (7) 26.6 (21)
University of California,
Los Angeles, United States
(Hinken)
12 100 (12)
46.2 (8.5)
26-57
75 (9) 50.0 (6) 604.1 (289.1) 8.3 (1) 33.3 (4)
University of California,
Los Angeles, United States
(Thames)
54 90.7 (49)
50.5 (12.9)
24-76
100 (54) 42.6 (23) 613.7 (284.5) 3.7 (2) 18.5 (10)
University of New South
Wales, Australia (Brew)
41 97.6 (40)
52.8 (8.1)
39-75
100 (41) 29.3 (12) 596.8 (271.5) 0 (0) 19.5 (8)
University of New South
Wales, Australia (Cysique)
68 100 (68)
55.3 (6.7)
44-69
100 (68) 0.01 (1) 549.46 (273.63) 7.4 (5) 0.01 (1)
SEARCH 011 Consortium,
Thailand
61 42.6 (26)
34.2 (7.0)
22-56
0 (0) 100 (61) 236.0 (139.0) 41.0 (25) 100 (61)
University of Cape Town,
South Africa
181 13.8 (25)
32.4 (5.0)
22-46
0 (0)
97.4 (148)
N=152
225.5 (149.2) 52.5 (95)
87.8 (129)
N=147
Nice University,
France
155 78.7 (122)
45.4 (10.0)
22-81
75.9 (126) 36.1 (56) 580.9 (277.6) 7.1 (11) 25.2 (39)
University of Novi Sad,
Serbia
61 90.2 (55)
44.3 (10.9)
25-66
100 (61) 16.39 (10) 616.28 (346.23) 1.5 (1) 16.4 (10)

43

2.2.2 METHODS

Participants and Clinical Assessments

T1-weighted magnetic resonance imaging (MRI) brain scans and clinical data from 1,044 HIV+ adult
participants (aged 22-81 years; 70.3% male) were collected at 18 sites as part of 12 independent studies.
The participating studies include the multisite HIVNC Consortium (N=229 scanned across 7 sites),
Brown University (N=79), University of Hawaii (N=53), University of California San Francisco (N=50)
and two groups from the University of California Los Angeles (Hinken N=12; Thames N=54) all in the
United States; Nice University Hospital in France (N=155); the University of Cape Town in South Africa
(N=181); two groups from the University of South Wales in Australia (Brew N=41; Cysique N=68); a
group from Serbia (N=61) and the SEARCH-011 study from Thailand (N=61). Clinical assessments at
the time of scan included current CD4+ T-cell counts (cells/mm
3
) and HIV plasma RNA viral load
(copies/mL). Viral load detection assay thresholds varied across sites, and included a range of thresholds
(for site-specific details, please see Supplementary Table S1). An individual was categorized as having
a detectable viral load according to the assay threshold at the collection site.
Participant demographic and clinical characteristics for each of the twelve studies are reported in
Table 1. Two transgender participants were excluded. Each study obtained approval from their local
ethics committee or institutional review board; participants signed an informed consent form at each
participating site.

Image Acquisition, Processing, and Quality Assurance

T1-weighted MRI brain images were acquired at each site, on a variety of 3 or 1.5 tesla scanner platforms.
Acquisition protocols are further detailed in Supplementary Table S2. Subcortical regions were
parcellated and quality control completed using publically available ENIGMA protocols
(http://enigma.usc.edu/protocols/). Intracranial volume estimation and regional segmentations were
conducted using FreeSurfer version 5.3 (Fischl 2012). Volumes were extracted from eight regions of
interest (ROIs): thalamus, caudate, putamen, pallidum, hippocampus, amygdala, nucleus accumbens, and
lateral ventricles. The average of the left and right volumes for each ROI was evaluated. Specific regions
for an individual were excluded from the analysis if the overall segmentation did not pass visual quality
44

assurance as detailed in ENIGMA’s harmonized quality control protocols. Histograms were also created
from each site’s data to investigate normality of the data distribution for each subcortical structure, and
statistical outliers were identified. If the mean of an individual’s subcortical volume was more than 3
standard deviations from the mean for the site, it was flagged for a more extensive quality control and
possible removal of the scan from the analysis. The 1,044 scans included in the overall study represent
only the scans for which all segmentations were of sufficient quality.

Statistical Analyses

Feasibility study of age effects: To ensure sufficient power to detect associations between brain measures
and variables of interest from data pooled across studies, a preliminary analysis tested for associations
between age and eight regional brain volumes. Random effects multiple linear regressions were
performed. Three fixed-effects covariates were included in the model: sex, the interaction between age
and sex, and total intracranial volume (ICV) to adjust for variability in head size. To account for scanner
effects, the data collection site was used as the random-effects grouping variable; there were 18 sites in
all. Statistical analyses were conducted with the ‘nlme’ package in R (version 3.2.3). Effect sizes were
estimated using the r-value after accounting for all covariates. Bonferroni correction was used to account
for the eight tests across brain regions and to control the family-wise error rate at 5%; an effect was
determined to be statistically significant if the corresponding p-value was ≤ 0.05/8 = 0.0063.

Associations between brain volumes and HIV plasma markers: Random effects multiple linear
regressions were performed to investigate the association between each of the eight regional brain
volumes and 1) current CD4+ T-cell count (cells/mm
3
) or 2) a binary variable indicating a detectable (1)
or undetectable (0) viral load. Site was used as the random-effects grouping variable and four fixed-
effects covariates were included in the model: age, sex, the interaction between age and sex, and ICV.
Effect sizes, after accounting for all covariates, were estimated using the r-value for continuous variables
and Cohen’s d for dichotomous factors. Significance was determined using the Bonferroni correction
threshold (p ≤ 0.0063).

Stratification by cART status and by sex: Post-hoc analyses tested for associations between brain volumes
and HIV plasma markers when the data were stratified by 1) cART status at the time of scan (cART+,
45

cART-), and 2) by sex. The same analytic framework was used, removing sex and sex-by-age interactions
from statistical models when stratifying by sex. Group differences in demographic and clinical factors
were evaluated with chi-squared tests for dichotomous factors, and two-tailed T-tests for continuous
measures.

Validation analyses comparing harmonized thresholds, and meta-analyses: Three sets of validation
analyses were performed: 1) dichotomizing CD4+ count based on the AIDS-defining threshold of 200
cells/mm
3
; 2) defining a common dVL threshold across sites (400 copies/mL, which was the highest
detection limit for any site); 3) comparing primary pooled statistical analyses to the meta-analytic
framework used in other ENIGMA studies (Hibar et al. 2016, Schmaal et al. 2016, van Erp et al. 2016),
where associations were conducted with data from each site separately, then aggregated using an inverse-
variance weighted fixed-effects meta-analysis from the ‘metafor’ package in R. Forest plots were created
to compare individual site effects and the aggregated effects.

2.2.3 RESULTS

Demographic and clinical characteristics of the entire ENIGMA-HIV cohort are summarized in Table 2.
Stratification by cART status revealed that compared to cART+ participants (N=756), cART- participants
(N=288) were significantly younger, had significantly lower CD4+ counts, and a greater percentage had
detectable viral loads. Similarly, compared to males (N=734), females (N=310) were significantly
younger, had significantly lower CD4+ counts, and a greater percentage had detectable viral loads.
Proportionally fewer females were on cART at the time of scan. As expected, on average females had
smaller regional brain volumes than males when unadjusted for ICV.

46

Table 2. Summary of demographic, clinical, and neuroanatomical information aggregated across all twelve
participating studies of HIV+ adults, and stratified by cART status and by sex. Subcortical volume comparisons
in this table are absolute and were not corrected for ICV, as was done in the formal analyses.

Total cART+ cART- cART+ vs
cART-
*P-Value
Male Female Male vs
Female
*P-Value
Total N 1,044 756 (72.41%) 288 (27.59%) 734 (70.3%) 310 (29.7%)
Mean Age [yrs]
(SD), Range
45.46 (11.60),
22-81
49.71 (9.91)
23-81
34.29 (7.62)
22-66
4.93x10
-108

48.44 (10.80),
22-81
38.40 (10.30),
22-70
1.16x10
-39

% on
cART (N)
72.41% (756) 100% (756) -- -- 87.87% (645) 35.81% (111) 2.71x10
-66

% with Detectable
Viral Load (N)
41.64% (421)
N=1011
22.21% (167)
N=752
98.07% (254)
N=259
3.20x10
-101

31.68% (230)
N=726
67.02% (191)
N=285
2.28x10
-25

% with Viral Load
> 400 copies/mL (N)
32.70% (329)
N=1006
12.63% (95)
N=752
92.13% (234)
N=254
1.40x10
-120

21.55% (156)
N=724
61.35% (173)
N=282
1.25x10
-33

Mean CD4+ Count
cells/mm
3
(SD)
441.40 (275.82) 501.30 (276.05) 284.26 (204.30) 1.01x10
-38
483.77 (278.40) 341.07 (242.00) 5.19x10
-16

% with CD4+ Count
< 200 cells/mm
3
(N)
19.35% (202) 10.71% (81) 42.01% (121) 2.57x10
-30
13.76% (101) 32.58% (101) 2.01x10
-12

Mean Thalamus
Volume mm
3
(SD)
7043.5 (979.5) 7189.2 (978.6) 6661.1 (874.4) 2.73x10
-16
7277.7 (963.4) 6489.0 (773.2) 2.44x10
-39

Mean Caudate
Volume mm
3
(SD)
3587.3 (498.0) 3616.1 (502.8) 3511.8 (477.8) 0.0020 3652.4 (500.1) 3433.2 (458.3) 1.57x10
-11

Mean Putamen
Volume mm
3
(SD)
5206.6 (740.0) 5183.3 (779.3) 5267.8 (622.2) 0.069 5270.0 (777.5)
5056.51
(617.99)
2.98x10
-6

Mean Pallidum
Volume mm
3
(SD)
1517.5 (251.7) 1478.6 (249.4) 1619.6 (228.4) 4.14x10
-17
1517.9 (251.8) 1516.6 (251.9) 0.94
Mean Hippocampus
Volume mm
3
(SD)
4002.1 (544.9) 4109.7 (498.0) 3719.6 (562.2) 1.11x10
-22
4152.0 (507.5) 3647.1 (459.8) 2.97x10
-47

Mean Amygdala
Volume mm
3
(SD)
152.8 (299.8) 1620.1 (299.7) 1376.2 (217.3) 6.12x10
-42
1640.9 (287.3) 1344.2 (213.5) 4.37x10
-63

Mean Accumbens
Volume mm
3
(SD)
541.0 (128.2) 527.4 (135.5) 576.8 (98.3) 1.66x10
-10
539.8 (135.8) 543.9 (108.1) 0.60
Mean Lateral
Ventricles Volume
mm
3
(SD)
19844.6
(14156.4)
21935.1
(14969.2)
14357.0
(9853.4)
2.07x10
-20

22204.9
(15009.2)
14256.2
(9873.2)
1.12x10
-22

*Differences in dichotomous factors were assessed using chi-squared tests; for continuous measures, differences
were assessed using two-tailed T-tests.

Feasibility Study of Age Effects

Across the entire sample (N=1,044), older age was associated with smaller volumes of all subcortical
structures and larger ventricular volumes (r-value range = 0.13-0.29), consistent with studies in
seronegative adults (Potvin et al. 2016) (Supplementary Table S3). Sex was a significant predictor in
models for associations between age and the thalamus, putamen, pallidum, hippocampus, and amygdala
(p ≤ 0.0063). Age effects in males mirrored that of the full group. Females showed significant negative
47

associations between age and volumes of the thalamus, putamen, pallidum, nucleus accumbens, and
lateral ventricles. No significant age by sex interaction was detected.
To further validate findings, associations were also separately conducted within each study, and
statistical summaries aggregated in a meta-analytic framework; findings were consistent regardless of
study design (Supplementary Table S4). Forest plots reflecting effect sizes from each site individually
relative to pooled (mega-) and meta-analysis effect sizes are shown in Supplementary Figures S1-S4.

Associations between Brain Volumes and HIV Plasma Measures

Across all participants (N=1,044), lower CD4+ counts were significantly associated with smaller
hippocampal volumes (r = 0.12; p = 0.0001; Table 3) and smaller thalamic volumes (r = 0.09; p = 0.003).
Detectable viral load (dVL; N=1,011) was significantly associated with smaller hippocampal (Cohen’s d
= -0.24; p = 0.0003; Table 4) and amygdalar volumes (d = -0.18; p = 0.0058). Associations between
CD4+ or dVL and brain volumes remained consistent after adjusting for cART status in the statistical
models. To ensure both CD4+ and dVL were independently associated with hippocampal volume, we
included both predictors in the same model and found both predictors remained significant after
adjustment for the other (dVL: d = -0.20, p = 0.0023; CD4+: r = 0.11; p = 0.0008).
CD4+ count associations were meta-analyzed to validate findings from pooled analyses; again,
findings were consistent regardless of study design (see Supplementary Table S5 for meta-analytic
results and Supplementary Figures S5-S8 for Forest plots). The distribution of detectable viral loads
was largely uneven across sites (see Table 1), preventing a comparable meta-analysis for dVL.
Validation analyses evaluating a common dVL threshold (400 copies/mL) across sites also
showed significant associations with hippocampal volume (d = -0.23; p = 0.0007; Supplementary Table
S6). Suggestive associations were detected between brain volumes and an AIDS-defining
immunosuppression status (CD4+ ≤ 200 cells/mm
3
), but no associations were significant after multiple
comparisons correction (Supplementary Table S7).

Stratification by cART Status and by Sex

Table 3 (CD4+) and Table 4 (dVL) detail regional findings within stratified groups. In the subset of
cART+ participants, smaller hippocampal volumes were significantly associated with lower CD4+ count
48

and dVL (CD4+: r = 0.12; p = 0.0008; dVL: d = -0.31, p = 0.0005). cART+ participants also showed
significant amygdalar volume associations with dVL (d = -0.30; p = 0.0005). In cART- participants,
smaller putamen volumes were significantly associated with lower CD4+ counts (r = 0.20; p = 0.0009);
we did not assess dVL as only 5 cART- participants had an undetectable VL. Associations in males
mirrored those found in the cART+ subgroup: hippocampal volumes were associated with both CD4+
and dVL (CD4+: r = 0.13; p = 0.0005; dVL: d = -0.29; p = 0.0004), and amygdalar volumes were
associated with dVL (d = -0.25; p = 0.0024). We found no statistically significant associations between
plasma measures and any brain volume in females; however we found a similar pattern in the ranking of
regional effect sizes between males and females (CD4+: Pearson’s r = 0.91; p = 0.002; dVL: r = 0.79; p
= 0.02; Supplementary Figure S9).

49

Table 3. R-values, beta-values (unstandardized regression slopes reflecting change in volume (mm
3
) for every 100 cells/mm
3
change in CD4+ count),
standard errors (SE), and p-values from associations between regional brain volumes and CD4+ count at the time of scan across all 1,044 HIV+
participants, and separately in the subset of 756 cART+ participants, 288 cART- participants, 734 males, and 310 females.
ROI
Total (n=1,044) cART+ (n=756) cART- (n=288) Male (n=734) Female (n=310)
r β SE p r β SE p r β SE p r β SE p r β SE p
Thalamus 0.093 29.32 9.86 0.0030** 0.10 25.04 9.57 0.0091* 0.14 59.52 24.61 0.016* 0.087 24.42 10.46 0.020* 0.12 43.26 21.39 0.043*
Caudate 0.026 4.60 5.47 0.40 0.020 3.25 5.63 0.56 0.023 5.81 15.16 0.70 0.030 4.75 5.90 0.42 0.007 1.36 12.00 0.91
Putamen 0.073 18.02 7.75 0.020* 0.050 10.71 8.36 0.20 0.20 65.10 19.39 0.0009** 0.066 15.69 8.87 0.077 0.11 28.45 15.26 0.063
Pallidum 0.056 4.95 2.74 0.071 0.050 4.05 2.95 0.17 0.11 12.64 7.06 0.075 0.058 4.71 3.05 0.12 0.050 5.00 5.81 0.39
Hippocampus 0.12 20.30 5.23 0.0001** 0.12 18.33 5.44 0.0008** 0.12 30.62 14.85 0.040* 0.13 20.19 5.78 0.0005** 0.11 20.13 11.05 0.069
Amygdala 0.075 5.91 2.47 0.017* 0.080 5.50 2.62 0.036* 0.040 4.20 6.52 0.52 0.066 4.92 2.77 0.076 0.10 9.10 5.19 0.081
Accumbens 0.050 1.94 1.21 0.11 0.040 1.37 1.29 0.29 0.11 6.21 3.26 0.056 0.041 1.52 1.40 0.28 0.094 4.01 2.50 0.11
Lateral
Ventricles
-0.070 -349.88 155.17 0.024* -0.070 -330.04 172.84 0.057 -0.14 -773.63 330.80 0.020* -0.072 -352.30 182.11 0.053 -0.12 -538.46 260.63 0.039*
**Significant at Bonferroni corrected threshold for tests in 8 regions of interest, p ≤ 0.0063.
*Suggestive at p ≤ 0.05.

50

Table 4. Cohen’s d effect sizes, standard errors (SE), and p-values for associations between regional brain volumes and detectable VL at the time
of scan across all 1,011 HIV+ participants with VL information, and separately in the subset of 752 cART+ participants, 726 males, and 285 females.
We did not assess VL in those off treatment due to the limited number of individuals in this subgroup with undetectable VL (n=5).
ROI
Total (n=1011) cART+ (n=752) cART- (n=259) Male (n=726) Female (n=285)
Cohen’s d SE p Cohen’s d SE p Cohen’s d SE p Cohen’s d SE p Cohen’s d SE p
Thalamus -0.08 0.060 0.21 -0.14 0.090 0.12 - - - -0.10 0.080 0.20 -0.080 0.13 0.56
Caudate -0.04 0.060 0.59 -0.02 0.090 0.83 - - - -0.020 0.080 0.81 -0.16 0.13 0.23
Putamen 0.03 0.060 0.60 0.02 0.090 0.83 - - - 0.030 0.080 0.71 -0.080 0.13 0.52
Pallidum 0.03 0.060 0.65 -0.01 0.090 0.94 - - - 0.030 0.080 0.72 -0.030 0.13 0.85
Hippocampus -0.24 0.060 0.0003** -0.31 0.090 0.0005** - - - -0.29 0.080 0.0004** -0.24 0.13 0.063
Amygdala -0.18 0.060 0.0058** -0.30 0.090 0.0008** - - - -0.25 0.080 0.0024** -0.13 0.13 0.31
Accumbens -0.05 0.060 0.42 -0.11 0.090 0.23 - - - -0.070 0.080 0.40 -0.060 0.13 0.62
Lateral Ventricles 0.08 0.060 0.20 0.12 0.090 0.19 - - - 0.12 0.080 0.13 -0.010 0.13 0.96
**Significant at Bonferroni corrected threshold for tests in 8 regions of interest, p ≤ 0.0063.
*Suggestive at p ≤ 0.05.

51

2.2.4 DISCUSSION

In one of the largest coordinated brain imaging studies of HIV-infected individuals, we pooled and
harmonized data from 1,044 individuals across six countries, and found associations between brain
volumes and plasma markers used to monitor HIV infection. In the full group, brain volume associations
specifically implicated the limbic system: lower CD4+ T-cell counts were associated with smaller
hippocampal and thalamic volumes; a detectable viral load was associated with smaller hippocampal and
amygdalar volumes. The limbic effects were largely driven by cART+ participants, the majority of whom
were male. In the subset of cART- participants, however, the putamen was the only region associated
with CD4+.
Plasma CD4+ T-cell count and viral load are routinely assessed in HIV clinics around the world.
These peripheral markers have been correlated with neuropsychological performance in HIV+
individuals (Marcotte et al. 2003, Heaton et al. 2010), as well as post mortem brain tissue HIV viral load
and pathology (Everall et al. 2009, Gelman et al. 2013). Unfortunately, the relationship between plasma
markers and MRI brain volume variation, assessed in vivo, has been inconsistent across studies.
Differences between studies, including clinical and demographic differences, methodological variability,
and/or insufficient power to estimate robust effect sizes, continue to complicate our understanding of the
neurological consequences of infection.
Brain signatures that generalize beyond individual studies are important to identify, to establish
neuroimaging biomarkers of HIV brain disruption. Literature-based meta-analyses assess consistency of
published findings, and offer initial insights into the generalizability of HIV-related findings. A recent
such meta-analysis of 19 published studies, spanning almost three decades, found significantly lower
total brain volumes and total gray matter volumes in HIV+ individuals compared to seronegative controls
(O'Connor et al. 2018); however, no significant serostatus association with basal ganglia volume was
found. In addition, effect sizes were found to be lower in more recent studies compared to older ones
(O'Connor et al. 2018), which may suggest a diverging profile of HIV-affected neurocircuitry in more
recent years.
The scope and extent of HIV-related neuropsychological impairments have evolved from the pre-
cART era. For example, while the prevalence of neurocognitive symptoms may be similar, reports have
highlighted an improvement in attention, verbal fluency, visuoconstructional deficits, but a deterioration
in learning efficiency and complex attention with treatment, even in patients with an undetectable plasma
52

viral load (Cysique et al. 2004, 2006). Brain deficits may persist despite cART due to factors such as
cART neurotoxicity (Robertson et al. 2012), an irreversible history of advanced disease (Ellis et al. 2011),
reservoirs of ongoing low-grade viral replication and persistent immune activation in the CNS (Lamers
et al. 2016, Martinez-Picado and Deeks 2016), or neurodegenerative processes that can occur with aging
(Brew et al. 2009).
In the pre-cART era, HIV-dementia was characterized as a “subcortical dementia”. In line with
reported post mortem neuropathological and viral protein distributions (Kure et al. 1990, Brew et al.
1995, Berger and Nath 1997, Morgello 2018), neuroimaging studies of HIV-infected individuals often
showed smaller basal ganglia volumes (Aylward et al. 1993). While often inclusive of the basal ganglia
regions, neuroimaging findings have extended beyond volume differences of the putamen, pallidum,
nucleus accumbens and caudate; serostatus related differences have been detected to varying degrees in
the thalamus (Pfefferbaum et al. 2012), amygdala (Ances et al. 2012) hippocampus (Pfefferbaum et al.
2014), and even cortical gray matter (Sanford et al. 2018), suggesting recent brain structural differences
are more dispersed. Once a staple of HIV-related brain alterations, basal ganglia volume associations
with CD4+ count or dVL were not detected in the full group of HIV+ participants assessed here, but only
in the cART- subset. cART- participants were not necessarily cART naive, but on average, their plasma
markers may be more in line with those recorded before cART initiation. While current CD4+ counts
were associated with limbic and not basal ganglia volumes in the full set of participants, it is possible
that their nadir CD4+ counts, reflecting a history of severe immunosuppression prior to treatment
initiation, would be associated with basal ganglia volumes. The shift in HIV-infection from a fatal to
chronic condition in the cART era appears to be accompanied by a shift in the profile of HIV-related
brain abnormalities, beyond the basal ganglia, frequently implicated in the pre-cART era, to limbic
structures; this shift in subcortical signatures may be contributing to the growing range of
neuropsychiatric and cognitive outcomes.
In conjunction with CD4 receptors, CXCR4 and CCR5 chemokine co-receptors mediate HIV cell
entry; neuroanatomical regions with greater receptor expression levels may be more susceptible to HIV-
related pathology. Both CCR5 and CXCR4 are expressed in astrocytes and microglia; however, unlike
CCR5, CXCR4 has also been found expressed in neurons. In immunohistochemical studies, the most
prominent regions of neuronal and glial CXCR4 chemokine receptor expression include not only the
basal ganglia, but also limbic structures: the hippocampus, amygdala, and thalamus (van der Meer et al.
2000). CCR5 co-receptors, generally expressed by memory CD4+ T-cells, are used by macrophage-tropic
53

HIV strains predominant in early infection; CXCR4 receptors, expressed by naïve T-cells, are used by
T-tropic HIV strains that appear later during infection. While memory cells divide more frequently than
naïve cells during early infection, the change in viral tropism may be driven by naïve cell division
approaching that of memory cells at low CD4+ counts associated with disease progression (Ribeiro et al.
2006). Limbic volume associations with lower CD4+ count and detectable viral loads may reflect this
phenotypic shift with advancing infection.
Males in our study may be more representative of the cART era: 90% of them were on cART
compared to only 35% of females, most of whom were recruited from Thailand and South Africa –
countries from which a substantial proportion of individuals eligible for treatment, may not receive it.
We found a similar pattern in the ranking of regional CD4+ count effect sizes between sexes, but regional
effects were less consistent for dVL; females showed larger effect sizes in the caudate and putamen,
while males showed larger effects in the amygdala and ventricles, further supporting a diverging cART
era profile. Despite being the largest neuroimaging evaluation of HIV+ adult women, we did not detect
any significant associations between plasma markers and brain volumes in females alone. This apparent
difference in power may be due to any number of confounding factors. For example, comorbidities,
including mental health conditions, may play a confounding role. Female-specific findings were noted in
an international study of post-traumatic stress disorder (PTSD) (Logue et al. 2018), where women with
PTSD were found to be driving case versus control differences in hippocampal volumes. Trauma,
particularly related to intimate partner violence, is overwhelmingly common among HIV-infected
women (Machtinger et al. 2012), so it is also possible that factors such as comorbid PTSD, may be
confounding limbic associations in women. Women constitute 52% of all individuals aged 15 years and
older living with HIV (AMFAR 2018). Nevertheless, women are under-represented in neuroHIV
research, impeding the reliability and generalizability of findings. Despite evaluating over 300 HIV-
infected women, twice as many participants included in our study were male; only two of the 12 studies
had recruited more than 40% women. A more extensive international effort assessing the neurological
effects of HIV-infection in women is needed.
Effects in limbic structures as seen here, are non-specific and have been detected in a wide-range
of such conditions studied in similar large-scale international efforts from the ENIGMA consortium
(Bearden and Thompson 2017). Serious mental illnesses (SMIs), such as depression or substance-use
disorder, have a high prevalence in HIV+ individuals and may increase the risk of HIV transmission
(Nurutdinova et al. 2012). As we assessed individuals with and without these comorbidities, limbic
54

associations detected in this study are not likely simply a reflection of such comorbid neuropsychiatric
conditions. However, viral-induced immunosuppression may also contribute to the risk of developing
SMIs by targeting the same neurocircuitry implicated across these disorders.
The hippocampus, in particular, showed the largest effect size with both CD4+ and viral load
measures. In post mortem studies, hippocampal tissue shows some of the highest viral concentrations
(Wiley et al. 1998). Hippocampal neurons also show increased gliosis and HIV chemokine co-receptors
and expression (Petito et al. 2001), and may be particularly susceptible to Tat induced apoptosis (Kruman
et al. 1998, Moore et al. 2006). Recent pathological studies from the cART-era suggest a potential shift
in HIV-related inflammation to the hippocampus and surrounding entorhinal cortex (Carnie et al. 2005,
Anthony and Bell 2008, Tate et al. 2010). Hippocampal atrophy is consistently reported across aging
populations, and accelerated atrophy is a hallmark of neurodegenerative diseases such as Alzheimer’s
disease (AD). Neuropathological hallmarks of aging and AD, including elevated levels of phosphorylated
tau and beta-amyloid deposits, have been detected in the hippocampus of cART treated HIV+ individuals
(Green et al. 2005, Anthony et al. 2006). Common age and HIV-related pathological processes, such as
inflammation and blood brain barrier impairment, may accelerate age-related neurodegenerative
processes (Cysique and Brew 2014, Cohen et al. 2015). Study participants ranged in age from 22 to 85
years, and while age was also significantly associated with reduced volumes in limbic structures, we did
not detect any interaction between age and CD4+ count or dVL. A better understanding of chronic
infection in the context of aging remains an important topic of research.
Literature based meta-analyses provide substantial insights into the reproducibility and
consistency of published findings. However, they are inherently limited by the fact that effects are likely
over-estimated due to publication biases substantially limiting the inclusion of studies with null findings.
Effect sizes reported in this study are considered small to moderate, yet given the large, diverse sample,
small effect sizes may not be clinically insignificant. Furthermore, methodological factors, including
image analysis methods and statistical design cannot be harmonized in such retrospective analyses. Here,
we partially address these limitations, by harmonizing the image processing and statistical analyses.
However, several limitations and challenges remain. This study from the ENIGMA-HIV working group
was comprised largely of studies that collected MRI data only from HIV-infected individuals, barring a
direct comparison to seronegative controls. Such serostatus comparisons may help elucidate the full
extent of HIV-related brain deficits, as opposed to highlighting regions more selectively affected by the
degree of immunosuppression and viral control. Plasma CD4+ count and VL were readily available
55

across studies, but these plasma markers are not comprehensive assessments of the full systemic impact
of HIV. Future efforts are needed to pool and harmonize additional immunological, cerebrovascular,
metabolic and inflammatory markers associated with infection. However, identifying clinical factors that
can be uniformly collected or interpreted across international studies remains challenging. For example,
differences in treatment regimens with varying CNS penetration effectiveness (CPE) scores, duration of
treatment, and standards of adherence that qualified a participant as cART+ at the time of scan, may vary
from study to study. Despite such variations, our study identified consistent and robust brain volume
associations with HIV VL and immunosuppression in a large and diverse study sample of HIV-infected
individuals from around the world.
This analysis demonstrates the feasibility and utility of a global collaborative initiative to
understand the neurological signatures of HIV infection. We invite other neuro-HIV researchers to join
the ENIGMA-HIV consortium; with a greater collaborative effort, we will be able to assess factors that
may modulate neurological outcomes —including cART treatment regimens, comorbidities, co-
infections, substance use, socioeconomic, and demographic factors— as well as the functional
implications of such structural brain differences in well-powered analyses. Understanding the
neurobiological changes that may contribute to neuropsychiatric and cognitive outcomes in HIV+
individuals is critical for identifying individuals at risk for neurological symptoms, driving novel
treatments that may protect the CNS, and monitoring treatment response.

2.2.5 ACKNOWLEDGEMENTS

Funding was provided in part by NIH grant U54 EB020403 from the Big Data to Knowledge Program,
to support ENIGMA’s big data analytics, and by P41 EB015922.

56

2.2.6 SUPPLEMENTARY APPENDIX

A. Detectable Viral Load Thresholds
Supplementary Table S1. Detectable plasma RNA viral load threshold by site.
Site
Detectable Viral
Load Threshold
(copies/mL)
Site
Detectable Viral Load
Threshold (copies/mL)
HIVNC Consortium
(7 Sites) , United States
50-400
Brown University,
United States
140
Site 1:
University of
California,
San Diego
50
University of California,
Los Angeles, United States
(Hinken)
48
Site 2:
Harbor UCLA Medical
Center
75
University of California,
Los Angeles, United States
(Thames)
19
Site 3:
Stanford University
50
University of New South
Wales, Australia (Brew)
20
Site 4:
Colorado
50
University of New South
Wales, Australia (Cysique)
49
Site 5:
Pittsburgh
50
SEARCH 011 Consortium,
Thailand
Unknown threshold;
VL always > 400
Site 6:
Rochester University
50
University of Cape Town,
South Africa
50
Site 7:
University of California,
Los Angeles
400
Nice University,
France
40
University of Hawaii,
United States
48
University of Novi Sad,
Serbia
2
University of California,
San Francisco, United States
50

57

B. MRI Protocols

Supplementary Table S2. T1-weighted MRI acquisition parameters
Site Scanner Acquisition parameters
HIVNC Consortium, United
States
1.5T GE Signa,
Siemens
Symphony/Sonata
GE SPGR; TR=2-23 ms; TE=3-9 ms; flip angle=30°; matrix
size = 256 × 128; voxel size=1x1x1.2-1.5 mm
Siemens: TR=20-24 ms; TE=10.1 ms; flip angle = 30°; matrix
size = 256 × 192; voxel size=1x1x1.2-1.3 mm
University of Hawaii,
United States
3T Phillips
Achieva
3D TFE; TR=6.9 ms; TE = 3.2 ms; flip angle = 8°; FOV =
256mm; voxel size=1x1x1.2 mm

University of California,
San Francisco, United States
3T Siemens
TrioTim
MP-RAGE; TR=2300 ms; TE=2.98 ms; flip angle=9°; FOV =
256mm; 160 slices; 240×256 matrix; voxel size=1x1x1 mm

Brown University,
United States
3T Siemens
TrioTim
MP-RAGE; TR=2250 ms; TE=3.06 ms; flip angle=9°; FOV =
220 mm; matrix = 256×256, slice thickness = 0.86mm
University of California,
Los Angeles, United States
(Hinken)
3T Siemens
TrioTim
MP-RAGE; TR=2200 ms; TE=2.2 ms; matrix size = 256 ×
256; FOV 240 mm; 176 slices; slice thickness = 1 mm

University of California,
Los Angeles, United States
(Thames)
3T Siemens
TrioTim
MP-RAGE; TR = 450.0 ms; TE = 10.0 ms; flip angle: 8°;
FOV 256 mm; matrix =256x219; voxel size =1.0 × 0.94 ×
0.94 mm.
University of New South Wales,
Australia (Brew)
3T Philips
Achieva
3D TFE; TR=5.3 ms; TE=2.4ms; flip angle = 18°, 256x256
matrix; 180 slices; voxel size=1x1x1 mm
University of New South Wales,
Australia (Cysique)
3T Philips
Achieva
3D TFE; TR= 6.39ms; TE=2.9 ms; flip angle: 8°; FOV 256
mm; 190 slices, voxel size=1x1x1 mm
SEARCH 011 Consortium,
Thailand
3T Siemens
Allegra
MP-RAGE; TR=2400 ms; TE=2.38 ms; flip angle=8°; 162
slices; voxel size=1x1x1 mm
University of Cape Town,
South Africa
3T Siemens MP-RAGE; TR=2400 ms; TE=2.38 ms; TI=1000 ms; flip
angle=8°; 162 slices; voxel size=1x1x1 mm
Nice University,
France
1.5T GE
Signa
SPGR; TR =12.4 ms, TE = 5.2 ms, flip angle = 18°; FOV =
240 mm; 256x256 matrix; voxel size=0.6 x 0.6 x 0.6 mm
University of Novi Sad,
Serbia
3T Siemens
TrioTim
MP-RAGE; TR = 2300 ms; TE=2.97 ms; FOV 256 mm;
voxel size=1 x 1 x 1 mm

58

C. Age Mega-Analysis

Supplementary Table S3. R-values, beta-values (reflecting change in mm
3
volume for every year of age), standard
errors (SE), and p-values from associations between age and regional brain volumes across 1,044 HIV+
individuals.
ROI
Total (n=1044) Male (n=734) Female (n=310)
r β SE p r β SE p r β SE p
Thalamus -0.24 -25.26 3.16 3.78E-15** -0.34 -28.62 2.96 7.08E-21** -0.21 -20.22 5.61 0.0004**
Caudate -0.13 -7.31 1.75 3.12E-5** -0.15 -6.96 1.68 3.68E-5** -0.16 -8.55 3.12 0.0065*
Putamen -0.29 -24.29 2.51 2.73E-21** -0.33 -24.01 2.54 4.35E-20** -0.36 -27.17 4.06 1.15E-10**
Pallidum -0.21 -5.99 0.88 1.76E-11** -0.22 -5.20 0.87 3.29E-9** -0.29 -7.75 1.51 5.51E-7**
Hippocampus -0.16 -9.01 1.70 1.34E-7** -0.24 -10.84 1.66 1.20E-10** -0.13 -6.30 2.93 0.032*
Amygdala -0.15 -3.77 0.80 3.15E-6** -0.20 -4.33 0.80 7.12E-8** -0.14 -3.38 1.39 0.016*
Accumbens -0.25 -3.19 0.39 1.15E-15** -0.26 -2.84 0.40 3.54E-12** -0.31 -3.75 0.67 4.35E-8**
Lateral
Ventricles
0.20 326.47 49.65 7.74E-11** 0.27 383.52 51.63 3.16E-13** 0.17 193.75 64.59 0.0029**
**Significant at Bonferroni corrected threshold for tests in 8 regions of interest, p ≤ 0.0063.
*Suggestive at p ≤ 0.05.

D. Age Meta-Analysis

Supplementary Table S4. To confirm findings from mega-analyses where data were pooled, we performed a
meta-analyses, where effects were found separately for each participating study and then aggregated using an
inverse-variance weighted meta-analysis. Associations between regional brain volumes and age meta-analyzed
across 18 sites reflect similar findings to pooled findings. We report r-values, beta-values (reflecting change in
mm
3
volume for every year of age), standard errors (SE), heterogeneity scores (I
2
) indicating the percentage of the
total variance in effect size explained by heterogeneity across sites, and p-values. Forest plots (Supplementary
Figures 1-4) show site-specific effects, compared with pooled mega, and meta-analytical effect sizes.
ROI
Meta-Analysis
r β SE I
2
p
Thalamus -0.27 -27.45 3.04 20.16 1.79E-19**
Caudate -0.097 -5.93 1.88 35.76 0.0016**
Putamen -0.25 -22.87 2.72 0 4.72E-17**
Pallidum -0.16 -4.77 0.92 59.19 2.22E-7**
Hippocampus -0.17 -9.81 1.81 49.53 6.01E-8**
Amygdala -0.11 -3.01 0.82 24.34 0.0002**
Accumbens -0.18 -2.32 0.40 0 4.77E-9**
Lateral Ventricles 0.21 311.88 44.6 21.99 2.69E-12**
**Significant at Bonferroni corrected threshold for tests in 8 regions of interest, p ≤ 0.0063
59

E. Age Forest Plots

Supplementary Figure S1. Forest plot of effect sizes (r-values and 95% confidence intervals) for associations
between age and putamen or pallidum volumes across all 18 sites.

Supplementary Figure S2. Forest plot of effect sizes (r-values and 95% confidence intervals) for associations
between age and hippocampal or thalamic volumes across 18 sites.

60

Supplementary Figure S3. Forest plot of effect sizes (r-values and 95% confidence intervals) for associations
between age and nucleus accumbens or amygdalar volumes across 18 sites.

Supplementary Figure S4. Forest plot of effect sizes (r-values and 95% confidence intervals) for associations
between age and caudate or ventricular volumes across 18 sites.

61

F. CD4+ Count Meta-Analysis

Supplementary Table S5. Associations between regional brain volumes and current CD4+ count meta-analyzed
across 18 sites reflect similar findings to pooled findings. We report r-values, beta-values (unstandardized
regression slopes reflecting change in mm
3
volume for every 100 cells/mm
3
change in CD4+ count), standard errors
(SE), heterogeneity scores (I
2
) indicating the percentage of the total variance in effect size explained by
heterogeneity between sites, and p-values.
ROI
Meta-Analysis
r β SE I
2
p
Thalamus 0.091 25.24 8.56 0 0.0031**
Caudate 0.01 1.73 5.43 2.99 0.75
Putamen 0.076 19.07 7.71 12.26 0.013*
Pallidum 0.079 6.76 2.63 5.54 0.010*
Hippocampus 0.11 17.21 5.02 32.24 0.0006**
Amygdala 0.071 5.25 2.27 0 0.021*
Accumbens 0.057 2.19 1.19 32.24 0.065
Lateral Ventricles -0.082 -341.09 128.34 0 0.0079*
**Significant at Bonferroni corrected threshold for tests in 8 regions of interest p ≤ 0.0063.
*Suggestive at p ≤ 0.05.

G. CD4+ Count Forest Plots

Supplementary Figure S5. Forest plot of effect sizes (r-values and 95% confidence intervals) for associations
between CD4+ count and putamen or pallidum volumes across 18 sites.

62

Supplementary Figure S6. Forest plot of effect sizes (r-values and 95% confidence intervals) for associations
between CD4+ count and hippocampal or thalamic volumes across 18 sites.

Supplementary Figure S7. Forest plot of effect sizes (r-values and 95% confidence intervals) for associations
between CD4+ count and nucleus accumbens or amygdalar volumes across 18 sites.

63

Supplementary Figure S8. Forest plot of effect sizes (r-values and 95% confidence intervals) for associations
between CD4+ count and caudate or ventricular volumes across 18 sites.

H. Effect Sizes in Males Compared to Females

Supplementary Figure S9. Effect sizes in males compared to females for each ROI were highly correlated
(CD4+: Pearson’s correlation r = 0.91, p =0.002; dVL: r = 0.79, p = 0.02).

64

I. Harmonized Viral Load Threshold Analysis

Supplementary Table S6. Cohen’s d effect sizes for associations between regional brain volumes and viral load
> 400 copies/mL across 1,006 HIV+ individuals.
ROI
Total (n=1006) Male (n=724) Female (n=282)
Cohen’s d SE p Cohen’s d SE p Cohen’s d SE p
Thalamus -0.12 0.067 0.078 -0.16 0.090 0.084 -0.037 0.12 0.77
Caudate -0.011 0.067 0.88 0.056 0.090 0.54 -0.14 0.12 0.26
Putamen 0.037 0.067 0.58 0.035 0.090 0.7 0.022 0.12 0.86
Pallidum 0.031 0.067 0.65 0.094 0.090 0.3 -0.11 0.12 0.39
Hippocampus -0.23 0.067 0.0007** -0.34 0.091 0.0003** -0.075 0.12 0.55
Amygdala -0.083 0.067 0.22 -0.11 0.090 0.22 -0.056 0.12 0.66
Accumbens -0.054 0.067 0.43 -0.11 0.090 0.23 0.074 0.12 0.56
Lateral Ventricles 0.11 0.067 0.11 0.21 0.091 0.022* -0.030 0.12 0.81
% with Viral Load >
400 copies/mL
32.70% 21.55% 61.35%
**Significant at Bonferroni corrected threshold for tests in 8 regions of interest p ≤ 0.0063.
*Suggestive at p ≤ 0.05.

J. Dichotomized CD4+ Count Threshold Analysis

Supplementary Table S7. Cohen’s d effect sizes for associations between regional brain volumes and AIDS-
defining immunosupression status (CD4+ count ≤ 200) across 1,044 HIV+ individuals.
ROI
Total (n=1044) Male (n=734) Female (n=310)
Cohen’s d SE p Cohen’s d SE p Cohen’s d SE p
Thalamus -0.092 0.078 0.24 0.020 0.11 0.85 -0.25 0.12 0.050*
Caudate -0.099 0.078 0.29 -0.12 0.11 0.26 -0.072 0.12 0.57
Putamen -0.20 0.078 0.011* -0.22 0.11 0.040* -0.22 0.12 0.074
Pallidum -0.17 0.078 0.036* -0.15 0.11 0.16 -0.2 0.12 0.11
Hippocampus -0.18 0.078 0.024* -0.21 0.11 0.050* -0.17 0.12 0.19
Amygdala -0.15 0.078 0.062 -0.098 0.11 0.37 -0.23 0.12 0.073
Accumbens -0.14 0.078 0.069 -0.030 0.11 0.78 -0.37 0.12 0.0035**
Lateral Ventricles -0.002 0.078 0.98 -0.11 0.11 0.30 0.28 0.12 0.028*
% with CD4+ Count
< 200 cells/mm
3

19.35% 13.76% 32.58%
**Significant at Bonferroni corrected threshold for tests in 8 regions of interest p ≤ 0.0063.
*Suggestive at p ≤ 0.05.
65

CHAPTER 3

Diffusion-Weighted MRI Microstructural White Matter Abnormalities

66

3.1 Novel Microstructural Measures Boost Power to Detect Neurodegenerative
Disease

This section is adapted from:

Nir TM, Jahanshad N, Villalon-Reina JE, Isaev D, Zavaliangos-Petropulu A, Zhan L, Leow AD, Jack Jr
CR, Weiner MW, Thompson PM, for ADNI (2017). Fractional anisotropy derived from the diffusion
Tensor Distribution Function boosts power to detect Alzheimer’s disease deficits. Magnetic Resonance in
Medicine, 78(6):2322-2333.

67

Fractional Anisotropy Derived from the Diffusion Tensor Distribution Function Boosts Power to
Detect Alzheimer’s Disease Deficits

Talia M. Nir
1
, Neda Jahanshad
1
, Julio E. Villalon-Reina
1
, Dmitry Isaev
1
, Artemis Zavaliangos-Petropulu
1
,
Liang Zhan
2
, Alex D. Leow
3
, Clifford R. Jack, Jr
4
, Michael W. Weiner
5
, Paul M. Thompson
1
,
for the Alzheimer’s Disease Neuroimaging Initiative (ADNI)

1
Imaging Genetics Center, University of Southern California, Marina del Rey, CA, USA
2
Computer Engineering Program, University of Wisconsin-Stout, Menomonie, WI, USA
3
Department of Psychiatry and Bioengineering, University of Illinois, Chicago, IL, USA
4
Department of Radiology, Mayo Clinic and Foundation, Rochester, MN, USA
5
Department of Radiology, UCSF School of Medicine, San Francisco, CA, USA

Abstract. In diffusion MRI (dMRI), fractional anisotropy derived from the single-tensor model
(FA
DTI
) is the most widely used metric to characterize white matter (WM) micro-architecture,
despite known limitations in regions with crossing fibers. Due to time constraints when scanning
patients in clinical settings, high angular resolution diffusion imaging (HARDI) acquisition
protocols, often used to overcome these limitations, are still rare in clinical population studies.
However, the tensor distribution function (TDF) may be used to model multiple underlying fibers
by representing the diffusion profile as a probabilistic mixture of tensors. We compared the ability
of standard FA
DTI
and TDF-derived FA (FA
TDF
), calculated from a range of dMRI angular
resolutions (41, 30, 15, and 7 gradient directions), to profile WM deficits in 251 individuals from
the Alzheimer’s Disease Neuroimaging Initiative (ADNI), and to detect associations with (1) AD
diagnosis, (2) Clinical Dementia Rating scores, and (3) average hippocampal volume. Across
angular resolutions and statistical tests, FA
TDF
showed larger effect sizes than FA
DTI
, particularly
in regions preferentially affected by AD, and was less susceptible to crossing fiber anomalies. The
TDF “corrected” form of FA may be a more sensitive and accurate alternative to the commonly-
used FA
DTI
, even in clinical quality dMRI data.

Keywords: Alzheimer’s Disease, White Matter, Diffusion-weighted Imaging, Fractional
Anisotropy, Tensor Distribution Function

68

3.1.1 INTRODUCTION

Diffusion-weighted MRI (dMRI) is a variant of standard MRI that can measure the diffusion of water
molecules in biological tissues, such as the brain. By characterizing the diffusion process at the voxel
level, we can make tentative inferences about the underlying white matter (WM) microstructure and
factors that affect it (Descoteaux and Poupon 2012). Since dMRI was developed, acquisition protocols
have improved to increase angular, spatial, and spectral resolution. Multiple mathematical models have
been developed to describe the diffusion process. One of the first - and still most popular - methods to
summarize diffusion properties in a specific voxel is the single tensor model (Stejskal and Tanner 1965)
commonly referred to as diffusion tensor imaging (DTI) (Basser et al. 1994). This model is limited as it
assumes diffusion is purely Gaussian; it can only model a single fiber population, with a single dominant
orientation, at every voxel. It cannot resolve complex WM architecture, such as dispersing, crossing or
kissing fibers. Yet, at the current resolution of dMRI, between one-third and two-thirds of WM voxels
contain fiber crossings (Behrens et al. 2007, Descoteaux et al. 2009). dMRI can also be used to evaluate
disease-related gray matter (GM) abnormalities, where the micro-architecture is even more complex
(Weston et al. 2015). Nevertheless, the tensor-derived fractional anisotropy (FA
DTI
) metric is still the most
widely used scalar measure to characterize tissue micro-architecture. It is widely used in research studies
of schizophrenia, depression, autism, HIV/AIDS, and other developmental, psychiatric, and
neurodegenerative disorders including Alzheimer’s disease (Nir et al. 2013, Nir et al. 2015b).
In recent years, many new models have been proposed to overcome limitations of the tensor model,
including q-ball imaging and diffusion orientation distribution functions (ODF) (Tuch 2004), constrained
spherical deconvolution (CSD) (Tournier et al. 2004), diffusion spectrum MRI (DSI) (Wedeen et al.
2005), multicompartment models such as the “ball and stick” model (Behrens et al. 2007), and neurite
orientation dispersion and density imaging (NODDI) (Zhang et al. 2012). Due to the numerous types of
biological, neuropsychiatric and imaging data often acquired for clinical populations, time constraints are
often placed on imaging protocols to reduce patient attrition or motion, and ensure adequate sample sizes.
This currently precludes state of the art models such as NODDI and those derived from diffusion spectrum
MRI, which require extremely dense or multi-shell acquisitions, and may prevent the reliable
reconstruction of many other higher-order diffusion models. However, the tensor distribution function
(TDF), as proposed by Leow et al. (2009), may still be feasible. The TDF is a probabilistic extension of a
multi-tensor model that describes crossing fibers mathematically as an ensemble of Gaussian tensors.
69

However, unlike other multi-compartment models (Behrens et al. 2007, Alexander 2008, Barazany et al.
2009, Alexander et al. 2010, Zhang et al. 2012) where one needs to specify in advance the total number
of compartments in the tissue, the authors propose a continuous distribution of tensors, with a profile of
“weights” or relative contributions estimated for tensors with a continuously varying range of shapes and
sizes, in the tensor space.
The Alzheimer's Disease Neuroimaging Initiative (ADNI) is a large longitudinal multisite study
of healthy elderly controls, individuals with mild cognitive impairment (MCI), and Alzheimer's disease
(AD). The goal is to identify sensitive imaging biomarkers that can be used to track or predict changes in
the brain, which is vital for drug trials to identify candidates for treatment and monitor effectiveness. In
addition to the battery of cognitive tests, cerebrospinal fluid (CSF) and blood tests, ADNI collects several
functional and structural MRI modalities including T1 and T2-weighted anatomical MRI, positron
emission tomography (PET), arterial spin labeling (ASL), and resting state functional MRI, and dMRI. In
such longitudinal studies, there is a real concern about patient attrition, especially in elderly individuals
who may not be able to tolerate being confined to a scanner for long periods. In an effort to collect such a
wide range of data-types and maintain patient enrollment, there are time constraints placed on possible
dMRI protocols, including debates as to whether or not to continue acquiring dMRI in the next phase of
ADNI. Clearly it is of great interest to maximize the power of the available scans and show that even
clinical quality diffusion scans can be powerful tools for uncovering disease related abnormalities in tissue
microstructure and WM neuro-circuitry.
Here we aimed to find whether FA metrics derived using the TDF model (FA
TDF
) may be more
sensitive to disease-related differences than the corresponding FA
DTI
measure that is now widely used.
Building upon preliminary findings in (Nir et al. 2016), our goal was to understand how the imaging
protocol may influence the sensitivity of the FA metrics, and further compared performance for each
metric computed from subsamples of the full ADNI dMRI 41 gradient direction angular resolution,
including subsets of 30, 15, and 7 gradient directions. Voxel-wise association tests were used to compare
FA
TDF
and FA
DTI
metrics computed from a range of angular resolutions, and their ability to detect
microstructural differences between AD patients and healthy elderly controls (CN). We also evaluated
associations between the two FA metrics and common AD biomarkers—hippocampal volume and
Clinical Dementia Rating (CDR) scores. Finally, we evaluated the test-retest reliability of each model’s
fit and the resulting scalar FA maps. In comparing models, there is interest in detecting clinical
70

associations with maximal sensitivity and power, ideally using improved metrics, which measure standard
properties more accurately.

3.1.2 METHODS

Subjects and Image Acquisition

Standard MRI, dMRI, and clinical data were downloaded from the publicly available Alzheimer’s Disease
Neuroimaging Initiative (ADNI) database (www.loni.usc.edu/ADNI). We analyzed baseline data from
251 participants: 53 healthy controls (CN; mean age: 72.6+/-6.1 yrs; 24M/29F), 28 with significant
memory concern (SMC; mean age: 72.4+/-4.5 yrs; 10M/18F), 121 with mild cognitive impairment (MCI;
mean age: 72.6+/-7.3 yrs; 75M/46F and 49 with AD (mean age: 75.0+/-8.6 yrs; 29M/20F). Of the 53 CN
participants, 33 returned for follow-up evaluations after 3 months and their scans were used for test-retest
analyses (mean age: 72.8+/-6.5 yrs; 16M/17F). All procedures were reviewed and approved by
institutional review boards. All participants gave written informed consent.
All subjects underwent whole-brain MRI scanning on 3 T GE Medical Systems scanners at 17
acquisition sites across North America. Anatomical T1-weighted SPGR (spoiled gradient echo) sequences
(256x256 matrix; voxel size = 1.2x1.0x1.0 mm
3
; TI=400 ms; TR = 6.98 ms; TE = 2.85 ms; flip angle=11°),
and dMRI (128x128 matrix; voxel size: 2.7x2.7x2.7 mm
3
; TR=9000 ms; dMRI scan time = 9 min) were
collected. 46 separate images were acquired for each dMRI scan: 5 T2-weighted images with no diffusion
sensitization (b0 images) and 41 diffusion-weighted images (DWI; b=1000 s/mm²).
Baseline hippocampal volume summary metrics, processed using the FreeSurfer package
(http://surfer.nmr.mgh.harvard.edu/), were downloaded from the ADNI database (n=243 available), as
was the “sum-of-boxes” Clinical Dementia Rating score (CDR-sob; n=238) (Berg 1988) .

Image Preprocessing

Raw images were preprocessed as in Nir et al. (2013). Extra-cerebral tissue was removed, raw DWI
images were corrected for motion and eddy current distortions, and T1-weighted images underwent
inhomogeneity normalization. Each T1-weighted image was linearly aligned to a standard brain template.
The diffusion images were linearly and then elastically registered (Leow et al. 2007) to their respective
71

T1-weighted scans to correct for echo-planar imaging (EPI) induced susceptibility artifacts. Gradient
tables were corrected for DWI linear registrations.

Diffusion Gradient Subsampling

To gain a better understanding of the dMRI parameters necessary to employ the TDF model, we used the
framework presented in (Zhan et al. 2010) to ‘down-sample’ the angular resolution from 41 gradient
directions to include only a subset of either 30, 15, or 7 gradient images. Gradient subsets were selected
by optimizing the spherical angular distribution energy. Briefly, the angular distribution energy, Eij, of a
pair of points, i and j, on the unit sphere may be defined as the inverse of the sum of the squares of (1) the
least spherical distance between point i and point j, and (2) the least spherical distance between point i and
point j's antipodal, symmetric point J. As in prior work, we acknowledge that protocols with fewer
gradients would be independently optimized for angular distribution rather than subsample directions from
an existing protocol, but the subsampling we use is designed to lead to the most equally distributed
sampling on the sphere possible. It also helps us to assess effects of gradient count while keeping other
factors of the patient’s scan constant (e.g., motion, artifacts).

dMRI Reconstruction Models and Scalar Maps

For each angular resolution (41, 30, 15, or 7 gradient directions), three different dMRI reconstruction
models were used to generate scalar FA maps. First, a single diffusion tensor (Basser et al. 1994) -
equivalent to a 3D ellipsoid capturing a single fiber orientation - was modeled at each voxel in the brain
from the corrected DWI scans. This model assumes that the diffusion is a 3D Gaussian process, fitted
using just six independent parameters of a tensor (3 eigenvalues describing its shape, and 3 Euler angles
describing its orientation). Scalar fractional anisotropy (FA
DTI
) maps were obtained from the resulting
diffusion tensor eigenvalues ( 1, 2, 3):

𝐹𝐴 = √
(𝜆 1
− 𝜆 2
)
2
+ (𝜆 1
− 𝜆 3
)
2
+ (𝜆 2
− 𝜆 3
)
2
2[𝜆 1
2
+ 𝜆 2
2
+ 𝜆 3
2
]

72

In contrast to the single tensor model, the tensor distribution function (TDF) represents the diffusion
profile as a probabilistic mixture of tensors that optimally explain the observed DWI data, allowing for
the reconstruction of multiple underlying fibers per voxel, together with a distribution of weights or
probabilities. We applied the framework proposed in (Leow et al. 2009, Zhan et al. 2009) to the angular
diffusion signal, to compute the voxel-wise optimal TDF, P
*
(D(θ, λ)) —the probability distribution
function of all feasible Gaussian tensors D(θ, λ) that best describes the observed signal. As in Leow et al.
(2009), to reduce the solution space, each tensor D(θ, λ) at spherical angle θ was assumed to be cylindrical
such that λ=(λ1,λ2=λ3) and λ1>=λ2. However, unlike the gradient descent approach in (Leow et al. 2009)
to solve for this optimal TDF, here we used a quadratic programming approach. The tensor orientation
distribution function (TOD), was then calculated by computing the marginal density function of the TDF
with the eigenvalues λ= (λ1, λ2) integrated out.

𝑇𝑂𝐷 (𝜃 ) = ∫ 𝑃 (𝐷 (𝜃 , 𝜆 ))𝑑𝜆 𝜆

For each θ, the eigenvalues are calculated by computing the expected value of each eigenvalue along θ,
from which a corresponding scalar FA metric is calculated:

𝜆 𝑖 ′
(𝜃 ) =
∫𝑃 (𝐷 (𝜃 , 𝜆 ))𝜆 𝑖 𝑑𝜆 ∫𝑃 (𝐷 (𝜃 , 𝜆 ))𝑑𝜆

𝐹𝐴 (𝜃 ) = √
(𝜆 1
′
(𝜃 ) − 𝜆 2
′
(𝜃 ))
2
+ (𝜆 1
′
(𝜃 ) − 𝜆 3
′
(𝜃 ))
2
+ (𝜆 2
′
(𝜃 ) − 𝜆 3
′
(𝜃 ))
2
2[𝜆 1
′
(𝜃 )
2
+ 𝜆 2
′
(𝜃 )
2
+ 𝜆 3
′
(𝜃 )
2
]

At each voxel, the final scalar FA
TDF
metric across all θ is then calculated as the sum of all FA(θ) weighted
by the probability that θ is the principal fiber direction, TOD(θ).

𝐹𝐴
𝑇𝐷𝐹 = ∫ 𝑇𝑂𝐷 (𝜃 ) ∗ 𝐹𝐴 (𝜃 )𝑑𝜃

73

A healthy control subject’s FA
DTI
and FA
TDF
maps, calculated from various angular resolutions,
are shown in Figure 1a and 1b for visual comparison. A voxel-wise two-tailed paired T-test was
performed to quantitatively compare TDF and DTI FA values in the healthy control group. All resulting
statistical maps were corrected for multiple comparisons using the standard false discovery rate (FDR)
method at q=0.05 (Benjamini and Hochberg 1995), and thresholded at the FDR critical P-value.
For comparison to an established HARDI technique also designed to reconstruct multiple fiber
orientations in a given voxel, we fitted orientation distribution functions (ODFs) at each voxel, with a non-
parametric q-ball reconstruction technique, using the normalized and dimensionless constant solid angle
(CSA) method (Tuch 2004, Aganj et al. 2010). The generalized FA (GFA
ODF
) was then calculated from
the CSA-ODF. GFA
ODF
is analogous to FA
DTI
, but calculated at each diffusion direction of the ODF (Tuch
2004) and is defined as:

𝐺𝐹𝐴 =
𝑠𝑡𝑑 (𝛹 )
𝑟𝑚𝑠 (𝛹 )
= √
𝑛 ∑ (𝛹 (𝑢 𝑖 ) − 〈𝛹 〉)
2 𝑛 𝑖 =1
(𝑛 − 1) ∑ 𝛹 (𝑢 𝑖 )
2 𝑛 𝑖 =1

Here, Ψ(u) is the ODF, i is each diffusion direction and 〈Ψ〉 =
1
𝑛 ∑ Ψ(𝑢 𝑖 )
𝑛 𝑖 =1
is the mean of the ODF.

Template Creation and Spatial Normalization

To avoid bias in the diffusion-based registrations, we created a multi-channel study-specific minimal
deformation template (MDT) with the ANTs registration software (Avants et al. 2011), equally weighting
FA
DTI
, FA
TDF
, GFA
ODF
and T1-weighted maps. Similarly, to spatially normalize each subject’s three FA
maps we performed a 3-channel linear then non-linear registration to the MDT. In this way all FA maps
were used to drive the registration and they were all normalized to the same space. To avoid differences
in registration accuracy, the deformations from the full angular resolution registration were applied to the
FA maps calculated from the various DWI gradient subsets for each individual.
Test-retest FA
DTI
and FA
TDF
maps generated from baseline and 3-month follow-up dMRI scans
were each linearly aligned to an intermediate space half way between each subject’s two time points
(Smith et al. 2002). Baseline and follow-up maps were each spatially normalized to the baseline MDT
74

with 2-channel linear then non-linear registrations. The deformations from the full angular resolution
registration were applied to the various test-retest DWI gradient subsets as well as the respective FA maps

Clinical Associations and Effect Sizes

To test for statistical effects of AD diagnosis on measures of white matter microstructure –FA
DTI
, FA
TDF
,
and GFA
ODF
maps calculated at various angular resolutions— we ran voxel-wise, random-effects linear
regressions, covarying for age and sex, and using the acquisition site as the random grouping variable. In
an effort to try and tease apart microstructural associations from those driven by atrophy and registration,
for each voxel, we also covaried for the log Jacobian determinant derived from the non-linear spatial
normalization of each map to the template. In addition to AD diagnosis, we also tested for voxel-wise
associations between FA and CDR-sob scores as well as average bilateral hippocampal volume (after
covarying for intracranial volume) across the entire study sample. All statistical tests were limited to
voxels present in all subject scans, as some scans had a slightly cropped field of view. As such, we did not
consider the inferior parts of the cerebellum and brain stem. All resulting statistical maps were corrected
for multiple comparisons using the standard false discovery rate (FDR) method at q=0.05 (Benjamini and
Hochberg 1995), and thresholded at the FDR critical P-value. We show regression coefficients (β-values)
only in regions where the false discovery rate was controlled.
We computed cumulative distribution function (CDF) plots to visualize and rank effect sizes across
voxel-wise tests. The cumulative observed voxel P-values from each regression were plotted against the
P-values from the expected null distribution. If the CDF curve initially rises at a rate steeper than 20 times
the null CDF (y = 20x), then the corresponding maps have supra-threshold or FDR significant voxels at q
= 0.05. Curves that rise at a steeper rate than that line represent significant voxels and larger deviations
represent larger effect sizes.
Effect sizes for detecting AD versus control group differences were also compared using Cohen's
d calculated as (μ AD−μCN)/spooled, where spooled=√[(s CN
2
+s AD
2
) / 2] (Cohen 1988). This metric has been
widely used in studies of disease effects on imaging measures (Hibar et al. 2016, Schmaal et al. 2016, van
Erp et al. 2016). For each FA metric and angular resolution, we used the average FA from the respective
statistical test’s significant cluster. To avoid overfitting, a 10-fold cross validation approach was used. In
each fold, 80% of the data were used for voxel-wise regressions to estimate the significant clusters
(training data), and the remaining test data was used to compute the Cohen's d effect sizes.
75

Test-Retest Reliability and Model Fit

We used the framework defined in (Pestilli et al. 2014, Rokem et al. 2015) to evaluate the goodness of the
fit of each dMRI model in healthy controls. We first compared voxel-wise root mean squared error
(RMSE) between the observed signal (A) and expected signal (B) from each model in each voxel:

RMSE(A, B) =
√
∑ (A
i
− B
i
)
2 N
i=1
N

where N is the number of gradient directions or DWIs, and Ai, Bi - the observed and expected signal
intensities in the given voxel in the i-th DWI. Additionally, baseline and 3-month follow-up test-retest
data was used to cross-validate each model’s fit—the model parameters were first estimated on baseline
control subjects’ DWI scans, and then used to predict the signal in the 3-month follow-up DWI scan. As
proposed in (Rokem et al. 2015), we defined the test-retest relative root mean squared error (rRMSE) in
each voxel as:

rRMSE =
(RMSE(M1, D2) + RMSE(M2, D1))
2RMSE(D1, D2)

Here, RMSE(M1,D2) is the RMSE between the data observed in the follow-up scan and predicted from
the first scan, RMSE(M2,D1) is the RMSE between the observed data in the first scan and that predicted
from the follow-up scan, and RMSE(D1,D2) is the RMSE between the observed data from both scans. A
model that predicts the repeated measurement more accurately than the original will result in an rRMSE
< 1 (Rokem et al. 2015). To compute RMSE (M1,D2), we used parameters learned from the first scan,
and the bvecs (scanner gradient directions) and b0 from the follow-up scan, and vice versa for
RMSE(M2,D1). A voxel-wise two-tailed paired T-test was performed to compare TDF and DTI baseline
RMSE and rRMSE values in the healthy control group. All resulting statistical maps were corrected for
multiple comparisons using the standard false discovery rate (FDR) method at q=0.05 (Benjamini and
Hochberg 1995); resulting maps were thresholded at the FDR critical P-value.
We also evaluated the test-retest reliability for FA
DTI
and FA
TDF
maps calculated from each angular
resolution by computing the voxel-wise intra-class correlation (ICC) between baseline and 3-month
76

follow-up healthy control FA maps, with the R PSYCH package (personality-
project.org/r/html/ICC.html). Again, the FDR method was used to correct for multiple comparisons.

3.1.3 RESULTS

Both a visual comparison of FA
DTI
and FA
TDF
maps (Figure 1a and 1b) and T-test between maps (Figure
1c) reveal that FA
TDF
maps have higher FA values not only in the core, coherent WM structures but
throughout the tissue, including near gray / white matter boundaries. The standard FA
DTI
measure tends
to show loss of signal near cortical boundaries and in regions with known fiber crossings and complex
gray matter architecture.

Figure 1. Diffusion fractional anisotropy (FA) maps, (a) FA
TDF
and (b) FA
DTI
, are shown for a single subject
calculated from 41, 30, 15 and 7 gradient direction sets. The FA
TDF
maps show more sharply defined white matter
(WM) boundaries, with much less signal drop-out in regions near the cortex that tend to have less coherent WM,
compared to FA
DTI
maps. (c) T-maps in regions where FA
DTI
and FA
TDF
maps are significantly different reveal
lower FA
DTI
values (negative association) throughout the tissue regardless of angular resolution (FDR critical P-
value for 41 gradients = 0.047, 30 gradients = 0.047, 15 gradients = 0.047, and 7 gradients = 0.046).
77

Figure 2. (a) Beta-maps show regions where lower FA
DTI
, FA
TDF
, and GFA
ODF
is significantly associated with AD
diagnosis, higher CDR-sob cognitive deficits, and lower average bilateral hippocampal volume. Across tests, FA
TDF

maps (middle row,) consistently show larger effect sizes in temporal lobe and hippocampal regions. This is denoted
by greater β-value magnitudes and more pervasive significant associations. The patterns are also more in line with
the expected topography of the disease effects. (b) Cumulative distribution function (CDF) plots show effect sizes
for FA
DTI
, FA
TDF
, and GFA
ODF
statistical associations. FA
TDF
maps (green curves) are consistently the most sensitive
metric (denoted by the higher critical P-values controlling the FDR, i.e., the highest non-zero x-coordinate where
the CDF crosses the y=20x line). (c) The absolute number and percentage of total significant voxels surviving FDR
correction, showing an association direction opposite to that traditionally accepted as showing impairment—dark
blue voxels highlighted by boxes in (a). FA
DTI
and GFA
ODF
associations show ~8-15%, while FA
TDF
tests show
<0.5% suggesting that FA
TDF
may be handling computations better in areas with crossing fibers.

Clinical Associations and Effect Sizes

As expected, across all of the FA metrics, AD diagnosis, greater cognitive impairment (higher CDR-sob
score), and lower average hippocampal volume, were associated with significant WM deficits (lower FA)
after correction for multiple comparisons (Figure 2a). However, across statistical tests, larger effect sizes,
as denoted by greater β-value magnitude and more widespread differences, were detected with FA
TDF

maps, as compared to FA
DTI
and GFA
ODF
. Moreover, FA
TDF
findings were highly localized to the temporal
78

lobe and hippocampal regions most vulnerable to early changes in AD. CDF plots further reflect the
increased sensitivity of FA
TDF
for differentiating disease groups, and for detecting clinical associations
(Figure 2b).
Across maps, some very small regions exhibited significant associations with FA in direction
contrary to what would traditionally be accepted as showing impairment (i.e., higher FA with impairment;
Figure 2a boxed regions). These regions were largely found at the junction of the corpus callosum
commissural fibers and the corona radiata, a region notorious for fiber crossings that may reduce FA as
computed from the tensor model (Oishi 2011). However, across analyses, FA
TDF
showed fewer
associations that were contrary to the hypothesized effects of the disease (Figure 2c). Relative to the total
number of significant voxels, FA
DTI
showed between ~13-15% of these voxels, GFA
ODF
showed ~8-10%,
while FA
TDF
showed <0.5% across tests, suggesting that FA
TDF
may be handling computations better for
crossing fibers.
A comparison of the same three clinical associations with FA
DTI
and FA
TDF
computed from a
subset of 30, 15 and 7 gradient directions revealed that even at 7 gradient directions, FA
TDF
was
consistently the most sensitive metric across statistical tests (Table 1; Figure 3). In fact FA
TDF
calculated
from 7 gradient directions had larger effect sizes than FA
DTI
calculated at the full angular resolution.
Across statistical tests performed at each angular resolution, FA
TDF
consistently showed less than 0.5% of
significant voxels with a direction of association opposite to that hypothesized, compared to FA
DTI
which
showed between 5% and 15% (Table 1). Finally, the mean Cohen's d effect sizes for picking up AD versus
CN group differences—which were calculated from test data from the 10-fold cross validation— once
again confirmed that across all angular resolutions, FA
TDF
showed larger effect sizes (Table 2).

79

Table 1. Comparison of FA
DTI
and FA
TDF
effect sizes, based on FDR critical P-value and percent of voxels surviving
FDR correction, for the full 41 gradient angular resolution, and each of the 30, 15, and 7 gradient subsets, followed
by the absolute number and percent of total significant voxels surviving FDR correction, showing an association
direction opposite to that traditionally accepted as showing impairment (i.e., higher FA with increased deficits).

FDR Critical
P-Value
Surviving
Voxels
Surviving Voxels w/
Opposite Association
Gradients 41 30 15 7 41 30 15 7 41 30 15 7
ADvCN
FA
TDF
0.0076 0.0085 0.0099 0.0105 15.17% 16.99% 19.83% 20.91%
0.22%
(59)
0.17%
(53)
0.10%
(34)
0.16%
(59)
FA
DTI

0.0009 0.0008 0.0005 0.0004 1.80% 1.63% 1.07% 0.71%
13.60%
(441)
13.75%
(404)
13.43%
(259)
5.92%
(76)
Avg Hipp
FA
TDF
0.0179 0.0208 0.0208 0.0165 35.88% 41.57% 41.53% 32.99%
0.18%
(117)
0.14%
(101)
0.10%
(76)
0.17%
(98)
FA
DTI

0.0018 0.0018 0.0014 0.0006 3.56% 3.65% 2.80% 1.21%
14.80%
(949)
13.81%
(909)
13.91%
(701)
10.11%
(220)
CDR-sob
FA
TDF
0.0161 0.0167 0.0177 0.0180 32.18% 33.42% 35.48% 36.00%
0.15%
(88)
0.16%
(97)
0.13%
(80)
0.18%
(117)
FA
DTI
0.0010 0.0009 0.0005 0.0004 1.98% 1.82% 1.05% 0.88%
14.20%
(506)
14.27%
(468)
13.70%
(260)
4.52%
(72)

Figure 3. Cumulative distribution function (CDF) plots of statistical associations between (a) AD diagnosis, (b)
average bilateral hippocampal volume, and (c) CDR-sob and FA
DTI
or FA
TDF
maps computed from 41, 30, 15 and
7 gradient direction sets. FA
TDF
maps (green curves) are consistently the most sensitive metric (denoted by the
higher critical P-values controlling the FDR, i.e., the highest non-zero x-coordinate where the CDF crosses the
y=20x line) across all gradient subsets. Curves correspond to values listed in Table 1.

Table 2. Mean and standard deviation (SD) Cohen's d effect sizes across 10 folds, for picking up FA
DTI
and FA
TDF

group differences between AD patients and healthy CN, across angular resolutions.
Gradients
FA
DTI
Mean (SD)
FA
TDF
Mean (SD)
41 1.64 (0.11) 1.90 (0.06)
30 1.63 (0.14) 1.88 (0.06)
15 1.64 (0.15) 1.87 (0.07)
7 1.77 (0.39) 1.95 (0.08)
80

Test-Retest Reliability and Model Fit

Mean maps of the RMSE calculated from the DTI and TDF model fit in the subset of 53 healthy control
subjects are shown in Figure 4a and 4b. A voxel-wise T-test revealed significantly lower error in the TDF
fit throughout the tissue regardless of angular resolution (Figure 5a; FDR critical P-value for 41 gradients
= 0.041, 30 gradients = 0.041, 15 gradients = 0.042, and 7 gradients = 0.043). The mean rRMSE maps
from DTI and TDF models from 33 healthy control individuals at two time points are shown in Figure 4c
and 4d. While the rRMSE was high in both the TDF and DTI models in the superior cortical gray matter
(mean rRMSE > 1), the fit was stable (< 1) in WM and overall temporal lobe regions where most of the
AD-related effects were detected. A T-test between the DTI and TDF rRMSE maps (Figure 5b) revealed
significantly lower rRMSE for the TDF fit, in not only the temporal lobes, but in the region of the superior
WM where commissural fibers and the corona radiata intersect, often leading to depleted FA
DTI
. The TDF
model showed higher error only in CSF. TDF rRMSE was progressively more similar to DTI (i.e., less
area of significant differences) with lower angular resolution (FDR critical P-value for 41 gradients =
0.024, 30 gradients = 0.021, 15 gradients = 0.017, and 7 gradients = 0.001).
In terms of test-retest reliability of the scalar FA maps, we found that across resolutions there was
an overall stable and strong ICC between baseline and follow-up FA
TDF
maps (mean ICC ~0.8 ; Figure
5c), while, as might be expected, there was a degradation in FA
DTI
ICC at the lowest angular resolutions
(Figure 5d).

81

Figure 4. The left two columns show the root mean squared error (RMSE) maps from the (a) DTI and (b) TDF
model fit, averaged across 53 healthy control subjects. The right two columns show the relative RMSE (rRMSE)
maps from the (c) DTI and (d) TDF model fit, trained on baseline scans and tested on 3-month follow-up scans in
each of 33 control subjects individually, and averaged across the group.
82

Figure 5. Statistical differences in reliability between TDF and DTI models. (a) T-maps in regions where the root
mean squared error (RMSE) maps of the TDF and DTI model fit in 53 healthy controls are significantly different
reveal higher error for DTI (positive association) throughout the tissue (red) regardless of angular resolution (FDR
critical P-value for 41 gradients = 0.041, 30 gradients = 0.041, 15 gradients = 0.042, and 7 gradients = 0.043). (b)
T-maps in regions where the relative RMSE (rRMSE) maps of the TDF and DTI model fit, trained on 33 healthy
controls’ baseline scans and tested on 3-month follow-up scans, are significantly different reveal higher error for
DTI (positive association) in the tissue (red), particularly in regions of known crossing fibers (FDR critical P-value
for 41 gradients = 0.024, 30 gradients = 0.021, 15 gradients = 0.017, and 7 gradients = 0.001); the TDF model
showed higher error only in CSF. (c) Intra-class correlation (ICC) maps in regions with a significant ICC between
baseline and 3-month follow-up FA
TDF
maps (FDR critical P -value for 41 gradients = 0.049, 30 gradients =0.049,
15 gradients = 0.049, and 7 gradients = 0.049) and (d) FA
DTI
maps (FDR critical P -value for 41 gradients = 0.050,
30 gradients = 0.050, 15 gradients = 0.049, and 7 gradients = 0.047). The mean ICC and standard deviation (SD) of
the ICC across all voxels are reported below each mapped coronal slice.

83

3.1.4 DISCUSSION

FA metrics derived from the tensor distribution function (TDF) may be more sensitive to disease related
microstructural abnormalities than corresponding single tensor-derived FA metrics that are now widely
used to assess clinical data. FA is highly affected by numerous factors including the number of dominant
fiber directions and orientation coherence as well as partial volume effects from neighboring gray matter.
By using the TDF approach, we can still employ an extension of the tensor model, adapted to identify
contributions to FA from separate crossing fiber compartments in tissue with more complex micro-
architecture and in voxels on tissue boundaries that are susceptible to partial voluming.
Alzheimer’s disease (AD) is characterized by cortical and hippocampal neuronal loss and
widespread gray matter (GM) atrophy driven in part by cortical amyloid plaque and neurofibrillary tangle
deposits, and vascular changes. Structural and diffusion MRI studies show white matter (WM)
abnormalities –perhaps due to myelin degeneration, and neuronal loss leading to progressive
disconnection of cortical and subcortical regions (Delbeuck et al. 2003, Oishi et al. 2011, Nir et al. 2013,
Toga and Thompson 2013, Jack et al. 2015). Standard anatomical MRI is still the imaging technique most
often used in AD studies and clinical trials, but dMRI is sensitive to microscopic changes in WM integrity
not always detectable with standard anatomical MRI (Xie et al. 2006, Canu et al. 2010). In addition to
WM, dMRI is an emerging tool for the evaluation of disease related gray matter (GM) abnormalities as
well (Cercignani et al. 2001, Bozzali et al. 2002, Kantarci et al. 2005, Whitwell et al. 2010, Chiapponi et
al. 2013, Weston et al. 2015). A growing number of studies assess cortical and subcortical GM diffusivity
changes in AD that may reflect GM cellular microstructure breakdown (Weston et al. 2015). Several
studies report microscopic changes in the hippocampi that may be detectable prior to volumetric changes
(Kantarci et al. 2005, Muller et al. 2005). Because dMRI changes may be detectable before (and therefore
predict) gross volume loss (Hugenschmidt et al. 2008, Nir et al. 2015a), it is important to maximize the
power to detect such changes.
In this study, we found that compared to both GFA
ODF
and FA
DTI
, FA
TDF
showed increased power
to detect subtle or diffuse disease effects, especially in hippocampal and temporal lobe regions. AD
pathology targets GM regions especially in the temporal lobe and hippocampus. In these regions, FA
DTI

might be sub-optimal, as it is best suited to detect differences in cohesive WM fiber bundles such as the
corpus callosum. We also found more significant FA
TDF
associations in voxels at GM or CSF boundaries
that may otherwise be susceptible to partial volume effects with FA
DTI
. Compared to FA
DTI
, larger FA
TDF

84

effect sizes were preserved even when the dMRI angular resolution was subsampled from 41 gradient
directions to 30, 15, or even 7 gradient directions. Perhaps surprisingly, FA
TDF
calculated from 7 gradient
directions had larger effect sizes than FA
DTI
calculated at the full angular resolution. While some higher
order models require extremely dense or multi-shell acquisitions, TDF may better extract the information
typically available in clinical settings, where time constraints limit scan times. It may also be helpful for
studies of valuable but lower-resolution legacy data. The TDF as proposed by Leow et al. (2009) makes
no assumptions about the number of compartments per voxel and, unlike the tensor distribution function
previously proposed by Jian et al. (2007), does not impose the same exact anisotropy profile on all fiber
compartments. This may lead to better estimates if there are higher levels of uncertainty in the data, such
as may arise with low resolution data.
Furthermore, FA
TDF
may also help to interpret apparent increases in FA
DTI
found in disease. In
many contexts, lower FA is hypothesized to reflect impairment. However, relative increases in FA have
been reported in FA
DTI
studies of AD, which may reflect a selective sparing or selective degeneration of
one of the pathways in a region with crossing fibers (Douaud et al. 2011). However, without histologic
data, we cannot be certain whether selective degeneration or increased “integrity”, or some mix of both,
is driving higher FA
DTI
values in a neuroimaging study. FA
TDF
, on the other hand, takes into account
crossing fiber compartments. A relatively higher FA may more consistently reflect healthier tissue, while
lower FA more consistently reflects deficits, making the direction of associations easier to interpret.
Across analyses, we found that FA
TDF
showed fewer ‘contrary to hypothesis’ regions (i.e., higher FA
associated with greater deficits). Across all statistical tests and angular resolutions, compared to FA
TDF
,
FA
DTI
showed both a higher absolute number of these types of significant voxels, and a higher percentage
relative to the total number of significant voxels (~5-15% of voxels versus <0.5% with FA
TDF
), suggesting
that FA
TDF
may in fact be resolving crossing fibers.
Analyses of test-retest reliability and model fit show that the TDF is quite reliable and robust in
regions that show disease effects in this analysis. Some instability in test-retest reliability is to be expected
when using the ADNI dataset - the scans are 3 months apart, and the healthy control participants are
elderly (mean age: 72.8 +/- 6.5 yrs). These individuals may show some biological aging and white matter
deterioration, even over the 3-month interscan interval. All registrations were visually evaluated, but age-
related changes and minor geometric miscalibration of the scanner may also contribute to minor
discrepancies in alignment between 2 scans from the same subject, contributing to both the final test-retest
ICC and rRMSE measures in both the DTI and TDF models.
85

A prior study also showed that FA
TDF
was a more stable metric with decreasing spatial resolution,
whereas FA
DTI
values decreased more rapidly due to more fiber incoherence and greater partial voluming
in larger voxels (GadElkarim et al. 2011). However, further analyses of FA
TDF
limitations on a wider range
of diffusion protocols and comparisons of performance to numerous other proposed scalar metrics are
necessary. In addition to FA, there is also a growing interest in assessing complementary diffusivity
metrics, including mean diffusivity (MD), axial diffusivity (AxD), and radial diffusivity (RD). As, FA is
an inherently normalized measure and diffusivity metrics are not, future work is necessary to define
analogous measures within the TDF framework.
Multi-shell and other DSI or q-space techniques may ultimately outperform tensor model metrics,
but they are often less feasible given the time constraints on dMRI protocols in clinical settings, as well
as for recovering information from valuable legacy data. The TDF model may ultimately allow us to take
advantage of such available clinical quality diffusion data with more sensitivity and fewer limitations than
the classic DTI model.

3.1.5 ACKNOWLEDGEMENTS

Data collection and sharing for this project was funded by the Alzheimer's Disease Neuroimaging
Initiative (ADNI) (National Institutes of Health Grant U01 AG024904) and DOD ADNI (Department of
Defense award number W81XWH-12-2-0012). ADNI is funded by the National Institute on Aging, the
National Institute of Biomedical Imaging and Bioengineering, and through generous contributions from
the following: Alzheimer’s Association; Alzheimer’s Drug Discovery Foundation; Araclon Biotech;
BioClinica, Inc.; Biogen Idec Inc.; Bristol-Myers Squibb Company; Eisai Inc.; Elan Pharmaceuticals, Inc.;
Eli Lilly and Company; EuroImmun; F. Hoffmann-La Roche Ltd and its affiliated company Genentech,
Inc.; Fujirebio; GE Healthcare; IXICO Ltd.; Janssen Alzheimer Immunotherapy Research &
Development, LLC.; Johnson & Johnson Pharmaceutical Research & Development LLC.; Medpace, Inc.;
Merck & Co., Inc.; Meso Scale Diagnostics, LLC.; NeuroRx Research; Neurotrack Technologies;
Novartis Pharmaceuticals Corporation; Pfizer Inc.; Piramal Imaging; Servier; Synarc Inc.; and Takeda
Pharmaceutical Company. The Canadian Institutes of Health Research is providing funds to support ADNI
clinical sites in Canada. Private sector contributions are facilitated by the Foundation for the National
Institutes of Health (www.fnih.org). The grantee organization is the Northern California Institute for
Research and Education, and the study is coordinated by the Alzheimer's Disease Cooperative Study at
86

the University of California, San Diego. ADNI data are disseminated by the Laboratory for Neuro Imaging
at the University of Southern California. Funding for the ENIGMA Center for Worldwide Medicine
Imaging and Genomics is provided as part of the BD2K Initiative under grant number U54 EB020403 to
support big data analytics.

87

3.2 Pooling and Harmonizing Multi-protocol Diffusion MRI Data in ADNI

This section is adapted from:

Zavaliangos-Petropulu A*, Nir TM*, Thomopoulos SI, Reid RI, Bernstein MA, Borowski B, Jack Jr CR,
Weiner MW, Jahanshad N, Thompson PM (2019). Diffusion MRI indices of cognitive impairment in brain
aging: The updated multi-protocol approach in ADNI3. Frontiers in NeuroInformatics, 13(2).

88

Diffusion MRI Indices and their Relation to Cognitive Impairment in Brain Aging: The Updated
Multi-Protocol Approach in ADNI3

*Artemis Zavaliangos-Petropulu
1
, *Talia M. Nir
1
*, Sophia I. Thomopoulos
1
, Robert I. Reid
2
, Matt A. Bernstein
3
,
Bret Borowski
3
, Clifford R. Jack, Jr.
3
, Michael W. Weiner
4
, Neda Jahanshad
1
, Paul M. Thompson
1
, for the
Alzheim
* These authors contributed equally to the manuscript

1
Imaging Genetics Center, Mark & Mary Stevens Neuroimaging & Informatics Institute, Keck School of Medicine, University of Southern
California, Marina del Rey, CA, USA
2
Department of Information Technology, Mayo Clinic and Foundation, Rochester, MN, USA
3
Department of Radiology, Mayo Clinic and Foundation, Rochester, MN, USA
4
Department of Radiology, University of California San Francisco School of Medicine, San Francisco, CA, USA

Abstract. Brain imaging with diffusion-weighted MRI (dMRI) is sensitive to microstructural
white matter changes associated with brain aging and neurodegeneration. In its third phase, the
Alzheimer’s Disease Neuroimaging Initiative (ADNI3) is collecting data across multiple sites and
scanners using different dMRI acquisition protocols, to better understand disease effects. It is vital
to understand when data can be pooled across scanners, and how the choice of dMRI protocol
affects the sensitivity of extracted measures to differences in clinical impairment. Here, we
analyzed ADNI3 data from 317 participants (mean age: 75.4±7.9 years; 143 men/174 women),
who were each scanned at one of 47 sites with one of six dMRI protocols using scanners from
three different manufacturers. We computed four standard diffusion tensor imaging (DTI) indices
including fractional anisotropy (FA
DTI
) and mean, radial, and axial diffusivity, and one FA index
based on the tensor distribution function (FA
TDF
), in 24 bilaterally averaged white matter regions
of interest. We found that protocol differences significantly affected dMRI indices, in particular
FA
DTI
. We ranked the diffusion indices for their strength of association with four clinical
assessments. In addition to diagnosis, we evaluated cognitive impairment as indexed by three
commonly used screening tools for detecting dementia and Alzheimer’s disease: the Alzheimer’s
Disease Assessment Scale (ADAS-cog), the Mini-Mental State Examination (MMSE), and the
Clinical Dementia Rating scale sum-of-boxes (CDR-sob). Using a nested random-effects
regression model to account for protocol and site, we found that across all dMRI indices and
clinical measures, the hippocampal-cingulum and fornix (crus) / stria terminalis regions most
consistently showed strong associations with clinical impairment. Overall, the greatest effect sizes
were detected in the hippocampal-cingulum and uncinate fasciculus for associations between axial
or mean diffusivity and CDR-sob. FA
TDF
detected robust widespread associations with clinical
measures, while FA
DTI
was the weakest of the five indices for detecting associations. Ultimately,
we were able to successfully pool dMRI data from multiple acquisition protocols from ADNI3 and
detect consistent and robust associations with clinical impairment and age.

Keywords: Alzheimer’s disease, ADNI3, White Matter, DTI, Multi-site, Harmonization, TDF,
ComBat
89

3.2.1 INTRODUCTION

Alzheimer’s disease (AD) is the most common type of dementia, affecting approximately 10% of the
population over age 65 (Alzheimer’s Association 2018). As life expectancy increases, there is an ever-
increasing need for sensitive biomarkers of AD - to better understand the disease, and to serve as surrogate
markers of disease burden for use in treatment and prevention trials. The Alzheimer’s Disease
Neuroimaging Initiative (ADNI) is an ongoing large-scale, multi-center, longitudinal study designed to
improve methods for clinical trials by identifying brain imaging, clinical, cognitive, and molecular
biomarkers of AD and aging. Now in its third phase (ADNI3), ADNI continues to incorporate newer
technologies as they become established (Jack et al. 2015); data from ADNI, collected at participating
sites across the U.S. and Canada, is publicly available and has been used in a diverse range of publications
(Veitch et al. 2018).
ADNI’s second phase (ADNI2) introduced to the initiative the use of diffusion-weighted MRI
(dMRI) as an additional approach for tracking AD progression (Jack et al. 2015). dMRI has since been
used in numerous studies to understand the effects of AD on white matter (WM) microstructure and brain
connectivity (Daianu et al. 2013a, Daianu et al. 2013b, Nir et al. 2013, Prasad et al. 2013b). Some of these
approaches use scalar dMRI measures to evaluate microstructural WM changes not detectable with
anatomical T1-weighted images (Giulietti et al. 2018), while others use tractography and graph-theory
analyses to study abnormalities in structural brain networks (Hu et al. 2016, Maggipinto et al. 2017,
Sulaimany et al. 2017, Powell et al. 2018, Sanchez-Rodriguez et al. 2018) (Nir et al. 2015a). In aggregate,
these studies point to WM abnormalities in AD, which may play a key role in early pathogenesis and
diagnosis (Sachdev et al. 2013).
ADNI2 acquired dMRI data with one acquisition protocol from approximately one third of
enrolled participants at the subset of ADNI sites that used 3 tesla General Electric (GE) scanners. To
ensure that dMRI could be collected from all enrolled participants, ADNI3 developed new dMRI protocols
for all GE, Siemens and Philips scanners used across ADNI sites. Now, data is being acquired with seven
different dMRI acquisition protocols (see methods for details;
http://adni.loni.usc.edu/methods/documents/mri-protocols/). ADNI3 began in October 2016, and has
already acquired data from over 300 participants. dMRI spatial resolution was improved between ADNI2
and ADNI3 by reducing the voxel size from 2.7x2.7x2.7 mm to 2.0x2.0x2.0 mm. While voxel size (i.e.,
spatial resolution) remains consistent across all seven ADNI3 protocols, angular resolution (the number
90

of gradient directions) varies across protocols to accommodate scanner restrictions and to ensure that the
multi-modal scanning session is completed in under 60 minutes. Although many large-scale multi-site
DTI studies have obtained consistent results even when acquisition protocols across sites are not
harmonized in advance (Jahanshad et al. 2013a, Kochunov et al. 2014, Acheson et al. 2017, Kelly et al.
2018), differences in dMRI acquisition parameters, including vendor, voxel size, and angular resolution,
are known to affect derived dMRI measures (Alexander et al. 2001, Cercignani et al. 2003, Zhan et al.
2010, Zhu et al. 2011). As a result, improved harmonization of multi-site diffusion data is of great interest
(Grech-Sollars et al. 2015, Pohl et al. 2016, Palacios et al. 2017). For example, ComBat - originally
developed to model and remove batch effects from genomic microarray data (Johnson et al. 2007) –was
one of the most effective methods for harmonizing DTI measures in a recent comparison of such
techniques (Fortin et al. 2017).
Here we tested whether standard diffusion tensor imaging (DTI)-derived anisotropy and diffusivity
indices, calculated from multiple imaging protocols in ADNI3, can be pooled and harmonized to show
robust associations with age and four clinical assessments. In addition to diagnosis, cognitive impairment
was assessed with three commonly used screening tools for detecting dementia and Alzheimer’s disease:
the Alzheimer’s Disease Assessment Scale (ADAS-cog) (Rosen et al. 1984), the Mini-Mental State
Examination (MMSE) (Folstein et al. 1975), and the Clinical Dementia Rating scale sum-of-boxes (CDR-
sob) (Berg 1988). For the rest of the paper we refer to these tools as “cognitive measures”. In addition to
standard DTI indices –fractional anisotropy (FA
DTI
), mean diffusivity (MD
DTI
), radial diffusivity (RD
DTI
),
and axial diffusivity (AxD
DTI
) –we also evaluated a modified measure of FA, derived from the tensor
distribution function (FA
TDF
) (Leow et al. 2009) which can be more sensitive to neurodegenerative
disease-related WM abnormalities than FA
DTI
across high- and low-angular resolution dMRI (Nir et al.
2017). The TDF model addresses well-established limitations of the standard single-tensor diffusion
model - which cannot resolve complex profiles of WM architecture such as crossing or mixing fibers,
present in up to 90% of WM voxels (Tournier et al. 2004, Descoteaux et al. 2007, Descoteaux et al. 2009,
Jeurissen et al. 2013).
In 24 WM regions of interest (ROIs), we ranked these five anisotropy and diffusivity indices, in
terms of their strength of association with key clinical measures, to identify dMRI indices that may help
understand and track AD progression. We hypothesized that the diffusion indices from ADNI2 (Nir et al.
2013, Nir et al. 2017) would still be associated with clinical measures of disease burden in ADNI3 - despite
the variation in protocols. We hypothesized that when data were pooled across ADNI3 protocols: (1)
91

higher diffusivity and lower anisotropy in the temporal lobe white matter would be most sensitive to
cognitive impairment, with highest effect sizes for associations with CDR-sob, and (2) FA
TDF
would detect
associations with clinical impairment with larger effect sizes than FA
DTI
.

3.2.2 METHODS

ADNI Participants

Baseline MRI, DTI, diagnosis, demographics, and cognitive measures were downloaded from the ADNI
database (https://ida.loni.usc.edu/). This analysis was performed when data collection for ADNI3 was still
ongoing (May 2018), and reflects the data available on April 30, 2018. Of the 381 participants scanned to
date, 55 were excluded after quality assurance: this included ensuring complete clinical and demographic
information, and image-level quality control (removing scans with severe motion, missing volumes, or
corrupt files). To ensure sufficient statistical power to assess differences in data collected with different
protocols, we evaluated only those protocols with complete available data for at least 10 participants at
the time of download; we did not assess protocol GE36, for which scans from 9 of 12 participants passed
quality assurance.
317 remaining participants –from 47 scanning sites - were included in the analysis (mean age:
75.4±7.9 yrs; 143 men, 174 women; Table 1): 211 were elderly cognitively normal controls (CN; mean
age: 74.5±7.3 yrs; 84 men, 127 women), 84 were diagnosed with mild cognitive impairment (MCI); mean
age: 76.3±8.1 yrs; 48 men, 36 women), and 22 were diagnosed with AD (mean age: 80.6±10.5 yrs; 11
men, 11 women). We note that two of the ADNI2 diagnostic categories - CN and Significant Memory
Concern (SMC) - are combined and identified as CN in ADNI3. ADNI2’s early and late MCI categories
are combined and identified as MCI in ADNI3.

Clinical Assessments

In addition to diagnosis, we indexed cognitive impairment using total scores from commonly used
screening tools for detecting dementia and AD (Table 1): the Alzheimer’s Disease Assessment Scale 13
(ADAS-cog), the Mini-Mental State Examination (MMSE), and the Clinical Dementia Rating scale sum-
of-boxes (CDR-sob). We refer to these tools as “cognitive measures”, but recognize the limitations of
92

these assessments as proxy measures of specific cognitive abilities (Balsis et al. 2015). The ADAS-cog is
frequently used in pharmaceutical trials, with scores ranging from 0-70; higher scores represent more
severe cognitive dysfunction (Rosen et al. 1984). MMSE is more often used by clinicians and researchers
in assessing cognitive aging. Scores for MMSE range from 0-30; lower scores typically indicate greater
cognitive dysfunction (Folstein et al. 1975). CDR-sob is used primarily in clinical trials and in clinical
practice for evaluating disease severity including the mild and early symptomatic stages of dementia. It is
calculated based on the sum of severity ratings in six domains (‘boxes’) - memory, orientation, judgment
and problem solving, community affairs, home and hobbies, and personal care. Scores range from 0 (no
dementia) to 3 (severe dementia) (Berg 1988). These evaluations are among the measures used in
diagnosing ADNI participants. Not all cognitive measures were available for every participant (MMSE,
N=315; CDR-sob, N=316, and ADAS-cog, N=278).

Table 1. Demographic and clinical measures for participants in ADNI3, subdivided by dMRI protocol. We report
the average age, MMSE, CDR-sob, and ADAS-cog measures, and their standard deviations.
Protocols Demographics
Clinical Assessments
Diagnosis Cognitive Measures
+

Name Total N Sites Age (yrs) Male CN MCI AD MMSE
*
CDR-sob
*
ADAS-cog
*

GE54 65 8 76.7 ± 7.3 32 (49.2%) 45 16 4 28.50 ± 3.26 0.78 ± 1.81 11.75 ± 6.81
P33 24 3 78.1 ± 7.1 13 (54.2%) 17 4 3 28.75 ± 2.03 1.31 ± 2.84 13.32 ± 6.76
P36 19 4 75.3 ± 6.6 7 (36.8%) 12 7 0 28.21 ± 2.39 0.76 ± 1.35 12.63 ± 5.12
S31 36 9 72.8 ± 8.6 15 (41.7%) 21 10 5 28.31 ± 2.77 0.79 ± 1.35 11.54 ± 5.25
S55 153 18 75.0 ± 8.4 66 (43.1%) 100 43 10 27.94 ± 3.28 0.95 ± 2.05 11.96 ± 5.65
S127 20 5 75.3 ± 5.4 10 (50.0%) 16 4 0 28.80 ± 1.70 0.33 ± 0.75 10.27 ± 2.83
TOTAL 317 47 75.4 ± 7.9 143 (45.1%) 211 84 22 28.23 ± 3.01 0.87 ± 1.91 11.89 ± 5.78
*
Data not available for all participants: MMSE N=315; CDR-sob N=316 and ADAS-cog N=278.
+
We recognize the limitations of these assessments as proxy measures of specific cognitive abilities (Balsis et al.
2015).

Diffusion MRI Acquisition Protocols

ADNI3 incorporated dMRI protocols for 3 tesla Siemens, Philips, and GE scanners. ADNI2, the first
phase of ADNI to include diffusion MRI, only prescribed dMRI protocols for GE scanners. The available
scanners span a wide range of software capabilities, such as support (or the lack of it) for custom diffusion
gradient tables and/or simultaneous multi-slice acceleration. Including additional scanners while staying
in a 7-10 minute scan duration resulted in data acquired with seven different acquisition protocols - of
which six had sufficient sample sizes to be evaluated here. Protocols varied in the number of diffusion
weighted imaging (DWI) directions (i.e., angular resolution), and the number of non-diffusion sensitized
gradients (b0 images), which serve as a reference to assess diffusion-related decay of the MR signal. Voxel
93

size across all ADNI3 protocols was 2.0x2.0x2.0 mm and 2.7x2.7x2.7 mm in ADNI2. Table 2 summarizes
the different protocols.
There is currently one multi-shell multiband protocol for Siemens Advanced Prisma scanners
(S127). As ADNI3 is still in its early stages, GE and Philips protocols for multi-shell acquisition have not
yet been finalized, so only 20 multi-shell scans were available for analysis at the time of writing. Here our
goal was to evaluate single-shell dMRI indices across protocols, so we used a subsample of the 127 DWI
volumes from the S127 multi-shell protocol to include only 13 b=0 and 48 b=1000 s/mm
2
DWI volumes
(removing 6 b=500 s/mm
2
and 60 b=2000 s/mm
2
volumes).
The Philips Basic Widebore R3 protocol (P36) included three b=2 s/mm
2
volumes and one b=0
s/mm
2
, because Philips scanners cannot acquire more than one b=0 s/mm
2
. The Philips Basic Widebore
(P33) was not a prescribed protocol, but rather acquired from Philips sites with a software version less
than 5.0 that could not acquire the b=2 s/mm
2
volumes.

Table 2. ADNI diffusion MRI acquisition protocols

Name Scanner Protocol
b 0
Volumes
DWI
Volumes
Total
Volumes
Time

(min)
Total
N
ADNI3
GE36 GE Basic Widebore 25x 4 b=0 s/mm
2
32 b=1000 s/mm
2
36 9:52 --
GE54 GE Basic 25x 6 b=0 s/mm
2
48 b=1000 s/mm
2
54 7:09 65
P33 Philips Basic Widebore 1 b=0 s/mm
2
32 b=1000 s/mm
2
33 7:32 24
P36 Philips Basic Widebore R3
1 b=0 s/mm
2

3 b=2 s/mm
2

32 b=1000 s/mm
2
36 6:54 19
P54 Philips Basic R5
1 b=0 s/mm
2

5 b=2 s/mm
2

48 b=1000 s/mm
2
54 8:05 --
S31 Siemens Basic VB17 1 b=0 s/mm
2
30 b=1000 s/mm
2
31 7:02 36
S55 Siemens
Basic Skyra E11 &
Prisma D13
7 b=0 s/mm
2
48 b=1000 s/mm
2
55 9:18 153
S127 Siemens Advanced Prisma VE11C 13 b=0 s/mm
2
48 b=1000 s/mm
2
61 7:25* 20
ADNI2
G46 GE
Discovery MR750 &
MR750w,
Signa HDx & HDxt
5 b=0 s/mm
2
41 b=1000 s/mm
2
46
7:00-
10:00
59
*Reflects the time to acquire the full multi-shell protocol (127 volumes), not the single-shell subset

dMRI Preprocessing and Scalar Indices

All DWI were preprocessed using the ADNI2 diffusion tensor imaging (DTI) analysis protocol as in Nir
et al. (2013). Briefly, we corrected for head motion and eddy current distortion, removed extra-cerebral
tissue, and registered each participant’s DWI to the respective T1-weighted brain to correct for echo planar
94

imaging (EPI) distortion. Details of the preprocessing steps may be found here:
https://adni.bitbucket.io/reference/docs/DTIROI/DTI-ADNI_Methods-Thompson-Oct2012.pdf. All DWI
and T1-weighted images were visually checked for quality assurance.
Scalar dMRI indices were derived from two reconstruction models: the single tensor model (DTI)
(Basser et al. 1994) and the tensor distribution function (TDF) (Leow et al. 2009). From the single tensor
model, FA
DTI
, AxD
DTI
, MD
DTI
, and RD
DTI
scalar maps were generated. In contrast to DTI, the TDF
represents the diffusion profile as a probabilistic mixture of tensors that optimally explain the observed
diffusion data, allowing for the reconstruction of multiple underlying fibers per voxel, together with a
distribution of weights, from which the TDF-derived form of FA (FA
TDF
) was calculated (Nir et al. 2017).

White Matter Tract Atlas ROI Summary Measures

ROI measures were generated as previously reported (Nir et al. 2013). Briefly, the FA image from the
Johns Hopkins University single subject Eve atlas (JHU-DTI-SS;
http://cmrm.med.jhmi.edu/cmrm/atlas/human_data/file/AtlasExplanation2.htm) was registered to each
participant’s corrected FA image using an inverse consistent mutual information based registration (Leow
et al. 2007); the transformation was then applied to the atlas WM parcellation map (WMPM) ROI labels
(as shown in Figure 7) (Mori et al. 2008) using nearest neighbor interpolation. Mean anisotropy and
diffusivity indices were extracted from 24 WM ROIs total (Table 3): 22 ROIs averaged bilaterally, the
full corpus callosum, and a summary across all ROIs (full WM).

Table 3. The following 24 ROIs from the JHU atlas (Mori et al. 2008) were analyzed.
CST Corticospinal tract SLF Superior longitudinal fasciculus
CP Cerebral peduncle SFO Superior fronto-occipital fasciculus
ALIC Anterior limb of internal capsule IFO Inferior fronto-occipital fasciculus
PLIC Posterior limb of internal capsule SS Sagittal stratum
RLIC Retrolenticular part of internal capsule EC External capsule
PTR Posterior thalamic radiation UNC Uncinate fasciculus
ACR Anterior corona radiata GCC Genu of corpus callosum
SCR Superior corona radiata BCC Body of corpus callosum
PCR Posterior corona radiata SCC Splenium of corpus callosum
CGC Cingulum (cingulate gyrus) CC Full corpus callosum
CGH Cingulum (hippocampal bundle) TAP Tapetum
Fx/ST Fornix (crus) / stria terminalis Full WM Full white matter

95

Comparing the ADNI2 and ADNI3 Protocols in Cognitively Normal Participants

Sample Sizes for the ADNI2 and ADNI3 Cognitively Normal Participants: We evaluated the six ADNI3
protocols and the ADNI2 protocol using scans from cognitively normal (CN) participants only. Of 85 CN
participants in ADNI2 with dMRI, 30 rolled over to ADNI3. To avoid duplication, and boost the number
of scans available for each protocol, we did not include all these roll-over participants in the ADNI3 group.
26 CN roll-over participants were included in the ADNI3 group. Four CN roll-over participants were
scanned with the S55 protocol, and due to the larger sample size already available for that protocol (N =
156), we included these four in the ADNI2 group. In total, 59 out of 85 ADNI2 CN participants were
included in the ADNI2 group and the remaining 26 were kept in the ADNI3 group for a total of 207
ADNI3 CN participants (see Supplementary Table S1 for CN demographics by ADNI phase and
protocol).

Assessing Age Effects: In CN participants, multivariate random-effects linear regressions were used to
assess whether dMRI indices from each ADNI protocol individually were associated with age, controlling
for sex and age*sex interactions as fixed variables, and acquisition site as a random variable. dMRI indices
for the CN group were subsequently pooled across ADNI3 protocols (N=207), or ADNI3 and ADNI2
protocols (N = 266) and tested for associations with age using an analogous model, but with protocol and
acquisition site as nested random variables (e.g., 8 sites used protocol GE54, and 3 sites used protocol
P33, so the acquisition site grouping variable is nested within the protocol grouping variable). We used
the false discovery rate (FDR) procedure to correct for multiple comparisons (q = 0.05) (Benjamini and
Hochberg 1995) across the 24 ROIs assessed for each dMRI index. Regions that survive a more stringent
Bonferroni correction at an alpha of 0.05 (p ≤ 0.05/24 = 0.0021) are also shown in the Supplements.

Effect of Protocol on dMRI Indices: In CN participants, we tested for significant differences in dMRI
indices between the seven ADNI protocols using analyses of covariance (ANCOVAs), adjusting for age,
sex, and age*sex interactions as fixed variables, and acquisition site as a random variable. For each dMRI
index, we used FDR to correct for multiple comparisons across the 24 ROIs assessed. Pairwise tests were
performed to directly compare protocols. In total, there were 504 tests per dMRI index: 24 ROIs * 21 pairs
of protocol comparisons (protocol 1 vs 2, protocol 1 vs 3, etc). As before, we used FDR to account for
multiple comparisons.
96

dMRI Harmonization with ComBat: ComBat uses an empirical Bayes framework to reduce unwanted
variation in multi-site data due to differences in acquisition protocol, while preserving the desired
biological variation in the data (Fortin et al. 2017). In the CN participants from ADNI2 and ADNI3, we
ran ComBat on each of the dMRI indices, including age, sex, age*sex, and information from all 24 ROIs
to inform the statistical properties of the protocol effects. Random-effects regressions tested for dMRI
microstructural associations with age, covarying for sex and age*sex as fixed variables and site as a
random variable; ANCOVAs and pairwise tests of dMRI differences between protocols were repeated for
the harmonized ROI data.

Clinical Assessments and their Relation to Pooled ADNI3 Diffusion Indices

Multivariate random-effects linear regressions were used to test associations between 5 dMRI indices in
each of the 24 WM ROIs and the three cognitive measures (ADAS, MMSE, CDR-sob), and with
diagnosis. Due to the limited available sample size of AD participants (N=22), and their uneven
distribution across the acquisition protocols tested here, we compared only groups of people with CN and
MCI diagnoses. Age, sex, and age*sex interactions were controlled for as fixed effects, and the protocol
and acquisition site were modeled as nested random variables. FDR was again used to correct for 24 ROI
tests (q = 0.05) (Benjamini and Hochberg 1995); Bonferroni corrections (p ≤ 0.05/24=0.0021) are
available in the Supplements. Effect sizes for associations were determined using the d-value standardized
coefficient (Rosenthal and Rosnow 1991):

𝑑 =
(2 ∗ 𝑇𝑣𝑎𝑙𝑢𝑒 )
√𝐷𝑒𝑔𝑟𝑒𝑒𝑠 𝑜𝑓 𝐹𝑟𝑒𝑒𝑑𝑜𝑚

3.2.3 RESULTS

Age Effects in Cognitively Normal Participants from ADNI2 and ADNI3 Protocols

When data were pooled across ADNI2 and ADNI3, significant associations with age were detected
throughout the WM. Figure 1a shows effect sizes for ROIs significantly associated with age after FDR
multiple comparisons correction (tabulated results and more stringent Bonferroni thresholds are shown in
97

Supplementary Table S2). Lower FA
TDF
and higher diffusivity indices were significantly associated with
older age in all 24 ROIs. For FA
DTI
, 22 ROIs were significantly associated with age. The largest effect
size was detected with FA
TDF
in the fornix (crus) / stria terminalis (Fx/ST; d = -1.459; p = 5.07x10
-21
).
The Fx/ST, genu of corpus callosum (GCC) and full WM consistently showed one of the 10 largest effect
sizes across dMRI indices.
The mean ages of the CN participants assessed in the two phases of ADNI were significantly
different (p = 0.049; ADNI2 mean age: 72.4±6.6 yrs; ADNI3 mean age: 74.5±7.4 yrs; demographics in
Supplementary Table S1). Pairwise tests comparing the mean age of CN participants scanned with each
protocol also showed significant differences between those scanned with S31 and two other protocols:
GE54 and S31 (p = 0.026); P33 and S31 (p = 0.0037). Due to differences in age and sample size between
protocols and phases, effect sizes could not be directly compared (Button et al. 2013), but the directions
of associations with age were largely consistent for ADNI2 and ADNI3 phases separately, and each
ADNI3 protocol (Figures 1-2). Each ADNI protocol showed directionally consistent associations in more
than 89% of tests (24 ROIs * 5 dMRI indices), except for P36 which was consistent in 81%, but had the
smallest sample size (N=12; Figure 2b). FA
TDF
and all three diffusivity indices were consistent in ≥ 96%
of tests (24 ROIs * 8 protocols/phases), while FA
DTI
was only consistent in 88% of tests. Most of the
associations detected in the unexpected direction for each protocol were driven by FA
DTI
. None of the
associations in the unexpected direction were significant after multiple comparisons correction, and only
2 had a p ≤ 0.05.
Figure 2 shows consistent associations in the full WM by protocol. As demographic and sample
size variability between protocols affect detected effect sizes, we also evaluated full WM dMRI
associations with age in an age- and sex-matched subset of 12 participants from each protocol (total N=84;
demographics in Supplementary Table S1). A comparison of the effect sizes between protocols suggests
that the protocols with greatest total number of diffusion-weighted (b=1000 s/mm
2
) and non-diffusion
sensitized (b0) gradients may detect larger effects (S127 followed by S55; Supplementary Figure S1).

Effect of Protocol on dMRI Indices from Cognitively Normal Controls

The influence of dMRI acquisition protocol on mean values of the diffusion indices is evident in boxplots
of dMRI indices in the full WM for each protocol (Figure 3). When modeling the mean full WM values
for each diffusion index, the residuals of the statistical model become more aligned after fitting the effect
98

of protocol and site (nested as a random variable with age, sex, and age*sex interactions as fixed effects)
than when we plot the residuals of just age, sex, and age*sex interactions.
ANCOVAs and pairwise tests for each ROI suggest there are significant differences between
protocols for all 5 dMRI indices across most ROIs (Figure 4). ANCOVAs revealed significant protocol
differences for 22 ROIs for FA
DTI
and FA
TDF
, with the highest overall effect size detected in the anterior
limb of the internal capsule (ALIC) and the external capsule (EC) for FA
DTI
(ALIC: d = 0.648; EC: d =
0.652). AxD
DTI
had the smallest effect size, overall, in the splenium of the corpus callosum (SCC; d =
0.106), and only 13 ROIs showed significant AxD
DTI
differences between protocols.
In pairwise analyses, AxD
DTI
was the most stable index across protocols, as significant protocol
differences were detected in only 20.6% of pairwise tests (24 ROIs * 21 pairwise tests), compared to
FA
DTI
, the most variable index, which showed significant protocol differences in 81.9% of tests (Figure
4b). ADNI2 was the most divergent protocol across dMRI indices, showing differences in 36.3% of tests.
99

Figure 1. (a) For each dMRI index, the absolute values of effect sizes (d-value) are plotted for regional WM
microstructural associations with age when all ADNI3 dMRI data are pooled, adjusting for any site or protocol
effects. For each test, we note the number of significant ROIs, as indicated by filled shapes, and the corresponding
FDR significance p-value threshold (q = 0.05). See Supplementary Table S2 for complete tabulated results. (b)
Here, we plot the residuals of diffusivity and anisotropy indices in the full WM (y-axis) against age (x-axis) after
regressing out the effects of sex in CN participants from each protocol separately. Individual level residuals from
each protocol are plotted with a different color. Despite protocol differences, age effects are evident across
protocols.

100

Figure 2. (a) Effect sizes (d-value) for each ADNI protocol and phase show the direction of dMRI associations
with age in the full WM are consistent. Due to differences in age and sample size between protocols and phases,
effect sizes could not be directly compared. (b) For each protocol and phase, the number of ROIs (out of 24), that
show the expected association direction, regardless of significance, are reported for each dMRI index; consistent
associations were detected across tests, except for protocol P36 which has the smallest sample size, and FA
DTI
,
which showed the smallest effect sizes and fewest significant associations across protocols when pooled.

101

Figure 3. Full WM mean (a) AxD
DTI
, MD
DTI
, and RD
DTI
and (b) FA
DTI
and FA
TDF
residuals for each protocol, after
fitting effects of age, sex, and age*sex interactions, are plotted here in the top rows (red). Protocol has an effect on
anisotropy and diffusivity measures. The lower panels (blue) show residuals after additionally fitting protocol and
site as nested random-effects, after which the residuals across protocols are more aligned.

102

Figure 4. (a) d-values from the ANCOVAs assessing differences in dMRI indices between protocols, for each of
the 24 ROIs; FA
DTI
showed the greatest significant differences (largest d-values; dark red) between protocols and
AxD
DTI
the fewest (dark green). (b) We report the number of times each protocol and each dMRI index showed
significant differences in pairwise tests between protocols (out of 504 tests per index and 720 tests per protocol);
AxD
DTI
was the most stable dMRI index across protocols, while FA
DTI
was the least stable.

103

Figure 5. Beta-values and error bars representing standard error from the association between each diffusion index
and age in CN participants, before and after ComBat harmonization. We show the three ROIs that consistently
showed one of the 10 largest effect sizes for associations with age across all five diffusion indices (see
Supplementary Figure S3 for all ROIs). Compared to pre-ComBat analyses, effect sizes are marginally different
across indices, but still within the standard error.

Diffusion MRI Harmonization with ComBat

After using ComBat to harmonize dMRI indices across protocols, ANCOVAs revealed that significant
protocol differences in dMRI indices were all but eliminated across ROIs (Supplementary Figure S2a);
significant protocol differences were detected only in the CST, for each of the dMRI indices. The number
of pairwise tests for which each protocol showed significant differences in dMRI indices decreased by
93.8% with ComBat (Supplementary Figure S2b).
After harmonization, we still detected significant associations between age and dMRI indices from
ADNI2 and ADNI3 pooled in the same number of ROIs (Supplementary Table S3). ComBat correction
did not significantly change effect sizes, while correcting for effects of protocol (Supplementary Figure
S3). In Figure 5 we show effect sizes before and after harmonization with ComBat in the full WM, Fx/ST,
and GCC, the three ROIs that consistently showed one of the 10 largest effect sizes for associations with
104

age across all five diffusion indices (for changes by protocol see Supplementary Figures S4-S6). As
harmonization with ComBat did not improve or change results found with random-effect linear
regressions, we proceeded to test clinical associations without applying the ComBat transformation.

Cognitive Measure Associations with Pooled ADNI3 dMRI Indices

Pooling data across ADNI3, we detected significant associations between all three cognitive measures and
regional dMRI indices throughout the WM. Greater cognitive impairment was associated with lower
anisotropy and higher diffusivity. Figure 6a-c shows effect sizes for ROIs significantly associated with
each cognitive measure after FDR multiple comparisons correction (for tabulated results and more
stringent Bonferroni corrections, please see Supplementary Tables S4-S6). Across tests (5 dMRI indices
* 3 cognitive measures), the hippocampal-cingulum (CGH), fornix (crus) / stria terminalis region (Fx/ST),
and the full WM consistently showed one of the 10 largest effect sizes (see Supplementary Figures S7-
S9 for associations with indices in the CGH, Fx/ST, and full WM, by protocol). In 14 of 15 tests, the CGH
consistently showed one of the top two largest effect sizes (CGH FA
DTI
association with CDR-sob was
the third largest), along with the uncinate fasciculus (UNC), which was top two in 12 of 15 tests (while
significant, cognitive associations with UNC FA
DTI
never showed one of the largest effect sizes).
FA
DTI
showed significant associations in the fewest ROIs: 55 out of 72 tests (76.4%; 24 ROIs * 3
cognitive measures) were significant. FA
TDF
showed more widespread associations with cognitive
measures throughout WM ROIs: 69 out of 72 tests (94.4%) were significant. Effect sizes were consistently
lower for FA
DTI
than for the other dMRI indices, across all 3 cognitive measures; the largest FA
DTI
effect
size was most consistently found in the Fx/ST, followed by the CGH or the GCC. The strongest FA
DTI

association overall was in the Fx/ST with CDR-sob (d = -0.681, p = 7.01x10
-8
). Compared to FA
DTI
,
FA
TDF
showed larger effect sizes; across cognitive tests, the strongest FA
TDF
associations were detected in
the uncinate fasciculus (UNC) with CDR-sob (d = -1.244; p = 1.39x10
-20
), followed by the CGH (d = -
1.213; p = 8.86x10
-20
). CDR-sob effect sizes for FA
DTI
and FA
TDF
in the CGH, UNC, Fx/ST, and full WM
are depicted by protocol in Supplementary Figure S10, revealing consistently larger effect sizes for
FA
TDF
across protocols.
Cognitive associations with all of the diffusivity indices were widespread: significant associations
were detected in 207 out of 216 tests (95.8%; 24 ROIs * 3 cognitive measures * 3 diffusivity indices).
Regional measures of AxD
DTI
consistently showed the largest effect sizes across all cognitive measures
105

(CDR-sob and the UNC: d = 1.344, p = 3.13x10
-23
; MMSE and the CGH: d = -1.178, p = 7.87x10
-19
;
ADAS-cog and the UNC: d = 1.048, p = 1.09x10
-13
).
Of the three cognitive measures, CDR-sob associations showed the largest effect sizes across
dMRI indices (in the UNC followed by the CGH for all indices except FA
DTI
); the largest effect sizes
across all tests were detected with AxD
DTI
(UNC: d = 1.344) and MD
DTI
(UNC: d = 1.342, p = 3.47x10
-
23
). Figure 7 shows the distribution of the effect sizes for CDR-sob throughout the brain. Temporal lobe
regions (UNC, CGH, IFO, SS) frequently showed greatest effect sizes (for ADAS-cog and MMSE figures,
see Supplementary Figures S11-S12). Effect size was not correlated with ROI size (Supplementary
Figure S13), consistent with prior studies of other disorders advance (Kelly et al. 2018).

Figure 6. For each dMRI index, the absolute values of effect sizes (d-value) are plotted for regional WM
microstructural associations with clinical measures. Lower anisotropy and higher diffusivity were significantly
associated with (a) higher CDR-sob, (b) lower MMSE, (c) higher ADAS-cog, and (d) an MCI diagnosis, when all
ADNI3 dMRI data are pooled, adjusting for any site or protocol effects. For each test, we note the number of
significant ROIs, as indicated by filled shapes, and the corresponding FDR significance p-value threshold (q = 0.05).
See Supplementary Tables 4-7 for complete tabulated results.
106

Figure 7. Effect size (absolute d-value) maps of WM regions that show significant associations with CDR-sob - the
cognitive measure with the largest effect sizes - reveal widespread associations throughout the WM, with
particularly strong associations in the temporal lobes (SS, IFO, UNC, and CGH; light green regions show the largest
effect sizes). As expected, positive associations were detected between CDR-sob and (a) AxD
DTI
(FDR critical
threshold p = 1.78x10
-4
) (b) MD
DTI
(FDR critical threshold p = 3.64x10
-4
) and (c) RD
DTI
(FDR critical threshold p
= 6.92x10
-3
); higher diffusivity was associated with greater cognitive impairment. Lower (d) FA
DTI
(FDR critical
threshold p = 0.025) and (e) FA
TDF
(FDR critical threshold p = 7.73x10
-3
) were also associated with greater
impairment, but FA
DTI
associations were detected in fewer regions with weaker effect sizes compared to FA
TDF
.

CN vs MCI Diagnosis Associations with Pooled ADNI3 dMRI Indices

For each diffusion index, Figure 6d shows the significant regional effect sizes for differences between
CN and MCI participants. Widespread diffusivity differences were detected, with significantly higher
diffusivity in MCI participants in 21 out of 24 ROIs (Supplementary Table S7 and Supplementary
Figure S14). Only three regions showed significantly lower FA
DTI
in MCI participants – Fx/ST (d =
-0.460; p = 3.89x10
-4
), CGH (d = -0.410; p = 1.53x10
-3
), and the posterior thalamic radiation (PTR; d =
0.367; p = 4.55x10
-3
). On the other hand, FA
TDF
was significant in 20 out of 24 ROIs, similar to diffusivity
107

indices. FA
TDF
and diffusivity indices in the CGH showed the largest effect sizes overall (AxD
DTI
d =
0.681; p = 2.26x10
-7
, MD
DTI
d = 0.700; p = 1.15x10
-7
; RD
DTI
d = 0.679; p = 2.41x10
-7
; FA
TDF
d = -
0.622; p = 2.00x10
-6
).
For all three cognitive measures, and in the comparison between CN and MCI participants, the
CGH and Fx/ST were the only regions that survived multiple comparisons correction across all dMRI
indices. The Fx/ST always had the largest effect size in FA
DTI
tests. The UNC showed either the first or
second largest effect size (alternating with CGH) across diffusivity indices and FA
TDF
tests, but was
significant only for cognitive measure associations with FA
DTI
(i.e., 3 of 4 clinical tests).

3.2.4 DISCUSSION

This study has three main findings: (1) When data were pooled from the six available diffusion MRI
protocols used in ADNI3, anisotropy and diffusivity indices showed robust associations with MCI
diagnosis, and with three common cognitive measures: MMSE, ADAS-cog, and CDR-sob; (2) When
using a higher-order diffusion model, the derived measure of anisotropy (FA
TDF
) showed stronger and
more widespread associations with clinical impairment than the standard DTI anisotropy measure (FA
DTI
);
(3) Despite significant differences in protocols, for each dMRI index, we were able to detect consistent
associations with clinical measures in ADNI3 participants, and age in ADNI2 and ADNI3 CN participants.
Accumulation of amyloid plaques and neurofibrillary tangles (NFT) in the brain (Braak and Braak
1991, 1996, Frank et al. 2003, Shaw et al. 2007) can directly impact WM (Lee et al. 2004, Roth et al.
2005), promoting myelin degeneration and axonal loss (Braak and Braak 1996, Kneynsberg et al. 2017).
While many factors drive anisotropy and diffusivity measures from DTI, higher anisotropy values may
indicate, in part, more coherent intact axons, while lower anisotropy and higher diffusivity may reflect
factors such as axonal injury and demyelination, among other factors (Beaulieu 2002, Song et al. 2003,
Song et al. 2005, Harsan et al. 2006, Le Bihan and Johansen-Berg 2012, Kantarci et al. 2017, Moore et al.
2018). In this paper, lower anisotropy values and higher diffusivity values were correlated with clinical
impairment most strongly in the hippocampal-cingulum and uncinate fasciculus. Along with the full WM,
reflecting global WM effects, the largest effect sizes were most frequently detected in the hippocampal-
cingulum and fornix (crus) / stria terminalis, WM bundles connecting hippocampal and parahippocampal
regions to the rest of the brain, consistent with patterns of AD pathology. The histopathological validity
of these findings has been supported, specifically in a recent study that compared NFT stages in autopsy
108

material along with ante-mortem MRI. Elevated MD
DTI
and lower FA
DTI
significantly correlated with
higher postmortem NFT stage, particularly in the crus of the fornix, the ventral cingulum tracts, the
precuneus, and entorhinal WM (Kantarci et al. 2017).
The participants recruited for ADNI3 tend to be younger and healthier, on average, than those in
ADNI2, as they were recruited with the intention of studying the transition from CN to AD (Jack et al.
2015). With few AD patients enrolled so far in ADNI3, the primary focus of this paper was to assess three
cognitive assessments (ADAS-cog, CDR-sob, and MMSE), and to compare CN to MCI participants. MCI
is now the focus of intense research; it is essential to find ways to clinically categorize the transitional
stages between normal aging and AD to evaluate targeted treatments, as pathophysiological mechanisms
may differ or change throughout the course of AD (Mueller et al. 2005). As in our prior analysis of ADNI2
(Nir et al. 2013), FA
DTI
was the least sensitive DTI measure. In ADNI3, AxD
DTI
and MD
DTI
showed the
largest effect sizes. Lower FA
DTI
and higher MD
DTI
are most frequently reported in studies of AD (Kavcic
et al. 2008, Clerx et al. 2012, Nir et al. 2013, Maggipinto et al. 2017, Mayo et al. 2017), but AxD
DTI
may
be more sensitive to unspecific microscopic cellular loss earlier in the disease (O'Dwyer et al. 2011),
perhaps making it more sensitive in the healthier participants of the ADNI3 dataset. Similarly, in ADNI2,
AxD
DTI
was the most sensitive to differences between CN and MCI diagnosis (Nir et al. 2013).
Among the three cognitive assessments, CDR-sob showed the strongest correlations with dMRI
indices, in line with prior ADNI brain imaging studies (Hua et al. 2009a, Nir et al. 2013). The largest of
these effects were found in temporal WM tracts including the hippocampal-cingulum, uncinate fasciculus,
sagittal stratum, and inferior fronto-occipital fasciculus. These are all regions that show early degenerative
changes in MCI and AD (Mielke et al. 2009, Nir et al. 2013, Maggipinto et al. 2017, Powell et al. 2018).
While associations with clinical impairment were detected throughout the WM, the region that most
frequently showed the lowest effect sizes and was significant in only 3 of the 20 clinical tests, was the
corticospinal tract (CST). However, the CST ROI from the JHU WMPM atlas is limited to a small region
in the inferior portion of the brain and has been shown to be the least reliable and reproducible ROI
(Jahanshad et al. 2013a, Acheson et al. 2017) suggesting alternate approaches, such as tractography-based
evaluations (Jin et al. 2017), or the use of the probabilistic JHU atlas (Hua et al. 2008), may be more
appropriate for studying the CST. Our analysis focused on white matter microstructure, but future work
assessing tract geometry and properties of anatomical brain networks using tractography may reveal more
detailed information. The validation and harmonization of tractography methods and derived network
metrics is a vast field of research with active ongoing work (Maier-Hein et al. 2017).
109

DTI is widely recognized as a useful tool for studying neurodegenerative disorders such as AD
(Oishi et al. 2011, Müller and Kassubek 2013, Abhinav et al. 2014, Acosta-Cabronero and Nestor 2014,
Maggipinto et al. 2017). However, at the spatial resolutions now used, a single voxel typically captures
partial volumes of different tissue compartments, e.g., the intra- and extra-cellular compartments, the
vascular compartment, the CSF and myelin; each affects water diffusion and the MR signal. The DTI
model cannot differentiate these components or even crossing fibers (Tuch et al. 2002, Jbabdi et al. 2010),
which are estimated to occur in up to 90% of WM voxels at the typical dMRI resolution (Descoteaux et
al. 2009, Jeurissen et al. 2013). In healthy tissue with crossing fibers, the DTI model may show low FA.
FA
DTI
may paradoxically appear to increase in regions where crossing fibers deteriorate in
neurodegenerative diseases such as AD (Douaud et al. 2011). FA
TDF
addresses this limitation even in low
angular resolution data (Nir et al. 2017). Here, compared to FA
DTI
, FA
TDF
showed more widespread
associations with cognitive measures and diagnosis throughout WM ROIs: FA
TDF
was significant in 89 of
the 96 tests (92.7%; 24 ROIs * 4 clinical tests), while FA
DTI
was only significant in 58 (60.4%). The
greatest difference was seen for diagnostic associations (CN vs MCI): FA
TDF
was significant in 20 out of
24 ROIs while FA
DTI
was only significant in 3. FA
TDF
also showed stronger effect sizes across the
protocols, suggesting that tensor limitations have likely confounded previous diffusion studies of cognitive
decline that have found little or no effects with FA (Acosta-Cabronero et al. 2010). Recently proposed
biophysical models of brain tissue may help to relate diffusion signals directly to underlying
microstructure and different tissue compartments (Harms et al. 2017). We may be able to further
disentangle questions of orientation coherence (dispersing and ‘kissing’ fibers), fiber diameter, fiber
density, membrane permeability, and myelination, which all influence classic anisotropy and diffusivity
measures derived from DTI. Several AD studies have already used multi-shell protocols to compute
diffusion indices from models that do not assume mono-exponential decay, such as diffusion kurtosis
imaging (DKI) (Jensen et al. 2005, Chen et al. 2017, Cheng et al. 2018, Wang et al. 2018a), and multi-
compartment models such as neurite orientation dispersion and density imaging (NODDI) (Zhang et al.
2012, Colgan et al. 2016, Slattery et al. 2017, Parker et al. 2018). To date, approximately 20 participants
in ADNI have been scanned with multi-shell diffusion protocols; in a future report, we will relate multi-
shell measures to those examined here.
Large-scale, multi-site neuroimaging studies can increase the power of statistical analyses and
establish greater confidence and generalizability for findings. Most multi-site neuroimaging studies are
susceptible to variability across sites. Variability in dMRI studies is due in part to heterogeneity in
110

acquisition protocols, scanning parameters, and scanner manufacturers (Zhu et al. 2009, Zhu et al. 2011,
Zhu et al. 2018). Anisotropy and diffusivity maps are affected by angular and spatial resolution (Alexander
et al. 2001, Kim et al. 2006, Zhan et al. 2010), the number of DWI directions (Giannelli et al. 2009), and
the number of acquired b-values (Correia et al. 2009). All five dMRI indices were significantly different
between protocols; AxD
DTI
was the most stable index, while FA
DTI
was the least stable, reflective of their
performance in detecting associations with cognitive measures. ADNI2 was the most divergent protocol
across dMRI indices, perhaps due to the larger voxel size in ADNI2 (2.7 mm
3
versus 2.0 mm
3
isotropic
voxels used in ADNI3). This is consistent with the notion that DTI measures vary with voxel size due to
partial voluming (Zhan et al. 2013a). Despite differences in protocols, the directions of associations were
consistent across protocols.
ADNI3 extends dMRI acquisitions across scanner manufacturers and platforms to maximize the
number of participants scanned with dMRI; this makes it necessary to account for site-related
heterogeneities and confounds in analytical models where data are pooled. Multi-site dMRI studies are
becoming increasingly common, and new data harmonization methods to adjust for site and acquisition
protocol are being developed and tested. A thorough investigation of dMRI harmonization methods is now
possible with ADNI3, one of the few publically available multi-site datasets acquired with multiple
protocols. As regional dMRI measures are available for download as part of the ADNI database, we
highlight two ways that the data may be pooled across sites: 1) performing statistical analyses with nested
random-effects models to account for site and acquisition protocol differences, and 2) harmonizing the
derived regional measures before aggregating the data across sites. In a preliminary analysis, we showed
that one harmonization method performed on these regional measures, ComBat, reduced cross-site
differences in dMRI indices, while preserving biological relationships with age in CN controls. The only
region where differences remained after ComBat, was the CST, the ROI with the weakest associations
with clinical measures, and previously identified as least reliable (Acheson et al. 2017). In Fortin et al.
(2017), compared to other harmonization methods, ComBat increased the number of voxels where
significant associations between age and FA
DTI
or MD
DTI
were detected. Here, the number of significant
ROIs and the magnitude of effect sizes were comparable for ComBat and nested random-effects model
approaches. This discrepancy between our findings and those of Fortin et al., may be due to several
differences between studies: 1) ADNI3 includes more sites and protocols, 2) in contrast to the number of
voxels, the number of ROIs is far less than the number of participants, and 3) the age effects in the elderly
populations tested here are stronger than the effects tested in adolescents in Fortin et al. When effects are
111

more readily detected, one harmonization approach may not be more advantageous than others. In addition
to exploring additional harmonization techniques, future work should evaluate voxel-wise ComBat
approaches and the effects of harmonization beyond CN participants (i.e., across the entire ADNI cohort).
In addition to ComBat, a number of harmonization approaches have recently been proposed at
various stages of analysis (Tax et al. 2018, Zhu et al. 2018). Site differences can be accounted for at the
time of overall group inference, such as with the random-effects regression level correction used here, or
by using a meta-analysis approach in lieu of pooling data (Thompson et al. 2014). The data may also be
transformed prior to multi-site group-level statistics. Some methods, such as ComBat and RAVEL, use
the distribution of derived features, such as diffusivity and anisotropy measures (Fortin et al. 2016, Fortin
et al. 2017). Alternatively, several proposed methods use information from the raw image to adjust for
acquisition variability (Zhu et al. 2018). For example, Kochunov et al. (2018) calculated the signal to
noise ratio for each protocol and include it in their regression models. Mirzaalian et al. (2018) used voxel-
wise spherical harmonic residual networks to derive local correction parameters. Finding the best method
to harmonize dMRI data is an active topic at ‘hackathons’ and technical challenges; in 2017 and 2018, the
International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI)
hosted a computational diffusion MRI challenge to explore approaches for data harmonization. With so
many available approaches, the preliminary random-effects regression and ComBat results from this paper
serve as a first step towards future work establishing robust approaches for combining data in ADNI3 and
other multi-site studies.
The current study is limited in that the sample sizes and sample demographics available for each
protocol vary, complicating direct comparison of the protocols (Button et al. 2013). A matched
comparison might be possible if a group of participants or a phantom were scanned using every protocol.
Even so, separating protocol differences from differences in scanner manufacturer is difficult. We also
could not directly compare all diagnostic groups in ADNI3, as few participants with AD were scanned.
A more complete picture of brain changes in aging and AD would include imaging metrics from
other modalities, such as perfusion imaging, resting state functional MRI (Wang et al. 2017), and
radiotracer methods such as FDG-PET (Popuri et al. 2018), or amyloid- and tau-sensitive PET (Grothe et
al. 2017, Phillips et al. 2018). Genetic and other ‘omics’ data could be analyzed as well, and may help to
predict diagnostic classification and brain aging, when combined with other neuroimaging markers (Ding
et al. 2018, Kauppi et al. 2018). While these data are all being collected as part of ADNI3 and other studies
of brain aging, our focus here was on the variety of available dMRI measures, calculated using different
112

protocols. With this in mind, the optimal dMRI indices to include in a multimodal study may be those that
contribute the greatest independent information beyond that available from anatomical MRI and other
standard imaging modalities. Multivariate methods –such as machine learning (Zhou et al. 2017, Wang et
al. 2018b) and even deep learning (Liu et al. 2017) –may also help to extract and capitalize on features
that predict clinical decline beyond those studied here.
In addition to providing a roadmap for the new ADNI3 dMRI data, these preliminary analyses
show that despite differences in the updated dMRI protocols, diffusion indices can be pooled to detect
white matter microstructural differences associated with aging and Alzheimer’s disease.

3.2.5 ACKNOWLEDGEMENTS

Data collection and sharing for ADNI was funded by National Institutes of Health Grant U01 AG024904
and the DOD (Department of Defense award number W81XWH-12-2-0012). Additional support was
provided by NIA grant RF1 AG04191, P01 AG026572-13, R56AG058854, RF1AG051710 and P41
EB015922. ADNI is funded by the National Institute on Aging, the National Institute of Biomedical
Imaging and Bioengineering, and through generous contributions from the following: AbbVie,
Alzheimer’s Association; Alzheimer’s Drug Discovery Foundation; Araclon Biotech; BioClinica, Inc.;
Biogen; Bristol-Myers Squibb Company; CereSpir, Inc.; Cogstate; Eisai Inc.; Elan Pharmaceuticals, Inc.;
Eli Lilly and Company; EuroImmun; F. Hoffmann-La Roche Ltd and its affiliated company Genentech,
Inc.; Fujirebio; GE Healthcare; IXICO Ltd.; Janssen Alzheimer Immunotherapy Research &
Development, LLC.; Johnson & Johnson Pharmaceutical Research & Development LLC.; Lumosity;
Lundbeck; Merck & Co., Inc.; Meso Scale Diagnostics, LLC.; NeuroRx Research; Neurotrack
Technologies; Novartis Pharmaceuticals Corporation; Pfizer Inc.; Piramal Imaging; Servier; Takeda
Pharmaceutical Company; and Transition Therapeutics. The Canadian Institutes of Health Research is
providing funds to support ADNI clinical sites in Canada. Private sector contributions are facilitated by
the Foundation for the National Institutes of Health (www.fnih.org). The grantee organization is the
Northern California Institute for Research and Education, and the study is coordinated by the Alzheimer’s
Therapeutic Research Institute at the University of Southern California. ADNI data are disseminated by
the Laboratory for Neuro Imaging at the University of Southern California. Samples from the National
Centralized Repository for Alzheimer's Disease and Related Dementias (NCRAD), which receives
government support under a cooperative agreement grant (U24 AG21886) awarded by the National
113

Institute on Aging (NIA), were used in this study. We thank contributors who collected samples used in
this study, as well as patients and their families, whose help and participation made this work possible.

3.2.6 SUPPLEMENTARY APPENDIX

A Demographic Data for Cognitively Normal Controls

Supplementary Table S1. Demographics for ADNI2 and ADNI3 cognitively normal (CN) participants,
by protocol.

Protocol
All Matched
Total
N
Age
Mean ± SD
Male
N (%)
Total
N
Age
Mean ± SD
Male
N (%)
ADNI3
GE54 45 75.8±7.0 20 (44.4) 12 76.4±3.8 4 (33.3)
P33 17 78.5±7.3 9 (52.9) 12 77.0±6.3 4 (33.3)
P36 12 75.9±6.4 4 (33.3) 12 75.9±6.4 4 (33.3)
S31 21 71.6±7.0 8 (38.1) 12 75.3±6.9 4 (33.3)
S55 96 73.5±7.6 36 (37.5) 12 75.6±3.5 4 (33.3)
S127 16 75.2±6.0 6 (37.5) 12 76.5±4.8 4 (33.3)
Total 207 74.5±7.4 83 (40.1) 72 76.1±5.3 24 (33.3)
ADNI2 GE46 59 72.4±6.6 24 (40.7) 12 76.5±3.8 4 (33.3)
Total 266 74.0±7.3 107 (40.2) 84 76.2±5.1 28 (33.3)

114

B. ADNI2 and ADNI3 Pooled dMRI ROI Associations with Age

Supplementary Table S2. P-values and corresponding effect sizes (d-values) are reported for associations between
age and 5 dMRI anisotropy and diffusivity indices in pooled ADNI2 and ADNI3 CN participants (N=266). For
each test, the ROIs are ordered by d-value. Regions that were significant after FDR (q = 0.05) or Bonferroni (α =
0.05) multiple comparisons correction are delineated by a dotted or solid line respectively.

115

C. dMRI Associations with Age by Protocol in Age and Sex Matched Subsets of N=12

Supplementary Figure S1. Effect sizes (d-values) from 12 CN sex- and age-matched CN controls from each ADNI
protocol (Supplementary Table S1) show the direction of dMRI associations with age in the full WM were
consistent across protocols. While direct comparisons of effect sizes were underpowered, findings suggested larger
effect sizes for protocol S127, the protocol with greatest total number of diffusion-weighted (b = 1000 s/mm
2
) and
non-diffusion sensitized (b 0) gradients, followed by S55, the protocol with the second greatest number of diffusion-
weighted and b 0 gradients.

116

D. Protocol Differences in dMRI Indices After ComBat

Supplementary Figure S2. (a) d-values from the ANCOVAs assessing differences in dMRI indices between
protocols, for each of the 24 ROIs, after ComBat harmonization; across dMRI indices, the CST was the only region
where significant protocol differences remain. (b) We report the number of times each protocol and each dMRI
index showed significant differences in pairwise tests between protocols after ComBat harmonization (out of 504
tests per index and 720 tests per protocol). After ComBat, the number of pairwise tests for which each protocol
showed significant differences in dMRI indices decreased by 93.8%.

117

E. Pooled Associations with Age after ComBat

Supplementary Table S3. P-values and corresponding effect sizes (d-values) are reported for ROI associations
between age and 5 dMRI anisotropy and diffusivity indices in ADNI2 and ADNI3 protocols pooled after
harmonization with ComBat. For each test, the ROIs are ordered by d-value. Regions that were significant after
FDR (q = 0.05) or Bonferroni (α = 0.05) multiple comparisons correction are delineated by a dotted or solid line
respectively.

118

Supplementary Figure S3. Effect sizes (beta-values with error bars that represent the standard error) are plotted
for the association between each diffusion index and age in CN controls from ADNI2 and ADNI3 protocols pooled
together before and after ComBat harmonization. Compared to pre-ComBat analyses, effect sizes were marginally
different across indices, but still within the standard error bounds. All associations were significant (FDR q = 0.05)
except for FA
DTI
in the CST and PCR.

119

F. dMRI ROI Associations with Age after ComBat by Protocol

Supplementary Figure S4. For each protocol, the beta-values (error bars represent the standard error) are plotted
for the association between each diffusion index in the full WM and age in CN controls, before and after ComBat
harmonization. Compared to pre-ComBat analyses, effect sizes were marginally different across indices, but still
within the standard error bounds.

120

Supplementary Figure S5. For each protocol, the beta-values (error bars represent the standard error) are plotted
for the association between each diffusion index in the fornix (crus) / stria terminalis (Fx/ST) and age in CN
controls, before and after ComBat harmonization. Compared to pre-ComBat analyses, effect sizes were marginally
different across indices, but still within the standard error bounds.

121

Supplementary Figure S6. For each protocol, the beta-values (error bars represent the standard error) are plotted
for the association between each diffusion index in the corpus callosum genu (GCC) and age in CN controls, before
and after ComBat harmonization. Compared to pre-ComBat analyses, effect sizes were only marginally different
across indices and protocols, but still within the standard error bounds.

122

G. ADNI3 Pooled Regional dMRI Measures: Associations with Cognitive Measures

Supplementary Table S4. P-values and corresponding effect sizes (d-values) are reported for ROI associations
between CDR-sob and dMRI indices across all pooled ADNI3 participants (N=316). For each test, the ROIs are
ordered by d-value. Regions that were significant after FDR (q=0.05) or Bonferroni (α=0.05) multiple comparisons
correction are delineated by a dotted or solid line respectively.

123

Supplementary Table S5. P-values and corresponding effect sizes (d-values) are reported for ROI associations
between ADAS-cog diagnosis and dMRI indices across all ADNI3 participants pooled together (N=278). For each
test, the ROIs are ordered by d-value. Regions that were significant after FDR (q=0.05) or Bonferroni (α=0.05)
multiple comparisons correction are delineated by a dotted or solid line respectively.

Supplementary Table S6. P-values and corresponding effect sizes (d-values) are reported for ROI associations
between MMSE and dMRI indices across all ADNI3 participants pooled together (N=315). For each test the ROIs
are ordered by d-value. Regions that were significant after correcting for multiple comparisons (FDR q=0.05) are
delineated by a dotted line.

124

H. dMRI Associations with Cognitive Measures by Protocol

Supplementary Figure S7. For each of six ADNI3 protocols individually and pooled, effect sizes (d-values) are
shown for associations between cognitive scores or diagnosis and dMRI indices in the full WM. We note that due
to differences in sample size between protocols, effect sizes should not be directly compared. However, the direction
of associations was consistent across protocols.

125

Supplementary Figure S8. For each of six ADNI3 protocols individually and pooled, effect sizes (d-values) are
shown for associations between cognitive scores or diagnosis and dMRI indices in the CGH—one of two ROIs that
consistently showed significant associations across all four clinical tests and dMRI indices. We note that due to
differences in sample size between protocols, effect sizes should not be directly compared. However, the direction
of associations was consistent across protocols.

126

Supplementary Figure S9. For each of six ADNI3 protocols individually and pooled, effect sizes (d-values) are
shown for associations between cognitive scores or diagnosis and dMRI indices in the Fx/ST—one of two ROIs
that consistently showed significant associations across all four clinical tests and dMRI indices. We note that due
to differences in sample size between protocols, effect sizes should not be directly compared. However, the direction
of associations was consistent across protocols.

127

Supplementary Figure S10. A comparison of FA
DTI
and FA
TDF
effect sizes (beta-values and standard error) for
CDR-sob associations in the CGH, UNC, Fx/ST, and full WM, for each of six ADNI3 protocols individually and
pooled; FA
TDF
consistently detected larger effect sizes across protocols.

128

I. Brain Maps of dMRI Associations with Cognitive Measures

Supplementary Figure S11. Effect size (absolute d-value) maps of WM regions that showed significant
associations with ADAS-cog (FDR q = 0.05). Associations were detected between ADAS-cog and (a) AxD
DTI
(b)
MD
DTI
and (c) RD
DTI
, where higher diffusivity was significantly associated with greater cognitive impairment.
Lower (d) FA
DTI
and (e) FA
TDF
were associated with greater impairment, but FA
DTI
associations were detected in
fewer regions with weaker effect sizes compared to FA
TDF
. Light green regions show the largest effect sizes.

129

Supplementary Figure S12. Effect size (absolute d-value) maps of WM regions that showed significant
associations with MMSE (FDR q = 0.05). Negative associations were detected between MMSE scores and (a)
AxD
DTI
(b) MD
DTI
and (c) RD
DTI
, where higher diffusivity was associated with lower MMSE (greater cognitive
impairment). Lower (d) FA
DTI
and (e) FA
TDF
were associated with lower MMSE, but FA
DTI
associations were
detected in fewer regions with weaker effect sizes compared to FA
TDF
. Light green regions show the largest effect
sizes.

130

J. ROI Size vs CDR-sob Effect Size

Supplementary Figure S13. CDR-sob effect sizes (d-values) for each ROI are ordered by ROI size (full WM, far
left, is the largest ROI and UNC, far right, is the smallest). Effect size and the square root of ROI size (number of
voxels) were not correlated (FA
DTI
: Pearson’s r = -1.1, p = 0.58; AxD
DTI
: r = 0.064 p = 0.77; MD
DTI
: r = 0.044, p =
0.84; RD
DTI
: r = -0.19, p = 0.93; FA
TDF
: r = -0.072, p = 0.74).

131

K. Regional dMRI Measures: Pooled Associations with Diagnosis

Supplementary Table S7. P-values and corresponding effect sizes (d-values) are reported for WM microstructural
differences between CN and MCI participants when all ADNI3 dMRI data are pooled together. For each test, the
ROIs are ordered by d-value. Regions that were significant after FDR (q = 0.05) or Bonferroni (α = 0.05) multiple
comparisons correction are delineated by a dotted or solid line respectively.

132

Supplementary Figure S14. Effect size (absolute d-value) maps of WM regions that showed significant differences
between CN participants and those with MCI (FDR q = 0.05). Positive associations were detected between MCI
diagnosis and (a) AxD
DTI
(b) MD
DTI
and (c) RD
DTI
, where higher diffusivity was associated with greater cognitive
impairment. Lower (d) FA
DTI
and (e) FA
TDF
were associated with greater impairment, but FA
DTI
associations were
detected in fewer regions with weaker effect sizes compared to FA
TDF
. Light green regions show the largest effect
sizes.

133

3.3 Pooling and Harmonizing Multi-site Diffusion MRI Data in an International
Multi-Cohort HIV Study

This section is adapted from:

Nir TM, Lam HY, Ananworanich J, Boban J, Brew BJ, Cysique L, Fouche JP, Kuhn T, Porges ES, Law
M, Paul R, Thames A, Woods AJ, Valcour VG, Thompson PM, Cohen RA, Stein DJ, Jahanshad N, for
the ENIGMA-HIV Working Group (2018). Effects of diffusion MRI model and harmonization on the
consistency of findings in an international multi-cohort HIV neuroimaging study. 2018 MICCAI
Workshop on Computational Diffusion MRI (CDMRI).

134

Effects of Diffusion MRI Model and Harmonization on the Consistency of Findings in an
International Multi-Cohort HIV Neuroimaging Study

Talia M. Nir
1
, Hei Y. Lam
1
, Jintanat Ananworanich
2
, Jasmina Boban
3
, Bruce J. Brew
4
, Lucette Cysique
4
,
J.P. Fouche
5
, Taylor Kuhn
6
, Eric S. Porges
7
, Meng Law
8
, Robert H. Paul
9
, April Thames
6,10
, Adam J. Woods
7
,
Victor G. Valcour
11
, Paul M. Thompson
1
, Ronald A. Cohen
7,12
, Dan J. Stein
5
, Neda Jahanshad
1
for the ENIGMA-HIV Working Group

1
Imaging Genetics Center, Mark and Mary Stevens Neuroimaging and Informatics Institute, University of Southern California, Marina del
Rey, CA, USA
2
HIV-NAT, Thai Red Cross AIDS Research Centre, Bangkok, Thailand
3
Faculty of Medicine, Diagnostic Imaging Center, University of Novi Sad, Novi Sad, Serbia
4
Department of Neurology, St Vincent’s Hospital, and University of New South Wales, Sydney, Australia
5
Department of Psychiatry and Mental Health, University of Cape Town, Cape Town, South Africa
6
Semel Institute for Neuroscience and Human Behavior, University of California, Los Angeles, CA, USA
7
Institute on Aging, Department of Aging and Geriatric Research, School of Medicine, University of Florida, Gainesville, FL, USA
8
Department of Radiology, Keck School of Medicine, University of Southern California, Los Angeles, CA, USA
9
Missouri Institute of Mental Health, University of Missouri in Saint Louis, Saint Louis, MO, USA
10
Department of Psychology, University of Southern California, Los Angeles, CA, USA
11
Memory and Aging Center, Department of Neurology, University of California, San Francisco, San Francisco, CA, USA
12
Department of Psychiatry and Human Behavior, Warren Alpert Medical School, Brown University, Providence, RI, USA

Abstract. Clinical and demographic heterogeneity of HIV-infected individuals world-wide, and
variations in diffusion MRI (dMRI) acquisition, processing, and analysis methods have led to
inconsistency in HIV-related white matter (WM) differences detected across studies. Therefore,
reliable neuroanatomical consequences of infection and therapeutic targets are difficult to identify.
Here, we pooled data from six existing HIV studies from around the world as part of the ENIGMA-
HIV consortium to evaluate 1) the effects of harmonization of dMRI measures across sites using
ComBat, and 2) whether an improved, higher-order tensor dMRI model, the tensor distribution
function (TDF), and derived scalar index (FA
TDF
) conferred higher sensitivity across
heterogeneous sites to understand the effect of HIV on WM microstructure. This study suggests
that improved dMRI indices and harmonization of these measures across cohorts, may be helpful
for detecting consistent effects of disease on the brain in international multi-site studies, while
preserving biological differences.

Keywords: HIV, Multi-site, Harmonization, TDF, Diffusion MRI, ComBat, DTI

135

3.3.1 INTRODUCTION

The human immunodeficiency virus type 1 (HIV) enters the brain soon after infection, instigating an
inflammatory cascade that typically leads to neural dysfunction, axonal damage, and diffuse myelin pallor
(Ellis et al. 2007). As a result, many brain imaging studies of HIV+ individuals have used measures
derived from diffusion tensor imaging (DTI), such as fractional anisotropy (FA
DTI
), to study white matter
(WM) microstructural abnormalities, and frequently identify lower FA
DTI
with infection. However,
inconsistencies in the effect sizes, regional distribution, and even direction of FA
DTI
changes reported
across studies have limited the generalizability of the conclusions drawn to date (O’Connor et al. 2017).
Understanding common neurobiological substrates of HIV could help lead to improved therapeutic
targets, and surrogate markers to evaluate treatment effects, but heterogeneity in study findings makes it
challenging to identify neuropathogenic pathways that generalize across international populations.
Sources of heterogeneity between findings of single cohort studies may be due to methodological as well
as population differences, including age, sex, and environmental, socioeconomic or lifestyle attributes of
the cohort. While this is true for studies across multiple diagnostic conditions, HIV is further complicated
by different viral load status, comorbidities and co-infections, drug use, age at infection, duration of
infection, treatment regimen, and degree of neurocognitive impairment, which can all differ drastically
across study populations. Other sources of heterogeneity may come from variations in study sample sizes
and analysis techniques that can also affect reported findings. It is important to account for technical and
methodological differences across studies without compromising or washing out true biological
differences.
To address methodological differences, and boost statistical power, the ENIGMA-HIV Working
Group was established to harmonize data analysis from neuroimaging studies around the world. However,
beyond coordinated diffusion MRI (dMRI) processing and analysis at each site, differences in scanners
and acquisition protocol variables, including b-values, angular and spatial resolution have been shown to
affect FA
DTI
and other standard diffusion indices (Kim et al. 2006, Zhan et al. 2010, Zhu et al. 2011).
Meta-analysis of effects across sites alleviates the need to directly pool data, and is convenient for
case/control studies; however, studies of rare modulators of effects (only a few participants per cohort)
may also be possible, retrospectively, if data could be reliably pooled across sites. Several harmonization
methods have been proposed, and a recent comparison of such techniques (Fortin et al. 2017) suggested
that ComBat (Johnson et al. 2007) is one of the most effective harmonization techniques for DTI measures.
136

The effect of harmonization techniques may not be independent of the measures being harmonized.
It is important to select dMRI models and derived scalar indices that can detect disease effects and their
modulators with maximal sensitivity and power. The DTI single diffusion tensor model (Basser et al.
1994) is the most frequently used dMRI model in clinical studies, yet the model is limited as it can only
model a single fiber population per voxel, with a single dominant orientation. At the typical resolution of
dMRI, estimates of the proportion of WM voxels containing crossing fibers are as high as 90%
(Descoteaux et al. 2009, Jeurissen et al. 2013). Among other possible models, the tensor distribution
function (TDF) as proposed by Leow et al. (2009), is a multi-tensor model that models crossing fibers as
a probabilistic ensemble of Gaussian tensors. FA derived from the TDF (FA
TDF
) has been shown to be
more sensitive to neurodegenerative disease-related differences than FA
DTI
, regardless of angular
resolution, and q‐ball imaging (QBI) generalized FA (GFA) (Tuch 2004, Nir et al. 2017). This offers a
potentially more powerful and suitable option for our ENIGMA-HIV study, where the dMRI data are from
clinical studies, limiting reliable reconstruction of many higher-order diffusion models that require high
angular resolution (HARDI) or multi-shell protocols (e.g., DSI (Wedeen et al. 2005), DKI (Wu and
Cheung 2010), NODDI (Zhang et al. 2012), and others). Prior studies have shown the TDF and resulting
FA
TDF
have less error in the fit and improved intra-study test-retest reliability across scans (Isaev et al.
2017, Nir et al. 2017), relative to FA
DTI
, but multi-site reliability and harmonization across studies have
not yet been evaluated.
Here, we pooled and analyzed dMRI data from six HIV neuroimaging studies from around the
world as part of the ENIGMA-HIV consortium, to evaluate 1) the effects of harmonization of dMRI
measures across sites using ComBat, and 2) whether FA
TDF
confers higher sensitivity across
heterogeneous sites to detect the effect of HIV on WM microstructure. While future studies are needed to
further quantify effects, we aimed to perform an initial investigation into the harmonization of diffusion
measures, to better detect consistent and cohort-specific effects across heterogeneous studies of HIV
infection.

137

3.3.2 METHODS

Study Samples

We analyzed data from six independent HIV studies from the U.S., Australia, and Serbia as part of the
ENIGMA-HIV DTI consortium. T1-weighted and diffusion-weighted MRI (DWI) were collected from
342 HIV-infected individuals (mean age: 50.6 ± 11.9 yrs; 301 M/46 F) and 243 seronegative controls
(mean age: 45.3 ± 14.3 yrs; 163 M/80 F). Study specific demographic and clinical data, scanner and
acquisition parameters are detailed in Tables 1-2.
As part of an effort to understand effects of scanner hardware and software upgrades during the
course of the Thai Resilience study of adolescents with perinatally acquired HIV (Paul et al. 2018), four
individuals were scanned twice, on average 2.8 months apart (±2.0 mo) using two different scanners and
protocols. These eight scans were used to better understand acquisition variability, as opposed to
individual or cohort level differences. DWI from the first protocol were acquired on a 1.5 T GE scanner
with 1 b0, 25 b=1000 s/mm
2
volumes, and 3.5x3.5x3.5 mm isotropic voxels acquired in duplicate, while
the second protocol DWI were acquired on a 3 T Philips scanner with 2 b0 and 32 b=1000 s/mm
2
volumes
with 2x2x2 mm isotropic voxels. Each site obtained approval from their local ethics committee and/or
institutional review board; participants signed an informed consent form at each participating site.

Image Preprocessing

DWI images were denoised using the LPCA filter (Manjon et al. 2013) and corrected for motion and eddy
current distortions. T1-weighted images were denoised using the nonlocal means filter (Coupe et al. 2008)
and underwent N3 intensity inhomogeneity normalization (Sled et al. 1998), and brain extraction (Fischl
2012). These images were linearly aligned to diffusion images, and diffusion images were then non-
linearly warped to their respective T1-weighted scans to correct for echo-planar imaging (EPI) induced
susceptibility artifacts (Avants et al. 2011); sites had not collected field maps and all protocols used single
phase acquisition. Diffusion gradient directions were rotated to accommodate linear registrations.

138

Table 1. Demographic characteristics of participants and DWI protocols, by study.
Study
Total
N
Age, years
(Mean±SD)
Sex
Male (%)
MRI
Scanner
Voxel Size
(mm
3
)
DWI
Protocol
UCSF, USA
(Watson et al. 2017)
52 HIV+
28 HIV-
64.6 ± 3.5 92.5%
3 T
Siemens
2x2x2
1 b 0 / 64 DWI
b=2000 s/mm
2

UNSW, Australia
(Cysique et al. 2017)
80 HIV+
40 HIV-
55.0 ± 6.6 100%
3 T
Philips
2.5x2.5x2.5
2 b 0 / 32 DWI
b=1000 s/mm
2

UCLA, USA
(Kuhn et al. 2018)
61 HIV+
33 HIV-
50.4 ± 13.2 73.4%
3 T
Siemens
1.98x1.98x2
7 b 0 / 64 DWI
b=1000 s/mm
2

Brown, USA
(Seider et al. 2016)
51 HIV+
27 HIV-
42.8 ± 10.6 62.8%
3 T
Siemens
1.77x1.77x1.8
11 b 0 / 64 DWI
b=1000 s/mm
2

ARCH
1
, USA
(Gullett et al. 2018)
26 HIV+
40 HIV-
39.3 ± 10.6 45.5%
3 T
Siemens
2x2x2
11 b 0 / 64 DWI
b=1000 s/mm
2

Serbia
(Boban et al. 2017)
72 HIV+
75 HIV-
40.0 ± 11.4 81.0%
3 T
Siemens
2.03x2x2
3 b 0 / 64 DWI
b=1500 s/mm
2

Table 2. Clinical characteristics of HIV+ participants, by study.
Study
% Detectable
Viral Load
Mean Current CD4+
(Mean±SD cells/mm
3
)
Mean Nadir CD4+
(Mean±SD cells/mm
3
)
% on
cART*
UCSF 32.6% 517.04 (220.48) 191.42 (146.81) 98.0%
UNSW 1.3% 556.49 (278.25) 177.09 (125.69) 100.0%
UCLA 35.7% 603.11 (297.77) 254.71 (224.57) 100.0%
Brown 32.0% 529.24 (271.10) 189.70 (171.43) 80.4%
ARCH 34.6% 499.92 (222.59) 182.08 (154.91) 92.0%
Serbia 26.4% 576.04 (330.56) 302.04 (187.32) 69.0%
*Combination antiretroviral therapy (cART)

Anisotropy Indices from DTI and TDF Models

A single diffusion tensor (DTI) was modeled at each voxel in the brain from the corrected DWI scans, and
fractional anisotropy (FA
DTI
) scalar maps were calculated (Basser et al. 1994). In contrast to the single
tensor model, the tensor distribution function (TDF) represents the diffusion profile as a probabilistic

1
Alcohol Research Center in HIV (ARCH)
139

mixture of tensors that optimally explain the observed DWI data, allowing for the reconstruction of
multiple underlying fibers per voxel, together with a distribution of weights (Leow et al. 2009). As in Nir
et al. (2017), the TDF corrected form of FA (FA
TDF
) was calculated from the resulting TDF.

ENIGMA-DTI Processing Protocol

Using publically available harmonized ENIGMA-DTI protocols (http://enigma.usc.edu/protocols/dti-
protocols/) (Jahanshad et al. 2013a), individual subject FA
DTI
maps were warped to the ENIGMA-DTI
FA template with ANTs (Avants et al. 2011) and the transformations applied to respective FA
TDF
maps.
FA indices were projected onto the ENIGMA-DTI template skeleton with TBSS (Smith et al. 2006). Using
the Johns Hopkins University (JHU) WM atlas (Mori et al. 2008), mean skeletonized FA indices were
extracted from 25 regions of interest (ROIs). Here, we focus on three ROIs in particular: the corpus
callosum (CC), a region with highly coherent fiber organization, the corona radiata (CR), a region with
large numbers of crossing fibers, and the full WM skeleton (Figure 1a).

Figure 1. (a) The ENIGMA skeletonized FA
DTI
template is displayed along with the three ROIs assessed in the
study: the corona radiata (CR), corpus callosum (CC), and full WM skeleton. (b) b 0 volumes from control
participants in each study visually highlight differences between dMRI acquisition protocols.

140

ComBat DWI Harmonization

ComBat, originally developed to model and remove batch effects from genomic microarray data (Johnson
et al. 2007), has recently been applied to multi-site cortical thickness (Fortin et al. 2018), functional MRI
connectivity (Meichen et al. 2018), and DTI measures (Fortin et al. 2017). ComBat uses an empirical
Bayes framework to remove variability from multi-site data introduced by differences in acquisition
protocols, while aiming to preserve the biological variability of site effect estimates. An empirical test of
this method on data with diagnostic heterogeneity as complex as global HIV infection has yet to be
performed. We used the ComBat algorithm to harmonize data across sites for each FA measure. For FA
TDF
and FA
DTI
, diagnosis, age, sex, and information from all 25 ROIs were used to inform the statistical
properties of the site effects.

Statistical Analyses

For each of the three ROIs, we fit a linear regression, covarying for age, sex, and age*sex interaction, to
test for differences in FA between HIV+ individuals and seronegative controls at each site. A fixed-effects,
inverse variance weighted meta-analysis of Cohen’s d effect sizes was conducted to combine effects across
sites. Heterogeneity scores (I
2
) for each test were also computed, indicating the percentage of the total
variance in effect size explained by heterogeneity between sites. We also pooled the data and performed
an analogous random effects mega-analysis across sites, grouping by site. After harmonization of data
across sites, meta- and mega-analyses were repeated for comparison with unharmonized results (Figure
2). The false discovery rate (FDR) procedure was used to correct for multiple comparisons across ROIs
(q=0.05) (Benjamini and Hochberg 1995).
For each site, a two-tailed, one-sample T-test was used to identify significant changes (%) in full
WM FA values after harmonization ((FAharmonized-FAoriginal)/FAoriginal); a two-sample paired T-test was used
to assess significant differences between DTI and TDF percent change in each site. To identify significant
differences between sites before and after harmonization, an ANCOVA was used controlling for
diagnosis, age, sex, age*sex, and site as a random variable, followed by post-hoc pairwise tests between
sites.

141

Figure 2. For each FA measure, a meta- and mega-analysis was conducted on both the original and harmonized
data.

3.3.3 RESULTS

DWI Harmonization

b0 volumes from control participants in each study allow us to visualize differences across dMRI protocols
from each site (Figure 1b). An ANCOVA testing for significant site differences in FA measures
confirmed significant differences in both full WM FA
DTI
(P=1.6*10
-76
; F=196.8) and FA
TDF
(P=1.1*10
-
75
; F=99.9) between sites. In pairwise tests between sites, significant differences in 12 of 15 pairwise
FA
DTI
tests were detected (FDR q<0.05; ARCH, UNSW, and Serbia were not different from each other,
despite differences in scanner manufacturer, b-value, angular and spatial resolution). FA
TDF
was
significantly different in 11 of 15 tests (Brown, Serbia, and UCLA were not significantly different from
each other, nor were Serbia and UCSF).
After harmonization with ComBat, an ANCOVA detected significant differences in both full WM
FA
DTI
(P=2.2*10
-15
; F=16.7) and FA
TDF
(P=1.2*10
-20
; F=22.5) between sites, but with reduced effect
sizes (F-value) compared to the original, unharmonized effect sizes. Pairwise tests revealed no significant
differences in FA
DTI
between sites after multiple comparisons correction, yet a significant difference
between FA
TDF
from Brown and all sites but ARCH remained.
142

Density plots of the pre-ComBat full WM TDF and DTI FA from HIV+ and seronegative control
participants from each site revealed the distribution of one site, Brown, deviated the most from the other
sites for FA
DTI
but not FA
TDF
(Figure 3a). This was rectified after harmonization with ComBat (Figure
3b). All sites showed significant changes in FA values after harmonization (Figure 3c). While FA
DTI

values did not consistently show a larger percent change in the full WM after harmonization compared to
FA
TDF
across sites, the largest percent change in FA was detected for FA
DTI
in Brown (24%), followed by
FA
DTI
in UCSF (10%). FA
TDF
changed at most 5% (UNSW). Similarly, box plots of TDF and DTI FA in
the full WM before and after ComBat, in controls only, show the greatest changes in Brown and UCSF
(Figure 3d).

Figure 3. Density plots of the mean TDF and DTI FA in the full WM skeleton from HIV+ and seronegative control
participants from each site (a) before and (b) after harmonization with ComBat. (c) The mean percent change
between original and harmonized full WM FA values are listed. (d) In controls only, box plots of the mean TDF
and DTI FA in the full WM before and after harmonization are shown to highlight differences across cohorts even
in controls.
143

HIV Diagnosis Effect Sizes

Significantly lower FA
TDF
was detected in HIV+ individuals compared to seronegative controls in the full
WM, CC, and CR in both mega- and meta-analyses (FDR q<0.05); the largest effects were detected in the
CR (Cohen’s d=-0.32; Figure 4a). Significantly lower FA
DTI
was only detected in the CC, a region with
highly coherent WM organization, where effect sizes were comparable to those for FA
TDF
. There were no
major differences between meta- or mega-analysis effect sizes. Effect sizes were marginally lower after
harmonization of FA measures across sites, but still within the standard error bounds; ComBat did not
alter the statistical significance for any ROI.
Heterogeneity scores (I
2
) revealed approximately two times the variance explained between sites
for FA
TDF
effect size estimates compared to FA
DTI
across ROIs, regardless of harmonization (Figure 4b).
There was effectively no variance across sites (I
2
=0) in effects detected by FA
DTI
in the CR. A comparison
of effect sizes from each site in the CR - the region with crossing fibers and the largest FA
TDF
effect size
- showed (1) no major differences between original and harmonized analyses except for Brown, where
FA
TDF
showed improved effect sizes, and (2) improvements in effect sizes from FA
TDF
compared to
standard FA
DTI
only for sites that already showed the strongest associations with FA
DTI
(Figure 4c).

Cross Scanner Variability

Despite differences in manufacturer, field strength, spatial and angular resolution, FA
DTI
and FA
TDF

measures were both highly correlated across Resilience study scanning protocols in the CC (FA
TDF

Pearson's r=0.98, FA
DTI
r=0.97), CR (FA
TDF
Pearson's r=0.97, FA
DTI
r=0.99), and full WM (FA
TDF

Pearson's r=0.96, FA
DTI
r=0.99). A two-tailed, paired T-test revealed a significantly greater percentage
increase in FA
TDF
values compared to FA
DTI
values between older and updated protocols, only in the CR
(P=0.004) and full WM (P=0.04).

144

Figure 4. (a) Effect sizes (Cohen’s d and standard error) from meta- and mega-analyses of TDF and DTI FA
differences between HIV+ and seronegative control participants before (FA
TDF
Mega FDR threshold P FDR=0.002,

Meta P FDR=0.001; FA
DTI
Mega P FDR=0.002,

Meta P FDR=0.001) and after harmonization with ComBat (FA
TDF
Mega
P FDR=0.017,

Meta P FDR=0.008; FA
DTI
Mega P FDR=0.005,

Meta P FDR=0.002) in the corona radiata (CR), corpus
callosum (CC), and full WM. (b) Heterogeneity scores (I
2
) reflect the percentage of total variance in effect sizes
explained by heterogeneity between sites. (c) Effect sizes (absolute value) of TDF and DTI FA differences between
HIV+ and seronegative control participants for each site, before and after harmonization in the CR.

3.3.4 DISCUSSION

Here we evaluated the effects of harmonization and higher-order tensor diffusion MRI scalar measures in
a large multi-site HIV study. Harmonization with ComBat allowed sites such as Brown and UCSF, which
deviated most from the other four sites, to achieve distributions more in line with the other cohorts. We
note that Brown was the only site with hepatitis C co-infection, while the UCSF study included the oldest
participants (Table 1). Although the distributions were more aligned, the differences between these
cohorts and the others were not completely washed out. ComBat appears to preserve biologically
145

meaningful variance between sites; the heterogeneity scores (I
2
) did not change drastically after
harmonization. We found no major effects of ComBat harmonization on meta- and mega-analysis
findings. We note that in Fortin et al. (2017), voxel-wise information was used to harmonize measures
while we used ROI information; fewer features may be a possible limitation of this approach, and future
analyses will evaluate harmonization at the voxel-wise level.
Improvements in scalar dMRI indices may play a major role in our ability to find consistent
diffusion parameters and related neurobiological substrates across sites, regardless of harmonization.
While many HIV studies report inconsistent differences in FA
DTI
, in addition to heterogeneity in the HIV
populations studied and methodology, this may be due to established limitations of the FA
DTI
measure
itself. Time constraints are often placed on dMRI protocols in clinical settings. HARDI or multi-shell
protocols, often used to overcome these limitations, are still rare in clinical population studies, although
this may change in the near future. In large multi-site clinical studies, such as ENIGMA - where reliable
reconstruction of many other proposed higher-order diffusion models may not be feasible –the TDF may
be a suitable option. HIV-related CC microstructural differences are some of the most consistently
reported findings (O’Connor et al. 2017), and both FA
TDF
and

FA
DTI
detected similar effect sizes in this
region with highly coherent WM organization. FA
TDF
was able to detect more widespread differences in
the full WM and larger effect sizes, in particular in the CR, a region with crossing fibers, which was not
significant with FA
DTI
. This suggests HIV may have more widespread and robust effects on WM
microstructure than previously reported.
Protocol, scanner, and field strength changes between the original and improved Resilience study
dMRI acquisitions resulted in similar changes of FA
TDF
and FA
DTI
values in coherent CC WM, but FA
TDF

changed significantly more than FA
DTI
in the CR. The DTI model is limited in regions with crossing fibers
regardless of the quality of DWI acquisition protocols; i.e., crossing fibers may result in microstructural
measures falsely representing those of unhindered diffusion. FA
TDF
, however, models such crossings even
in relatively low angular resolution data (Nir et al. 2017), and may still be sensitive to biological variability
in cases where the DTI model does not capture anisotropic diffusion. This suggests FA
TDF
could be more
sensitive to biological as well as protocol differences. This may explain the fact that in pairwise tests,
Brown (with its high rate of patient comorbidities) remains significantly different from other sites after
harmonization for FA
TDF
, but not FA
DTI
. This is further suggested by the higher I
2
estimate in FA
TDF
effect
size variance between sites, particularly in the CR where FA
DTI
showed effectively no variability. The
largest differences between HIV+ participants and controls were detected with FA
TDF
in the CR in both
146

meta- and mega-analyses across sites, whereas no significant differences could be detected with FA
DTI
.
This suggests that differences and changes in FA
TDF
between sites and protocols are capturing real
biological effects and not just noise. Larger studies with dMRI data collected with various protocols from
the same participants or phantom are necessary to draw further conclusions.
Statistical approaches such as ComBat may be helpful for harmonizing measures derived in multi-
cohort imaging studies, yet parallel efforts into improving diffusion metrics to more accurately and
consistently represent the underlying white matter microstructure are also critical. Advances in both of
these lines of research will ultimately help us uncover common neurobiological consequences of HIV
infection, and allow us to dig deeper into modulators of disease effects on the brain in HIV+ individuals
around the world.

3.3.5 ACKNOWLEDGEMENTS

Funding for ENIGMA is provided as part of the BD2K Initiative U54 EB020403 to support big data
analytics, and by P41 EB015922. Work from each site was funded by: (1) UCLA: K23MH095661,
Clinical and Translational Research Center Grants UL1RR033176 and UL1TR000124 (ADT); MH19535
(TK); (2) Serbia: Provincial Secretariat for Higher Education and Scientific Research 114-451-2730/2016-
02; (3) UNSW: NHMRC APP568746 (LC); (4) Brown and ARCH: R01MH074368, the
Lifespan/Tufts/Brown Center for AIDS Research P30 AI042853, P01AA019072 (RC); (5) UCSF:
K23AG032872 (VV), R01AG048234, and R01AG032289; (6) Resilience: R01MH102151 (JA). This
work was also supported by R01MH085604 Neuropathogenesis of clade C HIV in South Africa.

147

CHAPTER 4

Multimodal Predictors of Cognitive Impairment

148

4.1 Multimodal Brain Imaging Predicts Neurocognitive Impairment in Chronic HIV

This section is adapted from:

Nir TM, Lam HY, Jahanshad N, Ching CRK, Harezlak J, Martinez K, Schifitto G, Zhu T, Cohen RA,
Thompson PM, Navia BA, for the HIVNC (2018). Multimodal brain imaging predicts neurocognitive
impairment in chronic HIV. Conference on Retroviruses and Opportunistic Infections, Boston, MA,
March 2018.

149

Multimodal Brain Imaging Predicts Neurocognitive Impairment in Chronic HIV

Talia M. Nir
1
, Kenia Martinez
2
, Hei Y. Lam
1
, Neda Jahanshad
1
, Ofer Pasternak
3
, Chris R.K. Ching
1
, Jaroslaw
Harezlak
4
, Giovanni Schifitto
5
, Tong Zhu
6
, Ronald A. Cohen
7
, *Paul M. Thompson
1
, *Bradford A. Navia
8
for the HIV Neuroimaging Consortium
* These authors contributed equally to the manuscript

1
Imaging Genetics Center, Mark and Mary Stevens Neuroimaging and Informatics Institute, Keck School of Medicine, University of
Southern California, Marina del Rey, CA, USA
2
Department of Child and Adolescent Psychiatry, Hospital General Universitario Gregorio Marañón, Madrid, Spain
3
Departments of Psychiatry and Radiology, Brigham and Women's Hospital, Harvard Medical School, Boston, MA, USA
4
Indiana University School of Public Health, Bloomington, IN, USA
5
Department of Neurology, University of Rochester, Rochester, NY, USA
6
Department Radiation Oncology, University of North Carolina, Chapel Hill, NC, USA
7
Department of Aging and Geriatric Research, University of Florida, Gainesville, FL, USA
8
Department of Public Health, Infection Unit, Tufts University School of Medicine, Boston, MA, USA

Abstract. Neurological injury may persist in individuals with chronic HIV infection despite viral
suppression with combined antiretroviral therapy (cART). Abnormal brain metabolite
concentrations, white matter (WM) microstructure, brain volume, and plasma markers of infection
severity (nadir/current CD4+ count and viral load) have all been individually linked with HIV-
associated neurocognitive impairment cross-sectionally. However it is important to identify
biomarkers that predict which individuals are more likely to suffer cognitive losses, which may be
used to enrich patient selection for clinical trials or track treatment response. In 44 chronically
HIV-infected individuals on stable cART (mean age: 48.18 ± 7.72 yrs; sex: 28 male /16 female),
we 1) compared the ability of baseline measures derived from five modalities —
neuropsychological evaluations, plasma measures, MRS metabolite concentrations, diffusion MRI
(dMRI) WM microstructural indices, and T1-weighted volumetric measures— to predict cognitive
decline, and 2) compared the goodness of fit of predictive models that included combinations of
modalities. We found that the most parsimonious model that explained the most variance in
cognitive changes (lowest AIC and highest adjusted R
2
) included both baseline MRS measures of
glutamate+glutamine/creatine and dMRI diffusivity measures. Inclusion of additional measures
did not significantly improve models. Multimodal neuroimaging biomarkers, reflecting metabolite
and microstructural brain changes, may offer a new source of information that could be beneficial
for prognosis.

Keywords: HIV, cART, MRS, Diffusion MRI, Brain Volume, Cognitive Impairment

150

4.1.1 INTRODUCTION

Combination antiretroviral treatment (cART) has dramatically decreased HIV-related mortality and
morbidity. Despite near normal longevity among treated adults (Rodger et al. 2013, Costagliola 2014), a
spectrum of HIV-associated neurocognitive disorders –that include a convergence of motor, behavioral,
and cognitive facets —occur in 15-50% of patients and remain a significant public health concern
(Cysique et al. 2004, McArthur 2004, Heaton et al. 2010, Simioni et al. 2010). Although the frequency of
HIV-associated dementia has decreased, the incidence of milder cognitive dysfunction may be on the rise
(Ances and Ellis 2007, Clifford and Ances 2013). Growing evidence suggests persistent neurological
injury in chronic HIV infection despite cART, and that neuro-asymptomatic HIV-infected individuals can
develop progressive decline in cognitive function (Grant et al. 2014). This indicates a critical unmet need
to identify novel therapies to protect the central nervous system (CNS), even in the era of effective cART
treatment.
Plasma measures, such as CD4+ T-cell count and viral load, are generally used to monitor HIV
infection and treatment response, but may no longer be strongly associated with neurocognitive
impairment (NCI) in the cART era (Clifford and Ances 2013). Neuroimaging markers remain important
for understanding HIV-related brain changes and how they relate to NCI. Abnormal proton magnetic
resonance spectroscopy (MRS) cerebral metabolite levels, reflecting neuronal and glial dysfunction and
neuroinflammation (Harezlak et al. 2011), altered diffusion MRI (dMRI) measures of white matter (WM)
microstructure (Zhu et al. 2013), and smaller brain volumes in T1-weighted MRI derived morphometry
maps (Cohen et al. 2010) have each been individually linked with NCI in treated asymptomatic HIV-
infected individuals. These studies, however, have been largely cross-sectional and there remains a need
to identify biomarkers that predict which individuals are more likely to suffer cognitive losses. Based on
available literature in other fields, notably Alzheimer’s disease (Reiman and Jagust 2012, Hua et al. 2016,
Veitch et al. 2018), imaging biomarkers have provided an important non-invasive, in vivo approach to
identify subgroups at higher risk for cognitive decline and to monitor the effects novel drug treatments in
clinical trials that aim to slow or halt such decline (McArthur 2012, Chang and Shukla 2018).
As part of the HIV Neuroimaging Consortium (HIVNC), we assessed chronically infected
individuals on stable cART, including virologically suppressed individuals, to discover neuroimaging
measures, at baseline, associated with cognitive changes. To assess how different modalities differ in their
predictive ability we 1) tested which of five modalities —neuropsychological evaluations, plasma
151

measures, MRS metabolite concentrations, dMRI WM microstructural indices, or T1-weighted volumetric
measures— best predicted cognitive changes and 2) assessed whether inclusion of a combination of
modalities in predictive models improved model fit. We hypothesized that using multimodal markers of
HIV burden and brain injury, in particular those derived from neuroimaging, would better predict
cognitive changes in chronically HIV-infected adults on stable cART. Neuroimaging biomarkers may
offer a new source of information that could be beneficial for prognosis.

4.1.2 METHODS

Subjects and Clinical Assessments

1.5 tesla GE MR spectroscopy, structural and diffusion MRI, clinical, and longitudinal
neuropsychological data (mean time interval 1.71 +/- 0.69 yrs) were obtained from 50 HIV+ participants
recruited at the University of Rochester as part of the HIV Neuroimaging Consortium (Zhu et al. 2013).
HIVNC inclusion criteria consisted of: nadir CD4 counts ≤ 200cells/mm
3
, a stable antiretroviral regimen
for at least 12 consecutive weeks prior to study screening, hemoglobin ≥ 9.0 gm/dL, serum creatine ≤ 3 ×
ULN, AST (SGOT), ALT (SGPT), and alkaline phosphatase ≤ 3 × ULN, negative serum or urine
pregnancy test and age ≥ 18 years. Exclusion criteria included premorbid or comorbid psychiatric
disorders, confounding focal or diffuse neurologic disorders such as chronic seizures, stroke, head trauma
resulting in loss of consciousness of more than 30 min., multiple sclerosis, brain infection other than HIV,
or brain neoplasms, including CNS lymphoma; active alcohol and drug abuse or related medical
complications within 6 months of study; and diabetes mellitus with a fasting glucose > 140 mg/dl. All
procedures were reviewed and approved by the local institutional review board. All participants gave
written informed consent.
Demographic and clinical characteristics are reported in Table 1. Clinical plasma measures
assessed included nadir CD4+ count (cells/mm
3
), baseline (i.e., current) CD4+ count, and HIV RNA viral
load. Plasma HIV RNA was defined as detectable when viral load was > 50 copies/mL. As described in
(Gongvatana et al. 2013) neuropsychological (NP) evaluation was performed in 7 cognitive domains:
processing speed, verbal fluency, executive functioning, working memory, learning, memory, and
psychomotor function (Heaton et al. 1995, Heaton et al. 2010, Heaton et al. 2011). Individual tests used
in each cognitive domain are listed in Supplementary Table S1. Individual test scores were converted
152

into demographically corrected T-scores (mean = 50, SD = 10). A Global Deficit Score (GDS) was
calculated reflecting overall impairment. Specifically, demographically corrected T-scores were converted
to deficit scores ranging from 0 (no impairment) to 5 (severe impairment) and averaged. GDS has been
shown to be effective for detecting cognitive impairment in HIV-infected individuals (Heaton et al. 1995,
Carey et al. 2004). Of 50 participants with available baseline GDS, 44 had follow-up GDS, and were
therefore included in this study. To evaluate longitudinal change in GDS, baseline scores were subtracted
from the follow-up scores. As the time interval was variable across subjects (mean: 1.71 +/- 0.69 yrs), the
difference was then divided by the time interval resulting in a score that reflected the annualized rate of
change.

Table1. Demographic and clinical characteristics of participants. Continuous variables are reported as mean
(standard deviation), and categorical variables as N (percent).
N 44
Age (years) 48.18 (7.72)
Sex (Male) 28 (63.64%)
Education (years) 12.75 (2.83)
Nadir CD4+ (cells/mm
3
) 64.34 (44.84)
Current CD4+ (cells/mm
3
) 366.91 (202.98)
Detectable Viral Load (> 50 copies/mL) 10 (22.73%)
GDS Time Interval (years) 1.71 (0.69)
Cognitive Deficit Scores Baseline Follow-Up P-value
Global Deficit Score (GDS) 0.41 (0.52) 0.43 (0.60) 0.017*
Processing Speed 0.60 (0.94) 0.44 (0.87) 0.021*
Verbal Fluency 0.34 (0.55) 0.38 (0.67) 0.014*
Learning 1.07 (1.17) 0.99 (1.23) 0.82
Memory 1.14 (1.17) 1.02 (1.28) 0.40
Executive Function 0.53 (0.80) 0.64 (1.05) 0.040*
Working Memory 0.70 (0.94) 0.60 (1.04) 0.66
Psychomotor 0.55 (1.15) 0.53 (0.98) 0.15
* p ≤ 0.05 (Significant longitudinal change; One-way repeated measures ANCOVA)

T1-weighted Tensor-Based Morphometry (TBM)

Volumetric differences in regional brain volumes were assessed using TBM as described in (Hua et al.
2013a). Anatomical T1-weighted (T1w) spoiled gradient echo sequences (SPGE; matrix= 256 × 120;
voxel size 0.97x0.97x1.30 mm; TR/TE=21/6 ms; flip angle = 30) were acquired. All T1w anatomic scans
were denoised using a nonlocal means filter (Coupe et al. 2008) and underwent N3 intensity
inhomogeneity normalization (Sled et al. 1998). FreeSurfer software was used to create brain tissue masks
(Fischl 2012). As in Jahanshad et al. (2015b), a 3-channel study specific minimal deformation template
153

(MDT) was created from all subjects’ T1w images and additional FreeSurfer cortical, and subcortical
segmentation channels, using ANTs (Reuter et al. 2012). To quantify 3D patterns of volumetric tissue
differences, masked baseline T1w scans were linearly and elastically registered to the MDT using inverse-
consistent elastic intensity-based registration (Leow et al. 2007). The log of the Jacobian determinant
(derived from the gradients of the deformation fields) was averaged in voxels within four tissue types
segmented using FreeSurfer (Fischl et al. 2004): cortical gray matter (CGM), white matter (WM),
subcortical gray matter (SGM) structures, and lateral ventricles. For each subject, these 4 brain tissue
compartments were further subdivided into 19 FreeSurfer derived anatomically refined regions of interest
(ROIs): the cortical GM and subcortical WM were assessed in each of the frontal, temporal, parietal, and
occipital lobes as well as the insula and cingulate; the subcortical GM was segmented into individual
structures including the thalamus, caudate, putamen, pallidum, amygdala, hippocampus, and nucleus
accumbens (Figure 1c).

Diffusion MRI Models and Scalar Indices

To assess microstructural differences in WM, dMRI were acquired (matrix=128x128; voxel size
0.95x0.94x5.0 mm
3
; TR/TE= 7000/75ms), with 22 volumes: 1 T2-weighted b0 image with no diffusion
sensitization, and 21 diffusion-weighted images (DWI) with b=1000 s/mm
2
. Raw dMRI images were
denoised using the LPCA filter (Manjon et al. 2013) and corrected for eddy current and head motion with
FSL's eddy-correct tool. Extra-cerebral tissue was removed with FSL's BET tool (Smith 2002). b0 images
were linearly aligned to processed T1w images, using respective WM masks for FSL FLIRT’s boundary-
based registration (BBR) (Greve and Fischl 2009); the registration was subsequently inverted to bring the
T1w image to the native dMRI space. Diffusion images were then non-linearly warped to their respective
T1w scans with ANTs (Avants et al. 2011) to correct for EPI-induced susceptibility artifacts. Diffusion
gradient directions were rotated to account for linear registrations.
A single diffusion tensor (DTI) (Basser et al. 1994) was modeled at each voxel in the brain from
the corrected dMRI scans using FSL dtifit with weighted least squares. DTI fractional anisotropy (FA
DTI
),
mean diffusivity (MD), axial diffusivity (AxD) and radial diffusivity (RD) scalar maps were calculated.
However, a single voxel may contain multiple fiber populations and/or multiple tissue compartments
which cannot be resolved by the single tensor model. To address this limitation, we additionally fit two
more proposed models that may be appropriate for low resolution clinical dMRI data, as in this study. The
154

free water (FW) model (Pasternak et al. 2009) is a bi-tensor model with one isotropic tensor modeling
freely diffusing water in CSF or extracellular spaces. The FW index, which quantifies the fractional
volume of free-water in each voxel, may offer improved specificity and interpretability as it may reflect
HIV-related inflammation and resulting edema. In contrast to the single tensor model, the tensor
distribution function (TDF) represents the diffusion profile as a probabilistic mixture of tensors that
optimally explain the observed dMRI data, allowing for the reconstruction of multiple underlying fibers
per voxel, together with a distribution of weights (Leow et al. 2009). As in Zhan et al. (2009) and Nir et
al. (2017), the TDF corrected form of FA (FA
TDF
) was calculated. FA
TDF
has previously been shown to be
both a more sensitive and accurate measure to detect neurodegenerative disease related changes in
microstructure with clinical quality dMRI data (Nir et al. 2017).
Using publically available ENIGMA-DTI protocols (http://enigma.usc.edu/protocols/dti-
protocols/), individual subject FA
DTI
maps were warped to the ENIGMA-DTI FA template (Avants et al.
2011). Transformations were applied to respective DTI, TDF and FW scalar maps and measures were then
projected onto the template skeleton using TBSS (Smith et al. 2006). Using the Johns Hopkins University
WM atlas (Mori et al. 2008), mean skeletonized dMRI measures were extracted from the full WM and 21
WM ROIs listed in (Figure 1b; Table 2). Indices were averaged across left and right hemispheres to
obtain bilateral dMRI measures.

Magnetic Resonance Spectroscopy Metabolite Concentrations

As previously described (Harezlak et al. 2011), single-voxel proton MRS spectra were acquired using the
GE PRESS sequence from 6cc voxels in 3 brain regions: mid-line frontal cortex GM (MFC), right (or left)
mid-frontal centrum semiovale WM (FWM), and right (or left) basal ganglia (BG; Figure 1a). Field
homogeneity and water suppression were adjusted using automated algorithms from GE. Water
suppressed spectra were collected with TE/TR = 35/3000ms, bandwidth = 2500 Hz, 128 averages,
NEX=8. Metabolite ratios of N-acetylaspartate (NAA), choline (Cho), myoinistol (MI), glutamate and
glutamine (Glx), relative to creatine (Cr) –NAA/Cr, Cho/Cr, MI/Cr, and Glx/Cr – were determined using
LC Model spectral analysis software and an unsuppressed water FID at TE=35ms for eddy-current
correction (Provencher 2001). This automated processing method yields metabolite ratios with Cramer-
Rao lower bound less than 20% (Lee et al. 2003). Of the 44 participants with longitudinal GDS, only 39
had all MRS measures available.
155

Figure 1 We evaluated baseline neuroimaging measures derived from three MRI modalities and their ability to
predict future cognitive decline. (a) Volumetric differences in T1-weighted images were evaluated using TBM
derived jacobian maps averaged in four tissue classes first across the full brain, then in 19 localized regions
including each lobe and each subcortical structure individually (averaged bilaterally). (b) Six indices of WM
microstructure were estimated from three dMRI models and averaged across the entire WM skeleton for preliminary
analyses and in 21 WM tracts (averaged bilaterally) for secondary analyses (see Table 2 for list of abbreviations).
(c) Four metabolite ratios were measured in three brain regions with MRS.

156

Table 2. The full WM skeleton was sub-divided into 21 WM regions of interest from the JHU atlas.
BCC Corpus Callosum Body PLIC Posterior Limb Internal Capsule
GCC Corpus Callosum Genu RLIC Retrolenticular Internal Capsule
SCC Corpus Callosum Splenium FX Fornix
ACR Anterior Corona Radiata FX/ST Fornix (Cres) / Stria Terminalis
SCR Superior Corona Radiata IFO Inferior Fronto-Occipital Fasciculus
PCR Posterior Corona Radiata PTR Posterior Thalamic Radiation
CGC Cingulum (Cingulate Gyrus) SFO Superior Fronto-Occipital Fasciculus
CGH Cingulum (Hippocampal Portion) SLF Superior Longitudinal Fasciculus
CST Corticospinal Tract SS Sagittal Striatum
EC External Capsule TAP Tapetum
ALIC Anterior Limb Internal Capsule

Associations between Individual Baseline Predictors and Cognitive Change

First, longitudinal change in GDS was tested using repeated measures ANCOVA, where baseline and
follow up deficit scores were the within subject factor and age, sex and time between both assessments
were covariates.
Next, multiple linear regressions tested for associations between annual change in GDS and 26
baseline predictors (Table 3; Figure 1) from 5 modalities: 1) NP (baseline GDS), 2) plasma (log nadir
CD4+ count, log current CD4+ count, detectable viral load), 3) MRS (NAA/Cr, Cho/Cr, MI/Cr, and
Glx/Cr), 4) dMRI (FA
DTI
, MD, RD, AxD, FW, FA
TDF
), and 5) T1w MRI (TBM volume). Sex and age
were set as covariates. We highlight variables that survived correction for 26 comparisons using the false
discovery rate (FDR) method at q=0.05, but include all those suggestively associated with cognitive
change for model selection (p≤0.05). Effect sizes for baseline predictors were evaluated using the r-value.
For each model, goodness of fit was evaluated by comparing the adjusted R
2
(proportion of variance
explained) and the Akaike Information Criterion (AIC) (Akaike 1998). MRS was only available in a subset
of participants, so predictors associated with cognitive change in the full group (n=44) were also tested in
the subset of 39 participants with MRS for direct comparison between modalities.
As the study sample size from Rochester was a relatively small, each test was run both in the full
group and 500 random bootstrapped samples. The mean and 95% confidence interval for resulting
157

statistics were compared to results from analyses run in the full sample to evaluate the consistency of
findings.

Model Selection

We compared different sets of models, always keeping age and sex in the model, in order to identify the
minimum number of modalities needed to account for the greatest proportion of variance in rates of
cognitive change (adjusted R
2
) in three stages. Stage 1: Of the predictors associated with change in GDS
(p≤0.05), the single best predictor from each modality was identified—i.e., the model with the highest
adjusted R
2
and the lowest AIC. Stage 2: Starting with a model that included only the best predictor from
each modality together, we compared models with subsets of the best predictors/modalities (i.e.,
multimodal models with a maximum of 1 measure from a given modality), and compared adjusted R
2
and
AIC. Stage 3: If a multimodal model from Stage 2 was better than models from Stage 1, we expanded the
multimodal model to include all significant predictors from the ‘surviving’ modalities (i.e., models with
multiple predictors per modality). Stage 4: As a final check, we compared our best parsimonious model
from Stages 1-3 to a model including all the predictors individually associated with GDS change. Full
WM dMRI measures were highly correlated (Pearson’s r = 0.86-0.99; Supplementary Table S2),
therefore only one dMRI measure was included in any given model. A χ
2
test was used to assess significant
improvements in R
2
between nested models, while as a rule of thumb, a reduction in AIC greater than 2-3
units indicated improved model fit while balancing complexity (Burnham and Anderson 2002).

Secondary Regional Analyses

To better understand regional contributions driving full brain dMRI and T1w TBM associations with
change in GDS, we evaluated anatomically refined ROI associations in 21 WM tracts (dMRI; Table 2)
and 19 GM/WM lobes and subcortical structures (T1w; Figure 1). Within each modality, we again
corrected for multiple comparisons using FDR at q=0.05. We compared our reference ‘best model’ derived
from full brain measures to models including the best ROI predictor, to see if regional specificity improved
model fit (higher adjusted R
2
and lower AIC).

158

4.1.3 RESULTS

Individual Baseline Predictors Associated with Cognitive Decline

Repeated measures ANCOVA revealed a significant longitudinal increase in GDS (increased impairment;
p = 0.017; Table 1). While baseline GDS and baseline plasma measures were not associated with rates of
longitudinal change in GDS (p > 0.05), associations were detected with 10 baseline brain imaging
measures (dMRI, T1w TBM, MRS; Table 3).
Of the 12 baseline MRS metabolite measures, lower baseline BG NAA/Cr, FWM Glx/Cr, and
MFC Glx/Cr were associated with increases in GDS (p ≤ 0.05; n = 39). Only MFC Glx/Cr survived
multiple comparisons correction for 26 tests (p = 0.006), and showed the largest effect size (r = -0.44).
MFC Glx/Cr explained a higher proportion of the variance compared to the other metabolites (adjusted R
2
= 0.19) and the AIC, while similar to FWM Glx/Cr, was significantly lower than BG NAA/Cr (i.e., more
than 3 units lower).
Of the 4 T1w TBM tissue compartment volumetric measures, higher ventricular volume and lower
CGM volume at baseline were associated with annual cognitive decline (increased GDS; p ≤ 0.05; n =
44). Only cortical GM associations survived FDR correction (p = 0.010), and showed the largest effect
size (r = -0.394). CGM explained the highest proportion of the variance (adjusted R
2
= 0.17) and had a
significantly lower AIC than ventricular volume (greater than 3 units).
Of the 6 dMRI full WM measures, only baseline FA
DTI
was not associated with change in GDS (n
= 44). The TDF corrected form of FA was lower at baseline in those that cognitively declined (increased
GDS; p ≤ 0.05). Baseline DTI diffusivity (MD/RD/AxD) and FW survived FDR correction (p = 0.007,
0.010, 0.006, 0.008 respectively) and were significantly higher in those that declined. While AxD was
marginally better, effect sizes and adjusted R
2
across dMRI measures were very similar (r = 0.38-0.42;
adjusted R
2
= 0.16-0.19; Table 3), which was expected given the measures were highly correlated
(Pearson’s r = 0.86-0.99). Furthermore, the difference in AIC across measures was always less than 2.
Follow-up dMRI index and TBM volumetric associations run in the subset of 39 participants that
also had MRS revealed the same trends within each modality (Table 4). Direct comparison across
modalities revealed the largest effect sizes and adjusted R
2
, and lowest AIC were detected with MFC
Glx/Cr followed by full WM AxD. However all diffusion indices were again very similar. Results from
159

the 500 bootstrapped random samples further confirmed findings both within and across modalities
(Supplementary Table S3).

Table 3. Associations between change in GDS and each baseline NP, plasma, dMRI, MRS, and T1w measure,
when controlling for sex and age. P-values and r-values are reported for the variable and AIC and adjusted R
2
are
reported for the model.
Modality
Measure Model
p-value r-value AIC Adj. R
2

dMRI
N=44
FA
DTI
0.096 -0.26 8.05 0.086
MD 0.007** 0.41 3.15 0.18
RD 0.010** 0.39 3.80 0.17
AxD 0.006** 0.42 2.83 0.19
FW 0.008** 0.41 3.22 0.18
FA
TDF
0.014* -0.38 4.44 0.16
MRS
N=39
BG NAA/Cr 0.039* -0.34 7.27 0.11
FWM NAA/Cr 0.13 -0.25 9.51 0.060
MFC NAA/Cr 0.85 0.033 12.02 -0.003
BG MI/Cr 0.36 -0.16 11.12 0.020
FWM MI/Cr 0.54 0.11 11.63 0.007
MFC MI/Cr 0.26 -0.19 10.65 0.032
BG CHO/Cr 0.90 0.021 12.04 -0.004
FWM CHO/Cr 0.46 0.13 11.44 0.012
MFC CHO/Cr 0.26 -0.19 7.59 0.029
BG Glx/Cr 0.11 -0.26 9.24 0.066
FWM Glx/Cr 0.014* -0.40 5.23 0.16
MFC Glx/Cr 0.006** -0.444 3.485 0.194
T1w
N=44
Lat. Ventricles 0.048* 0.31 6.80 0.11
Cort. GM 0.010** -0.39 3.74 0.17
WM 0.64 0.075 10.90 0.025
Subcort. GM 0.63 0.077 10.88 0.026
Plasma
N=44
Nadir CD4+ (log) 0.77 0.046 11.05 0.022
Current CD4+ (log) 0.81 -0.039 11.08 0.021
Viral Load 0.38 -0.14 10.28 0.039
NP
N=44
BL GDS 0.65 -0.073 10.91 0.025
* p≤ 0.05 ** p≤ FDR critical threshold p=0.010

160

Model Selection

In the subset of n=39 participants matched across modalities (Table 4), within each respective modality,
CGM volume was the best T1w TBM predictor (AIC: 7.30; adjusted R
2
=0.11), MFC Glx/Cr the best MRS
predictor (AIC: 3.49; adjusted R
2
=0.19), and AxD marginally the best dMRI predictor (AIC: 5.29; adjusted
R
2
=0.16). The most parsimonious model that was significantly better than any model including 1 predictor,
included full WM AxD and MFC Glx/Cr (AIC: -0.14; adjusted R
2
: 0.28; Table 5). A comparison of the
AIC when swapping out AxD with one of the remaining dMRI measures (MD, RD, FW, FA
TDF
), confirms
no significant differences between AxD and other dMRI indices in this model (AIC range: 0.38-0.94;
adjusted R
2
: 0.27-0.26; Supplementary Table S4). Compared to the model with AxD and MFC Glx/Cr,
AxD and FWM Glx/Cr was slightly but not significantly worse (AIC: 1.55; adjusted R
2
: 0.25); however,
further swapping AxD for the remaining dMRI measures in this case did result in a significantly worse fit
(AIC increase consistently greater than 2 units). There was no difference between our reference ‘best
model’ and the model including all the best measures from each modality or the one that included all
individually associated measures.

Table 4. Statistics for baseline dMRI and T1w TBM predictors significantly associated with change in GDS in all
44 participants, tested in the subset of 39 individuals who also have MRS measures available.
Modality
Measure Model
p-value r-value AIC Adj. R
2

dMRI
N=39
MD 0.016* 0.39 5.48 0.15
RD 0.020* 0.38 5.97 0.14
AxD 0.014* 0.40 5.29 0.16
FW 0.017* 0.39 5.58 0.15
FA
TDF
0.025* 0.37 6.37 0.13
T1w
N=39
Lat. Ventricles 0.080 0.29 8.60 0.081
Cort. GM 0.040* -0.34 7.30 0.11
MRS
N=39
BG NAA/Cr 0.039* -0.34 7.27 0.11
FWM Glx/Cr 0.014* -0.40 5.23 0.16
MFC Glx/Cr 0.006* -0.44 3.49 0.19
* p < 0.05

161

Table 5. Statistical comparison of single and multimodal models predicting annual change in GDS. P-values are
reported for chi-squared tests between our reference ‘best model’ and other models evaluated; a p<0.05 signifies
that our ‘best model’ fit significantly better than the given model. Key: Red: dMRI, Green: MRS, Blue: T1w.
Model Selection (N=39) AIC Adj. R
2
χ
2
p-value
Best Model AxD + MFC Glx/Cr -0.14 0.28 --
Other Models
AxD 5.29 0.16 0.008*
MFC Glx/Cr 3.49 0.19 0.022*
Cort. GM 7.20 0.11 0.002*
AxD + MFC Glx/Cr + Cort. GM -2.13 0.33 0.059
AxD + MFC Glx/Cr + FWM Glx/Cr + BG
NAA/Cr + Cort. GM + Lat. Ventricles
0.26 0.33 0.17
* p < 0.05

Secondary Regional Analyses

No regional TBM ROI was significantly associated with cognitive change after multiple comparisons
correction. Local dMRI ROI analyses confirmed no significant association between FA
DTI
in any ROI and
cognitive change after FDR correction. Higher DTI diffusivity and FW, and lower FA
TDF
at baseline
throughout the WM were significantly associated with increases in cognitive impairment (Figure 2a). As
shown in Figure 2b, lower FA
TDF
in 15 ROIs (p≤critical FDR p=0.030; ALIC largest r=-0.43), higher
MD in 16 ROIs (p≤0.027; SFO largest r=0.54), higher RD in 14 ROIs (p≤0.028; SFO largest r=0.51),
higher AxD in 11 ROIs (p≤0.020; SS largest r=0.47), and higher FW in 16 ROIs (p≤0.025; SFO largest
r=0.54) were significantly associated with increases in GDS. MD in the SFO showed the largest effect
sizes overall. A post-hoc analysis tested whether the largest regional effects differed significantly from
the global full WM effect, by including the full WM as an additional covariate. SFO MD (p=0.011;
r=0.39), RD (p=0.020; r=0.36), and FW (p=0.011; r=0.39) remained significant. Compared to the full
WM the SFO, FX/ST, PCR, RLIC and SLF most frequently showed larger effect sizes (i.e., 4 out of the 5
indices).
Comparisons between the ‘best model’ (MFC Glx/Cr and full WM dMRI indices) to models
including 1 local WM ROI revealed that the model with MFC Glx/Cr and SFO MD was the best predictor
of change in GDS (AIC: -5.33; adjusted R
2
:0.37), which was not different from SFO AxD or FW. However
it was significantly better than models including full WM dMRI measures.

162

Figure 2. Effect sizes (r-value) of associations between mean dMRI measures in 21 ROIs and change in global
deficit scores. (a) Brain maps of effect sizes in ROIs significant at the FDR critical p-value which are demarcated
by the dotted line in (b). Lower FA
TDF
and higher DTI diffusivity (MD/RD/AxD) and FW at baseline were
associated with faster rates of cognitive decline, while FA
DTI
did not survive multiple comparisons correction.
163

Figure 3. Goodness of fit (adjusted R
2
) comparison of predictive models that include MFC Glx/Cr and one regional
dMRI measure (covarying for age and sex). An AIC 2 or 3 units lower than our reference ‘best model’—MFC
Glx/Cr + full WM AxD— is dilineated by dotted and solid lines respectively.

4.1.4 DISCUSSION

Chronically HIV-infected individuals may be at risk for brain injury and progressive NCI despite cART.
Identifying predictors of cognitive decline is important for forming a better understanding of disease
neuroprogression, and may provide tools and markers to empower future studies that aim to monitor
disease progression and treatment effects. In a cohort of HIV-infected participants on stable cART we
found: 1) a significant longitudinal increase in NCI, and 2) that the best predictive models of cognitive
change were those that included both frontal lobe dMRI diffusivity and MRS Glx/Cr neuroimaging
164

measures. Multimodal approaches, that include MRI measures, may therefore help us better understand
HIV-associated neuropathology and help identify those at risk for developing NCI.
MRI measures may offer better predictive power over baseline neuropsychiatric assessments,
which may reflect factors such as cognitive reserve (Tucker-Drob et al. 2009), or plasma measures, in line
with reports suggesting they but may no longer be strongly associated NCI in the cART era (Clifford and
Ances 2013). Moreover, metabolite and WM microstructural brain changes may precede gross volumetric
changes and cognitive change. This is in line with previous studies that report HIV-related differences in
MRS markers of glial function (Wu et al. 2015) and dMRI WM microstructure (Filippi et al. 2001, Pomara
et al. 2001) in the absence of differences in measures derived from anatomical MRI. Predictive models
including both MRS and dMRI measures significantly improved model fit, suggesting these modalities
offer complementary information about HIV-related brain processes that may contribute to NCI.
When evaluated individually, across measures the best predictor of cognitive decline was reduced
levels of glutamate/glutamine (Glx/Cr) in the mid-line frontal cortex GM, followed by mid-frontal WM.
Another study similarly found that lower frontal GM glutamate was associated with greater cognitive
impairment (Ernst et al. 2010). Excessive extracellular concentrations of glutamate, due to increased
production by HIV-infected macrophages and decreased reuptake by activated astrocytes, can lead to
excitotoxicity and contribute to brain injury in HIV (Kaul et al. 2001, Wang et al. 2004, Erdmann et al.
2007). Ernst et al. (2010) suggest that reduced glutamate findings may be driven by 1) an increased
demand for amino-acid precursors for synthetic pathways that, for example, repair HIV-related damage
to cell membranes; 2) an overall loss of GM and/or glutamatergic neurons; or 3) reduced astrocyte
reuptake and incomplete recycling of glutamate which may, in turn, reduce the total amount of
intracellular glutamate from which the MRS glutamate signal is largely driven. Reduced levels of Glx in
the frontal lobe of neuro-asymptomatic HIV+ individuals compared to controls have also been reported
(Harezlak et al. 2011). Associations between reduced Glx and both detectable plasma HIV RNA
concentrations and low nadir CD4+ counts in neuro-asymptomatic individuals point to the role of
glutamate/glutamine in HIV-related neuropathogenesis (Harezlak et al. 2014), while reported longitudinal
Glx decreases in these individuals suggest that changes in Glx may reflect early HIV-related changes in
the brain prior to the onset of cognitive symptoms (Gongvatana et al. 2013).
After Glx/Cr, dMRI measures were the next best predictors. As expected lower anisotropy and
greater diffusivity and free water throughout the WM at baseline were associated with an increase in
impairment. While the full WM was associated with cognitive decline, anatomically refined ROIs showed
165

larger effects and greater predictive power than the full WM; in particular, models including the superior
frontal-occipital fasciculus (SFO), a region delineated in the medial frontal WM in the JHU atlas, showed
the best model fit. The largest dMRI effects were detected with DTI AxD and MD, followed closely by
FW, although all measures were highly correlated. DTI derived MD is a non-specific average of radial
and axial diffusivities, however, histological rodent studies suggest that higher AxD may be correlated
with axonal damage and degeneration (Kinoshita et al. 1999, Song et al. 2003), or reduced axonal diameter
(Schwartz et al. 2005), as opposed to factors such as dys- or de-myelination that have been correlated with
higher RD (Nair et al. 2005, Harsan et al. 2007). DTI is widely recognized as a useful tool for studying
neurodegenerative disorders, but the single tensor model, cannot differentiate partial-volumes from
different tissue compartments present in a single voxel, or even crossing fibers which are estimated to
occur in up to 90% of WM voxels at the typical dMRI resolution (Tuch et al. 2002, Descoteaux et al.
2009, Jbabdi et al. 2010, Jeurissen et al. 2013). Models beyond DTI may offer further sensitivity or insight
into underlying pathology. Higher FW, for example, may reflect HIV-related inflammation and resulting
edema (Pasternak et al. 2009, Pasternak et al. 2012, Salminen et al. 2016). While the most widely used
measure of WM microstructure, FA
DTI
, was not significantly associated cognitive changes, FA
TDF
showed
widespread associations, similar to those detected with diffusivity and FW measures.
Given the recent evidence of progressive brain injury in chronically infected and treated HIV+
individuals, there will be a great need for novel therapies and for markers that could be used to enrich
patient selection or track treatment response. As in Alzheimer’s disease, where altered MRI markers have
been shown to precede clinical presentation of the disease (Jack et al. 2013), the HIV field is now poised
to identify imaging biomarkers that predict or correlate with decline in cognitive function, and that may
be used as measures of disease progression. Future studies are needed to evaluate whether our results
generalize to other HIV cohorts and to further evaluate these predictors for specific cognitive domains.

4.1.5 ACKNOWLEDGEMEMNTS

The study was funded by NIH NINDS R01 NS080655 and U54 EB020403.

166

4.1.6 SUPPLEMENTARY APPENDIX

A. Supplementary Methods

Supplementary Table S1. Neuropsychological evaluations in 7 domains as described in (Gongvatana et al. 2013).
Cognitive Domain Tests
Processing Speed
WAIS-III Digit Symbol
WAIS-III Symbol Search
Trail Making: Part A

Verbal Fluency
Controlled Oral Word Association (FAS)
Animal Naming - Correct Words
Action Fluency - Correct Words

Executive Function
Trail Making: Part B
Stroop Test

Working Memory
Paced Auditory Serial Addition Task
WAIS-III Letter Number Sequencing

Learning
Hopkins Verbal Learning Test - Immediate Recall
Benton Visual Memory Test - Immediate Recall

Memory
Hopkins Verbal Learning Test - Delayed Recall
Benton Visual Memory Test - Delayed Recall

Psychomotor Grooved Pegboard

B. Supplementary Results

Supplementary Table S2. Pearson’s correlation coefficients between dMRI, MRS, and TBM measures
associated with change in GDS (n=39).

MD RD AxD FW FA
TDF

BG
NAA/Cr
FWM
Glx/Cr
MFC
Glx/Cr
Cort.
GM
Lat.
Ventricle
MD 1 0.99 0.94 0.99 -0.97 -0.32 -0.11 -0.17 -0.13 0.74
RD 0.99 1 0.88 0.99 -0.98 -0.32 -0.15 -0.19 -0.18 0.71
AxD 0.94 0.88 1 0.93 -0.86 -0.30 -0.03 -0.10 -0.02 0.76
FW 0.99 0.99 0.93 1 -0.97 -0.32 -0.12 -0.16 -0.16 0.70
FA
TDF
-0.97 -0.98 -0.86 -0.97 1 0.34 0.16 0.16 0.20 -0.65
BG NAA/Cr -0.32 -0.32 -0.30 -0.32 0.34 1 0.28 0.33 0.12 -0.22
FWM Glx/Cr -0.11 -0.15 -0.03 -0.12 0.16 0.28 1 0.56 0.02 -0.02
MFC Glx/Cr -0.17 -0.19 -0.10 -0.16 0.16 0.33 0.56 1 0.25 -0.12
Cort. GM -0.13 -0.18 -0.02 -0.16 0.20 0.12 0.02 0.25 1 -0.15
Lat. Ventricle 0.74 0.71 0.76 0.70 -0.65 -0.22 -0.02 -0.12 -0.15 1

167

Supplementary Table S3. Mean statistics ± 1.96 x standard error (95% confidence interval) for baseline cognitive,
dMRI, MRS, and TBM measures associated with the change in GDS, computed on 500 random samples generated
by bootstrapping from the full and matched samples.
Modality
r-value AIC Adj. R
2

N=39 N=44 N=39 N=44 N=39 N=44
MRS

BG NAA/Cr -0.36 ±0.013 -- -0.72 ±1.15 -- 0.19 ±0.011 --
FWM Glx/Cr -0.39 ±0.011 -- -1.49 ±1.06 -- 0.21 ±0.011 --
MFC Glx/Cr -0.44 ±0.012 -- -3.70 ±1.06 -- 0.25 ±0.011 --
TBM

Lat. Ventricle 0.30 ±0.014 0.31±0.013 0.77±1.23 -0.98 ±1.22 0.16 ±0.013 0.18±0.013
Cort. GM 0.34±0.012 0.40±0.011 -0.11 ±1.16 -3.69 ±1.20 0.18±0.011 0.23±0.011
dMRI

MD 0.38 ±0.016 0.39 ±0.015 -2.38 ±1.21 -4.42 ±1.22 0.21 ±0.016 0.23 ±0.015
RD 0.37 ±0.016 0.37 ±0.015 -1.87 ±1.20 -3.77 ±1.21 0.21 ±0.016 0.22 ±0.015
AxD 0.39 ±0.016 0.40 ±0.015 -2.61 ±1.20 -4.79 ±1.20 0.22 ±0.016 0.24 ±0.015
FW 0.37 ±0.016 0.38 ±0.015 -2.25 ±1.18 -4.32 ±1.20 0.21 ±0.016 0.23 ±0.015
FA
TDF
-0.36 ±0.016 -0.36 ±0.015 -1.48 ±1.16 -3.10 ±1.18 0.20 ±0.015 0.21 ±0.014

Supplementary Table S4. Swapping dMRI indices in the best and most parsimonious predictive model shows no
significant difference in model fit.
Models AIC Adj. R
2

AxD + MFC Glx/Cr -0.14 0.28
FW + MFC Glx/Cr 0.38 0.27
MD + MFC Glx/Cr 0.39 0.27
FA

+ MFC Glx/Cr 0.86 0.26
RD + MFC Glx/Cr 0.94 0.26

168

CHAPTER 5

Future Work: Advanced Diffusion MRI Techniques

169

5.1 Multi-shell Diffusion MRI Measures of Brain Aging

This section is adapted from:

Nir TM, Thomopoulos SI, Villalon-Reina JE, Zavaliangos-Petropulu A, Reid RI, Bernstein MA,
Borowski B, Jack Jr CR, Weiner MW, Jahanshad N, Thompson PM (2019). Multi-shell diffusion MRI
measures of brain aging: A preliminary comparison from ADNI3. 2019 IEEE International Symposium
on Biomedical Imaging (ISBI). © 2019 IEEE. Reprinted, with permission.

170

Multi-shell Diffusion MRI Measures of Brain Aging: A Preliminary Comparison from ADNI3

Talia M. Nir
1
, Sophia I. Thomopoulos
1
, Julio E. Villalon-Reina
1
, Artemis Zavaliangos-Petropulu
1
, Robert I. Reid
2
,
Matt A. Bernstein
3
, Bret Borowski
3
, Clifford R. Jack, Jr.
3
, Michael W. Weiner
4
,
Neda Jahanshad
1
, Paul M. Thompson
1
, for the Alzheimer’s Disease Neuroimaging Initiative (ADNI)

1
Imaging Genetics Center, Mark & Mary Stevens Neuroimaging & Informatics Institute, Keck School of Medicine, University of Southern
California, Marina del Rey, CA, USA
2
Department of Information Technology, Mayo Clinic and Foundation, Rochester, MN, USA
3
Department of Radiology, Mayo Clinic and Foundation, Rochester, MN, USA
4
Department of Radiology, University of California San Francisco School of Medicine, San Francisco, CA, USA

Abstract. The Alzheimer’s Disease Neuroimaging Initiative (phase 3; ADNI3) is collecting
multisite diffusion MRI (dMRI) data using protocols optimized for different scanner vendors,
including one multi-shell protocol, to better understand disease effects. Here, we analyzed multi-
shell scans from 56 ADNI3 participants (age: 74.3±7.5 yrs; 17 female / 49 male). We evaluated
whether multi-shell dMRI measures computed from neurite orientation dispersion and density
imaging (NODDI), and diffusion kurtosis imaging (DKI) differentiated people with mild cognitive
impairment from healthy controls with higher sensitivity than standard diffusion tensor imaging
(DTI) measures. We also assessed the effects of various multi-shell derived dMRI samples on the
sensitivity of DTI measures. While we did not identify large differences in effect sizes among
tensor-based, NODDI, or DKI measures, we did detect greater effect sizes from DTI measures
estimated using multi-shell data converted to single-shell HARDI compared to those fit using the
subset of 48 b=1000 s/mm
2
volumes, typical of single-shell DTI.

Keywords: ADNI3, DTI, Multi-shell, NODDI, DKI

171

5.1.1 INTRODUCTION

The Alzheimer’s Disease Neuroimaging Initiative (ADNI) is an ongoing large-scale, multi-center,
longitudinal study, to improve methods for clinical trials by identifying brain imaging, clinical, cognitive,
and molecular biomarkers of AD and aging. In particular, identifying biomarkers sensitive to mild
cognitive impairment (MCI) is important to better categorize the transitional stages between normal aging
and AD, and to evaluate targeted treatments. To date, amyloid and tau biomarkers from CSF and PET, are
perhaps the most widely accepted biomarkers of AD and MCI for epidemiological, genetic studies, and
for clinical trials. Still, there is growing interest in diffusion (dMRI) measures, which offer additional
sensitivity to microstructural properties not evident on standard anatomical MRI.
The first phase of ADNI acquired dMRI data from approximately one third of enrolled participants
on GE scanners. In its third phase, ADNI3 has incorporated dMRI protocols for 3 T Siemens, Philips, and
GE scanners to allow dMRI data collection at all sites. The available scanners span a wide range of
software capabilities, such as support for custom diffusion gradient tables and/or simultaneous multi-slice
acceleration. Including additional scanners in ADNI3, while accommodating scanner restrictions and
limiting scan duration to 7-10 minutes, resulted in data acquired with 7 different acquisition protocols
(http://adni.loni.usc.edu/methods/documents/mri-protocols/).
Of the seven protocols, there is currently only one multi-shell protocol —for sites with Siemens
Advanced Prisma scanners. Several AD-related studies have already used multi-shell protocols to compute
diffusion measures from models that do not assume mono-exponential decay, such as diffusion kurtosis
imaging (DKI) (Jensen et al. 2005, Yuan et al. 2016), and multi-compartment models such as neurite
orientation dispersion and density imaging (NODDI) (Zhang et al. 2012, Colgan et al. 2016). Such multi-
shell biophysical and advanced signal based models of water diffusion in brain tissue may allow us to
relate diffusion signals directly to underlying microstructure and resolve multiple dominant fiber
directions, potentially offering greater specificity and sensitivity to disease processes.
Multi-shell scans are now available for 56 of over 600 scanned ADNI3 participants. To boost
power and pool data across all ADNI3 protocols and participants, single-shell models need to be evaluated
as well. The single-tensor model (DTI) (Basser et al. 1994) is still the most widely used to detect alterations
in white matter (WM) micro-architecture in clinical populations, despite known limitations in regions with
crossing fibers and the lack of specificity to characterize microstructural environments. For DTI, the
optimal b-value lies between b=900-1200 s/mm
2
(Kingsley and Monahan 2004, Alexander and Barker
172

2005), and at least 30 unique sampling orientations are recommended to robustly estimate anisotropy,
mean diffusivity, and tensor orientations (Jones 2004). Frequently, such studies use simulations or
phantom data, which are subject to different sources of noise than human clinical data and may not be
directly comparable. In practice, including sampling orientations from the other shells may help boost the
signal to noise ratio (SNR), improving sensitivity to disease related effects. One recent approach, proposed
by Yeh and Verstynen (2016), converts multi-shell data to single-shell high angular resolution diffusion
imaging (HARDI). In (Yeh and Verstynen 2016), diffusion signals and dMRI measures from the
converted data were highly correlated with those collected from a true HARDI acquisition. The clinical
utility of this method, however, has not yet been evaluated.
Here, we analyzed dMRI data from 56 ADNI3 participants scanned with a multi-shell protocol to
evaluate the sensitivity of various microstructural measures to detect differences between cognitively
normal (CN) elderly individuals and those with MCI. We hypothesized that multi-shell dMRI measures
derived from NODDI and DKI models would confer higher sensitivity than DTI, and that DTI measure
effects sizes would depend on how multi-shell data are used for single tensor estimation— i.e., using a
subset of b-values, the entire acquisition, or conversion to HARDI, as in Yeh and Verstynen (2016).

5.1.2 METHODS

Subjects and Image Acquisition

Baseline MRI, dMRI, diagnosis, and demographic data were downloaded from the ADNI database
(https://ida.loni.usc.edu/). Here we analyzed data from the 56 participants scanned with a multi-shell
dMRI protocol: 39 cognitively normal elderly controls (CN; mean age: 73.2±7.2 yrs; 25M/14F) and 17
with mild cognitive impairment (MCI; mean age: 76.8±7.5 yrs; 14M/3F). (Data collection for the study is
ongoing, and this constitutes an initial report.)
All subjects underwent whole-brain MRI scanning on 3 T Siemens Advanced Prisma scanners at
9 acquisition sites across North America. Anatomical T1-weighted MPRAGE sequences (256x256 matrix;
voxel size = 1.0x1.0x1.0 mm; TI=900 ms; TR = 2300 ms; TE = 2.98 ms; flip angle=9°), and multi-shell
multiband dMRI (116x116 matrix; voxel size: 2x2x2 mm; TR=3400 ms; scan time = 7.25 min) were
collected. 127 separate images were acquired for each dMRI scan: 13 b0 images, 48 b=1000 s/mm
2

diffusion-weighted images (DWI), 6 b=500 s/mm
2
DWI, and 60 b=2000 s/mm
2
DWI.
173

For comparison with dMRI measures, hippocampal volume measures, processed by the UCSF
team using the FreeSurfer package pipeline (Fischl et al. 2004), were also downloaded from the ADNI
database.

Image Preprocessing

DWI images were denoised using the LPCA filter (Manjon et al. 2013) and corrected for Gibbs ringing
with MRtrix (Kellner et al. 2016). Extra-cerebral tissue was removed, and eddy correction performed with
FSL’s EDDY tool (Andersson and Sotiropoulos 2016) including repol outlier replacement (Andersson et
al. 2016). DWI then underwent B1 field inhomogeneity corrections (Tustison et al. 2010). T1-weighted
images were preprocessed using the standard FreeSurfer pipeline (Fischl et al. 2004). Corrected b0 images
were linearly aligned to resulting FreeSurfer corrected T1 images, using the WM mask for FSL FLIRT’s
boundary-based registration (BBR) (Greve and Fischl 2009); the registration was subsequently inverted
to bring the T1 to the native DWI space. Diffusion images were then non-linearly warped to their
respective T1-weighted scans with ANTs (Avants et al. 2011) to correct for EPI-induced susceptibility
artifacts. Diffusion gradient directions were rotated to account for linear registrations.

dMRI Reconstruction Models and Scalar Measures

Fractional anisotropy (FA1000) and mean diffusivity (MD1000) were estimated by fitting a single diffusion
tensor (DTI) (Basser et al. 1994) to the subset of 48 b=1000 s/mm
2
DWI volumes. To compare
performance with a measure derived from a HARDI technique—non-parametric q-ball imaging (QBI)—
we also fit constant solid angle orientation distribution functions (CSA-ODFs) (Tuch 2004, Aganj et al.
2010) using the DiPy package (Garyfallidis et al. 2014) to generate generalized FA (GFA1000) maps.
Two multi-shell models were fit across all shells. Three diffusion kurtosis imaging (DKI) measures
(Jensen et al. 2005) were estimated using the DiPy package: axial kurtosis (AK), radial kurtosis (RK), and
mean kurtosis (MK). Neurite orientation dispersion and density imaging (NODDI) (Zhang et al. 2012)
was fit using the AMICO toolbox (Daducci et al. 2015) yielding maps of the orientation dispersion index
(ODI), intra-cellular volume fraction (ICVF), and isotropic volume fraction (ISOVF).

174

Multi-shell Samples for Single-shell Measures

For comparison with measures derived from the subset of 48 b=1000 s/mm
2
DWI volumes, DTI and QBI
CSA-ODFs were subsequently fit to the full multi-shell dMRI acquisition with b=500, 1000, and 2000
s/mm
2
and FAALL, MDALL and GFAALL maps estimated.
Additionally, multi-shell dMRI were interpolated into corresponding single-shell HARDI using
the generalized q-sampling method (GQI) framework proposed in Yeh and Verstynen (2016), available
through DSI Studio. The GQI model provides a linear relation between dMRI signal and the diffusing
spins via a spin density function (SDF) (Yeh et al. 2010), consequently enabling a direct conversion
between spins and dMRI signals acquired from shell or grid sampling schemes. Here, multi-shell data
were converted to single-shell HARDI data with b=1000 s/mm
2
and again FAHARDI, MDHARDI and
GFAHARDI maps were estimated.

White Matter Region of Interest Measures

Using publically available ENIGMA-DTI protocols (http://enigma.usc.edu/protocols/dti-protocols/)
(Jahanshad et al. 2013a), each subject’s FA map was warped to the ENIGMA-DTI FA template with
ANTs (Avants et al. 2011) and the transformations applied to all respective dMRI scalar maps. DTI, QBI,
DKI and NODDI measures were projected onto the ENIGMA-DTI template skeleton with TBSS (Smith
et al. 2006). Using the JHU WM atlas (Mori et al. 2008), mean skeletonized measures were extracted from
25 regions of interest (ROIs; Table 1).

Table 1. Index of 25 JHU atlas WM ROIs analyzed.
CST Corticospinal tract FX Fornix (body)
IC Internal capsule SLF Sup. longitudinal fasciculus
ALIC Ant. limb of IC SFO Sup. fronto-occipital fasciculus
PLIC Post. limb of IC IFO Inf. fronto-occipital fasciculus
RLIC Retrolenticular part of IC SS Sagittal stratum
PTR Post. thalamic radiation Fx/ST Fornix (crus)/Stria terminalis
CR Corona radiata UNC Uncinate fasciculus
ACR Ant. corona radiata CC Full corpus callosum
SCR Sup. corona radiata GCC Genu of CC
PCR Post.corona radiata BCC Body of CC
CGC Cingulum (cingulate) SCC Splenium of CC
CGH Cingulum (hippocampal) Full WM Full white matter
EC External capsule

175

Statistical Analyses

Random-effects linear regressions were used to test for associations between MCI diagnosis and mean
dMRI measures in each ROI or hippocampal volume, covarying for age, sex and their interaction, and
grouping the data by acquisition site. Effect sizes were estimated with Cohen’s d. The false discovery rate
(FDR) procedure was used to correct for multiple comparisons across ROIs (q=0.05) (Benjamini and
Hochberg 1995).

5.1.3 RESULTS

Multi-shell versus Single-shell dMRI Measures

Compared to CN participants, the MCI group showed significantly lower FA1000, GFA1000, AK, RK, MK,
and ICVF, and higher MD1000 (FDR q<0.05; Figure 1). No significant associations were detected with
ODI. The largest effect size across measures and ROIs was detected with MD in the Full WM (d = 1.72).
Except for ISOVF, every measure significantly associated with MCI included the Full WM ROI (Figure
2). For those measures where the Full WM was significant, secondary analyses tested for significant
regional associations when the Full WM was included as an additional covariate. Except for a now positive
association between ICVF in the body of the corpus callosum (BCC) and MCI (i.e., less “neuronal loss”
in the BCC than the rest of the WM), ROI associations were no longer significant indicating that no
regional effect differed significantly from the global effect detected in the Full WM. dMRI effect sizes in
the full WM exceeded those detected with hippocampal volume in the same subjects (d = -0.95).

176

Figure 1. Absolute value of effect sizes for associations between MCI diagnosis and single and multi-shell dMRI
measures from the DTI (blue colors), QBI (orange), DKI (reds) and NODDI (greens) models. Significant
associations (FDR q<0.05) are indicated by filled circles.

Single-shell Measures from Multi-shell Data

MD1000, MDALL, and MDHARDI all showed significant associations in 24 of 25 ROIs (FDR q<0.05), and
effect sizes were comparable (within the standard error). FA1000 and FAALL showed similar effect sizes,
and were significantly associated with 11 and 9 ROIs respectively. FAHARDI, like MD, was significantly
associated with MCI diagnosis in 24 ROIs and showed larger effect sizes (Figure 3a). Similarly, GFA1000
and GFAALL showed similar effect sizes, and were significantly associated with diagnosis in 13 and 11
ROIs respectively, but GFAHARDI was associated with diagnosis in 22 ROIs and again offered larger effect
sizes (Figure 3b).

177

Figure 2. Absolute value of effect sizes (and standard error bars) for associations between MCI diagnosis and dMRI
measures in the Full WM; the effect size for hippocampal volume (mean of left and right) is shown for comparison.
For each measure, the total number of ROIs significantly associated with diagnosis is also noted, as a measure of
the extent of effects (we note this is just a heuristic as it depends on the available sample size).

Figure 3. Effect sizes (and standard error bars) for associations between MCI diagnosis and (a) DTI and (b) QBI
derived fractional anisotropy measures.

178

5.1.4 DISCUSSION

This study has three main findings: 1) Compared to DKI and NODDI multi-shell measures, DTI MD in
the Full WM showed similar if not larger effect sizes for detecting differences between CN and MCI
diagnostic groups; 2) dMRI measure effect sizes exceeded those detected with hippocampal volume, a
more standard anatomical measure; 3) Compared to single-shell measures fit on only the subset of 48
b=1000 s/mm
2
sampling directions, more pervasive associations with larger effect sizes were detected for
QBI GFA and DTI FA when fit to multi-shell data converted to single-shell HARDI.
DTI is widely used to study neurodegenerative disorders such as AD; lower FA and higher MD
are still the most frequently reported dMRI measures in studies of AD. The DTI model, however, cannot
differentiate crossing fibers, and captures partial volumes of different tissue compartments (e.g., intra- and
extra-cellular compartments, ‘free water’ from CSF or inflammation, and myelin volume fraction). This
may reduce the sensitivity and specificity of single tensor measures. While multi-shell models such as
DKI and NODDI may overcome some of these limitations, data from clinical studies, including most sites
in ADNI3, may be limited to single-shell acquisitions, constraining the available models to simpler models
such as DTI. A number of AD-related pathological processes are thought to drive changes in DTI measures
including demyelination and axonal degeneration, gliosis, neuro-inflammation and swelling (Laurent et
al. 2018). Biophysical models such as NODDI may offer greater insight into underlying pathology (e.g.,
in our study, higher MD in the Full WM detected in MCI may in part be driven by neuronal loss indicated
by lower ICVF, the second largest Full WM effect size detected). Even so, our findings so far suggest that
similar if not greater effect sizes to distinguish patient groups (i.e., MCI vs. CN) may be obtained with
DTI measures, perhaps because they collapse all pathological changes in the WM under a Gaussian
modeling frame. There are, however, several implementations of NODDI (e.g., NODDI Toobox and
Dmipy ), and further comparisons are necessary. In addition, future work should focus on the ICVF, with
which the second largest Full WM effect size was detected. Here, as in the original NODDI model, the
‘Stick’ model (Behrens et al. 2003) was used, but several other intra-axonal diffusion models have been
proposed (e.g., ‘Soderman’ (Söderman and Jönsson 1995) and ‘Callaghan’ (Callaghan 1995)).
In multi-site studies such as ADNI3, where data from single and multi-shell dMRI are combined
or compared, the question remains how best to use multi-shell data for fitting mono-exponential models
such as QBI and DTI. While theoretically it may suffice to fit DTI with 30 volumes with b=1000 s/mm
2
,
we found that converting full multi-shell acquisitions to single-shell HARDI data using the framework
179

proposed by Yeh et al. (2016) improved sensitivity to detect disease related effects: the largest anisotropy
effect sizes were detected with the multi-shell data when it was converted to HARDI. GFAHARDI and
FAHARDI detected the largest effect sizes in the hippocampal cingulum and fornix (crus) / stria terminalis
region (followed by the full WM), WM bundles connecting hippocampal and parahippocampal regions to
the rest of the brain, consistent with patterns of AD pathology.
As ADNI3 progresses and as sample sizes grow, future studies are needed to verify which
measures offer greater diagnostic sensitivity as well as better microstructural interpretability.

5.1.5 ACKNOWLDEGEMENTS

Data collection and sharing for ADNI was funded by NIH grant U01 AG024904 and DOD award
W81XWH-12-2-0012. Additional support was provided by NIA grant RF1 AG04191 and P41 EB015922.
In reference to IEEE copyrighted material which is used with permission in this thesis, the IEEE does not
endorse any of the Univeristy of Southern California’s products or services. Internal or personal use of
this material is permitted. If interested in reprinting/republishing IEEE copyrighted material for
advertising or promotional purposes or for creating new collective works for resale or redistribution,
please go to http://www.ieee.org/publications_standards/publications/rights/rights_link.html to learn how
to obtain a License from RightsLink. If applicable, University Microfilms and/or ProQuest Library, or the
Archives of Canada may supply single copies of the dissertation.

180

5.2 Connectivity Network Measures Predict Brain Atrophy in Aging and AD

This section is adapted from:

Nir TM, Jahanshad N, Toga AW, Jack Jr CR, Weiner MW, Thompson PM, ADNI (2015). Connectivity
network measures predict volumetric atrophy in mild cognitive impairment. Neurobiology of Aging,
36(Suppl 1):S113-S120.

181

Connectivity Network Measures Predict Volumetric Atrophy in Mild Cognitive Impairment

Talia M. Nir
1
, Neda Jahanshad
1
, Arthur W. Toga
1
, Matt A. Bernstein
2
, Clifford R. Jack Jr.
2
, Michael W. Weiner
3
,
Paul M. Thompson
1,4
, for the Alzheimer’s Disease Neuroimaging Initiative (ADNI)

1
Imaging Genetics Center, Institute for Neuroimaging & Informatics, University of Southern California, Los Angeles, CA, USA
2
Department of Radiology, Mayo Clinic and Foundation, Rochester, MN, USA
3
Department of Radiology and Biomedical Imaging, UCSF School of Medicine, San Francisco, CA, USA
4
Departments of Neurology, Psychiatry, Radiology, Engineering, Pediatrics, and Ophthalmology, University of Southern California, Los
Angeles, CA, USA

Abstract. Alzheimer’s disease (AD) is characterized by cortical atrophy and disrupted anatomical
connectivity, and leads to abnormal interactions between neural systems. Diffusion weighted
imaging (DWI) and graph theory can be used to evaluate major brain networks, and detect signs
of a breakdown in network connectivity. In a longitudinal study using both DWI and standard
MRI, we assessed baseline white matter connectivity patterns in 30 subjects with mild cognitive
impairment (MCI; mean age: 71.8+/-7.5 yrs; 18 male/12 female) from the Alzheimer's Disease
Neuroimaging Initiative (ADNI). Using both standard MRI-based cortical parcellations and
whole-brain tractography, we computed baseline connectivity maps from which we calculated
global “small-world” architecture measures, including mean clustering coefficient (MCC) and
characteristic path length (CPL). We evaluated whether these baseline network measures predicted
future volumetric brain atrophy in MCI subjects, who are at risk for developing AD, as determined
by 3D Jacobian “expansion factor maps” between baseline and 6-month follow-up anatomical
scans. This study suggests that DWI-based network measures may be a novel predictor of AD
progression.

Keywords: Graph Theory, Brain Networks, White Matter, DTI, Tractography, ADNI, TBM,
Small Worldness, Connectivity.

182

5.2.1 INTRODUCTION

Alzheimer’s disease (AD), the most common form of dementia, is characterized by memory loss in its
early stages, typically followed by a progressive decline in other cognitive domains. People with mild
cognitive impairment (MCI) - a transitional stage between normal aging and AD - convert to AD at a rate
of about 10–15% per year (Petersen et al. 2001, Bruscoli and Lovestone 2004). The Alzheimer's Disease
Neuroimaging Initiative (ADNI) is one of several major efforts worldwide to identify sensitive biomarkers
that may help track or predict brain tissue loss due to AD progression.
AD is marked by pervasive gray matter atrophy, but the brain’s white matter (WM) pathways also
progressively decline (Braak and Braak 1996, Bartzokis 2011, Braskie et al. 2011, Hua et al. 2013b).
Recent models of AD suggest that cognitive deficits arise from the progressive disconnection of cortical
and subcortical regions, promoted by neuronal loss and white matter injury (Delbeuck et al. 2003, Pievani
et al. 2011). Many MRI-based image analysis methods have been used to track structural atrophy of the
brain, but diffusion tensor imaging (DTI) is sensitive to microscopic WM injury not always detectable
with standard anatomical MRI. DTI may be used to track the highly anisotropic diffusion of water along
axons, revealing microstructural WM fiber bundles connecting cortical and subcortical regions and
allowing for characterization of the brain’s WM structural network (Hagmann et al. 2008).
Graph theory network topology measures have been used increasingly to analyze brain networks
and characterize network organization. “Small-world” network properties have been regarded as typical
properties of many kinds of communication networks, and are found in social networks, efficient
biological networks, and even in healthy mammalian brain networks (Hilgetag et al. 2000, Achard and
Bullmore 2007, Reijneveld et al. 2007, Iturria-Medina et al. 2008). Networks with a small-world
organization can have both functional segregation and specialization of modules and a ‘low wiring cost’
that supports easy communication across an entire network. Small-world networks are marked by low
characteristic path length (CPL) and high mean clustering coefficient (MCC), so they are both integrated
and segregated. Studies using various modalities, including cortical thickness analyses, fMRI, and EEG,
suggest that AD patients have abnormal small-world architecture in their large-scale structural and
functional brain networks, with differences in MCC and CPL that may imply less optimal network
topology (Stam et al. 2007, He et al. 2008, Sanz-Arigita et al. 2010, Brown et al. 2011, Toga and
Thompson 2013).
183

In this study, we assessed 30 ADNI participants showing signs of mild cognitive impairment
(MCI). MCI subjects are the target of many clinical trials that aim to slow disease progression, before
brain changes are so pervasive that they are irremediable. However, predictors of decline in MCI are sorely
needed, as mildly impaired subjects do not usually exhibit drastic changes in most standard biomarkers of
AD. Here, we combined DTI with longitudinally acquired standard anatomical MRI (across a 6-month
interval) to measure the microstructure and connectivity of white matter tracts, and assess whether
variations in the degree and extent of connections might predict future brain decline. We created 68×68
structural connectivity matrices, or graphs, that describe the strength of connections between any pair of
brain regions based on baseline structural cortical parcellations and whole-brain tractography. In these
graphs, nodes designate brain regions, which are thought of as being connected by edges representing
WM fibers. We then used graph theory to describe general properties of the anatomical networks and to
characterize connectivity patterns.
Given the recent interest in “small world” phenomena as a characteristic of biological networks,
we examined whether global “small-world architecture” network measures, MCC and CPL, calculated
from baseline connectivity maps were able to predict future volumetric brain atrophy (dynamic tissue loss)
over a 6-month follow-up period, as determined by 3D Jacobian “expansion factor maps” of T1-weighted
structural scans. That is, we tested whether the intactness of the brain’s anatomical network predicted
ongoing brain decline in the future, assessed using the more accepted anatomical MRI methods. In follow-
up analyses, we additionally assessed whether several baseline local nodal measures (efficiency, clustering
and betweenness) were associated with volumetric brain atrophy. We found that global and nodal network
measures may offer a potentially useful biomarker for predicting longitudinal atrophy, at this critical time
before the onset of AD.

5.2.2 METHODS

Subject Information and Image Acquisition

Data collection for the ADNI2 project (the second phase of ADNI) is still in progress. Here we performed
an initial analysis of 30 MCI subjects who had returned for a follow-up evaluation at 6-months (mean age
at baseline: 71.8 ± 7.5 yrs; 18 men / 12 women). We note that in ADNI2 MCI participants include the
enrollment of a new early MCI (e-MCI) cohort, with milder episodic memory impairment than the MCI
184

group of ADNI1, now called late MCI (l-MCI) in ADNI2 (Table 1). We additionally analyzed baseline
data from 29 cognitively healthy control subjects (CTL) to create a study-specific brain template (mean
age at baseline: 73.4 ± 5.2 yrs; 15 men/14 women). Detailed inclusion and exclusion criteria are found in
the ADNI2 protocol (http://adni-info.org/Scientists/Pdfs/ADNI2_Protocol_FINAL_20100917.pdf).
All subjects underwent whole-brain MRI scanning on 3-Tesla GE Medical Systems scanners, on
at least one of two occasions (baseline and 6 months). T1-weighted IR-FSPGR (spoiled gradient echo)
sequences (256×256 matrix; voxel size = 1.2×1.0×1.0 mm
3
; TI = 400 ms; TR = 6.98 ms; TE = 2.85 ms;
flip angle = 11°), were collected as well as diffusion-weighted images (DWI; 35 cm field of view, 128×128
acquired matrix, reconstructed to a 256×256 matrix; voxel size: 2.7×2.7×2.7 mm
3
; scan time = 9 min;
more imaging details may be found at http://adni.loni.usc.edu/wp-
content/uploads/2010/05/ADNI2_GE_3T_22.0_T2.pdf). 46 separate images were acquired for each DTI
scan: 5 T2-weighted images with no dedicated diffusion sensitization (b0 images) and 41 diffusion-
weighted images (b=1000 s/mm
2
). All T1-weighted MR and DWI images were checked visually for
quality assurance to exclude scans with excessive motion and/or artifacts after preprocessing corrections;
all scans were included.

Table 1. Demographics and clinical scores for the participants
e-MCI l-MCI p-value for group difference
(n=21) (n=9) e-MCI vs l-MCI
Age 71.6 +/- 8.1 72.1 +/- 6.6 0.87
Sex 11 M / 10 F 7 M / 2 F --
Education (yrs) 15.8 +/- 2.7 16.2 +/- 3.1 0.73
MMSE 27.9 +/- 1.8 27.6 +/- 1.7 0.63

Image Preprocessing

Preprocessing of Baseline and 6-month Follow-up Anatomical Scans

All extra-cerebral tissue was removed from both baseline and 6 month T1-weighted anatomical scans
using a number of software packages, primarily ROBEX, a robust automated brain extraction program
trained on manually “skull-stripped” MRI data (Iglesias et al. 2011) and FreeSurfer (Fischl et al. 2004).
Skull-stripped volumes were visually inspected, and the best one selected and sometimes further manually
edited. Anatomical scans subsequently underwent N3 intensity inhomogeneity normalization (Sled et al.
185

1998). To align data from different subjects into the same 3D coordinate space, each anatomical image
was linearly aligned to a standard brain template (the Colin27) (Holmes et al. 1998).

Baseline DWI Preprocessing

For each subject, all raw DWI volumes were aligned to the average b0 image using the FSL eddy-correct
tool (www.fmrib.ox.ac.uk/fsl) to correct for head motion and eddy current distortions. Non-brain tissue
was removed from the diffusion-weighted images using the Brain Extraction Tool (BET) from FSL (Smith
et al. 2002). To correct for echo-planar induced (EPI) susceptibility artifacts, which can cause distortions
at tissue-fluid interfaces, skull-stripped b0 images were linearly aligned and then elastically registered to
their respective baseline T1-weighted structural scans using an inverse consistent registration algorithm
with a mutual information cost function (Leow et al. 2007). The resulting linear registration matrices and
3D deformation fields were then applied to the remaining 41 DWI volumes. FA maps were subsequently
calculated using FSL dtifit and overlaid on T1 anatomical scans to ensure proper alignment.

Tractography

At each voxel, orientation distribution functions (ODFs) were computed using the normalized and
dimensionless ODF estimator, derived for q-ball imaging (QBI) as in Aganj et al. (2010). The angular
resolution of the ADNI data is somewhat limited to avoid long scan times that may tend to increase patient
attrition, but the use of an ODF model makes best use of the available angular resolution. Tractography
was performed on the linearly aligned sets of DWI volumes by probabilistically seeding voxels with a
prior probability based on the FA value. Curves through a seed point received a score estimating the
probability of the existence, computed from the ODFs. We used a voting process provided by the Hough
transform to determine the best fitting curves through each point (Figure 1a) (Aganj et al. 2011). Elastic
deformations obtained from the EPI distortion correction, mapping the average b0 image to the T1-
weighted image, were then applied to the resulting tracts’ 3D coordinates. Each subject’s dataset contained
approximately 10,000 non-duplicated fibers (3D curves). In prior work, we have determined that this is a
sufficient number of fibers to determine most of the common network topology measures accurately
(Prasad et al. 2013b). We removed any erroneous fibers traced on the edge of the brain due to high intensity
noise. To limit small noisy tracts, we filtered out fibers with less than 10 points.
186

Automated Cortical Segmentation

Using FreeSurfer, 34 cortical labels from the Desikan-Killiany atlas (Table 2) were automatically
extracted in each hemisphere from baseline T1-weighted structural MRI scans (Fischl et al. 2004, Desikan
et al. 2006). The resulting T1-weighted images and were then aligned to the corrected T1 images, and the
linear transformation matrix was applied to the cortical parcellations using nearest neighbor interpolation
(to avoid intermixing of labels). This placed the cortical labels in the same space as the tractography,
calculated from the DWIs that were elastically registered to the corrected T1 space (Figure 1b). To ensure
tracts would intersect cortical labeled boundaries, labels were dilated with an isotropic box kernel of
5×5×5 voxels (Figure 1c) (Jahanshad et al. 2011b). Proper alignment of each subject’s cortical
parcellations, T1-weighted image, and tractography was verified by visually inspecting the 3 overlaid
images.

Table 2. Index of cortical labels extracted from the anatomical MRI scans by FreeSurfer (Fischl et al. 2004)
1 Banks of the superior temporal sulcus 19 Pars orbitalis
2 Caudal anterior cingulate 20 Pars triangularis
3 Caudal middle frontal 21 Peri-calcarine
4 -N/A- 22 Postcentral
5 Cuneus 23 Posterior cingulate
6 Entorhinal 24 Precentral
7 Fusiform 25 Precuneus
8 Inferior parietal 26 Rostral anterior cingulate
9 Inferior temporal 27 Rostral middle frontal
10 Isthmus of the cingulate 28 Superior frontal
11 Lateral occipital 29 Superior parietal
12 Lateral orbitofrontal 30 Superior temporal
13 Lingual 31 Supra-marginal
14 Medial orbitofrontal 32 Frontal pole
15 Middle temporal 33 Temporal pole
16 Parahippocampal 34 Transverse temporal
17 Paracentral 35 Insula
18 Pars opercularis

NxN Matrices Representing Structural Connectivity

As in Jahanshad et al. (2011b), for each subject, a baseline 68×68 (34 right hemisphere labels and 34 left)
connectivity matrix was created. Each element described the estimated proportion of the total number of
fibers, in that subject, connecting each of the labels to each of the other labels (Figure 1d).

187

Figure 1. (a) EPI-corrected whole-brain tractography calculated from the DWI. (b) Anatomical cortical
parcellations in one hemisphere are shown, registered to the same subject’s DWI space. (c) Red fiber density map,
where each voxel represents the total number of streamlines that pass through it, overlaid on the dilated labels. (d)
Connectivity matrix, in which each colored element represents the proportion of detected fibers connecting each of
the colored labels in each hemisphere to each of the other colored labels in (c) – computed as a proportion of the
total number of extracted fibers in the brain. This general method was used by us in (Jahanshad et al. 2012,
Jahanshad et al. 2013b), to which the reader is referred for further details.

Graph Theory Network Analyses

We applied the Brain Connectivity Toolbox (https://sites.google.com/site/bctnet/) to our weighted
baseline connectivity matrices generated above, to compute the measures whose values contribute to small
world architecture. In weighted measures, a path between two neighbors with strong connections
contributes more than a path between two weakly connected neighbors. Characteristic path length (CPL)
is an average measure (across the whole network) of the minimum number of edges necessary to travel
from one node to another in the network (i.e., average minimum path length) (Watts and Strogatz 1998).
Mean clustering coefficient (MCC) is an average measure (across the whole network) of how many
neighbors of a given node are also connected to each other, relative to the total possible number of
connections in the network (Onella et al., 2005). Small-worldness, which measures the balance between
network differentiation and network integration, is a ratio of the MCC and CPL of a network. As the small
worldness measure may falsely report small world topology in highly segregated, but poorly integrated
networks (Rubinov and Sporns 2010), we chose to assess MCC and CPL as joint predictors instead.
In a post hoc analysis, we additionally evaluated several weighted nodal measures to assess the
extent to which local connectivity measures can also drive prediction: nodal clustering coefficient (CC),
188

which parallels nodal efficiency (EFF), and nodal betweenness centrality (BTW). The CC measures how
many neighbors of a given node are also connected to each other, relative to the total possible connections,
while the EFF of a node is the average inverse shortest path length calculated on the neighborhood of a
given node. BTW is the fraction of all shortest paths in the network that contain a given node. Nodes with
high values of betweenness centrality participate in a large number of shortest paths. The equations to
calculate each of these measures can be found in (Rubinov and Sporns 2010).

Study-specific Template Creation

A study-specific minimal deformation template (MDT; (Gutman et al. 2010) was created using 29
cognitively healthy elderly control (CTL) subjects’ baseline spatially-aligned corrected anatomical
volumes. Using a customized template from subjects in the study (rather than a standard atlas or a single
optimally chosen subject) can reduce bias in the registrations. The MDT is the template that deviates least
from the anatomy of the subjects, and, in some circumstances, it can improve statistical power (Leporé et
al. 2007). The MDT was generated by creating an initial affine mean template from all 29 subjects, then
warping all the aligned individual scans to that mean (Leow et al. 2007) while regularizing the Jacobians
(Yanovsky et al. 2007). A new mean was created from the registered scans; this process was iterated
several times.

Tensor-Based Morphometry

To quantify 3D patterns of volumetric brain atrophy in MCI, each subject’s 6 month preprocessed T1-
weighted scan was elastically registered to its respective corrected baseline T1-weighted scan (Leow et
al. 2007). A separate 3D Jacobian map (i.e., volumetric expansion factor map) was created for each subject
to characterize the local volume differences between their baseline scan and 6 month scan. To ensure the
Jacobians had common anatomical coordinates for statistical analysis, each subject’s respective 3D
deformation field - from the elastic registration of the baseline T1-weighted scan to the MDT - was applied
to each Jacobian map.

189

Statistics

We ran voxel-wise multiple linear regressions, covarying for sex and age, and a partial F test, using
baseline MCC and CPL as predictors – both jointly and independently – of the longitudinal volumetric
changes. Computing thousands of association tests at a voxel-wise level can introduce a high false positive
error rate in neuroimaging studies, if not corrected. To correct for these errors, we used the searchlight
method for false discovery rate correction (sFDR) (Langers et al. 2007). All statistical maps are
thresholded at a corrected p-value to show regression coefficients only in regions that controlled the false
discovery rate (q=0.05).
In post hoc analyses, we further ran voxel-wise linear regressions, covarying for sex and age, to
detect any associations between baseline CC/EFF and BTW in each of the 68 nodes and the Jacobian
maps. To correct for multiple comparisons for each of 68 nodes, we used the sFDR correction at q=0.05/68
or q=0.00074 .

5.2.3 RESULTS

We found a significant association between the baseline global network measures, CPL and MCC, used
together as predictors in the same regression model, and 3D volumetric changes over the 6-month follow-
up interval (Figure 2a; corrected p<0.05). Separately, MCC was significantly negatively associated with
CSF volume changes surrounding the frontal, parietal, temporal and occipital lobes and positively
associated with regional volumetric changes around the right angular gyrus, left posterior orbital gyrus,
left precuneus and left fusiform (Figure 2b). CPL was negatively associated with regional volume changes
in the right and left anterior corona radiata and left superior corona radiata, as well as the left fusiform
and temporal lobe (Figure 2c). This suggests that lower MCC and increased CPL at baseline are associated
with decreases in tissue volume and increases in CSF expansion (implying tissue loss) in these regions
after 6 months.
In a post hoc analysis, we also found the right pars opercularis (inferior frontal gyrus) node’s local
EFF and CC are significantly positively associated with right internal capsule and temporal lobe and
negatively associated with the right insular sulcus/lateral fissure and chiasmatic cistern (Figure 3a;
corrected p<0.00074). The left superior parietal node EFF and CC are significantly negatively associated
with CSF volume around the left and right frontal lobe extending towards the right temporal lobe (Figure
190

3b; corrected p<0.00074). CC of the left peri-calcarine node was negatively associated with CSF volume
around the left frontal lobe (Figure 3c; corrected p<0.00074). Finally, the right temporal pole BTW was
positively associated with the volume of the left temporal lobe, angular gyrus, and posterior corona
radiata (Figure 3d; corrected p<0.00074). Overall, these measures suggest that decreased local CC/EFF
and BTW at baseline are associated with atrophy between baseline and a follow-up scan 6 months later.

Figure 2. (a) These p-maps show regions where CPL and MCC are joint predictors of volumetric changes on
standard anatomical MRI between baseline and a 6-month follow-up scan (corrected p<0.05) (b) These maps show
T-values within regions where only MCC has a significant correlation with volumetric changes (corrected p<0.05).
(c) These maps show T-values within regions where only CPL has a significant negative correlation with volumetric
changes (corrected p<0.05). Lower MCC and higher CPL at baseline are associated with greater volumetric atrophy
after 6 months.
191

Figure 3. Nodal clustering coefficient (CC) in the (a) right pars opercularis (inferior frontal gyrus) (b) left superior
parietal node and (c) left peri-calcarine node are significantly associated with 3D patterns of volumetric brain
atrophy, implying that increased clustering in these regions is associated with greater future atrophy. These same
patterns are associated with efficiency in these nodes, a measure that parallels CC. (d) Right temporal pole
betweenness (BTW) is positively associated with volume. These maps show T-values within regions that show a
significant association (corrected p<0.00074).

5.2.4 DISCUSSION

There is great interest in predicting which subjects with MCI are likely to decline, as well as in
understanding what patterns of organizational decline in the brain may be harbingers of brain tissue loss.
Rather than evaluating gross anatomical structures of the brain independently, brain connectivity analyses
can evaluate how integrated each region is with others, and thus may be more sensitive to alterations in
brain systems as a whole. Several recent studies have suggested that AD progression may involve a loss
of small world characteristics in the brain’s structural and functional networks (Stam et al. 2007, He et al.
2008, Sanz-Arigita et al. 2010). This is consistent with theoretical notions that small-world topology may
be functionally beneficial and efficient. In this study, we assessed whether abnormalities in small
192

worldness, the balance between network segregation and network integration, at baseline were predictive
of volumetric brain decline over a 6-month period.
We found an association between baseline small-world global network measures and volumetric
changes in T1-weighted structural scans. Moreover, we found that lower mean clustering (MCC) and
higher characteristic path length (CPL) at baseline are associated with greater atrophy. Networks with
lower CPL, a measure reflecting speed or ease of functional integration of distributed brain regions, and
higher levels of clustering or dense connections within regions across the network (MCC), may indicate
a more functionally coherent neural system (Bullmore and Sporns 2009).
To further investigate which regions or nodes may be driving global MCC associations we assessed
nodal clustering (CC), which parallels local nodal efficiency, and found that decreased clustering in the
right pars opercularis, left superior parietal node and left peri-calcarine node were significantly associated
with patterns of volumetric brain atrophy. To assess which nodes might help facilitate lower global CPL,
we evaluated nodal betweenness centrality (BTW), which measures whether a node participates in a large
number of shortest paths, facilitating integration between anatomically unconnected regions. We found
that lower BTW in the right temporal pole was associated with atrophy.
These regions have been implicated in other DTI network studies. The superior parietal cortex, for
example, is known to be affected by AD pathology early on (Jacobs et al. 2012), and is one of few ‘rich
club’ hubs, the set of most highly interconnected nodes, that play a central role in global network
integration (van den Heuvel et al. 2012). A DTI study by Lo et al. (2010) also revealed nodal efficiency
reductions in several prefrontal areas including the orbital part of the inferior frontal gyrus, and the
temporal pole. In a connectivity study involving gray matter volume correlations, the temporal pole,
fusiform, cingulate, superior parietal region, and orbital frontal gyrus showed significant changes in the
interregional correlations between the normal control and AD groups (Yao et al. 2010).
This study could be extended in several ways. There is a great deal of work in brain connectivity
analyses trying to identify subnetworks that are more sensitive to picking up differences in disease. Rather
than pick a fixed partition of the cortical surface, other work has attempted to adaptively refine and alter
the cortical partition to better sensitize the analysis to group differences in disease (Prasad et al. 2013a).
Although such adaptive approaches are elegant, they have the limitation that the cortical connectivity
matrices from different studies and cohorts would be quite difficult to compare, as they are not defining
connectivity for the same regions of interest. A second line of work has argued that connectivity can be
defined in different ways, some of which may be better sensitized to pick up disease-related differences.
193

For instance, some have defined lattice networks where every voxel is considered connected to all its
immediate neighbors, and the angular diffusion signal at each voxel is used to define a dense weighted
network that is amenable to connectivity analysis (Li et al. 2013). Other approaches use statistical methods
to pre-select fibers likely to show associations with disease (Jahanshad et al. 2015a). A third line of work
has attempted to threshold the connectivity networks to focus on nodes that have very high connectivity
to others, or that might be important hubs or highly connected “centers” for the network as a whole. This
leads to concepts such as network filtrations, k-cores, and rich club coefficients (Dennis et al. 2013), which
have begun to be tested for DTI based analysis of connectivity in disease (Daianu et al. 2013a, Daianu et
al. 2013b). When the ADNI2 dataset is much larger, it should be possible to compare many of these
methods head-to-head.
How the raw data have been acquired and processed prior to any statistical analysis can have large
effects on results as each step is susceptible to sources of error and bias (Jones and Cercignani 2010). For
example, connectivity studies comparing networks derived from 3T and 7T scans have revealed
differences between field strengths (Zhan et al. 2013b). Additional limitations may include the limited
angular resolution of the ADNI dataset, selected to avoid long scan times that may increase patient
attrition. However, the use of an ODF model makes best use of the available angular resolution. The
standard single-tensor model is somewhat limited in regions with extensive fiber crossing and mixing,
while the ODF model can better resolve multi-fiber trajectories.
TBM voxel-wise analyses assume that a specific voxel location in the brain is identical across all
subjects. However, registration accuracy from one subject to another may vary, in particular in aging
studies where structures atrophy. Similarly, while tracts were corrected for susceptibility-induced artifacts,
remaining distortion could cause misalignment and can lead to spurious results (Jahanshad et al. 2011a).
Different parcellation schemes may also affect graph theory metrics. We used the FreeSurfer Desikan–
Killiany atlas (Desikan et al. 2006) for cortical parcellation, which has been widely used for structural
connectivity analysis (Hagmann et al. 2008, Honey et al. 2009, Daianu et al. 2013a, Daianu et al. 2013b).
However, other parcellations are possible and there is still work being done to understand how different
parcellation templates and resolutions may influence different kinds of network measures (Hagmann et
al. 2008, Zalesky et al. 2010, Bassett et al. 2011, Prasad et al. 2013a).
It appears that the degree of integration and efficiency both across distributed brain regions- CPL
and BTW- and locally within regions- MCC, CC- is an important indication of a coherent neural system at
baseline, and may be predictive of future decline. These results are preliminary and need to be replicated
194

as ADNI2 progresses and new subjects are scanned. As the longitudinal study progresses, we can later
investigate which of these subjects eventually develops AD, and if these early aberrations in connectivity
can help to predict a patient’s conversion to AD, future brain tissue loss, and cognitive decline. This study
offers evidence that DTI-based network measures may be a novel predictor of AD progression.

5.2.5 ACKNOWLEDGEMENTS

Algorithm development and image analysis for this study was funded, in part, by grants to PT from the
NIBIB (R01 EB008281, R01 EB008432) and by the NIA, NIBIB, NIMH, the National Library of
Medicine, and the National Center for Research Resources (AG016570, AG040060, EB01651,
MH097268, LM05639, RR019771 to PT). Data collection and sharing for this project was funded by
ADNI (NIH Grant U01 AG024904). ADNI is funded by the National Institute on Aging, the National
Institute of Biomedical Imaging and Bioengineering, and through contributions from the following:
Abbott; Alzheimer's Association; Alzheimer's Drug Discovery Foundation; Amorfix Life Sciences Ltd.;
AstraZeneca; Bayer HealthCare; BioClinica, Inc.; Biogen Idec Inc.; Bristol-Myers Squibb Company;
Eisai Inc.; Elan Pharmaceuticals Inc.; Eli Lilly and Company; F. Hoffmann-La Roche Ltd and its affiliated
company Genentech, Inc.; GE Healthcare; Innogenetics, N.V.; IXICO Ltd.; Janssen Alzheimer
Immunotherapy Research & Development, LLC.; Johnson & Johnson Pharmaceutical Research &
Development LLC.; Medpace, Inc.; Merck & Co., Inc.; Meso Scale Diagnostics, LLC.; Novartis
Pharmaceuticals Corporation; Pfizer Inc.; Servier; Synarc Inc.; and Takeda Pharmaceutical Company. The
Canadian Institutes of Health Research is providing funds to support ADNI clinical sites in Canada.
Private sector contributions are facilitated by the Foundation for the National Institutes of Health. The
grantee organization is the Northern California Institute for Research and Education, and the study is
coordinated by the Alzheimer's Disease Cooperative Study at the University of California, San Diego.
ADNI data are disseminated by the Laboratory for Neuro Imaging at the University of Southern California.
This research was also supported by NIH grants P30 AG010129 and K01 AG030514 from the National
Institute of General Medical Sciences.

195

REFERENCES

Abhinav K, Yeh F-C, Pathak S, Suski V, Lacomis D, Friedlander RM and Fernandez-Miranda JC (2014).
Advanced diffusion MRI fiber tracking in neurosurgical and neurodegenerative disorders and
neuroanatomical studies: A review. Biochimica et Biophysica Acta (BBA) - Molecular Basis of Disease,
1842(11): 2286-2297.

Achard S and Bullmore E (2007). Efficiency and cost of economical brain functional networks. PLoS
Comput Biol, 3(2): e17.

Acheson A, Wijtenburg SA, Rowland LM, Winkler A, Mathias CW, Hong LE, Jahanshad N, Patel B,
Thompson PM, McGuire SA, Sherman PM, Kochunov P and Dougherty DM (2017). Reproducibility of
tract-based white matter microstructural measures using the ENIGMA-DTI protocol. Brain Behav, 7(2):
e00615.

Acosta-Cabronero J, Williams GB, Pengas G and Nestor PJ (2010). Absolute diffusivities define the
landscape of white matter degeneration in Alzheimer's disease. Brain, 133: 529-539.

Acosta-Cabronero J and Nestor PJ (2014). Diffusion tensor imaging in Alzheimer's disease: insights into
the limbic-diencephalic network and methodological considerations. Front Aging Neurosci, 6: 266.

Adle-Biassette H, Chretien F, Wingertsmann L, Hery C, Ereau T, Scaravilli F, Tardieu M and Gray F
(1999). Neuronal apoptosis does not correlate with dementia in HIV infection but is related to microglial
activation and axonal damage. Neuropathol Appl Neurobiol, 25(2): 123-133.

Aganj I, Lenglet C, Sapiro G, Yacoub E, Ugurbil K and Harel N (2010). Reconstruction of the orientation
distribution function in single- and multiple-shell q-ball imaging within constant solid angle. Magn Reson
Med, 64(2): 554-566.

Aganj I, Lenglet C, Jahanshad N, Yacoub E, Harel N, Thompson PM and Sapiro G (2011). A Hough
transform global probabilistic approach to multiple-subject diffusion MRI tractography. Med Image Anal,
15(4): 414-425.

Akaike H (1998). Information Theory and an Extension of the Maximum Likelihood Principle. Selected
Papers of Hirotugu Akaike. E. Parzen, K. Tanabe and G. Kitagawa. New York, NY, Springer New York:
199-213.

Alexander AL, Hasan KM, Lazar M, Tsuruda JS and Parker DL (2001). Analysis of partial volume effects
in diffusion-tensor MRI. Magnetic Resonance in Medicine, 45(5): 770-780.

Alexander DC and Barker GJ (2005). Optimal imaging parameters for fiber-orientation estimation in
diffusion MRI. Neuroimage, 27(2): 357-367.

196

Alexander DC (2008). A general framework for experiment design in diffusion MRI and its application
in measuring direct tissue-microstructure features. Magn Reson Med, 60(2): 439-448.

Alexander DC, Hubbard PL, Hall MG, Moore EA, Ptito M, Parker GJ and Dyrby TB (2010).
Orientationally invariant indices of axon diameter and density from diffusion MRI. Neuroimage, 52(4):
1374-1389.

AMFAR (2018). Statistics: Women and HIV/AIDS. From https://www.amfar.org/about-hiv-and-
aids/facts-and-stats/statistics--women-and-hiv-aids/.

Ances BM and Ellis RJ (2007). Dementia and neurocognitive disorders due to HIV-1 infection. Semin
Neurol, 27(1): 86-92.

Ances BM, Ortega M, Vaida F, Heaps J and Paul R (2012). Independent effects of HIV, aging, and
HAART on brain volumetric measures. J Acquir Immune Defic Syndr, 59(5): 469-477.

Ances BM and Hammoud DA (2014). Neuroimaging of HIV-associated neurocognitive disorders
(HAND). Current opinion in HIV and AIDS, 9(6): 545-551.

Anderson AM, Harezlak J, Bharti A, Mi D, Taylor MJ, Daar ES, Schifitto G, Zhong J, Alger JR, Brown
MS, Singer EJ, Campbell TB, McMahon DD, Buchthal S, Cohen R, Yiannoutsos C, Letendre SL, Navia
BA and Consortium HIVN (2015). Plasma and Cerebrospinal Fluid Biomarkers Predict Cerebral Injury
in HIV-Infected Individuals on Stable Combination Antiretroviral Therapy. J Acquir Immune Defic Syndr,
69(1): 29-35.

Andersson JLR, Graham MS, Zsoldos E and Sotiropoulos SN (2016). Incorporating outlier detection and
replacement into a non-parametric framework for movement and distortion correction of diffusion MR
images. Neuroimage, 141: 556-572.

Andersson JLR and Sotiropoulos SN (2016). An integrated approach to correction for off-resonance
effects and subject movement in diffusion MR imaging. Neuroimage, 125: 1063-1078.

Andres MA, Feger U, Nath A, Munsaka S, Jiang CS and Chang L (2011). APOE ε 4 allele and CSF APOE
on cognition in HIV-infected subjects. Journal of neuroimmune pharmacology : the official journal of the
Society on NeuroImmune Pharmacology, 6(3): 389-398.

Anthony IC, Ramage SN, Carnie FW, Simmonds P and Bell JE (2006). Accelerated Tau deposition in the
brains of individuals infected with human immunodeficiency virus-1 before and after the advent of highly
active anti-retroviral therapy. Acta Neuropathol, 111(6): 529-538.

Anthony IC and Bell JE (2008). The Neuropathology of HIV/AIDS. Int Rev Psychiatry, 20(1): 15-24.

Antinori A, Arendt G, Becker JT, Brew BJ, Byrd DA, Cherner M, Clifford DB, Cinque P, Epstein LG,
Goodkin K, Gisslen M, Grant I, Heaton RK, Joseph J, Marder K, Marra CM, McArthur JC, Nunn M, Price
RW, Pulliam L, Robertson KR, Sacktor N, Valcour V and Wojna VE (2007). Updated research nosology
for HIV-associated neurocognitive disorders. Neurology, 69(18): 1789-1799.
197

Assaf Y and Basser PJ (2005). Composite hindered and restricted model of diffusion (CHARMED) MR
imaging of the human brain. Neuroimage, 27(1): 48-58.

Alzheimer’s Association (2018). 2018 Alzheimer's disease facts and figures includes a special report on
the financial and personal benefits of early diagnosis. 14: 367-429.

Avants BB, Tustison NJ, Song G, Cook PA, Klein A and Gee JC (2011). A reproducible evaluation of
ANTs similarity metric performance in brain image registration. Neuroimage, 54(3): 2033-2044.

Aylward EH, Henderer JD, McArthur JC, Brettschneider PD, Harris GJ, Barta PE and Pearlson GD
(1993). Reduced basal ganglia volume in HIV-1-associated dementia: results from quantitative
neuroimaging. Neurology, 43(10): 2099-2104.

Baker LM, Cooley SA, Cabeen RP, Laidlaw DH, Joska JA, Hoare J, Stein DJ, Heaps-Woodruff JM,
Salminen LE and Paul RH (2017). Topological organization of whole-brain white matter in HIV infection.
Brain Connect, 7(2): 115-122.

Balsis S, Benge JF, Lowe DA, Geraci L and Doody RS (2015). How do scores on the ADAS-Cog, MMSE,
and CDR-SOB correspond? Clin Neuropsychol, 29(7): 1002-1009.

Barazany D, Basser PJ and Assaf Y (2009). In vivo measurement of axon diameter distribution in the
corpus callosum of rat brain. Brain, 132(Pt 5): 1210-1220.

Bartzokis G (2011). Alzheimer's disease as homeostatic responses to age-related myelin breakdown.
Neurobiol Aging, 32(8): 1341-1371.

Basser PJ, Mattiello J and LeBihan D (1994). MR diffusion tensor spectroscopy and imaging. Biophys J,
66(1): 259-267.

Bassett DS, Brown JA, Deshpande V, Carlson JM and Grafton ST (2011). Conserved and variable
architecture of human white matter connectivity. Neuroimage, 54(2): 1262-1279.

Bearden CE and Thompson PM (2017). Emerging Global Initiatives in Neurogenetics: The Enhancing
Neuroimaging Genetics through Meta-analysis (ENIGMA) Consortium. Neuron, 94(2): 232-236.

Beaulieu C (2002). The basis of anisotropic water diffusion in the nervous system - A technical review.
NMR in Biomedicine, 15: 435-455.

Becker JT, Kingsley L, Mullen J, Cohen B, Martin E, Miller EN, Ragin A, Sacktor N, Selnes OA, Visscher
BR and Multicenter ACS (2009). Vascular risk factors, HIV serostatus, and cognitive dysfunction in gay
and bisexual men. Neurology, 73(16): 1292-1299.

Becker JT, Sanders J, Madsen SK, Ragin A, Kingsley L, Maruca V, Cohen B, Goodkin K, Martin E,
Miller EN, Sacktor N, Alger JR, Barker PB, Saharan P, Carmichael OT, Thompson PM and Multicenter
ACS (2011). Subcortical brain atrophy persists even in HAART-regulated HIV disease. Brain Imaging
Behav, 5(2): 77-85.
198

Becker JT, Maruca V, Kingsley LA, Sanders JM, Alger JR, Barker PB, Goodkin K, Martin E, Miller EN,
Ragin A, Sacktor N, Selnes O and Multicenter ACS (2012). Factors affecting brain structure in men with
HIV disease in the post-HAART era. Neuroradiology, 54(2): 113-121.

Behrens TE, Woolrich MW, Jenkinson M, Johansen-Berg H, Nunes RG, Clare S, Matthews PM, Brady
JM and Smith SM (2003). Characterization and propagation of uncertainty in diffusion-weighted MR
imaging. Magn Reson Med, 50(5): 1077-1088.

Behrens TE, Berg HJ, Jbabdi S, Rushworth MF and Woolrich MW (2007). Probabilistic diffusion
tractography with multiple fibre orientations: What can we gain? Neuroimage, 34(1): 144-155.

Bell RP, Barnes LL, Towe SL, Chen NK, Song AW and Meade CS (2018). Structural connectome
differences in HIV infection: brain network segregation associated with nadir CD4 cell count. J
Neurovirol, 24(4): 454-463.

Benjamini Y and Hochberg Y (1995). Controlling the false discovery rate: a practical and powerful
approach to multiple testing. Journal of the Royal Statistical Society Series B-Methodological, 57(1): 289-
300.

Berg L (1988). Clinical Dementia Rating (CDR). Psychopharmacology bulletin, 24: 637-639.

Berger JR and Nath A (1997). HIV dementia and the basal ganglia. Intervirology, 40(2-3): 122-131.

Bhaskaran K, Mussini C, Antinori A, Walker AS, Dorrucci M, Sabin C, Phillips A, Porter K and
Collaboration C (2008). Changes in the incidence and predictors of human immunodeficiency virus-
associated dementia in the era of highly active antiretroviral therapy. Ann Neurol, 63(2): 213-221.

Blair C and Raver CC (2016). Poverty, stress, and brain development: New directions for prevention and
intervention. Academic pediatrics, 16(3 Suppl): S30-S36.

Boban J, Kozic D, Turkulov V, Ostojic J, Semnic R, Lendak D and Brkic S (2017). HIV-associated
neurodegeneration and neuroimmunity: multivoxel MR spectroscopy study in drug-naive and treated
patients. Eur Radiol, 27(10): 4218-4236.

Bozzali M, Cercignani M, Sormani MP, Comi G and Filippi M (2002). Quantification of brain gray matter
damage in different MS phenotypes by use of diffusion tensor MR imaging. AJNR Am J Neuroradiol,
23(6): 985-988.

Braak H and Braak E (1991). Neuropathological stageing of Alzheimer-related changes. Acta
neuropathologica, 82: 239-259.

Braak H and Braak E (1996). Evolution of the neuropathology of Alzheimer's disease. Acta neurologica
Scandinavica. Supplementum, 165: 3-12.

199

Braskie MN, Jahanshad N, Stein JL, Barysheva M, McMahon KL, de Zubicaray GI, Martin NG, Wright
MJ, Ringman JM, Toga AW and Thompson PM (2011). Common Alzheimer's disease risk variant within
the CLU gene affects white matter microstructure in young adults. J Neurosci, 31(18): 6764-6770.

Brew BJ, Rosenblum M, Cronin K and Price RW (1995). AIDS dementia complex and HIV-1 brain
infection: clinical-virological correlations. Ann Neurol, 38(4): 563-570.

Brew BJ, Pemberton L, Blennow K, Wallin A and Hagberg L (2005). CSF amyloid beta42 and tau levels
correlate with AIDS dementia complex. Neurology, 65(9): 1490-1492.

Brew BJ, Crowe SM, Landay A, Cysique LA and Guillemin G (2009). Neurodegeneration and ageing in
the HAART era. J Neuroimmune Pharmacol, 4(2): 163-174.

Brew BJ and Cysique L (2017). Does HIV prematurely age the brain? Lancet HIV, 4(9): e380-e381.

Brown JA, Terashima KH, Burggren AC, Ercoli LM, Miller KJ, Small GW and Bookheimer SY (2011).
Brain network local interconnectivity loss in aging APOE-4 allele carriers. Proc Natl Acad Sci U S A,
108(51): 20760-20765.

Bruscoli M and Lovestone S (2004). Is MCI really just early dementia? A systematic review of conversion
studies. Int Psychogeriatr, 16(2): 129-140.

Bullmore E and Sporns O (2009). Complex brain networks: graph theoretical analysis of structural and
functional systems. Nat Rev Neurosci, 10(3): 186-198.

Burnham KP and Anderson DR (2002). Model selection and multimodel inference: A practical
information-theoretic approach. New York, Springer-Verlag.

Burt TD, Agan BK, Marconi VC, He W, Kulkarni H, Mold JE, Cavrois M, Huang Y, Mahley RW, Dolan
MJ, McCune JM and Ahuja SK (2008). Apolipoprotein (apo) E4 enhances HIV-1 cell entry in vitro, and
the APOE epsilon4/epsilon4 genotype accelerates HIV disease progression. Proc Natl Acad Sci U S A,
105(25): 8718-8723.

Button KS, Ioannidis JPA, Mokrysz C, Nosek BA, Flint J, Robinson ESJ and Munaf (2013). Power failure:
why small sample size undermines the reliability of neuroscience. Nature Reviews Neuroscience, 14(5):
365-376.

Callaghan PT (1995). Pulsed-gradient spin-echo NMR for planar, cylindrical, and spherical pores under
conditions of wall relaxation. Journal of Magnetic Resonance, Series A, 113(1): 53-59.

Canu E, McLaren DG, Fitzgerald ME, Bendlin BB, Zoccatelli G, Alessandrini F, Pizzini FB, Ricciardi
GK, Beltramello A, Johnson SC and Frisoni GB (2010). Microstructural diffusion changes are
independent of macrostructural volume loss in moderate to severe Alzheimer's disease. J Alzheimers Dis,
19(3): 963-976.

200

Cardenas VA, Meyerhoff DJ, Studholme C, Kornak J, Rothlind J, Lampiris H, Neuhaus J, Grant RM,
Chao LL, Truran D and Weiner MW (2009). Evidence for ongoing brain injury in human
immunodeficiency virus-positive patients treated with antiretroviral therapy. J Neurovirol, 15(4): 324-
333.

Carey CL, Woods SP, Gonzalez R, Conover E, Marcotte TD, Grant I and Heaton RK (2004). Predictive
validity of global deficit scores in detecting neuropsychological impairment in HIV infection. J Clin Exp
Neuropsychol, 26(3): 307-319.

Carnie FW, Ramage SN, Bell JE, Anthony IC and Simmonds P (2005). Influence of HAART on HIV-
Related CNS Disease and Neuroinflammation. Journal of Neuropathology & Experimental Neurology,
64(6): 529-536.

Carvalhal A, Gill MJ, Letendre SL, Rachlis A, Bekele T, Raboud J, Burchell A and Rourke SB (2016).
Central nervous system penetration effectiveness of antiretroviral drugs and neuropsychological
impairment in the Ontario HIV Treatment Network Cohort Study. J Neurovirol, 22(3): 349-357.

Castelo JM, Courtney MG, Melrose RJ and Stern CE (2007). Putamen hypertrophy in nondemented
patients with human immunodeficiency virus infection and cognitive compromise. Arch Neurol, 64(9):
1275-1280.

Centers for Disease Control and Prevention (2016) HIV and Substance Use. from
https://www.cdc.gov/hiv/pdf/risk/cdc-hiv-substanceuse.pdf.

Centers for Disease Control and Prevention (2017) HIV Surveillance Report, 2016. from
http://www.cdc.gov/hiv/library/reports/hiv-surveillance.html.

Cercignani M, Bozzali M, Iannucci G, Comi G and Filippi M (2001). Magnetisation transfer ratio and
mean diffusivity of normal appearing white and grey matter from patients with multiple sclerosis. Journal
of Neurology, Neurosurgery, and Psychiatry, 70(3): 311-317.

Cercignani M, Bammer R, Sormani MP, Fazekas F and Filippi M (2003). Inter-sequence and inter-
imaging unit variability of diffusion tensor MR imaging histogram-derived metrics of the brain in healthy
volunteers. American Journal of Neuroradiology, 24(4): 638-643.

Chang L, Lee PL, Yiannoutsos CT, Ernst T, Marra CM, Richards T, Kolson D, Schifitto G, Jarvik JG,
Miller EN, Lenkinski R, Gonzalez G, Navia BA and Consortium HM (2004). A multicenter in vivo proton-
MRS study of HIV-associated dementia and its relationship to age. Neuroimage, 23(4): 1336-1347.

Chang L, Wong V, Nakama H, Watters M, Ramones D, Miller EN, Cloak C and Ernst T (2008). Greater
than age-related changes in brain diffusion of HIV patients after 1 year. Journal of neuroimmune
pharmacology : the official journal of the Society on NeuroImmune Pharmacology, 3(4): 265-274.

Chang L, Andres M, Sadino J, Jiang CS, Nakama H, Miller E and Ernst T (2011). Impact of apolipoprotein
E epsilon4 and HIV on cognition and brain atrophy: antagonistic pleiotropy and premature brain aging.
Neuroimage, 58(4): 1017-1027.
201

Chang L and Shukla DK (2018). Imaging studies of the HIV-infected brain. Handb Clin Neurol, 152: 229-
264.

Chen Y, Sha M, Zhao X, Ma J, Ni H, Gao W and Ming D (2017). Automated detection of pathologic
white matter alterations in Alzheimer's disease using combined diffusivity and kurtosis method.
Psychiatry Research: Neuroimaging, 264: 35-45.

Cheng J-X, Zhang H-Y, Peng Z-K, Xu Y, Tang H, Wu J-T and Xu J (2018). Divergent topological
networks in Alzheimer's disease: a diffusion kurtosis imaging analysis. Translational Neurodegeneration,
7(1): 10.

Cherner M, Ellis RJ, Lazzaretto D, Young C, Mindt MR, Atkinson JH, Grant I and Heaton RK (2004).
Effects of HIV-1 infection and aging on neurobehavioral functioning: preliminary findings. Aids, 18 Suppl
1: S27-34.

Chiang MC, Dutton RA, Hayashi KM, Lopez OL, Aizenstein HJ, Toga AW, Becker JT and Thompson
PM (2007). 3D pattern of brain atrophy in HIV/AIDS visualized using tensor-based morphometry.
Neuroimage, 34(1): 44-60.

Chiapponi C, Piras F, Piras F, Fagioli S, Caltagirone C and Spalletta G (2013). Cortical grey matter and
subcortical white matter brain microstructural changes in schizophrenia are localised and age independent:
a case-control diffusion tensor imaging study. PLoS One, 8(10): e75115.

Clerx L, Visser PJ, Verhey F and Aalten P (2012). New MRI markers for Alzheimer's disease: A meta-
analysis of diffusion tensor imaging and a comparison with Medial temporal lobe measurements. Journal
of Alzheimer's Disease, 29(2): 405-429.

Clifford DB, Fagan AM, Holtzman DM, Morris JC, Teshome M, Shah AR and Kauwe JSK (2009). CSF
biomarkers of Alzheimer disease in HIV-associated neurologic disease. Neurology, 73(23): 1982-1987.

Clifford DB and Ances BM (2013). HIV-associated neurocognitive disorder. The Lancet. Infectious
diseases, 13(11): 976-986.

Clifford KM, Samboju V, Cobigo Y, Milanini B, Marx GA, Hellmuth JM, Rosen HJ, Kramer JH, Allen
IE and Valcour VG (2017). Progressive brain atrophy despite persistent viral Suppression in HIV patients
older than 60 years. J Acquir Immune Defic Syndr, 76(3): 289-297.

Cohen J (1988). Statistical power analysis for the behavioral sciences. Hillsdale, N.J., L. Erlbaum
Associates.

Cohen RA, Harezlak J, Schifitto G, Hana G, Clark U, Gongvatana A, Paul R, Taylor M, Thompson P,
Alger J, Brown M, Zhong J, Campbell T, Singer E, Daar E, McMahon D, Tso Y, Yiannoutsos CT and
Navia B (2010). Effects of nadir CD4 count and duration of human immunodeficiency virus infection on
brain volumes in the highly active antiretroviral therapy era. J Neurovirol, 16(1): 25-32.

202

Cohen RA, Seider TR and Navia B (2015). HIV effects on age-associated neurocognitive dysfunction:
premature cognitive aging or neurodegenerative disease? Alzheimer's Research & Therapy, 7(1): 37.

Colgan N, Siow B, O'Callaghan JM, Harrison IF, Wells JA, Holmes HE, Ismail O, Richardson S,
Alexander DC, Collins EC, Fisher EM, Johnson R, Schwarz AJ, Ahmed Z, O'Neill MJ, Murray TK, Zhang
H and Lythgoe MF (2016). Application of neurite orientation dispersion and density imaging (NODDI)
to a tau pathology model of Alzheimer's disease. NeuroImage, 125: 739-744.

Corder EH, Robertson K, Lannfelt L, Bogdanovic N, Eggertsen G, Wilkins J and Hall C (1998). HIV-
infected subjects with the E4 allele for APOE have excess dementia and peripheral neuropathy. Nat Med,
4(10): 1182-1184.

Correa DG, Zimmermann N, Tukamoto G, Doring T, Ventura N, Leite SC, Cabral RF, Fonseca RP, Bahia
PR and Gasparetto EL (2016). Longitudinal assessment of subcortical gray matter volume, cortical
thickness, and white matter integrity in HIV-positive patients. J Magn Reson Imaging, 44(5): 1262-1269.

Correia MM, Carpenter TA and Williams GB (2009). Looking for the optimal DTI acquisition scheme
given a maximum scan time: are more b-values a waste of time? Magnetic Resonance Imaging, 27(2):
163-175.

Costagliola D (2014). Demographics of HIV and aging. Curr Opin HIV AIDS, 9(4): 294-301.

Coupe P, Yger P, Prima S, Hellier P, Kervrann C and Barillot C (2008). An optimized blockwise nonlocal
means denoising filter for 3-D magnetic resonance images. IEEE Trans Med Imaging, 27(4): 425-441.

Cysique LA, Maruff P and Brew BJ (2004). Prevalence and pattern of neuropsychological impairment in
human immunodeficiency virus-infected/acquired immunodeficiency syndrome (HIV/AIDS) patients
across pre- and post-highly active antiretroviral therapy eras: a combined study of two cohorts. J
Neurovirol, 10(6): 350-357.

Cysique LA, Maruff P and Brew BJ (2006). The neuropsychological profile of symptomatic AIDS and
ADC patients in the pre-HAART era: a meta-analysis. J Int Neuropsychol Soc, 12(3): 368-382.

Cysique LA, Moffat K, Moore DM, Lane TA, Davies NWS, Carr A, Brew BJ and Rae C (2013). HIV,
vascular and aging injuries in the brain of clinically stable HIV-infected adults: A (1)H MRS study. PLoS
ONE, 8(4): e61738.

Cysique LA and Brew BJ (2014). The effects of HIV and aging on brain functions: proposing a research
framework and update on last 3 years' findings. Curr Opin HIV AIDS, 9(4): 355-364.

Cysique LA, Soares JR, Geng G, Scarpetta M, Moffat K, Green M, Brew BJ, Henry RG and Rae C (2017).
White matter measures are near normal in controlled HIV infection except in those with cognitive
impairment and longer HIV duration. J Neurovirol, 23(4): 539-547.

203

Daducci A, Canales-Rodriguez EJ, Zhang H, Dyrby TB, Alexander DC and Thiran JP (2015). Accelerated
microstructure imaging via convex optimization (AMICO) from diffusion MRI data. Neuroimage, 105:
32-44.

Daianu M, Dennis EL, Jahanshad N, Nir TM, Toga AW, Jack CR, Jr., Weiner MW, Thompson PM and
the Alzheimer's Disease Neuroimaging I (2013a). Alzheimer's disease disrupts rich club organization in
brain connectivity networks. Proc IEEE Int Symp Biomed Imaging: 266-269.

Daianu M, Jahanshad N, Nir TM, Toga AW, Jack CR, Jr., Weiner MW, Thompson PM and Alzheimer's
Disease Neuroimaging I (2013b). Breakdown of brain connectivity between normal aging and Alzheimer's
disease: a structural k-core network analysis. Brain Connect, 3(4): 407-422.

Deeks SG, Tracy R and Douek DC (2013). Systemic effects of inflammation on health during chronic
HIV infection. Immunity, 39(4): 633-645.

Delbeuck X, Van der Linden M and Collette F (2003). Alzheimer's disease as a disconnection syndrome?
Neuropsychol Rev, 13(2): 79-92.

Dennis EL, Jahanshad N, Toga AW, McMahon KL, de Zubicaray GI, Hickie I, Wright MJ and Thompson
PM (2013). Development of the "Rich Club" in brain connectivity networks from 438 adolescents and
adults aged 12 to 30. Proc IEEE Int Symp Biomed Imaging: 624-627.

Descoteaux M, Angelino E, Fitzgibbons S and Deriche R (2007). Regularized, fast, and robust analytical
Q-ball imaging. Magn Reson Med, 58(3): 497-510.

Descoteaux M (2008). High angular resolution diffusion MRI: from local estimation to segmentation and
tractography, Université Nice Sophia Antipolis.

Descoteaux M, Deriche R, Knosche TR and Anwander A (2009). Deterministic and probabilistic
tractography based on complex fibre orientation distributions. IEEE Trans Med Imaging, 28(2): 269-286.

Descoteaux M and Poupon C (2012). Diffusion-weighted MRI. Comprehensive Biomedical Physics.
Elsevier, 2: 3.

Desikan RS, Segonne F, Fischl B, Quinn BT, Dickerson BC, Blacker D, Buckner RL, Dale AM, Maguire
RP, Hyman BT, Albert MS and Killiany RJ (2006). An automated labeling system for subdividing the
human cerebral cortex on MRI scans into gyral based regions of interest. Neuroimage, 31(3): 968-980.

Ding L, Zhu AH, Saremi A, Faskowitz JI, Hberg A, Thompson PM and Jahanshad N (2018). Voxelwise
meta-analysis of brain structural associations with genome-wide polygenic risk for Alzheimer 's disease.
4.

Ding Y, Lin H, Shen W, Wu Q, Gao M and He N (2017). Interaction effects between HIV and aging on
selective neurocognitive impairment. J Neuroimmune Pharmacol, 12(4): 661-669.

204

Douaud G, Jbabdi S, Behrens TE, Menke RA, Gass A, Monsch AU, Rao A, Whitcher B, Kindlmann G,
Matthews PM and Smith S (2011). DTI measures in crossing-fibre areas: increased diffusion anisotropy
reveals early white matter alteration in MCI and mild Alzheimer's disease. Neuroimage, 55(3): 880-890.

Duncan GJ, Ziol-Guest KM and Kalil A (2010). Early-childhood poverty and adult attainment, behavior,
and health. Child Dev, 81(1): 306-325.

Ellis R, Langford D and Masliah E (2007). HIV and antiretroviral therapy in the brain: neuronal injury
and repair. Nature Reviews Neuroscience, 8: 33-44.

Ellis RJ, Badiee J, Vaida F, Letendre S, Heaton RK, Clifford D, Collier AC, Gelman B, McArthur J,
Morgello S, McCutchan JA, Grant I and Group C (2011). CD4 nadir is a predictor of HIV neurocognitive
impairment in the era of combination antiretroviral therapy. AIDS, 25(14): 1747-1751.

Erdmann N, Zhao J, Lopez AL, Herek S, Curthoys N, Hexum TD, Tsukamoto T, Ferraris D and Zheng J
(2007). Glutamate production by HIV-1 infected human macrophage is blocked by the inhibition of
glutaminase. Journal of Neurochemistry, 102(2): 539-549.

Ernst T, Jiang CS, Nakama H, Buchthal S and Chang L (2010). Lower brain glutamate is associated with
cognitive deficits in HIV patients: A new mechanism for HIV-associated neurocognitive disorder. Journal
of magnetic resonance imaging : JMRI, 32(5): 1045-1053.

Ersche KD, Williams GB, Robbins TW and Bullmore ET (2013). Meta-analysis of structural brain
abnormalities associated with stimulant drug dependence and neuroimaging of addiction vulnerability and
resilience. Current Opinion in Neurobiology, 23(4): 615-624.

Everall I, Barnes H, Spargo E and Lantos P (1995). Assessment of neuronal density in the putamen in
human immunodeficiency virus (HIV) infection. Application of stereology and spatial analysis of
quadrats. J Neurovirol, 1(1): 126-129.

Everall I, Vaida F, Khanlou N, Lazzaretto D, Achim C, Letendre S, Moore D, Ellis R, Cherner M, Gelman
B, Morgello S, Singer E, Grant I, Masliah E and National Neuro ATC (2009). Cliniconeuropathologic
correlates of human immunodeficiency virus in the era of antiretroviral therapy. Journal of Neurovirology,
15(5-6): 360-370.

Falagas ME, Zarkadoulia EA, Pliatsika PA and Panos G (2008). Socioeconomic status (SES) as a
determinant of adherence to treatment in HIV infected patients: a systematic review of the literature.
Retrovirology, 5: 13.

Fedorov A, Beichel R, Kalpathy-Cramer J, Finet J, Fillion-Robin JC, Pujol S, Bauer C, Jennings D,
Fennessy F, Sonka M, Buatti J, Aylward S, Miller JV, Pieper S and Kikinis R (2012). 3D Slicer as an
image computing platform for the Quantitative Imaging Network. Magn Reson Imaging, 30(9): 1323-
1341.

Fernández AL and Abe J (2018). Bias in cross-cultural neuropsychological testing: problems and possible
solutions. Culture and Brain, 6(1): 1-35.
205

Filippi CG, Ulug AM, Ryan E, Ferrando SJ and van Gorp W (2001). Diffusion tensor imaging of patients
with HIV and normal-appearing white matter on MR images of the brain. AJNR Am J Neuroradiol, 22(2):
277-283.

Fischl B, van der Kouwe A, Destrieux C, Halgren E, Segonne F, Salat DH, Busa E, Seidman LJ, Goldstein
J, Kennedy D, Caviness V, Makris N, Rosen B and Dale AM (2004). Automatically parcellating the
human cerebral cortex. Cereb Cortex, 14(1): 11-22.

Fischl B (2012). FreeSurfer. Neuroimage, 62(2): 774-781.

Folstein MF, Folstein SE and McHugh PR (1975). “Mini-mental state”: A practical method for grading
the cognitive state of patients for the clinician. Journal of Psychiatric Research, 12(3): 189-198.

Forloni G, Demicheli F, Giorgi S, Bendotti C and Angeretti N (1992). Expression of amyloid precursor
protein mRNAs in endothelial, neuronal and glial cells: modulation by interleukin-1. Brain Res Mol Brain
Res, 16(1-2): 128-134.

Fortin J-P, Parker D, Tunç B, Watanabe T, Elliott MA, Ruparel K, Roalf DR, Satterthwaite TD, Gur RC,
Gur RE, Schultz RT, Verma R and Shinohara RT (2017). Harmonization of multi-site diffusion tensor
imaging data. Neuroimage, 161: 149-170.

Fortin J-P, Cullen N, Sheline YI, Taylor WD, Aselcioglu I, Cook PA, Adams P, Cooper C, Fava M,
McGrath PJ, McInnis M, Phillips ML, Trivedi MH, Weissman MM and Shinohara RT (2018).
Harmonization of cortical thickness measurements across scanners and sites. Neuroimage, 167: 104-120.

Fortin JP, Sweeney EM, Muschelli J, Crainiceanu CM and Shinohara RT (2016). Removing inter-subject
technical variability in magnetic resonance imaging studies. Neuroimage, 132: 198-212.

Frank RA, Galasko D, Hampel H and Hardy Ja (2003). Biological markers for therapeutic trials in
Alzheimer's disease: Proceedings of the biological markers working group; NIA initiative on
neuroimaging in Alzheimer's disease. Neurobiology of Aging, 24(4): 521-536.

GadElkarim JJ, Zhan L, Yang SL, Zhang AF, Altshuler L, Lamar M, Ajilore O, Thompson PM, Kumar
A and Leow A (2011). TDF-TRACT: Probabilistic tractography using the tensor distribution function.
2011 IEEE International Symposium on Biomedical Imaging: From Nano to Macro.

Garyfallidis E, Brett M, Amirbekian B, Rokem A, van der Walt S, Descoteaux M, Nimmo-Smith I and
Dipy C (2014). Dipy, a library for the analysis of diffusion MRI data. Front Neuroinform, 8: 8.

Gelman BB, Lisinicchia JG, Morgello S, Masliah E, Commins D, Achim CL, Fox HS, Kolson DL, Grant
I, Singer E, Yiannoutsos CT, Sherman S, Gensler G, Moore DJ, Chen T and Soukup VM (2013).
Neurovirological correlation with HIV-associated neurocognitive disorders and encephalitis in a HAART-
era cohort. J Acquir Immune Defic Syndr, 62(5): 487-495.

206

Giannelli M, Cosottini M, Michelassi MC, Lazzarotti G, Belmonte G, Bartolozzi C and Lazzeri M (2009).
Dependence of brain DTI maps of fractional anisotropy and mean diffusivity on the number of diffusion
weighting directions. J Appl Clin Med Phys, 11(1): 2927.

Giulietti G, Torso M, Serra L and Span (2018). Whole brain white matter histogram analysis of diffusion
tensor imaging data detects microstructural damage in mild cognitive impairment and Alzheimer's disease
patients. Journal of Magnetic Resonance Imaging.

Gongvatana A, Schweinsburg BC, Taylor MJ, Theilmann RJ, Letendre SL, Alhassoon OM, Jacobus J,
Woods SP, Jernigan TL, Ellis RJ, Frank LR and Grant I (2009). White matter tract injury and cognitive
impairment in human immunodeficiency virus-infected individuals. J Neurovirol, 15(2): 187-195.

Gongvatana A, Cohen RA, Correia S, Devlin KN, Miles J, Kang H, Ombao H, Navia B, Laidlaw DH and
Tashima KT (2011). Clinical contributors to cerebral white matter integrity in HIV-infected individuals.
Journal of Neurovirology, 17(5): 477-486.

Gongvatana A, Harezlak J, Buchthal S, Daar E, Schifitto G, Campbell T, Taylor M, Singer E, Algers J,
Zhong J, Brown M, McMahon D, So YT, Mi D, Heaton R, Robertson K, Yiannoutsos C, Cohen RA,
Navia B and Consortium HIVN (2013). Progressive cerebral injury in the setting of chronic HIV infection
and antiretroviral therapy. J Neurovirol, 19(3): 209-218.

Gonzalez-Scarano F and Martin-Garcia J (2005). The neuropathogenesis of AIDS. Nat Rev Immunol, 5(1):
69-81.

Goodkin K, Wilkie FL, Concha M, Hinkin CH, Symes S, Baldewicz TT, Asthana D, Fujimura RK, Lee
D, van Zuilen MH, Khamis I, Shapshak P and Eisdorfer C (2001). Aging and neuro-AIDS conditions and
the changing spectrum of HIV-1-associated morbidity and mortality. J Clin Epidemiol, 54 Suppl 1: S35-
43.

Grant I, Franklin DR, Jr., Deutsch R, Woods SP, Vaida F, Ellis RJ, Letendre SL, Marcotte TD, Atkinson
JH, Collier AC, Marra CM, Clifford DB, Gelman BB, McArthur JC, Morgello S, Simpson DM,
McCutchan JA, Abramson I, Gamst A, Fennema-Notestine C, Smith DM, Heaton RK and Group C
(2014). Asymptomatic HIV-associated neurocognitive impairment increases risk for symptomatic decline.
Neurology, 82(23): 2055-2062.

Grech-Sollars M, Hales PW, Miyazaki K, Raschke F, Rodriguez D, Wilson M, Gill SK, Banks T, Saunders
DE, Clayden JD, Gwilliam MN, Barrick TR, Morgan PS, Davies NP, Rossiter J, Auer DP, Grundy R,
Leach MO, Howe FA, Peet AC and Clark CA (2015). Multi-centre reproducibility of diffusion MRI
parameters for clinical sequences in the brain. NMR in Biomedicine, 28: 468-485.

Green DA, Masliah E, Vinters HV, Beizai P, Moore DJ and Achim CL (2005). Brain deposition of beta-
amyloid is a common pathologic feature in HIV positive patients. AIDS, 19(4): 407-411.

Greve DN and Fischl B (2009). Accurate and robust brain image alignment using boundary-based
registration. Neuroimage, 48(1): 63-72.

207

Grothe MJ, Barthel H, Sepulcre J, Dyrba M, Sabri O, Teipel SJ and Initiative AsDN (2017). In vivo
staging of regional amyloid deposition. Neurology, 89: 2031-2038.

Gullett JM, Lamb DG, Porges E, Woods AJ, Rieke J, Thompson P, Jahanshad N, Nir TM, Tashima K and
Cohen RA (2018). The impact of alcohol use on frontal white matter in HIV.

Gutman B, Svarer C, Leow A, Yanovsky I, Toga A and Thompson P (2010). Creating unbiased minimal
deformation templates for brain volume registration. Organization for Human Brain Mapping (OHBM),
Barcelona, Spain.

Hackman DA, Farah MJ and Meaney MJ (2010). Socioeconomic status and the brain: mechanistic insights
from human and animal research. Nat Rev Neurosci, 11(9): 651-659.

Hagmann P, Cammoun L, Gigandet X, Meuli R, Honey CJ, Wedeen VJ and Sporns O (2008). Mapping
the structural core of human cerebral cortex. PLoS Biol, 6(7): e159.

Harezlak J, Buchthal S, Taylor M, Schifitto G, Zhong J, Daar E, Alger J, Singer E, Campbell T,
Yiannoutsos C, Cohen R, Navia B and Consortium HIVN (2011). Persistence of HIV-associated cognitive
impairment, inflammation, and neuronal injury in era of highly active antiretroviral treatment. AIDS,
25(5): 625-633.

Harezlak J, Cohen R, Gongvatana A, Taylor M, Buchthal S, Schifitto G, Zhong J, Daar ES, Alger JR,
Brown M, Singer EJ, Campbell TB, McMahon D, So YT, Yiannoutsos CT, Navia BA and Consortium
HIVN (2014). Predictors of CNS injury as measured by proton magnetic resonance spectroscopy in the
setting of chronic HIV infection and CART. J Neurovirol, 20(3): 294-303.

Harms RL, Fritz FJ, Tobisch A, Goebel R and Roebroeck A (2017). Robust and fast nonlinear optimization
of diffusion MRI microstructure models. Neuroimage, 155: 82-96.

Harsan LA, Poulet P, Guignard B, Steibel J and Parizel Na (2006). Brain dysmyelination and recovery
assessment by noninvasive in vivo diffusion tensor magnetic resonance imaging. Journal of Neuroscience
Research, 83(January): 392-402.

Harsan LA, Poulet P, Guignard B, Parizel N, Skoff RP and Ghandour MS (2007). Astrocytic hypertrophy
in dysmyelination influences the diffusion anisotropy of white matter. J Neurosci Res, 85(5): 935-944.

He Y, Chen Z and Evans A (2008). Structural insights into aberrant topological patterns of large-scale
cortical networks in Alzheimer's disease. J Neurosci, 28(18): 4756-4766.

Heaps JM, Sithinamsuwan P, Paul R, Lerdlum S, Pothisri M, Clifford D, Tipsuk S, Catella S, Busovaca
E, Fletcher JL, Raudabaugh B, Ratto-Kim S, Valcour V, Ananworanich J and groups Ss (2015).
Association between brain volumes and HAND in cART-naive HIV+ individuals from Thailand. J
Neurovirol, 21(2): 105-112.

208

Heaton RK, Grant I, Butters N, White DA, Kirson D, Atkinson JH, McCutchan JA, Taylor MJ, Kelly MD,
Ellis RJ and et al. (1995). The HNRC 500--neuropsychology of HIV infection at different disease stages.
HIV Neurobehavioral Research Center. J Int Neuropsychol Soc, 1(3): 231-251.

Heaton RK, Clifford DB, Franklin DR, Jr., Woods SP, Ake C, Vaida F, Ellis RJ, Letendre SL, Marcotte
TD, Atkinson JH, Rivera-Mindt M, Vigil OR, Taylor MJ, Collier AC, Marra CM, Gelman BB, McArthur
JC, Morgello S, Simpson DM, McCutchan JA, et al. (2010). HIV-associated neurocognitive disorders
persist in the era of potent antiretroviral therapy: CHARTER Study. Neurology, 75(23): 2087-2096.

Heaton RK, Franklin DR, Ellis RJ, McCutchan JA, Letendre SL, Leblanc S, Corkran SH, Duarte NA,
Clifford DB, Woods SP, Collier AC, Marra CM, Morgello S, Mindt MR, Taylor MJ, Marcotte TD,
Atkinson JH, Wolfson T, Gelman BB, McArthur JC, et al. (2011). HIV-associated neurocognitive
disorders before and during the era of combination antiretroviral therapy: differences in rates, nature, and
predictors. J Neurovirol, 17(1): 3-16.

Heaton RK, Franklin DR, Jr., Deutsch R, Letendre S, Ellis RJ, Casaletto K, Marquine MJ, Woods SP,
Vaida F, Atkinson JH, Marcotte TD, McCutchan JA, Collier AC, Marra CM, Clifford DB, Gelman BB,
Sacktor N, Morgello S, Simpson DM, Abramson I, et al. (2015). Neurocognitive change in the era of HIV
combination antiretroviral therapy: the longitudinal CHARTER study. Clin Infect Dis, 60(3): 473-480.

Heinen R, Bouvy WH, Mendrik AM, Viergever MA, Biessels GJ and de Bresser J (2016). Robustness of
automated methods for brain volume measurements across different MRI field strengths. PLoS One,
11(10): e0165719.

Hibar DP, Westlye LT, van Erp TG, Rasmussen J, Leonardo CD, Faskowitz J, Haukvik UK, Hartberg
CB, Doan NT, Agartz I, Dale AM, Gruber O, Kramer B, Trost S, Liberg B, Abe C, Ekman CJ, Ingvar M,
Landen M, Fears SC, et al. (2016). Subcortical volumetric abnormalities in bipolar disorder. Mol
Psychiatry.

Hilgetag CC, Burns GA, O'Neill MA, Scannell JW and Young MP (2000). Anatomical connectivity
defines the organization of clusters of cortical areas in the macaque monkey and the cat. Philos Trans R
Soc Lond B Biol Sci, 355(1393): 91-110.

Hinkin CH, van Gorp WG, Mandelkern MA, Gee M, Satz P, Holston S, Marcotte TD, Evans G, Paz DH,
Ropchan JR and et al. (1995). Cerebral metabolic change in patients with AIDS: report of a six-month
follow-up using positron-emission tomography. J Neuropsychiatry Clin Neurosci, 7(2): 180-187.

Hoare J, Ransford GL, Phillips N, Amos T, Donald K and Stein DJ (2014). Systematic review of
neuroimaging studies in vertically transmitted HIV positive children and adolescents. Metab Brain Dis,
29(2): 221-229.

Hogg R, Strathdee S, Craib K, V O'Shaughnessy M, Montaner JSG and Schechter M (1994). Lower
socioeconomic status and shorter survival following HIV infection.

Holmes CJ, Hoge R, Collins L, Woods R, Toga AW and Evans AC (1998). Enhancement of MR images
using registration for signal averaging. J Comput Assist Tomogr, 22(2): 324-333.
209

Holt JL, Kraft-Terry SD and Chang L (2012). Neuroimaging studies of the aging HIV-1-infected brain. J
Neurovirol, 18(4): 291-302.

Honey CJ, Sporns O, Cammoun L, Gigandet X, Thiran JP, Meuli R and Hagmann P (2009). Predicting
human resting-state functional connectivity from structural connectivity. Proc Natl Acad Sci U S A,
106(6): 2035-2040.

Hu C, Hua X, Ying J, Thompson PM, Fakhri GE and Li Q (2016). Localizing Sources of Brain Disease
Progression with Network Diffusion Model. IEEE J Sel Top Signal Process, 10(7): 1214-1225.

Hua K, Zhang J, Wakana S, Jiang H, Li X, Reich DS, Calabresi PA, Pekar JJ, van Zijl PCM and Mori S
(2008). Tract probability maps in stereotaxic spaces: Analyses of white matter anatomy and tract-specific
quantification. Neuroimage, 39: 336-347.

Hua X, Lee S, Yanovsky I, Leow AD, Chou YY, Ho AJ, Gutman B, Toga AW, Jack CR, Jr., Bernstein
MA, Reiman EM, Harvey DJ, Kornak J, Schuff N, Alexander GE, Weiner MW, Thompson PM and
Alzheimer's Disease Neuroimaging I (2009a). Optimizing power to track brain degeneration in
Alzheimer's disease and mild cognitive impairment with tensor-based morphometry: an ADNI study of
515 subjects. Neuroimage, 48(4): 668-681.

Hua X, Leow AD, Levitt JG, Caplan R, Thompson PM and Toga AW (2009b). Detecting brain growth
patterns in normal children using tensor-based morphometry. Human Brain Mapping, 30(1): 209-219.

Hua X, Boyle CP, Harezlak J, Tate DF, Yiannoutsos CT, Cohen R, Schifitto G, Gongvatana A, Zhong J,
Zhu T, Taylor MJ, Campbell TB, Daar ES, Alger JR, Singer E, Buchthal S, Toga AW, Navia B, Thompson
PM and Consortium HIVN (2013a). Disrupted cerebral metabolite levels and lower nadir CD4 + counts
are linked to brain volume deficits in 210 HIV-infected patients on stable treatment. Neuroimage Clin, 3:
132-142.

Hua X, Hibar DP, Ching CRK, Boyle CP, Rajagopalan P, Gutman BA, Leow AD, Toga AW, Jack Jr CR,
Harvey D, Weiner MW and Thompson PM (2013b). Unbiased tensor-based morphometry: Improved
robustness and sample size estimates for Alzheimer's disease clinical trials. Neuroimage, 66: 648-661.

Hua X, Thompson PM, Leow AD, Madsen SK, Caplan R, Alger JR, O'Neill J, Joshi K, Smalley SL, Toga
AW and Levitt JG (2013c). Brain growth rate abnormalities visualized in adolescents with autism. Human
Brain Mapping, 34(2): 425-436.

Hua X, Ching CR, Mezher A, Gutman BA, Hibar DP, Bhatt P, Leow AD, Jack CR, Jr., Bernstein MA,
Weiner MW, Thompson PM and Alzheimer's Disease Neuroimaging I (2016). MRI-based brain atrophy
rates in ADNI phase 2: acceleration and enrichment considerations for clinical trials. Neurobiol Aging,
37: 26-37.

Hugenschmidt CE, Peiffer AM, Kraft RA, Casanova R, Deibler AR, Burdette JH, Maldjian JA and
Laurienti PJ (2008). Relating imaging indices of white matter integrity and volume in healthy older adults.
Cereb Cortex, 18(2): 433-442.

210

Hult B, Chana G, Masliah E and Everall I (2008). Neurobiology of HIV. Int Rev Psychiatry, 20(1): 3-13.

Iglesias JE, Liu CY, Thompson PM and Tu Z (2011). Robust brain extraction across datasets and
comparison with publicly available methods. IEEE Trans Med Imaging, 30(9): 1617-1634.

Isaev DY, Nir TM, Jahanshad N, Villalon-Reina JE, Zhan L, Leow AD and Thompson PM (2017).
Improved clinical diffusion MRI reliability using a tensor distribution function compared to a single
tensor. 12th International Symposium on Medical Information Processing and Analysis, SPIE.

Iturria-Medina Y, Sotero RC, Canales-Rodriguez EJ, Aleman-Gomez Y and Melie-Garcia L (2008).
Studying the human brain anatomical network via diffusion-weighted MRI and Graph Theory.
Neuroimage, 40(3): 1064-1076.

Jack CR, Jr., Knopman DS, Jagust WJ, Petersen RC, Weiner MW, Aisen PS, Shaw LM, Vemuri P, Wiste
HJ, Weigand SD, Lesnick TG, Pankratz VS, Donohue MC and Trojanowski JQ (2013). Tracking
pathophysiological processes in Alzheimer's disease: an updated hypothetical model of dynamic
biomarkers. Lancet Neurol, 12(2): 207-216.

Jack CR, Jr., Barnes J, Bernstein MA, Borowski BJ, Brewer J, Clegg S, Dale AM, Carmichael O, Ching
C, DeCarli C, Desikan RS, Fennema-Notestine C, Fjell AM, Fletcher E, Fox NC, Gunter J, Gutman BA,
Holland D, Hua X, Insel P, et al. (2015). Magnetic resonance imaging in Alzheimer's Disease
Neuroimaging Initiative 2. Alzheimers Dement, 11(7): 740-756.

Jacobs HI, Van Boxtel MP, Jolles J, Verhey FR and Uylings HB (2012). Parietal cortex matters in
Alzheimer's disease: an overview of structural, functional and metabolic findings. Neurosci Biobehav Rev,
36(1): 297-309.

Jahanshad N, Aganj I, Lenglet C, Joshi A, Jin Y, Barysheva M, McMahon K, de Zubicaray G, G. Martin
N, Wright M, W. Toga A, Sapiro G and Thompson P (2011a). Sex differences in the human connectome:
4-Tesla high angular resolution diffusion imaging (HARDI) tractography in 234 young adult twins.
Proceedings of the 8th IEEE International Symposium on Biomedical Imaging: From Nano to Macro,
ISBI.

Jahanshad N, Aganj I, Lenglet C, Joshi A, Jin Y, Barysheva M, McMahon KL, De Zubicaray GI, Martin
NG, Wright MJ, Toga AW, Sapiro G and Thompson PM (2011b). Sex differences in the human
connectome: 4-Tesla high angular resolution diffusion imaging (HARDI) tractography in 234 young adult
twins. Proceedings - International Symposium on Biomedical Imaging.

Jahanshad N, Valcour VG, Nir TM, Kohannim O, Busovaca E, Nicolas K and Thompson PM (2012).
Disrupted brain networks in the aging HIV+ population. Brain connectivity, 2(6): 335-344.

Jahanshad N, Kochunov PV, Sprooten E, Mandl RC, Nichols TE, Almasy L, Blangero J, Brouwer RM,
Curran JE, de Zubicaray GI, Duggirala R, Fox PT, Hong LE, Landman BA, Martin NG, McMahon KL,
Medland SE, Mitchell BD, Olvera RL, Peterson CP, et al. (2013a). Multi-site genetic analysis of diffusion
images and voxelwise heritability analysis: a pilot project of the ENIGMA-DTI working group.
Neuroimage, 81: 455-469.
211

Jahanshad N, Rajagopalan P, Hua X, Hibar DP, Nir TM, Toga AW, Jack CR, Jr., Saykin AJ, Green RC,
Weiner MW, Medland SE, Montgomery GW, Hansell NK, McMahon KL, de Zubicaray GI, Martin NG,
Wright MJ and Thompson PM (2013b). Genome-wide scan of healthy human connectome discovers
SPON1 gene variant influencing dementia severity. Proc Natl Acad Sci U S A, 110(12): 4768-4773.

Jahanshad N, Nir TM, Toga AW, Jack CR, Jr., Bernstein MA, Weiner MW, Thompson PM and
Alzheimer's Disease Neuroimaging I (2015a). Seemingly unrelated regression empowers detection of
network failure in dementia. Neurobiology of aging, 36 Suppl 1(0 1): S103-S112.

Jahanshad N, Roshchupkin G, Faskowitz J, Hibar DP, Gutman BA, Adams HH, Niessen WJ, Vernooij
MW, Ikram MA and Zwiers MP (2015b). Multi-site meta-analysis of image-wide genome-wide
associations of morphometry. MICCAI Imaging Genetics Workshop.

Jbabdi S, Behrens TEJ and Smith SM (2010). Crossing fibres in tract-based spatial statistics. Neuroimage,
49: 249-256.

Jensen JH, Helpern JA, Ramani A, Lu H and Kaczynski K (2005). Diffusional kurtosis imaging: the
quantification of non-gaussian water diffusion by means of magnetic resonance imaging. Magn Reson
Med, 53(6): 1432-1440.

Jernigan TL, Gamst AC, Archibald SL, Fennema-Notestine C, Mindt MR, Marcotte TD, Heaton RK, Ellis
RJ and Grant I (2005). Effects of methamphetamine dependence and HIV infection on cerebral
morphology. Am J Psychiatry, 162(8): 1461-1472.

Jernigan TL, Archibald SL, Fennema-Notestine C, Taylor MJ, Theilmann RJ, Julaton MD, Notestine RJ,
Wolfson T, Letendre SL, Ellis RJ, Heaton RK, Gamst AC, Franklin DR, Jr., Clifford DB, Collier AC,
Gelman BB, Marra C, McArthur JC, McCutchan JA, Morgello S, Simpson DM, Grant I and Group C
(2011). Clinical factors related to brain structure in HIV: the CHARTER study. J Neurovirol, 17(3): 248-
257.

Jeurissen B, Leemans A, Tournier JD, Jones DK and Sijbers J (2013). Investigating the prevalence of
complex fiber configurations in white matter tissue with diffusion magnetic resonance imaging. Hum
Brain Mapp, 34(11): 2747-2766.

Jian B, Vemuri BC, Ozarslan E, Carney PR and Mareci TH (2007). A novel tensor distribution model for
the diffusion-weighted MR signal. Neuroimage, 37(1): 164-176.

Jin Y, Huang C, Daianu M, Zhan L, Dennis EL, Reid RI, Jack CR, Zhu H and Thompson PM (2017). 3D
tract-specific local and global analysis of white matter integrity in Alzheimer's disease. Human Brain
Mapping, 38(3): 1191-1207.

Johnson WE, Li C and Rabinovic A (2007). Adjusting batch effects in microarray expression data using
empirical Bayes methods. Biostatistics, 8(1): 118-127.
Jones DK (2004). The effect of gradient sampling schemes on measures derived from diffusion tensor
MRI: a Monte Carlo study. Magn Reson Med, 51(4): 807-815.

212

Jones DK and Cercignani M (2010). Twenty-five pitfalls in the analysis of diffusion MRI data. NMR
Biomed, 23(7): 803-820.

Jovicich J, Czanner S, Han X, Salat D, van der Kouwe A, Quinn B, Pacheco J, Albert M, Killiany R,
Blacker D, Maguire P, Rosas D, Makris N, Gollub R, Dale A, Dickerson BC and Fischl B (2009). MRI-
derived measurements of human subcortical, ventricular and intracranial brain volumes: Reliability effects
of scan sessions, acquisition sequences, data analyses, scanner upgrade, scanner vendors and field
strengths. Neuroimage, 46(1): 177-192.

Kallianpur KJ, Shikuma C, Kirk GR, Shiramizu B, Valcour V, Chow D, Souza S, Nakamoto B and
Sailasuta N (2013). Peripheral blood HIV DNA is associated with atrophy of cerebellar and subcortical
gray matter. Neurology, 80(19): 1792-1799.

Kamkwalala A and Newhouse P (2017). Mechanisms of cognitive aging in the HIV-positive adult. Curr
Behav Neurosci Rep, 4(3): 188-197.

Kantarci K, Petersen RC, Boeve BF, Knopman DS, Weigand SD, O'Brien PC, Shiung MM, Smith GE,
Ivnik RJ, Tangalos EG and Jack CR, Jr. (2005). DWI predicts future progression to Alzheimer disease in
amnestic mild cognitive impairment. Neurology, 64(5): 902-904.

Kantarci K, Murray ME, Schwarz CG, Reid RI, Przybelski SA, Lesnick T, Zuk SM, Raman MR, Senjem
ML, Gunter JL, Boeve BF, Knopman DS, Parisi JE, Petersen RC, Jack CR and Dickson DW (2017).
White-matter integrity on DTI and the pathologic staging of Alzheimer's disease. Neurobiology of Aging,
56: 172-179.

Kaul M, Garden GA and Lipton SA (2001). Pathways to neuronal injury and apoptosis in HIV-associated
dementia. Nature, 410(6831): 988-994.

Kauppi K, Fan CC, McEvoy LK, Holland D, Tan CH, Chen C-H, Andreassen OA, Desikan RS and Dale
AMa (2018). Combining polygenic hazard score with volumetric MRI and cognitive measures improves
prediction of progression from mild cognitive impairment to Alzheimer's disease. Frontiers in
Neuroscience, 12: 260.

Kavcic V, Ni H, Zhu T, Zhong J and Duffy CJ (2008). White matter integrity linked to functional
impairments in aging and early Alzheimer's disease. Alzheimer's & Dementia, 4(6): 381-389.

Kellner E, Dhital B, Kiselev VG and Reisert M (2016). Gibbs-ringing artifact removal based on local
subvoxel-shifts. Magn Reson Med, 76(5): 1574-1581.

Kelly S, Jahanshad N, Zalesky A, Kochunov P, Agartz I, Alloza C, Andreassen OA, Arango C, Banaj N,
Bouix S, Bousman CA, Brouwer RM, Bruggemann J, Bustillo J, Cahn W, Calhoun V, Cannon D, Carr V,
Catts S, Chen J, et al. (2018). Widespread white matter microstructural differences in schizophrenia across
4322 individuals: results from the ENIGMA Schizophrenia DTI Working Group. Mol Psychiatry, 23(5):
1261-1269.

213

Kim M, Ronen I, Ugurbil K and Kim DS (2006). Spatial resolution dependence of DTI tractography in
human occipito-callosal region. Neuroimage, 32(3): 1243-1249.

Kingsley PB and Monahan WG (2004). Selection of the optimum b factor for diffusion-weighted magnetic
resonance imaging assessment of ischemic stroke. Magn Reson Med, 51(5): 996-1001.

Kinoshita Y, Ohnishi A, Kohshi K and Yokota A (1999). Apparent diffusion coefficient on rat brain and
nerves intoxicated with methylmercury. Environ Res, 80(4): 348-354.

Kneynsberg A, Combs B, Christensen K, Morfini G and Kanaan NM (2017). Axonal degeneration in
tauopathies: Disease relevance and underlying mechanisms. Frontiers in Neuroscience, 11: 572.

Kochunov P, Jahanshad N, Sprooten E, Nichols TE, Mandl RC, Almasy L, Booth T, Brouwer RM, Curran
JE, de Zubicaray GI, Dimitrova R, Duggirala R, Fox PT, Elliot Hong L, Landman BA, Lemaitre H, Lopez
LM, Martin NG, McMahon KL, Mitchell BD, et al. (2014). Multi-site study of additive genetic effects on
fractional anisotropy of cerebral white matter: Comparing meta and megaanalytical approaches for data
pooling. Neuroimage, 95: 136-150.

Kochunov P, Dickie EW, Viviano JD, Turner J, Kingsley PB, Jahanshad N, Thompson PM, Ryan MC,
Fieremans E, Novikov D, Veraart J, Hong EL, Malhotra AK, Buchanan RW, Chavez S and Voineskos
AN (2018). Integration of routine QA data into mega-analysis may improve quality and sensitivity of
multisite diffusion tensor imaging studies. Hum Brain Mapp, 39(2): 1015-1023.

Kruman, II, Nath A and Mattson MP (1998). HIV-1 protein Tat induces apoptosis of hippocampal neurons
by a mechanism involving caspase activation, calcium overload, and oxidative stress. Exp Neurol, 154(2):
276-288.

Kuhn T, Schonfeld D, Sayegh P, Arentoft A, Jones JD, Hinkin CH, Bookheimer SY and Thames AD
(2017). The effects of HIV and aging on subcortical shape alterations: A 3D morphometric study. Hum
Brain Mapp, 38(2): 1025-1037.

Kuhn T, Kaufmann T, Doan NT and Westlye LT (2018). An augmented aging process in brain white
matter in HIV. 39(6): 2532-2540.

Kure K, Weidenheim KM, Lyman WD and Dickson DW (1990). Morphology and distribution of HIV-1
gp41-positive microglia in subacute AIDS encephalitis. Pattern of involvement resembling a multisystem
degeneration. Acta Neuropathol, 80(4): 393-400.

Lamers SL, Rose R, Maidji E, Agsalda-Garcia M, Nolan DJ, Fogel GB, Salemi M, Garcia DL, Bracci P,
Yong W, Commins D, Said J, Khanlou N, Hinkin CH, Sueiras MV, Mathisen G, Donovan S, Shiramizu
B, Stoddart CA, McGrath MS and Singer EJ (2016). HIV DNA is frequently present within pathologic
tissues evaluated at autopsy from combined antiretroviral therapy-treated patients with undetectable viral
loads. J Virol, 90(20): 8968-8983.

Langers DR, Jansen JF and Backes WH (2007). Enhanced signal detection in neuroimaging by means of
regional control of the global false discovery rate. Neuroimage, 38(1): 43-56.
214

Laurent C, Buée L and Blum D (2018). Tau and neuroinflammation: What impact for Alzheimer's Disease
and Tauopathies? Biomedical Journal, 41(1): 21-33.

Le Bihan D and Johansen-Berg H (2012). Diffusion MRI at 25: exploring brain tissue structure and
function. Neuroimage, 61(2): 324-341.

Lee J-T, Xu J, Lee J-M, Ku G, Han X, Yang D-I, Chen S and Hsu CY (2004). Amyloid-β peptide induces
oligodendrocyte death by activating the neutral sphingomyelinase–ceramide pathway. The Journal of Cell
Biology, 164: 123-131.

Lee PL, Yiannoutsos CT, Ernst T, Chang L, Marra CM, Jarvik JG, Richards TL, Kwok EW, Kolson DL,
Simpson D, Tang CY, Schifitto G, Ketonen LM, Meyerhoff DJ, Lenkinski RE, Gonzalez RG and Navia
BA (2003). A multi-center 1H MRS study of the AIDS dementia complex: validation and preliminary
analysis. J Magn Reson Imaging, 17(6): 625-633.

Leow AD, Yanovsky I, Chiang MC, Lee AD, Klunder AD, Lu A, Becker JT, Davis SW, Toga AW and
Thompson PM (2007). Statistical properties of Jacobian maps and the realization of unbiased large-
deformation nonlinear image registration. IEEE Trans Med Imaging, 26(6): 822-832.

Leow AD, Zhu S, Zhan L, McMahon K, de Zubicaray GI, Meredith M, Wright MJ, Toga AW and
Thompson PM (2009). The tensor distribution function. Magn Reson Med, 61(1): 205-214.

Leporé N, Brun C, Pennec X, Chou Y-Y, Lopez OL, Aizenstein HJ, Becker JT, Toga AW and Thompson
PM (2007). Mean template for tensor-based morphometry using deformation tensors, Berlin, Heidelberg,
Springer Berlin Heidelberg.

Letendre S, Ellis RJ, Best B, Bhatt A, Marquie-Beck J, LeBlanc S, Rossi S, Capparelli E and McCutchan
A (2009). Penetration and effectiveness of antiretroviral therapy in the central nervous system. Anti-
Inflammatory & Anti-Allergy Agents in Medicinal Chemistry, 8(2): 169-183.

Li J, Jin Y, Shi Y, Dinov ID, Wang DJ, Toga AW and Thompson PM (2013). Voxelwise spectral
diffusional connectivity and its applications to Alzheimer's disease and intelligence prediction. Med Image
Comput Comput Assist Interv, 16(Pt 1): 655-662.

Liao YF, Wang BJ, Cheng HT, Kuo LH and Wolfe MS (2004). Tumor necrosis factor-alpha, interleukin-
1beta, and interferon-gamma stimulate gamma-secretase-mediated cleavage of amyloid precursor protein
through a JNK-dependent MAPK pathway. J Biol Chem, 279(47): 49523-49532.

Liu M, Zhang J, Adeli E and Shen D (2017). Deep multi-task multi-channel learning for joint classification
and regression of brain status. Medical image computing and computer-assisted intervention : MICCAI ...
International Conference on Medical Image Computing and Computer-Assisted Intervention, 10435: 3-
11.

Lo CY, Wang PN, Chou KH, Wang J, He Y and Lin CP (2010). Diffusion tensor tractography reveals
abnormal topological organization in structural cortical networks in Alzheimer's disease. J Neurosci,
30(50): 16876-16885.
215

Logue MW, van Rooij SJH, Dennis EL, Davis SL, Hayes JP, Stevens JS, Densmore M, Haswell CC, Ipser
J, Koch SBJ, Korgaonkar M, Lebois LAM, Peverill M, Baker JT, Boedhoe PSW, Frijling JL, Gruber SA,
Harpaz-Rotem I, Jahanshad N, Koopowitz S, et al. (2018). Smaller Hippocampal Volume in Posttraumatic
Stress Disorder: A Multisite ENIGMA-PGC Study: Subcortical Volumetry Results From Posttraumatic
Stress Disorder Consortia. Biol Psychiatry, 83(3): 244-253.

Lucas GM (2011). Substance abuse, adherence with antiretroviral therapy, and clinical outcomes among
HIV-infected individuals. Life sciences, 88(21-22): 948-952.

Lysandropoulos AP, Absil J, Metens T, Mavroudakis N, Guisset F, Van Vlierberghe E, Smeets D, David
P, Maertens A and Van Hecke W (2016). Quantifying brain volumes for Multiple Sclerosis patients
follow-up in clinical practice - comparison of 1.5 and 3 Tesla magnetic resonance imaging. Brain Behav,
6(2): e00422.

Machtinger EL, Wilson TC, Haberer JE and Weiss DS (2012). Psychological trauma and PTSD in HIV-
positive women: a meta-analysis. AIDS Behav, 16(8): 2091-2100.

Maggipinto T, Bellotti R, Amoroso N, Diacono D, Donvito G, Lella E, Monaco A, Antonella Scelsi M
and Tangaro S (2017). DTI measurements for Alzheimer's classification. Phys Med Biol, 62(6): 2361-
2375.

Maier-Hein KH, Neher PF, Houde JC and Ct (2017). The challenge of mapping the human connectome
based on diffusion tractography. Nature Communications, 8(1).

Mani A, Mullainathan S, Shafir E and Zhao J (2013). Poverty impedes cognitive function. Science,
341(6149): 976-980.

Manjon JV, Coupe P, Concha L, Buades A, Collins DL and Robles M (2013). Diffusion weighted image
denoising using overcomplete local PCA. PLoS One, 8(9): e73021.

Marcotte TD, Deutsch R, McCutchan JA, Moore DJ, Letendre S, Ellis RJ, Wallace MR, Heaton RK and
Grant I (2003). Prediction of Incident Neurocognitive Impairment by Plasma HIV RNA and CD4 Levels
Early After HIV Seroconversion. Archives of Neurology, 60(10): 1406-1412.

Marek K, Jennings D, Lasch S, Siderowf A, Tanner C, Simuni T, Coffey C, Kieburtz K, Flagg E,
Chowdhury S, Poewe W, Mollenhauer B, Klinik P-E, Sherer T, Frasier M, Meunier C, Rudolph A,
Casaceli C, Seibyl J, Mendick S, et al. (2011). The Parkinson Progression Marker Initiative (PPMI).
Progress in Neurobiology, 95(4): 629-635.

Martin-Thormeyer EM and Paul RH (2009). Drug abuse and hepatitis C infection as comorbid features of
HIV associated neurocognitive disorder: neurocognitive and neuroimaging features. Neuropsychology
review, 19(2): 215-231.

Martinez-Picado J and Deeks SG (2016). Persistent HIV-1 replication during antiretroviral therapy.
Current opinion in HIV and AIDS, 11(4): 417-423.

216

Masters MC and Ances BM (2014). Role of neuroimaging in HIV-associated neurocognitive disorders.
Seminars in neurology, 34(1): 89-102.

Mayo CD, Mazerolle EL, Ritchie L, Fisk JD and Gawryluk JR (2017). Longitudinal changes in
microstructural white matter metrics in Alzheimer's disease. Neuroimage: Clinical, 13: 330-338.

McArthur JC (2004). HIV dementia: an evolving disease. J Neuroimmunol, 157(1-2): 3-10.

McArthur JC, Steiner J, Sacktor N and Nath A (2010). Human immunodeficiency virus-associated
neurocognitive disorders: Mind the gap. Ann Neurol, 67(6): 699-714.

McArthur RA (2012). Translational neuroimaging: Tools for CNS drug discovery, development and
treatment, Elsevier Science.

Meichen Y, A. LK, A. CP, L. PM, Melvin M, Maurizio F, H. TM, M. WM, T. SR and I. SY (2018).
Statistical harmonization corrects site effects in functional connectivity measurements from multi-site
fMRI data. Human Brain Mapping, [Epub ahead of print].

Mielke MM, Kozauer NA, Chan KC, George M, Toroney J, Zerrate M, Bandeen-Roche K, Wang MC,
Vanzijl P, Pekar JJ, Mori S, Lyketsos CG and Albert M (2009). Regionally-specific diffusion tensor
imaging in mild cognitive impairment and Alzheimer's disease. Neuroimage, 46(1): 47-55.

Milanini B and Valcour V (2017). Differentiating HIV-associated neurocognitive disorders from
Alzheimer's disease: An emerging issue in geriatric NeuroHIV. Current HIV/AIDS reports, 14(4): 123-
132.

Mirzaalian H, Ning L, Savadjiev P, Pasternak O, Bouix S, Michailovich O, Karmacharya S, Grant G,
Marx CE, Morey RA, Flashman LA, George MS, McAllister TW, Andaluz N, Shutter L, Coimbra R,
Zafonte RD, Coleman MJ, Kubicki M, Westin CF, Stein MB, Shenton ME and Rathi Y (2018). Multi-site
harmonization of diffusion MRI data in a registration framework. Brain Imaging Behav, 12(1): 284-295.

Moore DJ, Masliah E, Rippeth JD, Gonzalez R, Carey CL, Cherner M, Ellis RJ, Achim CL, Marcotte TD,
Heaton RK and Grant I (2006). Cortical and subcortical neurodegeneration is associated with HIV
neurocognitive impairment. Aids, 20(6): 879-887.

Moore EE, Hohman TJ, Badami FS, Pechman KR, Osborn KE, Acosta LMY, Bell SP, Babicz MA,
Gifford KA, Anderson AW, Goldstein LE, Blennow K, Zetterberg H and Jefferson AL (2018).
Neurofilament relates to white matter microstructure in older adults. Neurobiology of Aging, 70: 233-241.

Morgello S (2018). Chapter 2 - HIV neuropathology. Handbook of Clinical Neurology. B. J. Brew,
Elsevier. 152: 3-19.

Mori S, Oishi K, Jiang H, Jiang L, Li X, Akhter K, Hua K, Faria AV, Mahmood A, Woods R, Toga AW,
Pike GB, Neto PR, Evans A, Zhang J, Huang H, Miller MI, van Zijl P and Mazziotta J (2008). Stereotaxic
white matter atlas based on diffusion tensor imaging in an ICBM template. Neuroimage, 40(2): 570-582.

217

Mueller SG, Weiner MW, Thal LJ, Petersen RC, Jack CR, Jagust W, Trojanowski JQ, Toga AW and
Beckett L (2005). Ways toward an early diagnosis in Alzheimer’s disease: The Alzheimer’s Disease
Neuroimaging Initiative (ADNI). Alzheimer's & Dementia, 1: 55-66.

Müller H-P and Kassubek J (2013). Diffusion tensor magnetic resonance imaging in the analysis of
neurodegenerative diseases. Journal of Visualized Experiments.

Muller MJ, Greverus D, Dellani PR, Weibrich C, Wille PR, Scheurich A, Stoeter P and Fellgiebel A
(2005). Functional implications of hippocampal volume and diffusivity in mild cognitive impairment.
Neuroimage, 28(4): 1033-1042.

Nair G, Tanahashi Y, Low HP, Billings-Gagliardi S, Schwartz WJ and Duong TQ (2005). Myelination
and long diffusion times alter diffusion-tensor-imaging contrast in myelin-deficient shiverer mice.
Neuroimage, 28(1): 165-174.

Nakagawa F, May M and Phillips A (2013). Life expectancy living with HIV: recent estimates and future
implications. Curr Opin Infect Dis, 26(1): 17-25.

Navia B, Harezlak J, Schifitto G, Taylor MJ, Heaton RK, Robertson K, Daar ES, Campbell T, Singer E
and Buchthal S (In Review). Predictors of HIV-associated cognitive decline in the era of combined
antiretroviral therapy. Social Neuroscience and Health.

Navia BA, Cho ES, Petito CK and Price RW (1986a). The AIDS dementia complex: II. Neuropathology.
Ann Neurol, 19(6): 525-535.

Navia BA, Jordan BD and Price RW (1986b). The AIDS dementia complex: I. Clinical features. Ann
Neurol, 19(6): 517-524.

Nebuloni M, Pellegrinelli A, Ferri A, Bonetto S, Boldorini R, Vago L, Grassi MP and Costanzi G (2001).
Beta amyloid precursor protein and patterns of HIV p24 immunohistochemistry in different brain areas of
AIDS patients. Aids, 15(5): 571-575.

Neuen-Jacob E, Arendt G, Wendtland B, Jacob B, Schneeweis M and Wechsler W (1993). Frequency and
topographical distribution of CD68-positive macrophages and HIV-1 core proteins in HIV-associated
brain lesions. Clin Neuropathol, 12(6): 315-324.

Nightingale S, Winston A, Letendre S, Michael BD, McArthur JC, Khoo S and Solomon T (2014).
Controversies in HIV-associated neurocognitive disorders. The Lancet. Neurology, 13(11): 1139-1151.

Nir TM, Jahanshad N, Villalon-Reina JE, Toga AW, Jack CR, Weiner MW, Thompson PM and
Alzheimer's Disease Neuroimaging I (2013). Effectiveness of regional DTI measures in distinguishing
Alzheimer's disease, MCI, and normal aging. Neuroimage Clin, 3: 180-195.

Nir TM, Jahanshad N, Busovaca E, Wendelken L, Nicolas K, Thompson PM and Valcour VG (2014).
Mapping white matter integrity in elderly people with HIV. Hum Brain Mapp, 35(3): 975-992.

218

Nir TM, Jahanshad N, Toga AW, Bernstein MA, Jack CR, Jr., Weiner MW and Thompson PM (2015a).
Connectivity network measures predict volumetric atrophy in mild cognitive impairment. Neurobiol
Aging, 36 Suppl 1: S113-120.

Nir TM, Villalon-Reina JE, Prasad G, Jahanshad N, Joshi SH, Toga AW, Bernstein MA, Jack CR, Jr.,
Weiner MW, Thompson PM and Alzheimer's Disease Neuroimaging I (2015b). Diffusion weighted
imaging-based maximum density path analysis and classification of Alzheimer's disease. Neurobiol
Aging, 36 Suppl 1: S132-140.

Nir TM, Zavaliangos-Petropulu A, Jahanshad N, Villalon-Reina JE, Zhan L, Leow AD, Bernstein MA,
Jack CR, Weiner MW and Thompson PM (2016). Diffusion tensor distribution function metrics boost
power to detect defecits in Alzhemier's disease. 2016 IEEE 13th International Symposium on Biomedical
Imaging (ISBI).

Nir TM, Jahanshad N, Villalon-Reina JE, Isaev D, Zavaliangos-Petropulu A, Zhan L, Leow AD, Jack CR,
Jr., Weiner MW and Thompson PM (2017). Fractional anisotropy derived from the diffusion tensor
distribution function boosts power to detect Alzheimer's disease deficits. Magn Reson Med, 78(6): 2322-
2333.

Nurutdinova D, Chrusciel T, Zeringue A, Scherrer JF, Al-Aly Z, McDonald JR and Overton ET (2012).
Mental health disorders and the risk of AIDS-defining illness and death in HIV-infected veterans. AIDS,
26(2): 229-234.

O'Connor EE, Zeffiro TA and Zeffiro TA (2018). Brain structural changes following HIV infection: Meta-
analysis. AJNR Am J Neuroradiol, 39(1): 54-62.

O'Dwyer L, Lamberton F, Bokde AL, Ewers M, Faluyi YO, Tanner C, Mazoyer B, O'Neill D, Bartley M,
Collins DR, Coughlan T, Prvulovic D and Hampel H (2011). Multiple indices of diffusion identifies white
matter damage in mild cognitive impairment and Alzheimer's disease. PLoS One, 6(6): e21745.

O’Connor EE, Jaillard A, Renard F and Zeffiro TA (2017). Reliability of white matter microstructural
changes in HIV infection: Meta-analysis and confirmation. American Journal of Neuroradiology, 38(8):
1510-1519.

Oh SW, Shin NY, Choi JY, Lee SK and Bang MR (2018). Altered White Matter Integrity in Human
Immunodeficiency Virus-Associated Neurocognitive Disorder: A Tract-Based Spatial Statistics Study.
Korean J Radiol, 19(3): 431-442.

Oishi K (2011). MRI atlas of human white matter. Amsterdam, Elsevier/Academic Press.

Oishi K, Mielke MM, Albert M, Lyketsos CG and Mori S (2011). DTI analyses and clinical applications
in Alzheimer's disease. J Alzheimers Dis, 26 Suppl 3: 287-296.

Ortega M, Heaps JM, Joska J, Vaida F, Seedat S, Stein DJ, Paul R and Ances BM (2013). HIV clades B
and C are associated with reduced brain volumetrics. Journal of Neurovirology, 19(5): 479-487.

219

Palacios EM, Martin AJ, Boss MA, Ezekiel F, Chang YS, Yuh EL, Vassar MJ, Schnyer DM, MacDonald
CL, Crawford KL, Irimia A, Toga AW, Mukherjee P and Investigators T-T (2017). Toward precision and
reproducibility of diffusion tensor imaging: A multicenter diffusion phantom and traveling volunteer
study. American Journal of Neuroradiology, 38: 537-545.

Parker TD, Slattery CF, Zhang J, Nicholas JM, Paterson RW, Foulkes AJM, Malone IB, Thomas DL,
Modat M, Cash DM, Crutch SJ, Alexander DC, Ourselin S, Fox NC, Zhang H and Schott JM (2018).
Cortical microstructure in young onset Alzheimer's disease using neurite orientation dispersion and
density imaging. Human Brain Mapping, 39(7): 3005-3017.

Pasternak O, Sochen N, Gur Y, Intrator N and Assaf Y (2009). Free water elimination and mapping from
diffusion MRI. Magn Reson Med, 62(3): 717-730.

Pasternak O, Shenton ME and Westin CF (2012). Estimation of extracellular volume from regularized
multi-shell diffusion MRI. Med Image Comput Comput Assist Interv, 15(Pt 2): 305-312.

Pathai S, Bajillan H, Landay AL and High KP (2014). Is HIV a model of accelerated or accentuated aging?
J Gerontol A Biol Sci Med Sci, 69(7): 833-842.

Paul R, Prasitsuebsai W, Jahanshad N, Puthanakit T, Thompson P, Aurpibul L, Hansudewechakul R,
Kosalaraksa P, Kanjanavanit S, Ngampiyaskul C, Luesomboon W, Lerdlum S, Pothisri M, Visrutaratna
P, Valcour V, Nir TM, Saremi A, Kerr S and Ananworanich J (2018). Structural neuroimaging and
neuropsychologic signatures in children with vertically acquired HIV. Pediatr Infect Dis J, 37(7): 662-
668.

Pedraza O and Mungas D (2008). Measurement in cross-cultural neuropsychology. Neuropsychology
review, 18(3): 184-193.

Peluso R, Haase A, Stowring L, Edwards M and Ventura P (1985). A Trojan Horse mechanism for the
spread of visna virus in monocytes. Virology, 147(1): 231-236.

Perry MJ (1998). Gender, race and economic perspectives on the social epidemiology of HIV infection:
Implications for prevention. Journal of Primary Prevention, 19(2): 97-104.

Pestilli F, Yeatman JD, Rokem A, Kay KN and Wandell BA (2014). Evaluation and statistical inference
for human connectomes. Nat Methods, 11(10): 1058-1063.

Petersen RC, Doody R, Kurz A, Mohs RC, Morris JC, Rabins PV, Ritchie K, Rossor M, Thal L and
Winblad B (2001). Current concepts in mild cognitive impairment. Arch Neurol, 58(12): 1985-1992.

Petito CK, Roberts B, Cantando JD, Rabinstein A and Duncan R (2001). Hippocampal injury and
alterations in neuronal chemokine co-receptor expression in patients with AIDS. J Neuropathol Exp
Neurol, 60(4): 377-385.

220

Pfefferbaum A, Rosenbloom MJ, Sassoon SA, Kemper CA, Deresinski S, Rohlfing T and Sullivan EV
(2012). Regional brain structural dysmorphology in human immunodeficiency virus infection: effects of
acquired immune deficiency syndrome, alcoholism, and age. Biological psychiatry, 72(5): 361-370.

Pfefferbaum A, Rogosa DA, Rosenbloom MJ, Chu W, Sassoon SA, Kemper CA, Deresinski S, Rohlfing
T, Zahr NM and Sullivan EV (2014). Accelerated aging of selective brain structures in human
immunodeficiency virus infection: a controlled, longitudinal magnetic resonance imaging study.
Neurobiol Aging, 35(7): 1755-1768.

Phillips JS, Das SR, McMillan CT, Irwin DJ, Roll EE, Da Re F, Nasrallah IM, Wolk DA and Grossman
M (2018). Tau PET imaging predicts cognition in atypical variants of Alzheimer's disease. Human Brain
Mapping, 39: 691-708.

Pievani M, de Haan W, Wu T, Seeley WW and Frisoni GB (2011). Functional network disruption in the
degenerative dementias. Lancet Neurol, 10(9): 829-843.

Platt L, Easterbrook P, Gower E, McDonald B, Sabin K, McGowan C, Yanny I, Razavi H and Vickerman
P (2016). Prevalence and burden of HCV co-infection in people living with HIV: a global systematic
review and meta-analysis. Lancet Infect Dis, 16(7): 797-808.

Pohl KM, Sullivan EV, Rohlfing T, Chu W, Kwon D, Nichols BN, Zhang Y, Brown SA, Tapert SF,
Cummins K, Thompson WK, Brumback T, Colrain IM, Baker FC, Prouty D, De Bellis MD, Voyvodic
JT, Clark DB, Schirda C, Nagel BJ and Pfefferbaum A (2016). Harmonizing DTI measurements across
scanners to examine the development of white matter microstructure in 803 adolescents of the NCANDA
study. Neuroimage, 130: 194-213.

Pomara N, Crandall DT, Choi SJ, Johnson G and Lim KO (2001). White matter abnormalities in HIV-1
infection: a diffusion tensor imaging study. Psychiatry Res, 106(1): 15-24.

Popuri K, Balachandar R, Alpert K, Lu D, Bhalla M, Mackenzie IR, Hsiung RG-Y, Wang L and Beg MFa
(2018). Development and validation of a novel dementia of Alzheimer's type (DAT) score based on
metabolism FDG-PET imaging. Neuroimage. Clinical, 18: 802-813.

Potvin O, Mouiha A, Dieumegarde L, Duchesne S and Alzheimer׳s Disease Neuroimaging I (2016).
FreeSurfer subcortical normative data. Data in brief, 9: 732-736.

Powell F, Tosun D, Sadeghi R, Weiner M, Raj A and Alzheimer's Disease Neuroimaging I (2018).
Preserved structural network organization mediates pathology spread in Alzheimer's disease spectrum
despite loss of white matter tract integrity. J Alzheimers Dis, 65(3): 747-764.

Prasad G, Burkart J, Joshi SH, Nir TM, Toga AW and Thompson PM (2013a). A dynamical clustering
model of brain connectivity inspired by the N -body problem. Multimodal Brain Image Anal (2013), 8159:
129-137.

Prasad G, Nir TM, Toga AW and Thompson PM (2013b). Tractography density and network measures in
Alzheimer's disease. Proc IEEE Int Symp Biomed Imaging, 2013: 692-695.
221

Price RW and Brew BJ (1988). The AIDS dementia complex. J Infect Dis, 158(5): 1079-1083.

Provencher SW (2001). Automatic quantitation of localized in vivo 1H spectra with LCModel. NMR
Biomed, 14(4): 260-264.

Raber J, Huang Y and Ashford JW (2004). ApoE genotype accounts for the vast majority of AD risk and
AD pathology. Neurobiol Aging, 25(5): 641-650.

Rahimian P and He JJ (2017). HIV/neuroAIDS biomarkers. Prog Neurobiol, 157: 117-132.

Raz N and Rodrigue KM (2006). Differential aging of the brain: patterns, cognitive correlates and
modifiers. Neurosci Biobehav Rev, 30(6): 730-748.

Reijneveld JC, Ponten SC, Berendse HW and Stam CJ (2007). The application of graph theoretical
analysis to complex networks in the brain. Clin Neurophysiol, 118(11): 2317-2331.

Reiman EM and Jagust WJ (2012). Brain imaging in the study of Alzheimer's disease. NeuroImage, 61(2):
505-516.

Rempel HC and Pulliam L (2005). HIV-1 Tat inhibits neprilysin and elevates amyloid beta. Aids, 19(2):
127-135.

Reuter M, Schmansky NJ, Rosas HD and Fischl B (2012). Within-subject template estimation for unbiased
longitudinal image analysis. Neuroimage, 61(4): 1402-1418.

Ribeiro RM, Hazenberg MD, Perelson AS and Davenport MP (2006). Naïve and Memory Cell Turnover
as Drivers of CCR5-to-CXCR4 Tropism Switch in Human Immunodeficiency Virus Type 1: Implications
for Therapy. Journal of Virology, 80(2): 802.

Robertson K, Liner J and Meeker RB (2012). Antiretroviral neurotoxicity. J Neurovirol, 18(5): 388-399.

Robertson KR, Smurzynski M, Parsons TD, Wu K, Bosch RJ, Wu J, McArthur JC, Collier AC, Evans SR
and Ellis RJ (2007). The prevalence and incidence of neurocognitive impairment in the HAART era.
AIDS, 21(14): 1915-1921.

Rodger AJ, Lodwick R, Schechter M, Deeks S, Amin J, Gilson R, Paredes R, Bakowska E, Engsig FN
and Phillips A (2013). Mortality in well controlled HIV in the continuous antiretroviral therapy arms of
the SMART and ESPRIT trials compared with the general population. AIDS, 27(6): 973-979.

Rokem A, Yeatman JD, Pestilli F, Kay KN, Mezer A, van der Walt S and Wandell BA (2015). Evaluating
the accuracy of diffusion MRI models in white matter. PLoS One, 10(4): e0123272.

Rosen WG, Mohs RC and Davis KL (1984). A new rating scale for Alzheimer's disease. American Journal
of Psychiatry, 141(11): 1356-1364.

Rosenthal R and Rosnow RL (1991). Essentials of behavioral research: methods and data analysis. 692.
222

Roth AD, Ramrez G and Alarcn Ra (2005). Oligodendrocytes damage in Alzheimer's disease: Beta
amyloid toxicity and inflammation. Biological Research, 38(4): 381-387.

Rottenberg DA, Moeller JR, Strother SC, Sidtis JJ, Navia BA, Dhawan V, Ginos JZ and Price RW (1987).
The metabolic pathology of the AIDS dementia complex. Ann Neurol, 22(6): 700-706.

Rubinov M and Sporns O (2010). Complex network measures of brain connectivity: uses and
interpretations. Neuroimage, 52(3): 1059-1069.

Sachdev PS, Zhuang L, Braidy N and Wen W (2013). Is Alzheimer's a disease of the white matter? Current
Opinion in Psychiatry, 26(3): 244-251.

Sacktor N, McDermott MP, Marder K, Schifitto G, Selnes OA, McArthur JC, Stern Y, Albert S, Palumbo
D, Kieburtz K, De Marcaida JA, Cohen B and Epstein L (2002). HIV-associated cognitive impairment
before and after the advent of combination therapy. J Neurovirol, 8(2): 136-142.

Sacktor N, Skolasky RL, Seaberg E, Munro C, Becker JT, Martin E, Ragin A, Levine A and Miller E
(2016). Prevalence of HIV-associated neurocognitive disorders in the Multicenter AIDS Cohort Study.
Neurology, 86(4): 334-340.

Salminen LE, Conturo TE, Bolzenius JD, Cabeen RP, Akbudak E and Paul RH (2016). Reducing CSF
partial volume effects to enhance diffusion tensor imaging metrics of brain microstructure. Technology
and innovation, 18(1): 5-20.

Sanchez-Rodriguez LM, Iturria-Medina Y, Baines EA, Mallo SC, Dousty M, Sotero RC and Initiative
oboTAsDN (2018). Design of optimal nonlinear network controllers for Alzheimer's disease. PLOS
Computational Biology, 14: e1006136.

Sanford R, Fellows LK, Ances BM and Collins DL (2018). Association of brain structure changes and
cognitive function with combination antiretroviral therapy in HIV-positive individuals. JAMA Neurol,
75(1): 72-79.

Sanz-Arigita EJ, Schoonheim MM, Damoiseaux JS, Rombouts SA, Maris E, Barkhof F, Scheltens P and
Stam CJ (2010). Loss of 'small-world' networks in Alzheimer's disease: graph analysis of FMRI resting-
state functional connectivity. PLoS One, 5(11): e13788.

Schmaal L, Hibar DP, Samann PG, Hall GB, Baune BT, Jahanshad N, Cheung JW, van Erp TG, Bos D,
Ikram MA, Vernooij MW, Niessen WJ, Tiemeier H, Hofman A, Wittfeld K, Grabe HJ, Janowitz D, Bulow
R, Selonke M, Volzke H, et al. (2016). Cortical abnormalities in adults and adolescents with major
depression based on brain scans from 20 cohorts worldwide in the ENIGMA Major Depressive Disorder
Working Group. Mol Psychiatry.

Schouten J, Cinque P, Gisslen M, Reiss P and Portegies P (2011). HIV-1 infection and cognitive
impairment in the cART era: a review. AIDS, 25(5): 561-575.

223

Schwartz ED, Cooper ET, Fan Y, Jawad AF, Chin CL, Nissanov J and Hackney DB (2005). MRI diffusion
coefficients in spinal cord correlate with axon morphometry. Neuroreport, 16(1): 73-76.

Scott JC, Woods SP, Carey CL, Weber E, Bondi MW and Grant I (2011). Neurocognitive consequences
of HIV infection in older adults: an evaluation of the "cortical" hypothesis. AIDS Behav, 15(6): 1187-
1196.

Seider TR, Gongvatana A, Woods AJ, Chen H, Porges EC, Cummings T, Psy.D, Correia S, Tashima K
and Cohen RA (2016). Age exacerbates HIV-associated white matter abnormalities. Journal of
Neurovirology, 22(2): 201-212.

Shaw LM, Korecka M, Clark CM, Lee VMY and Trojanowski JQ (2007). Biomarkers of
neurodegeneration for diagnosis and monitoring therapeutics. Nature Reviews Drug Discovery, 6(4): 295-
303.

Shimizu S, Hirose D, Hatanaka H, Takenoshita N, Kaneko Y, Ogawa Y, Sakurai H and Hanyu H (2018).
Role of neuroimaging as a biomarker for neurodegenerative diseases. Frontiers in Neurology, 9: 265-265.

Simioni S, Cavassini M, Annoni JM, Rimbault Abraham A, Bourquin I, Schiffer V, Calmy A, Chave JP,
Giacobini E, Hirschel B and Du Pasquier RA (2010). Cognitive dysfunction in HIV patients despite long-
standing suppression of viremia. AIDS, 24(9): 1243-1250.

Slattery CF, Zhang J, Paterson RW, Foulkes AJM, Carton A, Macpherson K, Mancini L, Thomas DL,
Modat M, Toussaint N, Cash DM, Thornton JS, Henley SMD, Crutch SJ, Alexander DC, Ourselin S, Fox
NC, Zhang H and Schott JM (2017). ApoE influences regional white-matter axonal density loss in
Alzheimer's disease. Neurobiology of Aging, 57: 8-17.

Sled JG, Zijdenbos AP and Evans AC (1998). A nonparametric method for automatic correction of
intensity nonuniformity in MRI data. IEEE Trans Med Imaging, 17(1): 87-97.

Smith SM (2002). Fast robust automated brain extraction. Hum Brain Mapp, 17(3): 143-155.

Smith SM, Zhang Y, Jenkinson M, Chen J, Matthews PM, Federico A and De Stefano N (2002). Accurate,
robust, and automated longitudinal and cross-sectional brain change analysis. Neuroimage, 17(1): 479-
489.

Smith SM, Jenkinson M, Johansen-Berg H, Rueckert D, Nichols TE, Mackay CE, Watkins KE, Ciccarelli
O, Cader MZ, Matthews PM and Behrens TE (2006). Tract-based spatial statistics: voxelwise analysis of
multi-subject diffusion data. Neuroimage, 31(4): 1487-1505.

Söderman O and Jönsson B (1995). Restricted diffusion in cylindrical geometry. Journal of Magnetic
Resonance, 117(1): 94-97.

Song S-K, Yoshino J, Le TQ, Lin S-J, Sun S-W, Cross AH and Armstrong RC (2005). Demyelination
increases radial diffusivity in corpus callosum of mouse brain. Neuroimage, 26: 132-140.

224

Song SK, Sun SW, Ju WK, Lin SJ, Cross AH and Neufeld AH (2003). Diffusion tensor imaging detects
and differentiates axon and myelin degeneration in mouse optic nerve after retinal ischemia. Neuroimage,
20(3): 1714-1722.

Soontornniyomkij V, Moore DJ, Gouaux B, Soontornniyomkij B, Tatro ET, Umlauf A, Masliah E, Levine
AJ, Singer EJ, Vinters HV, Gelman BB, Morgello S, Cherner M, Grant I and Achim CL (2012). Cerebral
beta-amyloid deposition predicts HIV-associated neurocognitive disorders in APOE epsilon4 carriers.
AIDS, 26(18): 2327-2335.

Spector SA, Singh KK, Gupta S, Cystique LA, Jin H, Letendre S, Schrier R, Wu Z, Hong KX, Yu X, Shi
C and Heaton RK (2010). APOE epsilon4 and MBL-2 O/O genotypes are associated with neurocognitive
impairment in HIV-infected plasma donors. AIDS, 24(10): 1471-1479.

Spudich S and Gonzalez-Scarano F (2012). HIV-1-related central nervous system disease: current issues
in pathogenesis, diagnosis, and treatment. Cold Spring Harb Perspect Med, 2(6): a007120.

Stam CJ, Jones BF, Nolte G, Breakspear M and Scheltens P (2007). Small-world networks and functional
connectivity in Alzheimer's disease. Cereb Cortex, 17(1): 92-99.

Stanley LC, Mrak RE, Woody RC, Perrot LJ, Zhang S, Marshak DR, Nelson SJ and Griffin WS (1994).
Glial cytokines as neuropathogenic factors in HIV infection: pathogenic similarities to Alzheimer's
disease. J Neuropathol Exp Neurol, 53(3): 231-238.

Stejskal EO and Tanner JE (1965). Spin Diffusion Measurements: Spin Echoes in the Presence of a Time‐
Dependent Field Gradient. The Journal of Chemical Physics, 42(1): 288-292.

Sulaimany S, Khansari M, Zarrineh P, Daianu M, Jahanshad N, Thompson PM and Masoudi-Nejad A
(2017). Predicting brain network changes in Alzheimer's disease with link prediction algorithms. Mol
Biosyst, 13(4): 725-735.

Tardieu M, Le Chenadec J, Persoz A, Meyer L, Blanche S and Mayaux MJ (2000). HIV-1-related
encephalopathy in infants compared with children and adults. French Pediatric HIV Infection Study and
the SEROCO Group. Neurology, 54(5): 1089-1095.

Tate DF, Conley JJ, Meier DS, Navia BA, Cohen R and Guttmann CRG (2009). Neuroimaging Among
HIV-Infected Patients: Current Knowledge and Future Directions. HIV and the Brain: New Challenges in
the Modern Era. H. R. Paul, C. N. Sacktor, V. Valcour and T. K. Tashima. Totowa, NJ, Humana Press:
75-107.

Tate DF, Paul RH, Kertesz K, Conley J and Russell T (2010). HIV–AIDS: The Neurologic and Cognitive
Consequences of HIV-1 Infection. Handbook of Medical Neuropsychology: Applications of Cognitive
Neuroscience. C. L. Armstrong and L. Morrow. New York, NY, Springer New York: 373-396.

Tate DF, Sampat M, Harezlak J, Fiecas M, Hogan J, Dewey J, McCaffrey D, Branson D, Russell T, Conley
J, Taylor M, Schifitto G, Zhong J, Daar ES, Alger J, Brown M, Singer E, Campbell T, McMahon D, Tso
225

Y, et al. (2011). Regional areas and widths of the midsagittal corpus callosum among HIV-infected
patients on stable antiretroviral therapies. J Neurovirol, 17(4): 368-379.

Tax C, Grussu F, Kaden E, Ning L, Rudrapatna U, Evans J, St-Jean S, Leemans A, Puch S, Rowe M,
Rodrigues P, Prčkovska V, Koppers S, Merhof D, Ghosh A, Tanno R, Alexander D, Charron C, Kusmia
S and Veraart J (2018). Cross-vendor and Cross-protocol harmonisation of diffusion MRI data: a
comparative study.

Thompson PM, Dutton RA, Hayashi KM, Toga AW, Lopez OL, Aizenstein HJ and Becker JT (2005).
Thinning of the cerebral cortex visualized in HIV/AIDS reflects CD4+ T lymphocyte decline. Proc Natl
Acad Sci U S A, 102(43): 15647-15652.

Thompson PM, Stein JL, Medland SE, Hibar DP, Vasquez AA, Renteria ME, Toro R, Jahanshad N,
Schumann G, Franke B, Wright MJ, Martin NG, Agartz I, Alda M, Alhusaini S, Almasy L, Almeida J,
Alpert K, Andreasen NC, Andreassen OA, et al. (2014). The ENIGMA Consortium: large-scale
collaborative analyses of neuroimaging and genetic data. Brain Imaging Behav, 8(2): 153-182.

Thompson PM and Jahanshad N (2015). Novel neuroimaging methods to understand how HIV affects the
brain. Current HIV/AIDS reports, 12(2): 289-298.

Toga AW and Thompson PM (2013). Connectomics sheds new light on Alzheimer's disease. Biol
Psychiatry, 73(5): 390-392.

Tournier JD, Calamante F, Gadian DG and Connelly A (2004). Direct estimation of the fiber orientation
density function from diffusion-weighted MRI data using spherical deconvolution. Neuroimage, 23(3):
1176-1185.

Tozzi V, Balestra P, Bellagamba R, Corpolongo A, Salvatori MF, Visco-Comandini U, Vlassi C,
Giulianelli M, Galgani S, Antinori A and Narciso P (2007). Persistence of neuropsychologic deficits
despite long-term highly active antiretroviral therapy in patients with HIV-related neurocognitive
impairment: prevalence and risk factors. J Acquir Immune Defic Syndr, 45(2): 174-182.

Tuch DS, Reese TG, Wiegell MR, Makris N, Belliveau JW and Wedeen VJ (2002). High angular
resolution diffusion imaging reveals intravoxel white matter fiber heterogeneity. Magn Reson Med, 48(4):
577-582.

Tuch DS (2004). Q-ball imaging. Magn Reson Med, 52(6): 1358-1372.

Tucker-Drob EM, Johnson KE and Jones RN (2009). The cognitive reserve hypothesis: a longitudinal
examination of age-associated declines in reasoning and processing speed. Developmental Psychology,
45(2): 431-446.

Tucker KA, Robertson KR, Lin W, Smith JK, An H, Chen Y, Aylward SR and Hall CD (2004).
Neuroimaging in human immunodeficiency virus infection. J Neuroimmunol, 157(1-2): 153-162.

226

Turchan-Cholewo J, Liu Y, Gartner S, Reid R, Jie C, Peng X, Chen KC, Chauhan A, Haughey N, Cutler
R, Mattson MP, Pardo C, Conant K, Sacktor N, McArthur JC, Hauser KF, Gairola C and Nath A (2006).
Increased vulnerability of ApoE4 neurons to HIV proteins and opiates: protection by diosgenin and L-
deprenyl. Neurobiol Dis, 23(1): 109-119.

Tustison NJ, Avants BB, Cook PA, Zheng Y, Egan A, Yushkevich PA and Gee JC (2010). N4ITK:
improved N3 bias correction. IEEE Trans Med Imaging, 29(6): 1310-1320.

UNAIDS (2018) Global HIV & AIDS statistics — 2018 fact sheet. 2018, from
http://www.unaids.org/en/resources/fact-sheet.

Valcour V, Shikuma C, Shiramizu B, Watters M, Poff P, Selnes O, Holck P, Grove J and Sacktor N
(2004a). Higher frequency of dementia in older HIV-1 individuals: the Hawaii Aging with HIV-1 Cohort.
Neurology, 63(5): 822-827.

Valcour V, Shikuma C, Shiramizu B, Watters M, Poff P, Selnes OA, Grove J, Liu Y, Abdul-Majid KB,
Gartner S and Sacktor N (2004b). Age, apolipoprotein E4, and the risk of HIV dementia: the Hawaii Aging
with HIV Cohort. J Neuroimmunol, 157(1-2): 197-202.

Valcour V, Paul R, Neuhaus J and Shikuma C (2011). The effects of age and HIV on neuropsychological
performance. Journal of the International Neuropsychological Society : JINS, 17(1): 190-195.

Valcour VG, Shikuma CM, Shiramizu BT, Williams AE, Watters MR, Poff PW, Grove JS, Selnes OA
and Sacktor NC (2005). Diabetes, insulin resistance, and dementia among HIV-1-infected patients. J
Acquir Immune Defic Syndr, 38(1): 31-36.

van den Heuvel MP, Kahn RS, Goni J and Sporns O (2012). High-cost, high-capacity backbone for global
brain communication. Proc Natl Acad Sci U S A, 109(28): 11372-11377.

van der Meer P, Ulrich AM, Gonźalez-Scarano F and Lavi E (2000). Immunohistochemical Analysis of
CCR2, CCR3, CCR5, and CXCR4 in the Human Brain: Potential Mechanisms for HIV Dementia.
Experimental and Molecular Pathology, 69(3): 192-201.

van Erp TG, Hibar DP, Rasmussen JM, Glahn DC, Pearlson GD, Andreassen OA, Agartz I, Westlye LT,
Haukvik UK, Dale AM, Melle I, Hartberg CB, Gruber O, Kraemer B, Zilles D, Donohoe G, Kelly S,
McDonald C, Morris DW, Cannon DM, et al. (2016). Subcortical brain volume abnormalities in 2028
individuals with schizophrenia and 2540 healthy controls via the ENIGMA consortium. Mol Psychiatry,
21(4): 585.

Veitch DP, Weiner MW, Aisen PS, Beckett LA, Cairns NJ, Green RC, Harvey D, Jack CR, Jr., Jagust W,
Morris JC, Petersen RC, Saykin AJ, Shaw LM, Toga AW, Trojanowski JQ and Alzheimer's Disease
Neuroimaging I (2018). Understanding disease progression and improving Alzheimer's disease clinical
trials: Recent highlights from the Alzheimer's Disease Neuroimaging Initiative. Alzheimers Dement.

227

von Giesen HJ, Antke C, Hefter H, Wenserski F, Seitz RJ and Arendt G (2000). Potential time course of
human immunodeficiency virus type 1-associated minor motor deficits: electrophysiologic and positron
emission tomography findings. Arch Neurol, 57(11): 1601-1607.

Walker SP, Wachs TD, Grantham-McGregor S, Black MM, Nelson CA, Huffman SL, Baker-Henningham
H, Chang SM, Hamadani JD, Lozoff B, Gardner JM, Powell CA, Rahman A and Richter L (2011).
Inequality in early childhood: risk and protective factors for early child development. Lancet, 378(9799):
1325-1338.

Wang B, Niu Y, Miao L, Cao R, Yan P, Guo H, Li D, Guo Y, Yan T, Wu J, Xiang J and Zhang H (2017).
Decreased complexity in Alzheimer's disease: Resting-state fMRI evidence of brain entropy mapping.
Frontiers in Aging Neuroscience, 9: 378.

Wang M-L, Wei X-E, Fu J-L, Li W, Yu M-M, Li P-Y and Li W-B (2018a). Subcortical nuclei in
Alzheimer’s disease: a volumetric and diffusion kurtosis imaging study. Acta Radiologica, 59: 1365-1371.

Wang X, Zhen X, Li Q, Shen D and Huang H (2018b). Cognitive assessment prediction in Alzheimer's
disease by multi-layer multi-target regression. Neuroinformatics: 1-10.

Wang Y, Zhang J, Gutman B, Chan TF, Becker JT, Aizenstein HJ, Lopez OL, Tamburo RJ, Toga AW
and Thompson PM (2010). Multivariate tensor-based morphometry on surfaces: application to mapping
ventricular abnormalities in HIV/AIDS. Neuroimage, 49(3): 2141-2157.

Wang Z, Trillo-Pazos G, Kim SY, Canki M, Morgello S, Sharer LR, Gelbard HA, Su ZZ, Kang DC,
Brooks AI, Fisher PB and Volsky DJ (2004). Effects of human immunodeficiency virus type 1 on astrocyte
gene expression and function: potential role in neuropathogenesis. J Neurovirol, 10 Suppl 1: 25-32.

Watson C, Busovaca E, Foley JM, Allen IE, Schwarz CG, Jahanshad N, Nir TM, Esmaeili-Firidouni P,
Milanini B, Rosen H, Carmichael OT, Thompson PM and Valcour VG (2017). White matter
hyperintensities correlate to cognition and fiber tract integrity in older adults with HIV. J Neurovirol,
23(3): 422-429.

Watts DJ and Strogatz SH (1998). Collective dynamics of 'small-world' networks. Nature, 393(6684):
440-442.

Wedeen VJ, Hagmann P, Tseng WY, Reese TG and Weisskoff RM (2005). Mapping complex tissue
architecture with diffusion spectrum magnetic resonance imaging. Magn Reson Med, 54(6): 1377-1386.

Weissenborn K, Tryc AB, Heeren M, Worthmann H, Pflugrad H, Berding G, Bokemeyer M, Tillmann
HL and Goldbecker A (2009). Hepatitis C virus infection and the brain. Metab Brain Dis, 24(1): 197-210.

Wendelken LA and Valcour V (2012). Impact of HIV and aging on neuropsychological function. J
Neurovirol, 18(4): 256-263.

Wendelken LA, Jahanshad N, Rosen HJ, Busovaca E, Allen I, Coppola G, Adams C, Rankin KP, Milanini
B, Clifford K, Wojta K, Nir TM, Gutman BA, Thompson PM and Valcour V (2016). ApoE epsilon4 is
228

associated with cognition, brain integrity, and atrophy in HIV over age 60. J Acquir Immune Defic Syndr,
73(4): 426-432.

Weston PSJ, Simpson IJA, Ryan NS, Ourselin S and Fox NC (2015). Diffusion imaging changes in grey
matter in Alzheimer’s disease: a potential marker of early neurodegeneration. Alzheimer's Research &
Therapy, 7(1): 47.

Whitwell JL, Avula R, Senjem ML, Kantarci K, Weigand SD, Samikoglu A, Edmonson HA, Vemuri P,
Knopman DS, Boeve BF, Petersen RC, Josephs KA and Jack CR, Jr. (2010). Gray and white matter water
diffusion in the syndromic variants of frontotemporal dementia. Neurology, 74(16): 1279-1287.

World Health Organization (2003). Diet, nutrition and the prevention of chronic diseases. World Health
Organ Tech Rep Ser, 916: i-viii, 1-149, backcover.

Wiley CA, Soontornniyomkij V, Radhakrishnan L, Masliah E, Mellors J, Hermann SA, Dailey P and
Achim CL (1998). Distribution of brain HIV load in AIDS. Brain Pathol, 8(2): 277-284.

Wing EJ (2016). HIV and aging. International Journal of Infectious Diseases, 53: 61-68.

Woods SP, Moore DJ, Weber E and Grant I (2009). Cognitive neuropsychology of HIV-associated
neurocognitive disorders. Neuropsychol Rev, 19(2): 152-168.

World Health Organization (2015) Guideline on when to start ART and on PrEP for HIV. from
http://apps.who.int/iris/bitstream/10665/186275/1/9789241509565_eng.pdf?ua=1.

Wright PW, Heaps JM, Shimony JS, Thomas JB and Ances BM (2012). The effects of HIV and
combination antiretroviral therapy on white matter integrity. AIDS, 26(12): 1501-1508.

Wu EX and Cheung MM (2010). MR diffusion kurtosis imaging for neural tissue characterization. NMR
Biomed, 23(7): 836-848.

Wu WE, Babb JS, Tal A, Kirov, II, George AE, Ratai EM, Gonzalez RG and Gonen O (2015). Early glial
activation precedes neurodegeneration in the cerebral cortex after SIV infection: a 3D, multivoxel proton
magnetic resonance spectroscopy study. HIV Med, 16(6): 381-387.

Xie S, Xiao JX, Gong GL, Zang YF, Wang YH, Wu HK and Jiang XX (2006). Voxel-based detection of
white matter abnormalities in mild Alzheimer disease. Neurology, 66(12): 1845-1849.

Yanovsky I, Thompson PM, Osher S and Leow AD (2007). Topology preserving log-unbiased nonlinear
image registration: theory and implementation. 2007 IEEE Conference on Computer Vision and Pattern
Recognition.

Yao Z, Zhang Y, Lin L, Zhou Y, Xu C and Jiang T (2010). Abnormal cortical networks in mild cognitive
impairment and Alzheimer's disease. PLoS Comput Biol, 6(11): e1001006.

229

Yeh F-C and Verstynen TD (2016). Converting multi-shell and diffusion spectrum imaging to high
angular resolution diffusion imaging. Frontiers in Neuroscience, 10: 418.

Yeh FC, Wedeen VJ and Tseng WY (2010). Generalized q-sampling imaging. IEEE Trans Med Imaging,
29(9): 1626-1635.

Yuan L, Sun M, Chen Y, Long M, Zhao X, Yin J, Yan X, Ji D and Ni H (2016). Non-Gaussian diffusion
alterations on diffusion kurtosis imaging in patients with early Alzheimer's disease. Neurosci Lett, 616:
11-18.

Zalesky A, Fornito A, Harding IH, Cocchi L, Yucel M, Pantelis C and Bullmore ET (2010). Whole-brain
anatomical networks: does the choice of nodes matter? Neuroimage, 50(3): 970-983.

Zhan L, Leow AD, Zhu S, Baryshev M, Toga AW, McMahon KL, de Zubicaray GI, Wright MJ and
Thompson PM (2009). A novel measure of fractional anisotropy based on the tensor distribution function.
Med Image Comput Comput Assist Interv, 12(Pt 1): 845-852.

Zhan L, Leow AD, Jahanshad N, Chiang MC, Barysheva M, Lee AD, Toga AW, McMahon KL, de
Zubicaray GI, Wright MJ and Thompson PM (2010). How does angular resolution affect diffusion
imaging measures? Neuroimage, 49(2): 1357-1371.

Zhan L, Jahanshad N, Ennis DB, Jin Y, Bernstein MA, Borowski BJ, Jack CR, Jr., Toga AW, Leow AD
and Thompson PM (2013a). Angular versus spatial resolution trade-offs for diffusion imaging under time
constraints. Hum Brain Mapp, 34(10): 2688-2706.

Zhan L, Mueller BA, Jahanshad N, Jin Y, Lenglet C, Yacoub E, Sapiro G, Ugurbil K, Harel N, Toga AW,
Lim KO and Thompson PM (2013b). Magnetic resonance field strength effects on diffusion measures and
brain connectivity networks. Brain connectivity, 3(1): 72-86.

Zhang H, Schneider T, Wheeler-Kingshott CA and Alexander DC (2012). NODDI: practical in vivo
neurite orientation dispersion and density imaging of the human brain. Neuroimage, 61(4): 1000-1016.

Zhou T, Thung K-H, Zhu X and Shen D (2017). Feature learning and fusion of multimodality
neuroimaging and genetic data for multi-status dementia diagnosis. Machine learning in medical imaging.
MLMI (Workshop), 10541: 132-140.

Zhu A, Moyer D, Nir T, Thompson P and Jahanshad N (2018). Challenges and opportunities in diffusion
MRI data harmonization. Computational Diffusion MRI and Brain Connectivity.

Zhu T, Liu X, Gaugh MD, Connelly PR, Ni H, Ekholm S, Schifitto G and Zhong J (2009). Evaluation of
measurement uncertainties in human diffusion tensor imaging (DTI)-derived parameters and optimization
of clinical DTI protocols with a wild bootstrap analysis. Journal of Magnetic Resonance Imaging, 29(2):
422-435.

230

Zhu T, Hu R, Qiu X, Taylor M, Tso Y, Yiannoutsos C, Navia B, Mori S, Ekholm S, Schifitto G and Zhong
J (2011). Quantification of accuracy and precision of multi-center DTI measurements: a diffusion phantom
and human brain study. Neuroimage, 56(3): 1398-1411.

Zhu T, Zhong J, Hu R, Tivarus M, Ekholm S, Harezlak J, Ombao H, Navia B, Cohen R and Schifitto G
(2013). Patterns of white matter injury in HIV infection after partial immune reconstitution: a DTI tract-
based spatial statistics study. Journal of Neurovirology, 19(1): 10-23.

Abstract (if available)

Abstract In the era of globally accessible combination antiretroviral therapy, morbidity and mortality rates have dramatically decreased in individuals who have contracted the human immunodeficiency virus (HIV). However, HIV-related comorbidities, including symptoms of brain dysfunction, remain common among individuals on suppressive treatment. In addition to neurocognitive testing, non-invasive, in vivo neuroimaging techniques will play an important role in better understanding the spectrum of central nervous system impairments in HIV+ individuals. The virus crosses the blood brain barrier soon after infection, potentially triggering brain aberrations before individuals show any signs or symptoms of disease. As in Alzheimer’s disease, where MRI markers have been shown to be abnormally altered prior to the clinical presentation of the disease, there is a need to identify reliable imaging biomarkers of HIV to understand, measure, and predict disease neuropathology and progression. Many factors—comorbidities, treatment history, aging, socioeconomic factors, etc.—may modulate how HIV impacts the brain. By using a combination of anatomical and diffusion MRI techniques, across a large collection of diverse HIV cohorts, it may become possible to map common brain deficits in the HIV+ population, while disentangling factors that may modulate these effects. Our understanding of these factors are essential to establishing better treatment guidelines and ultimately improving quality of life for infected individuals.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Neuroimaging markers of risk & resilience to brain aging and dementia

PDF

Using neuroinformatics to identify genomic and proteomic markers of suboptimal aging and Alzheimer's disease

PDF

Unveiling the white matter microstructure in 22q11.2 deletion syndrome with diffusion magnetic resonance imaging

PDF

Representation problems in brain imaging

PDF

Reproducibility and management of big data in brain MRI studies

PDF

Vascular contributions to brain aging along the Alzheimer's disease continuum

PDF

Neuroimaging in complex polygenic disorders

PDF

Independent and interactive effects of depression genetic risk and household socioeconomic status on emotional behavior and brain development

PDF

Relationships between lifetime chronic stress exposure, vascular risk, cognition, and brain structure in HIV

PDF

Morphological and microstructural models of typical development in the hippocampus and white matter

PDF

Decoding the neurological and genetic underpinnings of chronic pain

PDF

Anemia and white matter: a diffusion MRI analysis of the human brain

PDF

Data modeling approaches for continuous neuroimaging genetics

PDF

Blood-brain barrier pathophysiology in cognitive impairment and injury

PDF

Estimation of cognitive brain activity in sickle cell disease using functional near-infrared spectroscopy and dynamic systems modeling

PDF

Discovery of mature microRNA sequences within the protein- coding regions of global HIV-1 genomes: Predictions of novel mechanisms for viral infection and pathogenicity

PDF

Brain-based prediction of chronic pain progression: a longitudinal study of urologic chronic pelvic pain syndrome using baseline resting state connectivity from the periaqueductal gray

PDF

A virtual reality exergaming system to enhance brain health in older adults at risk for Alzheimer’s disease

PDF

Mapping water exchange rate across the blood-brain barrier

PDF

Heart, brain, and breath: studies on the neuromodulation of interoceptive systems

Asset Metadata

Creator Nir, Talia Miriam (author)

Core Title Novel multi-site brain imaging approaches to map HIV-related neuropathology

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Neuroscience

Publication Date 05/10/2019

Defense Date 03/15/2019

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

Tag brain volume,diffusion MRI,ENIGMA,HIV,neuroimaging,OAI-PMH Harvest,TBM,white matter

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Jahanshad, Neda (committee chair), Thompson, Paul M. (committee member), Ver Steeg, Greg (committee member)

Creator Email Talia.M.Nir@gmail.com,TNir@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c89-155401

Unique identifier UC11660193

Identifier etd-NirTaliaMi-7416.pdf (filename),usctheses-c89-155401 (legacy record id)

Legacy Identifier etd-NirTaliaMi-7416.pdf

Dmrecord 155401

Document Type Dissertation

Format application/pdf (imt)

Rights Nir, Talia Miriam

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