Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
Using neuroinformatics to identify genomic and proteomic markers of suboptimal aging and Alzheimer's disease
(USC Thesis Other)
Using neuroinformatics to identify genomic and proteomic markers of suboptimal aging and Alzheimer's disease
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
University of Southern California
SUBMITTED TO THE FACULTY OF THE
USC GRADUATE SCHOOL
IN PARTIAL FULFILLMENT OF THE REQUIREMENTS
for the degree
DOCTOR OF PHILOSOPHY (NEUROSCIENCE)
Dissertation Committee:
Paul M. Thompson, PhD
Helena Chui, MD
Sook-Lei Liew, PhD
Neda Jahanshad, PhD
December 2018
USING NEUROINFORMATICS TO IDENTIFY GENOMIC AND
PROTEOMIC MARKERS OF SUBOPTIMAL AGING AND
ALZHEIMER'S DISEASE
By
Brandalyn C. Riedel
University of Southern California
ii
EPIGRAPH
“I AM NOT ONE WHO BELIEVES THAT IT IS ANY NECESSARY VIRTUE IN THE
PHILOSOPHER TO SPEND HIS LIFE DEFENDING A CONSISTENT POSITION. IT IS
SURELY A KIND OF SPIRITUAL PRIDE TO REFRAIN FROM ‘THINKING OUT LOUD,’
AND TO BE UNWILLING TO LET A THESIS APPEAR IN PRINT UNTIL YOU ARE
PREPARED TO CHAMPION IT TO THE DEATH. PHILOSOPHY, LIKE SCIENCE, IS A
SOCIAL FUNCTION, FOR A MAN CANNOT THINK RIGHTLY ALONE, AND THE
PHILOSOPHER MUST PUBLISH HIS THOUGHTS AS MUCH TO LEARN FROM
CRITICISM AS TO CONTRIBUTE TO THE SUM OF WISDOM. IF, THEN, I SOMETIMES
MAKE STATEMENTS IN AN AUTHORITATIVE AND DOGMATIC MANNER, IT IS FOR
THE SAKE OF CLARITY RATHER THAN FROM THE DESIRE TO POSE AS AN
ORACLE.”
ALAN WATTS
University of Southern California
iii
DEDICATION
TO MY DEAR HUSBAND AND BEST FRIEND, ANDREW. SHARING OUR LOVE AND LIFE ALONG
THIS CRAZY JOURNEY OF PURSUING MY DOCTORATE HAS BEEN A BLESSING BEYOND
WORDS.
TO ALL THE PEOPLE I HAVE MET ALONG THE WAY WHO ARE SUFFERING FROM OR CARING
FOR A LOVED ONE WITH ALZHEIMER’S
University of Southern California
iv
ACKNOWLEDGEMENTS
We are trained as scientists to first be critical and skeptical, and then pierce through the barriers of
our own limitations in order to ask seemingly impossible questions. In my case, these questions have largely
revolved around understanding the genetic and proteomic interplay between optimal aging and Alzheimer’s
disease. There are many times that I would have thrown up my hands in surrender during this process had it
not been for the support of those around me. Thus, this work would be incomplete without the direct
acknowledgement of these people. The past five and a half years at the University of Southern California have
been a fulfilling, yet challenging, intellectual adventure. I am extremely grateful for the opportunity to study
here and become part of the Trojan family.
I am thankful for the great interdisciplinary environment I was immersed in through working in two
different labs during my Ph.D. This experience afforded me exposure to work with and learn from leading
researchers on the molecular etiology of Alzheimer’s and the process of implementing translational research
through Roberta Brinton’s lab, and neuroimaging, machine learning, and advanced statistics through Paul
Thompson’s lab. This experience, as well as the world-wide partnerships through ENIGMA demonstrated the
importance of collaborative efforts; I am grateful to my collaborators, both local and global, for all their
intellectual support along the way. From designing plans of attack to combat tough reviewer questions on
papers, to sharing my excitement for the results of various experiments and preparing me for presenting
work to a wider scientific audience at conferences, I will forever be grateful for the mentorship and guidance,
and opportunities for growth and independence that I have received from both mentors. Likewise, I would
like to thank the team of accomplished and driven scientists of my qualifying exam and dissertation
committees. Their valuable feedback and support helped to lay the groundwork and improve this thesis.
I want to acknowledge some of the unsung heroes that make this work possible: the people who
develop open access databases such as the Alzheimer’s Disease Neuroimaging Initiative (ADNI), which much
of my research uses, and open source software that I have used on a daily basis, especially R and the
corresponding libraries. In particular I wish to thank those who tirelessly find ways to improve
computational speeds of existing mathematical functions, such as for hierarchical clustering or Cholesky
University of Southern California
v
decomposition, without which much of my work would be greatly delayed. Moreover, I am thankful for the
countless individuals who share their knowledge and insight over websites like “Stack Overflow”, “Stack
Exchange”, and “Bioconductor”. Finally, much of this work would not be possible without the LONI grid
computing resources.
On a personal note, I wish to thank the friends who went through the process of graduate school with
me, including both my lab-mates and classmates. Our activities together, from intense discussions of science,
to studying together in coffee shops, and climbing literal and figurative mountains, have been memorable and
kept me going. I am also thankful to my long-distance friends who did not forget me while I lived the life of an
academic hermit. Although moving to California may have taken me away from them these past five and a half
years, I know that they are proud that I went after my dreams to pursue these crazy goals. Thank you to my
parents for their encouragement and indirectly engineering me to study science. Thank you to my dog Parker
for reminding me periodically during this writing process that it was time to take a break and go for a walk!
Finally, I am forever grateful to my loving husband, Andrew. When science seemed to spoil the taste of life,
you were always there to help me find a new sense of perspective and encouragement.
Co-author Acknowledgements
I would like to thank all of my co-authors and collaborators. Some of these projects listed below
are not included within this thesis for the sake of ensuring a consistent theme, but I wish to thank all of
the following co-authors for their invaluable contributions to works completed during my doctoral
research.
Riedel, B.C. & Thompson, P. M. (2018). A Preliminary Investigation on the Relationship between
Predicted Brain Age and Genetics in Alzheimer's disease, Alzheimer’s Association International
Conference, Chicago, IL. Oral Presentation.
University of Southern California
vi
Riedel, B. C., Daianu M., Ver Steeg, G., Mezher, A., Salminen, L. E., Galstyan A. & Thompson, P. M.
Uncovering biologically coherent peripheral signatures of health and risk for Alzheimer’s disease
in the aging brain. To be submitted to Frontiers in Aging Neuroscience, 2018.
Riedel, B. C.*, Zhu, D*, Jahanshad, N, Harrison M. B., … & Thompson P. M. MRI based classification of
Major Depressive Disorder in 16 Cohorts Worldwide: An ENIGMA machine learning study. Under
review at Molecular Psychiatry, 2018.
Riedel, B. C., Jahanshad, N., & Thompson, P. M. (2017, April). Graph theoretical approaches towards
understanding differences in frontoparietal and default mode networks in Autism. In Biomedical
Imaging (ISBI 2017), 2017 IEEE 14th International Symposium on Biomedical Imaging (pp. 460-
463). IEEE. Oral Presentation and Paper.
Harrison, M. B. *, Riedel, B. C.*, Prasad, G., Jahanshad, N., Faskowitz, J., & Thompson, P. M. (2017,
January). Utilizing brain measures for large-scale classification of autism applying EPIC. In 12th
International Symposium on Medical Information Processing and Analysis (Vol. 10160, p.
101600W). International Society for Optics and Photonics.
Riedel, B. C., Thompson, P. M., & Brinton, R. D. (2016). Age, APOE and sex: triad of risk of Alzheimer’s
disease. The Journal of Steroid Biochemistry and Molecular Biology, 160, 134-147.
* Equal contribution
University of Southern California
vii
ABBREVIATIONS
Aβ Amyloid β
ABO ABO Blood Type
AD Alzheimer’s disease
ADAS-cog Alzheimer’s Disease Assessment Scale - Cognitive Subscale
AIC Akaike Information Criterion
ApoE Apolipoprotein E (protein)
APOE Apolipoprotein E (gene)
BBB Blood brain barrier
CDR Clinical Dementia Scale
CERAD Neuropathology Task Force of the Consortium to Establish a Registry for
Alzheimer’s Disease
CorEx Correlation Explanation
CNS Central nervous system
CSF Cerebrospinal fluid
DNA Deoxyribonucleic acid
DSM Diagnostic and Statistical Manual of Mental Disorders
EM Expectation maximization
EOAD Early-onset Alzheimer’s disease
FAD Familial Alzheimer’s disease
GBM Gradient Boosting Machine
GWAS Genome-wide association study
HWE Hardy–Weinberg equilibrium
H2 Heritability
ICV Intracranial volume
LD Linkage disequilibrium
LMM Linear mixed model
LRT Likelihood ratio test
LOAD Late-onset Alzheimer’s disease
MAF Minor allele frequency
MCI Mild Cognitive Impairment
MDD Major depressive disorder
MDS Multidimensional Scaling
MMSE Mini-Mental State Examination
NFT Neurofibrillary tangle
NINCDS-ARDRA National Institute of Neurological and Communicative Disorders and
Stroke and Alzheimer's Disease and Related Disorders Association
Alzheimer’s Criteria
OR Odds ratio
PC Principal component
SNP Single nucleotide polymorphism
TBM Tensor based morphometry
X^2 Chi-square; goodness of fit metric
University of Southern California
viii
LIST OF FIGURES AND TABLES
Table 2-1 ................................................................................................................................................................. 65
Table 3-1 ................................................................................................................................................................. 83
Table 3-2 ................................................................................................................................................................. 93
Table 4-1 .............................................................................................................................................................. 111
Table 4-2 .............................................................................................................................................................. 117
Table 5-1 .............................................................................................................................................................. 130
Table 5-2 .............................................................................................................................................................. 134
Table 5-3 .............................................................................................................................................................. 138
Table 6-1 .............................................................................................................................................................. 155
Table 6-2 .............................................................................................................................................................. 168
Figure 2-1 ............................................................................................................................................................... 43
Figure 2-2 ............................................................................................................................................................... 62
Figure 2-3 ............................................................................................................................................................... 63
Figure 3-1 ............................................................................................................................................................... 80
Figure 3-2 ............................................................................................................................................................... 85
Figure 3-3 ............................................................................................................................................................... 86
Figure 3-4 ............................................................................................................................................................... 88
Figure 3-5 ............................................................................................................................................................... 89
Figure 3-6 ............................................................................................................................................................... 92
Figure 3-7 ............................................................................................................................................................... 94
Figure 3-8 ............................................................................................................................................................ 104
Figure 4-1 ............................................................................................................................................................ 120
Figure 4-2 ............................................................................................................................................................ 121
Figure 4-3 ............................................................................................................................................................ 123
Figure 4-4 ............................................................................................................................................................ 124
Figure 5-1 ............................................................................................................................................................ 134
Figure 5-2 ............................................................................................................................................................ 140
Figure 5-3 ............................................................................................................................................................ 142
Figure 5-4 ............................................................................................................................................................ 143
Figure 5-5 ............................................................................................................................................................ 144
Figure 5-6 ............................................................................................................................................................ 146
Figure 5-7 ............................................................................................................................................................ 147
Figure 5-8 ............................................................................................................................................................ 149
Figure 6-1 ............................................................................................................................................................ 160
Figure 6-2 ............................................................................................................................................................ 164
Figure 6-3 ............................................................................................................................................................ 165
Figure 6-4 ............................................................................................................................................................ 167
Figure 6-5 ............................................................................................................................................................ 167
U n i v e r s i t y o f S o u t h e r n C a l i f o r n i a
9
TABLE OF CONTENTS
EPIGRAPH........................................................................................................................................................ II
DEDICATION ................................................................................................................................................... III
ACKNOWLEDGEMENTS ................................................................................................................................... IV
ABBREVIATIONS ............................................................................................................................................ VII
LIST OF FIGURES AND TABLES ........................................................................................................................ VIII
TABLE OF CONTENTS ........................................................................................................................................ 9
OUTLINE OF THE THESIS AND GENERAL GOALS................................................................................................ 10
1 ALZHEIMER’S DISEASE .......................................................................................................................... 13
2 APOE, SEX, AND AGE: THE TRIAD OF ALZHEIMER’S DISEASE ................................................................... 37
3 UNCOVERING BIOLOGICALLY COHERENT PERIPHERAL SIGNATURES OF HEALTH AND RISK FOR
ALZHEIMER’S DISEASE IN THE AGING BRAIN .......................................................................................... 68
4 ABO BLOOD TYPE IS ASSOCIATED WITH EARLIER AGE OF ALZHEIMER’S DISEASE ONSET AND
DIFFERENCES IN BRAIN STRUCTURE .....................................................................................................105
5 MAPPING THE HERITABILITY CONTENT OF DEVIATIONS FROM OPTIMAL BRAIN AGING ..........................127
6 INVESTIGATING THE RELATIONSHIP BETWEEN BRAIN PREDICTED AGE AND GENETICS ...........................152
7 SUMMARY ..........................................................................................................................................172
8 FUTURE DIRECTIONS ...........................................................................................................................177
REFERENCES .................................................................................................................................................179
U n i v e r s i t y o f S o u t h e r n C a l i f o r n i a
10
OUTLINE OF THE THESIS AND GENERAL GOALS
The work outlined in this thesis involves methods that were designed, accelerated, and
applied in order to solve problems in the following two areas as they pertain to aging and
Alzheimer’s disease (AD): 1) biomarker discovery or refinement in plasma or brain, 2)
exploration of genetic contributions, mechanisms, and pathways in mediating deviations
from optimal aging and risk for Alzheimer’s disease. The overarching goal of this thesis is to
address complex neuroinformatics and high-dimensional questions related to the brain in a
more clinically relevant way, such that the gap between the two disciplines may potentially
be lessened, and therefore, more salient risk factors identified, and targeted outcomes
become more approachable. While each of these fields has the potential to inform the other,
they are often treated separately.
In the first chapter of this thesis, I provide an extended review of background literature
relevant to the studies that follow. This chapter provides an overview of aging and
Alzheimer’s disease, and lays the groundwork for the necessary genetic concepts and
methods used in chapter three to chapter six. Chapter two is adapted from our published
review paper on apolipoprotein E (APOE), the greatest genetic risk factor for late-onset AD.
This chapter discusses the interactions with APOE, age, and sex, in mediating this risk, both
in consideration of amyloid and other AD hallmarks, but also through amyloid-independent
mechanisms, such as lipid metabolism and the immune system. Chapter three is adapted
from a paper that is currently under review at Frontiers in Aging Neuroscience and explores
how a multivariate network approach using brain and proteomic measures at baseline can
University of Southern California
11
inform prediction of disease across the trajectory of AD, improving our ability to detect
progression to mild cognitive impairment and AD, beyond simple univariate methods.
All participants included in chapters three and four were from the the Alzheimer’s
Disease Neuroimaging Initiative (ADNI) or AddNeuroMed. The diagnosis of probable AD for
both of these studies was performed according to NINCDS-ADRDA clinical AD criteria.
Participants included in chapters five and six were from the United Kingdom (UK) Biobank.
To improve training of brain-age for older individuals (i.e. 75 and older), stable (> 6 months)
cognitively normal individuals from ADNI and the Open Access Series of Imaging Studies
(OASIS) were additionally included in chapter six.
The work discussed herein contributes to the state of the field by exploring the role of
ABO blood type in AD, which to date has not been directly established. This work is presented
in chapter four. Given the established interplay between age and risk for AD, the latter two
studies in this thesis focus on increasing our understanding of the transitional state of
optimal aging and deviations from an optimal aging trajectory with the idea that while
separate from AD directly, studying health informs potential protective mechanisms from
disease; while the reverse may often be inferred, it is indirect and less clear. The work
presented in chapter five contributes to what is known about heritability of structural brain
regions through important considerations of optimal aging, rather than assuming age-
associated diseases do not make important impacts on heritability across the aging
trajectory. A novel framework is laid out in chapter six for future genetic association studies,
which considers the concept of brain-age and how genetics may predict deviations between
chronological and brain-predicted age using structural magnetic resonance (MRI) measures.
The goal of this work was to then identify the phenotypic impact of functional genetic
University of Southern California
12
variants that operate within convergent aging and AD-risk pathways, and in doing so, to shed
light on the roles of those pathways in suboptimal aging and AD. Although modeling aging
more broadly, we advance knowledge on how brain-age interacts with AD by then
performing gene-based scoring, pathway enrichment, and considering the genetic overlap
with AD. The final chapters of this thesis provide a summary of what has been identified
through this work, and possible future directions.
U n i v e r s i t y o f S o u t h e r n C a l i f o r n i a
13
1 ALZHEIMER’S DISEASE
History and Definition of Alzheimer’s Disease
lzheimer’s disease (AD) is the costliest disease in America, accounting for 60-
80% of all cases of dementia (Alzheimer’s Association, 2015; Barker et al.,
2002). There are an estimated 47.5 million people living with dementia
worldwide (World Health Organization, 2015). While the diagnostic term
“Alzheimer’s disease” has existed for just over 100 years, the concept of
dementia – a loss of ability to act or reason on one’s own volition and increasing
difficulty with memory – has roots in ancient times. Until around the 19
th
century, dementia was considered a normal, but exaggerated part of the aging
process, exhibiting progressive memory loss and difficulties with executive
function, changes in mood, and often impairments in speech. Early thinking of
AD as a normal part of aging, coined “senile dementia”, came about from
macroscopic investigations of postmortem brains, showing overlapping and
widespread deterioration in size in both aging and AD. In 1893, Arnold Pick,
known for discovering Pick’s disease, or frontotemporal dementia, identified
A
University of Southern California
14
focal atrophic brain changes that lead to specific cognitive disturbances in
individuals with senile dementia (Kertesz & Munoz, 1998; Pick, 1906). Shortly
thereafter, Emil Redlich observed sclerotic plaques in a case of senile dementia,
suggesting the role of proliferated glial cells in the disease (Redlich, 1898). The
particular case for which AD received its name arose from a “peculiar severe
disease process of the cerebral cortex” identified by the neuroanatomist Alois
Alzheimer in 1906 (Hippius & Neundörfer, 2003). Alzheimer witnessed the
progression of Auguste Deter shortly after she was admitted to the Royal
Psychiatric Hospital for untreatable paranoia. At death, the autopsy Alzheimer
performed revealed miliary bodies and dense bundles of fibrils, now known as
amyloid plaques and neurofibrillary tangles, that constitute the
neuropathological hallmarks of AD (Stelzmann et al., 1995; Schachter & Davis,
2000). Alzheimer was the first to identify neurofibrillary tangles using a silver
staining technique. Shortly thereafter, Emil Kraeplin, a colleague of Alzheimer,
named the illness “Alzheimer disease” when he published his Psychiatrie
textbook in 1910 (Hippius & Neundörfer, 2003). It would be several decades
until it was first discovered that amyloid plaques and neurofibrillary tangles
were present in old-age individuals both with and without dementia, and likely
the most common cause of dementia (Blessed, Tomlinson, & Roth, 1968;
Tomlinson, Blessed, & Roth, 1970; Ross, Tomlinson, & Blessed, 1966).
University of Southern California
15
These discoveries ultimately led to the realization that AD would present
as a major socioeconomic burden with population aging on the rise (Katzman,
1976). The coming decades then saw the emergence of international criteria
and standardized pathologic procedures for AD diagnosis at autopsy (Mirra et
al., 1991; Moms et al., 1989). AD has since come to be understood as a spectrum
of disease, instead of a singular process (McKhann et al., 2011). Given the
insidious and gradual onset of AD symptoms, the recognition and definition of
the intermediate stage between healthy aging and dementia lead to the term
“mild cognitive impairment” to refer to this stage (Reisberg et al., 1988). In
2010/2011 the International Working Group (IWG) and the National Institutes
of Health and Alzheimer Association (NIH-AA) met to establish and update AD
diagnostic criteria, particularly for these early stages, based on the expanding
knowledge acquired through research on AD (Dubois et al., 2010; McKhann et
al., 2011). This meeting identified the importance of biomarkers, such as
cerebrospinal fluid and brain imaging, to assist with understanding clinical
presentation of AD and reach a more conclusive diagnosis before autopsy. Most
recently, the NIH-AA research roundtable has expanded on these results,
particularly in the research setting, suggesting the latest definition of AD be
based on biomarkers assessing pathophysiological processes in the brain,
combining measures of amyloid, tau, and neurodegenerative pathology
University of Southern California
16
(A/T/N), rather than requiring clinical presentation of dementia (Knopman et
al., 2018). This definition additionally expands on the idea of AD as a continuum
and considers 6 stages of dementia: 2 stages with a cognitively unimpaired state
but with subjective cognitive impairment, 1 stage of mild cognitive impairment,
and 3 stages of clinical dementia with increasing severity.
Outside of research criteria, a diagnosis of MCI due to AD is made when
individuals demonstrate a consistent degree of cognitive decline, beyond age-
established norms in AD-related domains such as memory. MCI can, however,
involve single, or multiple cognitive domains beyond memory. These
individuals do not exhibit symptoms to the same degree as AD and are able to
maintain normal activities of daily living, and therefore do not fulfill clinical
criteria for dementia. Vascular or other medical conditions must be first ruled
out as the cause before a diagnosis of MCI is made (Petersen et al., 2014;
Winblad et al., 2004). While MCI is often considered a transition state to AD, one
multiethnic study of 2,364 individuals found that only 23% of individuals with
MCI progressed to AD, and 31% reverted to a status of cognitively normal
(Manley et al., 2008). Clinical AD may be diagnosed as either probable or
possible AD (McKhann et al., 2011). In either case, a diagnosis of dementia must
first be made, which requires cognitive or behavioral symptoms that interfere
with daily functioning, a demonstrated decline in performance or functioning,
University of Southern California
17
and symptoms that are not explained by delirium or a major psychiatric
disorder. In the case of probable AD, the individual must meet criteria for
dementia and show an insidious and gradual onset (typically months to years),
a clear history of worsening cognition (typically by an informant), and finally a
predominantly amnestic presentation with possible deficits in language,
visuospatial presentation, or executive dysfunction. Cerebrovascular disease
must be ruled out as the primary causal factor under both conditions. Although
typically sporadic and occurring after the age of 65, an individual presenting
with these symptoms and harboring a genetic variant in one of several highly-
penetrant mutations known to cause early-onset AD (EOAD), such as APP or
PSEN1/PSEN2 will likely be diagnosed with probable AD with an increased
level of certainty. A definite diagnosis of AD is still reserved following post-
mortem analyses and identification of neurofibrillary tangles and amyloid
plaques with immunohistochemistry at autopsy.
For the remainder of this thesis, AD will refer specifically to probable or
possible AD of the late-onset sporadic form (occurring after age 65), unless
stated otherwise. Although research on the early-onset form of AD has yielded
insights about the role of amyloid in the etiology of disease more broadly, and
Auguste Deter may have harbored variants for this form (PSEN1), EOAD
accounts for less than 1% of all cases, exhibits a different genetic architecture,
University of Southern California
18
as well as problems more predominantly with increased production of Aβ42
relative to Aβ40, instead of decreased clearance as is more often seen with late-
onset AD (LOAD) (Kim et al., 2014; Müller et al., 2013).
The Role of Amyloid in Alzheimer’s Disease
The Amyloid Cascade Hypothesis was first proposed by John Hardy and
Gerald Higgins in 1992 as a result of work showing that the plaques defining AD
pathology were made largely of the Aβ peptide, and that mutations on
chromosome 21 associated with familial EOAD were suggestive of mutations in
the Aβ precursor protein (APP) (Hardy & Higgins, 1992). This hypothesis states
that neurodegeneration in AD is likely a result of accumulation and aggregation
of the Aβ peptide, resulting in deposition of plaques. In LOAD, this accumulation
is largely the result of deficiencies in the clearance of Aβ from the brain.
According to the original hypothesis, this imbalance of Aβ production and
clearance subsequently drives all other AD-pathological changes, including
neurofibrillary tangle deposition, microglial and astrocytic activation, chronic
inflammation, and cerebrovascular disease (Hardy & Higgins, 1992; Hardy &
Selkoe, 2002).
Under non-amyloidogenic processing, synthesized APP is quickly
transported to the neuronal surface at synaptic terminals, and proteolyzed by α-
University of Southern California
19
secretase, forming soluble non-pathologic sAPPα (Koo et al., 1990). However, in
amyloidogenic conditions, APP is instead internalized into endosomes and
cleaved by both β-secretase (BACE1) and γ-secretase to form Aβ (Kinoshita et
al., 2003).
Aβ1-40 accounts for about 90% of Aβ released from neurons and
increases significantly only during the later stages of AD. Aβ1-42 accounts for
10% of secreted Aβ, is increased during the early stages of disease, and is found
in greatest concentrations in the neuritic plaques, with a high propensity for
aggregating into oligomers and insoluble fibrils (Citron et al., 1996; Roher et al.,
1993). CSF levels and plasma levels of Aβ1-40 and Aβ1-42 correlate in healthy
individuals, but typically not in those with AD, and a clear direction of change in
plasma has not been sufficiently determined (Fukumoto et al., 2003; Giedraitis
et al., 2007). Although other species of Aβ, such as Aβ1-56 have been associated
with AD and cognition, these two species are the primary focus of the majority
of AD-related research (Frydman-Marom et al., 2011; Quist et al., 2005). Aβ can
also be transported across the blood brain barrier by receptors such as the low-
density lipoprotein receptor-related protein (LRP) and receptor for advanced
glycation end products (RAGE) (Wang et al., 2006). Aside from Aβ, plaques
consist of dystrophic neurites, activated microglia, and cytokine and
inflammatory factors, such as complement activation products and TNF .
University of Southern California
20
However, the model that more Aβ is bad and “less is necessarily good” does not
accurately explain the complexities of this protein. As reviewed by Atwood et al.
(2003), evidence that conditions of ischemia, hypoglycemia, and traumatic
brain injury can induce the overexpression of APP and a shift from non-
amyloidogenic to amyloidogenic processing support a potentially protective
role of amyloid in disease (Jendroska et al., 1995; Murakami et al., 1998; Shi et
al., 1997). Moreover, APP has been shown to modulate neurotrophic signaling
and is necessary for the maintenance of neuronal integrity in the hippocampus
of mice (Hasebe et al., 2013; Tyan et al., 2012). While aggregated Aβ peptide
accumulation induces oxidative stress and results in toxicity to mitochondria
via mechanisms independent of the effects of reactive oxygen species, it also has
anti-oxidant properties thought to be a response to intracellular increases in
transition metals (i.e. copper) that catalyze the generation of reactive oxygen
species (ROS) (Opazo et al., 2002; Sinha et al., 2013).
The Role of Tau in Alzheimer’s Disease
In 1991, Heiko and Eva Braak used an advanced silver staining technique
to study the brains of 83 demented and non-demented postmortem brains and
subsequently describe the topological and temporal changes in AD. They
identified distinct changes in both amyloid and tau tangle deposition, with
University of Southern California
21
results corresponding to the previous cognitive status of the individuals (Braak
& Braak, 1991). They noticed that the presence of tangles often preceded
neuritic plaques and that these tangles exhibited a more consistent pattern of
progression than amyloid, allowing them to identify six (“Braak”) stages of
pathological disease progression that correspond to disease status. The stages
include predominantly “transentorhinal” (I/II; typically without clinical
manifestation of dementia), “limbic” (III/IV; MCI or early AD), and “isocortical”
(V/VI; later-stage AD dementia).
The normal function of tau in the brain is promoting the assembly of
tubulin and maintaining the stability of resulting microtubules, which act as
essential structural scaffolds and intracellular transport networks for
organelles, growth factors, and other components within neurons (Drubin &
Kirschner, 1986). When tau is hyperphosphorylated (by any of several kinases
including glycogen synthase kinase-3 (GSK3), cyclin-dependent protein kinase-
5 (cdk5), and mitogen activated protein ERK- 1 or 2), it fails to interact with
tubulin, thereby destabilizing microtubules and subsequently leading to axonal
degredation (Alonso et al.,1994; Singh et al., 1994).
Aβ and tau may act synergistically to mediate neuronal toxicity. For
instance, both Aβ and tau have been shown to impair mitochondrial function,
independently (Hauptman et al., 2006; Reddy et al., 2008). Similarly, 3xTgAD
University of Southern California
22
mice expressing APP/PS2/human-tau show mitochondrial dysfunction to a
greater extent than mice expressing APP or tau alone (Rhein et al., 2009; Ittner
et al., 2011). Additionally, tau has been shown to mediate Aβ toxicity, with tau -
/- neuronal cultures being resistant to Aβ toxicity and Aβ induced dysfunctions
in axonal transport (Rapoport et al., 2002; Vossel et al., 2010).
Brain Imaging in Alzheimer’s Disease
The uncertainty inherent in a clinical diagnosis of AD has propelled a
search for sensitive diagnostic markers (Johnson et al., 2012). Due to the brain’s
inaccessibility, imaging plays a vital role in this search, serving as quantitative
biomarkers, and a “window” to the brain. Magnetic resonance imaging (MRI)
has helped to establish that there is a long pre-clinical and pre-symptomatic
period where the pathological effects of AD are detectable, before clinical
manifestation of the disease. The model of AD progression proposed by Clifford
Jack and colleagues (Jack Model), an extension of the amyloid cascade
hypothesis, provides a framework for hypothesis testing of the temporal
evolution in Alzheimer’s disease (AD) biomarkers in relation to each other and
the onset and progression of clinical AD (Jack & Holtzman, 2013). Most of these
markers are derived from neuroimaging using structural MRI or positron
emission tomography (PET). The model focuses on the five most well-
University of Southern California
23
established biomarkers of AD, which are divided into two categories: measures
of brain Aβ deposition, and measures of neurodegeneration. Brain Aβ markers
include cerebral amyloid angiopathy and neuritic plaques, measured through
PiB (Pittsburgh compound B) or Florbetapir imaging or CSF analysis of Aβ1-42
levels. Tau neurofibrillary tangles (NFT) follow a stereotypic topographic
progression pattern, first appearing in the brainstem and entorhinal cortex and
progressing to the hippocampus and paralimbic region (Braak & Braak, 1991).
Neurodegeneration, which maps onto NFT distribution, is characterized as
atrophy and loss of neurons and neuronal processes (Braak et al., 1994).
Measures of neurodegeneration include increased concentrations of CSF total
tau and phosphorylated tau, hypometabolism on FDG- PET, and atrophy on
structural MRI. Neurodegeneration generally starts in the entorhinal cortex and
hippocampus and progresses to the frontal lobes and parietal neocortex and is
primarily measured through volumetric and thickness differences. Notably,
while neurodegeneration in the medial temporal lobe is a useful marker of AD,
entorhinal and hippocampal volume are more predictive of conversion from
MCI to AD in non- APOE-ε4carriers (Mosconi et al., 2004). Although used in
research settings, tau PET has not yet been incorporated into this model.
Likewise, diffusion tensor imaging of white matter macrostructural differences
has largely not been incorporated, particularly as current tensor models have
University of Southern California
24
low power to detect the more specific microstructural differences associated
with early glial cell changes in AD. Atrophy on MRI correlates with neuron loss,
Braak NFT stage, and tau immunostaining, and does not correlate well with Aβ
load (Josephs et al., 2008). Therefore, it is believed that MRI offers a measure
related to tau-related neurodegeneration. FDG PET (glucose) hypometabolism
is also correlated with NFT burden. The Jack model posits that amyloid markers
(e.g. CSF and PiB/Florbetapir imaging) change earlier than all other markers in
the pure form of AD, and that neurodegenerative markers appearing first only
occurs for the late onset form of AD, brought on by other pathologies (Jack et al.,
2013). For instance, measures of hippocampal atrophy lack specificity as they
are also reduced in conditions such as diabetes as well as in other forms of
dementia (Fotuhi & Jack, 2012). Evidence of hippocampal sparing has also been
documented in some cases of early-onset AD, but less so in LOAD (Murray et al.,
2011). Aβ positivity as measured by PiB or Florbetapir is also associated with
substantial decline in episodic memory over 18 months in healthy older adults
(Chetelat et al., 2012). Longitudinal clinical follow-up for 3 years on people with
mild cognitive impairment (MCI), believed to be a transition state between
healthy aging and AD, showed that 57 of the 155 MCI patients progressed to AD,
and 93% of those who progressed were amyloid positive at baseline, while only
7% of those who progressed to AD were amyloid negative (Jagust et al., 2010).
University of Southern California
25
Aging Associated Changes Predispose an Individual to AD
Oxidation of proteins, lipids, and nucleic acids, as occurs increasingly
with age, activates cytokines, such as IL-1, IL-6, and TNF , production of
advanced glycation end-products and subsequent substrate binding to receptor
for advanced glycation end products (RAGE). These processes form a positive
feedback loop and have all been highly implicated in risk for AD (Deane et al.,
2004; Singh et al., 2001). Importantly, RAGE serves as a central component
regulating the cross-talk between the innate and adaptive immune systems.
Moreover, the first longevity mutant to be identified was the C. elegans gene
age-1, that encodes phosphatidylinositol 3-kinase (PI3K) and is downstream of
RAGE. Inhibition of PI3K has been shown to activate glycogen synthase kinase-3
and subsequent phosphorylation of tau (Friedman et al., 1988; Lui et al., 2003).
As we age, the innate immune system becomes dysregulated and is
characterized by persistent inflammatory responses that involve multiple
immune and non-immune cell types. For instance, neutrophils in old age show
both a decrease and impairment in phagocytosis, as well as a decrease in
superoxide generation, important for protection against free radicals (Fortin et
a., 2008; Radford et al., 2010; Simell et al., 2011). Additionally, eotaxin-1, a β-
chemokine responsible for inducing migration of monocytes and other immune
University of Southern California
26
cells, increases with age, and a functional variant has been shown to delay the
onset of AD through decreased accumulation of eotaxin (Lalli et al., 2015; Van
Coillie et al., 1999).
Finally, a gene expression microarray study of the hippocampus,
entorhinal cortex, superior frontal gyrus, and post-central gyrus in young (20-
59 years old), aged (60-99 years old), and individuals with Alzheimer’s disease
(74-95 years old), revealed marked upregulation of toll-like receptors, the
inflammasome, genes associated with microglial activation, and many other
immune-associated genes in older adults. These changes were partially
exacerbated, and largely overlapped with changes in individuals with AD,
compared to older adults without AD (Cribbs et al., 2012). Genes with
expression showing the greatest change in the transition from normal cognition
to AD were in the superiorfrontal gyrus (SFG) and hippocampus (HC). These
results indicate that the extent of innate immune gene upregulation in AD was
modest relative to the robust response apparent in the aged brain, consistent
with the emerging idea of a critical involvement of inflammation and immune
dysregulation in the earliest stages, perhaps even in the preclinical stage, of AD.
Notably, a greater number of genes exhibiting significant changes in expression
levels were observed in females relative to males in AD, especially in the
entorhinal cortex (EC) and HC. This finding is in line with previous results
University of Southern California
27
showing a sex-dependent effect of IL6 on synaptic dysfunction in the
hippocampus (Gordon et al., 2001; Walsh et al., 2014). Only two chemokines
showed a significant progressive pattern of change across aging and AD:
CXCL16, which was upregulated in the EC and SFG, and CXCL14, which was
progressively downregulated in the HC across aging and AD. These results also
showed an induction of MHC class I and II molecules across both aging groups
across the four brain regions. Markedly, MHC molecules are upregulated on
microglia and macrophages in response to elevated levels of cytokines
associated with inflammation, and in response to chronic pathology and
neurodegeneration, such as occurs in AD (McGeer et al., 1993; O'Keefe et al.,
2002). Indeed, increased gene expression of MHC II has been extensively
documented in both AD and transgenic mouse models of AD (Cuello et al.,
2010).
The Role of the Immune System in Alzheimer’s Disease
The concept of hormesis may play a pivotal role in bridging the
connection between age-associated changes and risk for AD. Specifically, the
systems-wide response to stress and immune activation as noted above can
lead to protection against pathogens, but this response is often awry along the
trajectory to AD. Moreover, infection burden increases the odds for developing
University of Southern California
28
AD (Bu et al., 2014). As individuals age, the brain accumulates levels of multiple
endogenous and exogenous factors that act as low-grade irritants and prime
microglial responses. AD develops when the accumulating effect of these factors
surpasses a particular threshold under a reduced capacity to temper microglial
reactivity. Together with increased levels of potentially harmful endogenous
factors, Aβ impairs growth factor signaling by activation of proinflammatory
cytokines, metabolic deficits, and other factors, converging to exceed the
threshold for the onset of AD. Thus, Aβ is likely to be both an initiating trigger
and chronic driver of innate immune activation, triggering the complement
system, activation of toll-like receptors, and activation of the inflammasome
(Fassbender et al., 2004; Halle et al., 2008; Jack et al., 2010; Rogers et al., 1992).
Additionally, alterations of the immune system are widespread in
neurodegenerative diseases including AD, Parkinson’s disease, multiple
sclerosis, and amyotrophic lateral sclerosis, and others. Although non-disease
specific alterations may be involved, this may support the role of the immune
system as an early factor whereby interactive effects with specific biological or
environmental stressors determine the particular neurodegenerative disease
that then develops. Indeed, viral infections have been linked to each of these
disease processes (Calne & Langston, 1983; Itzhaki et al., 1997; Pakpoor et al.,
2013). Moreover, network approaches have implicated both immune and
University of Southern California
29
microglial dysfunction as the leading perturbed pathways in AD, in such a way
that was specific to AD, but not neurodegeneration more generally, suggesting
that while there may be overlap broadly with these systems and
neurodegeneration, the specific components may be different for a particular
neurodegenerative disease (Zhang et al., 2013). Finally, this finding is
corroborated by pathway enrichment analyses using results from the
International Genomics of Alzheimer’s Disease Consortium (IGAP) on data from
over 74,000 individuals, with the most significant enrichment for genes
involved in the immune response (Jones et al., 2015).
A complex reshaping of the immune system occurs with age and
immunosenescence, and is a dynamic process. Indeed, AD heritability is
enriched for gene-sets involved in functional annotations for cells in the
innate/myeloid and adaptive/B-lymphoid lineages (Huang et al., 2017).
Likewise, genome-wide association studies in thousands of patients and
controls have identified susceptibility variants in innate-immune-related genes
that are expressed by myeloid cells, including CD33, TYROBP, and TREM2
(Griciuc et al., 2013; Guerreiro et al., 2013; Hollingsworth et al., 2011; Naj et al.,
2011; Replogle et al., 2015; Zhang et al., 2013). Additionally, under an
amyloidogenic environment, the brains resident immune cells, microglia, are
found nearby and phagocytose plaques. Likewise, post mortem studies of AD
University of Southern California
30
brains have revealed abnormal expression of cytokines, complement proteins,
and other immune and inflammatory mediators (Akiyama, et al., 2000). Finally,
acute phase proteins of the immune system including IL-6 and C-reactive
protein (CRP) have been found in both the amyloid plaques and neurofibrillary
tangles that develop in AD (McGeer & McGeer, 2002).
Discussion of the immune system in AD would be incomplete without
mention of cancer. Although dismal, it has been stated that if one lives long
enough, the development of either AD or cancer is certain. Yet, these two
diseases bear inverse relations with each other. Cancer survivors have a
decreased risk for AD (Roe et al., 2005). Moreover, neoplasia and
neurodegeneration share many genes and biological pathways, largely involving
the innate immune and inflammatory systems, though they are often regulated
in opposing directions (Behrens et al., 2009). APOE ε2, which may be protective
against AD, increases the risk and aggressiveness of certain cancers (Ifere et al.,
2013). Immune genes that have been linked to both AD and cancer include
multiple variants in HLA, SMAD7, TMEM173, and TREM2 (Jonsson et al., 2013;
Yan-Shi et al., 2017). Indeed, repurposing of cancer chemotherapeutics through
immune checkpoint inhibition is currently being explored for AD therapeutics
(Baruch et al., 2016).
University of Southern California
31
Alzheimer’s Genomics and Proteomics
The universal language that unites all living beings involves the interplay
between protein and deoxyribonucleic acid (DNA). Proteins, which are large
macromolecules of amino acid residues involved in enzymatic reactions of
metabolism, DNA replication, molecular transport, and other functions, is a
gene’s way of making another gene, while a gene, which is a sequence of DNA
that encodes for a molecule with a particular function, is a protein’s way of
making another protein. As such, considering the two jointly is important. The
first human genetic discovery, showing alterations at the genetic and proteomic
levels, came with the identification of the ABO blood type, which has newly
emerging implications in AD, as discussed in chapter 4 (Landstein & Levine,
1927). While biological risk pathways involved in AD may originate at the
genetic level, the consequences are far-reaching and span mRNA, protein, brain
structure, cognition, and behavior. Discussed in more detail in chapter two,
APOE, the greatest known genetic risk factor for late-onset AD, has been shown
to influence clinical course of AD and possibly faster rates of cognitive decline
through alterations in protein structure and function (Corder et al., 1993;
Martins et al., 2005). The first AD risk genes were identified through linkage
studies and included APP and presinilin 1, which are both highly penetrant risk
genes for EOAD. To date several risk genes have been identified that are
University of Southern California
32
associated with increased susceptibility to AD and a list of some of these top
genes are outlined in Table 1 (Bertram et al., 2007). These genes can largely be
categorized into production/degradation/clearance of amyloid, lipid
metabolism, innate immunity, and cell signaling. The majority of the more
recent discoveries have used genome-wide association studies (GWAS), which
examine a large portion of the genome through a hypothesis-free approach
using thousands, or millions, of single-nucleotide polymorphisms (SNPs) to
determine whether there are consistent differences between patients with AD
and controls. The first ten GWAS studies in AD used data from fewer than 2,000
AD cases, typically resulting in significant association for APOE only or results
that failed to replicate. However, based on twin studies, AD is highly heritable at
up to 80%, yet many of the discovered variants, including APOE, only contribute
a small portion to this picture, indicating that multivariate analyses considering
SNPs jointly may be an important avenue to aid in the “missing heritability”
problem (Gatz et al., 2006).
Failures of reproducibility are not uncommon in science, and genome-
wide association studies are no exception. Some of the causative factors in this
regard include small sample sizes and population structure. To address the
latter, multidimensional scaling or a related variation, principal components
analysis has been included in GWAS models, which capture the genetic
University of Southern California
33
correlation or relationships among individuals (Patterson & Reich 2006; Li et
al., 2008). Many studies have additionally chosen to focus on single ethnic
populations, with the idea that this reduces genetic variability due to non-causal
effects. Yet, studies have shown that the majority of genetic differences between
the average members of different races are small, while the aggregate
differences between any two individuals of the same or different race tend to be
much larger. In fact, only 7% of the genetic differences between individuals can
be attributed to the fact that they are of different races (Jorde et al., 2000).
Additionally, studies now recognize the importance of larger sample sizes to
increase power to detect consistent differences between groups. Although
computationally more complex, mixed effects models have been more recently
introduced to allow for addressing both population structure and cryptic
relatedness that have allowed us to move beyond many of the issues of
race/ethnicity and confounding due to relatedness. As a result, some of the
reproducibility crises have been mitigated, notably leading to reduced issues of
inflated significance and improved power to detect true signal (Hang et al.,
2010; Price et al., 2010; Yang et al., 2014).
University of Southern California
34
Gene Chromosomal Odds ratio Protein
location (95% CI)*
1. APOE 19q13.32 3.68 (3.30, 4.11) Apolipoprotein
2. BIN1 2q14.3 1.15 (1.10, 1.20) Bridging integrator 1
3. CLU 8p21.1 0.88 (0.86, 0.91) Clusterin
4. ABCA7 19p13.3 1.23 (1.18, 1.28) ATP-binding cassette, subfamily A,
member 7
5. CR1 1q32.1 1.14 (1.08, 1.20) Complement component (3b/4b)
receptor 1 (Knops blood group)
6. PICALM 11q14.2 0.88 (0.85, 0.91) Phosphatidylinositol-binding clathrin
assembly protein
7. MS4A6A 11q12.2 0.9 (0.88, 0.93) Membrane-spanning 4-domains,
subfamily A, member 6A
8. CD33 19q13.33 0.89 (0.86, 0.92) CD33
9. MS4A4E 11q12.2 1.08 (1.05, 1.11) Membrane-spanning 4-domains,
subfamily A, member 4E
10. CD2AP 16p12.3 1.12 (1.08, 1.16) CD2-associated protein
*Odds ratio is calculated from all studies including the initial study. (Bertram et al. The AlzGene
Database. Alzheimer Research Forum. Available at: http://www.alzgene.org. Updated
2011.04.18).
University of Southern California
35
Multivariate Models to Improve Detection and Understanding of
Alzheimer’s Disease
The majority of genes are expressed during embryogenesis, and genes
involved in the developmental process are typically regulated in a coordinated
way (Hong et al., 2010). Likewise, common human diseases originate from
complex interactions between genetic variations and environmental factors,
such as diet, age, and sex. Hence, for many diseases, the underlying mechanisms
are very complex and poorly understood. Additionallly, much of the missing
heritability in single-locus genome-wide studies can be explained by a large
number of loci that have a joint effect on the phenotype but may only provide a
weak signal on their own (Yang et al., 2010). Master regulatory genes involved
in multiple pathways, such as PGC1 , the master regulator of mitochondrial
biogenesis, PP2A, the master regulator of the cell cycle, and GSK-3, the master
regulator of neuronal progenitor homeostasis all have been implicated in risk
for AD and serve as salient examples for the importance of considering
networks and multivariate genetics (Hooper et al., 2008; Qin et al., 2009;
Vogelsberg-Ragaglia et al., 2001). Targeted analyses of these genes have
identified an extensive network of molecular interactions that are involved in
AD pathogenesis, functioning not as switches but more as part of a complex
interacting symphony. Moreover, chromosome 19, which harbors the APOE
University of Southern California
36
gene and many other AD-risk genes, as discussed in chapter two, holds many
keys to increasing our understanding of AD, which as we currently understand
it, speaks to prominent roles of lipid metabolism, mitochondrial functioning,
and the innate and adaptive immune systems. In the context of multivariate
genetics, many different approaches may be used from haplotype-based
methods, network analyses, gene and pathway enrichment, and methodology
specifically designed to target interactions. The work presented in this thesis
considers the majority of these approaches by constructing proteomic networks
and using these to determine risk for progression to AD in chapter three,
examining the role of ABO blood genotype and phenotype, and interactions with
FUT2 secretor status and APOE, in mediating risk for AD, as presented in
chapter four, determining SNP-based heritability of deviations from optimal
aging using a genetic relatedness matrix in chapter six, and mixed-effects GWAS
and subsequent pathway enrichment analyses of variants associated with brain
predicted age in chapter seven.
U n i v e r s i t y o f S o u t h e r n C a l i f o r n i a
37
2 APOE, SEX, AND AGE: THE TRIAD OF
ALZHEIMER’S DISEASE
Abstract
Age, apolipoprotein E ε4 (APOE) and chromosomal sex are well-established risk factors for late-
onset Alzheimer’s disease (LOAD; AD). Over 60% of persons with AD harbor at least one APOE-ε4
allele. The sex-based prevalence of AD is well documented with over 60% of persons with AD being
female. Evidence indicates that the APOE-ε4 risk for AD is greater in women than men, which is
particularly evident in heterozygous women carrying one APOE-ε4 allele. Paradoxically, men
homozygous for APOE-ε4 are reported to be at greater risk than homozygous APOE-ε4 women for
mild cognitive impairment and AD. Herein, we discuss the complex interplay between the three
greatest risk factors for Alzheimer’s disease, age, APOE-ε4 genotype and female sex. We propose that
the convergence of these three risk factors, and specifically the bioenergetic aging perimenopause to
menopause transition unique to the female, creates a risk profile for AD unique to the female. Further,
we discuss the unique risk of the APOE- ε4 positive male which appears to emerge early in the aging
process. Evidence for impact of the triad of AD risk factors is most evident in the temporal trajectory
of AD progression and burden of pathology in relation to APOE genotype, age and sex. Collectively,
the data indicate complex interactions between age, APOE genotype and sex that belies a one size fits
all approach and argues for a precision medicine approach that integrates the three main risk factors
for AD.
University of Southern California
38
This chapter was modified and adapted from
Riedel, B. C., Thompson, P. M., & Brinton, R. D. (2016). Age, APOE and sex: triad of risk of Alzheimer’s
disease. The Journal of Steroid Biochemistry and Molecular Biology, 160, 134-147.
1.1. Introduction
The greatest risk factors for Alzheimer’s disease are age (Brookmeyer et al., 1998; Brookmeyer
et al., 2007; Hebert et al., 2013), the ApoE4 allele (Liu et al., 2013; Mahley et al., 2009; Mayeux et al.,
1993) and female sex (Brookmeyer et al., 1998; Damoiseaux et al., 2012; Farrer et al., 1997;
Mortensen and Høgh, 2001; Payami et al., 1994). Prevalence of AD is greater in women (Brookmeyer
et al., 1998) whereas the incidence for AD is reported to be comparable in women and men (Barnes
et al., 2003) until later age when the incidence is greater in women (Ruitenberg et al.,
2001). Typically, the greater risk of AD in females is attributed to their greater longevity of, on
average, 4.5 years.
Herein, we review the three greatest risk factors for AD and propose that it is the interaction
between these risk factors that either increase or decrease risk of Alzheimer’s. The complexity of the
interaction between this triad of risk factors belies simple on factor causality but instead more closely
approaches the complexity of factors leading to late onset Alzheimer’s disease (LOAD). While this
approach considers three factors that have broad systems biology implications, it still does not fully
capture the complexity of the processes leading to risk of LOAD. It is, however, a step in the right
direction.
1. AGING
Age remains the greatest risk factor for Alzheimer’s and is thus a fundamental driver for
University of Southern California
39
development of the disease (Aging, 2015). Aging at the global population level is unprecedented and
without parallel in the history of humanity (United Nations, 2001). During the 20th century the
proportion of older persons continued to rise and this trend has continued into the twenty-first
century (United Nations, 2001). By 2050, the number of older persons in the world will exceed the
number of young for the first time in history (United Nations, 2001). Between 2015 and 2030, the
number of people in the world aged 60 years or over is projected to grow by 56% and by 2050 the
global aged population is projected to more than double (World Health Organization, 2015). Globally,
the number of people aged 80 years or over, the “oldest-old” persons, is growing the fastest. People
aged 80 or older currently constitute more than 3% of the population of Northern America and nearly
3% of the population of Europe. Population aging is a pervasive global phenomenon that will endure
for the foreseeable future. By 2050, one in every five people will be aged 60 years or over (15).
The female prevalence of AD is well documented and is generally attributed to the greater life
span of women relative to men (Brookmeyer et al., 1998). Globally, women outlive men by an average
of 4.5 years (World Health Organization, 2015). However, survival of males is projected to be
comparable to females with near equality in longevity between females and males (DoEaSAPD). Thus,
the greatest risk factor for AD will be equally distributed between the sexes in the near future. Data
from the multiple studies show that the incidence of Alzheimer's disease rises exponentially after the
sixth decade of life (Masters et al., 2015).
Aging is a complex progressive process involving every organ and cell system in the body and is
the result of coordinated systems biology events that can span decades (Blalock et al., 2011; Blalock
et al., 2003; Blalock et al., 2004; Brinton, 2013; Brinton et al., 2015; Chow and Herrup, 2015; Curtis
et al., 2005; Dorshkind et al., 2009; Petanceska, 2003; Shaw et al., 2013; Yeoman et al., 2012). Further,
aging is typified by a coordinated series of sequential steps involving specific pathways at each phase
in the aging process (Brinton et al., 2015; Yin et al., 2015). Systems driving brain aging are
contributors to risk of AD and include glucose hypometabolism and mitochondria dysfunction, innate
University of Southern California
40
immune and inflammatory reactions, beta amyloid processing, dysregulation of cholesterol
homeostasis, white matter degeneration and decline in regenerative capacity (Blalock et al., 2003;
Blalock et al., 2004; Brinton, 2013; Brinton et al., 2015; Curtis et al., 2005; Dorshkind et al., 2009;
Klosinski et al., 2015; Masters et al., 2015; Puglielli et al., 2003; Shaw et al., 2013).
While aging is typically described and depicted as a linear process, in reality the program of
senescence is nonlinear and better represented as step functions between transition states. Aging is
typified by transition states and because aging is occurring throughout multiple organ systems the
process of aging is highly complex (Blalock et al., 2011; Blalock et al., 2003; Blalock et al., 2004;
Brinton et al., 2015; Curtis et al., 2005). Each phase of the aging process can be considered a transition
state that can be modified to either accelerate or delay progression to the next phase of aging
(Brinton, 2013; Brinton et al., 2015; Dorshkind et al., 2009).
It is increasingly clear that complex systems often have critical thresholds, often referred to as
tipping points, when a system shifts from one state to the other (Brinton et al., 2015). Analyses of
transition state dynamics predict three health states, healthy, pre-disease and disease (Brinton et al.,
2015). The pre-disease transition state is typically defined as the limit of the normal state that
precedes the tipping point into disease. Importantly, the pre-disease state is unstable and is thus
potentially reversible. However, the duration of reversibility is limited. The bifurcation point
between pre-disease and disease is characterized by a critical slowing of the system, in which it
becomes increasingly slow to recover from small perturbations to equilibrium (Brinton et al., 2015).
Transition states are inherently unstable and in the case of neurological transition states,
indicators of dysfunction at the limits of normal can be signals of instability and tipping points. The
presence, variability, intensity and duration of neurological symptoms provide potential advance
warning signs of impending risk of later health risks, particularly neurodegenerative diseases.
Multiple conditions that emerge during aging, such as metabolic dysregulation, cholesterol
dyshomeostasis, insomnia, depression, subjective memory complaints and cognitive decline are
University of Southern California
41
associated with increased risk of neurodegenerative diseases later in life such as Alzheimer’s
(Brinton et al., 2015; Rettberg et al., 2016). Alzheimer’s, like multiple neurodegenerative diseases, is
characterized by a long prodromal period, 20 years, during which the disease progresses to clinically
diagnosed dysfunction. Identifying persons that are at risk for AD while still in a modifiable transition
states is critical for reversing or delaying development of Alzheimer’s (Rettberg et al., 2016).
Individual genetics form the foundation for and can modify the complex multidimensional
trajectory of aging (Holmans et al., 2005). Considered below are the two major genetic risk factors of
AD, APOE genotype and chromosomal sex.
2. OVERVIEW OF APOE
2.1. Biological Role and Evolution of APOE
Apolipoprotein E (ApoE) is a 34-kDa lipid binding protein that functions in the transport of
triglycerides and cholesterol in multiple tissues, including brain, by interacting with lipoprotein
receptors on target cells (Bu, 2009; Wang et al., 2006). ApoE is a cholesterol transporter and
functions as a key regulator to coordinate the mobilization of cholesterol between cells and to
redistribute cholesterol within cells. These functions are particularly critical for the nervous system
where ApoE transport of cholesterol is critical for maintenance of myelin and neuronal membranes
both in the central and peripheral nervous systems (Leduc et al., 2010). In the CNS, ApoE functions
in conjunction with APOJ and APOC1, which together deliver cholesterol necessary for membrane
remodeling, required for synaptic turnover and dendritic reorganization (Leduc et al., 2010). ApoE
is particularly critical to the brain as other cholesterol transporters abundant in the plasma, such as
ApoA1 and ApoB, are virtually absent in brain thus making the brain particularly reliant upon ApoE
for cholesterol transport (Leduc et al., 2010). The most recent evidence suggests that ApoE has roles
beyond cholesterol metabolism, serving as an important transcriptional repressor within numerous
University of Southern California
42
systems, including repression of NFκB, which is important for immune regulation, and repression of
kinases involved in apoptosis, such as MAP Kinase Activating Death Domain (Theendakara et al.,
2016).
APOE, a 3.6 kb-long gene, is located on chromosome 19 and (Lyall et al., 2014) encodes for
apolipoprotein E (ApoE), a 299 amino acid-long lipoprotein. Three ApoE isoforms exist in humans:
ApoE2, ApoE3, and ApoE4, which differ from one another by single amino acid substitutions at
positions 112 and 158, ApoE2 (Cys-112, Cys-158), ApoE3 (Cys-112, Arg-158), and ApoE4 (Arg-112,
Arg-158) (Mahley et al.). Substitution of cysteine at position 158 in ApoE2 results in
hypocholesterolemia caused by low levels of low-density lipoprotein (LDL), cholesterol (Mahley et
al., 2009). In contrast, substitution of cysteine with arginine at position 112 in ApoE4 results in
elevation of plasma cholesterol and LDL levels and predisposes the carrier to cardiovascular disease
and neurodegenerative disorders, including Alzheimer’s disease (AD) (Mahley et al., 2009).
APOE evolved from the common hominid ancestor of humans and the great apes (Mortensen et
al., 2001). While there are three main isoforms of APOE in humans (i.e. ε2, ε3, and ε4), all great apes
carry the APOE-ε4 allele (Hanlon and Rubinsztein, 1995). The ε3 allele is the most common in
humans, especially in regions with a long-established agricultural economy. However, the ancestral
ε4 allele remains high in regions where an economy of foraging still exists or where food-supply is
often scarce (Corbo and Scacchi, 1999). Approximately 75 million Americans are heterozygous for
the ε4 allele and around 7 million are homozygous. Although generally more common among African
populations, the risk in cognitive decline conferred by carrying the ε4 allele is greater among
individuals of European descent, particularly amongst northern European regions (Kuller et al.,
1998).
University of Southern California
43
FIGURE 2-1
ApoE NMR protein structure. Created from the RCSB Protein Databank and the PyMol graphics viewer (PDB
ID: PO2649) (Bu, 2009; Chen et al., 2011a), this figure shows the nuclear magnetic resonance protein
structure of apoE ε3. Lipoprotein binding (Roses et al. 2010) and Aβ-binding (2) motif regions are delineated
(Bu, 2009; Garai et al., 2011). Residues 112 and 158, which are altered between (and therefore determine)
the differing isoforms, are also labeled (Huang et al., 2011; Zhong and Weisgraber, 2009).
2.2. APOE Protein Structure: Lipoprotein and Beta Amyloid Binding Motifs
As depicted in Figure 1, the ApoE protein has an N-terminal receptor-binding region and C-
terminal hydrophobic lipid-binding region located at amino acids 244-272. It is the N-terminal
region, which contains the two polymorphic positions (i.e. positions 112 and 158) that distinguish
the differing isoforms in many ways, such as differing anti-oxidant profiles (where ε2 ≥ ε3> ε4)
(Miyata and Smith, 1996; Su et al., 2008). In APOE-ε4, Arginine-61 interactions with Arginine-112
lead to a conformational change causing Arginine-61 to interact with Glutamine-255 in the aqueous
environment, resulting in N- and C-terminal domain interactions that do not exist in the ε2 or ε3
isoforms, due to the less hydrophobic Cysteine-112 (Zhong and Weisgraber, 2009). These
interactions are particularly evident when ApoE is delipidated, preventing lipoprotein receptor
docking and internalization of unlipidated ApoE. APOE-ε4 also has an arginine-112 to glutamic acid-
109 salt bridge, which causes the arginine-61 side chain to extend away from the 4-helix bundle in
its structure. In turn, the arginine-61 domain interacts with glutamic acid-255 causing the structure
University of Southern California
44
of ε4 to become more compact compared to ε3. Structural differences lead to regions in ε4 that are
less protected and less stable than the same regions in ε3 (Huang et al., 2011). This difference also
leads to a binding profile that favors triglyceride-rich very low-density lipoproteins (VLDL) and low-
density lipoproteins (LDL) over small phospholipid-rich high-density lipoproteins (HDL) particles.
Compared to the structure of ε3, ε2 also causes changes in lipoprotein preference, due to impaired
lipolytic processing of VLDL, creating increases in VLDL and decreases in LDL cholesterol (Louhija et
al., 1994). These differences in the ε2 and ε4 structures lead to increases in pro-atherogenic
lipoproteins, an accelerated trajectory towards atherogenesis compared to ε3, and therefore an
increased representation in patients with hyperlipidemia and cardiovascular disease (Mahley and
Rall Jr, 2000). However, despite this high prevalence and the strength of its association with both
cardiovascular and cognitive risk, APOE testing is still used most exclusively for research purposes
only.
2.3 ApoE and Cholesterol Transport
The APOE gene is expressed most highly in the liver and brain (Bu, 2009; Elshourbagy et al., 1985;
Leduc et al., 2010). Over 75% of plasma ApoE protein is derived from hepatic synthesis (Elshourbagy
et al., 1985). Second to the liver, is brain synthesis of ApoE which predominantly occurs in astrocytes,
followed by microglia and neurons (Xu et al., 2006). Given this high production, ApoE is the most
prevalent brain lipoprotein. The level of APOE expression varies by genotype, with ε2 typically having
the greatest and ε4 having the lowest expression (Saunders, 1993; Schiele et al., 2000). APOE isoform
effects on plasma ApoE concentration accounts for 20% of inter-individual variability of ApoE
concentration across all ages, with other factors including BMI, waist-to-hip ratio, oral
contraceptives, and sex, also playing prominent role (Reilly et al., 1992).
APOE isoforms differ in their rates of catabolism such that ε2 is catabolized more slowly than ε3
University of Southern California
45
or ε4 (Gregg et al., 1981). The increased rate of catabolism of ApoE ε4 leads to reduced availability of
ApoE to serve as a clearance protein, required for the clearance of cholesterol and toxic amyloid-beta
(Aβ) oligomers (Vance and Hayashi, 2010).
Excess plasma ApoE stimulates hepatic VLDL production and impairs lipolysis whereas a low
level of ApoE impairs plasma clearance of triglyceride-rich lipoproteins. Accordingly, there is about
a 3-fold increase in AD-risk associated with the lowest tertile of ApoE plasma concentrations,
independent of APOE genotype (Rasmussen et al., 2015). Notably, ApoE plasma levels are lowest in
MCI patients who progress to AD. Given that ε2 carriers generally show the highest protein
concentration, there is some level of reduction in AD-risk in these individuals, while ε4 carriers have
an increased risk. Studies of cultured astrocytes show that ApoE-ε3 containing astrocytes are able to
release more cholesterol to supply neurons than ApoE-ε4 containing astrocytes (Gong et al., 2002).
ApoE-ε4 protein is also less efficient at delivering essential fatty acids such as docosahexaenoic
acid (DHA) to neurons, thereby altering the function of glucose transporters and possibly playing a
role in the development of insulin resistance, a state commonly attributed to an increased risk of AD
(Lane and Farlow, 2005). Given this alteration, it has been suggested that eiscosapentaenoic acid
(EPA)-rich oils may be more suitable in ε4 individuals, especially for use as a hypotriglyceridemic
agent, showing a greater hypercholesterolaemic response to DHA, and greater catabolism, than EPA
(Leigh-Firbank et al., 2002; Quinn et al., 2010). Indeed, there is a significant interaction between
APOE genotype and plasma lipid response to both diet and pharmacological therapies (Ordovas and
Schaefer, 1999). Since a ketogenic diet has shown to increase HDL and decrease LDL cholesterol in
both obese and normal weight individuals, it has been proposed that ε4 carriers would benefit more
from a ketogenic diet than non-carriers, particularly in regard to protection from AD (Cunnane et al.,
2011; Dashti et al., 2004; Sharman et al., 2002). Furthermore, diets higher in saturated fats or diets
rich in olive oil and high in monounsaturated fats both show more favorable outcomes in LDL-
particle count in ε4 individuals than ε3 individuals (Lahoz et al., 2001; Mente et al., 2009; Wood et
University of Southern California
46
al., 2006; Yancy et al., 2004); LDL-particle count is seen as a better measure of atherogenicity than
LDL cholesterol itself (Shalaurova et al., 2014; Wallenfeldt et al., 2004).
Due to the altered protein structure, ApoE-ε4 is poorly lipidated by ABCA1 and ABCG1, causing it
to be more rapidly catabolized than the other isoforms. Furthermore, APOE polymorphisms show a
differential impact on competition between fasting triglyceride-rich lipoproteins and LDL's for the
LDL receptor (LDL-R) (Olano-Martin et al., 2010), ultimately leading to downregulation of LDL-R in
states of high VLDL production, as seen in carriers of the ε2 and ε4 allele. These LDL-R interactions
are necessary for remnant clearance of the partially metabolized atherogenic chylomicrons and VLDL
particles. Interestingly, APOE-ε4 in non-human primates does not contain the domain interactions
that gives human ε4 its characteristic salt-bridge, thus it behaves more like ε3 in humans in that it
does not show preferential binding to triglyceride-rich lipoproteins like the human variant (Chen et
al., 2011a). In line with this, it has been proposed that gene targeting to replace arginine-61 with
threonine-61 would likely correct the lipid binding preference that leads to the toxic effects of ε4
(Mahley et al., 2006).
The cholesterol efflux regulatory protein (CERP), also known as ABCA1 (ATP-binding cassette
transporter), is a major regulator of cholesterol and phospholipid transport and homeostasis. ABCA1
is promotes lipidation of ApoE, which allows it to efficiently bind Aβ and facilitate Aβ uptake through
LRP1. Studies of APP mice with an ABCA1 deficiency have increased amyloid deposition in the brain,
paralleled by decreased levels of ApoE. Microdialysis assays indicate that ABCA1 deficiency
significantly decreases the clearance of Aβ in ApoE-ε4-expressing mice, while there are no significant
effects of ABCA1 on Aβ clearance in ApoE-ε3-expressing mice (Liu et al.).
Liver X receptor (LXR) is a nuclear receptor transcription factor that plays an essential role in
regulation of cholesterol, triglycerides, fatty acids, and glucose homeostasis. Bexarotene, which
activates LXR/retinoid X receptor (RXR) heterodimers, increases the expression of ApoE, reduces
amyloid plaques, and improves memory performance in an AD mouse model harboring the ε4 allele
University of Southern California
47
(Cramer et al., 2012). Likewise, treatment with natural LXR agonists, such as the algae derived
Fucosterol has shown to induce transcriptional activation of ApoE, and the cholesterol efflux
regulator and HDL mediator, ABCA1, without increasing cellular accumulation of triglycerides, a side-
effect of other LXR agonists (Sharman et al., 2002). LXR agonists have also been shown to increase
the activities of mitochondrial mediated antioxidant enzymes, such as SOD and glutathione (Lee et
al. 2003).
Polymorphisms in the Translocase of the Outer Mitochondrial Membrane (TOMM40) gene, which
is in linkage disequilibrium with APOE, are associated with late-onset AD risk and age of disease onset
(Roses et al., 2010). The expression of both apolipoprotein E and TOMM40 are significantly increased
with AD. APOE expression and TOMM40 expression mRNA levels in the temporal and occipital
cortexes are reportedly higher in carriers of the very long TOMM40 variant (523 poly-T), compared
to those with the short variant, both in normal individuals and those with AD. This increased
TOMM40 expression leads to an increased risk of AD in ε3/ε4 carriers and not ε3/ε3 carriers (Roses
et al., 2010). However, it is unclear what this means in terms of TOMM40 protein levels. For instance,
although both show alteration associated with AD pathology, mRNA of APOE expression does not
always correlate with ApoE protein levels (Soares et al., 2012). An additional variant at the
TOMM40/APOE gene cluster (rs2075650) has been significantly associated with LDL and C-reactive
protein (CRP) levels, with a trend toward association with HDL and triglyceride levels, proteins that
have all been associated with modulation of risk for heart disease and AD (Soares et al., 2012).
Beyond TOMM40, most genes involved in the APOE gene cluster (i.e. APOC1, APOC2, and APOC4) have
been associated with risk for AD. For instance, apoC2, which serves as a measure of triglyceride rich
lipoproteins, has been found to co-localize with Aβ (Middelberg et al., 2011; Shachter et al.). Similar
findings of increased triglycerides with increased apoC1 have been found, along with strong
associations with risk of dementia and insulin resistance (Medeiros et al., 2004; Serra-Grabulosa et
al., 2003). Taken together these findings indicate that there are structural and widespread metabolic
University of Southern California
48
differences between the APOE isoforms. These widespread differences play a prominent role in
conferring risk for Alzheimer's disease.
2.4. ApoE, Beta Amyloid and Tau
APOE is the best-characterized amyloid-β (Aβ) chaperone, with ApoE containing an Aβ-binding
motif around region 230-243 (see Figure 1) (Bu, 2009; Garai et al., 2011). Functional APOE can
induce intracellular Aβ degradation through trafficking of amyloid to lysosomes. However, the three
isoforms differ in their binding affinities. Specifically, the direct interaction between Aβ and ApoE
facilitates Aβ oligomerization in rank order ε4> ε3> ε2 (LaDu et al., 2012; Manelli et al., 2004).
As lipidated forms of APOE-ε4 bind poorly to soluble Aβ, it is suggested that lipoproteins may
contribute in other ways to the dynamics of Aβ clearance along the cerebral vasculature (LaDu et al.,
1995). Indeed, APOE enables cerebrovascular integrity through the cyclophilin A-NFκB-MMP
pathway. Astrocytes in ε3 mice suppress the cyclophilin A-NFκB-MMP pathway, while in ε4 mice they
do not, indicating the ε3 isoform may have greater vascular protective capabilities (Bell et al., 2012).
Further, APOE knockout mice experience progressive blood-brain-barrier (BBB) breakdown (Hafezi-
Moghadam et al., 2007).
APOE-ε4 has been shown to increase Aβ deposition and Aβ oligomer formation. Additionally,
studies of PiB binding, a technique used to image amyloid, have shown that ε4 increases brain
amyloid burden in a dose-dependent manner in cognitively normal individuals and individuals in the
prodromal stages of AD (Reiman et al., 2009; Villemagne et al., 2011). Furthermore, ε4 is associated
with greater membrane disruption and lysosomal leakage in the presence of Aβ compared to ε3.
Interestingly, reductions in glucose metabolism have been shown to correlate better with APOE
genotype than amyloid load, showing that genotype contributes to reduced glucose metabolism in
aging independently of Aβ. Indeed, APOE-ε4 individuals show less variation in CSF Aβ levels from a
cognitively normal status to mild cognitive impairment (MCI), believed to be a precursor to AD, than
University of Southern California
49
do non-carriers (Apostolova et al., 2014).
Mice with an ε4 allele insertion show increased tau accumulation, the main component of
neurofibrillary tangles seen within somatodendritic and intra-axonal compartments in those
diagnosed with AD (Bennett et al., 2013). Moreover, ApoE-ε3 is able to bind tau to prevent tau
phosphorylation and assembly into paired-helical filaments, while ε4 is not (Strittmatter et al., 1994).
Additionally, ε4 carriers consistently show higher tau, p-tau, and tau/Aβ-42 ratios (Mattsson et al.,
2009). Although these associations are present in both sexes, females show a greater association with
CSF tau, pointing to possible sex differences along the disease trajectory (Damoiseaux et al., 2012).
Future studies involving the use of the lipid-probe theta-toxin, which binds cholesterol, in 3xTgAD
and App-NL-F mice, may help further elucidate the sex interactions between ApoE and AD
pathogenesis (Ohno-Iwashita et al., 2004).
Under conditions of oxidative and excitotoxic stress or neuronal damage, ApoE is produced by
neurons (Xu et al., 2006). Due to the domain interaction of ε4, when ApoE is synthesized in this state
it undergoes proteolytic cleavage to a much greater extent than ε3 (Harris et al., 2003), with resulting
truncated C-terminal fragments entering the cytosol where they become neurotoxic (Brecht et al.,
2004) as well as toxic to mitochondria (Chang et al., 2005). Importantly, during oxidative
phosphorylation, mitochondria are one of the primary sources of oxidative stress [92-93]. In line with
this, post-mortem microarray analysis of brains of middle-aged cognitively normal APOE-ε4 and
APOE-ε3 carriers without pathological evidence of AD have revealed differences in gene transcripts
that are associated with both mitochondrial function and AD, including complex-I gene NADH
dehydrogenase, as well as genes associated with insulin signaling, such as CTNNB1 (Conejero-
Goldberg et al., 2011). Interestingly, no differences in transcripts directly involved in amyloid
processing were found between the groups. Support for the early role of mitochondrial bioenergetics
in the progression of AD is strengthened by evidence that young ε4 carriers show reductions in
cytochrome-c oxidase (COX) activity, a measure of oxidative metabolism, in brain tissue from the
University of Southern California
50
posterior cingulate, even in the absence of fibrillar amyloid (Crivello et al., 2010). Likewise, AD
patients with the ε4 allele show higher levels of hydroxyl radicals than AD patients without an E4
allele and reduced levels of glutathione peroxidase, an enzyme with antioxidant capacity that is
produced in mitochondria (Ihara et al., 2000; Lee et al., 2003). These differences may partially be
explained in that mass spectrometry analysis have revealed the methionine-108 region of ApoE is
considerably more reactive toward free radical labeling in ε4 (Gau et al., 2011). APOE-ε4 carriers also
have higher levels of isoprostanes, a marker of oxidative stress, than non-ε4 carriers when matched
for cholesterol levels (Ramassamy et al., 1999). Finally, decreases in mitochondrial respiratory
complexes are seen in neurons of ε4 mice and not ε3. In fact, treatment of these cells with GIND25, a
small molecule that disrupts the detrimental ε4 structural domain interaction has shown to restore
mitochondrial respiratory complex IV levels to a level similar to ε3 neurons (Conejero-Goldberg et
al., 2011). A similar compound, IAH, is currently being studied to elucidate its therapeutic potential
in AD patients (Chen et al., 2011b). These findings support the early role of energy metabolism in the
progression of AD and suggest that compromises in mitochondrial dysfunction may precede plaque
formation in ε4 carriers. It is likely that amyloid fibrils later amplify this mitochondrial dysfunction.
Taken together, these findings indicate that the structural differences in ApoE between the
isoforms result in a cascade of physiological effects. These differences result in both Aβ-dependent
and Aβ-independent changes that serve to effect overall risk for Alzheimer's disease. In the ε4
isoform, these changes include an increased avidity to bind pro-atherogenic lipoproteins, alterations
in the bioenergetic profile, reduced ability for ApoE to clear Aβ, and increased accumulation of tau.
Furthermore, the ApoE ε4 isoform leads to an increased production of reactive oxygen species (ROS)
and therefore increased neurotoxic and toxic mitochondrial environments under conditions of
oxidative stress.
2.5. APOE-ε4 Specific Risks for Alzheimer’s Disease
University of Southern California
51
Among persons diagnosed with AD, up to 60% carry at least one ε4 allele (Farrer et al., 1997).
APOE ε4 exerts its maximal effect on AD-risk by the early 70's (Jarvik et al., 1995), with a reduction
in risk after age 85 in both sexes (Farrer et al., 1997). Similarly, the accelerating effect of ε4 on rates
of decline diminishes with advancing disease stages, such that APOE genotype has no significant
effect on overall disease duration. The APOE-ε4 allele accounts for as much as 50% of the genetic
attributable AD risk (Farrer et al., 1997; Raber et al., 2004). In persons diagnosed with AD, APOE-ε4
homozygotes carriers are diagnosed at a mean age 68 years whereas in ε4 heterozygotes mean age
is 76 years, and 84 years in ε4 noncarriers (Liu et al., 2013). The APOE ε4 gene dose effect on risk and
age of AD onset indicates that APOE ε4 dramatically increases risk of AD and is associated with an
earlier age of onset.
In women, carrying one APOE ε4 allele shifts the AD risk curve five years earlier, while two alleles
shifts the curve to 10 years earlier in both women and men (Noguchi et al., 1993). In addition, the
risk odds ratio for AD in women with one allele is 3.5-4, while it is 12-15 in APOE ε4 homozygote
women and men (Coon et al., 2007; Corder et al., 1993).
APOE may serve as an interesting example of antagonistic pleiotropy – a gene that confers
advantage in one period of life but later presents as disadvantage. Specifically, young individuals with
the ε4 allele show greater neural efficiency on tasks of episodic memory as measured through a more
rapid decline in blood-oxygen-level dependent (BOLD) response over learning trials and therefore
more efficient use of memory resources (Mondadori et al., 2007). They also show better performance
in speed of processing, attention and verbal fluency (Marchant et al. 2010). This cognitive benefit
extends into middle-age (i.e. 45-55) on measures of attention and prospective memory (Evans et al.
2014). However, this benefit does not extend to verbal memory at this age. In fact, a study comparing
individuals 45-57, and 58-68, and age-matched controls found that the 45-57 ApoE-ε4 group
performed better on verbal memory tests than the control group, while the 58-68 group showed
impairments. At this later age, ε4 older adults also show increased fMRI activation in the association
University of Southern California
52
cortex during memory challenge and during tasks involving verbal fluency tests compared to non-ε4
age-matched controls (Bookheimer et al., 2000; Smith et al., 2002), as well as altered connectivity in
the default mode network. Further, stressful life events have greater impact on cognition of ε4
individuals of Caucasian-American descent compared to African Americans (Sheffler et al., 2014).
Despite the potential early benefits, older ε4 individuals show evidence of early reduction in
entorhinal cortex volume (Shaw et al., 2007), greater rate of hippocampal volume loss (Moffat et al.,
2000), and early rise in rate of myelin breakdown (Bartzokis et al., 2007). These individuals also
show characteristic patterns of parietal, cingulate, and temporal hypometabolism, consistent with an
AD-like pattern in brain glucose utilization (Reiman et al., 2004, 2005; Sheline et al., 2010). When
stratifying by APOE genotype, positron emission tomography (PET) imaging is predictive of AD
conversion with 100% sensitivity and 90% specificity (Mosconi et al., 2004). Conversely, APOE ε2
carriers show evidence of protective mechanisms in the brain, particularly in regard to white matter
compared to ε3 carriers. Specifically, they exhibit higher fractional anisotropy, a measure of fiber
integrity, and lower radial diffusivity, a measure of myelin integrity, in the posterior cingulate and
anterior corpus callosum, regions relevant to the progression of AD (Chiang et al., 2012; Holtzman et
al., 2012).
3. Sex Differences in APOE Genotype and Risk of Alzheimer’s
3.1. Evidence for Greater Pathology and Accelerated Degeneration Rates in Females relative to Males:
Impact of APOE-ε4.
As cholesterol biosynthesis is a precursor to the production of hormones, a physiologic
relationship between apolipoprotein E and sex steroids have been established (Struble et al., 2007).
Postmenopausal women constitute >60% of the affected Alzheimer population and are those who
will bear the greatest burden of the disease (Aging, 2015; Brookmeyer et al., 1998; Brookmeyer et
al., 2007; Prince et al., 2015). Twenty years ago Farrer and colleagues reported a sex difference in the
University of Southern California
53
lifetime risk of AD in women (Farrer et al., 1997). In his seminal report, women with a single copy of
the ApoE4 allele was sufficient to increase disease risk associated with two copies of the ApoE4 gene
in men (Farrer et al., 1997). This finding was confirmed a year later in a subsequent report by Payami,
Schellenberg and colleagues (Payami et al., 1994) who found that ApoE4 heterozygote men had lower
risk than ApoE4 homozygotes; there was not a significant difference between ε4 heterozygote males
and those without ε4. In contrast, ε4 heterozygote women had the same significant twofold increased
risk as homozygote men (Payami et al., 1994). Multiple studies indicate that the APOE-ε4 risk for AD
is greater in women, especially in heterozygous individuals carrying one APOE-ε4 allele (Altmann et
al., 2014; Damoiseaux et al., 2012; Farrer et al., 1997; Payami et al., 1994). Women with one copy of
the APOE-ε4 allele have a 4-fold increase in the risk of AD, (Altmann et al., 2014; Farrer et al., 1997;
Payami et al., 1994) whereas women and men with two copies of the ApoE4 allele exhibit a 15-fold
increase in risk (Altmann et al., 2014; Farrer et al., 1997; Payami et al., 1994) and a significantly lower
age of onset compared with AD patients carrying ApoE2 or 3 alleles. Consistent with greater AD risk,
Barnes, Bennet and colleagues found an even greater sex difference in the impact of pathology and
risk of AD (Liu et al., 2013). Each additional unit of AD pathology was associated with a nearly 3-fold
increase in the odds of clinical AD in men compared with a more than 22-fold increase in the odds of
clinical AD in women (Barnes et al., 2005).
Women diagnosed with mild cognitive impairment progressed with faster rates of cognitive
decline than men (Lin et al., 2015). This interaction has been observed in association with C-reactive
protein levels over a 10-year period, such that CRP variability is significantly associated with
cognitive decline for women ε4 carriers, but not men ε4 carriers (Metti et al., 2014). Further, women
had an accelerated rate of cognitive decline relative to men. The sex difference effect was greatest in
ApoE ε4 female carriers. Overall, women ε3/ε4 carriers often show faster age-related decline and
greater deterioration of cognition than elderly ε3/ε4 men (Farrer et al., 1997). Indeed, as measured
by rates of atrophy and clinical presentation, MCI women decline more rapidly (Smith et al., 2002)
University of Southern California
54
while at the same time experiencing a longer survival rate following diagnosis in ε3/ε4 and ε4/ε4
women (Corder et al., 1995).
In a study of 131 participants, decreased connectivity in the default mode network in healthy
older APOE-ε4 carriers was identified, with a significant sex interaction in the precuneus, a major hub
for the default mode. Specifically, female ε4 carriers show significantly reduced connectivity
compared to female ε3 carriers and male ε4 carriers in the anterior cingulate cortex. Additional
studies have revealed that cognitively normal female ε4 carriers show greater reduction of functional
connectivity in the precuneus compared to ε4 males. This region is structurally connected to the
medial temporal lobe, the initial site of tau pathology in AD. This region also shows reduced glucose
metabolism in early AD and asymptomatic ε4 carriers (Reiman et al., 1996). Furthermore, MCI female
ε4 carriers show more prominent phenotypic features than their male counterparts, such as reduced
hippocampal volumes and worse cognitive scores (Fleisher et al., 2005). Similar to the differences in
cognitively normal ε4 carriers, female ε4 MCI patients show decreased precuneus activity compared
to non-ε4 women and ε4 men. In MCI, this decreased connectivity extends to the posterior cingulate
(Reiman et al., 1996).
By the age of 40, 15% of ε4 homozygous cognitively normal individuals are amyloid positive, a
frequency which occurs 10-15 years earlier than for ε2/ε4 or ε3/ε4 individuals. By 90, more than
80% of ε4 are amyloid positive, although no sex differences are present (Jack et al., 2015). Given the
lack of sex differences in amyloid by ε4, this supports the role of tau in mediating sex differences. In
line with this, female ε4 carriers with MCI show greater CSF tau and tau/Aβ ratios compared to ε4
males with MCI (Payami et al., 1994). Similarly, ε4 women with mild AD are at a higher risk of having
both neurofibrillary tangles and Aβ plaques than ε4 men with mild AD (Corder et al., 2004). These
findings support the idea that women ε4 carriers have a greater extent of AD pathology as measured
by CSF and pathology analysis at autopsy. While other studies have failed to find a sex difference in
AD risk by APOE genotype, this can be explained in that there appears to be a differential effect on
University of Southern California
55
risk by carrying one ε4 allele compared to individuals who are homozygous for ε4, such that women
with one allele are at a greater risk compared to ε3/ε3 women, while men have no increased risk
with just one ε4 compared to ε3/ε3 men.
3.2. APOE Genotype and Response to Estrogen
On average, ApoE ε4 negative women have better cognitive performance than ApoE ε4 positive
women (Yaffe et al., 2000). Further, ApoE ε4 positive women were at greater risk for cognitive
impairment (Yaffe et al., 2000). The impact of estrogen therapy in postmenopausal women is
complex. ApoE ε4 negative women receiving estrogen or hormone therapy had the highest level of
cognitive performance, whereas women positive for ApoE ε4 receiving estrogen or hormone therapy
performed worse than ApoE ε4 carriers not receiving therapy (Yaffe et al., 2000). Notably, treatment
with tamoxifen, a selective estrogen receptor modulator and estrogen receptor antagonist,
ameliorated cognitive deficits observed during the menopausal transition, especially in ApoE ε4
women (Newhouse et al., 2013).
The contradictory effects of estrogenic agonists and an estrogen receptor antagonist in ApoE ε4
carriers suggest that the female ApoE ε4 brain is different from the ApoE ε3 carriers. We propose
that one fundamental difference between the ApoE ε4 brain and Apoe ε2 and ε3 brains is in reliance
of the ApoE ε4 brain on ketone bodies as a bioenergetic fuel. Reliance of the ApoE ε4 brain on ketone
bodies as a bioenergetic fuel puts that brain at risk for use of its own white matter as fuel (Klosinski
et al., 2015). In the female brain, estrogen activates the system biology of glucose metabolism while
simultaneously suppressing the ketogenic system in brain thereby promoting brain reliance on
glucose as its primary fuel to generate ATP (Brinton et al., 2015; Ding et al., 2013a; Ding et al., 2013b;
Rettberg et al., 2014; Yao and Brinton, 2012; Yao et al., 2009; Yao et al., 2010; Yao et al., 2011; Yao et
al., 2012; Yin et al., 2015). If the ApoE ε4 brain is a dual fuel dependent brain, being dependent upon
glucose and ketone bodies, then suppression of the ketogenic system in the ApoE ε4 brain would
University of Southern California
56
result in reduction in a critical fuel to generate ATP. In the case ApoE ε2 and ε3 carriers, estrogenic
activation of the glucose metabolism system and suppression of the ketogenic system is beneficial as
estrogen is promoting and sustaining brain utilization of the primary fuel system glucose. In contrast,
in a dual fuel dependent brain, one dependent upon both glucose and ketone bodies, suppression of
the ketogenic pathway would put that brain at metabolic risk. Reliance of the ApoE ε4 brain on ketone
bodies as a bioenergetic fuel is consistent wtih evidence from multiple laboratories demonstrating
that ApoE ε4 carriers are glucose hypometabolic prior to decline in cognitive function (Langbaum et
al., 2010; Mosconi et al., 2004; Reiman et al., 2005).
Estrogen receptors play a key and contrasting role in regulation of ApoE gene and protein
expression (Wang et al., 2006) and risk of AD (Ryan et al., 2014). ER up-regulated ApoE mRNA and
protein expression whereas in contrast, ER down-regulated ApoE mRNA and protein expression
(Wang et al., 2006). These data suggest that use of ER-selective ligands could provide therapeutic
benefit to reduce the risk of AD by increasing ApoE expression in ApoE ε2 or ε3 allele carriers and
decreasing ApoE expression in ApoE ε4 allele carriers. Polymorphisms in both estrogen receptor
alpha (rs4986938) and estrogen receptor β (rs2234693) have been associated with increased risk of
dementia and AD (Ryan et al., 2014).
Estrogen therapy may prove beneficial in APOE-ε4 women who show a favorable response by
increasing ABCA1 production. In fact, estrogen has been shown to increase ABCA1 mRNA expression
in mice as well as postmenopausal women receiving hormone therapy (Darabi et al., 2011;
Srivastava, 2002). Competitive interactions between estrogen receptor β and LXRβ for
transcriptional coactivator RAP250 likely modulate estrogens effects on ABCA1 levels (Srivastava,
2002).
Postmenopausal APOE-ε4 positive women age 49-69 who discontinued their hormone therapy
regimen exhibited telomere shortening, an index of biological aging, to a greater extent than women
who do not carry an ε4 allele or ε4 women not receiving hormone therapy (Jacobs et al., 2013).
University of Southern California
57
Estrogen exposure of mitochondria is implicated in regulation of mitochondrial functioning and
activates manganese superoxide dismutase (MnSOD) antioxidant activity (Pedram et al., 2006).
Conversely, APP and Aβ disrupt mitochondrial function and this is partially mediated by Tom40
interactions (Gabriel et al., 2003). The Tom40 protein is an active channel for protein sorting, and is
crucial to healthy mitochondria (Gabriel et al., 2003). As previously mentioned, the TOMM40 gene is
in linkage disequilibrium with APOE (Roses et al., 2010). Notably, estrogen treatment has been shown
to increase both ApoE and Tom40 levels through activation of estrogen receptors (Srivastava et al.,
1997). Estrogen treatment also modulates the ApoE receptor LDL related receptor protein (LRP1)
(Cheng et al., 2007). Estrogen treatment in ovariectomized mice also causes increases in LRP, an
ApoE binding protein, in the hippocampus and neocortex (Ivanova et al., 2013). Along these lines,
ApoE synthesis is required for estrogen-induced neuroprotection and neurite outgrowth (Cheng et
al., 2007; Nathan et al., 2004; Struble et al., 2007) and is lost in the presence of ε4 (Nathan et al.,
2004).
Collectively, these data indicate a complex interaction between the estrogenic and ApoE systems
that deserve greater investigation at the systems biology level of analysis and particularly regarding
the interaction between these two systems during aging.
3.6. Lack of Sex Differences in Young May Support a Role of Hormone Loss During Menopause.
Infants harboring the APOE-ε4 allele have lower gray matter volume and lower white matter
myelin water fractions, indicating reduced myelin integrity in AD-relevant regions compared to non-
carriers. These include the precuneus, posterior cingulate, lateral temporal, and occipitotemporal
regions (Dean et al., 2014). These infants also present with greater water fraction and gray matter
volume in frontal regions, with no observable sex differences. Interestingly, they have greater myelin
water fraction in regions that myelinate later and reduced water fraction in regions that myelinate
early, potentially indicating a condensed white matter development trajectory. As AD is first
University of Southern California
58
associated with myelin loss in regions that myelinate late in development, a reduction in white matter
development would put APOE-ε4 carriers at greater risk. In line with this, differences in APOE
genotypes are associated with a steeper rate of myelin breakdown in late-myelinating frontal regions,
and APOE has shown to mediate myelin maintenance and repair mechanisms (Bartzokis et al., 2006).
While there are observable differences in brains of APOE-ε4 carriers quite early in development, the
impact of sex on APOE related changes in white matter or gray matter are not always found in studies
looking at children or adolescents (Dean et al., 2014; Shaw et al., 2007). This evidence suggests that
sex differences conferring greater risk of AD in ε4 women could be synergistic with events that occur
during mid-life or later.
Interestingly, ApoE concentration shows sex-specific alterations consistent with the time-scale
of puberty, as well as menopause in women (see Figure 3). Specifically, females show higher ApoE
concentrations than males until around age 17, the age of puberty completion in women (Lee et al.,
2001; Vincent-Viry et al., 1998). These differences are then reversed around the average age of
menopause (Burger et al., 2002; Schiele et al., 2000). Importantly, this perimenopausal transition is
marked by a bioenergetic shift in the brain to utilizing ketone bodies as fuel (Brinton et al., 2015;
Cunnane et al., 2011; Yao et al., 2010) with estrogen loss post-menopause resulting in decreased
metabolic function in the brain (Li et al., 2014; Santoro and Sutton-Tyrrell, 2011; Yao et al., 2012).
3.7. Evidence for Protection of ε2/ε2 and ε3/ε3 Genotypes in Women.
Over 87% of centenarians are ε2/ε3 or ε3/ε3 among the majority of populations studied,
including those in France, Japan, Spain, Italy, and Finland (Asada et al., 1996; Blanché et al., 2001;
Castro et al., 1999; Garatachea et al., 2014). While the ε3 allele is the most common, the prevalence
among this age group is greater than in the general population. Notably, interactions with ACE
polymorphisms, shown to impact cardiovascular risk, likely impact the role of APOE in promoting
longevity (Schachter et al., 1994). Although ε2 is generally believed to be protective against AD,
University of Southern California
59
evidence exists to suggest this is true in ε2/ε3 individuals of both sexes, but only in ε2 homozygous
females (Sheffler et al., 2014; Yao et al., 2012). In line with the role of ε2 in promoting longevity, one
study of Swedish individuals found that ε2 was associated with a reduced risk of AD only in females
until age 85, when the allele stops being protective (Qiu et al., 2004). Additionally, there is a
decreased mortality rate in ε2 women, while this is not always observed in men (Rosvall et al., 2009).
However, it is important to note that carriage of the ε2 allele is associated with a greater risk for
hyperinsulinemia and diabetes in both sexes, both of which increase the risk for Alzheimer's (Kuhel
et al., 2013; Satirapoj et al., 2013).
The APOC1 gene, which is contained in the APOE gene cluster, has also been associated with
longevity. Specifically, one SNP in this gene (rs4420638) has been identified as protecting against AD
and promoting resilience. Interestingly, this SNP is in linkage disequilibrium with APOE and shows
sex specific differences (Nebel et al., 2011). In fact, APOC1 allele and genotype frequencies have been
identified as significantly different in elderly women age 84 and older compared to younger women
(Galinsky et al., 1997).
CSF analysis of people with MCI revealed higher levels of total tau in female ε4 carriers, and
lowest levels of phosphorylated tau in ε3 females over ε3 males, potentially pointing to a protective
factor in ε3 females over males (Altmann et al., 2014). Additionally, cognitively normal ε3
homozygote females had higher CSF amyloid than ε3 homozygous males (Altmann et al., 2014).
Further, ApoE serum concentrations continue to rise with age in ε3 women to a greater extent than
in ε4 women or ε3 men (Schiele et al., 2000). Given the role of APOE in Aβ sequestration and
clearance, this increase likely provides for protection in ε3 women (Altmann et al., 2014).
Unfortunately, most studies assessing APOE genotype effects on the brain and CSF levels group all
non-ε4 carriers together, making it difficult to more adequately assess whether or not the ε3 allele is
protective in homozygous women.
In the latest study of over 8,000 individuals by Altman and colleagues comparing the risk of ε4
University of Southern California
60
status by sex, an increased relative risk for female ε4 carriers (both heterozygous and homozygous)
was identified over female ε3 homozygotes in progressing from normal cognition to MCI, or from MCI
to AD. The absolute risk values were higher for male ε4 carriers with 22.4% of ε4 heterozygous male
carriers progressing from normal cognition to MCI, whereas 19.8% of female ε4 heterozygous
carriers did (Altmann et al., 2014). Rather than indicating that female ε4 carriers are at a greater risk,
this might indicate that the impact of having an ε4 is greater in females, likely because there is some
protective mechanism for the ε3 females. Indeed, the ε3 females showed the lowest risk of all groups.
Specifically, only 13.9% of ε3 homozygous females progress from normal cognition to MCI, while
20.7% of males in this same category progress; additionally, 24.2% of females progress from MCI to
AD, while 28% of males progress, indicating protection in the female ε3 carriers (Altmann et al.,
2014). Likewise, there is a higher proportion of male APOE ε3 homozygotes than females who have
progressed to AD in the NACC Uniform Data Set, a database which involves data collected from 34 AD
centers across the US (N=11,654) (Beekly et al., 2007; Morris et al., 2006).
3.8. Sex Differences in Cardiovascular Risk by APOE Genotype and Effects on Vascular Dementia.
In a Framingham Heart Study of 3,413 participants, age-adjusted period prevalence of
cardiovascular disease (CVD) was related to APOE genotype with a higher rate for men than women
(18.6% in the ε4 group for men and 9.9% for ε4 women) (Lahoz et al., 2001). Notably, while the ε2
allele was protective for women (4.9%), this protection was not seen in males with the ε2 allele
(18.2%), indicating that the presence of either the ε2 or ε4 alleles in men is associated with greater
CVD risk. Furthermore, the overall odds of CVD for men within the ε2 group was 1.94-fold greater
than that for men within the ε3 group, while ε2 women were at a .91 reduced risk compared to the
ε3 women (Lahoz et al., 2001). In a separate Framingham Heart Study of 7901 participants, mortality
due to cardiovascular disease that was 6 times higher in men than women at ages 45-54 (Chêne et
al., 2015). The increased period prevalence in risk for heart disease for men was still evident when
University of Southern California
61
comparing the lowest risk group of males (i.e. ε3/ε3) to the highest risk group of women (i.e. ε4/ε4).
However, after the age of 65, the sex difference in risk between males and females was less than 2-
fold.
The incidence of vascular dementia is higher in men than women (Ruitenberg et al., 2001). In
fact, results from a recent study examining sex differences by APOE genotype, couldn’t rule out
whether greater vascular pathologies could have been present in the male participants. While female
ε4 carriers showed greater tau levels than ε4 males with MCI, the differences are in line with a
potential vascular dementia in the males, as vascular dementia presents with similar CSF Aβ but
lower total tau and phosphorylated-tau levels than AD (Ruitenberg et al., 2001). Importantly,
participants presenting with potential vascular disease are often not included in studies of AD. While
this is beneficial in that it limits potential confounds, there is also a growing understanding of the
connection between AD and cerebrovascular health. Given that men are at a greater risk for heart
disease and other cardiovascular events, especially towards the young-old age spectrum (e.g. 65-74),
understanding this group would shed light on the sex differences in ε4 carriers.
Although cardiovascular deaths in late mid-life cannot explain all differences in dementia rate by
sex, a substantial part of the difference in dementia risk between females and males could derive
from incidence of cardiovascular disease before the age of 65. At mid-life, lifetime risk of AD is not
different for men and women whereas after midlife for women and men and subsequent to
menopause in women, from 65 years of age onward, women have a 2-fold greater lifetime risk of AD
then men. One could speculate that lower risk of AD in males over the age of 65 is due to survivor
effects in the cohorts. This hypothesis will be testable at the population level as males gain longevity
comparable to females in the near future (World Health Organization, 2015).
University of Southern California
62
FIGURE 2-2
Sex-specific incidence estimates of Alzheimer's disease per 1,000-person years. Obtained with data from
the Cache County Study (Ruitenberg et al., 2001). Data reported in Ruitenberg et al., (2001) indicate that men
are at greater risk than women for developing earlier onset AD. However, this sex difference is reversed by age
75, with women at a 2-fold greater risk for AD, thereafter.
Men with ε2/ε3 alleles exhibit higher insulin levels compared to controls, indicative of a state
of insulin resistance that has been associated with impaired clearance of atherogenic triglycerides.
In women, estrogen therapy improves insulin sensitivity, potentially providing protection prior to
menopause (Lahoz et al., 2001). Men also show greater levels of SREBP2 expression, a protein that
interacts with ApoE to further regulate lipid homeostasis (De Marinis et al., 2008). Moreover,
estrogen can upregulate ApoE gene expression by increasing ApoE mRNA (Srivastava et al., 1997). In
fact, female rats also show higher expression levels than males (Brussaard et al., 1997). Given the
impact of estrogen on ApoE levels, this higher expression would allow for greater protection in terms
of both cognition and heart health in females until menopause, when estrogen levels decline
dramatically. This increase likely proves protective in ε2 women against CVD, while detrimental in
ε4 women towards the risk of AD due to the pathological structure and function of ε4.
University of Southern California
63
FIGURE 2-3
ApoE serum concentrations by sex and age (N=6,934). From data collected in the ApoEurope project,
obtained with permission from (Schiele et al., 2000). Data indicate that men have greater apoE concentrations
than women until age 50-54, consistent with the timing of menopause. While men experience progressive
declines in apoE concentrations following this age, apoE concentrations rise in women. As low serum levels of
apoE are associated with an increased risk for AD, this tipping point might explain the discordant findings in
risk by sex, such that men are at risk due to a reduced production of apoE, while ε4 women produce an
overabundance of as isoform with impaired function.
3.9. Evidence for Worse Effects in Men or No Sex Difference.
Most intervention studies that include a separate analysis of possible therapeutic benefit in ε4
individuals, fail to consider that carrying an ε4 allele is associated with an earlier age of diagnosis and
accelerated pathology. Therefore, age-matched studies often yield negative results in ε4 individuals.
Similarly, AD shows an altered trajectory in time towards diagnosis between the sexes (see Figure
2); (Miech et al., 2002) likely blurring the sex interactions in ε4 carriers. For instance, the association
between APOE genotype and cognitive decline is significant only in women over the age of 70
(Mortensen and Høgh, 2001). Additionally, ε2 participants are often excluded from analysis, though
studies that have included them indicate that having the ε2/ε4 allele is associated with an increased
risk for AD compared to ε3/ε3 (Blalock et al., 2003; Sharman et al., 2002; Yao et al., 2012). Given the
differential impact of this genotype on women and men for cardiovascular health, it is likely that
surviving men would exhibit an increased risk for AD with this genotype compared to females (Kaerst
et al., 2013). However, there is likely to be a selection pressure on men ε4 carriers, such that those
University of Southern California
64
who overcome the prominent impact of this allele on cardiovascular risk likely possess other factors
that compensate for the presence of an otherwise 'frail' genotype.
Low plasma ApoE is associated with a decrease in hippocampal volume, and shows less of a sex
difference in concentration levels in ε4 individuals than between the other genotypes (Teng et al.,
2015). However, normal ε4 men show greater AD-pathology and worse memory performance than
normal ε4 women, pointing to an earlier incidence of vulnerability (Jack et al., 2015). Notably, years
of life lost in individuals with at least one ε4 allele are 2.5-fold greater for men than for ε4 women
(Rosvall et al., 2009). Furthermore, there is some evidence of a sex by APOE gene dose interaction,
such that the significant effects of ε4 on risk for decline in episodic memory are generally less
dependent of zygosity in females than males (Lehmann et al., 2006). Likewise, ε4/ε4 men are at
higher risk of AD than ε4/ε4 women (Farrer et al., 1997). For example, in a recent study, a greater
proportion of ε4/ε4 men progressed from normal cognition to a state of MCI than did ε4/ε4 women.
Similar trends were seen for individuals starting as MCI at baseline and progressing to AD (Altmann
et al., 2014). Additionally, episodic memory as measured through the Kendrick Object Learning Test
has been reportedly worse in normal ε4 homozygous men than women (Lehmann et al., 2006).
Furthermore, studies on the effect of APOE-ε4 homozygosity in men with MCI have shown that ε4
men have smaller hippocampal volume than ε4 homozygous females (Fleisher et al., 2005).
4. Conclusion
Collectively, the data indicate a complex interaction between the trial of greatest risk
factors for Alzheimer’s disease, age, APOE genotype and sex. We provide a table outlining the
intricacies discussed herein (Table 1). The complexity of this interaction remains to be fully
and specifically characterized. While investigating the interaction between three systems
across multiple transition states of aging is incredibly challenging at the basic and clinical
University of Southern California
65
levels of analysis, it is feasible using a big data bioinformatic approach. Clearly there are sex
differences in risk of AD that are modified by APOE genotype. But even at this level there is
a fair degree of variability that is not explained by either APOE genotype or by sex, suggesting
that the systems biology of aging is a driving factor.
TABLE 2-1
Summary of Sex Differences by APOE Genotype
Genotype Sex Findings
ε2/ε2
Males More severe lipidemia and atherosclerosis due to high plasma apoE (Lahoz et al., 2001)
Females
Reduced AD risk in women <85, increased after age 85 (Castro et al., 1999)
General
Increased prevalence of hyperlipidemia and cardiovascular disease (Mahley and Rall Jr, 2000)
ApoE concentrations are generally the highest in this group, protecting from AD (Rasmussen et al., 2015; Schiele et al.,
2000)
ε2/ε3
Males Increased AD risk due to higher incidence of insulin resistance (Lahoz et al., 2001)
Females
Estrogen and ERT improve insulin sensitivity, compensating for the increased risk in males (Lahoz et al., 2001)
Reduced cardiovascular risk and AD risk in women <85, increased after age 85 (Lahoz et al., 2001) (Castro et al., 1999)
General
Associated with increased longevity and protection from AD (Asada et al., 1996; Blanché et al., 2001; Garatachea et al.,
2014; Li et al., 2014; Santoro and Sutton-Tyrrell, 2011)
ε3/ε3
Males
Potentially protective against AD, although at higher risk than genotype-matched females (Altmann et al., 2014)
Females
MCI patients show lowest levels of p-tau, associated with delayed disease progression (Altmann et al., 2014)
More favorable CSF amyloid profiles than ε3/ε3 men (Altmann et al., 2014)
Lowest AD risk compared to males and all other APOE genotypes in females (Altmann et al., 2014)
Increases in apoE concentration with age provide protection from AD risk (Altmann et al., 2014; Schiele et al., 2000)
Associated with increased longevity and protection from AD (Asada et al., 1996; Blanché et al., 2001; Garatachea et al.,
2014; Li et al., 2014; Santoro and Sutton-Tyrrell, 2011)
General
Protective against AD (Altmann et al., 2014; Castro et al., 1999; Jack et al., 2015; Kuller et al., 1998; Mahley et al.,
2006; Santoro and Sutton-Tyrrell, 2011; Sheffler et al., 2014)
ε2/ε4
Males Increased risk of cardiovascular disease compared to ε3/ε3 individuals (Lahoz et al., 2001)
Females
Protection from cardiovascular risk and AD (Altmann et al., 2014; Castro et al., 1999)
University of Southern California
66
General
Greater AD risk in individuals of European descent (Kuller et al., 1998)
Evidence of brain amyloid accumulation occurs 10-15 years later than in ε4/ε4 carriers (Jack et al., 2015)
ε3/ε4
Males Increased risk for MCI and AD, although slower decline (Altmann et al., 2014; Reiman et al., 2005)
Females
Significant effects on risk for decline in episodic memory (Kaerst et al., 2013)
Higher MCI and AD risk than non-ε4 carriers and male ε3/ε4, comparable to ε4/ε4 men (Altmann et al., 2014; Payami
et al., 1994)
Increased likelihood of pathological levels of CSF tau and tau/Aβ ratio (Altmann et al., 2014)
Compared to ε3/ε4 men, faster age-related cognitive decline and longer survival rates (Corder et al., 1995; Farrer et
al., 1997; Smith et al., 2002)
General
In 70
+
normal ε4 carriers have greater amyloid PET than other genotypes (Jack et al., 2015)
Greater AD risk in individuals of European descent (Kuller et al., 1998)
Evidence of brain amyloid accumulation occurs 10-15 years later than for ε4/ε4 carriers (Jack et al., 2015)
ε4/ε4
Males
Higher risk of MCI and AD than ε3/ε4 or ε2/ε4 males or ε3/ε4 and ε4/ε4 females (Altmann et al., 2014; Castro et al.,
1999; Farrer et al., 1997)
Females
Comparable AD risk to ε3/ε4 females (Liu et al., 2013); longer survival rates following AD diagnosis (Reiman et al.,
2005)
General
Over 30% develop AD by age 75, with greater risk in individuals of European descent (Kuller et al., 1998)
ApoE concentrations are generally the lowest in this group, a predictor of AD risk (Rasmussen et al., 2015; Schiele et
al., 2000)
Highest risk for brain Aβ accumulation (Reiman et al., 2009; Villemagne et al., 2011)
General ε4
findings
Males
Years of life lost in men ε4 carriers are greater than in genotype-matched women (Schachter et al., 1994)
Females
Greater prevalence of AD in female ε4 carriers than males (Farrer et al., 1997; Payami et al., 1994)
Greater alterations in precuneus and anterior cingulate cortex connectivity (Damoiseaux et al., 2012; Reiman et al.,
1996)
Associations between APOE and cognitive decline evident later in females (Jack et al., 2015; Mortensen and Høgh,
2001)
More prominent phenotypic features in female MCI ε4 carriers – reduced hippocampal volume and worse cognitive
scores (Fleisher et al., 2005)
CSF tau, p-tau, and tau/Aβ-42 levels highest in MCI ε4 carriers compared to non-carriers with MCI, particularly for
females (Damoiseaux et al., 2012; Mattsson et al., 2009; Payami et al., 1994; Ruitenberg et al., 2001)
Women ε4 with mild-AD are more likely to have both neurofibrillary tangles and amyloid than ε4 men, indicating
greater pathology (Corder et al., 2004)
Postmenopausal ε4 women on ERT exhibit signs of neuroprotection and preservation of telomere length compared to
ε4 women not receiving treatment (Jacobs et al., 2013; Newhouse et al., 2013; Struble et al., 2007)
University of Southern California
67
General
APOE-ε4 show greater neural efficiency on episodic memory tasks (Mondadori et al., 2007) and better performance in
speed of processing, attention, and verbal fluency until mid 50’s when declines are evident (Evans et al., 2014; Lahoz
et al., 2001; Marchant et al., 2010)
Reduced myelin integrity in AD-relevant regions evident in infancy (Dean et al., 2014)
Increased dementia risk (AD, PDD, VaD, and DLB) and cardiovascular risk (Altmann et al., 2014; Lahoz et al., 2001;
Ruitenberg et al., 2001)
Carrying one ε4 allele shifts the AD risk curve 5 years earlier (OR=3.5-4) (Coon et al., 2007; Corder et al., 1993; Noguchi
et al., 1993)
Two alleles shifts the AD risk curve by 10 years (OR=12-15) (Coon et al., 2007; Corder et al., 1993; Noguchi et al., 1993)
APOE-ε4 exerts its maximal effects on AD risk by the early 70’s (Jarvik et al., 1995)
Up to 65% of individuals with AD harbor at least one ε4 allele (Farrer et al., 1997)
CSF Aβ less predictive of conversion from normal to MCI than for other genotypes (Apostolova et al., 2014)
By 40, 15% cognitively normal APOE-ε4 carriers are amyloid positive (Jack et al., 2015)
Summary of the results outlined in this review. Both protective and risk factors associated with APOE
genotype by sex are outlined. Aβ: Amyloid-beta 1-42;
AD: Alzheimer's disease; DLB: Dementia with Lewy Bodies; ERT: Estrogen replacement therapy; MCI: Mild
cognitive impairment; OR: Odds ratio; PDD: Parkinson's dementia; P-tau: Phosphorylated tau; VaD: Vascular
dementia.
U n i v e r s i t y o f S o u t h e r n C a l i f o r n i a
68
3 UNCOVERING BIOLOGICALLY COHERENT
PERIPHERAL SIGNATURES OF HEALTH AND
RISK FOR ALZHEIMER’S DISEASE IN THE AGING
BRAIN
Abstract
Brain aging is a multifaceted process that remains poorly understood. Despite
significant advancements in technology, progress towards identifying reliable risk factors
for suboptimal brain health has been limited in part by the use of methodological approaches
that rely on parametric statistics and univariate designs to explain complex, non-parametric
relationships between genetics, biology, and the environment. Here we demonstrate the
clinical utility of a novel unsupervised machine learning technique, Correlation Explanation
(CorEx), to discover how individual measures of structural imaging, genetics, plasma, and
CSF markers jointly inform risk for Alzheimer’s disease (AD). We examined 829 participants
(M age: 75.3 ± 6.9 years; 350 women and 479 men) from the Alzheimer’s Disease
Neuroimaging Initiative database to identify multivariate predictors of cognitive decline and
brain atrophy over a one-year period. Our sample included 231 cognitively normal
individuals, 397 with mild cognitive impairment, and 201 with AD as their baseline
diagnosis. CorEx discovered latent factors that predicted longitudinal disease trajectories
more accurately than the original features, and that align with established biological
University of Southern California
69
pathways. By combining a subset of plasma and brain features with CorEx factors jointly
representing plasma and brain measures, we were able to improve disease prediction along
the trajectory from normal cognition, mild cognitive impairment, to AD, with an area under
the receiver operating curve of up to 99%, and prediction accuracy of up to 89.9% on
independent “held out” testing data. This approach identified signatures of cardiovascular,
immune, and bioenergetic functions that were important predictors of AD, and supports the
utility of network measures in boosting detection and prediction of AD.
This chapter was modified and adapted from
Riedel, B. C., Daianu M., Ver Steeg, G., Mezher, A., Salminen, L. E., Galstyan A. & Thompson, P. M.
Uncovering biologically coherent peripheral signatures of health and risk for Alzheimer’s disease
in the aging brain. Under review at Frontiers in Aging Neuroscience, 2018.
Introduction
Alzheimer’s disease (AD) affects approximately 10% of the population over the age
of 65, and several lines of evidence suggest that there is an extended preclinical phase during
which treatments are most likely to be effective (Brookmeyer et al., 2011). The growing list
of potential biomarkers (e.g., neuroimaging, genetics, proteomics, cognition) offers
increasing potential for early diagnosis of AD and better prognosis of age-associated
diseases. The most widely accepted etiological model for AD suggests there is a temporal
order in brain changes that are characteristic of AD pathology, and map onto the typical
cognitive profile of amnestic cognitive decline (Braak & Braak, 1991; Jack et al., 2016).
University of Southern California
70
Accumulation of beta-amyloid “plaques” in the brain has been traditionally identified
as a hallmark of AD that begins decades before the onset of clinical symptoms. Amyloid
accumulation can be measured using positron emission tomography (PET) with amyloid
tracers, or via cerebrospinal fluid levels of Aβ1-42 (Hardy & Allsop, 1991; Jack et al., 2016;
McKhann et al., 2011). However, amyloid may not directly cause clinical symptoms, but may
trigger neuronal injury and subsequent degeneration through disruption of the tau protein
– the key component of neurofibrillary tangles (Hampel et al., 2010; McKhann et al., 2011;
Mudher & Lovestone, 2002). Importantly, neuronal and synaptic loss are key determinants
of cognitive impairment across domains, which are accompanied by brain atrophy, declines
in brain glucose metabolism in the temporal and parietal cortices, and increases of CSF tau
and phosphorylated-tau (Brookmeyer et al., 2007; Mapstone et al., 2014; Mosconi et al.,
2008). Coupled with carrier status of the APOE-ε4 allele, which increases AD risk up to four-
fold per allele, these changes can be broadly categorized into the hallmarks of AD risk and
pathology (Corder et al., 1993).
The primary goal of biomarker research is to better diagnose at-risk individuals well
before the onset of clinical symptoms to optimize the success of potential disease-modifying
treatments. Thus, classical hallmarks of AD, such as amyloid and tau, may not represent ideal
biomarkers of AD. Unfortunately, blood tests cannot diagnose AD with sufficient sensitivity
or specificity to be clinically useful (Mapstone et al., 2014). Machine learning methods have
been widely used to improve prediction of disease and identify signatures of risk (Jensen &
Bateman, 2011). Earlier work used a set of 18 plasma signaling proteins to discriminate AD
from other forms of dementia and cognitively normal controls, achieving 89% accuracy in
classifying Alzheimer’s disease from people without AD when a machine learning method
University of Southern California
71
was trained and tested on non-overlapping data (Ray et al., 2007). Similarly, Doecke et al.
(2012) classified AD patients versus healthy controls with 85% accuracy using a
combination of plasma inflammatory markers, participant demographics and clinical
information (Doecke et al., 2012). Although promising, most of this work has focused on
cross-sectional measures to classify current disease status.
As AD pathology develops decades before the onset of clinical symptoms, identifying
biomarkers that predict conversion to AD in cognitively normal individuals and those with
mild cognitive impairment (MCI) is of high clinical relevance. Even so, studies that combined
structural MRI and CSF measures to classify people with stable mild cognitive impairment
(MCI) relative to those that progressed to AD reported an accuracy ranging from 58.6 to
66.4%, depending on the prediction time frame (12-36 months) (Westman et al., 2012). To
improve these results and better understand the etiology of AD, it is important to determine
the best set of variables (features) for predicting clinical progression, and how these features
cluster with each other in the early versus late stages of disease progression. While the
crosstalk between the brain and periphery, crucial for shaping neuronal survival and
function, shows bi-directional interactions between neuroendocrine, neuroimmune, and
bioenergetic systems, and amyloid and tau pathways, the role of these markers in AD
etiology remains unclear (Alam et al., 2016; Blennow 2010; Chakrabarty et al., 2015;
Engelhart et al., 2004; Gonzalez et al., 2017; Zheng et al., 2014). In fact, some analytes have
shown different directions of effects across cohorts, perhaps depending on the specific
diagnostic groups being compared, measurement noise, as well as sex and genetic
background differences (Hu et al., 2012).
University of Southern California
72
Machine learning methods can identify predictors of diagnosis and prognosis, but
most methods (‘supervised’ algorithms) require diagnostic groups to be defined a priori and
this is counterintuitive to what is known about the AD continuum. A systems biology
approach has the potential to move beyond discrete biomarkers to complex networks of
predictors, which may better reflect disease heterogeneity and higher-order interactions
among predictors, improving predictive and explanatory power (Padmanabhan et al., 2017).
Here, we introduce a novel method - Correlation Explanation (CorEx) - to overcome several
issues imposed by traditional machine learning techniques. CorEx leverages principles of
information theory and unsupervised learning to distil the joint predictive power of classical
AD biomarkers across a range of diverse data types. Using this approach, we can better
understand predictors of decline in this disease in a tractable and principled way. Here we
implemented this method to study the discriminative power of more than 400 genetic,
plasma proteomic, CSF, imaging, and demographic measures from 829 participants from the
Alzheimer’s Disease Neuroimaging Initiative (ADNI), phase 1. We hypothesized that CorEx
would discover latent factors of variables across various data types that would enhance
predictive accuracy to identify individuals who progressed from cognitively normal status
to MCI, and MCI to AD. We identified the longitudinal diagnosis of each individual and used
this information to distinguish (1) stable CN individuals, from CN and MCI individuals who
progress to AD, (2) stable MCI, from CN and MCI individuals who progress to AD, (3) people
with a diagnosis of AD from individuals who do not progress to AD over the time frame of
the study. For each of these prediction problems, we determined the feature relevance. We
hypothesized that classical hallmarks of AD - namely APOE4 and CSF Aβ1-42 and tau levels,
would be among the most consistent and important factors for predicting clinical
University of Southern California
73
progression across all analyses. We further hypothesized that CorEx would boost the
predictive accuracy of the signal from these hallmarks by construction of latent factors
representing these measures, compared to the individual measures. Finally, we assessed the
relevance of these features in predicting a longitudinal cognitive composite score and
measures of brain atrophy.
Methods
Participants
Data were collected from 829 predominantly Caucasian individuals (350 women, 479
men) participating in ADNI - a longitudinal study of biomarkers of AD. Diagnosis of probable
AD is based on the NINCDS-ADRDA Alzheimer’s Criteria (McKhann et al., 1984). Inclusion
and exclusion criteria may be found in Petersen et al. (2010). Visits occurred every 6 months
for the first 2 years, and every 12 months thereafter. All ADNI data are publicly available
online (http://www.adni-info.org/) (Weiner et al., 2017). The study was conducted
according to the Good Clinical Practice guidelines, the Declaration of Helsinki, and the US 21
CFR Part 50-Protection of Human Subjects, and Part 56-Institutional Review Boards (IRB).
Written informed consent was obtained from all participants in advance. All study
procedures were approved by the local and participating IRBs of the ADNI study.
Biomarker Quantification and Analysis
An extensive panel of 203 laboratory tests was collected for all participants at the
baseline assessment. The panel consisted of plasma protein markers from the Luminex
XMAP platform by Rules-Based Medicine (Myriad RBM, Austin, TX). Proteins included
markers of liver function, cytokines, lipoproteins, oxidative stress, growth factors, hormone
University of Southern California
74
levels, glucose metabolism, and amyloid and tau levels, among others. Full protocol details
are available through the ADNI website
(http://adni.loni.usc.edu/wpcontent/uploads/2010/11/BC_Plasma_Proteomics_Data_Prim
er.pdf). We examined cerebrospinal fluid (CSF) levels from lumbar punctures of tau,
phosphorylated tau 1-81, and amyloid-β-1-42 levels collected at the baseline visit. All quality
control procedures were described previously by the ADNI Biomarker Core (Shaw et al.,
2009; Soares et al., 2012). Genomic analyses were completed according to the ADNI protocol.
Our primary genetic marker of interest was APOE ϵ4 carrier status, which was coded as the
number of ϵ4 alleles. For clarity in the following sections, we group measures into: (1)
Demographics (height, weight, sex, age), (2) Hallmarks of AD pathology (CSF measures and
APOE ϵ4 count), and (3) Plasma proteomics (all other measures, including a urine test of
kidney functions).
Scan Acquisition and Image Processing
All participants underwent whole-brain magnetic resonance imaging (MRI) on 1.5
Tesla GE, Siemens, or Philips scanners at 59 sites across North America. Although various
software platforms were used, a standardized MRI protocol ensured cross-site comparability
(Jack et al., 2008). A typical 1.5 T MR protocol involved a 3D sagittal MP-RAGE scan with
repetition time (TR): 2400 ms, minimum full TE, inversion time (TI): 1000 ms, flip angle: 8°,
24 cm field of view, and a 192×192×166 acquisition matrix in the x-, y-, and z- dimensions,
yielding a voxel size of 1.25×1.25×1.2 mm
3
that was later reconstructed to 1-mm isotropic
voxels. Using standard image preprocessing to correct for motion, intensity normalization,
affine registration of volumes to MNI space, skull stripping, non-linear registration using the
University of Southern California
75
Gaussian Classifier Atlas (GCA), and brain parcellation, we measured 68 cortical and 8
subcortical regions of interest (ROI) from baseline images with the Desikan-Killiany atlas
(11) from FreeSurfer version 5.3, processed locally
(https://surfer.nmr.mgh.harvard.edu/) (Fischl et al., 2002). We extracted measures of
cortical thickness and surface area for each of the 68 cortical ROIs, and regional volumes for
each subcortical ROI in the left and right hemispheres. To address potential confounds we
regressed out the effects of age, sex, and education on surface area, thickness, and volume
measures, as well as intra-cranial volume (ICV) on surface area and volume measures. All
subsequent analyses use these residualized measures.
Tensor-Based Morphometry Measures of Atrophy
We computed MRI-derived measures of structural brain atrophy by comparing each
subject’s one-year follow-up MRI to their baseline scan and measuring temporal lobe tissue
loss and ventricular expansion using tensor-based morphometry (TBM). To do this, we
registered each participant’s preprocessed follow-up scan to their baseline scan with a
nonlinear inverse-consistent elastic intensity-based registration algorithm, optimizing a
joint cost-function from the mutual information and elastic deformation of the images (Hua
et al., 2013). Then, representations of the degree of local contraction or expansion of the 3D
registration from the one-year scan to baseline, also known as the Jacobian determinant
map, was computed at the voxel level. Values in these maps represent relative tissue volume
differences expressed as positive or negative percentages of their baseline. Using this metric,
we assessed a bilateral temporal lobe region of interest as in Hua et al.; we next extracted
ventricular surfaces for each subject and as above registered these between timepoints for
University of Southern California
76
each subject to determine ventricular expansion or contraction between scans (Gutman et
al., 2013; Hua et al., 2011).
Correlation Explanation
Correlation Explanation (CorEx) is an information-theoretic optimization method
that constructs a low-dimensional hierarchy of latent factors that progressively explain more
nonlinear dependencies in the observations X 1…X N as measured by maximizing the
multivariate mutual information - also called total correlation (TC)
(https://github.com/gregversteeg/Bio_CorEx) (Ver Steeg et al., 2014; Ver Steeg et al.,
2015).
The special case of two variables is more commonly known as mutual information (I) and is
defined as the difference between the sum of the individual entropies (H) and the entropy of
the variables considered together.
I(X 1; X 2) ≡ H(X 1) + H(X 2) -H(X 1, X 1)
The dependence in the data as measured by TC(X) can be reduced or “explained” by
conditioning on some constructed factors Y
1
…Y
M
. The conditional TC goes to zero if and only
if all variables are independent of each other after conditioning on Y. “Explanation” refers to
this latter phenomenon, with the constructed factors containing all the information about
University of Southern California
77
causes of dependence in the data. As opposed to similar learning approaches, CorEx naturally
decomposes information in a hierarchical way: lower layers capture more local
relationships, and higher layers reflect more global interactions. The framework of CorEx is
in line with the ideas of network biology. An example construction of the CorEx framework
is presented in Supplementary Figure 1. Additional details of the optimization parameters
and the overall framework of CorEx are in (Ver Steeg et al., 2014, 2015) with some practical
considerations for biological data discussed in (Pepke et al., 2017).
We applied CorEx to discover shared information among plasma and demographic
measures, and the hallmarks of AD across all participants. We separately applied CorEx to
discover shared information within residualized brain measures. A total of 25 latent factors
were chosen for each CorEx model, representing optimal total correlation for the
corresponding measures. For clarity, we refer to these as “CorEx plasma” and “CorEx brain”.
This number of factors represents the simplest explanation of the data with maximum
multivariate mutual information across factors. These constructions resulted in generally
non-overlapping groups of latent factors that are maximally informative and robust to noise.
The values obtained for each factor represent a decomposition of common information and
correspond to the maximum likelihood labels for that specific latent factor and particular
participant (Ver Steeg, 2017).
Longitudinal Cognitive Composite Scoring
To facilitate comparisons with the joint set of plasma markers without needing to
address challenges raised by testing multiple hypotheses by considering cognitive tests
separately, and to determine which markers predicted the cognitive decline seen in AD, we
University of Southern California
78
created a cognitive composite score. Cognition was measured using three different cognitive
assessments that broadly represent executive function, orientation, attention, verbal, short-
term, and working memory – the Mini-Mental State Examination (MMSE), the Alzheimer’s
Disease Assessment Scale-cognitive subscale 13 item (ADAS-Cog), and the Clinical Dementia
Rating scale (CDR: sum of boxes) (Folstein et al., 1975; Hughes et al., 1982; Mohs 1996;
Reitan 1955). We calculated a composite z-score for each participant at their baseline and
one-year follow-up visits. Aggregate scores are often more reliable than individual items, and
composite scores can reduce the effect of measurement errors (Ayutyanont et al., 2014;
Langbaum et al., 2015). The three cognitive assessments were evenly weighted but were
inversely coded for the CDR sum of boxes and ADAS-Cog so that lower scores represented
greater cognitive impairment across tests. In this way, all measures were aligned in direction
before converting to z-scores. These z-scores were then summed within visit for each
subject, representing the composite score for that time point. A similar approach was
assessed in prior work on neurodegenerative diseases (Crane et al., 2013; Cutter et al., 1999;
Donohue et al., 2015). Finally, we calculated a longitudinal composite score by subtracting
the baseline composite from the one-year follow-up. This longitudinal composite score was
used in subsequent analyses.
Diagnostic Groupings
Rather than classify current diagnosis, we wanted to determine how well we could
predict future disease progression. To do this, we identified two stable groups: (1)
cognitively normal (CNs) and (2) MCI (MCIs) individuals with at least one-year follow-up
and stable diagnosis across all visits. In ADNI, individuals have been followed for repeat
University of Southern California
79
assessments for an average of 1-5 years. We also identified CN and MCI individuals at
baseline who progressed to AD and considered these as the progressing groups (CNp &
MCIp). MCI individuals without at least one-year follow-up or who reverted to CN status
were excluded in analyses comparing stable and progressing CN and MCI individuals.
Overall, we identified 22 CNp and 211 MCIp patients, compared to 165 CNs and 147 MCIs
patients after a mean follow-up of 52.5 months. This difference represents a 42.8%
prevalence rate of AD, corresponding to an annual conversion rate of 9.8%. Finally, we
considered more broadly all individuals receiving a diagnosis of AD (at baseline or later)
compared to a non-AD diagnosis, with an average follow-up for CN and MCI groups of 49.5
months. Individuals who did not receive an AD diagnosis at any time point were identified
as the “non-AD” group, while all other individuals were included in the AD group.
Approach
To contribute to the framework of the temporal evolution in AD biomarkers in
relation to each other and the onset and progression of clinical AD, our analyses were carried
out with three subgroups: (1) CNs vs CNp/MCIp, (2) MCIs vs CNp/MCIp, (3) non-AD vs. AD.
For each diagnostic grouping, we performed feature selection across CorEx factors and the
original features to determine the top 10 features within each class (CorEx plasma, CorEx
brain, plasma, brain measures). We then used those features for diagnostic prediction with
gradient boosting, or with bootstrap stepwise regression for the continuous outcomes of
cognition and TBM measures. We outline the steps involved in the analyses in Figure 1. To
determine the signifiance of model improvement using differing feature types, we used the
McNemar test to compare the top two models within each diagnostic prediction.
University of Southern California
80
FIGURE 3-1
Methods included in the current analyses. After constructing latent factors using CorEx within plasma and
brain measures separately, we then performed feature selection for each feature type to determine the top 10
features for each outcome measure and each feature type. We then combined these top features across feature
types within each corresponding outcome and either (1) performed disease prediction with gradient boosting
classification or (2) used bootstrap stepwise regression to determine the importance of the top features for
predicting the continuous outcome measures (i.e. cognition and TBM atrophy measures).
Feature Importance and Selection
To reduce training time and data dimensionality, we employed feature selection prior
to our disease predictions (Bell & Wang, 2000; Guyon et al., 2003). For each diagnostic
grouping: (1) stable CN vs progressors (CN and MCI), (2) stable MCI vs progressors (CN and
MCI), (3) No AD vs. AD, we applied a random split of the data stratified by diagnosis using
70% of the data for training, and the remaining 30% for hold-out testing. We then used an
ensemble of machine learning methods to determine feature importance within each
diagnostic grouping and increase stability and robustness of the selected features (Seijo-
Pardo et al., 2017). Binary diagnostic predictions included the General Linear Model (GLM),
Gradient Boosting Machines (GBM), Treebag, Linear Discriminant Analysis (LDA), and K-
Nearest Neighbors (KNN), while regression tasks included GLM, GBM, KNN, ridge regression,
University of Southern California
81
and the least absolute shrinkage and selection operator (LASSO) method. For each model, a
weighted-average mean squared error was used across methods to estimate prediction error
on unseen test data, and its association reported using results across 10-times repeated 10-
fold cross-validation (Pacławski et al., 2015). For all tasks, variable importance rankings for
feature selection were carried out for each feature type separately (CorEx plasma, CorEx
brain, plasma, brain). We then combined the top 10 measures for each feature type and
performed the same feature importance scheme to determine the relative importance of
these features across the joint set of all top measures. All statistical analyses were performed
using R, version 3.4.4 (R Core Team, 2013).
Disease Prediction
Using the top features across all data types from our feature selection step, we
performed disease conversion prediction using gradient boosting machine (GBM) learning
(Freund & Schapire 1997; Friedman 2001). This technique is based on the principle that
combined learners (decision trees) can outperform single learners, and thus is aligned with
the combinatoric network approach of CorEx. We used GBM for predicting diagnosis as it is
more immune to collinearity issues than other common machine learning models, and
therefore suited for prediction with both CorEx factors and the corresponding original
features. In GBM, trees are grown sequentially to reduce the errors of the previous trees, but
the residuals are resampled, and a fraction of the data is available at each iteration to reduce
overfitting. Learning is regularized through shrinkage on the learning rate. For our
prediction tasks, data was randomly stratified by diagnosis with 70% used for training and
30% used for hold-out testing. Parameters were optimized using a grid-search under a 10-
University of Southern California
82
times repeated 10-fold cross-validation framework on the training data only. Results on
hold-out test data were used to determine the average accuracies across the repeated cross-
validations. Given the disproportionate number of individuals between diagnostic groups,
we applied model weights to balance the groups for disease predictions. Area under the
receiver operating curve (AUC) was calculated to determine the models’ overall
discriminative power on the test data. Within each diagnostic group prediction, we
iteratively applied these methods to identify which combinations of data types yielded the
best predictive AUC, and to understand how CorEx features contributed to the overall
results.
Association of Measures with Cognition and Brain Atrophy
To estimate the stability and relative contributions of the original measures and
CorEx factors in predicting outcome measures, we performed feature selection and then
separately ran 1,000 bootstrap stepwise regressions for each outcome measure: (1)
temporal lobe TBM, (2) ventricle TBM, (3) cognitive composite score. For the cognitive
analyses we included all data types (CorEx plasma, CorEx brain, plasma, brain). However, for
predicting the temporal lobe and ventricular TBM measures, we only included the CorEx
plasma and plasma measures as predictors.
Results
Demographics for the 829 ADNI phase-1 participants in the current study are
categorized by baseline diagnosis and sex in Table 1.
University of Southern California
83
TABLE 3-1
Demographics. Basic demographic information for the ADNI participants included here, grouped by baseline
diagnosis and sex.
CorEx Networks
We used CorEx to learn low-dimensional representations and reconstruct meaningful
biological and hierarchical structures using data across a wide range of >200 plasma, CSF,
genetic, and demographic measures, and separately for >200 cortical brain measures. We
then built more robust predictors that we display in the form of a tree-based network in
Figure 2 and Figure 3. Measures are labeled with text and color-coded based on the
measurement type, indicated in the key. Latent factors are illustrated as ‘nodes’ in the graph
(and factors at the first level of the hierarchy (k=1 as in Supplementary Figure 1) are
numbered 0, …, 24). Links reflect learned functional relationships between variables and the
gray shade of an edge reflects the shared mutual information (darker indicates more shared
mutual information). The size of a latent factor node is based on the amount of multivariate
mutual information among its children nodes. Hierarchical groups constructed by CorEx
University of Southern California
84
were biologically coherent, such as a cluster of apolipoproteins (factor 15) and the hallmarks
of AD (factor 14).
University of Southern California
85
FIGURE 3-2
Hierarchical representation of biomarkers constructed by CorEx. Using CorEx across a panel of >200
markers of plasma, demographic, and CSF measures, we constructed a hierarchical network based on the joint
information shared between measures. We identified 25 latent factors that represent the optimal total
correlation across measures and factors. Latent factors are represented by circular nodes and numbered
accordingly. Colors indicate variable type, as defined in the bottom legend, and gray shaded edges reflect the
amount of mutual information shared between connecting nodes, where darker edges indicate more shared
University of Southern California
86
information. The size of each node is a function of the amount of mutual information shared among the
connected variables.
FIGURE 3-3
Hierarchical representation of brain measures constructed by CorEx. Using CorEx across a panel of >200
residualized gray matter measures we constructed a hierarchical network based on the joint information
shared between measures. We identified a set of 25 latent factors that represent optimal total correlation
across measures and factors. Latent factors are represented by circular nodes and numbered accordingly. The
colors indicate measurement type, defined in the legend on the bottom right, with thickness measures in blue,
cortical volumes in green, and surface area in purple. Gray shaded edges reflect the amount of mutual
information shared between connecting nodes; darker edges indicate more shared information. Only the top
measurements are shown for each node.
Predicting Cognitive Decline
After adjusting for age, sex, education, and APOE4 status, our cognitive composite
score was significantly predictive of progression to AD within the baseline CN and MCI
University of Southern California
87
groups (p < 0.001) and of AD longitudinally across all groups (p < 0.001). As presented in
Figure 4, bootstrap stepwise regression analyses predicting our composite score revealed
that the CorEx factor representing the accepted hallmarks of AD (factor 14) was the most
predictive. Other highly important features included CD5 Molecule Like (CD5L), and the
CorEx brain factor which includes thickness and volume measures of the limbic lobe
(amygdala, hippocampus, and entorhinal cortex) (factor 6). Multiple CorEx plasma factors
(e.g. factor 10, 12, 14, 16) show greater predictability, as indicated by the number of times
the features were selected, than the individual measures that the factors represent.
University of Southern California
88
FIGURE 3-4
Bootstrap regression results for a longitudinal cognitive composite score. As outlined in the Methods, we
created a longitudinal cognitive composite score broadly encompassing executive function, orientation,
attention, verbal, short-term, and working memory. Scores are based on change from baseline to one-year
follow-up and were significantly associated with progression to AD within the baseline CN and MCI groups (p
< 0.001) and of AD longitudinally across all groups (p < 0.001). To understand the most predictive features of
this score, we performed feature selection across feature types and then ran bootstrap stepwise regressions
using 1,000 permutations. The top features are shown on the left; arrows indicate the direction of effect for
each measure. Corresponding regions included in the CorEx factors are outlined in Figure 2 and Figure 3.
University of Southern California
89
FIGURE 3-5
Bootstrap regression results for atrophy measures using tensor-based morphometry in the temporal
lobes (top) and ventricles (bottom). As outlined in the Methods, we created tensor-based morphometry
(TBM) measures of atrophy between baseline scanning and one-year follow-up. To understand the most
predictive features of these atrophy measures, we performed feature selection using CorEx plasma and the
original plasma measures, and then ran bootstrap stepwise regressions using 1,000 permutations for each TBM
University of Southern California
90
measure. The top features are shown on the left; arrows indicate the direction of effect for each measure.
Corresponding regions included in the CorEx factors are outlined in Figure 2 and Figure 3.
Predicting Brain Atrophy
Following feature selection, we ran bootstrap stepwise analyses on two tensor-based
morphometry summary measures representing brain change over time: in the temporal
lobes and the lateral ventricles. As shown in Figure 5, the majority of the CorEx factors that
were selected following our ensemble feature selection step were the same for these two
TBM measures. Specifically, factors 4, 7, 12, and 14 were among the most predictive for both
TBM measures of atrophy. CorEx factors 5 and 17 were maintained only for the ventricles,
and CorEx factors 1, 16, and 24, were maintained only for the temporal lobe measures. While
separate, all of these measures include proteins involved in the immune response, such as
cell surface markers CD40, complement C3, monocyte chemotactic protein 3, and
macrophage colony stimulating factor. Additionally, while there was overlap among some of
the top CorEx factors and the original features they represent, this was not the case for many
of the features. The CorEx factor including many classically accepted hallmarks of AD (factor
14), was the most predictive single measure for both brain atrophy measures, and much
more predictive than the original features that the factor represents. The factor representing
multiple apolipoproteins and cholesterol measures (factor 15), was also highly predictive
for the temporal lobe atrophy measure. Chromogranin A was highly predictive of temporal
lobe atrophy, but much less so for the ventricles. Moreover, of the features selected with the
bootstrap stepwise regression at least 50% of the time, 60% of the measures were CorEx
plasma factors for the ventricles, and 80% were CorEx plasma factors for the temporal lobe,
University of Southern California
91
corresponding to greater predictive performance overall for the CorEx factors than for the
original measures.
Predicting Longitudinal Diagnosis
Feature selection results showing the relative importance of the top CorEx factors and
the individual features for predicting longitudinal progression are presented in Figure 6.
Using our ensemble feature selection technique, the top features performed well across
prediction groups, with an accuracy using all measures ranging from 71.9-88.2%, as shown
in Table 2.
University of Southern California
92
FIGURE 3-6
Feature selection results for longitudinal prediction of AD. Using our ensemble feature selection technique
outlined in the Methods section, we identified the top 10 features within each feature type (CorEx plasma,
CorEx brain, plasma, brain) across each of our three diagnostic groups used for prediction: (1) stable
cognitively normal (CN) vs. CN and individuals with mild cognitive impairment (MCI) who progress to AD, (2)
stable MCI vs. CN and MCI participants who progress to AD, and (3) ‘No AD’ vs. AD. We then combined the top
10 features across feature types and within diagnostic groups to identify the relative importance of these 40
features for each diagnostic group. “Raw” measures denote the individual features, while “CorEx” measures
denote the factors identified with CorEx. Latent factors are defined in Figure 2 for CorEx plasma measures, and
Figure 3 for CorEx brain measures.
University of Southern California
93
TABLE 3-2
Tests of prediction results using gradient boosting machines across diagnostic groups and feature
types. Gradient boosted machines (GBMs) were used to predict longitudinal diagnosis for stable CN vs. CN and
MCI participants who progressed to AD, stable MCI vs. CN and MCI participants who progress to AD, and all
individuals who had or progressed to AD vs. those who did not. Results were obtained following feature
selection, as shown in Figure 6, and a 70/30 split for training and testing, respectively. Additional details are
outlined in the Methods. We report the area under the ROC curve (AUC) for the testing sets averaged across
repeats. We included individual groups of measures (plasma, hallmarks of AD, demographics, and grey matter
measures) and CorEx features in the prediction and explored different combinations of these feature groups to
understand the benefit of including CorEx factors. Among the original features, plasma measures showed high
AUC for the stable CN vs. CN and MCI participants who progress to AD, while brain measures were more optimal
for predicting a longitudinal diagnosis of AD vs. those who do not have or develop AD.
We generated receiver operating characteristic (ROC) curves to determine how the testing
performance was affected by different combinations of feature types, as shown in Figure 7.
University of Southern California
94
FIGURE 3-7
University of Southern California
95
University of Southern California
96
ROC plots for the average testing sets across diagnostic groups and feature types. We show the area under
the ROC curve (AUC) averaged across testing repeats. We denote the progressing groups by “CNp” or “MCIp”.
We included individual groups of measures (plasma, hallmarks of AD, demographics, and grey matter
measures) and CorEx features in the prediction and explored different combinations of these feature groups to
understand the benefit of including CorEx factors. Across diagnostic groups we largely see improved AUC with
inclusion of CorEx factors. CorEx factors are outlined in Figure 2 and Figure 3.
Results were generally improved by combining all feature types, particularly for AUC
metrics. We were able to predict progression to AD among CN and MCI participants
University of Southern California
97
compared to stable CN with up to 88.4% test accuracy and 99% AUC. McNemar’s test
indicated significantly different performance with plasma compared to CorEx plasma
measures (p<4.3x10
-14
), while the combined feature-set was significantly different from
plasma measures (p<1.5x10
-10
) indicating optimal performance for the combined features
overall. The greatest benefit of incorporating CorEx factors was observed for discriminating
CN and MCI progression to AD versus MCI participants who remained stable, with 71.9% test
accuracy and 96% AUC. Here, CorEx plasma measures were the top individual feature type
and performed significantly different from the original plasma measures (p<0.0001), while
performance using the combined feature types was significantly different from CorEx plasma
features alone (p<6.3x10
-06
). Finally, comparing individuals who have or progress to AD and
those who do not, the structural brain measures appeared to perform the best for individual
measures (p<1.8x10
-09
), while the combined set of features performed the best overall
(p<5.6x10
-15
).
CorEx plasma factor 14, representing the hallmarks of AD, was consistently more
predictive across all analyses than any of the individual hallmark measures. For instance, in
the stable CN vs. CN and MCI progressors analysis, CorEx plasma factor 14 was almost twice
as important than either Aβ1-42 or tau CSF measures, and this difference was even greater
for the stable MCI vs. CN and MCI progressors or the longitudinal AD vs. ‘No AD’ analyses.
However, this factor was not the most important factor in the stable CN vs. CN and MCI
progressors. Interestingly, CSF total tau was more important than phosphorylated tau in
each of these analyses. Not all important individual features were contained within the most
important CorEx plasma factors, such as apolipoprotein AII for the stable CN vs. CN/MCI
progressors. Additionally, some of the top individual features were not captured with CorEx,
University of Southern California
98
such as CD5L, cortisol, and leptin, supporting the importance of combining original analytes
and CorEx factors due to their somewhat independent contributions to predicting AD.
Discussion
In this study, we used information theory through a novel method, CorEx, to discover
strong and biologically relevant relationships among the hallmarks of AD, plasma markers,
demographics, and MRI-derived brain measures, and distilled these relationships into new
and more powerful predictors of disease progression. The latent factor representing the key
hallmarks of AD (factor 14), CSF Aβ-1-42 and tau measures and APOE, was consistently
included in the top features across all analyses. Moreover, many of the features that formed
a CorEx factor are consistent with known mechanisms involved in the etiology and
expression of AD. Specifically, plasma factors involved in established biological pathways
included multiple Apolipoproteins (factor 15), and two latent factors representing white
blood cell count measures that were connected within the CorEx network (factor 12 and 18).
Each of these three factors were also among the most predictive in multiple analyses.
Most latent factors of the brain represent the bilateral measures for a particular
region, such as factors 22 and 24, which represent the cerebellar cortex and pallidum,
respectively. Additional factors contain bilateral measures across the limbic lobe, such as the
amygdala, entorhinal cortex, and hippocampus, encompassing both thickness and
volumetric measurements (factor 6). Aside from this factor, most factors contained only one
measurement type (surface area, volume, or thickness), or a combination of surface area and
volume, but not all three. Thickness measures were less likely to be combined with either
University of Southern California
99
surface area or volume within a factor, and this is supported by cross-sectional and
longitudinal studies of brain development in adolescents and adults showing that within
most regions, cortical thickness exhibits linear decreases with age, whereas both cortical
volume and surface area show overlapping curvilinear trajectories (Rimol et al., 2012;
Sanabria-Diaz et al., 2010; Storsve et al., 2014).
Although the brain measures grouped within a factor were generally what we would
expect, certain plasma measures were not. For instance, we would not have expected that
apolipoprotein H or clusterin would be found within a separate factor from other
apolipoproteins, given their shared role in lipid transport, or that the ratio of apolipoprotein
B to apolipoprotein A1 would be separate either, and that this ratio would be within a factor
consisting primarily of immune markers such as monocytes and immunoglobulin E.
However, in line with this, prior studies have shown that regulatory T cells promote the
clearance of apolipoprotein B-containing lipoproteins via increasing the expression of
sortilin-1 and lipid-modifying enzymes by the liver, pointing to the interplay between the
immune system, lipid metabolism, and the liver, which may synergistically mediate risk for
AD (Getz & Reardon, 2014). Here we found apolipoprotein B to be predictive of expansion of
the ventricles.
As shown in Figure 6, the most important features in predicting progression to AD
varied depending on the comparison group (i.e., stable cognitively normal, stable MCI, or
more generally individuals who do not progress to AD). Volume of the left hippocampus -
and the latent factor containing medial temporal and amygdala structures were highly
University of Southern California
100
important for predicting clinical progression from cognitively normal status, and those with
MCI. Conversely, latent factors of plasma markers were the strongest predictors of
progression to AD in CN/MCI participants, suggesting that they may be more sensitive
markers of early disease progression than structural imaging markers. This supports the
utility of plasma measures as an early biomarker tool, and also corroborates the crosstalk
between the peripheral and central nervous system in mediating risk for disease. Although
previously associated with AD, certain plasma markers were not retained in latent factors
that were deemed important for any of our analyses (e.g., latent plasma factor 6 and 9), which
includes measures of bilirubin and blood pressure related measures respectively, and factor
11 which includes very-low-density lipoproteins, triglycerides, and adiponectin. These
findings may suggest that these particular markers are only nominally associated with AD,
and that compared to other measures included here, they do not provide additional
predictive power.
The most important CorEx factor predicting our cognitive composite were volume
and thickness measures of the medial temporal lobe (factor 6), which are among the earliest
regions to exhibit neuronal degeneration, neurofibrillary tangle deposition, and
accumulation of amyloid in AD (Braak & Braak, 1991; Scheff & Price, 2003). The most
important plasma measures for our cognitive composite were the latent factor representing
the hallmarks of AD, and CD5L, a soluble immune effector expressed primarily by mature
macrophages that is involved in fatty acid metabolism and lipid biosynthesis (Sanjurjo et al.,
2015). Importantly, CD5L is involved in regulating the inflammatory response to pathogens
and in the development and maintenance of the lymphoid compartment and may have
University of Southern California
101
additional relevance to AD through its regulatory roles of apoptosis and autophagy
(Miyazaki et al., 1999; Nixon et al., 2011). Additionally, although identified in a small sample
of 78 subjects, plasma levels of this protein have been associated with neocortical amyloid
burden (Ashton et al., 2015). Given the established association between Th17 cells and CD5L,
and the interaction between Th17 cells and neurodegeneration, these results suggest follow-
up studies elucidating the specific role for CD5L in AD is warranted (Wang et al., 2015; Zhang
et al., 2013).
Although CorEx discovered latent factors that showed greater predictive accuracy for
AD progression than the individual hallmarks of AD, no individual plasma marker showed
greater importance than the latent factor representing the AD hallmarks. This further points
to the importance of considering plasma markers together instead of individually. Other
factors were consistently important for predicting clinical progression across all of our
analyses, such as chemokines and white blood cell markers of the innate and adaptive
immune system (factors 12 and 13). An impaired immune response to toxic amyloidogenic
substances may make it easier for amyloid to accumulate in the brain (Richartz-Salzburger
et al., 2007). Likewise, amyloid precursor protein (APP) mRNA expression is increased in
peripheral lymphocytes in AD, and alterations in APP expression leading to amyloid
deposition may also cause changes in peripheral immune cells creating a feedback loop that
ultimately may lead to cognitive decline and neurodegeneration (Jiang et al., 2003). A
subsequent decline in lymphocyte levels and lymphocyte percentages may indicate immune
dysregulation or immunosenescence in AD. Similarly, neutrophils defend tissue against
invading pathogens during sterile inflammation and increases in neutrophil levels are also
University of Southern California
102
associated with blood brain barrier disruptions and chronic inflammation in AD
(Kolaczkowska & Kubes, 2013; Phillipson & Kubes, 2011; Zenaro et al., 2015). Moreover, a
set of dietary, immune, and inflammatory markers (factors 16 and 24) were more important
predictors of diagnosis than the hallmarks of AD when predicting CN/MCI individuals who
progress vs. the stable CN group. These factors included variables that are known to be
modifiable by dietary and lifestyle changes, such as vitamin B12 levels and C-reactive protein
(CRP). Importantly, factor 16 also included amyloid P component (AP), an important
component of all amyloid deposits including those typically found in the brains of patients
with AD (Yasojima et al., 2000). These relationships are also supported by prior literature.
For instance, one Finnish-community dwelling longitudinal study found that serum levels of
B12 were protective against AD in cognitively normal individuals, and that there were
potential interactions with homocysteine serum levels, another modifiable marker
(Hooshman et al., 2010). Likewise, CRP is predictive of later cognitive decline in mid-life
(Laurin et al., 2009). Moreover, CRP and AP are both acute phase proteins of the innate
immune response and colocalize in neurons in individuals with AD (Steel & Whitehead,
1994). As part of the innate immune response, CRP and AP activate the classical complement
pathway in an antibody independent fashion, and indeed factor 16 includes complement c3,
further supporting the network approach achieved with CorEx. Activation of complement is
important for opsonizing targets for phagocytosis and the subsequent destruction of
pathogens, such as amyloid beta plaques (Hicks et al., 1992; Wolbink et al., 1996). The
autodestructive association of CRP and AP with activated complement fragments attached
to host tissue has been seen in degenerative conditions, such as atherosclerosis and AD
(Lagrand et al., 1997; McGeer et al., 2001; Torzewski et al., 1998).
University of Southern California
103
A few aspects of the data on individual predictors deserve comment. Firstly, our data
are based on a variable length of follow-up (roughly 1-5 years), which optimizes the available
study information. A homogeneous follow-up interval would not have been ideal as it would
have reduced the number of available participants progressing to AD predictions based on a
smaller sample size may generalize poorly. Secondly, the cross-sectional design of the
plasma markers is an important limitation and we could not follow up on the potential
cyclical or prandial changes of some of these measures, so we are unable to make conclusions
about the trajectory of these measures. It will also be important to replicate these findings
in future cohorts. Finally, sex was used as a demographic variable included in the plasma
measures when constructing the CorEx plasma latent factors. As sex mediates multiple
pathways involved in AD risk, such as APOE, some of these factors may have been grouped
differently if we had constructed the factors separately for women and men (Fisher et al.,
2018; Riedel et al., 2016). Indeed, the CorEx factor 0 - which includes sex - was among the
most important features for predicting longitudinal diagnosis of AD, though body weight and
hormone levels may also be driving this association. Future work will seek to address this
potential confound.
Summary
We tested CorEx, a novel model-free data-driven approach to combine relevant
groups of >400 potential biomarkers from brain imaging, genetics, plasma, and demographic
information. We were able to discover a small set of tractable relationships in 829
participants across the trajectory from normal cognition to MCI and AD. While the
University of Southern California
104
relationships between some of the measures have been previously documented and several
measures were already known to be associated with AD, the clustering achieved with CorEx
and the subsequent results provide more direct evidence for a network of related measures
and how these measures jointly predict disease progression, brain atrophy, and cognitive
decline. These results also demonstrate the power of CorEx to identify clusters of variables
that involve synergistic and coherent sets of the original features, revealing stronger
combinations of variables that may be only weakly predictive when examined as individual
predictors. Our results point to the consistent importance of amyloid and tau across the
disease trajectory, but also to the timepoint specific contributions of the immune and
inflammatory systems, and to the role of cardiovascular health, hormone levels and glucose
metabolism.
FIGURE 3-8
Supplementary Figure 1. Example CorEx mathematical representation of latent factor construction. The
graphical model depicted is optimized to learn latent factors, Y, that minimize TC(X|Y ) + TC(Y) = 0 (Ver Steeg,
2017), where TC = total correlation, X are the input variables X 1…X N, and Y are the constructed latent factors.
In other words, starting from the top layer, each layer learns to explain dependence in the layer below. The
lowest layer, represented X 1…X N signify the original observed measures, while all subsequent layers represent
latent factors learned through CorEx.
U n i v e r s i t y o f S o u t h e r n C a l i f o r n i a
105
4 ABO BLOOD TYPE IS ASSOCIATED WITH
EARLIER AGE OF ALZHEIMER’S DISEASE ONSET
AND DIFFERENCES IN BRAIN STRUCTURE
Abstract
There is an increasing understanding that cardiovascular disease (CVD) and
cognitive impairment share an overlapping etiology, including alterations in lipid
homeostasis. The ABO blood genotype has implications for many CVDs including stroke
and coronary heart disease, with O blood type generally being found protective.
Furthermore, fucosyltransferase 1 and 2 (FUT1 and FUT2), which encode for the H antigen
that serves as the precursor to all ABO blood antigens, is epistatic to the ABO locus, and is
located upstream of apolipoprotein E (APOE), a key protein in the regulation of lipid
homeostasis, as well as the greatest identified genetic risk factor for Alzheimer's. Despite
clear evidence for ABO in CVD, the role of ABO in mediating risk for dementia is just
beginning to emerge. In this large-scale study, analyzing data from the Alzheimer’s Disease
Neuroimaging Initiative and the AddNeuroMed cohort (N=1,501, M age 74.5 ± 6.9; 844 men,
657 women), we show that O blood genotype interacts with APOE and secretor status,
defined by rs601338, leading to larger volume in the cerebellum, the occipital fusiform
gyrus and regions of the temporal and frontal lobe. Paradoxically, we show that O genotype
University of Southern California
106
and phenotype are also associated with increased risk for Alzheimer’s disease compared to
A individuals (Odds Ratio: 1.27), and an earlier age of AD-onset (p = 0.054). Finally, we
show evidence for interactions with plasma levels of receptor for advanced glycation end
products (RAGE), in mediating risk for AD in O blood type carriers. Overall, these results
indicate that while O blood type may be protective for brain health at a younger age, this
protection may not translate to individuals over the age of 65.
Introduction
There is an increasing understanding that cardiovascular disease (CVD) and
cognitive impairment share an overlapping etiology, including alterations in lipid
homeostasis (Kivipelto et al., 2001). The main blood type classification, ABO blood
genotype, can be broken down into four basic phenotypes: A, B, AB, and O. ABO blood
phenotype has implications for many CVDs including stroke and coronary heart disease,
with O blood type, the most common phenotype, generally being found protective for both
(Wiggins et al., 2009; Liumbrino et al., 2013). Furthermore, the H antigen, encoded by the
fucosyltransferase 1 and 2 (FUT1 and FUT2) genes (Galactoside 2-alpha-L-
fucosyltransferase 1 and 2), serve as the precursors to all ABO blood antigens in red blood
cells, or epithelium and mucus-secreting tissues, respectively, and these genes are epistatic
to the ABO locus. FUT1 and FUT2 are both located on chromosome 19 upstream of APOE,
which has shown high-levels of linkage disequilibrium and pleiotropy with nearby genes
(Lin et al., 2016; Theendakara et al., 2016; Yu et al., 2007). ApoE is a key protein in the
regulation of lipid homeostasis and APOE has been identified as the greatest genetic risk
factor for late-onset Alzheimer's (AD); (Pierce et al., 2010; Riedel et al., 2016).
University of Southern California
107
Aside from the potential protection from cardiovascular diseases exhibited in O
blood type individuals, these individuals are also protected from multiple types of cancer,
and show greater survival times (Aird et al., 1957; Beckman, L., & Ängqvist, 1987; Vasan et
al., 2016). Yet cancer has an inverse relationship with AD, such that cancer survivors have a
decreased risk for AD (Roe et al., 2005). Additionally, APOE ε2 carriers may be protected
from AD, but show an increased risk of aggressive cancers (Ifere et al., 2013). This
interaction may speak to the importance of the immune system in understanding the role
of ABO in disease. ABO blood type also serves as a regulator of immune effector cells, with
links between O blood type and increased susceptibility to Escherichia coli, Helicobacter
pylori, Norwalk virus, and other pathogens (Hutson et al., 2002; Kanbay et al., 2005; Linden
et al., 2008). O blood type individuals with positive secretor status, defined by the FUT2
gene, have higher levels of natural anti-TF (Thomsen-Friedenreich hapten) IgM and IgG
production (Kurtenkov et al., 2005). Importantly, IgM ABO antibodies have been shown to
activate complement, a critical pathway in mediating risk for AD, and circulating immune
complexes of plasma derived amyloid-β and IgM are increased in AD (Marcello et al., 2009;
McGeer et al., 1989).
Despite clear evidence for ABO in CVD and cancer, the specific role of ABO in
mediating cognitive function and risk for dementia is just beginning to emerge. Recently, a
small study, of 189 young and middle-aged adults, reported that O blood type carriers
might be at a reduced risk for Alzheimer's disease due to greater cerebellar gray matter
and hippocampal volume (De Marco & Venneri, 2015). Additionally, a possible reduction in
risk for cognitive decline, as measured by verbal fluency and word list recall, was reported
for middle-aged O carriers (Alexander et al., 2014). Most of these studies have been
University of Southern California
108
speculative as they have not been carried out within an older population at greater risk for
AD, are against what might be predicted given findings relating to the immune system and
have not asked more targeted questions about how blood type might mediate risk for AD,
leading to a large knowledge gap in the field. Moreover, most studies exploring blood type
have focused on ABO phenotype, ignoring the complexities of ABO genotype. Specifically, as
O is recessive, an individual must carry two O alleles in order to have the O phenotype,
while an A or B individual may, and often do carry one O allele (Von Dungern & Hirszfeld,
1911). To bridge these knowledge gaps, we report the first study of this nature in a large
sample of older adults (N=1,501; M age 74.5 ± 6.9; 844 men, 657 women), exploring the role of
both ABO blood genotype and phenotype, and the interaction with FUT2 secretor status
and APOE, on brain imaging in older individuals from the Alzheimer’s Disease
Neuroimaging Initiative (ADNI) and the AddNeuroMed cohort (Lovestone et al., 2009;
Weiner et al., 2017). We investigate the role of ABO in mediating age-of-AD-onset and
explore the interaction between plasma proteomics and ABO blood type as they relate to
risk for AD. We hypothesized that the immune system likely plays an important role in
understanding ABO blood type interactions with Alzheimer’s, and that given this and the
links with ABO and cancer, O might actually increase the risk for AD. In line with this, we
also hypothesized that we might see plasma interactions between the number of O alleles
and the complement protein or other proteins involved in the innate immune system in
mediating risk for AD.
Methods
Participants
University of Southern California
109
The data used here were obtained from two large multi-center longitudinally
followed cohorts, the Alzheimer’s Disease Neuroimaging Initiative (ADNI) from phases
ADNI-1, ADNI-GO, and ADNI-2, and AddNeuroMed. Both studies involve similar designs,
with diagnosis of probable AD based on the NINCDS-ADRDA Alzheimer’s Criteria
(McKhann et al., 1984). The goals of these studies are to identify early biomarkers for
clinical trials and improve understanding of mechanisms of AD. All ADNI data are publicly
available online (http://www.adni-info.org/) (Weiner et al., 2017). ADNI was launched in
2003 by the National Institute on Aging (NIA), the National Institute of Biomedical Imaging
and Bioengineering (NIBIB), the Food and Drug Administration (FDA), private
pharmaceutical companies and non-profit organizations, as a private-public partnership
across 59 sites in the United States and Canada. The primary goal of ADNI is to test whether
serial magnetic resonance imaging (MRI), positron emission tomography (PET), and other
biologically relevant markers are useful in clinical trials of mild cognitive impairment (MCI)
and early stages of AD. All ADNI participants were 55-90 years of age at baseline
recruitment. Inclusion and exclusion criteria have been previously described (Petersen et
al. 2010). AddNeuroMed, part of InnoMed (Innovative Medicines in Europe), is an
Integrated Project funded by the European Union Sixth Framework program (Lovestone et
al., 2009). The goals of AddNeuroMed were to develop and validate biomarkers of AD and
identify therapeutic targets and outcomes, based on converging evidence with in vitro and
in vivo analysis. Human data was collected from six different sites across Europe, five of
which were analyzed for the current study: the University of Kuopio in Finland, the
University of Perugia in Italy, Aristotle University of Thessaloniki in Greece, King’s College
London, and the University of Lodz in Poland (Lovestone et al., 2009). Both studies were
University of Southern California
110
conducted according to the Good Clinical Practice guidelines and the Declaration of
Helsinki and approved by the sites’ Institutional Review Boards (IRB). Written informed
consent was obtained from all participants in advance.
Estimating AD Age of Onset
A time variable was created using age at first AD diagnosis within ADNI to estimate
age of AD onset for participants diagnosed with AD before their baseline visit or at some
follow-up point during the study. For all other participants without AD at baseline, or who
did not progress to AD, age at their most current visit was used to determine censor status
and time. This censor variable was used for survival analyses to indicate (1) AD
participants or (0) non-AD participants.
Identifying ABO Blood Type and Secretor Status
Both ADNI and AddNeuroMed study participants were genotyped using one of two
genotype arrays. ADNI-1 and AddNeuroMed batch 1 participants were genotyped using the
Human 610-Quad BeadChip (Illumina, Inc., San Diego, CA, USA) array by TGen (Phoenix,
AZ). ADNI-2 and AddNeuroMed batch 2 participants were genotyped with the OmniExpress
BeadChip by the Center for Applied Genomics of Children’s Hospital of Philadelphia
(Philadelphia, PA). Detailed genotyping protocols have been described previously (Saykin
et al., 2010). We used genotype information to determine ABO blood type, number of O
alleles, and secretor status (a single variant: rs601338) using previously published criteria
(Wolpin et al., 2010). Although the variants sequenced were similar between arrays, not all
ABO-relevant variants were sequenced with the OmniExpress BeadChip (rs5743357 and
University of Southern California
111
rs505922). In order to determine ABO and other metrics for the participants genotyped
with this chip, we imputed missing alleles using variants 400 kb upstream and downstream
of the ABO gene that have been shown to be in high linkage disequilibrium (> 0.8) with the
missing variants. The specific variants as they are used to define blood type for each allele
are presented in Table 1. ABO genotype for each allele is then used to define ABO
phenotype. For example, an individual with an A allele and an O allele has an A phenotype,
an individual with an A and B allele has the AB phenotype, while an individual with two O
alleles has the O phenotype. Secretor status was determined using rs601338, where
number of secretor alleles is defined by number of G alleles.
TABLE 4-1
ABO defining SNPs used to determine number of O alleles and ABO genotype. ABO phenotype is
determined by considering the joint set of ABO genotype alleles. Secretor status was determined using a
single SNP, rs601338, where the G allele encodes the secretor status and A encodes non-secretors.
MRI Acquisition and Image Correction
All participants included in the current study underwent brain MRI scanning at
baseline. As imaging acquisition for AddNeuroMed was designed to be compatible with
ADNI, and to maximize power, we processed and ultimately combined scans across both
ADNI and AddNeuroMed (Jack et al., 2008; Westman et al., 2011). The imaging protocol for
both studies included a high-resolution sagittal 3D T1-weighted MPRAGE (voxel size: 1.1 x
University of Southern California
112
1.1 x 1.2 mm
3
) using either a 1.5 or 3 T scanner, acquired using a custom pulse sequence
designed to ensure compatibility across scanners (Jack et al., 2008). Using an affine 9-
parameter registration to adjust for global brain scale and alignment all scans were
corrected for intensity nonuniformities and linearly registered to the International
Consortium for Brain Mapping standardized brain template (ICBM-53) (Mazziotta et al.,
2001). A nine-parameter affine registration was applied to each scan to account for global
position and scale differences across individuals, such as differences in head size (Hua et
al., 2008). Globally aligned images were re-sampled in an isotropic space of 220 voxels
along each x-y-z axis, resulting in a final voxel size of 1 mm
3
.
Group Average Template Creation – Minimal Deformation Template (MDT)
A minimum deformation template (MDT), representing an unbiased “average” brain
template was created to allow for automated image registration, reduce statistical bias, and
to optimize detection of statistically meaningful effects. This was carried out using a mutual
information (MI)-based inverse-consistent algorithm and applying the inverse of the
average displacement field from all subjects to the MDT (Hua et al., 2008). As our study was
focused on blood type, we created an MDT from 39 cognitively normal individuals across
both ADNI and AddNeuroMed in accordance with the relative number of participants from
each study and using ABO phenotype distributions in line with the established population
averages (Garratty et al., 2004; Mourant et al., 1954). Specifically, our MDT was created
from ADNI: 13 O, 12 A, (7 A/A, 5 A/O) 2 B (1 B/B, 1 B/O), 2 AB, AddNeuroMed: 4 O, 4 A (all
A/O), 1 B (B/O), 1 AB. We used an almost equal distribution by sex as well, with 20 men,
and 19 females used for our MDT.
University of Southern California
113
Three-dimensional Jacobian Maps
All scans were non-linearly aligned to the MDT optimized using the regularity of the
deformation by quantifying the inverse consistent symmetric Kullback–Leibler (KL)-
distance between the MDT and the resulting deformations, such that all scans were in a
common space (Hu et al., 2011). The local expansion or contraction of the 3D elastic
warping transform, also known as the Jacobian determinant, was determined for each
subject. These 3D Jacobian maps indicate the relative volume differences between an
individual’s scan and the average MDT brain, revealing volumetric differences (i.e., relative
expansions or contractions for a given region). Of our original 1,526 participants, we
excluded 25 whose scans or deformation maps did not meet quality control standards due
to poor registrations to the MDT or significant artifacts.
Voxelwise Comparisons
Cross-sectional tensor-based morphometry (TBM) was used within the LONI
pipeline environment to analyze our 3D Jacobian maps (Dinov et al., 2010). We tested for
associations between brain volume and number of O alleles after accounting for all pair-
wise interactions between number of O alleles and number of secretor alleles using
multivariate linear mixed-effects regression with the following additional covariates: age,
number of APOE-ε4 alleles, sex, education, diagnosis, number of secretor alleles, number of
O alleles, and intracranial volume. Additionally, scanner field strength was used as a
random effect variable. We analyzed this model across all individuals, and separately for
only cognitively normal individuals (excluding diagnosis as a covariate). All interaction
University of Southern California
114
variables were coded as factors and represent all pairwise relationships (e.g. 0/1/2 O
alleles by 0/1/2 Secretor alleles, thus 8 total interactions with no O alleles and no secretor
alleles used as the referent). All maps were corrected for multiple comparisons using the
Benjamini-Hochberg false discovery rate (FDR); (Benjamini & Hochberg, 1995). The FDR
procedure assigns significance values to a statistical map based on the expected proportion
of voxels with statistics exceeding a given threshold under the null hypothesis and was set
at q < 0.05.
Quantifying Alzheimer’s-like Patterns of Atrophy via SPARE-AD
The Spatial Pattern of Abnormalities for Recognition of Early AD is an imaging
composite measure that provides individual scores indicating an AD-like pattern of
atrophy. Negative values indicate brain structure that is closer to normal, while positive are
closer to AD. This metric has been described elsewhere and was calculated with linear
support vector machines using multi-atlas structural parcellations representing regions of
interest (Davatzikos et al., 2011; Doshi et al., 2016). SPARE-AD has been found to be
predictive of cognitive decline in MCI and to differentiate individuals with MCI that
progress from those that remain stable (Fan et al., 2008; Misra et al., 2009).
Post-mortem brain Expression
We used data from six publicly available transcriptomic datasets generated using post-
mortem human brain across multiple brain regions. The regions selected varied between
studies and details have been published previously (Colantuoni et al., 2011; Freytag et al.,
2017; Hawrylycz et al., 2012; Hernandez et al., 2012; Kang et al., 2011; Trabzuni et al.,
University of Southern California
115
2013; Zhang et al., 2013; ). All datasets contained healthy brains across middle-age to late
adulthood. Using methods previously described, all outlier samples were removed, and
remaining transcripts were normalized using the Removal of Unwanted Variation (RUV)
method to control for systematic noise using multiple negative control genes (Freytag et al.,
2013; Gagnon-Bartsch et al., 2012). Transcriptopmic datasets were then subset by age
ranges (range=40-106; N=3,470) and analyzed for gene-prioritization based on co-
expression with known disease genes. Highly-penetrant AD genes were used for training
(APOE, APP, PSEN2), while ABO, FUT2, and the 20 most established AD-risk genes were
included as candidate disease genes (Lambert et al., 2013).
Adjusted Odds Ratio and Cox Proportional Hazards Regression
Cox regression and Kaplan-Meier survival analyses were conducted using the
estimated age of AD onset or age at most recent visit for non-AD participants as the
outcome variable with ABO blood type as the independent variable of interest, adjusted for
sex, education, and whether or not the participant had one or more APOE ε4 alleles (Nieto
& Coresh, 1996). Similarly, to determine the effect size of ABO, we performed logistic
regression with diagnostic status as the outcome variable: AD vs. non-AD, with ABO blood
type, APOE ε4 allele positivity, education, sex, and age as covariates to determine adjusted
odds-ratios by blood type.
Plasma Processing and Analyses
In order to determine if interactions between baseline plasma proteins and blood
type mediate potential risk for AD, we first pre-processed plasma markers within each
University of Southern California
116
study separately for a subset of CN and MCI participants that had these measures (N=474).
This was accomplished by box-cox normalizing each marker within baseline diagnostic
groups (i.e. CN and MCI); (Box & Cox, 1964). We then combined these results within each
study and converted values to z-scores across diagnostic groups in order to compare across
groups with potentially different distributions, and finally combined results across ADNI
and AddNeuroMed (Lavrakas, 2008). After quality control, this allowed us to combine and
subsequently perform interaction analyses with plasma and ABO blood type and number of
secretor alleles on the outcome of progression to AD using logistic-regression with 60
different plasma markers. We sought to determine whether or not any plasma markers
showed an interaction between ABO blood type and risk for AD (i.e. AD_progression ~
ABO*Plasma), and expected that both number of secretor alleles and ABO blood type would
also be significant predictors of risk for AD. To help validate our findings, we performed the
same procedure within ADNI-1 CN and MCI participants who also had 12-month follow-up
plasma measures (N=401).
Results
Demographics for the 1501 ADNI and AddNeuroMed participants in the current
study are categorized by baseline diagnosis and sex in Table 2. The majority of participants
were of western European descent with 96.0% reporting their race as white and 97.6%
were of non-Hispanic ethnicity.
University of Southern California
117
TABLE 4-2
Demographics. Basic demographic information for the ADNI and AddNeuroMed participants included here,
grouped by baseline diagnosis and sex.
Tensor Based Morphometry (TBM)
Across the combined datasets, and within cognitively normal individuals only, we
found no regions that were significantly associated with number of O alleles after FDR
correction. However, we found consistent significant interactions with number of APOE ε4
alleles and number of O alleles, as well as with number of secretor alleles and number of O
alleles. We focus our results on the regions consistently showing FDR corrected significant
associations for both the cognitively normal individuals as well as across diagnostic groups.
Our results are presented in Figure 1. Significant positive interactions were found between
number of APOE ε4 alleles and number of O alleles in bilateral regions of the cerebellum,
though more widespread in the right hemisphere. Effects, as measured using
unstandardized B values, were larger in the cognitively normal group across all significant
regions, potentially indicating controlling for diagnosis may mask out some of the effects.
University of Southern California
118
We also found significant positive interactions in the right frontal-pole and right middle
frontal gyrus, while we found negative interactions with the left occipital fusiform gyrus.
Significant interactions between number of secretor alleles and number of O alleles were
positively associated with the right parahippocampal gyrus and the left temporal fusiform
cortex. Negative associations were seen for the right genu of the corpus callosum and the
left precentral gyrus. Again, all significant effects were stronger in the cognitively normal
group.
SPARE-AD metric
In the subset of ADNI participants with this metric available (N=725), we used a
linear regression model with age, sex, number of APOE alleles, and the interaction between
number of secretor alleles and number of O alleles to determine the overall effect of
number of O alleles on an AD-like pattern of atrophy. The median score for this metric was
-0.07 (range= -4.58 to 6.25). Number of O alleles was positively associated with SPARE-AD,
indicating an increased AD-like pattern of atrophy with a greater number of O alles
(B=0.33; p=0.04). The interaction between number of O alleles and number of secretor
alleles was also significant but in the opposite direction, such that having more O alleles
and more secretor alleles was protective (B=-0.19, p=0.045). Number of secretor alleles
was also positively associated with SPARE-AD (B=0.29, p=0.038).
Brain Expression Prioritization
Results from our gene prioritization are presented in Figure 4. ABO was not among
the top predictors exhibiting co-expression with APOE, APP, or PSEN2. However, FUT2 was
University of Southern California
119
among the top prioritized genes, exhibiting greater consistent patterns of co-expression
with established AD genes than many of the most promising candidate genes identified
through GWAS. FUT2 showed co-expression patterns with ABCA7 and PSEN2 across
multiple datasets, while possible co-expression also exists with FUT2 and APOE, FERMT2,
and BIN1, although the consensus with these was less consistent between studies. ABO
showed consistent co-expression with FUT2, and possible co-expression with ABCA7.
Adjusted Odds Ratios and Risk of AD
Using A blood type as the comparator group due to sample size considerations for B
and AB blood types, we found a significant positive association with O blood type and risk
for AD (Odds Ratio: 1.27, CI: 1.02-1.61, p = 0.048). B and AB blood types were not
significantly associated with risk for AD (B Odds Ratio: 0.86, CI: 0.59-1.24; AB Odds Ratio:
0.91, CI: 0.50-1.62). These results are presented in Figure 2. To verify these associations
were driven by differences in risk for O specifically, we ran O as the reference level and
found that A was significantly associated with reduced odds of AD (Odds Ratio: 0.78, CI:
0.62-0.99, p = 0.048). Using O as the reference level also reduced the odds ratios for both
AB and B blood types, with B showing significantly lower risk (B Odds Ratio: 0.68, CI: 0.47-
0.97, p = 0.037; AB Odds Ratio: 0.72, CI: 0.39-1.28). Using number of O alleles instead of
ABO phenotype resulted in a significant positive relationship, such that more O alleles were
associated with greater risk for AD (Odds Ratio: 1.19, CI: 1.00-1.40, p = 0.042).
University of Southern California
120
FIGURE 4-1
Tensor-based morphometry results showing significant associations for 2 O alleles, indicating O
genotype. All colors correspond to unstandardized B values where significant after FDR correction (q < 0.05),
with red-yellow denoting negative associations, and blue indicating positive associations. Results for the
cognitively normal individuals are shown on the bottom row, while results across all diagnostic groups are
shown in the top row. Covariates for all models were age, number of APOE ε4 alleles, sex, education, number
of secretor alleles, intracranial volume, the interaction between number of O alleles and number of secretor
alleles, and diagnosis for the analyses including all diagnostic groups, scanner field strength was included as a
random effect.
University of Southern California
121
FIGURE 4-2
Adjusted odds-ratio by ABO phenotype. ABO O blood type carriers were found to be at a significantly
greater increased risk of AD, compared to all other blood types. Odds ratios shown for each variable on the y-
axis are after adjusting for all other variables included in the model.
Adjusted Cox Proportional Hazards Regression
Our unadjusted results with Cox Proportional Hazards regression with ABO blood
type showed a significant difference between age of AD onset, such that O carriers were at
risk for earlier onset of AD (p = 0.005). These results were particularly driven by
differences between A and O, likely as a result of lower sample sizes for the other blood
types. After adjusting for sex, education, and the presence of APOE ε4 alleles our analyses
showed a nominal association of ABO blood type with age of AD onset, again with O
carriers showing an earlier onset (p=0.054). The median age-of-onset for O blood type
University of Southern California
122
carriers was 82.1, 84.8 for A carriers, 85.5 for B carriers, and 86.9 for AB carriers. These
results are presented in Figure 3. Our overall model was significant both before and
nominally after adjusting for confounds.
Plasma Interactions
After adjusting for age, sex, education, number of secretor alleles, and whether or
not an individual had any APOE ε4 alleles, we found a significant negative interaction
between the plasma marker receptor for advanced glycation end products (RAGE) and O
blood type in predicting risk for AD (p = 0.014). Number of secretor alleles was significant
(p = 0.018) while the individual predictors O blood type and RAGE were not significant (p =
0.078, p = 0.14, respectively). We followed up these findings by confirming this interaction
in a subsample of ADNI-1 individuals (N=401) with plasma measurements at 12-months.
Again, RAGE was not significant (p = 0.23), but number of secretor alleles was (p = 0.01), as
was O blood type (p = 0.009), and the interaction between O blood type and RAGE
remained significant (p = 0.019).
University of Southern California
123
FIGURE 4-3
Cox Proportional Hazards Ratio for Alzheimer’s age of onset by ABO blood type. Dotted lines
correspond to the median age-of-onset for the connected blood type. O blood type was associated with earlier
age of onset than all other blood types, both in a univariate model and after adjusting for sex, education, and
presence of the ε4 allele.
University of Southern California
124
FIGURE 4-4
Brain Expression Analysis. Networks were derived by training with the top established highly-penetrant
genes for Alzheimer’s, as shown in green (APP, PSEN2, APOE); PSEN1 was not available across datasets. The
top 20 established AD-risk genes were used along with FUT2 and ABO for gene prioritization based on co-
expression across datasets and brain regions. FUT2 was among the top 5 prioritized genes. Here we show
these top 5 genes along with the highly-penetrant genes to understand their interactions. ABO, while not
among the top prioritized genes, is included here to understand the specific interactive pathway of this gene.
Discussion
As it is one of the most pleiotropic genes in the genome, ABO blood type has been
associated with a number of diseases, largely implicating the role of blood type in the
cardiovascular and immune systems (Franchini & Liumburno, 2013; Yang et al., 2014).
Although blood type has been investigated in stroke research, showing that O blood type is
protective, few studies have directly assessed how ABO blood type affects brain structure
during the aging process, in both healthy and disease conditions. Moreover, to date, no
studies have looked specifically at ABO genotype, rather than ABO phenotype, nor at the
University of Southern California
125
interaction with FUT2 and ABO. In this study, we carried out an extensive analysis,
including tensor-based morphometry, brain expression, and a composite AD brain metric
to determine the impact of O alleles on brain structure in the aging brain. We applied these
analyses to cognitively normal, MCI, and AD participants within ADNI and AddNeuroMed.
We then determined whether or not ABO blood type was an important predictor of risk for
AD. We found evidence of significant brain structure differences by number of O alleles
after accounting for interactions with FUT2, which defines secretion status of the ABO
precursor h antigen across tissues. These results are contrary to prior literature showing
positive associations between O blood type compared to A or other blood types and
temporal lobe volume (De Marco & Venneri, 2015). However, these results are consistent
with where in the brain FUT2 is known to be expressed (Lipovich et al., 2013; Xu et al.,
1998; Yamamoto et al., 2013). Although it was previously speculated that O blood type
might lead to protection from AD, given the negative association with O blood type and
hippocampal volume presented here as well as our subsequent results showing an
increased odds-ratio of risk for AD in O blood type individuals, our results provide evidence
to the contrary (De Marco & Venneri, 2015).
More broadly, the majority of regions identified aside from the hippocampus are
involved in coordinated activity involving visual perception and recognition, and possibly
related motor planning and execution (Hanakawa et al., 2008; Horn 2006). How these
regions specifically subserve risk for AD is not immediately evident. However, prior
research suggests a role of decreased motor function in AD (Buchman & Bennett 2011). It
may be that blood type increases risk for mixed pathology dementia involving Lewy bodies,
which has more commonly been shown to accumulate in the identified regions (Braak et
University of Southern California
126
al., 2003). Indeed, individuals with MCI exhibiting Parkinsonism have a higher risk of
subsequent development of AD (Aggarwal et al., 2006).
A number of limitations to our study deserve mention. Although we found
significant interactions with RAGE and O blood type, we could not assess cerebrospinal
fluid (CSF) levels of RAGE as these were not available in AddNeuroMed and did not pass
established quality control procedures for ADNI. Yet, CSF measures have shown to provide
more direct insight into AD pathology. Thus, future follow-up exploring the interactions
between RAGE and ABO blood type in mediating risk for AD may offer improved
mechanistic insights by including these measures. Moreover, it will be important to assess
whether these findings hold for all-cause dementia, or if they are specific to AD. Finally, as
our population was racially and ethnically homogeneous, we could not assess interactions
with race, although these may play an important role given known differences by race and
cardiovascular disease and AD (Sheffler et al., 2014; Tang et al., 1998; Yusuf et al., 2001).
In conclusion, we found that studying ABO genotype offers insights into the role of
blood type in mediating risk for AD and helps clarify confusion with previous literature
finding larger brain volumes in O phenotype carriers. We showed that both genotype and
phenotype, however, are important risk factors for AD, with O carriers and number of O
alleles being significantly associated with increased risk for AD. Higher plasma levels of
RAGE were found to potentially mitigate some of these risks in O carriers. This study is the
first to show an interaction between blood type and RAGE and offers novel mechanistic
avenues for future research.
U n i v e r s i t y o f S o u t h e r n C a l i f o r n i a
127
5 MAPPING THE HERITABILITY CONTENT OF
DEVIATIONS FROM OPTIMAL BRAIN AGING
Abstract
Studying human brain aging allows us to understand how healthy an individual’s
brain appears for their age, while also offering insights into deviations from the normal
trajectory. For instance, debate exists as to whether neurodegenerative diseases follow this
trajectory at an accelerated rate, or whether conditions such as Alzheimer's disease
indicate an altered aging trajectory altogether. Understanding heritability, or the
proportion of phenotypic variance that is due to genetic factors, provides further insights
into whether or not optimal brain aging is itself genetically influenced, or whether ancillary
genetically linked conditions, such as cardiovascular disease, are the primary drivers in
deviations from this normal trajectory and consequently drivers of what is known about
the heritability of brain structure. To better answer these questions requires large-scale
studies across the aging spectrum. In this work we use a healthy group of 1,365
participants (M age 59.81 ± 7.19), who we call “optimal agers”, to learn models of brain
trajectories using quantile regression across 234 structural brain regions, networks, and
laterality measures, and apply these learned models to a larger dataset from the UK
Biobank (N=8,307, M age 63.06 ± 7.35). Using these learned models, we determined the
residual differences for each subject and each region, allowing us to then determine how
heritable deviations from an optimal aging trajectory are using mixed effects models and a
University of Southern California
128
genetic relatedness matrix. Overall, our results suggest that deviations in laterality based-
measures are largely not heritable, though many measures of surface area, thickness, and
volume are. Many regions show differences by age group or sex. The regions showing the
largest heritability were thickness of the left and right lingual gyrus (h
2
: 0.27 and 0.26,
respectively), thickness of the left isthmus cingulate (h
2
: 0.23), and composite thickness
measures of the default mode network and frontoparietal networks (h
2
: 0.21 and 0.19,
respectively). Analyses of the residuals indicated that much of the variation was accounted
for by the presence of broad-scale cardiovascular diseases, supporting the important role
of cardiovascular disease in deviations from an optimal aging trajectory, and the important
consideration of disease processes in understanding narrow-sense heritability of the brain.
Introduction
Studying human brain aging across a population allows us to understand how
healthy an individual’s brain appears for their age, while also providing insights into
deviations from a normal trajectory. For instance, debate exists as to whether
neurodegenerative diseases follow a normal trajectory at an accelerated rate, or whether
conditions like Alzheimer's disease indicate an altered aging trajectory altogether.
Moreover, aging has largely been medicalized, while age-associated diseases are not
inevitable results of the aging process. The influence of genetic variations in understanding
the aging brain in both health and disease are thus of great interest. Studies of brain
imaging and genetics seek to understand the degree to which genetics can explain variation
in structural or functional imaging measures. Aside from genome-wide association studies,
this has traditionally been assessed using pedigree information in twin or family-based
University of Southern California
129
studies but can also be determined by measuring heritability across populations of
unrelated individuals (Bartels et al., 2003). Heritability is defined as the proportion of
observed phenotypic variance that can be explained by variation in genetic factors, most
often using genome-wide single nucleotide polymorphisms (SNPs) to assess additive, or
narrow-sense heritability (Dou et al., 2017; Visscher et al., 2008). As SNP heritability does
not capture non-additive genetic variation, narrow-sense heritability provides a lower-
bound on defining how genetically determined variation in a phenotype is and can thus be
useful in guiding downstream targeted analyses or for understanding the utility of
therapeutic interventions.
Importantly, heritability is not static, and many heritability estimates, particularly
those relevant to the aging process show differences over time, such that many traits show
lower heritability in younger populations (Brooks-Wilson, 2013). Heritability differences
have also been observed between sexes, under certain environmental conditions, or for
particular ethnic backgrounds (Guadalupe et al., 2017; Newsome et al., 2016). Yet, as
sample size plays a large role in the amount of phenotypic variance observed, and thus the
amount of variance that can be explained by genetics, addressing these complexities has
been difficult. The United Kingdom (UK) Biobank provides a rare opportunity to address
targeted questions related to heritability of traits across middle-age and older adults
(Sudlow et al., 2015). This is particularly pertinent to understanding heritability of brain
metrics, as the UK Biobank offers one of the largest single cohort neuroimaging datasets to
date. In this work, we sought to determine how heritable deviations from optimal aging
are, using a set of healthy individuals to learn models across brain regions and then applied
these learned models to a wider range of participants. We used the residual differences
University of Southern California
130
from the models to address the question of how heritable deviations from the optimal
aging trajectories are across 234 different regions of interest, including volume, thickness,
surface area, laterality, and network measures. We hypothesized that not all regions would
show significant heritability, that certain regions would show significance only in the
younger or only in the older participants, and that some regions would show significance in
only one sex. As the frontoparietal and default mode networks have been implicated in
important changes during normal aging, as well as diseases such as Alzheimer’s, we
hypothesized that these composite network measures would show more consistent
heritability results than most individual regions (Fair et al., 2008; Li et al., 2015; Raichle et
al., 2015).
TABLE 5-1
Demographics
Optimal Agers Heterogenous Agers
N 1,365 8,307
Age (Range) 59.81 (46-77.4) 63.06 (45.2-79.3)
Sex (% F) 70.11% 49.81%
Education (ISCED) 16.95 (3.5) 16.5 (3.9)
APOE4 Count (0/1/2) 973/366/26 6013/2108/186
Demographics. Basic demographic information for the UK Biobank participants included here, grouped by the
optimal agers or heterogeneous agers we identified.
Methods
Participants
All data used here were from the UK Biobank (Application ID#11559; July 2017
release). At the time of analyses, we had access to magnetic resonance imaging (MRI) and
University of Southern California
131
genetic data for 9,672 participants (M age 62.6 ± 7.4). Basic demographic information by
subgroup is provided in Table 1. To date, the UK Biobank is the largest prospective study
of aging, collecting detailed health, physical, and lifestyle information on more than
500,000 middle aged and older adults. All participants were recruited between 2006-2010
with a subset of individuals completing neuroimaging at a follow-up visit around four years
after their baseline visit.
In this study, we examined structural brain measures in the full population of
participants with both imaging and genetic data but divided these data into two groups:
‘optimal agers’, and a more heterogeneous aging cohort. Our optimal aging group was used
to more adequately identify aging trajectories in order to establish trajectories that were
separate from disease, such that we could then apply them to the broader heterogeneous
aging cohort. Subject exclusion for the broader cohort were limited to safety factors that
prevented neuroimaging from being performed (e.g. participants with a cardiac pacemaker
or cochlear implant, pregnancy, metal fragments, etc.), or participants who had an MRI that
did not pass quality control (QC). Optimal agers were defined as those without chronic age-
associated diseases using a comprehensive biopsychosocial model similar to established
studies on optimal aging (Anstey et al., 2007; Crooks et al., 2008; Erikson et al., 2016).
Specifically, exclusion criteria were neurological conditions, a history of psychiatric
conditions identified through self-report and the 10
th
revision of the International
Statistical Classification of Diseases and Related Health Problems (ICD-10; all Chapter V
and VI conditions), current or past history of malignant neoplasms, human
immunodeficiency virus, thyroid disease, diabetes, endocrine conditions, ischemic heart
disease, cardiovascular or cerebrovascular disease, atherosclerosis, hypercholesterol,
University of Southern California
132
hypertension, arthritis, chronic kidney disease, chronic obstructive pulmonary disease,
waist-hip ratios above established norms for the participant’s sex, a history of alcohol
consumption beyond the UK recommended guidelines and current smokers. Additional
exclusion criteria as measured using self-report from a mental-health questionnaire
included poor health satisfaction and poor family or friends satisfaction. We also excluded
for history of significant head or cranial injury and contraindications for MRI. All
participants provided informed consent prior to study participation.
Genotyping
All UK Biobank samples were genotyped using either the UK Believe platform
(N=49,979) or the UK Biobank axiom array (N=102,750).
Neuroimaging Acquisition
Participants completed a 31-minute neuroimaging protocol at the UK Biobank
imaging center in Cheadle, Manchester, UK. All scans included here were collected on a
single scanner. Sagittal 3D structural T1-weighted MPRAGE scans were acquired using a
Siemens Skyra 3 T scanner using VD13A SP4 software with a 32-channel RF receiver head
coil (TI/TR: 800/2000 ms; voxel size: 1.0 x 1.0 x 1.0 mm
3
). Scans were pre-scan normalized
using an on-scanner bias-field correction filter and standard shimming was applied. More
detailed descriptions of the scanning protocol may be found in Miller et al. (2016).
Image Processing
At the time of this analysis, cortical surface area and thickness were not available
from the UK Biobank data showcase. We therefore independently extracted 68 cortical
University of Southern California
133
thickness and surface area measures across the left and right hemispheres using FreeSurfer
v.5.3 (https://surfer.nmr.mgh.harvard.edu/) and the Desikan-Killiany atlas (Fischl et al.,
2002). We used established ENIGMA protocols for cortical extraction and QC on all
FreeSurfer segmentations (http://enigma.usc.edu/protocols/imaging-protocols/).
Participants with scans that failed the visual QC entirely, due to either major anatomical
abnormalities or scan artifacts, were excluded from further analysis. Scans with minor
segmentation errors were included in the analysis after the affected ROIs were removed
and replaced with imputation using Expectation Maximization (Moon 1996). For
subcortical volume measures, we used the 14 bilateral imaging-derived phenotypes (IDPs)
released by UK Biobank (thalamus, pallidum, putamen, caudate, amygdala, hippocampus,
and nucleus accumbens). These measures were derived using FMRIB’s Integrated
Registration and Segmentation Tool (FSL) (Smith et al., 2004). The laterality index (LI) was
calculated for all of these bilateral regions, using the established formula: (right –
left)/(right + left) for each corresponding region (Bullmore et al., 1995).
Network Composite Measures
To better understand network level structural deviations, we applied a functionally
defined atlas of the default mode and frontoparietal networks as in Riedel et al. (2017) to
our thickness measures, allowing us to create composite measures across both
hemispheres, as well as for each hemisphere separately. More detailed information on the
construction of this network can be found in Yeo et al. (2014). A list of the regions involved
in construction of these networks used here are outlined in Table 2 and displayed in
Figure 1.
University of Southern California
134
TABLE 5-2
Regions used for our default mode network and frontoparietal network composite measures. We
constructed 3 measures for each of the 2 networks, a bilateral composite measure derived from the summation
of thickness measures listed for the corresponding network, and one for each hemisphere separately.
FIGURE 5-1
Visual representation of regions used for our default mode network and frontoparietal network
composite measures. Specific regions are outlined in Table 2.
Determining Residuals
We used imaging data from our optimal agers to fit models using quantile median
regression for each brain metric separately. All statistical analyses were performed using R
University of Southern California
135
version 3.4.4. (R Core Team, 2013). For all brain measures, we calculated 7 different
models with the brain measure as the outcome variable and then applied the learned
model to the heterogeneous agers as follows (note: intracranial volume (ICV) was not used
as a covariate for thickness measures):
Model 1: ICV + Sex + Age + (Age centered)
2
+ Education + APOE-ε4 allele
number + Waist hip ratio
Model 2: ICV + Sex + Education + APOE-ε4 allele number + Waist hip ratio
Model 3: ICV + Age + (Age centered)
2
+ Education + APOE-ε4 allele number +
Waist hip ratio
Model 4: ICV + Education + APOE-ε4 allele number + Waist hip ratio
Model 5: ICV + Sex + Education + Waist hip ratio
Model 6: ICV + Age + (Age centered)
2
+ Education + Waist hip ratio
Model 7: ICV + Education + Waist hip ratio
Heritability Analysis
For each model that resulted in significant prediction of the brain measure in the
optimal agers group, we determined the residual difference between predicted and actual
brain measures for each individual, and then used these residuals to run heritability
analyses. We first constructed a genetic relationship matrix (GRM) using 81,667 genotyped
markers across all autosomes that were selected to be independent R2 < 0.2 with a minor-
allele frequency > 0.05 using PLINK (Purcell et al., 2007). The GRM was then calculated
using RareMetalWorker, a forerunner of RareMetal (Feng et al., 2014). Covariance
components were estimated using the Newton-Raphson iterative procedure with the
Average Information Restricted Maximum Likelihood (AIREML) mixed effects algorithm.
Fixed effects included 4 multidimensional scaling (MDS) components to account for
population structure and ethnicity differences, as calculated using an independent set of
clustered SNPs derived from genome-wide data. Models 2-7 included grouping variables, in
University of Southern California
136
which separate residual variance components are fit for each grouping variable (Conomos
et al., 2015). These grouping variables were as follows:
Model 2: 3 age bins of equal sample size (44-56, 56-66, 66-79)
Model 3: sex
Model 4: age bins x sex
Model 5: age bins x APOE-ε4 (+/-)
Model 6: APOE-ε4 (+/-) x sex
Model 7: sex x age bins x APOE-ε4 (+/-)
Within the models, the GRM was included as a random effect to model heritability. The
significance of each model, and corresponding heritability, for each region, was determined
by comparing a null model, in which the random effects were not included, to the full
model. These statistics were conducted using likelihood ratio tests (LRT).
Explaining Residual Differences using ICD Codes
Using the residuals from model 1 where heritability was significant, we sought to
determine how broad-scale ICD disease codes, coded as presence/absence of disease
conditions, accounted for the residual differences. These codes included: A/B) Infectious
and parasitic diseases, C) Neoplasms, D) Benign Neoplasms and Diseases of the blood and
blood-forming organs and certain disorders involving the immune mechanism, E)
Endocrine, nutritional and metabolic diseases, F) Mental and behavioral disorders, G)
Diseases of the nervous system, H) Diseases of the ear and mastoid process, I) Diseases of
the circulatory system, J) Diseases of the respiratory system, K) Diseases of the digestive
system, L) Diseases of the skin and subcutaneous tissue, M) Diseases of the musculoskeletal
system and connective tissue, N) Diseases of the genitourinary system, O) Pregnancy,
childbirth and the puerperium, P) Certain conditions originating in the perinatal period, Q)
Congenital malformations, deformations and chromosomal abnormalities, R) Symptoms,
University of Southern California
137
signs and abnormal clinical and laboratory findings, NOS, S/T) Injury, poisoning and
certain other consequences of external causes, V/W/X) External causes of morbidity and
mortality, Y) Other, Z) Factors influencing health status and contact with health services.
Measures of proportion of variance in the residuals explained for a given region were
determined by partial eta squared effects across all binary ICD codes. Partial eta squared
measures the effects of any one variable after partialing out all other variables and
interactions (Cohen, 1973).
Results
For the sake of brevity, and the more nuanced interpretations required for the latter
models due to interactions with grouping variables and sample size differences, we focus
our results and discussion on models 1-3. Briefly, these latter models showed larger
heritability in individuals with the APOE-ε4 than individuals without APOE-ε4 for nearly all
brain regions and age groups. For instance, in model 5 the average heritability, where
significant, ranged from 0.13-0.21 in APOE-ε4 negative participants, showing a decreasing
trend with increasing age, and ranged from 0.25-0.32 in the APOE-ε4 participants, with no
trend in change with increasing age. In model 4, males also showed greater average
heritability than females across age groups, with declining trends with increasing age for
both sexes (M: 0.20-0.41; F: 0.15-0.19). These trends were maintained in model 6
considering APOE-ε4 by sex but showed the largest average heritability in female APOE-ε4
carriers (M x APOE: 0.18 vs. 0.25; F x APOE: 0.12 vs. 0.28). In model 7, increased average
heritability across regions for female APOE-ε4 carriers vs. male APOE-ε4 carriers was most
prominent in the oldest group (M: 0.20; F: 0.44).
University of Southern California
138
The top overall results or top group-specific results for models 1-3 are displayed in
Table 3.
TABLE 5-3
Top significantly heritable regions for the first 3 models. Specific covariates included in each model are
as outlined in the methods. Significance is shown in the middle column in red under each model and specific
group, while the corresponding heritability is shown in the far-right column for each model.
Model 1
Results for all significantly heritable regions are displayed in Figure 2. The most
significantly heritable regions included bilateral thickness of the lingual gyrus (h
2
L: 0.27,
R: 0.26, (p < 1x10
-12
)), thickness of the left isthmus cingulate (h
2
L: 0.23, (p < 1x10
-9
)),
surface area and thickness of the left superior temporal gyrus (h
2
Thk/SA: 0.23, (p < 1x10
-
University of Southern California
139
9
)), thickness of the right parahippocampal gyrus (h
2
0.22, (p < 1x10
-10
)), the default mode
network composite measure (h
2
L: 0.21, (p < 1x10
-8
)), and the left and right hemispheric
composite measures of the default mode network (h
2
L: 0.21 (p < 1x10
-8
), R: 0.19, (p <
1x10
-7
)). The regions showing the weakest significance at p < 0.05 were the volume of the
right parahippocampal gyrus (h
2
=0.06, p=0.04) and the thickness of the right rostral
anterior cingulate gyrus (h
2
=0.07, p=0.035).
University of Southern California
140
FIGURE 5-2
Significantly heritable regions for model 1. Heritability is shown on the outside of the circular plot, with
brighter yellow regions indicating higher heritability. The middle circle denotes significance, with darker
purple regions having higher significance.
Model 2
Results for all significantly heritable regions are displayed in Figures 3-5. The
following regions were significantly heritable across all age groups: thickness of the right
University of Southern California
141
parahippocampal gyrus, surface area of the right middle temporal gyrus, and thickness of
the left superior temporal gyrus. Additionally, thickness of the left isthmus cingulate was
significantly heritable across age groups but showed a trend towards declining heritability
with age (44-56: 0.40, 56-66: 0.28, 66-79: 0.18), as did surface area of the left precuneus
(44-56: 0.43, 56-66: 0.19, 66-79: 0.17). Surface areas of the left inferior parietal lobe, left
paracentral gyrus, left precentral gyrus, left superior parietal lobe, and volumes of the left
amygdala and left and right putamen showed significance only in the young group (H
2
0.31-0.49). However, more regions showed significant heritability in the older age groups
than the younger group. Exhibiting significant heritability in both of the older 2 age groups
(e.g. 56-66 & 66-79), but not the young group, were thickness of the left and right inferior
parietal lobe, thickness of the left postcentral gyrus, and surface area of the right precuneus
and right entorhinal cortex (h
2
0.17-0.29). Finally, regions only showing significance in the
oldest age group included surface area of the left caudal middle frontal gyrus, the
frontoparietal network, surface area of the left caudal anterior cingulate gyrus (h
2
0.18-
0.25).
University of Southern California
142
FIGURE 5-3
Significantly heritable regions for model 2 for the young group (44-56). Heritability is shown on the
outside of the circular plot, with brighter yellow regions indicating higher heritability. The middle circle
denotes significance, with darker purple regions having higher significance.
University of Southern California
143
FIGURE 5-4
Significantly heritable regions for model 2 for the middle-aged group (56-66). Heritability is shown on
the outside of the circular plot, with brighter yellow regions indicating higher heritability. The middle circle
denotes significance, with darker purple regions having higher significance.
University of Southern California
144
FIGURE 5-5
Significantly heritable regions for model 2 for the old-aged group (66-79). Heritability is shown on the
outside of the circular plot, with brighter yellow regions indicating higher heritability. The middle circle
denotes significance, with darker purple regions having higher significance.
Model 3
Results for all significantly heritable regions are displayed in Figures 6-7. Regions
showing significant heritability in both sexes were often more heritable for males than
females. These regions predominantly included measures of surface area and thickness,
University of Southern California
145
such as thickness of the right parahippocampal gyrus (h
2
M: 0.27, F: 0.23 (p < 0.005),
thickness of the right isthmus cingulate (h
2
M: 0.26 (p < 0.005), F: 0.15 (p = 0.014)), surface
area of the right insula (h
2
M: 0.28 (p < 0.0002), F: 0.11 (p = 0.045)), left insula (h
2
M: 0.25
(p < 0.0008), F: 0.12 (p = 0.029)), and thickness of the left isthmus cingulate (h
2
M: 0.33 (p
< 0.0001), F: 0.26 (p < 0.0006)). Many regions across surface area, thickness, and volume
measures were significant in only one sex. For instance, surface area of the right precuneus
(h
2
M: 0.33 (p < 0.0001), F: 0.08 (p = 0.09)) and surface area of the right banks of the
superior temporal sulcus (h
2
M: 0.06 (p = 0.26), F: 0.25 (p < 0.0002)). The frontoparietal
and default mode network measures were consistently heritable for both sexes but showed
a trend for greater heritability in females (h
2
M: 0.16, 0.18; F: 0.21, 0.25).
University of Southern California
146
FIGURE 5-6
Significantly heritable regions for model 3 for males. Heritability is shown on the outside of the circular
plot, with brighter yellow regions indicating higher heritability. The middle circle denotes significance, with
darker purple regions having higher significance.
University of Southern California
147
FIGURE 5-7
Significantly heritable regions for model 3 for females. Heritability is shown on the outside of the circular
plot, with brighter yellow regions indicating higher heritability. The middle circle denotes significance, with
darker purple regions having higher significance.
Most and Least Consistent Regions
Regions that were significant in our quantile regression step, but never significantly
heritable primarily included laterality measures, especially LI measures of thickness,
though LI surface area and volume measures also showed a low frequency of significant
University of Southern California
148
heritability. Other measures that were consistently not found to be heritable included
bilateral surface area measures of the frontal pole, and thickness of the left temporal pole.
The most consistently heritable regions across groups included thickness of the left and
right isthmus cingulate, surface area of the right superior temporal lobe, thickness of the
right parahippocampal gyrus, surface area of the left insula, surface area of the middle
temporal lobe, LI for surface area of the precuneus, and our composite network regions of
the default mode and frontoparietal networks.
Effects of ICD Codes on Residual Differences
The impact of ICD codes varied between significantly heritable brain regions, but
the majority of regions showed association with multiple ICD codes. A subset of these
results with high variance explained are displayed in Figure 8. ICD codes P and Q showed
the lowest prevalence (< 1% of participants) and did not explain much of the residual
variance. ICD code I, representing cardiovascular diseases, and present in 23.4% of the UK
Biobank participants included here, most consistently showed high explanatory power and
partial eta squared values across regions. Diseases of the genitourinary system, (N codes;
22.3% of participants), showed high eta squared values for our network measures, as did E
codes (11.7 % of participants), endocrine, nutritional and metabolic diseases. Thickness
and surface area of the right pericalcarine showed the least association with any ICD codes.
University of Southern California
149
FIGURE 5-8
Partial eta squared effect size by ICD code on brain region residuals. Each ICD category is as outlined in
the methods. I category codes, representing cardiovascular disease, consistently showed higher effects than
most other diagnostic groupings.
Discussion
In this work, we presented the first study to date assessing heritability of deviations
from an optimal aging trajectory. We included a large sample of 9,672 total subjects and
explored heritability across 234 different structural brain metrics, including surface area,
thickness, volume, frontoparietal and default mode network composite thickness
measures, and laterality index measures for surface area, thickness, and volume. We
determined that many of the deviations from an optimal aging trajectory were heritable
across different subgroup analyses, but that specific age groups and sex are important
considerations that moderate heritability levels. Heritability largely increased with age, but
University of Southern California
150
the reverse was found for multiple regions as well. Moreover, we determined that
deviations in laterality metrics were largely not heritable. Finally, we identified a trend for
sex differences in heritability of brain metrics, such that males largely showed greater
heritability across age groups. However, when considering interactions between sex and
APOE-ε4 positivity, we found a reverse trend, such that females showed greater
heritability. These latter results are novel in that no study on heritability has considered
interactions with APOE on brain structure. Yet, much is known about the importance of this
gene on the brain, and APOE-ε4 has been shown to have a greater impact in females in
mediating risk for AD, supporting the results of these analyses (Fisher et al., 2018; Riedel et
al., 2016).
A few limitations of our study deserve comment. Although this is to be expected
given our stringent exclusion criteria, our optimal aging cohort was smaller than our
heterogeneous aging cohort. Thus, it will be important to replicate these findings with a
larger group of healthy participants. However, the single-site design of the scanning
protocols within UK Biobank offers an unmatched level of consistency that likely addresses
this drawback to some extent. We also did not explore medication use. It is possible that
using ICD diagnoses to identify individuals meeting our inclusion and exclusion criteria for
our optimal aging group, led to the inclusion of participants who were taking medications
in line with the presence of one of the chronic diseases used for our exclusion criteria, or
who were taking medications known to impact brain structure. Finally, the average age of
the UK Biobank cohort is young for an aging focused study, with an average age of 62.5.
Given the sample distributions across age, we evaluated participants in 3 age bins, where
our oldest bin was 66-79. It is likely that exploring the heritability of regions showing age
University of Southern California
151
specific trends will be improved further by including more individuals over age 75 and
exploring groups in smaller age bins.
In conclusion, deviations from an optimal aging trajectory are largely, but
differentially, heritable. Heritability was generally consistent with prior studies of broad-
scale heritability of brain structures, but the lower results we present here also suggest
that the higher estimates may be confounded by age-associated diseases, such as
cardiovascular disease and endocrine disorders (Bryant et al., 2013; Pol et al., 2006). By
identifying an optimal aging cohort, we were able to help inform disease-independent
trajectories that may guide future follow-up studies and more targeted approaches.
U n i v e r s i t y o f S o u t h e r n C a l i f o r n i a
152
6 INVESTIGATING THE RELATIONSHIP
BETWEEN BRAIN PREDICTED AGE AND
GENOMICS
Abstract
Research shows that people age at unique rates, likely due to genetic and
environmental influences, and their interactions. Neuroimaging and genetic characteristics
of different aging rates may yield an important suboptimal brain aging biomarker,
improving early predictions of age-associated diseases, such as Alzheimer’s disease (AD).
Using structural T1-weighted brain scans from the United Kingdom (UK) Biobank study
and gradient boosting machine learning methods, we estimated brain predicted age in
N=10,091 middle-age and older adults (M age 62.4 ± 7.4). Using an ‘optimal aging’ subset of
the cohort for both cross-validation and hold-out validation, we show improved
generalizability of the brain-age metric. We then performed a mixed-effects genome-wide
association study across 9,903 of these individuals with brain-age as the phenotype and
performed downstream gene-scoring and pathway enrichment. Our results revealed a
significant variant in the KALRN gene (p < 1.4x10
-8
) and significant gene-enrichment in
genes overlapping with Alzheimer’s disease, such as ADAMTS10 (p < 6.8x10
-5
), IL34 (p <
3.6x10
-4
) and PIK3IP1 (p < 9.1x10
-4
). Pathway-enrichment revealed enriched pathways
involved in developmental biology, axon guidance, glucose and lipid metabolism, VEGF
signaling, apoptosis, and other pathways involved in brain development and maintenance.
University of Southern California
153
These results support the utility of our brain-age framework and suggests that this
measure might be useful in identifying brain health and suboptimal brain aging
trajectories.
Introduction
There is substantial heterogeneity in how aging processes affect different
individuals, and people age at distinct rates. These differences are due to both genetic and
environmental influences, and their interactions. Identifying individual deviations from
normal aging would allow for improved understanding of the biological characteristics of
these. Indeed, metrics such as heart-age serve this purpose in understanding
cardiovascular health (Lopez-Gonzalez et al., 2015). Not only has heart-age been shown to
be a reliable metric for research purposes, it has also proven motivational to the patient
and improved cardiovascular outcomes (Lopez-Gonzalez et al., 2015). Determining such a
metric of brain-age may potentially serve a similar role in understanding brain health and
improving patient outcomes. Moreover, these biomarkers, representing deviations from
normal brain aging may ultimately help improve early predictions of age-related diseases,
such as Alzheimer’s disease (AD) or Parkinson’s disease, potentially before irreversible
neurodegenerative processes have occured.
In this study, we sought to determine brain-age using structural T1-weighted scans
from the United Kingdom (UK) Biobank by using a method developed in our lab,
COMPOSED, with gradient boosting machines to determine brain-age in a large population-
based sample (N=10,091). We hypothesized that deviations in brain-age would be
associated with specific genetic variants that might predispose certain individuals to
suboptimal brain aging. A better understanding of the subsequent genes and pathways that
University of Southern California
154
underlie variation in brain-age and thus risk to diseases may lead to better treatment
outcomes and mechanistic insights, we performed both gene and pathway enrichment with
our results.
Methods
Participants
The data used here were from the UK Biobank (Application ID #11559; July 2017
release). At the time of analyses, we had access to magnetic resonance imaging (MRI) and
genetic data for 10,091 participants (M age 62.4 ± 7.4). Basic demographic information by
subgroup is provided in Table 1. To date, the UK Biobank is the largest prospective study
of aging, collecting detailed health, physical, and lifestyle information on more than
500,000 middle aged and older adults. All Biobank participants were recruited between
2006-2010; a subset of which had neuroimaging completed during their follow-up visit
approximately four years after their baseline visit.
Additional Training Cohorts
To improve training and subsequent prediction of brain-age on older individuals
within UK Biobank, we incorporated a subset of stable ( > 6 months) cognitively normal
individuals from the Alzheimer’s Disease Neuroimaging Initiative (ADNI; N = 32) and the
Open Access Series of Imaging Studies (OASIS; N = 38), M age 80.5 ± 3.4. These participants
were used for training only.
University of Southern California
155
TABLE 6-1
Demographics
Training (UKB)
Training
(ADNI/OASIS)
Hold Out Group
N 1,152 70 8,939
Age (SD) 60.2 (7.4) 80.5 (3.4) 62.7 (7.4)
Sex (% F) 69.3% 47.1% 50.3%
Education (ISCED) 16.9 (3.5) 16.6 (2.5) 16.5 (3.9)
APOE4 Count
(0/1/2)
769/304/23 56/12/2 6197/2169/188
Demographics. Basic demographic information for the UK Biobank participants included here, grouped by the
optimal agers and the more heterogeneous agers group, as defined in the methods section.
In this study, we examined structural brain measures in the full population of
participants with both imaging and genetic data (N=9,652) but divided these data into two
groups: optimal agers, and a more heterogenous aging cohort. A subset of both groups was
used for identifying associations between chronological age and brain metrics. Exclusion
for the broader cohort was limited to safety factors that prevented neuroimaging from
being performed (e.g., participants with a cardiac pacemaker or cochlear implant,
pregnancy, metal fragments, etc.), or participants who had an MRI that did not pass quality
control (QC). ‘Optimal agers’ were defined as those without chronic age-associated diseases
using a comprehensive biopsychosocial model similar to established studies on optimal
aging and in line with prior literature on health outcomes in aging (Anstey et al., 2007;
Christensen et al., 2006; Crooks et al., 2008; Erikson et al., 2016). Specifically, exclusion
criteria were neurological conditions, a history of psychiatric conditions identified through
self-report and the 10
th
revision of the International Statistical Classification of Diseases
University of Southern California
156
and Related Health Problems (ICD-10; all Chapter V and VI conditions), current or past
history of malignant neoplasms, human immunodeficiency virus, thyroid disease, diabetes,
endocrine conditions, ischemic heart disease, cardiovascular or cerebrovascular disease,
atherosclerosis, hypercholesterol, hypertension, arthritis, chronic kidney disease, chronic
obstructive pulmonary disease, waist-hip ratios above established norms for the
participant’s sex, a history of alcohol consumption beyond the UK recommended
guidelines, poor health satisfaction, poor family or friends satisfaction, and current
smokers. We also excluded for history of significant head or cranial injury and
contraindications for MRI. The UK Biobank study was approved by the North West Multi-
Centre Research Ethics Committee and all participants provided informed consent prior to
study participation.
Neuroimaging Acquisition
All participants completed a 31-minute neuroimaging protocol at the UK Biobank
imaging center in Cheadle, Manchester, UK. Scans included here were conducted on a
single scanner. 3D structural T1-weighted sagittal MPRAGE scans were acquired using a
Siemens Skyra 3 T scanner with VD13A SP4 software and a 32-channel RF receiver head
coil (TI/TR: 800/2000 ms; voxel size: 1.0 x 1.0 x 1.0 mm
3
). Using an on-scanner bias-field
correction and standard shimming, scans were pre-scan normalized. More detailed
information on the scanning protocol may be found in Miller et al. (2016). Participants
from OASIS and ADNI were scanned following similar protocols with 3 T scanners; details
for these studies have been described elsewhere (Jack et al., 2008; Marcus et al., 2007).
Image Processing
University of Southern California
157
At the time of this analysis, cortical surface area and thickness measures were not
available from the UK Biobank data showcase. We therefore, independently extracted 68
cortical thickness and surface area measures across the left and right hemispheres using
FreeSurfer v.5.3 (https://surfer.nmr.mgh.harvard.edu/) and the Desikan-Killany atlas
(Fischl et al., 2002). Established ENIGMA protocols were used for cortical extraction and
for performing QC on all FreeSurfer segmentations
(http://enigma.usc.edu/protocols/imaging-protocols/). Participants with scans that
failed the visual QC step entirely, due to either scan artifacts or major anatomical
abnormalities, were excluded from further analysis. Scans with minor segmentation errors
in particular regions only were included in the analysis after the affected ROIs were
removed and replaced using Expectation Maximization for imputation (Moon 1996). We
used the 14 bilateral subcortical volume measures from the imaging-derived phenotypes
(IDPs) released by UK Biobank (thalamus, pallidum, putamen, caudate, amygdala,
hippocampus, and nucleus accumbens). These measures were derived using FMRIB’s
Integrated Registration and Segmentation Tool (FSL) (Smith et al., 2004). Similar
procedures were applied for the OASIS and ADNI scans.
Brain-age Prediction
Within each study, we first regressed out intracranial volume (ICV) for all surface
area and volume measures to account for confounds in head size before using gradient
boosting machines (GBM) to determine relationships between chronological age and brain
metrics in our training and testing sets. We divided our data into a training/testing set and
a hold-out validation/prediction set. For training/testing 94% were optimal agers and the
remaining were cognitively normal older adults from OASIS and ADNI, for a total of 1,222
University of Southern California
158
participants used for training. This breakdown represented 80% of the total proportion of
optimal agers. The remaining portions of optimal agers and heterogeneous agers were
used for hold-out validation and prediction (N=8,939). This particular selection was made
as our early modeling indicated including a portion of optimal agers for hold-out returned
less overfit results, yielding improved performance on our prediction set and a more
robust externally valid model. We performed 10-times repeated 10-fold cross-validation,
following the methods outlined in Figure 1 to determine brain-age. Both sex and age were
used to learn relationships with chronological age and brain measures. These metrics were
used to then determine optimal merged sets of features within each measurement type
(i.e., thickness, surface area, volume), representing composite measures that served to
improve predictions of brain-age. To determine optimal GBM parameters we performed a
grid search.
Bias Correction
One common issue in machine learning methods that use model averaging to
improve overall prediction is a tendency to overestimate small values and to
underestimate the large values of a given distribution. This issue has been documented
across domains, including for brain-age predictions in adolescent and older age cohorts
(Chung et al., 2018; Cole et al., 2017; Song, 2015). Although simple linear regression has
been used to reduce this bias, an alternative approach that has shown improved
performance by Song (2015) has been to use the same machine learning method, such as
random forest, or gradient boosting as used here, to learn this bias with the training set and
correct for it. This is achieved here by first learning associations between chronological age
and brain structures to predict brain-age in the training group, calculating the residual
University of Southern California
159
difference between brain-age and chronological age, running gradient boosting with this
residual difference as the outcome using the same brain structures, and then combining the
two predictions (i.e. brain predicted age + residual difference prediction). This combined,
bias corrected prediction can then be applied to the holdout/validation data.
University of Southern California
160
FIGURE 6-1
Outline of the COMPOSED algorithm used for determining brain-age. COMPOSED stands for Classification
optimization using merged partitions over sex and disease and is a supervised machine learning method
designed with the goal of identifying patterns of optimally merged brain regions that improve classification or
prediction over the individual regions alone. X denotes structural brain measures derived from FreeSurfer, f
the selected structural brain measures, M the merged sets of structural brain measures to generate composite
regions, and Y represents either brain-age or the brain-age gap for the bias correction step; dCor: distance
correlation; AIC: Akaike information criterion.
University of Southern California
161
Genome-wide association analyses of brain-age
All UK Biobank samples were genotyped using either the UK Believe platform (N=49,
979) or the UK Biobank axiom array y (N=102,750). Individual markers were removed from the
analysis that did not meet the following quality control criteria based on previously established
parameters: genotype call rate < 95%, individual missingness rate > 10%, significant deviation
from Hardy-Weinberg equilibrium p < 5.7x10
-7
, minor allele frequency < 0.05, and a quality
control score < 0.15 (Wellcome Trust Case Control Consortium, 2007). After this quality control
5,400,389 SNPs remained for genome-wide association. We next calculated a genetic
relationship matrix (GRM) by identifying 81,667 genotyped markers across all autosomes
that were selected to be independent R
2
< 0.2, with a minor-allele frequency > 0.05 using
PLINK (Purcell et al., 2007). The GRM was then calculated using RareMetalWorker, a
forerunner of RareMetal (Feng et al., 2014). To control for population stratification in our
subsequent analyses we determined the continuous axes of genetic variation with
multidimensional scaling (MDS) in PLINK. Using the genotyped data, we calculated 4 MDS
components. These components, along with covariates of age, age-centered
2
and sex were
used as fixed effects in our mixed-effects genome-association models using the genome-
wide efficient mixed model association (GEMMA) algorithm (Zhou & Stephens, 2012). The
GRM was included as a random effects variable and the phenotype represented our
predicted brain-ages.
Gene-scoring and pathway-enrichment analysis
Gene-based tests reduce issues of multiple-testing and arguably give greater
biological meaning to association results (Ma et al., 2013). Therefore, we used the Pathway
University of Southern California
162
scoring algorithm, Pascal, to determine gene-scoring and subsequently pathway
enrichment (Lamparter et al., 2016). All variants that were included in our mixed-effects
genome-wide association analyses were included to determine gene scores. These scores
are based on the weighted sum of chi-squared statistics for a given gene, corrected for
linkage disequilibrium based on reference data from the European 1000 genomes
reference set (Durbin et al., 2012; Liu et al., 2010). Genes that are known to be in a
functionally related cluster within 50kb of each other are incorporated into fusion genes,
termed meta-genes. Pathway enrichment was then performed with these gene-based and
meta-gene scores. Pathway enrichment scoring involves converting gene scores with a chi-
squared quantile function, summing across all genes in a given pathway, and then obtaining
a p-value estimate using a Monte Carlo simulation (1,000,000 simulations) by sampling an
equivalent number of gene sets for the corresponding pathway. All pathways analyzed
include those that are defined by the Kyoto Encyclopedia of Genes and Genomes (KEGG),
Reactome, and Biocarta (D'Eustachio, 2011; Kanehisa, 2000; Nishimura, 2001).
Results
Brain-age Predictions
We show early results in Figure 2, which display the importance of including a
portion of the optimal agers in the holdout set to reduce overfitting as observed with the
high-density peaks in the training/testing group, but uniform distribution of the hold-out
set. Figure 3 shows the improved results after applying a bias correcting step during the
training/testing phase. After determining optimal parameters and validating these
parameters with our validation set of optimal agers, our brain-age prediction results
University of Southern California
163
returned a root-mean-square error of 6.40 in our training/testing data, a root-mean-square
error of 6.68 for our optimal agers validation group, and a root-mean-square error of 7.45
across all groups. A total of 90 different brain features were used for determining brain-age
after our feature selection step within COMPOSED. Parameters used were: 100 trees,
learning rate of 0.08, max depth of 5, 25 minimum sample split, a minimum of 5 samples at
a leaf node, 0.65 max features considered for a split, Friedman-MSE for determining GBM
splits and the Huber loss-function. The highest weighted features were surface areas of the
left lateral orbitofrontal cortex, the right pericalcarine, the left parahippocampal gyrus, as
well as composite measures of surface area of the right and left insula, and a composite
measure of total gray and white matter volume.
University of Southern California
164
FIGURE 6-2
Distribution of brain predicted age differences by training and hold-out sets using the optimal aging
group only and applying the model to the heterogeneous aging cohort. Although improved predictive
performance was achieved including a subset of the optimal group for hold-out validation rather than just using
this group for train/test cross-validation, these results still showed high level of overfit, as shown by the density
peak for the optimal training/testing group (i.e., > 40% peak at 0) and wide spread predictions for both hold-
out groups (i.e., range -22:18) with little evidence of a consistent peak around zero. This finding led us to
incorporate additional data from older individuals from two datasets, and to incorporate a bias correction step
prior to applying learned models to the hold-out and prediction groups.
University of Southern California
165
FIGURE 6-3
Distribution of brain predicted age differences for our final results. Training/testing was performed with
a group consisting of optimal agers from the UK Biobank and cognitively normal older (75+) controls from the
Alzheimer’s Disease Neuroimaging Initiative (ADNI) and the Open Access Series of Imaging Studies (OASIS).
The training set represented 80% of the total proportion of optimal agers from UK Biobank. The remaining
portions of optimal agers and heterogenous agers from the UK Biobank were used for hold-out validation (i.e.
optimal agers) and prediction of brain-age i.e. heterogeneous agers). Overall root-mean square error was 7.46
and differences between predicted and chronological age ranged from -19 to 26, with a tighter distribution
observed for the hold-out validation optimal agers than for the heterogeneous aging group.
Genome-wide Association
Results from our genome-wide association mixed-effects model of brain-age are
shown in Figures 4 and 5. The top variant identified was a common variant in the Kalirin
(KALRN) gene on chromosome 3, rs9843963 (p < 1.4x10-
08
). This gene was also significant
in our gene-scoring analysis (p < 3.3x10-
03
). However, the top genes from our gene-scoring
step were ADAMTS10 (p < 6.8x10-
05
), AHCTF1 (p < 8.7x10-
05
), and LINC00173 (p < 9.4x10-
05
). The top fusion-genes were SLC35A3-SLC30A7 (p < 4.5x10-
04
), STT3A-SRPR (p < 6.1x10-
University of Southern California
166
04
), UST-NUP43 (p < 7.0x10-
04
), and a family of olfactory receptor genes on chromosome 7
(p < 8.3x10-
04
). The top results from our pathway-enrichment analysis are shown in Table
2. This step revealed enriched pathways for developmental biology, axon guidance, glucose
and lipid metabolism, and components of the adaptive immune system.
University of Southern California
167
FIGURE 6-4
Quantile-quantile (QQ) plot of brain-age. Plot represents the observed –log10 significance values of the
tested variants on brain-age on the vertical axis and the expected -log10 values for brain-age on the
horizontal axis.
FIGURE 6-5
Manhattan plot showing genome-wide association results for brain-age. Our top variant was in the KALRN
gene and showed significance at the 5x10
-8
level, with multiple other variants across the genome showing
nominal significance at the 1x10
-7
level. Chromosome segment densities, indicating number of variants
included in the analysis for the corresponding chromosomal region, are displayed on the bottom portion of the
Manhattan plot and defined in the legend on the right.
University of Southern California
168
TABLE 6-2
Significantly enriched pathways (p < 0.05) discovered in the UK Biobank dataset for genetic markers
associated with predicted brain-age. To control the false discovery rate, significance was derived following
1,000,000 Monte Carlo simulations for each pathway, separately. Many of the enriched pathways are known to
be involved in developmental programming or maintenance of brain structure and function.
University of Southern California
169
Discussion
In this work, we presented the first study to date assessing the genome-wide single-
nucleotide variants, genes, and pathways involved in mediating differences in predicted
brain-age. We included a large sample of 10,091 individuals to model and predict brain-age
using structural brain measures of surface area, thickness, and volume. We then carried out
genome-wide associations of brain-age with a subset of these individuals (N=9,903). Our
brain-age predictions resulted in a root-mean-square error of 6.40 on our optimal agers
training set, 6.68 on our optimal agers hold-out set, and 7.45 overall. These results are
comparable to other brain-age prediction models using structural MRI that report this
metric on older adults (Cole et al., 2017; Le et al., 2018; Löwe et al., 2016; Richard et al.,
2018). Importantly, many models use the Pearson correlation to validate the results of
their methodology, but our preliminary analyses showed this to be a poor metric given the
more sigmoid shape of age prediction distributions, particularly when determining
machine learning parameters only using cross-validation. Specifically, young individuals
are more often predicted to be older, and old individuals more often predicted to be
younger than their corresponding chronological ages. Bias correction techniques have been
introduced to mitigate these effects, but still may not fully account for these differences and
thus evaluating results on a hold-out set, as done here, is necessary.
Our results provide information on coefficient weights across regions, allowing
determination of the relative importance of a given region or a set of combined regions for
predicting brain-age. The regions implicated were wide spread, but the most important
regions, as determined by GBM coefficient weight, were generally surface area measures
and summary measures. These included surface areas of the left lateral orbitofrontal
University of Southern California
170
cortex, the right pericalcarine, the left parahippocampal gyrus, as well as composite
measures of surface area of the right and left insula, and a composite measure of total gray
and white matter volume. These results point to the importance of considering the whole
brain in understanding brain aging and suggest that incorporation of other modalities that
are more sensitive to different structures or tissue types, such as diffusion imaging for
white matter, or FDG PET for glucose metabolism, may lead to further improvements in
predicting brain-age.
The most significant variant found to be associated with brain-age, rs9843963, is on
the Kalirin (KALRN) gene (p < 1.4x10-
08
). Further, our gene scoring results indicated this
gene to be significant overall (p < 3.3x10-
03
). The kalirin gene is most widely expressed in
the brain and has been associated with multiple neurodegenerative diseases including
Huntinton’s disease, Alzheimer’s disease, and Amyotrphic Lateral Sclerosis (Fagerburg et
al., 2014; Murray et al., 2012; Remmers et al., 2014; Youn et al., 2007). Additionally, this
gene has been associated with neuropsychiatric disorders, such as Schizophrenia, and is
involved in cytoskeleton remodeling, spine and dendritic morphology, and trafficking of
synaptic proteins in the brain (Powell, 2018; Remmers et al., 2014). Together, this
information supports the important and widespread roles of kalirin in both brain structure
and function.
Aging is a heterogeneous process, which is particularly important to consider when
modeling brain-age. Thus, it will be important to replicate our findings with a larger sample
of optimal agers used for training and validation across a wide range of ages. However, our
results show that robust modeling of brain-age is possible, and the genetic associations,
University of Southern California
171
genes, and pathways identified with brain-age support the role of the brain-age framework
in measuring brain health and suboptimal brain aging trajectories.
U n i v e r s i t y o f S o u t h e r n C a l i f o r n i a
172
7 SUMMARY
The work outlined in this thesis covers a wide range of topics that all converge on
understanding optimal aging and risk for AD. Broadly, this work has shown that multivariate
models can contribute to the field of AD research and what is known about the risk for
deviations from optimal aging and subsequently AD. In chapter two, I provided an overview
of the important genomic and proteomic interactions between apolipoprotein E (APOE), and
how these interact with sex to mediate accelerated risk for AD more strongly in women.
Importantly, I discuss roles for this protein and gene that go beyond associations with
amyloid, but also involve alterations in lipid metabolism, dietary considerations, the immune
system, and subsequently inflammation. In chapter three, I discovered that considering
multivariate hierarchical network representations of plasma and brain measures aids in our
understanding and ability to detect future progression of AD. Notably, this work showed that
combining latent factor representations of baseline plasma measures and structural brain
metrics allowed for improved predictive accuracy of disease over the original features. This
work also implicated multiple immune-related markers as important predictive markers
along the disease trajectory. In chapter four, I show how interactions with ABO blood type,
both its phenotype and genotype, are associated with interactions with number of secretor
alleles (FUT2), and subsequently relative differences in brain volume in the frontal and
University of Southern California
173
temporal lobes and risk for AD. This work also showed that the role of ABO in mediating risk
for AD may involve interactions with other plasma immune markers, such as receptor for
advanced glycation end products (RAGE). Notably, identification of the interaction between
ABO and RAGE is novel, yet supported by established literature, as RAGE is the primary
transporter of amyloid from the periphery to the brain, and an association with glycation
end products and ABO have been established in diabetes. Indeed, Alzheimer’s has at times
been referred to as type 3 diabetes. While ABO O blood type may put an individual at an
increased risk for AD, high plasma levels of RAGE were shown to be protective for these
individuals, as were 2 secretor alleles. Thus, this work identifies potential risks and
protective mechanisms for the role of blood type in AD. The work in these two chapters
provide support for a systems biology and network approach to advance our mechanistic
understanding of AD.
In chapter five, I show that deviations from optimal aging are differentially heritable.
Although alterations in brain laterality have been previously proposed as an important
aspect in modeling aging that potentially explains differences in resilience to age-associated
changes and cognitive reserve, here I show that this metric is largely not heritable. This work
also shows that deviations from optimal aging and the heritability of these metrics, depends
for many regions, on the particular sex or age being considered, with the latter point
indicating that heritability in deviations from optimal aging are not static. Many of the
regions that were discovered to be heritable are involved in age-associated cognitive decline,
or risk for AD, pointing to the potential targetability of these particular regions in behavioral
or therapeutic interventions. However, this is not the case for all regions, and so future
studies will need to consider alternative approaches if the goal is to target brain regions
University of Southern California
174
which show age-associated deviations that are largely not heritable. The approach carried
out in this study allowed us to understand deviations from optimal aging, instead of more
broadly identifying how heritable brain metrics are in a general framework. This is an
important distinction as heritability of a brain region more broadly is determined across an
individual’s entire lifespan, and thus may offer less insight into how effective behavioral or
therapeutic interventions may be for a particular individual later in life. Moreover, we found
that age-associated diseases, such as cardiovascular disease, can affect the heritability of
brain metrics, leading to confounds on established levels of heritability of these metrics.
Modeling optimal aging to then predict brain metrics across a range of health states allows
us to move beyond medicalizing aging and understand health to instead inform disease.
In chapter six I show that modeling brain-age more robustly is possible. However,
there are a number of considerations that must be made in order to do so successfully, and
thus allow for more reproducible and robust predictions. Importantly, many of these
considerations have not been made in recently published work incorporating brain-age. A
significant portion of prior work on brain-age has failed to utilize hold-out data to validate
their model, instead using cross-validation frameworks with a healthy group and assessing
the Pearson correlation between chronological and predicted brain-age from this group. Our
preliminary work revealed that over-fitting with machine learning is quite easy and thus
using the Pearson correlation for validation is biased and akin to Simpson’s paradox, as
results tend to exhibit regression towards the mean for held-out data, such that younger
individuals often appear to have an older brain-age, and older individuals tend to have a
younger brain-age, while middle-aged individuals tend to have brain-ages that are more
closely aligned with their chronological age. Some of these issues are reduced through
University of Southern California
175
incorporation of a larger training set, but our work showed that this was highly dependent
on the distribution of ages within the healthy training group, and that the maximum and
minimum age ranges of this group must at least match, if not exceed the testing group. Bias
correction is also useful, but further improvements can still be made in future work.
Moreover, the mantra “all models are wrong, but some are useful” is especially suitable here
as our work also showed that good performance metrics in cross-validation does not often
lead to a good model fit on new data, even using a healthy group for hold-out that largely
overlaps in status with the training group. These results indicated that failing to validate
optimal machine learning parameters through hold-out data, instead of simply using cross-
validation, will nearly always lead to a difference in predicted brain-age for the latter group
compared to the training group, thereby biasing any conclusions made. Furthermore, much
of the previous work on brain-age has performed subsequent analyses of brain-age by
violating Lord’s paradox through incorporation of chronological age as a covariate while
using the predicted brain-age difference as the outcome measure. In this regard, our
preliminary analyses indicated that, as expected, the amount of variance explained is highly
skewed and results are biased when these two measures are included in the same model.
This is likely a result of the collinearity that exists when using an outcome measure that was
largely derived from a covariate in the same model. If the model additionally contains other
variables it is typically assumed they have a relationship with the outcome variable, and so,
they also have multi-collinearity with the age covariate, violating an important assumption
in linear regression. In our modeling of brain-age, we used a hold-out set of optimal agers
with a similar age distribution to validate the GBM parameters, which were then applied to
the larger set of participants. This still yielded high Pearson correlation and low root-mean-
University of Southern California
176
square error between predicted and chronological age, but also gave more robust
predictions to the hold-out sets and exhibited less over-fitting on the training sets. A genome-
wide association study was then performed, including kinship and using a mixed-effects
model. We further accounted for potential non-linear age effects by including centered-age-
squared as a covariate in our GWAS model. These results showed that pathway enrichment
analysis, while incorporating information about linkage disequilibrium, identified several
relevant pathways involved in axon guidance, developmental biology, glucose and lipid
metabolism, and the adaptive immune system. Moreover, these results revealed multiple
significant genes that have been consistently implicated in AD, such as ADAMTS10, and were
in line with our preliminary brain-age work with ADNI in which we also found a gene in the
ADAMTS family, that is known to be involved in cell migration, and neuroplasticity. Members
of this family serve as an α-secretase protease during cleavage of amyloid precursor protein.
In conclusion, this work shows that by incorporating neuroimaging with bioinformatics, and
using a systems biology framework, we can enhance our knowledge of the aging process,
ability to detect risk for AD, and mechanistic underpinnings of this risk.
U n i v e r s i t y o f S o u t h e r n C a l i f o r n i a
177
8 FUTURE DIRECTIONS
Much of my research has come full-circle. When I started graduate school, I had just
finished working as a clinical coordinator for part of the Human Microbiome Project.
Understanding the link between the gastrointestinal system and the brain is becoming an
increasingly important endeavor, with supporting evidence across health, disease, and
neuropsychiatric states. Much of these interactions are known to be mediated by the gut’s
microbiome and subsequently by dysbiosis. This work largely converges on understanding
interactions between lipid metabolism and the immune system in mediating health. Notably,
the gut contains nearly 70% of our body’s immune system, and interactions between the
microbiome and how the immune system mediates risk for Alzheimer’s are emerging as
critical fields of research. Indeed, much of the work presented in this thesis finds strong
support for the role of the immune system in early changes associated with an altered aging
trajectory and AD, both through plasma and genomic analyses. Moreover, much of the
systemically-derived neurotransmitters our brain relies on, particularly for communication
with the gut, are produced by the microbiome, from dopamine to serotonin, which have
important roles in age-associated changes in cognitive function, mood, and
neurodegenerative disorders. Thus, future work includes planning and logistics for a multi-
site collaborative effort across the United States to advance our understanding of the
University of Southern California
178
microbiome in the earliest stages of Alzheimer’s disease and how changes in the relative
distribution of, and specific species of bacteria maps onto neuroimaging and cognitive
metrics.
Beyond this, future work will include a continued focus on systems biology, considering
not just the brain, but the peripheral system, and the interplay between the two, in risk for
AD and deviations from optimal aging. Multivariate genetic analyses, including further
studies of single nucleotide polymorphisms, with incorporation of gene expression and RNA-
sequencing, are important fields of research that will advance our knowledge of how aging
interacts with risk for AD. Although polygenic risk scores may improve our ability to predict
AD in the clinic, understanding pathways and networks more specifically will be a continued
area of focus, that has the potential to yield more fruitful insights into how deviations can be
targeted for therapeutic interventions. Likewise, combining information across each of these
areas to understand phenotypes of risk in recognition that AD exists on a continuum, rather
than focusing on simple binary disease states, will increase power to further expand our
knowledge of the pathways and potential targetable mechanisms of AD.
U n i v e r s i t y o f S o u t h e r n C a l i f o r n i a
179
REFERENCES
1. Aggarwal, N. T., Wilson, R. S., Beck, T. L., Bienias, J. L., & Bennett, D. A. (2006). Motor
dysfunction in mild cognitive impairment and the risk of incident Alzheimer
disease. Archives of Neurology, 63(12), 1763-1769. National Institute on Aging.
(2015).
2. Alzheimer's Disease Progress Report 2014-2015: Advancing Research toward a Cure. In
Bethesda: National Institute on Aging, National Institute of Health, US Department of
Health and Human Services.
3. Aird, I., Bentall, H. H., & Roberts, J. F. (1953). Relationship between cancer of stomach and
the ABO blood groups. British Medical Journal, 1(4814), 799.
4. Alam, Q., Zubair Alam, M., Mushtaq, G., Damanhouri, G. A., Rasool, M., Amjad Kamal, M., &
Haque, A. (2016). Inflammatory process in Alzheimer’s and Parkinson's diseases:
central role of cytokines. Current Pharmaceutical Design, 22(5), 541-548.
5. Alexander, K. S., Zakai, N. A., Gillett, S., McClure, L. A., Wadley, V., Unverzagt, F., &
Cushman, M. (2014). ABO blood type, factor VIII, and incident cognitive impairment
in the REGARDS cohort. Neurology, 83(14), 1271-1276.
6. Alonso, A. D. C., Zaidi, T., Grundke-Iqbal, I., & Iqbal, K. (1994). Role of abnormally
phosphorylated tau in the breakdown of microtubules in Alzheimer
disease. Proceedings of the National Academy of Sciences, 91(12), 5562-5566.
7. Altmann A, Tian L, Henderson VW, and Greicius MD. (2014). Sex Modifies the Apoe‐
Related Risk of Developing Alzheimer Disease. Annals of Neurology, 75:563-573.
8. Alzheimer’s Association. (2015). 2015 Alzheimer's disease facts and figures. Alzheimer's
& Dementia: The Journal of the Alzheimer's Association, 11(3), 332.
9. Anstey, K. J., von Sanden, C., Salim, A., & O'kearney, R. (2007). Smoking as a risk factor for
dementia and cognitive decline: a meta-analysis of prospective studies. American
Journal of Epidemiology, 166(4), 367-378.
10. Apostolova LG, Hwang KS, Kohannim O, Avila D, Elashoff D, Jack CR, Shaw L,
Trojanowski JQ, Weiner MW, and Thompson PM. (2014). Apoe4 Effects on
Automated Diagnostic Classifiers for Mild Cognitive Impairment and Alzheimer's
Disease. NeuroImage: Clinical, 4:461-472.
11. Asada T, Kariya T, Yamagata Z, Kinoshita T, and Asaka A. (1996). Apolipoprotein E
Allele in Centenarians. Neurology, 46:1484-1484.
12. Ashton, N. J., Kiddle, S. J., Graf, J., Ward, M., Baird, A. L., Hye, A., ... & Miller, B. L. (2015).
Blood protein predictors of brain amyloid for enrichment in clinical trials?
Alzheimer's & Dementia: Diagnosis, Assessment & Disease Monitoring, 1(1), 48-60.
13. Atwood, C. S., Obrenovich, M. E., Liu, T., Chan, H., Perry, G., Smith, M. A., & Martins, R. N.
(2003). Amyloid-β: a chameleon walking in two worlds: a review of the trophic and
toxic properties of amyloid-β. Brain Research Reviews, 43(1), 1-16.
14. Ayutyanont, N., Langbaum, J. B., Hendrix, S. B., Chen, K., Fleisher, A. S., Friesenhahn, M., ...
& Muñoz, C. (2014). The Alzheimer’s Prevention Initiative composite cognitive test
score: sample size estimates for the evaluation of preclinical Alzheimer’s disease
University of Southern California
180
treatments in presenilin 1 E280A mutation carriers. The Journal of Clinical
Psychiatry, 75(6), 652.
15. Baimbridge, K. G., & Miller, J. J. (1982). Immunohistochemical localization of calcium-
binding protein in the cerebellum, hippocampal formation and olfactory bulb of the
rat. Brain Research, 245(2), 223-229.
16. Barker, W. W., Luis, C. A., Kashuba, A., Luis, M., Harwood, D. G., Loewenstein, D., ... &
Graff-Radford, N. (2002). Relative frequencies of Alzheimer disease, Lewy body,
vascular and frontotemporal dementia, and hippocampal sclerosis in the State of
Florida Brain Bank. Alzheimer Disease & Associated Disorders, 16(4), 203-212.
17. Barlow, H. (1961). Initial remarks. In Neurophysiologie und Psychophysik des Visuellen
Systems/The Visual System: Neurophysiology and Psychophysics (pp. 375-376).
Springer, Berlin, Heidelberg.
18. Barnes L, Wilson R, Schneider J, Bienias J, Evans D, and Bennett D. (2003). Gender,
Cognitive Decline, and Risk of AD in Older Persons. Neurology, 60:1777-1781.
19. Barnes LL, Wilson RS, Bienias JL, Schneider JA, Evans DA, and Bennett DA. (2005). Sex
Differences in the Clinical Manifestations of Alzheimer Disease Pathology. Archives
of General Psychiatry, 62:685-691.
20. Bartels, M., de Geus, E. J., Kirschbaum, C., Sluyter, F., & Boomsma, D. I. (2003).
Heritability of daytime cortisol levels in children. Behavior Genetics, 33(4), 421-433.
21. Bartzokis G, Lu PH, Geschwind DH, Edwards N, Mintz J, and Cummings JL. (2006).
Apolipoprotein E Genotype and Age-Related Myelin Breakdown in Healthy
Individuals: Implications for Cognitive Decline and Dementia. Archives of General
Psychiatry, 63:63-72.
22. Bartzokis G, Lu PH, Geschwind DH, Tingus K, Huang D, Mendez MF, Edwards N, and
Mintz J. (2007). Apolipoprotein E Affects Both Myelin Breakdown and Cognition:
Implications for Age-Related Trajectories of Decline into Dementia. Biological
Psychiatry, 62:1380-1387.
23. Baruch, K., Deczkowska, A., Rosenzweig, N., Tsitsou-Kampeli, A., Sharif, A. M.,
Matcovitch-Natan, O., ... & Schwartz, M. (2016). PD-1 immune checkpoint blockade
reduces pathology and improves memory in mouse models of Alzheimer's
disease. Nature Medicine, 22(2), 135.
24. Bastianelli, E. (2003). Distribution of calcium-binding proteins in the cerebellum. The
Cerebellum, 2(4), 242-262.
25. Beckman, L., & Ängqvist, K. A. (1987). On the mechanism behind the association
between ABO blood groups and gastric carcinoma. Human Heredity, 37(3), 140-143.
26. Beekly DL, Ramos EM, Lee WW, Deitrich WD, Jacka ME, Wu J, Hubbard JL, Koepsell TD,
Morris JC, and Kukull WA. (2007). The National Alzheimer's Coordinating Center
(NACC) Database: The Uniform Data Set. Alzheimer Disease & Associated Disorders,
21:249-258.
27. Behrens, M. I., Lendon, C., & Roe, C. M. (2009). A common biological mechanism in
cancer and Alzheimer's disease? Current Alzheimer Research, 6(3), 196-204.
28. Bell RD, Winkler EA, Singh I, Sagare AP, Deane R, Wu Z, Holtzman DM, Betsholtz C,
Armulik A, and Sallstrom J. (2012). Apolipoprotein E Controls Cerebrovascular
Integrity Via Cyclophilin A. Nature, 485:512-516.
University of Southern California
181
29. Bell, D. A., & Wang, H. (2000). A formalism for relevance and its application in feature
subset selection. Machine Learning, 41(2), 175-195.
30. Benjamini, Y., & 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), 289-300.
31. Bennett RE, Esparza TJ, Lewis HA, Kim E, Mac Donald CL, Sullivan PM, and Brody DL.
(2013). Human Apolipoprotein E4 Worsens Acute Axonal Pathology but Not
Amyloid-Β Immunoreactivity after Traumatic Brain Injury in 3xtg-Ad Mice. Journal
of Neuropathology & Experimental Neurology, 72:396-403.
32. Bertram, L., McQueen, M. B., Mullin, K., Blacker, D., & Tanzi, R. E. (2007). Systematic
meta-analyses of Alzheimer disease genetic association studies: the AlzGene
database. Nature Genetics, 39(1), 17.
33. Blalock EM, Buechel HM, Popovic J, Geddes JW, and Landfield PW. (2011). Microarray
Analyses of Laser-Captured Hippocampus Reveal Distinct Gray and White Matter
Signatures Associated with Incipient Alzheimer's Disease. Journal of Chemical
Neuroanatomy, 42:118-126.
34. Blalock EM, Chen K-C, Sharrow K, Herman JP, Porter NM, Foster TC, and Landfield PW.
(2003). Gene Microarrays in Hippocampal Aging: Statistical Profiling Identifies
Novel Processes Correlated with Cognitive Impairment. The Journal of
Neuroscience, 23:3807-3819.
35. Blalock EM, Geddes JW, Chen KC, Porter NM, Markesbery WR, and Landfield PW.
(2004). Incipient Alzheimer's Disease: Microarray Correlation Analyses Reveal
Major Transcriptional and Tumor Suppressor Responses. Proceedings of the
National Academy of Sciences of the United States of America, 101:2173-2178.
36. Blanché H, Cabanne L, Sahbatou M, and Thomas G. (2001). A Study of French
Centenarians: Are Ace and Apoe Associated with Longevity? Comptes Rendus de
l'Académie des Sciences-Series III-Sciences de la Vie, 324:129-135.
37. Blennow, K. (2010). Biomarkers in Alzheimer's disease drug development. Nature
Medicine, 16(11), 1218.
38. Blessed, G., Tomlinson, B. E., & Roth, M. (1968). The association between quantitative
measures of dementia and of senile change in the cerebral grey matter of elderly
subjects. The British Journal of Psychiatry, 114(512), 797-811.
39. Bookheimer SY, Strojwas MH, Cohen MS, Saunders AM, Pericak-Vance MA, Mazziotta JC,
and Small GW. (2000). Patterns of Brain Activation in People at Risk for Alzheimer's
Disease. New England Journal of Medicine, 343:450-456.
40. Box, G. E., & Cox, D. R. (1964). An analysis of transformations. Journal of the Royal
Statistical Society. Series B (Methodological), 211-252.
41. Braak, H., & Braak, E. (1991). Neuropathological stageing of Alzheimer-related changes.
Acta Neuropathologica, 82(4), 239-259.
42. Braak, H., Braak, E., Yilmazer, D., de Vos, R. A., Jansen, E. N., Bohl, J., & Jellinger, K.
(1994). Amygdala pathology in Parkinson's disease. Acta Neuropathologica, 88(6),
493-500.
43. Braak, H., Del Tredici, K., Rüb, U., De Vos, R. A., Steur, E. N. J., & Braak, E. (2003). Staging
of brain pathology related to sporadic Parkinson’s disease. Neurobiology of
Aging, 24(2), 197-211.
University of Southern California
182
44. Brecht WJ, Harris FM, Chang S, Tesseur I, Yu G-Q, Xu Q, Fish JD, Wyss-Coray T, Buttini M,
and Mucke L. (2004). Neuron-Specific Apolipoprotein E4 Proteolysis Is Associated
with Increased Tau Phosphorylation in Brains of Transgenic Mice. The Journal of
Neuroscience, 24:2527-2534.
45. Brinton RD, Yao J, Yin F, Mack WJ, and Cadenas E. (2015). Perimenopause as a
Neurological Transition State. Nature Reviews Endocrinology.
46. Brinton RD. (2013). Neurosteroids as Regenerative Agents in the Brain: Therapeutic
Implications. Nature Reviews Endocrinology, 9:241-250.
47. Brookmeyer R, Gray S, and Kawas C. (1998). Projections of Alzheimer's Disease in the
United States and the Public Health Impact of Delaying Disease Onset. American
Journal of Public Health, 88:1337-1342.
48. Brookmeyer R, Johnson E, Ziegler-Graham K, and Arrighi HM. (2007). Forecasting the
Global Burden of Alzheimer’s Disease. Alzheimer's & Dementia, 3:186-191.
49. Brookmeyer, R., Evans, D. A., Hebert, L., Langa, K. M., Heeringa, S. G., Plassman, B. L., &
Kukull, W. A. (2011). National estimates of the prevalence of Alzheimer’s disease in
the United States. Alzheimer's & Dementia: The Journal of the Alzheimer's
Association, 7(1), 61-73.
50. Brookmeyer, R., Johnson, E., Ziegler-Graham, K., & Arrighi, H. M. (2007). Forecasting the
global burden of Alzheimer’s disease. Alzheimer's & dementia: The Journal of the
Alzheimer's Association, 3(3), 186-191.
51. Brooks-Wilson, A. R. (2013). Genetics of healthy aging and longevity. Human
Genetics, 132(12), 1323-1338.
52. Brussaard H, Leuven JG, Frölich M, Kluft C, and Krans H. (1997). Short-Term Oestrogen
Replacement Therapy Improves Insulin Resistance, Lipids and Fibrinolysis in
Postmenopausal Women with Niddm. Diabetologia, 40:843-849.
53. Bryant, C., Giovanello, K. S., Ibrahim, J. G., Chang, J., Shen, D., Peterson, B. S., ... &
Alzheimer's Disease Neuroimaging Initiative. (2013). Mapping the genetic variation
of regional brain volumes as explained by all common SNPs from the ADNI
study. PLos One, 8(8), e71723.
54. Bu G. (2009). Apolipoprotein E and Its Receptors in Alzheimer's Disease: Pathways,
Pathogenesis and Therapy. Nature Reviews Neuroscience, 10:333-344.
55. Bu, X. L., Yao, X. Q., Jiao, S. S., Zeng, F., Liu, Y. H., Xiang, Y., ... & Yi, X. (2015). A study on the
association between infectious burden and Alzheimer's disease. European Journal of
Neurology, 22(12), 1519-1525.
56. Buchman, A. S., & Bennett, D. A. (2011). Loss of motor function in preclinical
Alzheimer’s disease. Expert Review of Neurotherapeutics, 11(5), 665-676.
57. Bullmore, E., Brammer, M., Harvey, I., & Ron, M. (1995). Against the laterality index as a
measure of cerebral asymmetry. Psychiatry Research: Neuroimaging, 61(2), 121-
124.
58. Burger HG, Dudley EC, Robertson DM, and Dennerstein L. (2002). Hormonal Changes in
the Menopause Transition. Recent Progress in Hormone Research, 57:257-276.
59. Calne, D., & Langston, J. W. (1983). Aetiology of Parkinson's disease. The
Lancet, 322(8365-8366), 1457-1459.
60. Castro E, Ogburn CE, Hunt KE, Tilvis R, Louhija J, Penttinen R, Erkkola R, Panduro A,
Riestra R, and Piussan C. (1999). Polymorphisms at the Werner Locus: I. Newly
University of Southern California
183
Identified Polymorphisms, Ethnic Variability of 1367cy/Arg, and Its Stability in a
Population of Finnish Centenarians. American Journal of Medical Genetics, 82:399-
403.
61. Chakrabarty, P., Li, A., Ceballos-Diaz, C., Eddy, J. A., Funk, C. C., Moore, B., ... & Sacino, A.
(2015). IL-10 alters immunoproteostasis in APP mice, increasing plaque burden and
worsening cognitive behavior. Neuron, 85(3), 519-533.
62. Chan-Palay, V., & Palay, S. L. (1972). The stellate cells of the rat's cerebellar
cortex. Zeitschrift für Anatomie und Entwicklungsgeschichte, 136(2), 224-248.
63. Chang S, ran Ma T, Miranda RD, Balestra ME, Mahley RW, and Huang Y. (2005). Lipid-
and Receptor-Binding Regions of Apolipoprotein E4 Fragments Act in Concert to
Cause Mitochondrial Dysfunction and Neurotoxicity. Proceedings of the National
Academy of Sciences of the United States of America, 102:18694-18699.
64. Chen H-K, Ji Z-S, Dodson SE, Miranda RD, Rosenblum CI, Reynolds IJ, Freedman SB,
Weisgraber KH, Huang Y, and Mahley RW. (2011). Apolipoprotein E4 Domain
Interaction Mediates Detrimental Effects on Mitochondria and Is a Potential
Therapeutic Target for Alzheimer Disease. Journal of Biological Chemistry,
286:5215-5221.
65. Chen J, Li Q, and Wang J. (2011). Topology of Human Apolipoprotein E3 Uniquely
Regulates Its Diverse Biological Functions. Proceedings of the National Academy of
Sciences, 108:14813-14818.
66. Chêne G, Beiser A, Au R, Preis SR, Wolf PA, Dufouil C, and Seshadri S. (2015). Gender
and Incidence of Dementia in the Framingham Heart Study from Mid-Adult Life.
Alzheimer's & Dementia, 11:310-320.
67. Cheng X, McAsey ME, Li M, Randall S, Cady C, Nathan BP, and Struble RG. (2007).
Estradiol Replacement Increases the Low-Density Lipoprotein Receptor Related
Protein (Lrp) in the Mouse Brain. Neuroscience Letters, 417:50-54.
68. Chetelat, G., Desgranges, B., Eustache, F., Perrotin, A., de la Sayette, V., Fouquet, M., &
Villain, N. (2012). Amyloid imaging in at risk cognitively normal individuals and
patients with subjective cognitive impairment. Alzheimer's & Dementia, 8(4), P607.
69. Chiang G, Zhan W, Schuff N, and Weiner M. (2012). White Matter Alterations in
Cognitively Normal Apoe Ɛ2 Carriers: Insight into Alzheimer Resistance? American
Journal of Neuroradiology, 33:1392-1397.
70. Chow H-m, and Herrup K. (2015). Genomic Integrity and the Ageing Brain. Nature
Reviews Neuroscience.
71. Christensen, K., Johnson, T. E., & Vaupel, J. W. (2006). The quest for genetic
determinants of human longevity: challenges and insights. Nature Reviews
Genetics, 7(6), 436.
72. Chung, Y., Addington, J., Bearden, C. E., Cadenhead, K., Cornblatt, B., Mathalon, D. H., ... &
Walker, E. (2018). Use of Machine Learning to Determine Deviance in
Neuroanatomical Maturity Associated With Future Psychosis in Youths at Clinically
High Risk. JAMA Psychiatry.
73. Citron, M., Diehl, T. S., Gordon, G., Biere, A. L., Seubert, P., & Selkoe, D. J. (1996). Evidence
that the 42-and 40-amino acid forms of amyloid β protein are generated from the β-
amyloid precursor protein by different protease activities. Proceedings of the
National Academy of Sciences, 93(23), 13170-13175.
University of Southern California
184
74. Cohen, J. (1973). Eta-squared and partial eta-squared in fixed factor ANOVA
designs. Educational and psychological measurement, 33(1), 107-112.
75. Colantuoni, C., Lipska, B. K., Ye, T., Hyde, T. M., Tao, R., Leek, J. T., ... & Kleinman, J. E.
(2011). Temporal dynamics and genetic control of transcription in the human
prefrontal cortex. Nature, 478(7370), 519.
76. Cole, J. H., Ritchie, S. J., Bastin, M. E., Hernández, M. V., Maniega, S. M., Royle, N., ... &
Wray, N. R. (2017). Brain age predicts mortality. Molecular Psychiatry.
77. Conejero-Goldberg C, Hyde T, Chen S, Dreses-Werringloer U, Herman M, Kleinman J,
Davies P, and Goldberg T. (2011). Molecular Signatures in Post-Mortem Brain Tissue
of Younger Individuals at High Risk for Alzheimer's Disease as Based on Apoe
Genotype. Molecular Psychiatry, 16:836-847.
78. Conomos, M. P., Thornton, T., & Conomos, M. M. P. (2015). Package ‘GENESIS’.
79. Coon KD, Myers AJ, Craig DW, Webster JA, Pearson JV, Lince DH, Zismann VL, Beach TG,
Leung D, and Bryden L. (2007). A High-Density Whole-Genome Association Study
Reveals That Apoe Is the Major Susceptibility Gene for Sporadic Late-Onset
Alzheimer's Disease. Journal of Clinical Psychiatry, 68:613.
80. Corbo R, and Scacchi R. (1999). Apolipoprotein E (Apoe) Allele Distribution in the
World. Is Apoe* 4 a ‘Thrifty’allele? Annals of Human Genetics, 63:301-310.
81. Corder E, Saunders A, Strittmatter W, Schmechel D, Gaskell P, Rimmler J, Locke P,
Conneally P, Schmader K, and Tanzi R. (1995). Apolipoprotein E, Survival in
Alzheimer's Disease Patients, and the Competing Risks of Death and Alzheimer's
Disease. Neurology, 45:1323-1328.
82. Corder EH, Ghebremedhin E, Taylor MG, Thal DR, OHM TG, and Braak H. (2004). The
Biphasic Relationship between Regional Brain Senile Plaque and Neurofibrillary
Tangle Distributions: Modification by Age, Sex, and Apoe Polymorphism. Annals of
the New York Academy of Sciences, 1019:24-28.
83. Corder, E. H., Saunders, A. M., Strittmatter, W. J., Schmechel, D. E., Gaskell, P. C., Small, G.,
... & Pericak-Vance, M. A. (1993). Gene dose of apolipoprotein E type 4 allele and the
risk of Alzheimer's disease in late onset families. Science, 261(5123), 921-923.
84. Corder, E. H., Saunders, A. M., Strittmatter, W. J., Schmechel, D. E., Gaskell, P. C., Small, G.,
... & Pericak-Vance, M. A. (1993). Gene dose of apolipoprotein E type 4 allele and the
risk of Alzheimer's disease in late onset families. Science, 261(5123), 921-923.
85. Cramer PE, Cirrito JR, Wesson DW, Lee CD, Karlo JC, Zinn AE, Casali BT, Restivo JL,
Goebel WD, and James MJ. (2012). Apoe-Directed Therapeutics Rapidly Clear Β-
Amyloid and Reverse Deficits in Ad Mouse Models. Science, 335:1503-1506.
86. Crane, P. K., Carle, A., Gibbons, L. E., Insel, P., Mackin, R. S., Gross, A., ... & Weiner, M.
(2012). Development and assessment of a composite score for memory in the
Alzheimer’s Disease Neuroimaging Initiative (ADNI). Brain Imaging and Behavior,
6(4), 502-516.
87. Cribbs, D. H., Berchtold, N. C., Perreau, V., Coleman, P. D., Rogers, J., Tenner, A. J., &
Cotman, C. W. (2012). Extensive innate immune gene activation accompanies brain
aging, increasing vulnerability to cognitive decline and neurodegeneration: a
microarray study. Journal of Neuroinflammation, 9(1), 179.
88. Crivello F, Lemaître H, Dufouil C, Grassiot B, Delcroix N, Tzourio-Mazoyer N, Tzourio C,
and Mazoyer B. (2010). Effects of Apoe-Ɛ4 Allele Load and Age on the Rates of Grey
University of Southern California
185
Matter and Hippocampal Volumes Loss in a Longitudinal Cohort of 1186 Healthy
Elderly Persons. Neuroimage, 53:1064-1069.
89. Crooks, V. C., Lubben, J., Petitti, D. B., Little, D., & Chiu, V. (2008). Social network,
cognitive function, and dementia incidence among elderly women. American Journal
of Public Health, 98(7), 1221-1227.
90. Cuello, A. C., Ferretti, M. T., Leon, W. C., Iulita, M. F., Melis, T., Ducatenzeiler, A., ... &
Canneva, F. (2010). Early-stage inflammation and experimental therapy in
transgenic models of the Alzheimer-like amyloid pathology. Neurodegenerative
Diseases, 7(1-3), 96-98.
91. Cunnane S, Nugent S, Roy M, Courchesne-Loyer A, Croteau E, Tremblay S, Castellano A,
Pifferi F, Bocti C, and Paquet N. (2011). Brain Fuel Metabolism, Aging, and
Alzheimer’s Disease. Nutrition, 27:3-20.
92. Curtis R, Geesaman BJ, and DiStefano PS. (2005). Ageing and Metabolism: Drug
Discovery Opportunities. Nature Reviews Drug Discovery, 4:569-580.
93. Cutter, G. R., Baier, M. L., Rudick, R. A., Cookfair, D. L., Fischer, J. S., Petkau, J., ... & Ellison,
G. W. (1999). Development of a multiple sclerosis functional composite as a clinical
trial outcome measure. Brain, 122(5), 871-882.
94. Damoiseaux JS, Seeley WW, Zhou J, Shirer WR, Coppola G, Karydas A, Rosen HJ, Miller
BL, Kramer JH, and Greicius MD. (2012). Gender Modulates the Apoe Ε4 Effect in
Healthy Older Adults: Convergent Evidence from Functional Brain Connectivity and
Spinal Fluid Tau Levels. The Journal of Neuroscience, 32:8254-8262.
95. Darabi M, Ani M, Panjehpour M, Rabbani M, Movahedian A, and Zarean E. (2011). Effect
of Estrogen Receptor Β A1730g Polymorphism on Abca1 Gene Expression Response
to Postmenopausal Hormone Replacement Therapy. Genetic Testing and Molecular
Biomarkers, 15:11-15.
96. Dashti HM, Mathew TC, Hussein T, Asfar SK, Behbahani A, Khoursheed MA, Al-Sayer
HM, Bo-Abbas YY, and Al-Zaid NS. (2004). Long-Term Effects of a Ketogenic Diet in
Obese Patients. Experimental & Clinical Cardiology, 9:200.
97. Davatzikos, C., Bhatt, P., Shaw, L. M., Batmanghelich, K. N., & Trojanowski, J. Q. (2011).
Prediction of MCI to AD conversion, via MRI, CSF biomarkers, and pattern
classification. Neurobiology of Aging, 32(12), 2322-e19.
98. Davies, T. L., Gompels, M., Johnston, S., Bovill, B., & May, M. T. (2013). Mind the gap:
difference between Framingham heart age and real age increases with age in HIV-
positive individuals–a clinical cohort study. BMJ Open, 3(10), e003245.
99. De Marco, M., & Venneri, A. (2015). ‘O’Blood Type is Associated with Larger Grey-
Matter Volumes in the Cerebellum. Brain Research Bulletin.
100. De Marinis E, Martini C, Trentalance A, and Pallottini V. (2008). Sex Differences in
Hepatic Regulation of Cholesterol Homeostasis. Journal of Endocrinology, 198:635-
643.
101. Dean DC, Jerskey BA, Chen K, Protas H, Thiyyagura P, Roontiva A, O'Muircheartaigh J,
Dirks H, Waskiewicz N, and Lehman K. (2014). Brain Differences in Infants at
Differential Genetic Risk for Late-Onset Alzheimer Disease: A Cross-Sectional
Imaging Study. JAMA Neurology, 71:11-22.
University of Southern California
186
102. Deane, R., Wu, Z., & Zlokovic, B. V. (2004). RAGE (yin) versus LRP (yang) balance
regulates Alzheimer amyloid β-peptide clearance through transport across the
blood–brain barrier. Stroke, 35(11 suppl 1), 2628-2631.
103. D'Eustachio P. Reactome knowledgebase of human biological pathways and processes.
Methods Mol. Biol. 2011;694(49-61.
104. Ding F, Yao J, Rettberg JR, Chen S, and Brinton RD. (2013). Early Decline in Glucose
Transport and Metabolism Precedes Shift to Ketogenic System in Female Aging and
Alzheimer's Mouse Brain: Implication for Bioenergetic Intervention. PLos One,
8:e79977.
105. Ding F, Yao J, Zhao L, Mao Z, Chen S, and Brinton RD. (2013). Ovariectomy Induces a
Shift in Fuel Availability and Metabolism in the Hippocampus of the Female
Transgenic Model of Familial Alzheimer's. PLos One, 8:e59825.
106. Dinov, I., Lozev, K., Petrosyan, P., Liu, Z., Eggert, P., Pierce, J., ... & Magsipoc, R. (2010).
Neuroimaging study designs, computational analyses and data provenance using the
LONI pipeline. PLos One, 5(9), e13070.
107. DoEaSAPD UN. (2001). World Population Ageing: 1950-2050. In. New York: United
Nations.
108. DoEaSAPD UN. (2015). World Population Ageing 2015. In. 5th ed. New York: United
Nations.
109. Doecke, J. D., Laws, S. M., Faux, N. G., Wilson, W., Burnham, S. C., Lam, C. P., ... & De
Ruyck, K. (2012). Blood-based protein biomarkers for diagnosis of Alzheimer
disease. Archives of Neurology, 69(10), 1318-1325.
110. Donohue, M. C., Sperling, R. A., Salmon, D. P., Rentz, D. M., Raman, R., Thomas, R. G., ... &
Aisen, P. S. (2014). The preclinical Alzheimer cognitive composite: measuring
amyloid-related decline. JAMA Neurology, 71(8), 961-970.
111. Dorshkind K, Montecino-Rodriguez E, and Signer RA. (2009). The Ageing Immune
System: Is It Ever Too Old to Become Young Again? Nature Reviews Immunology,
9:57-62.
112. Doshi, J., Erus, G., Ou, Y., Resnick, S. M., Gur, R. C., Gur, R. E., ... & Alzheimer's
Neuroimaging Initiative. (2016). MUSE: MUlti-atlas region Segmentation utilizing
Ensembles of registration algorithms and parameters, and locally optimal atlas
selection. NeuroImage, 127, 186-195.
113. Dou, J., Sun, B., Sim, X., Hughes, J. D., Reilly, D. F., Tai, E. S., ... & Wang, C. (2017).
Estimation of kinship coefficient in structured and admixed populations using
sparse sequencing data. PLoS Genetics, 13(9), e1007021.
114. Drubin, D. G., & Kirschner, M. W. (1986). Tau protein function in living cells. The
Journal of Cell Biology, 103(6), 2739-2746.
115. Dubois, B., Feldman, H. H., Jacova, C., Cummings, J. L., DeKosky, S. T., Barberger-Gateau,
P., ... & Gauthier, S. (2010). Revising the definition of Alzheimer's disease: a new
lexicon. The Lancet Neurology, 9(11), 1118-1127.
116. Durbin, R. M., Abecasis, G. R., Bentley, D. R., Chakravarti, A., Clark, A. G., Donnelly, P., ...
& Schmidt, J. P. (2012). An integrated map of genetic variation from 1,092 human
genomes.
117. Elshourbagy NA, Liao WS, Mahley RW, and Taylor JM. (1985). Apolipoprotein E Mrna
Is Abundant in the Brain and Adrenals, as Well as in the Liver, and Is Present in
University of Southern California
187
Other Peripheral Tissues of Rats and Marmosets. Proceedings of the National
Academy of Sciences, 82:203-207.
118. Engelhart, M. J., Geerlings, M. I., Meijer, J., Kiliaan, A., Ruitenberg, A., van Swieten, J. C., ...
& Breteler, M. M. (2004). Inflammatory proteins in plasma and the risk of dementia:
the rotterdam study. Archives of Neurology, 61(5), 668-672.
119. Erikson, G. A., Bodian, D. L., Rueda, M., Molparia, B., Scott, E. R., Scott-Van Zeeland, A. A.,
... & Torkamani, A. (2016). Whole-genome sequencing of a healthy aging
cohort. Cell, 165(4), 1002-1011.
120. Evans S, Dowell NG, Tabet N, Tofts PS, King SL, and Rusted JM. (2014). Cognitive and
Neural Signatures of the Apoe E4 Allele in Mid-Aged Adults. Neurobiology of Aging,
35:1615-1623.
121. Fair, D. A., Cohen, A. L., Dosenbach, N. U., Church, J. A., Miezin, F. M., Barch, D. M., ... &
Schlaggar, B. L. (2008). The maturing architecture of the brain's default
network. Proceedings of the National Academy of Sciences, 105(10), 4028-4032.
122. Fagerberg, L., Hallström, B. M., Oksvold, P., Kampf, C., Djureinovic, D., Odeberg, J., ... &
Asplund, A. (2014). Analysis of the human tissue-specific expression by genome-
wide integration of transcriptomics and antibody-based proteomics. Molecular &
Cellular Proteomics, 13(2), 397-406.
123. Fan, Y., Batmanghelich, N., Clark, C. M., Davatzikos, C., & Alzheimer’s Disease
Neuroimaging Initiative. (2008). Spatial patterns of brain atrophy in MCI patients,
identified via high-dimensional pattern classification, predict subsequent cognitive
decline. Neuroimage, 39(4), 1731-1743.
124. Farfara, D., Lifshitz, V., & Frenkel, D. (2008). Neuroprotective and neurotoxic
properties of glial cells in the pathogenesis of Alzheimer's disease. Journal of
Cellular and Molecular Medicine, 12(3), 762-780.
125. Farrer LA, Cupples LA, Haines JL, Hyman B, Kukull WA, Mayeux R, Myers RH, Pericak-
Vance MA, Risch N, and van Duijn CM. (1997). Effects of Age, Sex, and Ethnicity on
the Association between Apolipoprotein E Genotype and Alzheimer Disease: A Meta-
Analysis. JAMA, 278:1349-1356.
126. Fassbender, K., Walter, S., Kühl, S., Landmann, R., Ishii, K., Bertsch, T., ... & Rivest, S.
(2004). The LPS receptor (CD14) links innate immunity with Alzheimer’s
disease. The FASEB Journal, 18(1), 203-205.
127. Feng, S., Liu, D., Zhan, X., Wing, M. K., & Abecasis, G. R. (2014). RAREMETAL: fast and
powerful meta-analysis for rare variants. Bioinformatics, 30(19), 2828-2829.
128. Fischl, B., Salat, D. H., Busa, E., Albert, M., Dieterich, M., Haselgrove, C., ... & Montillo, A.
(2002). Whole brain segmentation: automated labeling of neuroanatomical
structures in the human brain. Neuron, 33(3), 341-355.
129. Fisher, D. W., Bennett, D. A., & Dong, H. (2018). Sexual Dimorphism in Predisposition
to Alzheimer’s Disease. Neurobiology of Aging.
130. Fleisher A, Grundman M, Jack CR, Petersen RC, Taylor C, Kim HT, Schiller DH, Bagwell
V, Sencakova D, and Weiner MF. (2005). Sex, Apolipoprotein E Ε4 Status, and
Hippocampal Volume in Mild Cognitive Impairment. Archives of Neurology, 62:953-
957.
University of Southern California
188
131. Folstein, M. F., Folstein, S. E., & McHugh, P. R. (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.
132. Fortin, C. F., McDonald, P. P., Lesur, O., & Fülöp Jr, T. (2008). Aging and neutrophils:
there is still much to do. Rejuvenation Research, 11(5), 873-882.
133. Fotuhi, M., Do, D., & Jack, C. (2012). Modifiable factors that alter the size of the
hippocampus with ageing. Nature Reviews Neurology, 8(4), 189-202.
134. Freytag, S., Burgess, R., Oliver, K. L., & Bahlo, M. (2017). brain-coX: investigating and
visualising gene co-expression in seven human brain transcriptomic
datasets. Genome Medicine, 9(1), 55.
135. Freund, Y., Schapire, R. E., Singer, Y., & Warmuth, M. K. (1997, May). Using and
combining predictors that specialize. In Proceedings of the Twenty-Ninth Annual
ACM Symposium on Theory of Computing (pp. 334-343). ACM.
136. Friedman, D. B., & Johnson, T. E. (1988). A mutation in the age-1 gene in
Caenorhabditis elegans lengthens life and reduces hermaphrodite
fertility. Genetics, 118(1), 75-86.
137. Friedman, J. H. (2001). Greedy function approximation: a gradient boosting machine.
Annals of Statistics, 1189-1232.
138. Frydman-Marom, A., Levin, A., Farfara, D., Benromano, T., Scherzer-Attali, R., Peled, S.,
... & Ovadia, M. (2011). Orally administrated cinnamon extract reduces β-amyloid
oligomerization and corrects cognitive impairment in Alzheimer's disease animal
models. PLos One, 6(1), e16564.
139. Fukumoto, H., Tennis, M., Locascio, J. J., Hyman, B. T., Growdon, J. H., & Irizarry, M. C.
(2003). Age but not diagnosis is the main predictor of plasma amyloid β-protein
levels. Archives of Neurology, 60(7), 958-964.
140. Gabriel K, Egan B, and Lithgow T. (2003). Tom40, the Import Channel of the
Mitochondrial Outer Membrane, Plays an Active Role in Sorting Imported Proteins.
The EMBO Journal, 22:2380-2386.
141. Gagnon-Bartsch, J. A., & Speed, T. P. (2012). Using control genes to correct for
unwanted variation in microarray data. Biostatistics, 13(3), 539-552.
142. Galinsky D, Tysoe C, Brayne CE, Easton DF, Huppert FA, Dening TR, Paykel ES, and
Rubinsztein DC. (1997). Analysis of the Apo E/Apo Ci, Angiotensin Converting
Enzyme and Methylenetetrahydrofolate Reductase Genes as Candidates Affecting
Human Longevity. Atherosclerosis, 129:177-183.
143. Garai K, Baban B, and Frieden C. (2011). Dissociation of Apolipoprotein E Oligomers to
Monomer Is Required for High-Affinity Binding to Phospholipid Vesicles.
Biochemistry, 50:2550-2558.
144. Garatachea N, Emanuele E, Calero M, Fuku N, Arai Y, Abe Y, Murakami H, Miyachi M,
Yvert T, and Verde Z. (2014). Apoe Gene and Exceptional Longevity: Insights from
Three Independent Cohorts. Experimental Gerontology, 53:16-23.
145. Garratty, G., Glynn, S. A., McEntire, R., & Retrovirus Epidemiology Donor Study. (2004).
ABO and Rh (D) phenotype frequencies of different racial/ethnic groups in the
United States. Transfusion, 44(5), 703-706.
University of Southern California
189
146. Gatz, M., Reynolds, C. A., Fratiglioni, L., Johansson, B., Mortimer, J. A., Berg, S., ... &
Pedersen, N. L. (2006). Role of genes and environments for explaining Alzheimer
disease. Archives of General Psychiatry, 63(2), 168-174.
147. Gau B, Garai K, Frieden C, and Gross ML. (2011). Mass Spectrometry-Based Protein
Footprinting Characterizes the Structures of Oligomeric Apolipoprotein E2, E3, and
E4. Biochemistry, 50:8117-8126.
148. Getz, G. S., & Reardon, C. A. (2014). The mutual interplay of lipid metabolism and the
cells of the immune system in relation to atherosclerosis. Clinical Lipidology, 9(6),
657-671.
149. Giedraitis, V., Sundelöf, J., Irizarry, M. C., Gårevik, N., Hyman, B. T., Wahlund, L. O., ... &
Lannfelt, L. (2007). The normal equilibrium between CSF and plasma amyloid beta
levels is disrupted in Alzheimer's disease. Neuroscience Letters, 427(3), 127-131.
150. Gong J-S, Kobayashi M, Hayashi H, Zou K, Sawamura N, Fujita SC, Yanagisawa K, and
Michikawa M. (2002). Apolipoprotein E (Apoe) Isoform-Dependent Lipid Release
from Astrocytes Prepared from Human Apoe3 and Apoe4 Knock-in Mice. Journal of
Biological Chemistry, 277:29919-29926.
151. Gonzalez, B., Abud, E. M., Abud, A. M., Poon, W. W., & Gylys, K. H. (2017). Tau Spread,
Apolipoprotein E, Inflammation, and More: Rapidly Evolving Basic Science in
Alzheimer Disease. Neurologic Clinics, 35(2), 175-190.
152. Gordon, C. M., LeBoff, M. S., & Glowacki, J. (2001). Adrenal and gonadal steroids inhibit
IL-6 secretion by human marrow cells. Cytokine, 16(5), 178-186.
153. Gregg RE, Zech LA, Schaefer EJ, and Brewer HB. (1981). Type Iii
Hyperlipoproteinemia: Defective Metabolism of an Abnormal Apolipoprotein E.
Science, 211:584-586.
154. Griciuc, A., Serrano-Pozo, A., Parrado, A. R., Lesinski, A. N., Asselin, C. N., Mullin, K., ... &
Tanzi, R. E. (2013). Alzheimer’s disease risk gene CD33 inhibits microglial uptake of
amyloid beta. Neuron, 78(4), 631-643.
155. Guadalupe, T., Mathias, S. R., Theo, G. M., Whelan, C. D., Zwiers, M. P., Abe, Y., ... &
Aribisala, B. S. (2017). Human subcortical brain asymmetries in 15,847 people
worldwide reveal effects of age and sex. Brain Imaging and Behavior, 11(5), 1497-
1514.
156. Guerreiro, R., Wojtas, A., Bras, J., Carrasquillo, M., Rogaeva, E., Majounie, E., ... & Hazrati,
L. (2013). TREM2 variants in Alzheimer's disease. New England Journal of
Medicine, 368(2), 117-127.
157. Gutman, B. A., Hua, X., Rajagopalan, P., Chou, Y. Y., Wang, Y., Yanovsky, I., ... & the
Alzheimer's Disease Neuroimaging Initiative. (2013). Maximizing power to track
Alzheimer's disease and MCI progression by LDA-based weighting of longitudinal
ventricular surface features. Neuroimage, 70, 386-401.
158. Guyon, I., & Elisseeff, A. (2003). An introduction to variable and feature selection.
Journal of Machine Learning Research, 3(Mar), 1157-1182.
159. Hafezi-Moghadam A, Thomas KL, and Wagner DD. (2007). Apoe Deficiency Leads to a
Progressive Age-Dependent Blood-Brain Barrier Leakage. American Journal of
Physiology-Cell Physiology, 292:C1256-C1262.
University of Southern California
190
160. Halle, A., Hornung, V., Petzold, G. C., Stewart, C. R., Monks, B. G., Reinheckel, T., ... &
Golenbock, D. T. (2008). The NALP3 inflammasome is involved in the innate immune
response to amyloid-β. Nature Immunology, 9(8), 857.
161. Hampel, H., Frank, R., Broich, K., Teipel, S. J., Katz, R. G., Hardy, J., ... & Sanhai, W. R.
(2010). Biomarkers for Alzheimer's disease: academic, industry and regulatory
perspectives. Nature Reviews Drug Discovery, 9(7), 560.
162. Hanakawa, T., Dimyan, M. A., & Hallett, M. (2008). Motor planning, imagery, and
execution in the distributed motor network: a time-course study with functional
MRI. Cerebral Cortex, 18(12), 2775-2788.
163. Hanlon CS, and Rubinsztein DC. (1995). Arginine Residues at Codons 112 and 158 in
the Apolipoprotein E Gene Correspond to the Ancestral State in Humans.
Atherosclerosis, 112:85-90.
164. Hardy, J. A., & Higgins, G. A. (1992). Alzheimer's disease: the amyloid cascade
hypothesis. Science, 256(5054), 184.
165. Hardy, J., & Allsop, D. (1991). Amyloid deposition as the central event in the aetiology
of Alzheimer's disease. Trends in Pharmacological Sciences, 12, 383-388.
166. Hardy, J., & Selkoe, D. J. (2002). The amyloid hypothesis of Alzheimer's disease:
progress and problems on the road to therapeutics. Science, 297(5580), 353-356.
167. Harris FM, Brecht WJ, Xu Q, Tesseur I, Kekonius L, Wyss-Coray T, Fish JD, Masliah E,
Hopkins PC, and Scearce-Levie K. (2003). Carboxyl-Terminal-Truncated
Apolipoprotein E4 Causes Alzheimer's Disease-Like Neurodegeneration and
Behavioral Deficits in Transgenic Mice. Proceedings of the National Academy of
Sciences, 100:10966-10971.
168. Hasebe, N., Fujita, Y., Ueno, M., Yoshimura, K., Fujino, Y., & Yamashita, T. (2013).
Soluble β-amyloid precursor protein alpha binds to p75 neurotrophin receptor to
promote neurite outgrowth. PLos One, 8(12), e82321.
169. Hauptmann, S., Keil, U., Scherping, I., Bonert, A., Eckert, A., & Müller, W. E. (2006).
Mitochondrial dysfunction in sporadic and genetic Alzheimer's
disease. Experimental Gerontology, 41(7), 668-673.
170. Hawrylycz, M. J., Lein, E. S., Guillozet-Bongaarts, A. L., Shen, E. H., Ng, L., Miller, J. A., ... &
Abajian, C. (2012). An anatomically comprehensive atlas of the adult human brain
transcriptome. Nature, 489(7416), 391.
171. Hebert LE, Weuve J, Scherr PA, and Evans DA. (2013). Alzheimer Disease in the United
States (2010–2050) Estimated Using the 2010 Census. Neurology, 80:1778-1783.
172. Hernandez, D. G., Nalls, M. A., Moore, M., Chong, S., Dillman, A., Trabzuni, D., ... &
Zonderman, A. B. (2012). Integration of GWAS SNPs and tissue specific expression
profiling reveal discrete eQTLs for human traits in blood and brain. Neurobiology of
Disease, 47(1), 20-28.
173. Hicks, P. S., Saunero-Nava, L., Du Clos, T. W., & Mold, C. (1992). Serum amyloid P
component binds to histones and activates the classical complement pathway. The
Journal of Immunology, 149(11), 3689-3694.
174. Hippius, H., & Neundörfer, G. (2003). The discovery of Alzheimer's disease. Dialogues
in Clinical Neuroscience, 5(1), 101.
University of Southern California
191
175. Hollingworth, P., Harold, D., Sims, R., Gerrish, A., Lambert, J. C., Carrasquillo, M. M., ... &
Dowzell, K. (2011). Common variants at ABCA7, MS4A6A/MS4A4E, EPHA1, CD33
and CD2AP are associated with Alzheimer's disease. Nature Genetics, 43(5), 429.
176. Holmans P, Hamshere M, Hollingworth P, Rice F, Tunstall N, Jones S, Moore P,
DeVrieze FW, Myers A, and Crook R. (2005). Genome Screen for Loci Influencing Age
at Onset and Rate of Decline in Late Onset Alzheimer's Disease. American Journal of
Medical Genetics Part B: Neuropsychiatric Genetics, 135:24-32.
177. Holtzman DM, Herz J, and Bu G. (2012). Apolipoprotein E and Apolipoprotein E
Receptors: Normal Biology and Roles in Alzheimer Disease. Cold Spring Harbor
Perspectives in Medicine, 2:a006312.
178. Hoogland, T. M., & Kuhn, B. (2010). Recent developments in the understanding of
astrocyte function in the cerebellum in vivo. The Cerebellum, 9(3), 264-271.
179. Hooper, C., Killick, R., & Lovestone, S. (2008). The GSK3 hypothesis of Alzheimer’s
disease. Journal of Neurochemistry, 104(6), 1433-1439.
180. Hooshmand, B., Solomon, A., Kåreholt, I., Leiviskä, J., Rusanen, M., Ahtiluoto, S., ... &
Kivipelto, M. (2010). Homocysteine and holotranscobalamin and the risk of
Alzheimer disease A longitudinal study. Neurology, 75(16), 1408-1414.
181. Horn, A. K. (2006). The reticular formation. Progress in brain research, 151, 127-155.
182. Hu, W. T., Holtzman, D. M., Fagan, A. M., Shaw, L. M., Perrin, R., Arnold, S. E., ... &
Pickering, E. (2012). Plasma multianalyte profiling in mild cognitive impairment and
Alzheimer disease. Neurology, 79(9), 897-905.
183. Hu, Y. S., Xin, J., Hu, Y., Zhang, L., & Wang, J. (2017). Analyzing the genes related to
Alzheimer’s disease via a network and pathway-based approach. Alzheimer's
Research & Therapy, 9(1), 29.
184. Hua, X., Gutman, B., Boyle, C. P., Rajagopalan, P., Leow, A. D., Yanovsky, I., ... &
Alexander, G. E. (2011). Accurate measurement of brain changes in longitudinal MRI
scans using tensor-based morphometry. Neuroimage, 57(1), 5-14.
185. Hua, X., Hibar, D. P., Ching, C. R., Boyle, C. P., Rajagopalan, P., Gutman, B. A., ... & Weiner,
M. W. (2013). Unbiased tensor-based morphometry: improved robustness and
sample size estimates for Alzheimer's disease clinical trials. Neuroimage, 66, 648-
661.
186. Hua, X., Leow, A. D., Parikshak, N., Lee, S., Chiang, M. C., Toga, A. W., ... & Alzheimer's
Disease Neuroimaging Initiative. (2008). Tensor-based morphometry as a
neuroimaging biomarker for Alzheimer's disease: an MRI study of 676 AD, MCI, and
normal subjects. Neuroimage, 43(3), 458-469.
187. Huang RY, Garai K, Frieden C, and Gross ML. (2011). Hydrogen/Deuterium Exchange
and Electron-Transfer Dissociation Mass Spectrometry Determine the Interface and
Dynamics of Apolipoprotein E Oligomerization. Biochemistry, 50:9273-9282.
188. Huang, K. L., Marcora, E., Pimenova, A. A., Di Narzo, A. F., Kapoor, M., Jin, S. C., ... &
Chouraki, V. (2017). A common haplotype lowers PU. 1 expression in myeloid cells
and delays onset of Alzheimer's disease. Nature Neuroscience, 20(8), 1052.
189. Hughes, C. P., Berg, L., Danziger, W., Coben, L. A., & Martin, R. L. (1982). A new clinical
scale for the staging of dementia. The British Journal of Psychiatry, 140(6), 566-572.
University of Southern California
192
190. Hutson, A. M., Atmar, R. L., Graham, D. Y., & Estes, M. K. (2002). Norwalk virus infection
and disease is associated with ABO histo-blood group type. The Journal of Infectious
Diseases, 185(9), 1335-1337.
191. Ifere, G. O., Desmond, R., Demark-Wahnefried, W., & Nagy, T. R. (2013). Apolipoprotein
E gene polymorphism influences aggressive behavior in prostate cancer cells by
deregulating cholesterol homeostasis. International Journal of Oncology, 43(4),
1002-1010.
192. Ihara Y, Hayabara T, Sasaki K, Kawada R, Nakashima Y, and Kuroda S. (2000).
Relationship between Oxidative Stress and Apoe Phenotype in Alzheimer's Disease.
Acta Neurologica Scandinavica, 102:346-349.
193. Ittner, L. M., & Götz, J. (2011). Amyloid-β and tau—a toxic pas de deux in Alzheimer's
disease. Nature Reviews Neuroscience, 12(2), 67.
194. Itzhaki, R. F., Lin, W. R., Shang, D., Wilcock, G. K., Faragher, B., & Jamieson, G. A. (1997).
Herpes simplex virus type 1 in brain and risk of Alzheimer's disease. The
Lancet, 349(9047), 241-244.
195. Ivanova MM, Radde BN, Son J, Mehta FF, Chung S-H, and Klinge CM. (2013). Estradiol
and Tamoxifen Regulate Nrf-1 and Mitochondrial Function in Mouse Mammary
Gland and Uterus. Journal of Molecular Endocrinology, 51:233-246.
196. Jack CR, Wiste HJ, Weigand SD, Knopman DS, Vemuri P, Mielke MM, Lowe V, Senjem
ML, Gunter JL, and Machulda MM. (2015). Age, Sex, and Apoe Ε4 Effects on Memory,
Brain Structure, and Β-Amyloid across the Adult Life Span. JAMA Neurology, 72:511-
519.
197. Jack Jr, C. R., & Holtzman, D. M. (2013). Biomarker modeling of Alzheimer's disease.
Neuron, 80(6), 1347-1358.
198. Jack Jr, C. R., Knopman, D. S., Jagust, W. J., Petersen, R. C., Weiner, M. W., Aisen, P. S., ... &
Trojanowski, J. Q. (2013). Tracking pathophysiological processes in Alzheimer's
disease: an updated hypothetical model of dynamic biomarkers. The Lancet
Neurology, 12(2), 207-216.
199. Jack Jr, C. R., Knopman, D. S., Jagust, W. J., Shaw, L. M., Aisen, P. S., Weiner, M. W., ... &
Trojanowski, J. Q. (2010). Hypothetical model of dynamic biomarkers of the
Alzheimer's pathological cascade. The Lancet Neurology, 9(1), 119-128.
200. Jack, C. R., Bennett, D. A., Blennow, K., Carrillo, M. C., Feldman, H. H., Frisoni, G. B., ... &
Petersen, R. C. (2016). A/T/N: an unbiased descriptive classification scheme for
Alzheimer disease biomarkers. Neurology, 87(5), 539-547.
201. Jack, C. R., Bernstein, M. A., Fox, N. C., Thompson, P., Alexander, G., Harvey, D., ... & Dale,
A. M. (2008). The Alzheimer's disease neuroimaging initiative (ADNI): MRI methods.
Journal of Magnetic Resonance Imaging, 27(4), 685-691.
202. Jacobs EG, Kroenke C, Lin J, Epel ES, Kenna HA, Blackburn EH, and Rasgon NL. (2013).
Accelerated Cell Aging in Female Apoe-Epsilon4 Carriers: Implications for Hormone
Therapy Use. PLos One, 8:e54713.
203. Jagust, W. J., Bandy, D., Chen, K., Foster, N. L., Landau, S. M., Mathis, C. A., ... & Koeppe, R.
A. (2010). The Alzheimer's Disease Neuroimaging Initiative positron emission
tomography core. Alzheimer's & Dementia, 6(3), 221-229.
University of Southern California
193
204. Jarvik G, Wijsman EM, Kukull W, Schellenberg G, Yu C, and Larson E. (1995).
Interactions of Apolipoprotein E Genotype, Total Cholesterol Level, Age, and Sex in
Prediction of Alzheimer's Disease a Case‐Control Study. Neurology, 45:1092-1096.
205. Jendroska, K., Poewe, W., Pluess, J., Iwerssen-Schmidt, H., Paulsen, J., Barthel, S., ... &
DeArmond, S. J. (1995). Ischemic stress induces deposition of amyloid β
immunoreactivity in human brain. Acta Neuropathologica, 90(5), 461-466.
2066. Jensen, L. J., & Bateman, A. (2011). The rise and fall of supervised machine learning
techniques. Bioinformatics 27(24), 3331-3332.
207. Jiang, S., Zhang, M., Ren, D., Tang, G., Lin, S., Qian, Y., ... & Wang, D. (2003). Enhanced
production of amyloid precursor protein mRNA by peripheral mononuclear blood
cell in Alzheimer's disease. American Journal of Medical Genetics Part B:
Neuropsychiatric Genetics, 118(1), 99-102.
208. Joachim, C. L., Morris, J. H., & Selkoe, D. J. (1989). Diffuse senile plaques occur
commonly in the cerebellum in Alzheimer's disease. The American Journal of
Pathology, 135(2), 309.
209. Johnson, K. A., Fox, N. C., Sperling, R. A., & Klunk, W. E. (2012). Brain imaging in
Alzheimer disease. Cold Spring Harbor Perspectives in Medicine, 2(4), a006213].
210. Jones, L., Lambert, J. C., Wang, L. S., Choi, S. H., Harold, D., Vedernikov, A., ... & Ibrahim-
Verbaas, C. A. (2015). Convergent genetic and expression data implicate immunity in
Alzheimer's disease. Alzheimer's & dementia: The Journal of the Alzheimer's
Association, 11(6), 658-671.
211. Jonsson, T., Stefansson, H., Steinberg, S., Jonsdottir, I., Jonsson, P. V., Snaedal, J., ... &
Rujescu, D. (2013). Variant of TREM2 associated with the risk of Alzheimer's
disease. New England Journal of Medicine, 368(2), 107-116.
212. Jorde, L. B., Watkins, W. S., Bamshad, M. J., Dixon, M. E., Ricker, C. E., Seielstad, M. T., &
Batzer, M. A. (2000). The distribution of human genetic diversity: a comparison of
mitochondrial, autosomal, and Y-chromosome data. The American Journal of Human
Genetics, 66(3), 979-988.
213. Josephs, K. A., Whitwell, J. L., Ahmed, Z., Shiung, M. M., Weigand, S. D., Knopman, D. S., ...
& Jack, C. R. (2008). β‐amyloid burden is not associated with rates of brain atrophy.
Annals of Neurology, 63(2), 204-212.
214. Kaerst L, Kuhlmann A, Wedekind D, Stoeck K, Lange P, and Zerr I. (2013).
Cerebrospinal Fluid Biomarkers in Alzheimer’s Disease, Vascular Dementia and
Ischemic Stroke Patients: A Critical Analysis. Journal of Neurology, 260:2722-2727.
215. Kanbay, M., Gür, G., Arslan, H., Yilmaz, U., & Boyacioĝlu, S. (2005). The relationship of
ABO blood group, age, gender, smoking, and Helicobacter pylori infection. Digestive
Diseases and Sciences, 50(7), 1214-1217.
216. Kanehisa M, Goto S. KEGG: kyoto encyclopedia of genes and genomes. Nucleic Acids
Research. 2000;28(1):27-30.
217. Kang, H. J., Kawasawa, Y. I., Cheng, F., Zhu, Y., Xu, X., Li, M., ... & Guennel, T. (2011).
Spatio-temporal transcriptome of the human brain. Nature, 478(7370), 483.
218. Kang, H. M., Sul, J. H., Zaitlen, N. A., Kong, S. Y., Freimer, N. B., Sabatti, C., & Eskin, E.
(2010). Variance component model to account for sample structure in genome-wide
association studies. Nature Genetics, 42(4), 348.
University of Southern California
194
219. Katzman, R. (1976). The prevalence and malignancy of Alzheimer disease: a major
killer. Archives of Neurology, 33(4), 217-218.
220. Kertesz, A., & Munoz, D. (1998). Pick's disease, frontotemporal dementia, and Pick
complex: emerging concepts. Archives of Neurology, 55(3), 302-304.
221. Kiddle, S. J., Thambisetty, M., Simmons, A., Riddoch-Contreras, J., Hye, A., Westman, E.,
... & Lunnon, K. (2012). Plasma based markers of [11C] PiB-PET brain amyloid
burden. PLos One, 7(9), e44260.
222. Kim, D. H., Yeo, S. H., Park, J. M., Choi, J. Y., Lee, T. H., Park, S. Y., ... & Cha, H. J. (2014).
Genetic markers for diagnosis and pathogenesis of Alzheimer's
disease. Gene, 545(2), 185-193.
223. Kinoshita, A., Fukumoto, H., Shah, T., Whelan, C. M., Irizarry, M. C., & Hyman, B. T.
(2003). Demonstration by FRET of BACE interaction with the amyloid precursor
protein at the cell surface and in early endosomes. Journal of Cell Science, 116(16),
3339-3346.
224. Kivipelto, M., Helkala, E. L., Laakso, M. P., Hänninen, T., Hallikainen, M., Alhainen, K., ...
& Nissinen, A. (2001). Midlife vascular risk factors and Alzheimer's disease in later
life: longitudinal, population based study. BMJ, 322(7300), 1447-1451.
225. Klosinski LP, Yao J, Yin F, Fonteh AN, Harrington MG, Christensen TA, Trushina E, and
Brinton RD. (2015). White Matter Lipids as a Ketogenic Fuel Supply in Aging Female
Brain: Implications for Alzheimer's Disease. EBioMedicine, 2:1888-1904.30.
226. Knopman, D. S., Haeberlein, S. B., Carrillo, M. C., Hendrix, J. A., Kerchner, G., Margolin,
R., ... & Murray, M. E. (2018). The National Institute on Aging and the Alzheimer's
Association Research Framework for Alzheimer's disease: Perspectives from the
Research Roundtable. Alzheimer's & Dementia, 14(4), 563-575.
227. Kolaczkowska, E., & Kubes, P. (2013). Neutrophil recruitment and function in health
and inflammation. Nature Reviews Immunology, 13(3), 159.
228. Koo, E. H., Sisodia, S. S., Archer, D. R., Martin, L. J., Weidemann, A., Beyreuther, K., ... &
Price, D. L. (1990). Precursor of amyloid protein in Alzheimer disease undergoes fast
anterograde axonal transport. Proceedings of the National Academy of
Sciences, 87(4), 1561-1565.
229. Kuhel DG, Konaniah ES, Basford JE, McVey C, Goodin CT, Chatterjee TK, Weintraub NL,
and Hui DY. (2013). Apolipoprotein E2 Accentuates Postprandial Inflammation and
Diet-Induced Obesity to Promote Hyperinsulinemia in Mice. Diabetes, 62:382-391.
230. Kuller LH, Shemanski L, Manolio T, Haan M, Fried L, Bryan N, Burke GL, Tracy R, and
Bhadelia R. (1998). Relationship between Apoe, Mri Findings, and Cognitive
Function in the Cardiovascular Health Study. Stroke, 29:388-398.
231. Kurtenkov, O., Klaamas, K., & Miljukhina, L. (1995). The lower level of natural anti‐
thomsen‐friedenreich antigen (TFA) agglutinins in sera of patients with gastric
cancer related to ABO (H) blood‐group phenotype. International Journal of
Cancer, 60(6), 781-785.
232. LaDu MJ, Munson GW, Jungbauer L, Getz GS, Reardon CA, Tai LM, and Yu C. (2012).
Preferential Interactions between Apoe-Containing Lipoproteins and Aβ Revealed
by a Detection Method That Combines Size Exclusion Chromatography with Non-
Reducing Gel-Shift. Biochimica et Biophysica Acta (BBA)-Molecular and Cell Biology
of Lipids, 1821:295-302.
University of Southern California
195
233. LaDu MJ, Pederson TM, Frail DE, Reardon CA, Getz GS, and Falduto MT. (1995).
Purification of Apolipoprotein E Attenuates Isoform-Specific Binding to-Amyloid.
Journal of Biological Chemistry, 270:9039-9042.
234. Lagrand, W. K., Niessen, H. W., Wolbink, G. J., Jaspars, L. H., Visser, C. A., Verheugt, F. W.,
... & Hack, C. E. (1997). C-reactive protein colocalizes with complement in human
hearts during acute myocardial infarction. Circulation, 95(1), 97-103.
235. Lahoz C, Schaefer EJ, Cupples LA, Wilson PW, Levy D, Osgood D, Parpos S, Pedro-Botet
J, Daly JA, and Ordovas JM. (2001). Apolipoprotein E Genotype and Cardiovascular
Disease in the Framingham Heart Study. Atherosclerosis, 154:529-537.
236. Lalli, M. A., Bettcher, B. M., Arcila, M. L., Garcia, G., Guzman, C., Madrigal, L., ... &
Coppola, G. (2015). Whole-genome sequencing suggests a chemokine gene cluster
that modifies age at onset in familial Alzheimer's disease. Molecular
Psychiatry, 20(11), 1294.
237. Lambert, J. C., Ibrahim-Verbaas, C. A., Harold, D., Naj, A. C., Sims, R., Bellenguez, C., ... &
Grenier-Boley, B. (2013). Meta-analysis of 74,046 individuals identifies 11 new
susceptibility loci for Alzheimer's disease. Nature Genetics, 45(12), 1452.
238. Lamparter, D., Marbach, D., Rueedi, R., Kutalik, Z., & Bergmann, S. (2016). Fast and
rigorous computation of gene and pathway scores from SNP-based summary
statistics. PLoS Computational Biology, 12(1), e1004714.
239. Landsteiner, K., & Levine, P. (1927). A New Agglutinable Factor Differentiating
Individual Human Bloods. Proceedings of the Society for Experimental Biology and
Medicine, 24(6), 600-602.
240. Lane RM, and Farlow MR. (2005). Lipid Homeostasis and Apolipoprotein E in the
Development and Progression of Alzheimer's Disease. Journal of Lipid Research,
46:949-968.
241. Langbaum JB, Chen K, Caselli RJ, Lee W, Reschke C, Bandy D, Alexander GE, Burns CM,
Kaszniak AW, and Reeder SA. (2010). Hypometabolism in Alzheimer-Affected Brain
Regions in Cognitively Healthy Latino Individuals Carrying the Apolipoprotein E Ε4
Allele. Archives of Neurology, 67:462-468.
242. Langbaum, J. B., Hendrix, S., Ayutyanont, N., Bennett, D. A., Shah, R. C., Barnes, L. L., ... &
Tariot, P. N. (2015). Establishing composite cognitive endpoints for use in
preclinical Alzheimer’s disease trials. The Journal of Prevention of Alzheimer's
Disease, 2(1), 2.
243. Laurin, D., Curb, J. D., Masaki, K. H., White, L. R., & Launer, L. J. (2009). Midlife C-
reactive protein and risk of cognitive decline: a 31-year follow-up. Neurobiology of
Aging, 30(11), 1724-1727.
244. Lavrakas, P. J. (2008). Encyclopedia of survey research methods. Sage Publications.
245. Le, T. T., Kuplicki, R., Yeh, H. W., Aupperle, R. L., Khalsa, S. S., Simmons, W. K., & Paulus,
M. P. (2018). Effect of ibuprofen on BrainAGE: a randomized, placebo-controlled,
dose-response exploratory study. Biological Psychiatry: Cognitive Neuroscience and
Neuroimaging.
246. Leduc V, Jasmin-Bélanger S, and Poirier J. (2010). Apoe and Cholesterol Homeostasis
in Alzheimer's Disease. Trends in Molecular Medicine, 16:469-477.
247. Lee PA, Guo SS, and Kulin HE. (2001). Age of Puberty: Data from the United States of
America. APMIS, 109:S156-S163.
University of Southern California
196
248. Lee S, Lee YS, Jung SH, Kang SS, and Shin KH. (2003). Anti-Oxidant Activities of
Fucosterol from the Marine Algaepelvetia Siliquosa. Archives of Pharmacal
Research, 26:719-722.
249. Lehmann D, Refsum H, Nurk E, Warden D, Tell G, Vollset S, Engedal K, Nygaard H, and
Smith A. (2006). Apolipoprotein E Ε4 and Impaired Episodic Memory in
Community-Dwelling Elderly People: A Marked Sex Difference. The Hordaland
Health Study. Journal of Neurology, Neurosurgery & Psychiatry, 77:902-908.
250. Leigh-Firbank EC, Minihane AM, Leake DS, Wright JW, Murphy MC, Griffin BA, and
Williams CM. (2002). Eicosapentaenoic Acid and Docosahexaenoic Acid from Fish
Oils: Differential Associations with Lipid Responses. British Journal of Nutrition,
87:435-445.
251. Li R, Cui J, and Shen Y. (2014). Brain Sex Matters: Estrogen in Cognition and
Alzheimer’s Disease. Molecular and Cellular Endocrinology, 389:13-21.
252. Li, Q., & Yu, K. A. I. (2008). Improved correction for population stratification in
genome‐wide association studies by identifying hidden population
structures. Genetic Epidemiology, 32(3), 215-226.
253. Li, X., Liang, Y., Chen, Y., Zhang, J., Wei, D., Chen, K., ... & Zhang, Z. (2015). Disrupted
frontoparietal network mediates white matter structure dysfunction associated
with cognitive decline in hypertension patients. Journal of Neuroscience, 35(27),
10015-10024.
254. Lin KA, Choudhury KR, Rathakrishnan BG, Marks DM, Petrella JR, Doraiswamy PM, and
Initiative AsDN. (2015). Marked Gender Differences in Progression of Mild Cognitive
Impairment over 8 Years. Alzheimer's & Dementia: Translational Research &
Clinical Interventions, 1:103-110.
255. Lin, R., Zhang, Y., Yan, D., Liao, X., Gong, G., Hu, J., ... & Cai, W. (2016). Association of
common variants in TOMM40/APOE/APOC1 region with human longevity in a
Chinese population. Journal of Human Genetics, 61(4), 323.
256. Linden, S., Mahdavi, J., Semino-Mora, C., Olsen, C., Carlstedt, I., Boren, T., & Dubois, A.
(2008). Role of ABO secretor status in mucosal innate immunity and H. pylori
infection. PLoS Pathogens, 4(1), e2.
257. Lipovich, L., Tarca, A. L., Cai, J., Jia, H., Chugani, H. T., Sterner, K. N., ... & Kuzawa, C. W.
(2013). Developmental changes in the transcriptome of human cerebral cortex
tissue: long noncoding RNA transcripts. Cerebral cortex, 24(6), 1451-1459.
258. Liu C-C, Kanekiyo T, Xu H, and Bu G. (2013). Apolipoprotein E and Alzheimer Disease:
Risk, Mechanisms and Therapy. Nature Reviews Neurology, 9:106-118.
259. Liu, S. J., Zhang, A. H., Li, H. L., Wang, Q., Deng, H. M., Netzer, W. J., ... & Wang, J. Z.
(2003). Overactivation of glycogen synthase kinase‐3 by inhibition of
phosphoinositol‐3 kinase and protein kinase C leads to hyperphosphorylation of tau
and impairment of spatial memory. Journal of Neurochemistry, 87(6), 1333-1344.
260. Liumbruno, G. M., & Franchini, M. (2013). Beyond immunohaematology: the role of the
ABO blood group in human diseases. Blood Transfusion, 11(4), 491.
261. Liu JZ, McRae AF, Nyholt DR, et al. A versatile gene-based test for genome-wide
association studies. Am. J. Hum. Genet. 2010;87(1):139-145.
262. Lopez-Gonzalez, A. A., Aguilo, A., Frontera, M., Bennasar-Veny, M., Campos, I., Vicente-
Herrero, T., ... & Tauler, P. (2015). Effectiveness of the Heart Age tool for improving
University of Southern California
197
modifiable cardiovascular risk factors in a Southern European population: a
randomized trial. European Journal of Preventive Cardiology, 22(3), 389-396.
263. Louhija J, Miettinen HE, Kontula K, Tikkanen MJ, Miettinen T, and Tilvis R. (1994).
Aging and Genetic Variation of Plasma Apolipoproteins. Relative Loss of the
Apolipoprotein E4 Phenotype in Centenarians. Arteriosclerosis, Thrombosis, and
Vascular Biology, 14:1084-1089.
264. Lovestone, S., Francis, P., Kloszewska, I., Mecocci, P., Simmons, A., Soininen, H., ... &
Ward, M. (2009). AddNeuroMed—the European collaboration for the discovery of
novel biomarkers for Alzheimer's disease. Annals of the New York Academy of
Sciences, 1180(1), 36-46.
265. Lyall DM, Harris SE, Bastin ME, Maniega SM, Murray C, Lutz MW, Saunders AM, Roses
AD, Hernández MdCV, and Royle NA. (2014). Alzheimer's Disease Susceptibility
Genes Apoe and Tomm40, and Brain White Matter Integrity in the Lothian Birth
Cohort 1936. Neurobiology of Aging, 35:1513. e25-1513. e33.
266. Ma, L., Clark, A. G., & Keinan, A. (2013). Gene-based testing of interactions in
association studies of quantitative traits. PLoS Genetics, 9(2), e1003321.
267. Mahley RW, and Rall Jr SC. (2000). Apolipoprotein E: Far More Than a Lipid Transport
Protein. Annual review of genomics and human genetics, 1:507-537.
268. Mahley RW, Weisgraber KH, and Huang Y. (2006). Apolipoprotein E4: A Causative
Factor and Therapeutic Target in Neuropathology, Including Alzheimer’s Disease.
Proceedings of the National Academy of Sciences, 103:5644-5651.
269. Mahley RW, Weisgraber KH, and Huang Y. (2009). Apolipoprotein E: Structure
Determines Function, from Atherosclerosis to Alzheimer's Disease to Aids. Journal
of Lipid Research, 50:S183-S188.
270. Manelli AM, Stine WB, Van Eldik LJ, and LaDu MJ. (2004). Apoe and Aβ1–42
Interactions. Journal of Molecular Neuroscience, 23:235-246.
271. Manly, J. J., Tang, M. X., Schupf, N., Stern, Y., Vonsattel, J. P. G., & Mayeux, R. (2008).
Frequency and course of mild cognitive impairment in a multiethnic
community. Annals of Neurology, 63(4), 494-506.
272. Mapstone, M., Cheema, A. K., Fiandaca, M. S., Zhong, X., Mhyre, T. R., MacArthur, L. H., ...
& Nazar, M. D. (2014). Plasma phospholipids identify antecedent memory
impairment in older adults. Nature Medicine, 20(4), 415.
273. Marcello, A., Wirths, O., Schneider-Axmann, T., Degerman-Gunnarsson, M., Lannfelt, L.,
& Bayer, T. A. (2009). Circulating immune complexes of Aβ and IgM in plasma of
patients with Alzheimer’s disease. Journal of Neural Transmission, 116(7), 913-920.
274. Marchant NL, King SL, Tabet N, and Rusted JM. (2010). Positive Effects of Cholinergic
Stimulation Favor Young Apoe &Epsiv; 4 Carriers. Neuropsychopharmacology,
35:1090-1096.
275. Marcus, D. S., Wang, T. H., Parker, J., Csernansky, J. G., Morris, J. C., & Buckner, R. L.
(2007). Open Access Series of Imaging Studies (OASIS): cross-sectional MRI data in
young, middle aged, nondemented, and demented older adults. Journal of Cognitive
Neuroscience, 19(9), 1498-1507.
276. Martinez, V. G., Escoda-Ferran, C., Simoes, I. T., Arai, S., Mascaró, M. O., Carreras, E., ... &
Lozano, F. (2014). The macrophage soluble receptor AIM/Api6/CD5L displays a
University of Southern California
198
broad pathogen recognition spectrum and is involved in early response to microbial
aggression. Cellular & Molecular Immunology, 11(4), 343.
277. Martins, C. A. R., Oulhaj, A., De Jager, C. A., & Williams, J. H. (2005). APOE alleles predict
the rate of cognitive decline in Alzheimer disease A nonlinear
model. Neurology, 65(12), 1888-1893.
278. Masters CL, Bateman R, Blennow K, Rowe CC, Sperling RA, and Cummings JL. (2015).
Alzheimer's Disease. Nature Reviews Disease Primers:15056.
279. Mattsson N, Zetterberg H, Hansson O, Andreasen N, Parnetti L, Jonsson M, Herukka S-
K, van der Flier WM, Blankenstein MA, and Ewers M. (2009). Csf Biomarkers and
Incipient Alzheimer Disease in Patients with Mild Cognitive Impairment. JAMA,
302:385-393.
280. Mayeux R, Stern Y, Ottman R, Tatemichi TK, Tang MX, Maestre G, Ngai C, Tycko B, and
Ginsberg H. (1993). The Apolipoprotein Ε4 Allele in Patients with Alzheimer's
Disease. Annals of Neurology, 34:752-754.
281. Mazziotta, J., Toga, A., Evans, A., Fox, P., Lancaster, J., Zilles, K., ... & Holmes, C. (2001). A
probabilistic atlas and reference system for the human brain: International
Consortium for Brain Mapping (ICBM). Philosophical Transactions of the Royal
Society B: Biological Sciences, 356(1412), 1293-1322.
282. McGeer PL, Kawamata T, Walker DG, Akiyama H, Tooyama I, McGeer EG: Microglia in
degenerative neurological disease. Glia 1993, 7:84–92
283. McGeer, E. G., Yasojima, K., Schwab, C., & McGeer, P. L. (2001). The pentraxins: possible
role in Alzheimer’s disease and other innate inflammatory diseases. Neurobiology of
Aging, 22(6), 843-848.
284. McGeer, P. L., & McGeer, E. G. (2002). Local neuroinflammation and the progression of
Alzheimer’s disease. Journal of Neurovirology, 8(6), 529-538.
285. McGeer, P. L., Akiyama, H., Itagaki, S., & McGeer, E. G. (1989). Activation of the classical
complement pathway in brain tissue of Alzheimer patients. Neuroscience
Letters, 107(1-3), 341-346.
286. McKhann, G. M., Knopman, D. S., Chertkow, H., Hyman, B. T., Jack, C. R., Kawas, C. H., ... &
Mohs, R. C. (2011). The diagnosis of dementia due to Alzheimer’s disease:
Recommendations from the National Institute on Aging-Alzheimer’s Association
workgroups on diagnostic guidelines for Alzheimer's disease. Alzheimer's &
Dementia: The Journal of the Alzheimer's Association, 7(3), 263-269.
287. McKhann, G., Drachman, D., Folstein, M., Katzman, R., Price, D., & Stadlan, E. M. (1984).
Clinical diagnosis of Alzheimer's disease Report of the NINCDS‐ADRDA Work
Group* under the auspices of Department of Health and Human Services Task Force
on Alzheimer's Disease. Neurology, 34(7), 939-939.
288. Medeiros LA, Khan T, El Khoury JB, Pham CL, Hatters DM, Howlett GJ, Lopez R, O'Brien
KD, and Moore KJ. (2004). Fibrillar Amyloid Protein Present in Atheroma Activates
Cd36 Signal Transduction. Journal of Biological Chemistry, 279:10643-10648.
289. Mente A, de Koning L, Shannon HS, and Anand SS. (2009). A Systematic Review of the
Evidence Supporting a Causal Link between Dietary Factors and Coronary Heart
Disease. Archives of Internal Medicine, 169:659-669.
University of Southern California
199
290. Metti, A. L., Yaffe, K., Boudreau, R. M., Simonsick, E. M., Carnahan, R. M., Satterfield, S., ...
& Cauley, J. A. (2014). Trajectories of inflammatory markers and cognitive decline
over 10 years. Neurobiology of Aging, 35(12), 2785-2790.
291. Middelberg RP, Ferreira MA, Henders AK, Heath AC, Madden PA, Montgomery GW,
Martin NG, and Whitfield JB. (2011). Genetic Variants in Lpl, Oasl and
Tomm40/Apoe-C1-C2-C4 Genes Are Associated with Multiple Cardiovascular-
Related Traits. BMC Medical Genetics, 12:123.
292. Miech R, Breitner J, Zandi P, Khachaturian A, Anthony J, and Mayer L. (2002).
Incidence of Ad May Decline in the Early 90s for Men, Later for Women the Cache
County Study. Neurology, 58:209-218.
293. Miller, K. L., Alfaro-Almagro, F., Bangerter, N. K., Thomas, D. L., Yacoub, E., Xu, J., ... &
Griffanti, L. (2016). Multimodal population brain imaging in the UK Biobank
prospective epidemiological study. Nature Neuroscience, 19(11), 1523.
294. Mirra, S. S., Heyman, A., McKeel, D., Sumi, S. M., Crain, B. J., Brownlee, L. M., ... & Berg, L.
(1991). The Consortium to Establish a Registry for Alzheimer's Disease (CERAD)
Part II. Standardization of the neuropathologic assessment of Alzheimer's
disease. Neurology, 41(4), 479-479.
295. Misra, C., Fan, Y., & Davatzikos, C. (2009). Baseline and longitudinal patterns of brain
atrophy in MCI patients, and their use in prediction of short-term conversion to AD:
results from ADNI. Neuroimage, 44(4), 1415-1422.
296. Miyata M, and Smith JD. (1996). Apolipoprotein E Allele–Specific Antioxidant Activity
and Effects on Cytotoxicity by Oxidative Insults and Β–Amyloid Peptides. Nature
Genetics, 14:55-61.
297. Miyazaki T, Hirokami Y, Matsuhashi N, Takatsuka H, Naito M. Increased susceptibility
of thymocytes to apoptosis in mice lacking AIM, a novel murine macrophage-derived
soluble factor belonging to the scavenger receptor cysteine-rich domain
superfamily. J Exp Med 1999; 189: 413–422.
298. Moffat S, Szekely C, Zonderman A, Kabani N, and Resnick S. (2000). Longitudinal
Change in Hippocampal Volume as a Function of Apolipoprotein E Genotype.
Neurology, 55:134-136.
299. Mohs, R. C. (1996). The Alzheimer's disease assessment scale. International
Psychogeriatrics, 8(2), 195-203.
300. Moms, J. C., Heyman, A., Mohs, R. C., Hughes, J. P., van Belle, G., Fillenbaum, G., ... &
Clark, C. (1989). The Consortium to Establish a Registry for Alzheimer's Disease
(CERAD). Part I. Clinical and neuropsychological assessment of Alzheimer's
disease. Neurology, 39(9), 1159-1159.
301. Mondadori CR, Dominique J-F, Buchmann A, Mustovic H, Wollmer MA, Schmidt CF,
Boesiger P, Hock C, Nitsch RM, and Papassotiropoulos A. (2007). Better Memory and
Neural Efficiency in Young Apolipoprotein E Ε4 Carriers. Cerebral Cortex, 17:1934-
1947.
302. Moon, T. K. (1996). The expectation-maximization algorithm. IEEE Signal Processing
magazine, 13(6), 47-60.
303. Morris JC, Weintraub S, Chui HC, Cummings J, DeCarli C, Ferris S, Foster NL, Galasko D,
Graff-Radford N, and Peskind ER. (2006). The Uniform Data Set (Uds): Clinical and
University of Southern California
200
Cognitive Variables and Descriptive Data from Alzheimer Disease Centers.
Alzheimer Disease & Associated Disorders, 20:210-216.
304. Mortensen EL, and Høgh P. (2001). A Gender Difference in the Association between
Apoe Genotype and Age-Related Cognitive Decline. Neurology, 57:89-95.
305. Mosconi, L., Perani, D., Sorbi, S., Herholz, K., Nacmias, B., Holthoff, V., ... & Pupi, A.
(2004). MCI conversion to dementia and the APOE genotype A prediction study with
FDG-PET. Neurology, 63(12), 2332-2340.
306. Mosconi, L., Tsui, W. H., Herholz, K., Pupi, A., Drzezga, A., Lucignani, G., ... & Diehl-
Schmid, J. (2008). Multicenter standardized 18F-FDG PET diagnosis of mild
cognitive impairment, Alzheimer's disease, and other dementias. Journal of Nuclear
Medicine, 49(3), 390-398.
307. Mourant, A. E., & MAD, P. (1954). The Distribution of the Human Blood Groups. The
Distribution of the Human Blood Groups.
308. Mudher, A., & Lovestone, S. (2002). Alzheimer's disease–do tauists and baptists finally
shake hands?. Trends in Neurosciences, 25(1), 22-26.
309. Müller, U., Winter, P., & Graeber, M. B. (2013). A presenilin 1 mutation in the first case
of Alzheimer's disease. The Lancet Neurology, 12(2), 129-130.
310. Murakami, N., Yamaki, T., Iwamoto, Y., Sakakibara, T., Kobori, N., Fushiki, S., & Ueda, S.
(1998). Experimental brain injury induces expression of amyloid precursor protein,
which may be related to neuronal loss in the hippocampus. Journal of
Neurotrauma, 15(11), 993-1003.
311. Murray, M. E., Graff-Radford, N. R., Ross, O. A., Petersen, R. C., Duara, R., & Dickson, D.
W. (2011). Neuropathologically defined subtypes of Alzheimer's disease with
distinct clinical characteristics: a retrospective study. The Lancet Neurology, 10(9),
785- 796.
312. Murray, P. S., Kirkwood, C. M., Gray, M. C., Ikonomovic, M. D., Paljug, W. R.,
Abrahamson, E. E., ... & Lopez, O. L. (2012). β-Amyloid 42/40 ratio and kalirin
expression in Alzheimer disease with psychosis. Neurobiology of Aging, 33(12),
2807-2816.
313. Naj, A. C., Jun, G., Beecham, G. W., Wang, L. S., Vardarajan, B. N., Buros, J., ... & Larson, E.
B. (2011). Common variants at MS4A4/MS4A6E, CD2AP, CD33 and EPHA1 are
associated with late-onset Alzheimer's disease. Nature Genetics, 43(5), 436.
314. Nathan BP, Barsukova AG, Shen F, McAsey M, and Struble RG. (2004). Estrogen
Facilitates Neurite Extension Via Apolipoprotein E in Cultured Adult Mouse Cortical
Neurons. Endocrinology, 145:3065-3073.
315. Nebel A, Kleindorp R, Caliebe A, Nothnagel M, Blanché H, Junge O, Wittig M, Ellinghaus
D, Flachsbart F, and Wichmann H-E. (2011). A Genome-Wide Association Study
Confirms Apoe as the Major Gene Influencing Survival in Long-Lived Individuals.
Mechanisms of Ageing and Development, 132:324-330.
316. Newhouse P, Albert K, Astur R, Johnson J, Naylor M, and Dumas J. (2013). Tamoxifen
Improves Cholinergically Modulated Cognitive Performance in Postmenopausal
Women. Neuropsychopharmacology, 38:2632-2643.
317. Newsome, J., Vaske, J. C., Gehring, K. S., & Boisvert, D. L. (2016). Sex differences in
sources of resilience and vulnerability to risk for delinquency. Journal of Youth and
Adolescence, 45(4), 730-745.
University of Southern California
201
318. Nieto, F. J., & Coresh, J. (1996). Adjusting survival curves for confounders: a review and
a new method. American Journal of Epidemiology, 143(10), 1059-1068.
319. Nishimura D. BioCarta. Biotech Software & Internet Report. 2001;2(3):117-120.
320. Nixon, R. A., & Yang, D. S. (2011). Autophagy failure in Alzheimer's disease—locating
the primary defect. Neurobiology of Disease, 43(1), 38-45.
321. Noguchi S, Murakami K, Yamada N, Payami H, Kaye J, Heston L, Bird T, and
Schellenberg G. (1993). Apolipoprotein E Genotype and Alzheimer's Disease. The
Lancet, 342:737-738.
322. O'Keefe, G. M., Nguyen, V. T., & Benveniste, E. N. (2002). Regulation and function of
class II major histocompatibility complex, CD40, and B7 expression in macrophages
and microglia: Implications in neurological diseases. Journal of Neurovirology, 8(6),
496-512.
323. Ohno-Iwashita Y, Shimada Y, Waheed AA, Hayashi M, Inomata M, Nakamura M, Maruya
M, and Iwashita S. (2004). Perfringolysin O, a Cholesterol-Binding Cytolysin, as a
Probe for Lipid Rafts. Anaerobe, 10:125-134.
324. Olano-Martin E, Anil E, Caslake MJ, Packard CJ, Bedford D, Stewart G, Peiris D, Williams
CM, and Minihane AM. (2010). Contribution of Apolipoprotein E Genotype and
Docosahexaenoic Acid to the Ldl-Cholesterol Response to Fish Oil. Atherosclerosis,
209:104-110.
325. Opazo, C., Huang, X., Cherny, R. A., Moir, R. D., Roher, A. E., White, A. R., ... & Bush, A. I.
(2002). Metalloenzyme-like activity of Alzheimer's disease β-amyloid Cu-dependent
catalytic conversion of dopamine, cholesterol, and biological reducing agents to
neurotoxic H2O2. Journal of Biological Chemistry, 277(43), 40302-40308.
326. Ordovas JM, and Schaefer EJ. (1999). Genes, Variation of Cholesterol and Fat Intake
and Serum Lipids. Current Opinion in Lipidology, 10:15-22.
327. Pacławski, A., Szlęk, J., Lau, R., Jachowicz, R., & Mendyk, A. (2015). Empirical modeling
of the fine particle fraction for carrier-based pulmonary delivery formulations.
International Journal of Nanomedicine, 10, 801.
328. Padmanabhan, K., Shpanskaya, K., Bello, G., Doraiswamy, P. M., & Samatova, N. F.
(2017). Toward Personalized Network Biomarkers in Alzheimer's Disease:
Computing Individualized Genomic and Protein Crosstalk Maps. Frontiers in Aging
Neuroscience, 9, 315.
329. Pakpoor, J., Disanto, G., Gerber, J. E., Dobson, R., Meier, U. C., Giovannoni, G., &
Ramagopalan, S. V. (2013). The risk of developing multiple sclerosis in individuals
seronegative for Epstein-Barr virus: a meta-analysis. Multiple Sclerosis
Journal, 19(2), 162-166.
330. Patterson, N., Price, A. L., & Reich, D. (2006). Population structure and
eigenanalysis. PLoS Genetics, 2(12), e190.
331. Payami H, Montee KR, Kaye JA, Bird TD, Yu C-E, Wijsman EM, and Schellenberg GD.
(1994). Alzheimer's Disease, Apolipoprotein E4, and Gender. JAMA, 271:1316-1317.
332. Pedram A, Razandi M, Wallace DC, and Levin ER. (2006). Functional Estrogen
Receptors in the Mitochondria of Breast Cancer Cells. Molecular Biology of the Cell,
17:2125-2137.
University of Southern California
202
333. Pepke, S., Ver Steeg, G. (2017). Comprehensive discovery of subsample gene
expression components by information explanation: therapeutic implications in
cancer. BMC Medical Genomics, 10, 12.
334. Petanceska SS. (2003). Age-Associated Androgen Deficiency and Alzheimer's Disease:
A Case in the Making? Journal of Alzheimer's Disease, 5:271-273.
335. Petersen, R. C., Aisen, P. S., Beckett, L. A., Donohue, M. C., Gamst, A. C., Harvey, D. J., ... &
Trojanowski, J. Q. (2010). Alzheimer's disease Neuroimaging Initiative (ADNI)
clinical characterization. Neurology, 74(3), 201-209.
336. Petersen, R. C., Caracciolo, B., Brayne, C., Gauthier, S., Jelic, V., & Fratiglioni, L. (2014).
Mild cognitive impairment: a concept in evolution. Journal of Internal
Medicine, 275(3), 214-228.
337. Phillipson, M., & Kubes, P. (2011). The neutrophil in vascular inflammation. Nature
Medicine, 17(11), 1381.
338. Pick, A. (1906). Über einen weiteren Symptomenkomplex im Rahmen der Dementia
senilis, bedingt durch umschriebene stärkere Hirnatrophie (gemischte
Apraxie). European Neurology, 19(2), 97-108.
339. Pierce, B. A. (2010). Genetics: A conceptual approach. Macmillan.
340. Pol, H. E. H., Schnack, H. G., Posthuma, D., Mandl, R. C., Baaré, W. F., van Oel, C., ... &
Bürgel, U. (2006). Genetic contributions to human brain morphology and
intelligence. Journal of Neuroscience, 26(40), 10235-10242.
341. Powell, T. R. (2018). Translating Schizophrenia Population Genetics Findings to
Neurobiological Mechanisms: The Case of KALRN-9. Biological Psychiatry, 83(6),
e41-e42.
342. Price, A. L., Zaitlen, N. A., Reich, D., & Patterson, N. (2010). New approaches to
population stratification in genome-wide association studies. Nature Reviews
Genetics, 11(7), 459.
343. Prince M, Wimo A, Guerchet M, Ali G, Wu Y, and Prina M. (2015). World Alzheimer
Report 2015. The Global Impact of Dementia. An Analysis of Prevalence, Incidence,
Cost and Trends. Alzheimer's Disease International, London.
344. Purcell, S., Neale, B., Todd-Brown, K., Thomas, L., Ferreira, M. A., Bender, D., ... & Sham,
P. C. (2007). PLINK: a tool set for whole-genome association and population-based
linkage analyses. The American Journal of Human Genetics, 81(3), 559-575.
345. Qin, W., Haroutunian, V., Katsel, P., Cardozo, C. P., Ho, L., Buxbaum, J. D., & Pasinetti, G.
M. (2009). PGC-1α expression decreases in the Alzheimer disease brain as a function
of dementia. Archives of Neurology, 66(3), 352-361.
346. Qiu C, Kivipelto M, Agüero-Torres H, Winblad B, and Fratiglioni L. (2004). Risk and
Protective Effects of the Apoe Gene Towards Alzheimer’s Disease in the
Kungsholmen Project: Variation by Age and Sex. Journal of Neurology, Neurosurgery
& Psychiatry, 75:828-833.
347. Quinn, J. F., Raman, R., Thomas, R. G., Yurko-Mauro, K., Nelson, E. B., Van Dyck, C., ... &
Shinto, L. (2010). Docosahexaenoic acid supplementation and cognitive decline in
Alzheimer disease: a randomized trial. JAMA, 304(17), 1903-1911.
348. Quist, A., Doudevski, I., Lin, H., Azimova, R., Ng, D., Frangione, B., ... & Lal, R. (2005).
Amyloid ion channels: a common structural link for protein-misfolding
University of Southern California
203
disease. Proceedings of the National Academy of Sciences of the United States of
America, 102(30), 10427-10432.
349. Raber J, Huang Y, and Ashford JW. (2004). Apoe Genotype Accounts for the Vast
Majority of Ad Risk and Ad Pathology. Neurobiology of Aging, 25:641-650.
350. Radford, D. J., Wang, K., McNelis, J. C., Taylor, A. E., Hechenberger, G., Hofmann, J., ... &
Lord, J. M. (2010). Dehydroepiandrosterone sulfate directly activates protein kinase
C-β to increase human neutrophil superoxide generation. Molecular
Endocrinology, 24(4), 813-821.
351. Raichle, M. E. (2015). The brain's default mode network. Annual Review of
Neuroscience, 38, 433-447.
352. Ramassamy C, Averill D, Beffert U, Bastianetto S, Theroux L, Lussier-Cacan S, Cohn JS,
Christen Y, Davignon J, and Quirion R. (1999). Oxidative Damage and Protection by
Antioxidants in the Frontal Cortex of Alzheimer’s Disease Is Related to the
Apolipoprotein E Genotype. Free Radical Biology and Medicine, 27:544-553.
353. Rapoport, M., Dawson, H. N., Binder, L. I., Vitek, M. P., & Ferreira, A. (2002). Tau is
essential to β-amyloid-induced neurotoxicity. Proceedings of the National Academy
of Sciences, 99(9), 6364-6369.
354. Rasmussen KL, Tybjærg‐Hansen A, Nordestgaard BG, and Frikke‐Schmidt R. (2015).
Plasma Levels of Apolipoprotein E and Risk of Dementia in the General Population.
Annals of Neurology, 77:301-311.
355. Ray, S., Britschgi, M., Herbert, C., Takeda-Uchimura, Y., Boxer, A., Blennow, K., ... & Kaye,
J. A. (2007). Classification and prediction of clinical Alzheimer's diagnosis based on
plasma signaling proteins. Nature Medicine, 13(11), 1359.
356. Reddy, P. H., & Beal, M. F. (2008). Amyloid beta, mitochondrial dysfunction and
synaptic damage: implications for cognitive decline in aging and Alzheimer's
disease. Trends in Molecular Medicine, 14(2), 45-53.
357. Redlich, E. (1898). Uber miliare Sklerose der Hirnrinde bei seniler Atrophie. Jahrb
Psychiatry Neurol, 17, 208-216.
358. Reilly S, Ferrell R, Kottke B, and Sing C. (1992). The Gender-Specific Apolipoprotein E
Genotype Influence on the Distribution of Plasma Lipids and Apolipoproteins in the
Population of Rochester, Minnesota. Ii. Regression Relationships with Concomitants.
American Journal of Human Genetics, 51:1311.
359. Reiman EM, Caselli RJ, Yun LS, Chen K, Bandy D, Minoshima S, Thibodeau SN, and
Osborne D. (1996). Preclinical Evidence of Alzheimer's Disease in Persons
Homozygous for the Ε4 Allele for Apolipoprotein E. New England Journal of
Medicine, 334:752-758.
360. Reiman EM, Chen K, Alexander GE, Caselli RJ, Bandy D, Osborne D, Saunders AM, and
Hardy J. (2004). Functional Brain Abnormalities in Young Adults at Genetic Risk for
Late-Onset Alzheimer's Dementia. Proceedings of the National Academy of Sciences,
101:284-289.
361. Reiman EM, Chen K, Alexander GE, Caselli RJ, Bandy D, Osborne D, Saunders AM, and
Hardy J. (2005). Correlations between Apolipoprotein E Ε4 Gene Dose and Brain-
Imaging Measurements of Regional Hypometabolism. Proceedings of the National
Academy of Sciences of the United States of America, 102:8299-8302.
University of Southern California
204
362. Reiman EM, Chen K, Liu X, Bandy D, Yu M, Lee W, Ayutyanont N, Keppler J, Reeder SA,
and Langbaum JB. (2009). Fibrillar Amyloid-Β Burden in Cognitively Normal People
at 3 Levels of Genetic Risk for Alzheimer's Disease. Proceedings of the National
Academy of Sciences, 106:6820-6825.
363. Reisberg, B., Ferris, S. H., De Leon, M. J., & Crook, T. (1988). Global deterioration scale
(GDS). Psychopharmacol Bull, 24(4), 661-3.
364. Remmers, C., Sweet, R. A., & Penzes, P. (2014). Abnormal kalirin signaling in
neuropsychiatric disorders. Brain Research Bulletin, 103, 29-38.
365. Replogle, J. M., Chan, G., White, C. C., Raj, T., Winn, P. A., Evans, D. A., ... & Bennett, D. A.
(2015). A TREM1 variant alters the accumulation of Alzheimer‐related amyloid
pathology. Annals of Neurology, 77(3), 469-477.
366. Rettberg JR, Dang H, Hodis HN, Henderson VW, John JAS, Mack WJ, and Brinton RD.
(2016). Identifying Postmenopausal Women at Risk for Cognitive Decline within a
Healthy Cohort Using a Panel of Clinical Metabolic Indicators: Potential for Detecting
an at-Alzheimer's Risk Metabolic Phenotype. Neurobiology of Aging, 40:155-163.
367. Rettberg JR, Yao J, and Brinton RD. (2014). Estrogen: A Master Regulator of
Bioenergetic Systems in the Brain and Body. Frontiers in Neuroendocrinology, 35:8-
30.
368. Rhein, V., Song, X., Wiesner, A., Ittner, L. M., Baysang, G., Meier, F., ... & Savaskan, E.
(2009). Amyloid-β and tau synergistically impair the oxidative phosphorylation
system in triple transgenic Alzheimer's disease mice. Proceedings of the National
Academy of Sciences, 106(47), 20057-20062.
369. Richard, G., Kolskaar, K., Sanders, A. M., Kaufmann, T., Petersen, A., Doan, N. T., ... &
Andreassen, O. (2018). Assessing distinct patterns of cognitive aging using tissue-
specific brain age prediction based on diffusion tensor imaging and brain
morphometry. bioRxiv, 313015.
370. Richartz-Salzburger, E., Batra, A., Stransky, E., Laske, C., Köhler, N., Bartels, M., ... &
Schott, K. (2007). Altered lymphocyte distribution in Alzheimer’s disease. Journal of
Psychiatric Research, 41(1), 174-178.
371. Riedel, B. C., Jahanshad, N., & Thompson, P. M. (2017, April). Graph theoretical
approaches towards understanding differences in frontoparietal and default mode
networks in Autism. In Biomedical Imaging (ISBI 2017), 2017 IEEE 14th
International Symposium on (pp. 460-463). IEEE.
372. Riedel, B. C., Thompson, P. M., & Brinton, R. D. (2016). Age, APOE and sex: triad of risk
of Alzheimer’s disease. The Journal of Steroid Biochemistry and Molecular Biology,
160, 134-147.
373. Rimol, L. M., Nesvåg, R., Hagler, D. J., Bergmann, Ø., Fennema-Notestine, C., Hartberg, C.
B., ... & Melle, I. (2012). Cortical volume, surface area, and thickness in schizophrenia
and bipolar disorder. Biological Psychiatry, 71(6), 552-560.Rogers, J., Li, R.,
Mastroeni, D., Grover, A., Leonard, B., Ahern, G., ... & Sabbagh, M. (2006). Peripheral
clearance of amyloid β peptide by complement C3-dependent adherence to
erythrocytes. Neurobiology of Aging, 27(12), 1733-1739.
374. Roe, C. M., Behrens, M. I., Xiong, C., Miller, J. P., & Morris, J. C. (2005). Alzheimer disease
and cancer. Neurology, 64(5), 895-898.
University of Southern California
205
375. Rogers, J., Cooper, N. R., Webster, S., Schultz, J., McGeer, P. L., Styren, S. D., ... & Ward, P.
(1992). Complement activation by beta-amyloid in Alzheimer disease. Proceedings
of the National Academy of Sciences, 89(21), 10016-10020.
376. Roher, A. E., Lowenson, J. D., Clarke, S., Woods, A. S., Cotter, R. J., Gowing, E., & Ball, M. J.
(1993). beta-Amyloid-(1-42) is a major component of cerebrovascular amyloid
deposits: implications for the pathology of Alzheimer disease. Proceedings of the
National Academy of Sciences, 90(22), 10836-10840.
377. Roses A, Lutz M, Amrine-Madsen H, Saunders A, Crenshaw D, Sundseth S, Huentelman
M, Welsh-Bohmer K, and Reiman E. (2010). A Tomm40 Variable-Length
Polymorphism Predicts the Age of Late-Onset Alzheimer's Disease. The
Pharmacogenomics Journal, 10:375-384.
378. Rosvall L, Rizzuto D, Wang H-X, Winblad B, Graff C, and Fratiglioni L. (2009). Apoe-
Related Mortality: Effect of Dementia, Cardiovascular Disease and Gender.
Neurobiology of Aging, 30:1545-1551.
379. Roth, M., Tomlinson, B. E., & Blessed, G. (1966). Correlation between scores for
dementia and counts of ‘senile plaques’ in cerebral grey matter of elderly
subjects. Nature, 209(5018), 109.
380. Ruitenberg A, Ott A, van Swieten JC, Hofman A, and Breteler MM. (2001). Incidence of
Dementia: Does Gender Make a Difference? Neurobiology of Aging, 22:575-580.
381. Ryan J, Carriere I, Carcaillon L, Dartigues J-F, Auriacombe S, Rouaud O, Berr C, Ritchie
K, Scarabin P-Y, and Ancelin M-L. (2014). Estrogen Receptor Polymorphisms and
Incident Dementia: The Prospective 3c Study. Alzheimer's & Dementia, 10:27-35.
382. Sanabria-Diaz, G., Melie-García, L., Iturria-Medina, Y., Alemán-Gómez, Y., Hernández-
González, G., Valdés-Urrutia, L., ... & Valdés-Sosa, P. (2010). Surface area and cortical
thickness descriptors reveal different attributes of the structural human brain
networks. Neuroimage, 50(4), 1497-1510.
383. Sanjurjo, L., Amézaga, N., Aran, G., Naranjo-Gómez, M., Arias, L., Armengol, C., ... &
Sarrias, M. R. (2015). The human CD5L/AIM-CD36 axis: A novel autophagy inducer
in macrophages that modulates inflammatory responses. Autophagy, 11(3), 487-
502.
384. Santoro N, and Sutton-Tyrrell K. (2011). The Swan Song: Study of Women’s Health
across the Nation’s Recurring Themes. Obstetrics and Gynecology Clinics of North
America, 38:417-423.
385. Satirapoj B, Supasyndh O, Dispan R, Punpanich D, Tribanyatkul S, and Choovichian P.
(2013). Apolipoprotein E Genetic Polymorphisms and the Development of
Nephropathy in Type 2 Diabetes. J Med Assoc Thai, 96:1119-1126.
386. Saunders A. (1993). Strittmatter Wj1 Schmechel D, George-Hyslop Ph1 Pericak-Vance
Ma1 Joo Sh, Rosi Bl1 Gusella Jf, Crappermaclachlan Dr, Alberts Mj, Hulette Cm, Crain
Bj, Goldgaber D, Roses Ad: Association of Apolipoprotein E Allele Epsilon 4 with
Late-Onset Familial and Sporadic Alzheimer's Disease. Neurology, 43:1467-1472.
387. Saykin, A. J., Shen, L., Foroud, T. M., Potkin, S. G., Swaminathan, S., Kim, S., ... &
Thompson, P. M. (2010). Alzheimer's Disease Neuroimaging Initiative biomarkers as
quantitative phenotypes: genetics core aims, progress, and plans. Alzheimer's &
dementia: The Journal of the Alzheimer's Association, 6(3), 265-273.
University of Southern California
206
388. Schachter F, Faure-Delanef L, Guénot F, Rouger H, Froguel P, Lesueur-Ginot L, and
Cohen D. (1994). Genetic Associations with Human Longevity at the Apoe and Ace
Loci. Nature Genetics, 6:29-32.
389. Schachter, A., & Davis, K. (2000). Research and practice in Alzheimer's Disease.
Volume 3. Research and Practice in Alzheimer's Disease, 3, 67.
390. Scheff, S. W., & Price, D. A. (2003). Synaptic pathology in Alzheimer’s disease: a review
of ultrastructural studies. Neurobiology of Aging, 24(8), 1029-1046.
391. Schiele F, De Bacquer D, Vincent-Viry M, Beisiegel U, Ehnholm C, Evans A, Kafatos A,
Martins M, Sans S, and Sass C. (2000). Apolipoprotein E Serum Concentration and
Polymorphism in Six European Countries: The Apoeurope Project. Atherosclerosis,
152:475-488.
392. Seijo-Pardo, B., Porto-Díaz, I., Bolón-Canedo, V., & Alonso-Betanzos, A. (2017).
Ensemble feature selection: homogeneous and heterogeneous approaches.
Knowledge-Based Systems, 118, 124-139.
393. Serra-Grabulosa J, Salgado-Pineda P, Junqué C, Sole-Padulles C, Moral P, López-Alomar
A, Lopez T, Lopez-Guillen A, Bargallo N, and Mercader J. (2003). Apolipoproteins E
and C1 and Brain Morphology in Memory Impaired Elders. Neurogenetics, 4:141-
146.
394. Shachter N, Hayek T, Leff T, Smith J, Rosenberg D, Walsh A, Ramakrishnan R, Goldberg
I, Ginsberg H, and Breslow J. (1994). Overexpression of Apolipoprotein Cii Causes
Hypertriglyceridemia in Transgenic Mice. Journal of Clinical Investigation, 93:1683.
395. Shalaurova I, Connelly MA, Garvey WT, and Otvos JD. (2014). Lipoprotein Insulin
Resistance Index: A Lipoprotein Particle–Derived Measure of Insulin Resistance.
Metabolic Syndrome and Related Disorders, 12:422-429.
396. Sharman MJ, Kraemer WJ, Love DM, Avery NG, Gómez AL, Scheett TP, and Volek JS.
(2002). A Ketogenic Diet Favorably Affects Serum Biomarkers for Cardiovascular
Disease in Normal-Weight Men. The Journal of Nutrition, 132:1879-1885.
397. Shaw AC, Goldstein DR, and Montgomery RR. (2013). Age-Dependent Dysregulation of
Innate Immunity. Nature Reviews Immunology, 13:875-887.
398. Shaw P, Lerch JP, Pruessner JC, Taylor KN, Rose AB, Greenstein D, Clasen L, Evans A,
Rapoport JL, and Giedd JN. (2007). Cortical Morphology in Children and Adolescents
with Different Apolipoprotein E Gene Polymorphisms: An Observational Study. The
Lancet Neurology, 6:494-500.
399. Shaw, L. M., Vanderstichele, H., Knapik‐Czajka, M., Clark, C. M., Aisen, P. S., Petersen, R.
C., ... & Dean, R. (2009). Cerebrospinal fluid biomarker signature in Alzheimer's
disease neuroimaging initiative subjects. Annals of Neurology, 65(4), 403-413.
400. Sheffler J, Moxley J, and Sachs-Ericsson N. (2014). Stress, Race, and Apoe:
Understanding the Interplay of Risk Factors for Changes in Cognitive Functioning.
Aging & Mental Health, 18:784-791.
401. Sheline YI, Morris JC, Snyder AZ, Price JL, Yan Z, D'Angelo G, Liu C, Dixit S, Benzinger T,
and Fagan A. (2010). Apoe4 Allele Disrupts Resting State Fmri Connectivity in the
Absence of Amyloid Plaques or Decreased Csf Aβ42. The Journal of Neuroscience,
30:17035-17040.
University of Southern California
207
402. Shi, J., Xiang, Y., & Simpkins, J. W. (1997). Hypoglycemia enhances the expression of
mRNA encoding β-amyloid precursor protein in rat primary cortical astroglial
cells. Brain Research, 772(1-2), 247-251.
403. Simell, B., Vuorela, A., Ekström, N., Palmu, A., Reunanen, A., Meri, S., ... & Väkeväinen, M.
(2011). Aging reduces the functionality of anti-pneumococcal antibodies and the
killing of Streptococcus pneumoniae by neutrophil phagocytosis. Vaccine, 29(10),
1929-1934.
404. Singh, J., Grundke-Iqbal, I., McDonald, B., & Iqbal, K. (1994). Comparison of the
phosphorylation of microtubule-associated protein tau by non-proline dependent
protein kinases. Molecular and Cellular Biochemistry, 131(2), 181-189.
405. Singh, R., Barden, A., Mori, T., & Beilin, L. (2001). Advanced glycation end-products: a
review. Diabetologia, 44(2), 129-146.
406. Sinha, M., Bhowmick, P., Banerjee, A., & Chakrabarti, S. (2013). Antioxidant role of
amyloid β protein in cell-free and biological systems: implication for the
pathogenesis of Alzheimerdisease. Free Radical Biology and Medicine, 56, 184-192.
407. Smith C, Andersen A, Kryscio R, Schmitt F, Kindy M, Blonder L, and Avison M. (2002).
Women at Risk for Ad Show Increased Parietal Activation During a Fluency Task.
Neurology, 58:1197-1202.
408. Smith, S. M., Jenkinson, M., Woolrich, M. W., Beckmann, C. F., Behrens, T. E., Johansen-
Berg, H., ... & Niazy, R. K. (2004). Advances in functional and structural MR image
analysis and implementation as FSL. Neuroimage, 23, S208-S219.
409. Soares H, Potter W, Pickering E, Kuhn M, Immermann F, Shera D, Ferm M, Dean R,
Simon A, and Swenson F. (2012). Biomarkers Consortium Alzheimer’s Disease
Plasma Proteomics Project. Plasma Biomarkers Associated with the Apolipoprotein
E Genotype and Alzheimer Disease. Arch Neurol, 69:1310-1317.
410. Soares, H. D., Potter, W. Z., Pickering, E., Kuhn, M., Immermann, F. W., Shera, D. M., ... &
Siuciak, J. A. (2012). Plasma biomarkers associated with the apolipoprotein E
genotype and Alzheimer disease. Archives of Neurology, 69(10), 1310-1317.
411. Song, J. (2015). Bias corrections for Random Forest in regression using residual
rotation. Journal of the Korean Statistical Society, 44(2), 321-326.
412. Srivastava RAK, Srivastava N, Averna M, Lin RC, Korach KS, Lubahn DB, and Schonfeld
G. (1997). Estrogen up-Regulates Apolipoprotein E (Apoe) Gene Expression by
Increasing Apoe Mrna in the Translating Pool Via the Estrogen Receptor Α-Mediated
Pathway. Journal of Biological Chemistry, 272:33360-33366.
413. Srivastava RAK. (2002). Estrogen-Induced Regulation of the Atp-Binding Cassette
Transporter A1 (Abca1) in Mice: A Possible Mechanism of Atheroprotection by
Estrogen. Molecular and Cellular Biochemistry, 240:67-73.
414. Steel, D. M., & Whitehead, A. S. (1994). The major acute phase reactants: C-reactive
protein, serum amyloid P component and serum amyloid A protein. Immunology
Today, 15(2), 81-88.
415. Stelzmann, R. A., Norman Schnitzlein, H., & Reed Murtagh, F. (1995). An English
translation of Alzheimer's 1907 paper,“Über eine eigenartige Erkankung der
Hirnrinde”. Clinical Anatomy, 8(6), 429-431.
416. Storsve, A. B., Fjell, A. M., Tamnes, C. K., Westlye, L. T., Overbye, K., Aasland, H. W., &
Walhovd, K. B. (2014). Differential longitudinal changes in cortical thickness, surface
University of Southern California
208
area and volume across the adult life span: regions of accelerating and decelerating
change. Journal of Neuroscience, 34(25), 8488-8498.
417. Strittmatter W, Weisgraber K, Goedert M, Saunders A, Huang D, Corder E, Dong L,
Jakes R, Alberts M, and Gilbert J. (1994). Hypothesis: Microtubule Instability and
Paired.
418. Struble RG, Cady C, Nathan BP, and McAsey M. (2007). Apolipoprotein E May Be a
Critical Factor in Hormone Therapy Neuroprotection. Frontiers in Bioscience: A
Journal and Virtual Library, 13:5387-5405.
419. Struble, R. G., Nathan, B. P., Cady, C., Cheng, X., & McAsey, M. (2007). Estradiol
regulation of astroglia and apolipoprotein E: an important role in neuronal
regeneration. Experimental Gerontology, 42(1-2), 54-63.
420. Su YR, Blakemore JL, Zhang Y, Linton MF, and Fazio S. (2008). Lentiviral Transduction
of Apoai into Hematopoietic Progenitor Cells and Macrophages Applications to Cell
Therapy of Atherosclerosis. Arteriosclerosis, Thrombosis, and Vascular Biology,
28:1439-1446.
421. Sudlow, C., Gallacher, J., Allen, N., Beral, V., Burton, P., Danesh, J., ... & Liu, B. (2015). UK
biobank: an open access resource for identifying the causes of a wide range of
complex diseases of middle and old age. PLoS Medicine, 12(3), e1001779.
422. Tang, M. X., Stern, Y., Marder, K., Bell, K., Gurland, B., Lantigua, R., ... & Mayeux, R.
(1998). The APOE-∊ 4 allele and the risk of Alzheimer disease among African
Americans, whites, and Hispanics. Jama, 279(10), 751-755.
423. Team, R. C. (2013). R: A language and environment for statistical computing.
424. Teng E, Chow N, Hwang KS, Thompson PM, Gylys KH, Cole GM, Jack Jr CR, Shaw LM,
Trojanowski JQ, and Soares HD. (2015). Low Plasma Apoe Levels Are Associated
with Smaller Hippocampal Size in the Alzheimer's Disease Neuroimaging Initiative
Cohort. Dementia and Geriatric Cognitive Disorders, 39:154-166.
425. Theendakara, V., Peters-Libeu, C. A., Spilman, P., Poksay, K. S., Bredesen, D. E., & Rao, R.
V. (2016). Direct transcriptional effects of apolipoprotein E. Journal of
Neuroscience, 36(3), 685-700.
426. Tomlinson, B. E., Blessed, G., & Roth, M. (1970). Observations on the brains of
demented old people. Journal of the Neurological Sciences, 11(3), 205-242.
427. Torzewski, J., Torzewski, M., Bowyer, D. E., Fröhlich, M., Koenig, W., Waltenberger, J., ...
& Hombach, V. (1998). C-reactive protein frequently colocalizes with the terminal
complement complex in the intima of early atherosclerotic lesions of human
coronary arteries. Arteriosclerosis, Thrombosis, and Vascular Biology, 18(9), 1386-
1392.
428. Trabzuni, D., Ramasamy, A., Imran, S., Walker, R., Smith, C., Weale, M. E., ... & North
American Brain Expression Consortium. (2013). Widespread sex differences in gene
expression and splicing in the adult human brain. Nature Communications, 4, 2771.
429. Tyan, S. H., Shih, A. Y. J., Walsh, J. J., Maruyama, H., Sarsoza, F., Ku, L., ... & Dickstein, D. L.
(2012). Amyloid precursor protein (APP) regulates synaptic structure and
function. Molecular and Cellular Neuroscience, 51(1), 43-52.
430. Van Coillie, E., Van Damme, J., & Opdenakker, G. (1999). The MCP/eotaxin subfamily of
CC chemokines. Cytokine & Growth Factor Reviews, 10(1), 61-86.
University of Southern California
209
431. Vance JE, and Hayashi H. (2010). Formation and Function of Apolipoprotein E-
Containing Lipoproteins in the Nervous System. Biochimica et Biophysica Acta
(BBA)-Molecular and Cell Biology of Lipids, 1801:806-818.
432. Vasan, S. K., Hwang, J., Rostgaard, K., Nyrén, O., Ullum, H., Pedersen, O. B., ... & Edgren,
G. (2016). ABO blood group and risk of cancer: a register-based cohort study of 1.6
million blood donors. Cancer Epidemiology, 44, 40-43.
433. Ver Steeg, G. (2017). Unsupervised learning via total correlation explanation. arXiv
Preprint arXiv:1706.08984.
434. Ver Steeg, G., & Galstyan, A. (2014). Discovering structure in high-dimensional data
through correlation explanation. In Advances in Neural Information Processing
Systems (pp. 577-585).
435. Ver Steeg, G., & Galstyan, A. (2015, February). Maximally informative hierarchical
representations of high-dimensional data. In Artificial Intelligence and Statistics (pp.
1004-1012).
436. Villemagne VL, Pike KE, Chetelat G, Ellis KA, Mulligan RS, Bourgeat P, Ackermann U,
Jones G, Szoeke C, and Salvado O. (2011). Longitudinal Assessment of Aβ and
Cognition in Aging and Alzheimer Disease. Annals of Neurology, 69:181-192.
437. Vincent-Viry M, Schiele F, Gueguen R, Bohnet K, Visvikis S, and Siest G. (1998).
Biological Variations and Genetic Reference Values for Apolipoprotein E Serum
Concentrations: Results from the Stanislas Cohort Study. Clinical Chemistry, 44:957-
965.
438. Visscher, P. M., Hill, W. G., & Wray, N. R. (2008). Heritability in the genomics era—
concepts and misconceptions. Nature reviews genetics, 9(4), 255.
439. Vogelsberg-Ragaglia, V., Schuck, T., Trojanowski, J. Q., & Lee, V. M. Y. (2001). PP2A
mRNA expression is quantitatively decreased in Alzheimer's disease
hippocampus. Experimental Neurology, 168(2), 402-412.
440. Von Dungern, E., & Hirszfeld, L. (1911). On the group-specific structures of the blood.
III. Z Immunitatsforsch, 8, 526-562.
441. Vossel, K. A., Zhang, K., Brodbeck, J., Daub, A. C., Sharma, P., Finkbeiner, S., ... & Mucke,
L. (2010). Tau reduction prevents Aβ-induced defects in axonal
transport. Science, 330(6001), 198-198.
442. Wallenfeldt K, Bokemark L, Wikstrand J, Hulthe J, and Fagerberg B. (2004).
Apolipoprotein B/Apolipoprotein Ai in Relation to the Metabolic Syndrome and
Change in Carotid Artery Intima-Media Thickness During 3 Years in Middle-Aged
Men. Stroke, 35:2248-2252.
443. Walsh KP, et al: Amyloid-beta and proinflammatory cytokines utilize a prion protein-
dependent pathway to activate NADPH oxidase and induce cofilin-actin rods in
hippocampal neurons. PLos One 2014;9: e95995.
444. Wang JM, Irwin RW, and Brinton RD. (2006). Activation of Estrogen Receptor Α
Increases and Estrogen Receptor Β Decreases Apolipoprotein E Expression in
Hippocampus in Vitro and in Vivo. Proceedings of the National Academy of Sciences,
103:16983-16988.
445. Wang, C., Yosef, N., Gaublomme, J., Wu, C., Lee, Y., Clish, C. B., ... & Kishi, Y. (2015).
CD5L/AIM regulates lipid biosynthesis and restrains Th17 cell pathogenicity. Cell,
163(6), 1413-1427.
University of Southern California
210
446. Wang, W. Y., Tan, M. S., Yu, J. T., & Tan, L. (2015). Role of pro-inflammatory cytokines
released from microglia in Alzheimer’s disease. Annals of Translational Medicine,
3(10).
447. Wang, Y. J., Zhou, H. D., & Zhou, X. F. (2006). Clearance of amyloid-beta in Alzheimer's
disease: progress, problems and perspectives. Drug Discovery Today, 11(19-20),
931-938.
448. Weiner, M. W., Veitch, D. P., Aisen, P. S., Beckett, L. A., Cairns, N. J., Green, R. C., ... &
Petersen, R. C. (2017). Recent publications from the Alzheimer's Disease
Neuroimaging Initiative: Reviewing progress toward improved AD clinical trials.
Alzheimer's & Dementia: The Journal of the Alzheimer's Association, 13(4), e1-e85.
449. Westman, E., Muehlboeck, J. S., & Simmons, A. (2012). Combining MRI and CSF
measures for classification of Alzheimer's disease and prediction of mild cognitive
impairment conversion. Neuroimage, 62(1), 229-238.
450. Westman, E., Simmons, A., Muehlboeck, J. S., Mecocci, P., Vellas, B., Tsolaki, M., ... &
Spenger, C. (2011). AddNeuroMed and ADNI: similar patterns of Alzheimer's
atrophy and automated MRI classification accuracy in Europe and North
America. Neuroimage, 58(3), 818-828.
451. Wiggins, K. L., Smith, N. L., Glazer, N. L., Rosendaal, F. R., Heckbert, S. R., Psaty, B. M., ...
& Lumley, T. (2009). ABO genotype and risk of thrombotic events and hemorrhagic
stroke. Journal of Thrombosis and Haemostasis, 7(2), 263-269.
452. Winblad, B., Palmer, K., Kivipelto, M., Jelic, V., Fratiglioni, L., Wahlund, L. O., ... & Arai, H.
(2004). Mild cognitive impairment–beyond controversies, towards a consensus:
report of the International Working Group on Mild Cognitive Impairment. Journal of
Internal Medicine, 256(3), 240-246.
453. Wolbink, G. J., Brouwer, M. C., Buysmann, S., Ten Berge, I. J., & Hack, C. E. (1996). CRP-
mediated activation of complement in vivo: assessment by measuring circulating
complement-C-reactive protein complexes. The Journal of Immunology, 157(1),
473-479.
454. Wolpin, B. M., Kraft, P. L., Xu, M., Steplowski, E., Olsson, M. L., Arslan, A. A., ... & LaCroix,
A. (2010). Variant ABO blood group alleles, secretor status and risk of pancreatic
cancer: results from the pancreatic cancer cohort consortium. Cancer Epidemiology
and Prevention Biomarkers, cebp-0751.
455. Wood RJ, Volek JS, Liu Y, Shachter NS, Contois JH, and Fernandez ML. (2006).
Carbohydrate Restriction Alters Lipoprotein Metabolism by Modifying Vldl, Ldl, and
Hdl Subfraction Distribution and Size in Overweight Men. The Journal of Nutrition,
136:384-389.
456. World Health Organization. (2015). World health statistics 2015. World Health
Organization.
457. Xu, P. T., Gilbert, J. R., Qiu, H. L., Rothrock-Christian, T., Settles, D. L., Roses, A. D., &
Schmechel, D. E. (1998). Regionally specific neuronal expression of human APOE
gene in transgenic mice. Neuroscience Letters, 246(2), 65-68.
458. Xu Q, Bernardo A, Walker D, Kanegawa T, Mahley RW, and Huang Y. (2006). Profile
and Regulation of Apolipoprotein E (Apoe) Expression in the Cns in Mice with
Targeting of Green Fluorescent Protein Gene to the Apoe Locus. The Journal of
Neuroscience, 26:4985-4994.
University of Southern California
211
459. Yaffe K, Haan M, Byers A, Tangen C, and Kuller L. (2000). Estrogen Use, Apoe, and
Cognitive Decline Evidence of Gene–Environment Interaction. Neurology, 54:1949-
1954.
460. Yamaguchi, H., Hirai, S., Morimatsu, M., Shoji, M., & Nakazato, Y. (1989). Diffuse type of
senile plaques in the cerebellum of Alzheimer-type dementia demonstrated byβ
protein immunostain. Acta Neuropathologica, 77(3), 314-319.
461. Yamamoto, M., Yamamoto, F., Luong, T. T., Williams, T., Kominato, Y., & Yamamoto, F.
(2003). Expression profiling of 68 glycosyltransferase genes in 27 different human
tissues by the systematic multiplex reverse transcription‐polymerase chain reaction
method revealed clustering of sexually related tissues in hierarchical clustering
algorithm analysis. Electrophoresis, 24(14), 2295-2307.
462. Yancy WS, Olsen MK, Guyton JR, Bakst RP, and Westman EC. (2004). A Low-
Carbohydrate, Ketogenic Diet Versus a Low-Fat Diet to Treat Obesity and
Hyperlipidemia: A Randomized, Controlled Trial. Annals of Internal Medicine,
140:769-777.
463. Yang, C., Wan, X., Yang, Q., Xue, H., & Yu, W. (2010). Identifying main effects and
epistatic interactions from large-scale SNP data via adaptive group Lasso. BMC
bioinformatics, 11(1), S18.
464. Yang, J., Zaitlen, N. A., Goddard, M. E., Visscher, P. M., & Price, A. L. (2014). Advantages
and pitfalls in the application of mixed-model association methods. Nature
Genetics, 46(2), 100.
465. Yang, X., Huang, Y., & Feng, J. F. (2014). Is there an association between ABO blood
group and overall survival in patients with esophageal squamous cell
carcinoma?. International Journal of Clinical and Experimental Medicine, 7(8), 2214.
466. Yao J, and Brinton RD. (2012). Estrogen Regulation of Mitochondrial Bioenergetics:
Implications for Prevention of Alzheimer’s Disease. Advances in Pharmacology (San
Diego, Calif.), 64:327.
467. Yao J, Hamilton RT, Cadenas E, and Brinton RD. (2010). Decline in Mitochondrial
Bioenergetics and Shift to Ketogenic Profile in Brain During Reproductive
Senescence. Biochimica et Biophysica Acta (BBA)-General Subjects, 1800:1121-
1126.
468. Yao J, Irwin R, Chen S, Hamilton R, Cadenas E, and Brinton RD. (2012). Ovarian
Hormone Loss Induces Bioenergetic Deficits and Mitochondrial Β-Amyloid.
Neurobiology of Aging, 33:1507-1521.
469. Yao J, Irwin RW, Zhao L, Nilsen J, Hamilton RT, and Brinton RD. (2009). Mitochondrial
Bioenergetic Deficit Precedes Alzheimer's Pathology in Female Mouse Model of
Alzheimer's Disease. Proceedings of the National Academy of Sciences, 106:14670-
14675.
470. Yao J, Rettberg JR, Klosinski LP, Cadenas E, and Brinton RD. (2011). Shift in Brain
Metabolism in Late Onset Alzheimer’s Disease: Implications for Biomarkers and
Therapeutic Interventions. Molecular Aspects of Medicine, 32:247-257.
471. Yasojima, K., Schwab, C., McGeer, E. G., & McGeer, P. L. (2000). Human neurons
generate C-reactive protein and amyloid P: upregulation in Alzheimer’s disease.
Brain Research, 887(1), 80-89.
University of Southern California
212
472. Yeo, B. T., Krienen, F. M., Chee, M. W., & Buckner, R. L. (2014). Estimates of segregation
and overlap of functional connectivity networks in the human cerebral
cortex. Neuroimage, 88, 212-227.
473. Yeoman M, Scutt G, and Faragher R. (2012). Insights into Cns Ageing from Animal
Models of Senescence. Nature Reviews Neuroscience, 13:435-445.
474. Yi, H., Xue, L., Guo, M. X., Ma, J., Zeng, Y., Wang, W., ... & Li, W. X. (2010). Gene expression
atlas for human embryogenesis. The FASEB Journal, 24(9), 3341-3350.
475. Yin F, Yao J, Sancheti H, Feng T, Melcangi RC, Morgan TE, Finch CE, Pike CJ, Mack WJ,
and Cadenas E. (2015). The Perimenopausal Aging Transition in the Female Rat
Brain: Decline in Bioenergetic Systems and Synaptic Plasticity. Neurobiology of
Aging, 36:2282-2295.
476. Youn, H., Jeoung, M., Koo, Y., Ji, H., Markesbery, W. R., Ji, I., & Ji, T. H. (2007). Kalirin is
under-expressed in Alzheimer's disease hippocampus. Journal of Alzheimer's
Disease, 11(3), 385-397.
477. Yu, C. E., Seltman, H., Peskind, E. R., Galloway, N., Zhou, P. X., Rosenthal, E., ... &
Schellenberg, G. D. (2007). Comprehensive analysis of APOE and selected proximate
markers for late-onset Alzheimer's disease: patterns of linkage disequilibrium and
disease/marker association. Genomics, 89(6), 655-665.
478. Yusuf, S., Reddy, S., Ôunpuu, S., & Anand, S. (2001). Global burden of cardiovascular
diseases: Part II: variations in cardiovascular disease by specific ethnic groups and
geographic regions and prevention strategies. Circulation, 104(23), 2855-2864.
479. Zenaro, E., Pietronigro, E., Della Bianca, V., Piacentino, G., Marongiu, L., Budui, S., ... &
Montresor, A. (2015). Neutrophils promote Alzheimer's disease–like pathology and
cognitive decline via LFA-1 integrin. Nature Medicine, 21(8), 880.
480. Zhang, B., Gaiteri, C., Bodea, L. G., Wang, Z., McElwee, J., Podtelezhnikov, A. A., ... &
Fluder, E. (2013). Integrated systems approach identifies genetic nodes and
networks in late-onset Alzheimer’s disease. Cell, 153(3), 707-720.
481. Zhang, J., Ke, K. F., Liu, Z., Qiu, Y. H., & Peng, Y. P. (2013). Th17 cell-mediated
neuroinflammation is involved in neurodegeneration of aβ1-42-induced Alzheimer’s
disease model rats. PLos One, 8(10), e75786.
482. Zheng, X., Zhang, X., Kang, A., Ran, C., Wang, G., & Hao, H. (2015). Thinking outside the
brain for cognitive improvement: Is peripheral immunomodulation on the way?.
Neuropharmacology, 96, 94-104.
483. Zhong N, and Weisgraber KH. (2009). Understanding the Association of
Apolipoprotein E4 with Alzheimer Disease: Clues from Its Structure. Journal of
Biological Chemistry, 284:6027-6031.
484. Zhou, X., & Stephens, M. (2012). Genome-wide efficient mixed-model analysis for
association studies. Nature Genetics, 44(7), 821.
Abstract (if available)
Abstract
The work outlined in this thesis involves methods that were designed, accelerated, and applied in order to solve problems in the following two areas as they pertain to aging and Alzheimer’s disease (AD): 1) biomarker discovery or refinement in plasma or brain, 2) exploration of genetic contributions, mechanisms, and pathways in mediating deviations from optimal aging and risk for Alzheimer’s disease. The overarching goal of this thesis is to address complex neuroinformatics and high-dimensional questions related to the brain in a more clinically relevant way, such that the gap between the two disciplines may potentially be lessened, and therefore, more salient risk factors identified, and targeted outcomes become more approachable. While each of these fields has the potential to inform the other, they are often treated separately. ❧ In the first chapter of this thesis, I provide an extended review of background literature relevant to the studies that follow. This chapter provides an overview of aging and Alzheimer’s disease, and lays the groundwork for the necessary genetic concepts and methods used in chapter three to chapter six. Chapter two is adapted from our published review paper on apolipoprotein E (APOE), the greatest genetic risk factor for late-onset AD. This chapter discusses the interactions with APOE, age, and sex, in mediating this risk, both in consideration of amyloid and other AD hallmarks, but also through amyloid-independent mechanisms, such as lipid metabolism and the immune system. Chapter three is adapted from a paper that is currently under review at Frontiers in Aging Neuroscience and explores how a multivariate network approach using brain and proteomic measures at baseline can inform prediction of disease across the trajectory of AD, improving our ability to detect progression to mild cognitive impairment and AD, beyond simple univariate methods. ❧ All participants included in chapters three and four were from the the Alzheimer’s Disease Neuroimaging Initiative (ADNI) or AddNeuroMed. The diagnosis of probable AD for both of these studies was performed according to NINCDS-ADRDA clinical AD criteria. Participants included in chapters five and six were from the United Kingdom (UK) Biobank. To improve training of brain-age for older individuals (i.e. 75 and older), stable (> 6 months) cognitively normal individuals from ADNI and the Open Access Series of Imaging Studies (OASIS) were additionally included in chapter six. ❧ The work discussed herein contributes to the state of the field by exploring the role of ABO blood type in AD, which to date has not been directly established. This work is presented in chapter four. Given the established interplay between age and risk for AD, the latter two studies in this thesis focus on increasing our understanding of the transitional state of optimal aging and deviations from an optimal aging trajectory with the idea that while separate from AD directly, studying health informs potential protective mechanisms from disease
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
Vascular contributions to brain aging along the Alzheimer's disease continuum
PDF
Neuroimaging markers of risk & resilience to brain aging and dementia
PDF
Age-related androgen depletion and the development of Alzheimer's disease
PDF
From risk mitochondrial and metabolic phenotype towards a precision medicine approach for Alzheimer's disease
PDF
Novel multi-site brain imaging approaches to map HIV-related neuropathology
PDF
Alzheimer’s disease: dysregulated genes, ethno-racial disparities, and environmental pollution
PDF
Affective neuropsychiatric symptoms and neural connectivity in the early stages of Alzheimer’s disease
PDF
Neuroinflammation and ApoE4 genotype in at-risk female aging: implications for Alzheimer's disease
PDF
Decoding the neurological and genetic underpinnings of chronic pain
PDF
TREM2 and C1q signaling regulates immunoproteostasis in Alzheimer's disease
PDF
Characterizing brain aging with neuroimaging, health, and genetic data
PDF
The role of inflammation in mediating effects of obesity on Alzheimer's disease
PDF
Brainstem structural integrity in the progression of Alzheimer's disease
PDF
Role of CD33 structure and function in Alzheimer’s disease
PDF
Memory abnormalities in Alzheimer's disease and anxiety models
PDF
Blood-brain barrier pathophysiology in cognitive impairment and injury
PDF
Independent and interactive effects of depression genetic risk and household socioeconomic status on emotional behavior and brain development
PDF
Blood-brain barrier breakdown and vessel integrity in Alzheimer’s disease
PDF
Estrogen and progesterone-based hormone therapy and the development of Alzheimer's disease
PDF
An overview of Alzheimer’s disease and the potential for APP-beta secretase-interaction as a therapeutic target
Asset Metadata
Creator
Riedel, Brandalyn Cherish
(author)
Core Title
Using neuroinformatics to identify genomic and proteomic markers of suboptimal aging and Alzheimer's disease
School
College of Letters, Arts and Sciences
Degree
Doctor of Philosophy
Degree Program
Neuroscience
Publication Date
05/05/2019
Defense Date
06/11/2018
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
aging,Alzheimer's,bioinformatics,brain imaging,MRI,Neuroscience,OAI-PMH Harvest,omics
Format
application/pdf
(imt)
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Liew, Sook-Lei (
committee chair
), Chui, Helena (
committee member
), Jahanshad, Neda (
committee member
), Thompson, Paul M. (
committee member
)
Creator Email
bcriedel@usc.edu,riedelbc@gmail.com
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-c89-101838
Unique identifier
UC11675938
Identifier
etd-RiedelBran-6929.pdf (filename),usctheses-c89-101838 (legacy record id)
Legacy Identifier
etd-RiedelBran-6929.pdf
Dmrecord
101838
Document Type
Dissertation
Format
application/pdf (imt)
Rights
Riedel, Brandalyn Cherish
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
Tags
Alzheimer's
bioinformatics
brain imaging
MRI
omics