University of Southern California Dissertations and Theses

Alzheimer’s disease: dysregulated genes, ethno-racial disparities, and environmental pollution

(USC Thesis Other)

Alzheimer’s disease: dysregulated genes, ethno-racial disparities, and environmental pollution

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content
Alzheimer’s disease: dysregulated genes, ethno-racial disparities, and environmental pollution

by
Nibal Arzouni

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

August 2022

Copyright 2022 Nibal Arzouni
ii

Dedication

This research work is dedicated to my family and all people who supported me, encouraged
me, and believed in my abilities and skills throughout these years.

This research work is also dedicated to all people who suffered and are suffering from
Alzheimer’s disease, cognitive impairment, any brain disease, and dementia type. It is also
dedicated to everyone who took care of loved ones who suffered from dementia.

iii

Acknowledgements

From my early school years and childhood, I had the curiosity to learn and explore new
scientific ideas. I had the passion to pursue and learn science which was my favorite topic in my
school years. This passion, curiosity, and thirst for scientific knowledge grew stronger and stronger
in my middle and high school years. While I was finishing my undergraduate degree in electrical
engineering and mathematics at the University of Michigan in Ann Arbor, I knew that just an
undergraduate degree won’t be enough for me to quench this thirst for science especially that I
graduated with highest distinction summa cum laude which gave me more courage, enthusiasm,
and confidence to pursue my dreams. I knew I wanted to continue a doctorate degree while I was
still finishing my undergraduate degree.
I decided to join the University of Southern California for my doctorate degree in
computational biology and bioinformatics which is an interdisciplinary program that spans my
topics and scientific fields like computer science, engineering, mathematics, statistics, and
medicine. This great combination of scientific fields allowed me to explore and enjoy many ideas
and work on the boundaries of all these scientific fields. My PhD journey was very enjoyable,
intellectually stimulating, and self-satiating. My PhD degree and journey won’t be possible
without the help and support of many people.
I am forever grateful and thankful to my advisor professor Arthur Toga. He is a typical
example of a great scientist and a perfect successful person. His scientific knowledge, experience,
intelligence, wisdom, and integrity greatly enhanced my experience while I was working on my
research. Professor Toga supported me in all possible ways to achieve my research endeavors. He
iv
provided the perfect scientific environment with all possible resources to stimulate my intellectual
abilities and make my scientific ideas come to fruitful conclusions. He believed in my intellectual
abilities and supported me to learn and advance in my research. He allowed me to pursue my own
ideas and research interests without limitations. I will never forget his very brilliant and intelligent
mind, his great character and personality, and his integrity. I was always fascinated by the human
brain and its abilities, and professor Toga and his lab allowed me to explore this amazing human
magical body part which is the most important organ in our body.
I am thankful to my parents and siblings who were always supportive in all ways. They
supported me to achieve and pursue my educational ambitions. They were always the first people
to provide me with emotional support when I needed it through my stressful challenging times.
My parents planted in me the seeds for love and passion to learn and to never give up achieving
my goals. They told me and taught me how to enjoy science.
I am thankful to all professors who taught me throughout all these years and believed in
my abilities and skills. I am thankful to my dissertation committee members, professors Fengzhu
Sun, Yonggang Shi, Ryan Cabeen, and Aiichiro Nakano for their encouragement and helpful
comments. I am also thankful to my lab mates in my imaging genetics group which is part of the
lab of neuroimaging (LONI) for their helpful scientific comments and discussions especially Dr.
Lu Zhao and Dr. Will Matloff. I am thankful to all my classmates and PhD doctoral candidates
who made my PhD years full of excitement and joy through many scientific discussions inside and
outside the classroom. They encouraged me and increased my confidence to achieve great things
through their continuous support and acknowledgement of my intellectual abilities and scientific
knowledge.
v
I am also very thankful to the Quantitative and Computational Biology department and
USC for providing a great program and providing all possible resources. I am thankful to my
department for the Viterbi fellowship they provided the first 2 years of my PhD and allowing me
to take a broad range of courses that helped my research and satisfied my scientific interests. I
have taken courses in electrical engineering, mathematics, computer science, statistics, and biology
which were very helpful to my research. I also want to thank Ashley Tozzi, the graduate services
advisor, for her endless help. I am very glad that I joined USC for my PhD degree. My experience
at USC could not be any better. USC is committed to excellence, and it is a great scientific
environment for satisfying the human thirst for knowledge and every step up the ladder of scientific
discoveries and technological advances. I hope my future will continue to be full of more scientific
challenges and ideas to tackle and explore.

vi
Table of Contents
Dedication ....................................................................................................................................... ii
Acknowledgements ........................................................................................................................ iii
List of Tables ............................................................................................................................... viii
List of Figures ................................................................................................................................ ix
Abbreviations .................................................................................................................................. x
Abstract .......................................................................................................................................... xi
Chapter 1: Introduction and Background ........................................................................................ 1
1.1 Alzheimer’s Disease ............................................................................................................. 1
1.2 Neuropathological Changes in AD ....................................................................................... 7
Chapter 2: Dysregulated Genes in Late-Onset AD ....................................................................... 12
2.1 Background and Study Purpose .......................................................................................... 12
2.2 Materials and Methods ........................................................................................................ 14
2.2.1 Data and Preprocessing ................................................................................................ 14
2.2.2 Linear Mixed-Effects Model (LMM) for Differential Gene Expression Analysis ...... 16
2.2.3 Gene Ontology and Gene Set Enrichment Analyses ................................................... 18
2.2.4 AD ML Binary Classification ...................................................................................... 19
2.3 Results ................................................................................................................................. 20
2.3.1 Differentially Expressed Genes (DEGs) ...................................................................... 20
2.3.2 Enrichment Analyses Results ...................................................................................... 25
2.3.3 AD Classification Results ............................................................................................ 29
2.4 Discussion ........................................................................................................................... 31
Chapter 3: Ethno-racial Disparities in Brain Biomarkers for Late-Onset AD .............................. 39
3.1 Background and Study Purpose .......................................................................................... 39
3.2 Materials and Methods ........................................................................................................ 41
3.2.1 Data and Participants ................................................................................................... 41
3.2.2 MRI Derived Brain Measures ...................................................................................... 43
3.2.3 Statistical Analysis ....................................................................................................... 44
3.3 Results ................................................................................................................................. 45
3.3.1 Ethno-Racial Differences in Brain Measures .............................................................. 45
3.3.2 The Changes of Brain Measures with Age and Race .................................................. 48
3.3.3 The Association of Race and Sex ................................................................................ 53
3.3.4 The Association of Race and APOE ............................................................................ 55
3.4 Discussion ........................................................................................................................... 57
Chapter 4: The Effects of Environmental Noise Exposure on Human Health and Brain ............. 61
4.1 Background and Study Purpose .......................................................................................... 61
vii
4.2 Material and Methods ......................................................................................................... 63
4.2.1 Study Population .......................................................................................................... 63
4.2.2 Assessment of Exposure to Noise and Air Pollution ................................................... 64
4.2.3 Brain Imaging Data and Processing ............................................................................. 65
4.2.4 Statistical Analyses ...................................................................................................... 66
4.3 Results ................................................................................................................................. 70
4.3.1 Neuroimaging Analysis ............................................................................................... 70
4.3.2 Cognitive and Hearing Scores ..................................................................................... 72
4.3.3 Association of Noise Exposure with Non-Imaging Phenotypes .................................. 73
4.4 Discussion ........................................................................................................................... 79
Chapter 5: Summary and Future Directions ................................................................................. 86
Papers and Abstracts ..................................................................................................................... 91
References ..................................................................................................................................... 92

viii

List of Tables

Table 2. 1 Summary of demographics and characteristics for subjects with and without AD ..... 15
Table 2. 2 Top 10 differentially expressed genes between disease and control samples ............. 22
Table 2. 3 GO annotations using the top genes and the enriched pathways ................................. 27

Table 3. 1 Summary of demographics and characteristics for AD subjects ................................. 42
Table 3. 2 Statistical differences in brain volumetric measures for AD participants ................... 47
Table 3. 3 Interaction effect between Race and Age, Sex, and APOE allele 4 status .................. 52

Table 4. 1 Summary of demographics and characteristics for subjects in G55 and G70 groups . 69
Table 4. 2 Statistical analysis on 11 selected brain volume measures between G55 and G70
groups ............................................................................................................................................ 71
Table 4. 3 Statistical Analysis on the cognitive and hearing scores between the two groups ...... 72
Table 4. 4 Significant associations of health-related phenotypes with noise exposure using model
1..................................................................................................................................................... 75
Table 4. 5 Significant associations of health-related phenotypes with noise exposure using model
2..................................................................................................................................................... 76
Table 4. 6 Significant associations of health-related phenotypes with noise exposure using model
3..................................................................................................................................................... 77

ix

List of Figures

Figure 1. 1: The hypothetical model of temporal changes in AD biomarkers ................................ 9
Figure 1. 2 : Brain atrophy in AD subjects ................................................................................... 11
Figure 1. 3: Comparison of MR images between AD and normal subjects ................................. 11

Figure 2. 1: The volcano plot of Log2 fold change versus B-statistic for all genes. .................... 23
Figure 2. 2: Log2 fold change comparison for the top ten genes in all 4 brain regions. .............. 24
Figure 2. 3: Network graph of the AD upregulated pathways ...................................................... 28
Figure 2. 4: Supervised ML classification using the top ten genes .............................................. 30

Figure 3. 1 The plots of the change of 6 brain measures with age and race ................................. 49
Figure 3. 2: The change of the white and gray matter volumes with age and race ....................... 50
Figure 3. 3: The change of the total brain volume with age and race ........................................... 51
Figure 3. 4: The effects of race and sex ........................................................................................ 54
Figure 3. 5: The effects of race and APOE ................................................................................... 56

Figure 4. 1: The relationship between noise and air pollution in UK Biobank data ..................... 78

x

Abbreviations

AD Alzheimer’s disease
LOAD Late onset Alzheimer’s disease
FAD Familial AD
MCI Mild cognitive impairment
MRI Magnetic resonance imaging
APOE Apolipoprotein E (gene)
AB Beta-Amyloid
NFT Neurofibrillary tangles
LMM Linear mixed-effects model
DEG Differentially Expressed Gene
NHW Non-Hispanic White
AA African American
HIS Hispanic
CDR Clinical Dementia Rating
TIV Total Intracranial Volume
MMSE Mini-Mental State Exam
MoCA Montreal Cognitive Assessment
xi

Abstract

Alzheimer’s disease (AD) is a progressive neurodegenerative disease that affects many
people and causes detrimental cognitive impairments and functional disabilities. Late onset AD
(LOAD) is the most common form of AD which appears after the age of 65 years old. It is a
multifactorial disease with mixed etiologies that are not fully understood. Aging is the highest risk
factor for developing AD besides many other causes including genetics, lifestyle factors, and
environmental pollution. Familial AD (FAD) is another rare form of AD that appears in early ages.
Genetics is the main cause of FAD. AD is considered the sixth leading cause of death in the United
States and the fifth leading cause of death in Americans who are older than 65 years. AD and
dementia in general are becoming a major health concern worldwide which stems the importance
of understanding AD especially with the continuing increase of the aging population. Normal
aging is not typically associated with dementia and AD symptoms, but the risk of developing AD
increases with aging. In this research work, we attempted to further understand AD and elucidate
its progression, development, and biological implications. Gene expression analysis was
performed using gene expression data from multiple brain regions to find dysregulated genes and
understand their biological implications and pathways. Ethno-racial disparities in brain biomarkers
were also studied for LOAD using a non-mixed AD subjects from three races: Non-Hispanic
White, African American, and Hispanic AD subjects. We also studied the effects of environmental
noise exposure on human health and brain using a large prospective cohort from the UK Biobank
project.
1

Chapter 1: Introduction and Background

1.1 Alzheimer’s Disease
Alzheimer’s disease (AD) is a progressive neurodegenerative brain disorder which is a
severe detriment to functional and cognitive abilities. The symptomatic cognitive decline and
impairments usually appear at the age of 65 or older. However, this age-related sporadic Late-
Onset of AD (LOAD) is not the only form. Early-Onset familial Alzheimer’s disease (FAD) is a
rare form of AD with symptoms developing in people in early ages that can occur in their 30s or
40s. Normal aging is not typically associated with dementia and AD symptoms, but the risk of
developing AD increases with aging. The number of deaths from Alzheimer’s dementia increased
by 89% between 2000 and 2014. AD is considered the sixth leading cause of death in the United
States and the fifth leading cause of death in Americans who are older than 65 years [1,2]. The
current estimate of AD patients in the US is around 6 million, and this number of Americans living
with AD is projected to increase to 13.8 million by 2050 which prioritizes the importance of
understanding the etiology of AD in its early stages before the onset of symptoms. The current
estimate of AD patients around the world is around 50 million, and it is expected to increase to
150 million by mid-century. The early symptom of AD is characterized by memory loss and
forgetfulness. However, the advanced stages of AD show very severe symptoms besides cognitive
impairments that are detrimental to person’s life in every way. Advanced AD patients may have
difficulty in speaking, learning, processing information, and executive functioning. They are
usually disoriented in the sense that they don’t have a clear awareness of themselves, time, and
place. They can’t recognize family members. Some AD patients may have difficulty eating or
2
swallowing, and they can even have difficulty moving. Even the activities that they used to do
daily, they won’t be able to do it or forget how to do it. These are only some major symptoms of
AD out of many adverse health effects.
The pathogenesis of Alzheimer’s disease is complex and not fully understood with causes
that include lifestyle and environmental factors besides genetics. Although it is more difficult to
characterize and validate the effects of non-genetic factors in the progression of AD, their whole
contribution as an AD risk is almost definite [3,4]. The genetic studies on the etiology of AD have
been successful in identifying genes with universal acceptance to their contributions in AD
progression. The early-onset FAD is believed to be passed entirely through genetics and caused
by mutations in genes like Presenilin 1 (PSEN1), Presenilin 2 (PSEN2), and Amyloid-β precursor
protein (AβPP) that are mainly involved in the formation of amyloid-β proteins[5–8].
Apolipoprotein E (APOE) allele 4 is also established as a high genetic risk for sporadic late-onset
AD [7,9,10] besides being reported as a risk factor for cardiovascular diseases [11].
Most AD studies have focused on the Non-Hispanic White (NHW) subjects. The
composition of the older population in the United States is expected to change ethnically with a
massive increase in the proportion that is racially other than NHW [12]. The US Census Bureau
reported in 2019 that minority ethno-racial populations with Hispanic (HIS) and African American
(AA) origins represented 18.5% and 13.4% of the total population respectively, and these
percentages are expected to increase [12,13]. This demographic change is not reflected in most
AD research studies, and minority groups are usually underrepresented in AD cohorts which are
mostly comprised of NHW participants. For example, the National Alzheimer’s Coordinating
Center (NACC), which maintains and coordinates the database that consists of data from ADRCs,
3
had dementia data of which AA and HIS subjects represented only 9% and 8% respectively
according to their June 2021 data freeze [14].
Substantial differences were reported in prevalence and incidence rates for AD among
different ethno-racial groups. It is often estimated that there are higher rates of dementia among
ethnic minorities [15,16]. The ethno-racial disparities in the risk of developing AD are attributed
to many factors and stressors besides genetics. Socioeconomic factors such as educational
attainment, occupational characteristics, and financial resources may influence the manifestation
of the disease in different ethno-racial groups [17,18]. Cultural differences, access to health care,
and exposure to environmental toxins may also influence the risk of AD [19–21]. Racial
differences in cardiovascular diseases may also explain the disparities in AD risk. Most
cardiovascular disease factors, that are associated with and contribute to the cognitive decline and
AD progression, are more prevalent in AA and Hispanics than NHW [22–24].
Alzheimer’s disease is a multi-factorial disease with multiple causes that are not fully
understood including environmental pollution and toxins. Environmental pollution has recently
become a focus of the scientific efforts for its role as a risk factor for many adverse health effects
including cardiovascular diseases and dementia. Several medical research studies on air and noise
pollution have found direct association of environmental pollution with increased risk of dementia
and cardiovascular diseases [25–27]. The association of air and noise pollution with cognitive
decline and dementia is often attributed to the cardiovascular risk factors and diseases as indirect
mediators [23,28–30]. However, some studies on air pollution have shown that air pollutants and
ultrafine particulate matter can reach the brain and cause neuroinflammation leading to
neurodegeneration and AD [31]. Animal studies have shown that ultrafine air pollutants can reach
the brain through the olfactory neuronal pathway and act as neurotoxins and cause
4
neuroinflammation which may lead to dementia and cognitive decline [32,33]. Many studies on
neuroimaging using magnetic resonance imaging scans have reported an association of increased
air pollution exposure with brain atrophy and smaller brain structures. An increased exposure to
PM2.5, which is ultrafine air pollution particles with aerodynamic diameter less than 2.5 µm, was
associated with lower brain structure, and residential living with lower air pollution has been
associated with lower neurodegeneration and brain atrophy [34–37].
Community noise is another environmental exposure risk factor which is receiving a recent
growing attention besides air pollution. Noise could cause adverse health effects, and it is
considered as a hazard to human health whether it is transportation noise or from industry and
construction [27]. Unlike air pollution, studies on noise pollution and its effects and relation to
dementia and human brain are very scarce. Research studies have linked noise exposure to
cardiovascular risk factors and diseases. Noise exposure was associated with hypertension,
obesity, and increased heart rate [38,39]. Many research studies on noise pollution have reported
cognitive decline and higher incidence and prevalence rates of dementia and Alzheimer’s
disease[40–42]. Neuroimaging studies on noise exposure and dementia in humans are very scarce
and mainly limited to animal experiments. Long term exposure to high level of noise in animal
studies has indicated changes in brain regions that are mostly affected by Alzheimer’s disease [43].
The motivation behind this work is to further understand Alzheimer’s disease and its
progression considering different biological processes and aspects which include transcriptional
genetic changes in AD which may precede the onset of symptoms, ethno-racial disparities, and
possible etiologies and causes like environmental pollution especially environmental noise
exposure. Alzheimer’s disease is a leading cause of mortality and disability among the elderly
population. AD and dementia in general are becoming a major health concern worldwide which
5
stems the importance of understanding AD especially with the continuing increase of the aging
population. In chapter one, I give an introduction about AD and its neuropathological changes that
serves as a background and motivation for this work. Chapter two addresses the transcriptional
genetic changes and dysregulated genes between AD and healthy control subjects by integrating
gene expression data from multiple brain regions for the same subject which has never been done
before. Most studies have limited their gene expression analysis to a specific brain region. The
dysregulated genes found in our study represented AD specific transcriptional changes in brain
regions that represent the cognitive system affected by AD. These changes may even precede any
onset of AD symptoms.
Chapter three discusses ethno-racial disparities for LOAD specifically disparities in brain
biomarkers using non-mixed AD subjects from three major ethnic groups in the US: Non-Hispanic
white, Hispanic, and African American AD subjects. While studying gene expression differences,
I learned that APOE allele 4 is a high-risk factor for AD, but it affects Non-Hispanic White subjects
more than any other race. This led me to get interested in studying whether there are any brain
disparities in the biological expression of AD in different races. Most AD studies on ethno-racial
disparities have focused on the incidence and prevalence rates. Some studies have also focused on
molecular biomarkers and the hippocampal region without using non-mixed AD subjects. We used
non-mixed AD participants to study most brain regions that can be affected by AD by eliminating
many variables and factors that may contribute to cognitive decline besides AD.
Chapter four addresses the effects of environmental noise exposure on human health and
brain. Alzheimer’s disease is a multi-factorial disease with mixed etiologies. Environmental toxins
and pollution are believed to be a possible etiology and cause for dementia and AD. The studies
of environmental pollution have considered air pollution effects, but fewer studies have considered
6
noise pollution. Neuroimaging studies on the effects of noise exposure on human brain are very
scarce, and neuroimaging studies were mainly limited to animal studies. We focused on studying
the effects of noise exposure on human brain using a large-scale UK Biobank data by considering
most brain regions. We also studied the effects of noise on a large-scale of health-related
phenotypes as an attempt to understand the effects of noise on human health and its relation to
dementia and Alzheimer’s disease.
Chapter five provides conclusions on what has been identified while analyzing the data
throughout this research work. It also provides some possible future directions based on the results
and identified conclusions.

7

1.2 Neuropathological Changes in AD

The main molecular biomarkers that constitute the neuropathological hallmarks of AD are
the aggregation and deposition of Beta-Amyloid (AB) peptides, tau, and phosphorylated tau in a
toxic process that leads to the neurodegeneration in the brain. The normal function of Beta-
Amyloid in the brain is that it plays an essential physiological role in neural growth and repair
[44]. However, it is believed that mutations in some genes like Presenilin 1 (PSEN1), Presenilin 2
(PSEN2), and Amyloid-β precursor protein (AβPP) that are mainly involved in the formation of
amyloid-β proteins[5–8] contribute to the deposition and aggregation of the AB neuritic plaques
especially in the case of early onset FAD. Apolipoprotein E (APOE) allele 4 is also established as
a high genetic risk for sporadic late-onset AD [7,9,10] which also contributes to the accumulation
of AB. In LOAD, it is believed that the imbalance in the production, degradation, and clearance of
the Amyloid peptides can contribute to the accumulation of the plaques in the brain[6,8,44,45] .
There are two major isoforms of AB: AB42 and AB40, and the only difference is that AB42 has
two extra residues at the C-terminus. The Amyloid neuritic plaques in AD consist mostly of AB42.
Both isoforms maybe present in AB plaques, but it seems that there is a preferential deposition of
AB42 at the extracellular space in AD[46]. The aggregation and the deposition of the Amyloid
plaques occur in the extracellular space between neuronal cells. When the plaques become a
massive aggregation, they start to form an obstacle at the synaptic clefts and hinder the
communications between neuronal cells after damaging the synapse’s structural integrity. When
the neuronal cell can’t communicate with its adjacent or neighboring neuronal cells, it will die.
That is what causes neurodegeneration and atrophy in AD brains [47,48].
8
Neurofibrillary tangles (NFTs) are also related to AD progression caused by abnormal tau
proteins. The main constituent of NFTs is tau protein which is encoded by the gene Microtubule
associated protein tau (MAPT). Abnormal tau phosphorylation is believed to be caused by
oxidative stress and iron [49]. NFTs are abnormal aggregation of tau proteins within the neuronal
cell. Tau proteins have a healthy functional role in the neurons that support the internal microtubule
structures. These microtubules play an important role in neuronal cell growth and transport of
nutrients in the cell body. However, abnormal accumulation of the NFTs within the neuron in
Alzheimer’s disease can harm the synaptic integrity and cause neuronal cell loss and
neurodegeneration. NFTs initially affect the entorhinal cortex and then spread to the neocortex and
hippocampus of the brain [50]. It is hypothesized that the neurofibrillary tangles of AD brains are
most likely formed and deposited after the initial formation of amyloid plaques [45,51] which can
induce the formation of NFTs. However, this hypothesis is still under scrutiny because AB and tau
accumulations might be initiated independently in sporadic AD. There was evidence from autopsy
data that suggested that AD tau pathology may precede AB deposition [52,53]. Staging of the
severity of AD using the aggregation of neurofibrillary tangles was mainly described by Braak
staging in 1991 [50]. Braak staging consists of 6 stages. The first two stages (I/II) are used when
the NFTs are mainly confined to the entorhinal region. Stages (III/IV) are used when the NFTs
spread to the limbic regions of the brain mainly the hippocampus, and the last two stages are used
when the NFTs spread to the rest of the brain. It is believed that the accumulations of the plaques
and NFTs precede the onset of AD symptoms in many years which can be around 10 years or
more. Jack et al. suggested a hypothetical model that described the temporal evolution of AD
biomarkers in relation to each other and to the progression and onset of AD and its clinical
symptoms [52]. The model was built on evidence presented from many research studies available
9
at that time, but they acknowledged that their assumptions may need more empirical assessment.
In their hypothetical model, they showed that Amyloid biomarkers become abnormal first, and
then it is followed by abnormality in tau and imaging biomarkers leading to neurodegeneration
and clinical symptoms. Figure 1.1 shows their hypothetical model and how the biomarkers exhibit
early acceleration followed by deceleration in a sigmoid shaped trajectory reaching a plateau.

Figure 1. 1: The hypothetical model of temporal changes in AD biomarkers
The hypothetical model of the temporal ordering of the dynamic biomarker
abnormalities involved in Alzheimer’s disease pathophysiological pathway. It shows
how Beta Amyloid biomarker abnormality precedes any other AD biomarker. This
model is adapted from Jack et al. (2010)
10

Neuroimaging studies on AD using magnetic resonance imaging have shown that
neurodegeneration and brain atrophy start in the hippocampal and entorhinal regions and then the
rest of the brain. The early symptom of AD is memory loss because AD affects the hippocampus
which is part of the limbic system and embedded in the medial temporal lobe. The hippocampus
plays an essential role in information processing and memory from short term to long term
memory. AD also initially affects the entorhinal region of the brain which is also located in the
medial temporal lobe. The entorhinal region plays an important role in memory and orientation.
AD patients have been reported to be disoriented in the sense that they don’t have awareness and
perception of themselves, time, and place. Figure 1.2 shows a coronal view of the differences
between a normal healthy control and a patient with advanced AD. We can notice that brain
atrophy and shrinkage affect the whole brain. Volume decrease is global and regional in AD brain
where we can see the atrophy in the hippocampal region as well. Figure 1.3 shows a comparison
of the differences between three groups: normal, MCI, and AD subjects using an axial view of T1
weighted MR images. We can notice the global and regional gradual brain shrinkage and volume
decrease between the three groups. The AD subjects show volume decrease and enlarged ventricles
as the result of AD neurodegeneration. AB and tau deposition can also be assessed via imaging
using positron emission tomography (PET) using PiB (Pittsburgh compound B) and FDG-PET
which can show the accumulation of AB plaques and NFTs in AD patients compared to healthy
controls.

11

Figure 1. 2 : Brain atrophy in AD subjects
This figure shows the brain atrophy and shrinkage of AD patients compared to healthy
subjects. There are regional and global brain atrophies and volume decrease especially in
the hippocampus. This figure is adapted from Alzheimer’s Association 2008.

Figure 1. 3: Comparison of MR images between AD and normal subjects
An axial view of MR images comparing AD subjects with MCI and Normal subjects. This
figure is adapted from Chandra et al. Journal of Neurology. 2019
12
Chapter 2: Dysregulated Genes in Late-Onset AD

2.1 Background and Study Purpose

Most previous research studies on AD genetics have used Genome wide association
studies (GWAS) techniques besides genetic linkage and gene expression studies. Gene expression
data has been used to detect AD biomarkers and investigate the biological processes and molecular
functions that are involved in AD progression. Gene expression profiling allows genome wide
measurements of transcriptomic data. This type of data analysis can give insight into relating AD
and its clinical symptoms with gene interactions that play an essential role in AD development.
Recent advances in the scale of gene expression data and genomics have allowed the collection of
genomic data from multiple tissues and brain regions of the same individual [54]. Since AD
progression is sequential and affects many brain regions, it is crucial to integrate gene expression
information from multiple brain regions that are part of the cognitive system in order to determine
the degree to which the cognitive system is affected [52]. Several research studies found
dysregulated levels of gene expression in both gray and white matters involved in AD [55,56]. In
this study, new dysregulated genes as AD biomarker candidates were identified using a linear
mixed model for differential expression analysis with repeated measures by integrating gene
expression data from multiple brain regions. This project analysis was not necessarily driven by a
certain hypothesis, and it was mainly focused to find dysregulated genes in AD. The dysregulated
genes can be defined as the AD associated genes which are differentially expressed and reflect
statistically significant transcriptional changes in the brain of AD subjects compared to healthy
controls. These AD associated genes and their underlying pathways can help further understand
the pathogenesis and progression of the disease. The identified genes definitely contribute to AD
13
progression and development and may open ways to explore new effective treatments at early
stages before the onset of clinical symptoms especially that AD has limited dementia at early stages
which can result in misleading diagnosis of the disease. The data used in this analysis was
previously published in Miller et al. [57]. However, the authors did not consider the integration of
the gene expression data from all four brain regions and their aggregate results. The novelty of the
study lies in the integration of multiple brain regions per subject in the analysis using linear mixed
model to resolve the correlation structure in contrary to previous research studies which mainly
focused on specific brain regions. The novelty of the study also lies in the new differentially
expressed genes that were found and not previously reported as AD associated genes. The main
question that motivated this analysis is finding genes with differential expression and
transcriptional changes that occur between healthy control and AD elderly patients especially that
gene expression changes can occur before the onset of AD symptoms. The differentially expressed
genes are not necessarily AD biomarkers or tissue specific genes. They are considered AD specific
or associated genes which are affected in AD using and integrating more than one brain region
which are part of the cognitive system affected by AD. We used linear mixed effects model to
combine the data to find differentially expressed genes. We also used gene ontology and gene set
enrichment analyses to understand the biological functions and processes behind those genes.
These reported genes can help understand AD development and progression and open new ways
for therapeutic treatments possibly before the onset of symptoms.

14
2.2 Materials and Methods

2.2.1 Data and Preprocessing

RNA sequencing data was obtained from the Aging, Dementia and Traumatic Brain Injury
(TBI) Study via the Allen Brain Institute (http://aging.brain-map.org). This dataset, derived from
a subset of the Adult Changes in Thought (ACT) cohort, consists of neuropathologic, molecular,
sand transcriptomic data for the postmortem brains of 55 TBI and 52 matched control subjects
[57]. We focused on the control group without previous TBI. Among the control group, 15 subjects
had diagnosed AD, 7 subjects had dementia from other causes, and 30 had no dementia. We used
these 15 AD and 30 control subjects to investigate the differential gene expression in AD. All
subjects have signed a consent form in order to participate in the study. The demographics of these
45 subjects used in the analysis are shown in Table 2.1. All 15 AD subjects were considered late-
onset AD cases with mean age of death of 89.6 years. For each subject, gene expression data was
available from four different brain regions: the hippocampus (HIP), the temporal cortex (TCx), the
parietal cortex (PCx), and the forebrain white matter (FWM). RNA sequencing data was missing
in some regions for some subjects yielding a total number of 165 samples with 111 samples in the
control group and 54 samples in the AD group.
Additionally, the data was analyzed using only 3 brain regions (HIP + TCx + PCx) with a
total number of 124 samples (84 control samples and 40 AD samples). The data was also analyzed
using only 2 regions (HIP + TCx) with a total number of 83 samples (57 control samples and 26
AD samples). The percentages of AD subjects and control subjects that have APOE e4 allele were
33.33% and 10% respectively. Advanced staging of the subjects was not available other than the
Braak stages. Table 2.1 shows the mean of Braak stages for control and AD subjects. The mean
Braak stage for AD subjects was around 4 with 6 subjects out of 15 had a Braak stage less than or
15
equal to 4. The mean Braak stage for control subjects was around 3 with 28 subjects out of 30 had
a Braak stage less than or equal to 4. The RNA sequencing data, which consisted of expression
data for 50,281 genes, was normalized into FPKM values across all samples to account for
processing batch and RNA quality. All 50,281 genes were used in the analysis without filtering in
order to get a wide range of genomic transcriptional candidates that can be associated with AD.

Table 2. 1 Summary of demographics and characteristics for subjects with and without AD
Control Subjects Alzheimer’s Subjects
Category Mean (SD) Mean (SD)
Age at death (yrs) 87.13 (5.62) 89.60 (6.21)
Education (yrs) 14.70 (3.13) 13.67 (2.76)
Braak stage 2.67 (1.24) 4.13 (1.84)
CERAD score 1.16 (0.94) 1.80 (1.32)
Count Count
Sex 14 F/ 16 M 6 F/ 9 M
APOE e4 alleles 3 Yes/ 27 No (10 %) 5 Yes/ 10 No (33.33%)
Abbreviations: APOE, Apolipoprotein; SD, standard deviation; F, female; M, male;
CERAD, Consortium to Establish a Registry for Alzheimer’s Disease; yrs, years.
The characteristics of the subjects that were used in the analysis are reported. All AD
subjects in our analysis were considered late-onset AD cases. There were 45 subjects in total
with 30 control and 15 AD subjects. The total number of samples was 165 samples with 111
control samples and 54 AD samples. The percentages of control and AD subjects having
APOE e4 alleles were 10 % and 33.33% respectively.

16
2.2.2 Linear Mixed-Effects Model (LMM) for Differential Gene Expression Analysis

Multiple samples were collected from different brain regions of the same subject which
suggested a repeated measure design for differential gene expression analysis because samples
from the same subject are correlated. This violates the common assumption for statistical tests that
the samples should be statistically independent. Ignoring the correlation between samples in a
repeated measure design will not yield consistent and effective estimates and will not control type
I error rate. Linear mixed models allow using both fixed and random effects that deal with the non-
independence arising from a hierarchical multilevel structure in the data. In our analysis,
individuals were taken as random effects in fitting the linear mixed model. The Limma package
[58–60] was used which is an open source R package available through the Bioconductor project.
Limma entails many methods that can handle complex experimental designs and overcome the
small sample size problem by borrowing information between genes using an empirical Bayes
approach [61] and resulting in moderated t-statistic.
The Voom method [62] was used that generates precision weights for each observation to
account for precision variations between different observations. It estimates a mean-variance trend
using locally weighted regression (LOWESS), and it predicts the variance of each observation.
The Voom precision weights are the inverse of the predicted observation variance. Squeezing the
gene wise variances to the common trended variance will reduce the false positive rate for genes
with small variances and improve the detection power for differentially expressed genes with
larger variances. The linear mixed model was estimated using the lmer function in the lme4
package [63]. The null hypothesis in LMM usually lies in whether one or more of the regression
coefficients as contrast estimators are equal to zero. In this analysis, our interest lies solely in the
contrast estimator that corresponds to the disease status to analyze the differential expression
17
between Alzheimer’s disease and healthy control samples. The gene expression data was first
adjusted for age and sex, and then a logarithmic transformation was performed on the data to
counteract the unequal variability between large and small values.

18
2.2.3 Gene Ontology and Gene Set Enrichment Analyses
The statistically significant differentially expressed genes (DEGs) with the smallest P-
values obtained by using LMM were used with gene ontology enrichment analysis to identify the
enriched dysfunctional biological implications associated with these genes. Gene Ontology tool
powered by Panther was used in this analysis [64,65]. The enriched biological processes,
molecular functions, and cellular components associated with the top ten genes and all
differentially expressed genes were identified using Bonferroni correction for multiple testing with
a P-value threshold of .05. In addition to Panther, the Database for Annotation, Visualization, and
Integrated Discovery [66,67] was used to identify some enriched biological themes and diseases
associated with the provided list of the top ten genes.
Gene set enrichment analysis was performed to find the significantly enriched co-regulated
gene sets or pathways. Each pathway represents a certain biological or molecular function of
interest. The gene sets are usually defined from external sources like Gene Ontology database [64]
or from previously established research studies. In this analysis, Broad institute GSEA [68] was
used with the Reactome gene sets derived from the Reactome pathway database [69]. The 50,281
genes were pre-ranked using the moderated t-statistic derived from the linear mixed model for
differential expression. All 50,281 pre-ranked genes were then used with GSEA to find the
significantly enriched pathways in both control and AD groups using a threshold P-value of .01
and false discovery rate (FDR) of .02.

19
2.2.4 AD ML Binary Classification
Supervised machine learning (ML) algorithms were used to identify the discriminative
power of the top ten differentially expressed genes in distinguishing between the two categories.
Support vector machines [70] using both linear and radial basis function kernels were selected for
the binary classification in addition to random forest and quadratic Bayes algorithms. The data was
divided into a training set and a test set. The training set consisted of approximately 70% of the
samples (114 samples), and the remaining 30% were used for testing to test and approximate how
the classifiers generalize to unknown data. There was no overlap in the subjects and samples
between the training and testing data to avoid any correlation between samples in both sets. The
genes were used as features for each sample with all different combinations of N genes (2£N£10)
out of 10 genes. For every combination of N genes, the combination with the maximum accuracy
on the training data was chosen, and the corresponding accuracy on the test data was reported on
those N genes to avoid data snooping. Precision and recall scores in addition to the F1 score were
reported for the testing accuracy that corresponded to the maximum training accuracy over all
combinations. The limitation in this classification analysis is that we violated the assumption of
IID in machine learning algorithms and ignored the inter-sample correlations. However, this
implicit assumption of independence is often violated in machine learning algorithms using many
datasets that reflect real life applications especially biological data. Machine learning algorithms
may still learn from non-independent data and give good accuracies. We mainly used these
samples to assess the discriminative power of those genes to differentiate between AD and control
samples.

20

2.3 Results

2.3.1 Differentially Expressed Genes (DEGs)

Using the linear mixed-effects model, the reported P-value associated with each gene was
used to assess its differential expression between Alzheimer’s disease and control groups. The
final P-values were adjusted for multiple comparison by Bonferroni correction. There were 602
genes out of 50,281 genes that showed significance at a 5% significance level using all four brain
regions in the analysis. The results for the top ten genes are shown in table 2.2 with columns
representing log2 fold change (logFC), average log2 expression (AvExp), moderated t-statistic,
corrected P-value, and B-statistic for each gene. The B-statistic is the posterior log odds of
differential expression derived in [61]. Log fold change was defined in our analysis as AD to
control ratio. A positive value indicated that the gene was upregulated in AD. Table 2.2 also gives
more information about the size and location of the top 10 genes in the genome. The gene category
can be protein coding, RNA gene (non-protein coding), and pseudogene. Among the 602 DEGs,
490 genes were protein coding genes, 74 genes were RNA genes, and 29 genes were pseudogenes.
The volcano plot in figure 2.1 shows how the log2 fold change is changing with the posterior log
odds for differential expression. The B-statistic for each gene generally increased with increased
significance in differential gene expression. The top 10 differentially expressed genes showed a
logFC > 0.5 and B-statistic > 50. Neurofilament heavy polypeptide (NEFH) gene showed the most
significant differential expression between Alzheimer’s and control subjects.
Most DEGs, 484 genes out of 602 genes, were upregulated in AD. The top ten
differentially expressed genes were all upregulated in AD using all four brain regions in the
21
analysis. Figure 2.2 shows the comparison of LogFC for the top ten genes in the 4 distinct brain
regions. It was noticeable how the LogFC values were high and showed upregulation in FWM for
all top ten genes. These top DEGs were mostly dysregulated in the same direction in all four brain
regions except for three genes (SNAP25, NEFL, RGS4) that were downregulated in the
hippocampus. The gene RGS4 was also downregulated in the temporal cortex. The top ten genes
were searched using Agora platform (https://agora.ampadportal.org) which was initially developed
by NIA-funded AMP-AD consortium that shares evidence in support of AD target discovery. None
of the top ten genes were previously reported to have any RNA expression changes in AD brains,
and three genes (NEFL, TESPA1, and SNAP25) were reported as nominated AD targets according
to Agora. The database Alzgene (www.alzgene.org), which is a collection of published AD genetic
association studies, was also searched, and none of the genes were previously reported in any AD
research study according to Alzgene. Few genes out of the top ten DEGs, in particular NEFL [71],
NEFH [71], RGS4 [72], and SNAP25 [71,73], were previously found in research studies to exhibit
altered expression levels in various brain cell types in AD brains.
The data was also analyzed using (HIP + TCx + PCx) samples excluding FWM samples.
Using LMM, there were 146 genes that showed statistical significance in differential expression
after Bonferroni correction. There were 90 downregulated genes in AD out of 146 DEGs. There
were 90 common genes with the 602 differentially expressed genes using all four brain regions.
Fifty-three genes out of 90 common genes were found as downregulated in AD in (HIP + TCx +
PCx) analysis. The same 53 common genes were also found to be downregulated in the analysis
of four brain regions. Additionally, the data was analyzed using only (HIP + TCx) samples with
the accepted hypothesis that AD affects HIP first followed by TCx [74,75]. The differentially
expressed genes were only 93 genes. There were 61 downregulated genes in AD out of 93 DEGs.
22
The (HIP + TCX) analysis had a limited number of samples since there were only 83 samples in
total.

Table 2. 2 Top 10 differentially expressed genes between disease and control samples
Entrez ID Symbol LogFC AvExp t P-value B
4744 NEFH 0.653 4.387 14.34 5.12E-42 94.387
5816 PVALB 0.631 2.839 9.60 7.80E-39 86.539
9840 TESPA1 0.587 3.437 12.91 6.00E-34 75.185
100873748 RNU6-33P 0.574 1.014 12.62 1.86E-32 71.555
6616 SNAP25 0.560 8.954 12.30 8.17E-31 67.650
222008 VSTM2A 0.551 4.364 12.11 6.87E-30 65.397
9118 INA 0.544 4.665 11.95 3.89E-29 63.556
1123 CHN1 0.529 7.890 8.43 7.51E-27 58.283
4747 NEFL 0.538 6.488 8.42 5.15E-26 56.246
5999 RGS4 0.512 5.509 11.26 9.18E-25 55.646
Symbol Gene Name Gene Length (BP) Chromosome
NEFH Neurofilament, heavy polypeptide 11160 22
PVALB Parvalbumin 18795 22
TESPA1 Thymocyte expressed, positive selection associated1 36747 12
RNU6-33P RNA, U6 small nuclear 33, pseudogene 106 4
SNAP25 Synaptosomal-associated protein, 25kDa 88588 20
VSTM2A V-set and transmembrane domain containing 2A 28755 7
INA Internexin neuronal intermediate filament protein, alpha 13208 10
CHN1 Chimerin 1 206068 2
NEFL Neurofilament, light polypeptide 6155 8
RGS4 Regulator of G-protein signaling 4 8027 1
Abbreviations: LogFC, log2 fold change; AvExp, average log2 expression value; t,
moderated t-statistic; B, B-statistic which is the posterior log odds of differential expression;
BP, base pairs.
The top 10 differentially expressed genes between AD and control samples are reported in
this table. The table shows the statistical results that were obtained from linear mixed-effects
model on gene expression data for each gene. The gene length and location in the genome are
also shown.
23

Figure 2. 1: The volcano plot of Log2 fold change versus B-statistic for all genes.
Genes with P-value < .05 are shown as red circles. Otherwise, they are blue crosses. The
NEFH gene is marked. The NEFH gene was the most statistically significant differentially
expressed gene.

24

Figure 2. 2: Log2 fold change comparison for the top ten genes in all 4 brain regions.
All these genes were upregulated in AD using the four-brain region LMM analysis. All the
genes were mostly upregulated in all four brain regions except for SNAP25, NEFL, RGS4
that were downregulated in the Hippocampus. RGS4 was also downregulated in the
Temporal cortex.

25
2.3.2 Enrichment Analyses Results
The gene ontology analysis was focused on the top ten DEGs as AD biomarker candidates
with the most significant differential expression. Gene Ontology database powered by Panther
revealed the enriched annotations associated with those genes. The underlying enriched GO terms
in biological processes, cellular components, and molecular functions were mainly related to
neurofilament and postsynaptic cytoskeleton organization and structural constituent of synapses.
These GO annotations suggested that the genes are associated with dysfunctional structural brain
connectivity in Alzheimer’s disease. Table 2.3 summarizes the GO terms associated with the top
ten genes using a threshold of 0.05 for the Bonferroni corrected P-values. The Database for
Annotation, Visualization, and Integrated Discovery clustered the enriched biological themes
associated with each gene. All top genes showed that they are mainly expressed in brain tissues.
The diseases associated with these genes were shown to be mainly related to neurological and
psychological disorders, motor neuron diseases, and brain structural connectivity. Gene ontology
analysis was also performed on all 602 DEGs and not just the top 10 DEGs. The Gene ontology
annotations using all 602 genes gave a more comprehensive list of terms which completely agreed
with terms that were found using the top ten genes in table 2.3. Gene ontology analysis on (HIP +
TCx) DEGs did not give any annotations after correction. GO analysis on (HIP + TCx + PCx)
DEGs gave few annotations in which “Structural constituent of myelin sheath” was the only
annotation in molecular functions that survived after correction, and “Neurofilament” was the only
annotation in cellular components that survived after correction.
Broad institute GSEA was used to reveal the most significant enriched pathways associated
with the disease using the Reactome pathways database and using all genes. The genes revealed
pathways that were mainly associated with neuronal systems, axon guidance, and neurotransmitter
26
release cycle across chemical synapses. The enriched pathway results asserted and agreed with the
GO terms and verified that these genes are associated with Alzheimer’s disease. Forty gene sets
and forty-nine gene sets were upregulated and downregulated in Alzheimer’s disease respectively
with a cutoff P-value of .01 and cutoff FDR of .02. Table 2.3 shows the top enriched pathways
with the resulting enrichment score and gene set size mapped to the 50,281 genes. Cytoscape
(3.7.1) [76] was used to visualize the network graphs of the enriched pathways. The upregulated
and downregulated pathways were clustered separately where there were no connections between
the nodes representing the upregulated and downregulated pathways. The AD upregulated
pathways formed a connected uncomplete graph excluding the axon guidance pathway cluster.
Figure 2.3 shows the top highlighted AD upregulated pathways that are mentioned in table 2.3
with red edges showing their connections. Neuronal system was the most enriched pathway, and
it was connected with the transmission across chemical synapses pathway with an edge similarity
coefficient of 0.83 since they share 186 genes. The highest similarity coefficient of 0.86 was
between the transmission across chemical synapses pathway and neurotransmitter receptor binding
and transmission in postsynaptic cell pathway where they share 137 genes. The downregulated
pathways in phenotype AD were mainly associated with DNA synthesis, translation, and
lipoprotein metabolism.

27
Table 2. 3 GO annotations using the top genes and the enriched pathways
Biological Process GO Terms P-value
Postsynaptic intermediate filament cytoskeleton organization 9.56E-06
Neurofilament cytoskeleton organization 5.73E-05
Postsynaptic cytoskeleton organization 2.17E-04
Intermediate filament cytoskeleton organization 8.73E-03
Intermediate filament-based process 9.28E-03
Neurofilament bundle assembly 1.43E-02
Molecular Function GO Terms P-value
Structural constituent of postsynaptic intermediate filament cytoskeleton 3.00E-06
Structural constituent of synapse 1.99E-04
Structural constituent of cytoskeleton 2.52E-02
Cellular Component GO Terms P-value
Neurofilament 2.82E-05
Postsynaptic intermediate filament cytoskeleton 1.40E-03
Schaffer collateral – CA1 synapse 8.07E-03
Postsynaptic cytoskeleton 1.05E-02
Upregulated pathways in phenotype AD Gene Set Size Enrichment Score
Neuronal system 276 2.66
Transmission across chemical synapses 184 2.50
Voltage gated potassium channels 43 2.44
Potassium channels 98 2.43
Neurotransmitter receptor binding and transmission in postsynaptic cell 135 2.34
Neurotransmitter release cycle 34 2.11
Axon guidance 243 1.70
Downregulated pathways in phenotype AD Gene Set Size Enrichment Score
Translation 209 -2.60
SRP dependent cotranslational protein targeting to membrane 169 -2.58
Peptide chain elongation 145 -2.50
Metabolism of RNA 316 -2.26
DNA strand elongation 30 -1.93
Synthesis of DNA 91 -1.91
Metabolism of proteins 484 -1.90
Lipoprotein metabolism 28 -1.88
This table shows the enriched Gene ontology (GO) terms associated with the top
differentially expressed genes and their corresponding P-values. The GO powered by
Panther was used to derive these biological processes, molecular functions, and cellular
components associated with the top 10 genes. The enriched pathways were found using
Broad Institute GSEA tool using a threshold P-value < .01 and a threshold FDR < .02.
28

Figure 2. 3: Network graph of the AD upregulated pathways
A zoomed-in network graph on the AD nodes representing the upregulated pathways in AD
with P-value < .01 and FDR <.02. The yellow highlighted nodes are the most significantly
enriched upregulated pathways with red edges showing the connections between those
pathways. This is a zoomed-in graph from supplemental figure 2.3.

29
2.3.3 AD Classification Results
Four supervised ML algorithms were used for this binary classification in which the
random forest classifier gave the best accuracy on the test data. Figure 2.4 (A) shows the maximum
training accuracy on N genes (2£N£10) for all four classifiers. Figure 2.4 (B) shows the testing
accuracy that corresponded to the maximum training accuracy on the same N genes. Although
SVM with gaussian kernel gave the highest accuracy on the training data, random forest classifier
gave the best generalization accuracy on the test data with an accuracy of 83% although we
violated the independence assumption. This 83% test accuracy corresponded to the maximum
training accuracy on the same six genes. The recall and precision scores for this test accuracy were
0.97 and 0.81 respectively with F1-score of 0.88. The high (100%) accuracy of SVM with gaussian
kernel on the training data can be due to overfitting which resulted in a lower accuracy on the test
data. The out of sample generalization error has an upper bound £ 27% [17% (test error) + 10%]
with a probability ³ 0.95 where the 10% is the error bar estimate derived from Hoeffding inequality
using a tolerance d = 0.05.

30
A B

Figure 2. 4: Supervised ML classification using the top ten genes
(A) Classification results on the training data using four algorithms. The plots correspond to
the maximum training accuracy achieved on a combination of N genes where 2£N£ 10. (B)
Classification results on the testing data. The plots correspond to the testing accuracy on N
genes that achieved the highest accuracy on the training data.

31
2.4 Discussion
The analysis in this study aimed to find dysregulated genes between control and AD elderly
subjects. The statistically significant differentially expressed genes were AD specific genes that
reflect transcriptional changes associated with AD rather than a certain brain tissue. The significant
dysregulated genes were mostly related to dysfunctional connectivity between neuronal cells in
the brain. They were related to neurofilament organization, structural constituent of the synapses,
axon guidance, and neurotransmitter release cycle. All these GO and pathways enrichment results
suggested that the significant dysregulated genes were associated with dysfunctional brain
connectivity that affects the cognitive system in the human brain leading to AD. Gene expression
analysis can reveal the biological implications underlying certain diseases. The recent advances in
the scale of acquiring genomic data have opened many ways to analyze and explore complex
diseases requiring paralleled advances in computational tools and methods. Alzheimer’s disease is
a complex disease with etiologies that are not fully understood. In this study, gene expression data
was collected from multiple brain regions which are usually affected by Alzheimer’s disease.
It was verified in our analysis using Wilcoxon rank sum test that there were no
statistically significant differentially expressed genes between dementia and control samples
within the same brain region as it was mentioned in Miller et al. [57]. All four brain regions have
been combined in the analysis because AD will affect multiple brain regions, so it is crucial to
integrate gene expression information from brain regions which are part of the cognitive system
to understand how the cognitive system is affected by AD. Even if these brain regions have distinct
anatomical or functional patterns, they are all involved in the development and progression of AD.
The genetic and transcriptional AD changes occur concordantly in multiple brain regions that
affect the cognitive systems. FWM has distinct cellular content and structure, but several research
32
studies examined and observed similar deregulated levels of gene expression in both gray and
white matters involved in AD [56]. Patel et al. [55] reported differentially expressed genes which
were common across multiple brain regions in AD using a meta-analysis of AD datasets. They
identified some AD specific genes which were expressed consistently in the same direction across
multiple brain regions which were both gray and white matter regions. The differentially expressed
genes in our analysis using the four brain regions represent AD associated transcriptional changes
in most of the brain rather than tissue specific changes limited to one brain region. The used four
brain regions cover most of the brain parts. Additionally, modeling multiple brain regions together
allows pooling all samples from the data set and consequently increases the statistical power.
Sample sizes for post-mortem brain gene-expression studies are typically very small, so being able
to combine samples from different regions is advantageous.
The hierarchical multilevel structure of the data can be resolved by using linear mixed
models for differential expression. The assumption of independent observations in statistical tests
is violated in such hierarchical data. LMMs can be used to resolve this issue with the full use of
the data. Other approaches exist in dealing with hierarchical structure, but they don’t take the
advantage of using information from all data [77]. Aggregate analysis takes the average of
observations from the same subject which can yield consistent results, but it doesn’t consider the
advantage of having more observations and may lose important differences by averaging all
samples within the same subject. Another alternative is to run separate analysis for each brain
region. This separate analysis method can also work, but it will produce many models and again
does not take the advantage of having information from other brain regions or subjects
simultaneously. Using all available data with LMM will increase the statistical power to detect
differentially expressed genes that are AD specific. Ignoring the correlation structure between the
33
samples from all brain regions using Wilcoxon rank sum test did not show any association with
neurological disorders or AD. In our study, the top differentially expressed genes produced by
LMM integrating all four brain regions showed high association with neurological disorders. The
results verified the effectiveness of using LMM on hierarchical structures of gene expression data.
Several previous studies have used post-mortem AD and control samples of different brain
regions to investigate differential gene expression in a specific region [57,78]. Typically, these
studies examined each region separately in order to investigate the specific changes that take place
in each brain region, which can dissect the pathology and impact over a brain region. The main
purpose of our study was to analyze AD pathology in most of the brain and not to study distinct
functions or transcriptional patterns of different and specific brain regions. Some studies attempted
to integrate all brain regions, but they did not use a single appropriate statistical method that
integrate all brain regions simultaneously. Wang et al. [78] found the differentially expressed genes
in each of six brain regions. In an attempt to integrate all data, they combined the differentially
expressed genes from all regions by taking the union and formed a differential co-expression
network representing all brain regions without using a statistical test that combines all regions.
The genes APOE, AβPP, PSEN1, and PSEN2 did not show any significant differential
expression in our analysis after correction for multiple comparisons using Bonferroni adjustment.
Some previously reported AD associated genes showed differential expression in this study. For
example, the gene RGS6 was found differentially expressed in our analysis of using four brain
regions. The gene RGS6 was previously reported as an AD associated gene [79]. The gene FIBCD1
was also found to be differentially expressed. The gene FIBCD1 encodes for Fibrinogen C
Domain-Containing protein 1 which was previously reported as associated with CNS
inflammation, cognitive decline, and inhibition of repair [80]. Some previously reported AD
34
biomarkers were not found to be among the 602 DEGS. For example, in addition to the genes
APOE, AβPP, PSEN1, and PSEN2, the gene A2M which encodes alpha-2-macroglobulin was
reported as an AD biomarker [81–83], but it was not found as differentially expressed in this study.
Similarly, the gene BDNF which encodes brain-derived neurotrophic factor protein was previously
reported as an AD biomarker [84], and it was not among the DEGs. Few genes out of the top ten
DEGs, in particular NEFL [37], NEFH [37], RGS4 [38], and SNAP25 [37,39], were previously
found to exhibit altered expression levels in various brain cell types in AD brains.
A more specific examination of the top ten genes showed that those top DEGs can be
highly linked to AD. The gene NEFH encodes the heavy neurofilament protein, and the gene NEFL
encodes the light chain neurofilament protein. The gene INA encodes internexin neuronal
intermediate filament protein alpha. Neurofilaments play an important role in axonal growth and
intracellular transport to axons and dendrites. The number of neurofilaments in the axon increases
as the axon becomes myelinated, mature, and connected to its target neuronal cell. The increase in
the neurofilament number determines the cross-sectional diameter of the axon and its transport.
Any abnormalities in the organization or structural constituents of the neurofilaments or synapses
can lead to abnormal axon or synaptic transmission and can signify a dysfunctional connectivity
between neuronal brain cells. It has been reported in previous research studies that neurofilament
mutations and post-translational modifications can lead to Alzheimer’s disease and neurological
diseases [85,86]. The gene PVALB encodes a high affinity calcium ion-binding protein that is
structurally and functionally similar to calmodulin and troponin C. Calcium binding proteins like
Calmodulin have been previously reported to be involved in Amyloid plaques formation and are
linked to Alzheimer’s disease [87]. TESPA1 is also a protein coding gene that codes for thymocyte
expressed positive selection associated 1. This gene was nominated by Agora as an AD target since
35
it was identified to be downregulated in the parahippocampal area in AD brains. RNU6-33P is a
non-protein coding gene, and it is a pseudogene although it was found to be dysregulated in our
analysis. The gene SNAP25 encodes synaptosome associated protein 25 which is involved in the
regulation of neurotransmitter release cycle from synapses, a process which is known to be altered
in AD subjects [88]. SNAP25 was also found in previous research studies to be associated with
AD [89]. The gene VSTM2A encodes V-set and transmembrane domain containing 2A which was
previously reported to be a tumor suppressor although not in brain tissues [90], and some studies
found that tumor suppressor responses can be associated with AD [91]. The gene CHN1 encodes
Chimerin1, a GTPase-activating protein for ras-related p21-rac and a phorbol ester receptor. It is
mainly expressed in neurons and plays an important role in neuronal signal transduction
mechanisms and axonal guidance. CHN1 was never reported as a linked gene to AD, but previous
research studies found that disturbances in signal transduction mechanisms are associated with
Alzheimer’s disease [92]. The gene RGS4 encodes regulator of G protein signaling family
members which act as GTPase activating proteins. RGS4 was previously reported as dysregulated
in AD [72]. It was also previously reported that there is evidence of dynamic regulations of RGS4
levels in neuronal systems, and many neuropsychiatric disorders are linked to dysfunctions of RGS
proteins [93]. The mean gene length (in bp) of the top 10 genes is 41760 bp. This mean length
agreed with some previous findings [94,95] on gene length with AD association where they
reported that the AD associated genes tend to be large on average.
The enriched GO annotations in biological processes, molecular functions, and cellular
components were all related to axonal growth, brain development, the structural constituent of the
synapses, and neurofilament cytoskeleton organization. Schaffer collateral-CA1 synapse also
showed significant enrichment as a cellular component affected in Alzheimer’s disease. As part of
36
the hippocampal structure, it plays a very crucial role in memory formation and information
processing. It was previously reported that this synapse is damaged in AD patients [96]. Schaffer
collaterals play an important role in the limbic system development that affects learning and
memory. Pathway enrichment analysis agreed with the GO annotations where the significantly
enriched pathways were mainly related to axon guidance and neuronal system. These pathways
were associated with axonal growth and transport. It also showed that the voltage gated potassium
channels were affected which signified dysfunctional action potential transmission across the
axon. Transmission across chemical synapses and neurotransmitters release cycle were also shown
to be affected in AD. All these pathways suggested that axon transmission and synaptic
connectivity with other neuronal cells are severely affected in AD. The loss of connectivity
between dysfunctional neurons will eventually spread in brain tissues, and the affected brain
regions will shrink and cause atrophy in the final stages of AD. The downregulated pathways were
related to DNA synthesis and elongation. Translation pathway was also downregulated in addition
to lipid and protein metabolism pathways which were previously reported in some research studies
[97]. The top genes showed a good discriminative power to distinguish between AD and control
samples. It was shown using supervised machine learning algorithms that a random forest classifier
gave a very good accuracy on the test data. There are many definitions of a “biomarker” which
can be broad, controversial, and overlapping at the same time [98]. The detected DEGs, especially
the top genes, may not be considered as genetic biomarkers for AD since the analysis was mainly
limited to late-onset AD subjects. However, they can be considered as biomarkers for axonal
transport and synaptic transmission in AD.
The main conclusions from both (HIP + TCx) and (HIP + TCx + PCx) analyses were
implicit within the main analysis of all four brain regions. Integrating the gene expression data
37
from four different brain regions gave a more comprehensive results and enabled a broader survey
of AD-related gene expression changes, without the constraint of a single brain region. Comparing
the three analyses, adding FWM brain region samples increased the number of differentially
expressed genes as well as the number of upregulated genes when looking at the brain changes. It
increased the statistical power by increasing the number of samples using all 4 brain regions. It
was also shown in previous research that some transcripts may increase after the onset of AD
symptoms. It is crucial to mention that previous research studies have shown that transcriptional
changes for some genes may precede any neuropathy, and transcriptional changes can follow up-
down or down-up changes [99]. This means that some genes can be downregulated or upregulated
before or after any AD symptoms or neuropathy.
limitations on the classification results are the assumption of independence in training the
classifiers and the imbalanced class learning where there were fewer AD samples compared to
control samples. This class imbalance may force the classifier to learn most of the target concepts
from the majority class with poor learning from the minority class. Synthetic oversampling
techniques like SMOTE [100] were not used in order to avoid the possibility of overfitting the
training data and to limit the analysis to true biological gene expression data. Undersampling the
majority class was also not used in order not to lose information from available data. Otherwise,
the accuracy results will be better. Another main limitation in machine learning algorithms is that
we ignored the assumption of independence in our data. However, this assumption is sometimes
violated in biological data, and classifiers can still learn from non-independent training data and
give good accuracies. The small sample size can also be considered a limitation although LMM
deals well with small sample sizes. Another limitation of the study is the inability to study different
stages of AD. Advanced staging of the subjects was not available other than the Braak stages. If
38
stratification of the subjects according to their Braak stages was performed, the analysis will be
very limited by the sample size especially for AD samples since there were only 15 AD subjects
in total, and it will not serve or add to the study main purpose of finding AD associated genes. As
mentioned before, concerted transcriptional changes can occur before any AD neuropathy,
accumulation of amyloid beta, and before the onset of AD symptoms [99]. This means that gene
expression changes in AD subjects can be detected in very early Braak stages before the onset of
AD symptoms [99]. Studying the progression of AD in different brain regions is also a study
limitation. It is beyond the scope and capability of this analysis to address the question in which
brain region the changes occur first given the fact that the data is not longitudinal and without
advanced staging.

39
Chapter 3: Ethno-racial Disparities in Brain Biomarkers for Late-
Onset AD

3.1 Background and Study Purpose
Most AD research studies have focused on the Non-Hispanic White (NHW) subjects. The
composition of the older population in the United States is expected to change ethnically with a
massive increase in the proportion that is racially other than NHW [12]. The US Census Bureau
reported in 2019 that minority ethno-racial populations with Hispanic (HIS) and African American
(AA) origins represented 18.5% and 13.4% of the total population respectively [13], and these
percentages are expected to increase [12]. This demographic change is not reflected in most AD
research studies, and minority groups are usually underrepresented in AD cohorts. Several studies
have investigated ethno-racial disparities in the risk of developing AD. It is often estimated that
there are higher rates of dementia among ethnic minorities [15,16]. AA are about two times higher
than NHW to contract AD in their life span, and similar estimates often suggest that Hispanics are
1.5 times higher than NHW to have AD [16,101,102]. Some research studies have also found racial
differences in mortality rates among AD patients. AA and Hispanics with AD had lower mortality
rates than NHW with AD [103]. These ethno-racial differences are controversial and often
conflicting because some research studies have reported that racial differences in the risk of AD
did not exist [30,104]. The controversy exists not only in the incidence, prevalence, and risk of
AD but also in the biological manifestation and expression of Alzheimer’s disease. Some studies
have shown racial disparities in AD molecular biomarkers like Beta-Amyloid and Tau proteins
[105–107]. Most of these studies have mainly focused on AA and NHW participants who were
not solely comprised of AD diagnosed subjects. Ethno-racial disparities in AD brain biomarkers
are very scarce. The main purpose of this research study was to investigate potential ethno-racial
40
differences in imaging-based brain biomarkers for non-mixed AD participants using three major
ethno-racial groups in the US: Non-Hispanic Whites, African Americans, and Hispanics. Non-
mixed AD was defined as having a clinicopathological diagnosis of AD without the presence of
any mixed etiologies that may contribute to cognitive decline besides AD. We focused on non-
mixed AD diagnosed participants in order to characterize and establish whether there exist any
racial disparities in the biological expression of AD. The main motivation behind studying ethno-
racial disparities in AD was APOE allele 4 which is a high genetic risk factor for AD. While
studying differential gene expressions between AD and control subjects, I learned that APOE allele
4 has stronger effects on non-Hispanic white subjects compared to other races. Studying ethno-
racial disparities became an interesting topic to investigate and understand the biological
expression of AD in different races. Some studies on ethno-racial disparities in AD have been
performed, but they mainly focused on incidence and prevalence rates. Some studies also
considered differences in the hippocampus and molecular biomarkers like beta amyloids or tau
proteins, but they mainly used two races like whites and African American subjects. Additionally,
they used MCI and AD patients. In this study, we focused on using non-mixed AD subjects only
and from three different races to study brain biomarkers from most of the brain that can be affected
by AD.

41
3.2 Materials and Methods
3.2.1 Data and Participants

The cross-sectional data was obtained from December 2020 data freeze of NACC. The data
was restricted to three main racial groups: NHW, AA, and HIS AD participants. Eligibility criteria
included participants 1-who were older than 50 years old to study late-onset AD, 2- who had at
least one T1-weighted MRI scan, and 3- who did not have any other psychological, cardiovascular,
and cerebrovascular diseases that may contribute to cognitive decline besides AD. In order to
obtain non-mixed AD data and meet the third eligibility criterion, AD participants were excluded
if they had self-reported TBI, seizures, stroke, Parkinson’s disease, atrial fibrillation, congestive
heart failure, blood pressure, diabetes, sleep apnea, bipolar disorder, Schizophrenia, PTSD, active
depression, anxiety, and obsessive-compulsive disorder.
Some subjects were further excluded if they did not have their Apolipoprotein E (APOE)
e4 allele status and clinical dementia rating (CDR® Dementia Staging Instrument) for staging the
severity of AD [108]. Informed consent forms were obtained from all participants according to
their AD research centers.
Table 3.1 shows the characteristics of all participants in this study. There were 42 AA, 42
NHW, and 41 Hispanic non-mixed AD subjects. There was no statistically significant difference
in the age distribution among all AD subjects from the three racial groups, but there was significant
difference in education level using one-way analysis of variance (ANOVA). There was no
significant difference in the CDR scores among AD participants using ANOVA. The percentage
of NHW AD subjects with a CDR value greater than 0 was around 95.24% with 11.9% having a
CDR value of 3. Similar to NHW AD subjects, AA AD subjects with CDR value greater than 0
were around 95.24% with 11.9% having a CDR value of 3. HIS AD subjects with CDR greater
42
than 0 were around 92.68% with 9.76% having a CDR value of 3. The percentages of AD
participants who had at least one allele of APOE e4 were 47.6%, 50.0%, and 36.6% for AA, NHW,
and HIS respectively.

Table 3. 1 Summary of demographics and characteristics for AD subjects
Participants with Alzheimer’s Disease
Characteristic AA (n=42) NHW (n=42) HIS (n=41) P Value
Mean Age (SD) in Yrs. 79.55 (8.08) 78.45 (7.76) 79.34 (8.73) 0.8908
Age Range [59-92] [63-94] [59-95]
Male Sex (%) 33.3% 54.7% 46.3%
Mean Ed. Lv. (SD) in Yrs. 13.33 (3.39) 15.91 (2.86) 9.44 (6.09) 2.54E-9
1 or 2 APOE e4 allele (%) 47.6% 50.0% 36.6%
Mean CDR (SD) 1.08 (0.87) 1.14 (0.89) 1.19 (0.85) 0.8442
CDR > 0 (%) 95.24% 95.24% 92.68%
TIV (CC) 1278.9 (135.01) 1515.5 (155.57) 1307.4 (139.95) 2.06E-12
Abbreviations: APOE, Apolipoprotein E; SD, standard deviation; Yrs, years; Ed. Lv.,
Education Level; CDR, Clinical Dementia Rating; TIV, Total Intracranial Volume.
The characteristics of the AD participants are reported. P value is used for testing the
hypothesis whether there is significant difference among AD subjects between the three
groups using one-way ANOVA. Results were deemed statistically significant at a P value <
0.05.

43
3.2.2 MRI Derived Brain Measures

All brain measures were derived from T1-weighted MR images and provided by NACC
where volume calculations were performed by UC Davis IDeA lab. Advanced MRI methods were
used for the removal of non-brain tissues, image intensity inhomogeneity correction, and ROI
segmentation and measurement [109–112]. Nine brain regions from both hemispheres, that are
usually affected by AD and including most of the brain, were selected for this analysis: total
hippocampal volume (HIP), total entorhinal gray matter volume (ENT GM), total brain volume
(TB), total brain white matter volume (WM), total brain gray matter volume (GM), total brain
cerebrospinal fluid (CSF) volume, temporal lobe cortical gray matter volume (TL GM), parietal
lobe cortical gray matter volume (PL GM), and total brain white matter hyperintensity volume
(WM HYP).

44
3.2.3 Statistical Analysis

Statistical analysis was conducted on the cross-sectional data for the non-mixed AD
individuals from the three ethno-racial groups. Each brain volume measure was fit into a linear
regression model with covariates as confounding factors including age, sex, education level, APOE
e4 allele status, CDR, scan site, and total intracranial volume (TIV). Race was also included in the
model as a covariate, but it was not regressed out since the effect of race was of main interest in
the analysis. The brain measures were adjusted for the confounding factors in the regression model
using the residual method to eliminate their effects [113,114]. After adjusting each brain measure
by regressing out the confounding factors, one-way ANOVA and pair-wise statistical comparisons
using Wilcoxon rank sum statistic were performed to test whether there were any statistical
differences between the three groups. P values of the pair-wise comparisons were further adjusted
for multiple comparison with the false discovery rate correction [115]. All P values were deemed
statistically significant at a P value < 0.05. Additionally, Cohen’s D effect size was also calculated
in each comparison.
The association and interaction effect between age and race were studied for AD
participants among the three ethno-racial groups. The effects of the confounding factors were
regressed out from the brain measures except for age effects in order to study how the AD brain
volumetric measures are racially affected by age. The association and interaction effects between
race and sex and between race and APOE e4 allele were also studied using two-way ANOVA. The
effects of sex and APOE e4 allele were not regressed out to understand their effects on AD
participants. Multiple pair-wise comparisons followed two-way ANOVA to give more detailed
results on the associations and know the pairs with significant differences.

45
3.3 Results
3.3.1 Ethno-Racial Differences in Brain Measures

Many brain measures showed statistically significant differences between NHW and AA
and between NHW and Hispanic AD participants after adjustment for confounding factors and
multiple comparisons correction as shown in Table 3.2. There were no significant differences
between AA and Hispanic AD subjects in any of the brain measures except the parietal lobe
cortical gray matter volume (PL GM). However, the difference in PL GM became significant only
after adjustment for scan site. The HIP, TL GM, WM HYP, and ENT GM volumes showed
significant differences between NHW and AA participants after adjustment for confounding
factors and multiple comparisons correction. The most significant difference between AA and
NHW was in the WM HYP with a P value of 0.0003 and effect size of 0.79 which is considered a
large effect. The HIP, TL GM, and WM HYP showed significant difference between NHW and
Hispanic AD subjects. The most significant difference was in the hippocampal volume (HIP) with
a P value of 0.002 and effect size of 0.6. The TB, WM, GM, and CSF did not show any significant
difference using one-way ANOVA or pair-wise comparisons between each pair of races as shown
in Table 3.2.
Table 3.2 shows the comparisons of the raw measures between the three racial groups
before adjustment using the mean value. Table 3.2 also shows the comparison of the adjusted
measures where the median values were used instead of the mean values in the comparison since
adjustment changed the scale of the values. The adjustment for confounding factors changed and
inverted the comparison for some measures especially the cerebrospinal fluid (CSF). This change
was mainly due to the TIV adjustment since NHW AD subjects had the highest TIV value in units
of cc with mean and standard deviation of 1515.5 and 155.57. The TIV mean (SD) values for AA
46
and HIS AD subjects were 1278.9 (135.01) and 1307.4 (139.95) respectively. One-way ANOVA
gave a P value of 2.06E-12 for the difference in TIV between the three groups as shown in table
3.1. As an example, the mean value of the raw CSF volume was higher for NHW AD subjects
followed by HIS and then AA subjects. However, the order of comparison was inverted after
adjustment for confounding factors due to TIV correction as shown in Table 3.2. The order of
comparison did not change in most cases without TIV correction.

47
Table 3. 2 Statistical differences in brain volumetric measures for AD participants
Measure N/A (D) N/H (D) A/H (D) 1-ANOVA Stat1 Stat2
1-HIP 0.01 (0.45) 0.002 (0.6) 0.16 (0.19) 0.026 N>H>A N>A>H
2-TB 0.055 (0.41) 0.17 (0.28) 0.55 (0.07) 0.18 N>H>A N>H>A
3-WM 0.63 (0.11) 0.17 (0.22) 0.53 (0.1) 0.61 N>H>A N>A>H
4-GM 0.11 (0.37) 0.57 (0.15) 0.33 (0.23) 0.21 N>H>A N>H>A
5-CSF 0.37 (0.2) 0.4 (0.14) 0.9 (0.03) 0.64 N>H>A A>H>N
6-TL GM 0.04 (0.4) 0.02 (0.47) 0.56 (0.06) 0.05 N>H>A N>A>H
7-PL GM 0.057 (0.45) 0.78 (0.03) 0.01 (0.53) 0.03 N>H>A N>H>A
8-WM HYP 0.0003 (0.79) 0.02 (0.54) 0.22 (0.12) 0.0046 A>H>N A>H>N
9-ENT GM 0.033 (0.49) 0.1 (0.32) 0.7 (0.19) 0.06 N>H>A N>H>A
Abbreviations: The abbreviations of the brain volumetric measures are stated in section 2.2;
N, Non-Hispanic White; A, African American; H, Hispanic; D, Cohen’s D Effect size in
absolute value; Race1/Race2 means the pair-wise comparison between race1 and race2; 1-
ANOVA, one-way ANOVA for the comparison of the three ethno-racial groups; Stat1, status
1 represents the comparison of the raw measures using the mean values before adjustment
for confounding factors; Stat2, status 2 represents the comparison of the measures using the
median values after adjustment for confounding factors.
This table shows the P values for the pairwise comparisons and one-way ANOVA for each
brain measure after adjustment for confounding factors: Age, Sex, Education level, APOE
Allele 4 status, CDR, Scan Site, and TIV. All P values were FDR corrected for multiple
comparisons and deemed statistically significant at a P value < 0.05.

48
3.3.2 The Changes of Brain Measures with Age and Race

All nine brain volume measures were adjusted for confounding factors except for age and
race. The adjusted measures revealed a volume decrease with advanced aging except for the CSF
and WM HYP volumes which showed increasing volumes with aging as shown in figure 3.1 C
and D respectively. Figure 3.1C shows how the CSF volume exhibited an increase with age in the
three racial groups with a faster increase for NHW AD subjects as seen from the slope of the linear
fit. Figure 3.1D shows the change of the WM hyperintensity volume with age in which AA and
HIS AD participants showed similar increase with age and faster than NHW subjects. The
hippocampal volume revealed a decrease in volume with advanced aging for all AD participants
as shown in figure 3.1A with slightly faster decrease for NHW AD subjects. The volume decrease
with age in the WM volume was the fastest for NHW AD participants as shown in figure 3.2A.
The GM volume showed similar volume decrease for all races, and the PL GM volume showed
faster decrease for AA AD subjects. The entorhinal volume showed very similar decrease with age
for all races with almost parallel linear fit as shown in figure 3.1B. The total brain volume showed
a very significant volume decrease with age for all races with a faster decrease for NHW than AA
and HIS AD subjects as shown in figure 3.3. All measures showed significant change with age
with P values less than 0.05 except for the entorhinal volume as shown in table 3.3. All brain
measures did not show any interaction effect between race and age which meant that the effects of
age did not depend on race as shown in table 3.3.

49

Figure 3. 1 The plots of the change of 6 brain measures with age and race
The plots include the linear fit that shows the rate of change of the measures with age for
each race as seen from the slope of the straight line. All brain measures were adjusted for
Sex, Education level, APOE, CDR, Scan site, and TIV as confounding factors except Age.
A A B
C
D
E
F
50

Figure 3. 2: The change of the white and gray matter volumes with age and race
The plots show the change of the total brain white matter (A) and gray matter (B) volumes
with age and race. The plots include the linear fit that shows the rate of change of the
measures with age for each race as seen from the slope of the straight line. All brain measures
were adjusted for Sex, Education level, APOE, CDR, Scan site, and TIV as confounding
factors except Age.

A
B
51

Figure 3. 3: The change of the total brain volume with age and race
The brain measure was adjusted for all confounding factors except age and race. NHW AD
subjects showed the fastest decrease with age, and HIS and AA subjects showed a similar TB
volume decrease with age.

52

Table 3. 3 Interaction effect between Race and Age, Sex, and APOE allele 4 status
Measure Age Age*Race Sex Sex*Race APOE APOE*Race
1-HIP 0.02 0.78 0.98 0.29 0.087 0.13
2-TB 3E-11 0.53 0.55 0.34 0.44 0.63
3-WM 1E-4 0.25 0.33 0.34 0.32 0.29
4-GM 1.8E-6 0.55 0.09 0.73 0.91 0.7
5-CSF 1.4E-8 0.34 0.94 0.29 0.44 0.69
6-TL GM 0.04 0.55 0.55 0.92 0.33 0.97
7-PL GM 1E-4 0.55 0.05 0.92 0.6 0.5
8-WM HYP 5.3E-8 0.73 0.07 0.43 0.75 0.46
9-ENT GM 0.1 0.99 0.71 0.43 0.27 0.14
This table shows the P values of the interaction effects between Race and each of Sex, Age,
and APOE status using 2-way ANOVA. It also shows the association effect of each factor. All
brain measures were adjusted for confounding factors except for Race and the factor of
interest (Sex, Age, or APOE). P values were deemed statistically significant at a P value <
0.05.

53
3.3.3 The Association of Race and Sex

The brain volume measures were adjusted for age, education level, APOE status, CDR,
scan site, and TIV without adjusting for sex and race. All adjusted brain measures did not show
any significant interaction effect between sex and race, and the effects of sex did not depend on
race as shown in table 3.3. Multiple pair-wise comparisons did not show any significant differences
in brain volume measures between males and females except in the GM and WM HYP volumes
within the same race. The GM volume showed significant differences in NHW and HIS AD
subjects between males and females with males having less median for the adjusted volumes than
females as shown in figures 3.4A and 3.4B. The P values were 0.038 with effect size of 0.459 and
0.048 with effect size of 0.408 in NHW and HIS subjects respectively. The white matter
hyperintensity (WM HYP) showed significance between males and females only in the Hispanic
AD subjects. The P value was 0.04 in HIS subjects for the WM HYP volume with effect size of
0.522. The raw volumes and the volumes that were adjusted for confounding factors without TIV
showed higher mean and median values for males, but adjustment for TIV caused the median to
be higher in females. There were significant differences in TIV between sexes within each race.
Males had significantly higher mean TIV values than females in the three groups with P values of
1.23E-5, 1.28E-4, and 2.1E-5 in NHW, AA, and HIS AD participants respectively.

54

Figure 3. 4: The effects of race and sex
The boxplots show the effects of race and sex on two brain measures in the non-mixed AD
participants. (A) GM volume; (B) WM HYP volume.
** means statistically significant difference was found between males and females within the
same race using Wilcoxon test in pair-wise comparisons. The volumes were adjusted for the
confounding factors including TIV but excluding Sex.
Abbreviations: M, Males; F, Females.

A
B
55

3.3.4 The Association of Race and APOE

The effects of race and APOE e4 allele on the brain measures in non-mixed AD participants
were also studied to analyze the association between race and APOE. The volume measures were
adjusted for the confounding factors except race and APOE e4 allele status. Two-way ANOVA
did not show any significant interaction effect between race and APOE for all brain measures even
for HIP and ENT GM volumes as shown in table 3.3 mainly because there were no APOE effects
on AA and HIS AD participants. Multiple pair-wise comparisons showed statistically significant
differences between APOE e4 carriers and non-carriers in the hippocampal volume (P = 0.0416,
D = 0.6469) and the entorhinal GM volume (P = 0.0497, D = 0.627) but only in NHW AD
participants. AD NHW APOE carriers showed lower median adjusted HIP and ENT GM volumes
than AD non-carriers as depicted in figures 3.5A and 3.5B. APOE e4 allele status did not appear
to have any significant effects on AA or HIS AD participants.

56

Figure 3. 5: The effects of race and APOE
The boxplots show the effects of race and APOE on two brain measures in non-mixed AD
participants. (A) HIP volume; (B) ENT GM volume.
** means statistically significant difference was found between APOE e4 allele carriers and
non-carriers within the same race. The volumes were adjusted for Age, Sex, Education level,
CDR, Scan site, and TIV excluding APOE status.
Abbreviations: C, AD Carriers who had at least one APOE e4 allele; NC, AD Non-Carriers
who did not have APOE e4 allele.

A B
57
3.4 Discussion
Most AD research studies have focused on the differences between healthy and AD
subjects using mainly NHW elderly subjects. Our study was mainly focused to analyze the
disparities in brain imaging biomarkers between AD elderly participants of three major ethno-
racial backgrounds in the nation. Studying the ethno-racial differences in brain biomarkers using
non-mixed AD subjects is very scarce and even don’t exist. Non-mixed elderly AD participants
were solely used in a matched analysis in this study to reduce the number of variables and eliminate
any etiologies that may contribute to cognitive decline and brain atrophy besides AD. This can
help understand the differences in the biological expression and manifestation of Alzheimer’s
disease in different racial groups regardless of cardiovascular, cerebrovascular, or any mixed
dementia type besides AD. The brain measures were adjusted for main confounding factors [113].
The measures were adjusted for CDR since different AD subjects may have different stages of
dementia, so the adjustment of CDR can correct for the inter-subject variability in the stage and
severity of AD. The measures were adjusted for education because higher education level has
been associated with lower dementia risk [116]. Higher education can also be an indicative for a
healthy lifestyle. The volume brain measures were also adjusted for age because age is considered
the first risk factor for AD. Sex was also used as a confounding factor since women were reported
to have higher incidence of AD [117]. Additionally, we used scan site as a confounding factor
since the used cross-sectional data was acquired from NACC which collects imaging data from
multiple scan sites at different ADRCs. In combining multi-site neuroimaging data, there is a need
to remove the non-biological variance introduced by different MRI scanner hardware and
acquisition protocols. The analysis proposed significant differences in many brain measures
between NHW and AA or between NHW and HIS AD participants after adjusting for the
58
confounding factors. None of the brain measures showed any statistically significant differences
between AA and HIS participants except the parietal lobe cortical gray matter volume and only
after adjusting for scan site effects. These results implied that global and regional brain atrophies
and neurodegeneration as an outcome of the biological manifestation of AD and race were more
similar between AA and HIS than NHW elderly AD patients.
Advancing age comes with inevitable consequences of deterioration in most biological
systems. Many studies have found a strong association between increasing age and the rate of brain
atrophy in normal aging [118], and the atrophy and neurodegeneration are even more exacerbated
in AD patients where age is the greatest risk factor for AD [119,120]. Most brain measures in this
study showed atrophy and volume decrease with advancing age except for the CSF and WM
hyperintensity which showed volume increase with aging. Most brain measures showed similar
volume decrease with age for the three racial groups which made it more difficult to assess or
determine which race has the fastest AD progression with aging. However, NHW AD subjects
showed the fastest decrease in the white matter volume and the fastest increase in CSF volume.
AA showed the fastest decrease in parietal lobe cortical gray matter volume and total brain gray
matter volume. Both AA and Hispanic AD subjects showed faster increase in white matter
hyperintensity than NHW AD subjects. WM HYP are highly correlated with cognitive decline and
increasing risk of dementia and stroke [121]. The hippocampal and the entorhinal volumes
exhibited almost the same change with advancing age among the three races.
Studies on sex differences in dementia have shown that women had higher incidence rates
and greater AD risk than men, and almost two thirds of AD cases are women [117]. Studies have
reported that women appeared to suffer greater cognitive decline and brain atrophy than men, and
males appeared to have more cognitive reserve possibly due to education and occupation levels
59
[117,122–124]. The effects of sex in our study appeared mainly on the total brain GM volumes
where significant differences were found between males and females in NHW and HIS AD
participants. Males showed less adjusted volumes than females when TIV correction was
performed. The raw and adjusted volumes without TIV adjustment showed higher volumes in
males, but TIV correction inverted the results since males had greater mean TIV than females in
all races. Some studies have previously reported that TIV adjustment will result in higher measures
in females which agreed with our results [125–127]. Significant sex differences also appeared in
WM HYP volumes in Hispanic AD subjects with females having higher adjusted volumes than
males when TIV correction is included.
APOE genotype with e4 allele polymorphism has been established in previous research
studies as a high risk for developing AD with lower age of onset of symptoms [128]. Previous
research studies have found that the risk of AD was significantly increased for NHW individuals
with e4 allele, but APOE e4 allele association with AD was weaker among AA and Hispanic AD
subjects [22,129]. APOE e4 allele was also previously found to have a strong effect in the
hippocampal and entorhinal cortex atrophy [130,131]. Our results confirmed and agreed with
previous studies. In this study, the effects of APOE e4 appeared in the hippocampal and entorhinal
GM volumes, and it mainly appeared in NHW AD participants. There were significant differences
in the hippocampal and entorhinal volumes between APOE e4 carriers and non-carriers in NHW
AD subjects. APOE e4 did not have significant effects on AA or Hispanic AD participants. AD
is known to affect entorhinal and hippocampal volumes, and both regions are involved in memory
dysfunction which is one of the earliest hallmarks of AD [130–132].

60
The most plausible explanation for these racial differences in brain imaging biomarkers
can be attributed to genetic differences between the three races. Genetics can be considered the
first reason to explain these racial differences since we have corrected for many variables and
matched the AD subjects to eliminate any factors that may contribute to cognitive decline. We
have used non-mixed AD subjects to eliminate the possible contributions of cardiovascular and
cerebrovascular diseases and risk factors in addition to some dementia types. Although genetics
can be considered the first factor or reason, it may not be the only reason to explain these ethno-
racial differences. There might be some hidden factors that may introduce some variability in the
contribution of AD pathogenesis especially that AD is a multifactorial disease with mixed
etiologies. Cultural differences, access to health care, environmental toxins, or socio-economic
status in addition to many other variables may contribute to brain atrophy and AD which were
unavailable to consider in the analysis. This study has few limitations. Molecular AD biomarkers
like Beta-Amyloid or Tau protein levels were not included in the analysis because few NACC
subjects had their neuropathy data. Another limitation is the sample size which is not necessarily
an enough representation for the whole ethno-racial groups, so the results may not be very
generalizing. Many subjects were also missing their MMSE and MoCA cognitive test scores
because NACC started using MoCA scores after 2015 [133]. However, there was no significant
difference in the available MoCA scores between the three races.

61
Chapter 4: The Effects of Environmental Noise Exposure on Human
Health and Brain

4.1 Background and Study Purpose
Environmental pollution has recently become a focus of the scientific efforts for its role as
a risk factor for many adverse health effects including cardiovascular diseases and dementia
including Alzheimer’s disease. While air pollution and its effects on the human health and brain
have been studied, the effects of noise pollution are still not very well understood or investigated.
Noise could cause adverse health effects, and it is considered as a hazard to human health whether
it is transportation noise or from industry and construction [27]. Unlike air pollution, studies on
noise pollution and its effects and relation to dementia and human brain are very scarce. It was
estimated in 2013 that more than 104 million individuals in the US had an average exposure level
of greater than 70 dBA over 24 hours which is considered harmful to human health and put them
at a greater risk of hearing loss according to the US environmental protection agency (EPA) [134].
Research studies have linked noise exposure to cardiovascular risk factors and diseases.
Noise exposure was associated with hypertension, obesity, and increased heart rate [27,38,39].
Noise exposure has also been associated with the release of stress hormones, sleep disturbances,
increased oxidative stress, endothelial dysfunction, and inflammation [135–138]. All these risk
factors may contribute to the onset of dementia and Alzheimer’s disease. Many research studies
on noise pollution have reported cognitive decline and higher incidence and prevalence rates of
dementia and Alzheimer’s disease [40–42]. Community noise is a modifiable risk factor at the
individual and population levels, and its effects can be alleviated through governmental regulations
and possibly technological advances. Studies on the effects of noise exposure on brain atrophy and
neuropathological brain changes have been limited to animal experiments. Long term exposure to
62
high level of noise in animal studies have indicated changes in brain regions that are mostly
affected by Alzheimer’s disease. High levels of noise exposure in rats have been linked to over
production of Beta Amyloid (AB) in addition to hyperphosphorylated tau and neurofibrillary
tangles (NFT) in the hippocampus and the prefrontal cortex [43,139,140]. These are considered
early signs and hallmarks in Alzheimer’s disease etiology which suggests that long term noise
exposure may lead to the onset and progression of AD. Noise induced AB and NFT in animals
may lead to neuronal apoptosis and synaptic malfunction. Noise exposure can cause inflammation
and oxidative stress which are associated with AD-like neuropathological changes [43,141]. A
number of animal studies on noise exposure have also reported cognitive decline in spatial learning
and memory [140,142–144]. Neuroimaging studies on noise exposure and dementia in humans are
very scarce and not well investigated. In this study, we attempted to investigate the effects of
environmental noise exposure on human health and dementia. Neuroimaging studies on noise
exposure in humans that relate noise pollution to brain atrophy and dementia are very scarce and
even don’t exist to the best of our knowledge although some studies have reported cognitive
decline with noise pollution in humans. We studied the effects of noise on the human brain using
neuroimaging data, and we also studied the association of noise exposure with many health-related
phenotypes and risk factors. The main motivation behind this study analysis is to investigate the
effects of environmental pollution especially that AD is a multifactorial disease with mixed
etiologies. Environmental pollution can be a factor that contribute to the pathogenesis of
Alzheimer’s disease. After studying gene expression and ethno-racial differences, environmental
pollution as an AD cause was an interesting topic to investigate.

63
4.2 Material and Methods
4.2.1 Study Population

The used data is from the UK biobank project (November 2020 release) which consists of
a prospective cohort with a rich variety of extensive phenotypic and health information available
on approximately 500,000 individuals across the United Kingdom, aged between 40 and 69 at
recruitment. At the initial recruitment visit between 2006 and 2010, the participants provided
signed electronic consent, provided health related information, and completed a range of physical
measures. Starting in 2014, a subset of participants attended a follow-up assessment visit during
which imaging data was collected [145–147]. The UK biobank data includes a wide range of health
information that includes genetic, data from blood samples, analyses of urine and saliva samples,
information about lifestyle and health, clinical assessments, cognitive tests, and many more.

64
4.2.2 Assessment of Exposure to Noise and Air Pollution

The annual average noise pollution estimates for the year 2009 were modelled using a
version of the CNOSSOS-EU noise model which represents a coherent and harmonized framework
to assess noise levels from the main sources of noise (road traffic, railway traffic, aircraft and
industrial) across Europe. These common noise assessment methods (CNOSSOS) are
recommended by the European Noise Directive 2002/49/EC. Noise mapping was performed using
a sound propagation model which considers the propagation paths between the noise sources and
the receiver points [148,149]. The average 24-hour sound level of noise pollution LDen was
mainly used in this study. LDen is the A-weighted noise level measure over the 24-hour period in
dBA (A-weighted decibels). The annual average air pollution concentration estimates were
modelled for each residential address using a land use regression model between 2010 and 2011.
The model was developed as part of the European Study of Cohorts for Air Pollution Effects
(ESCAPE) (http://www.escapeproject.eu). GIS-derived predictor variables (e.g., traffic
intensity, population, and land-use) were evaluated to model small scale spatial variation of annual
average air pollution concentrations for each study area [150]. This model can estimate individual
exposure to air pollution concentration based on the participant’s residential address. PM2.5
(particulate matter 2.5 with aerodynamic diameter less than 2.5 µm) was used in this study since
it is ultrafine air pollution particles and considered to have the most serious health adverse effects.

65
4.2.3 Brain Imaging Data and Processing

The UK biobank project has used a variety of imaging modalities to estimate anatomical
brain structures (T1-weighted, T2-weighted, and swMRI), tissue microstructure (diffusion MRI),
and brain activity (functional MRI). T1-weighted MR scans were mostly used to derive the main
anatomical brain structures and volumes [151]. T1-derived IDPs (image derived phenotypes) can
estimate and provide crucial markers of brain atrophy whether it is global or local brain structures.
The T1 structural images were defaced first to keep subjects’ anonymization, and then the removal
of non-brain tissues were performed using Brain Extraction Tool (BET) and FMRIB’s Linear
Image Registration Tool (FLIRT) [152–154]. The data was then nonlinearly warped to the
MNI152 template space using FMRIB’s Nonlinear Image Registration Tool (FNIRT)[155,156].
T1 images are further processed to extract derived measures of a variety of brain tissues and
structures via the FreeSurfer software using the Desikan-Killiany atlas [157,158]. The FreeSurfer
derived measures and outputs were then Quality Control checked using the Qoala-T approach
[159]. The total white matter hyperintensity volume was derived from T2 FLAIR images and T1
images. The white matter lesion segmentation was processed using the BIANCA tool [160].

66
4.2.4 Statistical Analyses

Subjects who have imaging data in addition to air and noise pollution data were divided
into two groups in order to investigate whether there exist any significant statistical differences in
brain measures between the two groups. The two groups were (i) G55 group containing subjects
who were exposed to noise levels below 55 dBA, and (ii) G70 group containing subjects who were
exposed to noise levels above 70 dBA since it is considered the threshold for having a higher risk
of health adverse effects according to the EPA [134]. Table 4.1 shows some demographic and
physical characteristics of the subjects in both groups. An initial statistical imaging analysis was
performed on 11 selected brain volume measures that include the left hippocampal volume, the
right hippocampal volume, the left entorhinal volume, the right entorhinal volume, the total brain
volume, the total brain gray matter volume, the total brain white matter volume, the total
cerebrospinal fluid volume, and the total white matter hyperintensity volume. These brain volume
measures were selected to assess the effects of noise exposure on brain regions that are typically
affected by Alzheimer’s disease particularly the hippocampal and the entorhinal volumes[130–
132]. Additionally, the transverse superior temporal gyrus (Heschl’s gyrus) volume in both left
and right hemispheres were selected since these brain regions are responsible for hearing [161].
A second statistical analysis was performed on 1286 image derived phenotypes (IDPs)
including volumes, thicknesses, and areas of brain regions. In both statistical analyses on the 11
selected IDPs and the 1286 IDPs, three models were created: (1) the first model did not adjust for
any confounding factors and mainly studied the differences between the G55 and G70 groups using
raw brain measures, (2) the second model adjusted for age, sex, APOE E4 status, education level,
and TIV if it was a volume measure, and (3) the third model adjusted for income, BMI, diabetes,
hypertension, and air pollution PM 2.5 in addition to the variables in model 2. In both statistical
67
analyses, a linear regression was fit in models 2 and 3 where the IDP and the confounding factors
were the dependent and the independent variables respectively. The imaging measures were then
adjusted by regressing out the confounding variables. Wilcoxon rank sum test was then performed
on the adjusted brain IDPs to assess any significant statistical differences between the G55 and
G70 groups.
Statistical analyses were also performed on several cognitive and hearing tests. Fluid
intelligence score, mean time to correctly identify matches, and maximum digits remembered
correctly were the cognitive scores that were used in the statistical analyses. We also used speech
reception threshold estimates on the left and right hemispheres as hearing tests. Wilcoxon rank
sum test was used to assess the differences between the two groups for all these cognitive and
hearing tests. P-values for all statistical analyses were corrected for multiple comparisons using
false discovery rate adjustment. P-values were deemed statistically significant below 0.05
significance level.
Multiple phenome scans were also performed to test the association of noise exposure as
the trait of interest with a comprehensive set of 20,622 non-imaging phenotypes using the phenome
scan analysis tool (PHESANT) [162]. PHESANT is implemented in R and can use a large
heterogenous subset of phenotypes with different data types. The non-imaging phenotype and the
noise exposure variable were the dependent and the independent variables of the regression
respectively. Continuous data type phenotypes were tested using linear regression (lm R function).
Binary, un-ordered categorical, and ordered categorical variables were tested in PHESANT using
binomial regression (glm R function), multinomial logistic regression (multinom R function), and
ordered logistic regression (polr R function) respectively. Three separate phenome scans were
performed on the non-imaging phenotypes with different confounding factors. The first model
68
used age, sex, and time spent at current residential address as confounding factors. The second
model used air pollution PM 2.5 measure as a confounding factor in addition to the factors in
model 1. The second model corrected for air pollution in order to remove the effects of air pollution
in the association study since air and noise pollution are usually strongly coupled. The third model
corrected for sleep duration, income, and education level in addition to the confounding factors in
models 1 and 2. The reason for sleep duration correction was to test if the significant associations
were due to noise exposure directly independent of sleep disturbances and fragmentation. Higher
socio-economic status is usually associated with better health, so the correction of income and
education levels will remove the effects of socio-economic status. Similarly, p-values were
corrected for multiple comparisons using false discovery rate adjustment. P-values were deemed
statistically significant below 0.05 significance level.

69
Table 4. 1 Summary of demographics and characteristics for subjects in G55 and G70
groups
G55 G70
(51.56 – 55 dBA) (70 – 82.01 dBA)
Category Mean (SD) or Percentage Mean (SD) or Percentage
Age (years) 63.73 (7.53) 63.35 (7.61)
Male (%) 47.26 47.22
Education Level
College degree (%) 48.13 45.56
A/AS levels (%) 12.41 11.73
GCSEs or equiv.(%) 18.37 18.8
CSEs or equiv. (%) 3.79 3.46
NVQ, HND, HNC (%) 5.91 6.17
Other professional qual.(%) 5.11 5.11
None / No answer (%) 6.28 9.17
BMI (Kg/m2) 26.51 (4.16) 26.9 (4.34)
Air PM 2.5 (µg/m3) 9.75 (0.83) 11.43 (1.58)
Have Diabetes (%) 2.49 2.86
APOE Allele 4 Status
Have no Allele 4 (%) 72.38 69.32
Have one Allele 4 (%) 25.36 29.02
Have 2 Allele 4 (%) 2.26 1.65
Blood Pressure
Normal (%) 22.52 22.41
Prehypertension (%) 38.9 39.4
Hypertension (%) 38.58 38.2
Average Total Income
Less than 18K Pds (%) 9.77 13.53
18K - 30,999 Pds (%) 19.61 21.2
31K – 51,999 Pds (%) 36.65 36.24
52K – 100K Pds (%) 26.85 24.66
Greater than 100K Pds(%) 7.12 4.36
TIV (mm3) 1.549E+6 (1.52E+5) 1.548E+6 (1.5E+5)
Abbreviations: SD, standard deviation; A/AS, advanced/advanced subsidiary; GCSE,
general certificate of secondary education; CSE, certificate of secondary education; NVQ,
national vocational qualifications; HND, higher national diploma; HNC, higher national
certificate; equiv., equivalent; qual., qualifications; APOE, Apolipoprotein E; Pds, Pounds;
TIV, total intracranial volume.
The characteristics of the subjects between the two groups that were used in the analysis
were reported. There were 19394 subjects in the G55 group and 665 subjects in the G70
group who had imaging data and other covariates information.
70
4.3 Results
4.3.1 Neuroimaging Analysis

After eliminating subjects who did not have imaging data and pollution data, some
participants were further excluded if no information was found for APOE allele 4 status. The
subjects were then divided into two groups according to their noise exposure levels. G55 and G70
were the two groups of subjects who were exposed to noise levels below 55 dB and above 70 dB
respectively. The total number of remained subjects in both groups was 20059 subjects with the
majority of 19394 subjects in G55 and 665 subjects in G70. G55 subjects were exposed to a noise
level between 51.56 and 55 dB A-weighted with mean and SD of 53.35 and 1.07 dB respectively.
G70 subjects were exposed to a noise level between 70 and 82.01 dB A-weighted with mean and
SD of 71.85 and 1.88 dB respectively. Table 4.1 shows some demographic and physical
characteristics of the subjects in both groups. Statistical analysis on the 11 selected brain volume
measures did not show any significant statistical differences between the two groups after using
the 3 models which corrected for multiple confounding factors. P values were also corrected for
multiple comparisons using false discovery rate adjustment. None of the p values were lower than
the significance level of 0.05. Table 4.2 shows the 11 selected IDPs with the p values from each
of the three models. Additional statistical analysis was performed on 1286 IDPs using the three
models. None of these 1286 IDPs showed any significant statistical difference between the two
groups using the 3 models after adjustments for confounding factors and correction for multiple
comparisons.

71

Table 4. 2 Statistical analysis on 11 selected brain volume measures between G55 and G70
groups
Brain ROI Volume Model 1 Model 2 Model 3
1-Left Hippocampus 0.407 0.364 0.622
2-Right Hippocampus 0.591 0.256 0.479
3-Left Entorhinal Vol. 0.942 0.882 0.988
4-Right Entorhinal Vol. 0.258 0.317 0.225
5-Total Brain Vol. 0.598 0.362 0.624
6-TB Gray Matter Vol. 0.479 0.113 0.373
7-TB White Matter Vol. 0.781 0.996 0.969
8-TB CSF Vol. 0.553 0.378 0.458
9-TB White Matter Hyp. Vol. 0.939 0.35 0.583
10-Left Transverse STG Vol. 0.94 0.834 0.922
11-Right Transverse STG Vol. 0.404 0.196 0.26
Abbreviations: ROI, region of interest; Vol., volume; TB, total brain; CSF, cerebrospinal
fluid; Hyp., hyperintensity; STG, superior temporal gyrus.
The p values of the statistical differences between the two groups are reported using 3
models. Model 1 used raw volume measures. Model 2 adjusted for sex, age, TIV, education,
APOE E4 status as confounding factors. Model 3 adjusted for sex, age, education, TIV,
APOE E4 status, income, BMI, diabetes, hypertension, and PM 2.5. All p values were
corrected for multiple comparisons using FDR adjustment.

72
4.3.2 Cognitive and Hearing Scores

Statistical analysis was performed on the cognitive and hearing scores using all half million
participants from the UK Biobank dataset with a maximum level of noise of 93 dBA. There were
no significant statistical differences between the two groups after correction for multiple
comparisons as shown in Table 4.3. Missing values of the cognitive and hearing test measures
were eliminated from the statistical analysis. All half million participants were initially used for
this analysis before eliminating the missing values.

Table 4. 3 Statistical Analysis on the cognitive and hearing scores between the two groups
G55 G70
Measure Subject # Mean (SD) Subject # Mean (SD) P Value
1-FIS 86298 5.97 (2.14) 3343 6.0 (2.13) 0.4

2-MTC 251407 559.6 (117.2) 9728 558.5 (116.0) 0.47

3-MDR 26510 6.47 (1.79) 923 6.39 (1.87) 0.2

4-SRTL 85000 -6.59 (2.07) 3344 -6.53 (2.26) 0.54

5-SRTR 84989 -6.56 (2.07) 3334 -6.63 (1.98) 0.19
Abbreviations: Subject #, number of subjects in group G55 or G70; SD, standard deviation;
FIS, fluid intelligence score; MTC, mean time to correctly identify matches; MDR, maximum
digits remembered; SRTL, speech reception threshold estimate on the left hemisphere;
SRTR, speech reception threshold estimate on the right hemisphere.
The table shows the statistics for 5 different cognitive and hearing scores between G55 and
G70 groups. There were no significant statistical differences between the two groups in any
of the tests. P values were FDR adjusted for multiple comparisons.

73
4.3.3 Association of Noise Exposure with Non-Imaging Phenotypes

Three separate phenome scans were performed using the half million participants from the
UK Biobank dataset in order to increase the power of detecting any association between noise
exposure and non-imaging phenotypes. A variety and large set of 20,622 phenotypes were used,
and most of them are health-related factors. The first model corrected for age, sex, and time spent
at current residential address. Many phenotypes showed strong association with noise exposure
after correction for multiple comparisons. Table 4.4 shows some important health-related
phenotypes using model 1. These measures were mostly related to red blood cells, diabetes, BMI,
inflammation, sleep disturbances, high blood pressure, and increased heart rate. They can be
considered as cardiovascular risk factors. Additionally, many significant associations were related
to air pollution measures, smoking and alcohol consumption, depression, and medications related
to pain killers and lowering cholesterol.
A second phenome scan was performed using model 2 which corrected for air pollution
PM 2.5 measure in addition to age, sex, and time spent at current address. This analysis was
performed to determine whether the significant associations in model 1 were caused by noise
exposure or air pollution since noise and air pollution are strongly coupled. Figure 4.1 shows that
noise and air pollution were strongly related especially after 70 dBA. High noise levels were also
associated with high air pollution level, and the data followed a cubic relationship as shown in
figure 4.1. Many phenotypes lost significance using model 2 especially phenotypes related to red
blood cells including diabetes. Noticeably, some phenotypes gained strong significant association
like white blood cell count which was not significant in model 1. This can indicate that red blood
cells phenotypes including diabetes are mainly affected by air pollution and not necessarily noise
exposure. On the other hand, white blood cells phenotypes are mainly affected by noise pollution
74
and not air pollution. Many vascular and heart problems diagnosed by a doctor remained
significant like high blood pressure, heart attack, stroke, and angina. Feeling of nervousness and
miserableness showed significant association with noise exposure. Medications for lowering
cholesterol, high blood pressure, and headache also showed strong significant associations. Table
4.5 shows some of the main important significant associations.
Additionally, a third model was used which corrected for sleep duration, income, and
education levels in addition to the confounding factors in model 2. Model 3 analysis was performed
to check whether the significant associations were caused solely by lack of sleep. Model 3 also
checked whether the significant noise associations were due to differences in socio-economic
status. Many phenotypes lost significance like high blood pressure, stroke, heart attack, angina,
white blood cell count, neutrophil count, and C-reactive protein. However, many phenotypes
remained significant as shown in table 4.6. Significant associations were still found with
phenotypes related to cholesterol, BMI, systolic and diastolic blood pressure, nervous feelings,
and monocyte and basophil counts. Medications like Rosuvastatin, Pravastatin, Paracetamol,
Nefopam, Warfarin, Amiloride showed significance which are related to cholesterol, pain killers,
high blood pressure, and stroke.

75

Table 4. 4 Significant associations of health-related phenotypes with noise exposure using
model 1
Phenotype P value Beta 95% CI
Glycated Hemoglobin 2.04E-14 0.012 0.0097, 0.0153
Hemoglobin Concentration 3.03E-14 -0.01 -0.012, -0.007
Hematocrit Percentage 6.0E-12 -0.009 -0.012, -0.007
BMI 1.52E-13 0.012 0.0094, 0.015
Diabetes Diagnosed by Doctor 8.17E-8 0.043 0.03, 0.05
Red Blood Cell Count 3.63E-8 -0.008 -0.011, -0.006
Mean Reticulocyte Volume 2.41E-5 0.008 0.006, 0.011
Total Bilirubin 2.99E-8 -0.009 -0.012, -0.007
Chest Pain or Discomfort 3.9E-8 0.026 0.019, 0.034
Pulse Rate 7.03E-5 0.0086 0.005, 0.011
Cystatin C 5.98E-3 0.0068 0.004, 0.009
C-reactive Protein 1.19E-2 0.007 0.004, 0.01
Sleep Duration 8.35E-4 -0.014 -0.019, -0.009
High Blood Pressure 3.32E-2 0.0157 0.009, 0.022
Abbreviations: Beta, Beta regression coefficient; CI, Confidence Interval for Beta.
The table shows some of the significant association between phenotypes and noise exposure
using model 1 which corrected for age, sex, and time spent at current address. P values were
corrected for multiple comparisons using FDR adjustment.

76

Table 4. 5 Significant associations of health-related phenotypes with noise exposure using
model 2
Phenotype P value Beta 95% CI
Cholesterol 3.45E-8 0.008 0.005, 0.011
White blood Cell Count 7.77E-7 0.007 0.004, 0.0109
Neutrophil Count 4.34E-7 0.008 0.004, 0.011
Lymphocyte Percentage 0.0019 0.004 0.001, 0.007
C-reactive Protein 0.00013 0.006 0.002, 0.009
Albumin 0.0005 0.005 0.002, 0.008
Pulse Rate 0.0017 0.008 0.003, 0.013
Sleeplessness/Insomnia 7.8E-12 0.019 0.014, 0.025
Chest Pain or Discomfort 0.0018 0.012 0.004, 0.021
Weight 0.00012 0.005 0.002, 0.007
Nervous Feelings 6.83E-15 0.029 0.021, 0.036
Heart Attack 0.0004 0.036 0.016, 0.057
High Blood Pressure 0.0019 0.011 0.004, 0.018
Stroke 0.022 0.028 0.004, 0.053
Angina 0.0001 0.034 0.016, 0.051
Headache 3.74E-5 0.016 0.008, 0.023
Cholesterol Lowering Medication 0.0009 0.021 0.008, 0.033
Blood Pressure Medication 0.0094 0.014 0.003, 0.025
The table shows some of the significant association between phenotypes and noise exposure
using model 2 which corrected for age, sex, and time spent at current address, air pollution
PM 2.5. P values were corrected for multiple comparisons using FDR adjustment.

77

Table 4. 6 Significant associations of health-related phenotypes with noise exposure using
model 3
Phenotype P value Beta 95% CI
Cholesterol 0.0042 0.014 0.004, 0.024
Body Mass Index 0.0002 0.018 0.008, 0.028
Basal Metabolic Rate 1.74E-5 0.014 0.007, 0.02
Systolic Blood Pressure 0.003 0.014 0.004, 0.024
Diastolic Blood Pressure 0.031 0.011 0.001, 0.02
Vitamin D 6.79E-5 -0.021 -0.03, -0.01
Vitamin B6 0.041 -0.025 -0.05, -0.0009
Total Bilirubin 0.025 -0.011 -0.02, -0.001
Whole Body Fat Mass 2.29E-5 0.013 0.007, 0.019
Weight 2.62E-5 0.018 0.009, 0.027
Nervous Feelings 0.013 0.032 0.006, 0.059
Monocyte Count 0.027 0.011 0.001, 0.021
Basophill Count 0.032 0.02 0.001, 0.039
Platelet Count 0.002 0.015 0.005, 0.025
Platelet Crit 0.006 0.013 0.003, 0.023
Apolipoprotein B 0.024 0.011 0.001, 0.022
The table shows some of the significant association between phenotypes and noise exposure
using model 3 which corrected for age, sex, and time spent at current address, air pollution
PM 2.5, sleep duration, income, and education. P values were corrected for multiple
comparisons using FDR adjustment.

78

Figure 4. 1: The relationship between noise and air pollution in UK Biobank data
This figure shows that air and noise pollution have a cubic relationship in this data. After 70
dBA, the air pollution estimates increased much faster. This means that high noise levels are
also related to high air pollution levels especially after 70 dBA.

79
4.4 Discussion
The studies of environmental noise exposure effects on human brain structural changes are
very scarce and not well understood although there is no longer any doubt that long term noise
pollution causes adverse health effects. Noise exposure has been significantly associated with
cardiovascular risk factors and dementia [39–41,137]. Using the large UK biobank cohort, we
found significant associations of environmental noise exposure with high risk factors of
cardiovascular diseases, cerebrovascular diseases, and dementia independent of air pollution
exposure, sleep disturbance, and socio-economic status. Brain structural differences were not
found between participants who were exposed to noise levels above an average of 70 dBA and
below 55 dBA with a maximum level of 82 dBA of noise exposure independent of many
confounding factors that may contribute to or affect brain atrophy.
Environmental noise exposure has received a growing interest in the scientific research
community to understand its effects on human health, but none of these studies considered the
effects of noise exposure on human brain possibly due to the difficulty of exposing humans to high
levels of noise for a long term. That may explain why most studies which considered the effect of
noise on neural systems were mostly limited to animal experiments. Animal experiments in rats
showed over production of Beta Amyloids (AB) and NFTs [43,139,140,142] which are considered
the early signs in Alzheimer’s disease etiology because the accumulation of these proteins will
eventually cause neurodegeneration and cognitive decline. Noise exposure has been associated
with higher incidence and risk of developing AD and cognitive decline in humans [40–42], but
none of these studies considered brain imaging data and brain structural changes that can be an
indicative of dementia and cognitive decline. In an attempt to investigate the structural changes of
the brain as the result of noise pollution, we considered 11 selected global and regional brain
80
measures that can be affected by AD including the hippocampal and entorhinal regions in both
hemispheres which are the first brain regions affected by AD. We also considered the transverse
superior temporal gyrus in both hemispheres which are responsible for hearing. It was expected to
find some significant structural differences in those 11 brain regions between the G55 and G70
groups because previous research found higher incidence of dementia and cognitive decline with
noise exposure. Additionally, noise levels above a 24 hour average of 70 dBA was reported to
have health and hearing loss risks according to EPA [134]. However, none of these 11 selected
brain regions showed any statistically significant differences between the two noise groups
independent of confounding factors that may contribute to or affect brain atrophy and cognitive
decline. Confounding factors included air pollution, sex, age, education level, income, BMI,
diabetes, hypertension, TIV, and APOE E4 status. We also considered 1286 image derived
phenotypes, and no significant differences were found between the G55 and G70 groups before
and after correction of the same confounding factors. Although the sample size was large which
increased the statistical power to detect any brain structural differences, no differences were found
possibly due to the maximum level of noise. It could mean that noise levels up to 82 dBA may not
cause any obvious brain structural differences, and it may take a long time to cause any brain
changes.
Decrements in cognitive performance are usually the early symptoms of any dementia type.
All half million participants were considered in analyzing the cognitive scores with the elimination
of missing values. The maximum level of noise was around 93 dBA. Significant statistical
differences in cognitive scores were not found between the G55 and G70 groups using the fluid
intelligence score, mean time to correctly identify matches, and maximum digits remembered.
Hearing tests on both hemispheres did not show any significant statistical differences between the
81
two groups. Our results were not in accordance with some research studies which reported
associations between noise exposure and some cognitive scores. For instance, children exposed to
chronic environmental noise have shown poorer cognitive performances and functions that involve
language comprehension, attention, and memory when compared to children living in quieter areas
and studying in quieter schools [163,164]. Some studies also reported an association between noise
and cognitive performance in adults. Weuve et al. showed that noise level was associated with
worse cognitive performance particularly in perceptual speed, but it was not associated with
cognitive decline in a consistent way using participants from the Chicago Health and Aging Project
(CHAP) [40]. The Heinz Nixdorf Recall study using a cohort of German adults also reported that
long term exposure to air pollution and traffic noise were associated with poorer cognitive abilities
related to memory and executive functions [41]. Both Heinz Nixdorf and CHAP studies had a
maximum noise exposure of around 80 dBA which is lower than the maximum noise exposure of
93 dBA in our study. We did not find any significant differences in the three cognitive scores
possibly because it may need chronic and steady long-term exposure to high noise levels beyond
93 dBA to cause any effects. The used three cognitive scores may not cover the whole cognitive
dimension and assess the cognitive abilities associated with noise exposure of the participants in
our study. However, the inverse association of noise and cognitive decline was not consistent with
all cognitive scores as reported in the CHAP study which reported association particularly with
perceptual speed and not necessarily with any other cognitive abilities. Noise induced hearing loss
is also a major public health concern related to environmental noise affecting all age groups. It was
estimated that around 1.3 billion people may be affected by hearing loss in 2010 [165]. It was also
estimated that 5% of the world population suffers from noise induced hearing loss due to the
auditory damage caused by high levels of noise exposure [166]. Hearing scores did not show any
82
significant difference between the G55 and G70 groups in our analysis. It could be that noise
exposure up to 93 dBA does not cause any significant auditory damage.
After correction for age, sex, and time spent at current residential address, strong
associations were found between noise and some health-related phenotypes. Many phenotypes
were related to red blood cells including diabetes. Some cardiovascular risk factors also showed
significant associations with noise like high blood pressure, BMI, and heart rate. Noise pollution
has been associated with cardiovascular diseases, and cardiovascular diseases are associated with
dementia and AD [23,27–30,38,39,167]. Noise exposure was also associated with sleep disruption
in our study which is known to have short- and long-term health consequences. Sleep
fragmentations are linked to oxidative stress, inflammation, hypertension, cardiovascular diseases,
and dementia [136,138,168–170]. C-reactive protein (CRP) was also found to be associated with
noise pollution. CRP is a marker of inflammation, and increased levels in brain and serum were
found to be associated with dementia and AD [171–173]. Many associations were also found
related to cardiovascular medications and diseases.
Noise and air pollution are strongly coupled and usually share common sources like
transportation or industrial sources. In the UK biobank data, air pollution increased in a cubic
relationship with noise after 70 dBA. Model 2 analysis was important to disentangle and separate
the effects of the two exposures to assess whether the associations with noise were mainly due to
noise and not air pollution. Many associations related to cardiovascular risk factors and diseases
remained significant independent of air pollution like hypertension, heart rate, heart attack, stroke,
sleep disruption, C-reactive protein, and angina. Cholesterol gained more significance after
correction for air pollution. Red blood cell phenotypes lost significance after air pollution
adjustment including diabetes. Some phenotypes gained strong significance which were not
83
significantly associated with noise before air pollution adjustment like white blood cell phenotypes
and nervous feelings. This could mean that air pollution has strong effects on red blood cells and
not white blood cells. Ultra-fine air particulate matter like PM 2.5 can enter the blood stream and
cause alterations to the blood cells, and they can even reach the brain [174,175]. It is possible that
the main effect of air pollution on blood cells involved red blood cells from our analysis results.
Strong associations were found with white blood cells phenotypes after air pollution adjustment.
This could mean that noise exposure has strong effects on white blood cells and not red blood
cells. White blood cells are considered a biomarker for inflammation and are related to vascular
endothelial dysfunction [176]. It was found in a recent study that aircraft noise exposure was
associated with white blood cells and endothelial dysfunction in mice [177]. Endothelial
dysfunction is associated with cardiovascular and cerebrovascular diseases [178,179] which can
also lead to dementia and AD. Our study is the first to report association between white blood cells
and noise exposure in humans.
Corrections for sleep duration and socio-economic status were also performed to assess
whether the significant associations with noise were independent of sleep disturbances and socio-
economic status in addition to air pollution. Higher socio-economic status is usually linked to
better lifestyle and access to health care which can be indicative for better health in general. It was
expected that higher income maybe positively correlated with lower noise levels. However, our
data did not show any consistent correlation between noise levels and income levels. Participants
with the highest income level (greater than 100K pounds per year) showed almost similar exposure
to noise levels as participants with the lowest income level (less than 18K pounds per year). Some
studies have shown that there may not be an association between noise and socio-economic status
because popular city centers of large metropolitan areas are attracting people with high socio-
84
economic status despite of high traffic noise [180]. Many significant associations with noise were
attenuated or lost after adjustment for socio-economic status, sleep duration, air pollution, time
spent at current address, age, and sex. C-reactive protein, high blood pressure, stroke, heart attack,
angina, white blood cell count, and neutrophil count lost significance. However, many
cardiovascular risk factors remained significant like cholesterol and BMI. Systolic and diastolic
blood pressures and nervous feelings remained significant. Additionally, some white blood cell
types remained significant like monocyte count and basophil count which indicated that white
blood cells were still associated with noise exposure. Vitamins D and B6 were also found to be
associated with environmental noise. Previous research studies reported that both vitamins were
associated with cardiovascular diseases, dementia, and Alzheimer’s disease [181–184].
Apolipoprotein B (APOB) was also associated with environmental noise in our study, and recent
studies have reported that APOB can be associated with cholesterol, cardiovascular diseases, and
dementia especially Alzheimer’s disease [185–187]. Basal metabolic rate and medications like
Rosuvastatin, Pravastatin, Paracetamol, Nefopam, Warfarin, and Amiloride showed significance
independent of air pollution, sleep disturbances, and socio-economic status. These medications
are mostly related to cholesterol, pain killers, high blood pressure, and stroke.
The most plausible explanation for the potential mechanism behind environmental noise
effects on human health can be due to the emotional stress, discomfort, and nervous feelings that
can arouse the endocrine system and the release of many stress hormones [163,188]. Stress
hormones have been linked to cardiovascular disease risk factors and dementia especially
AD[189,190]. Noise exposure can also arouse the autonomic nervous system through the
interactions with the central auditory system which may cause unconscious stress even during
sleeping and at low noise levels [163,191]. The main strong point in this study was the large sample
85
size which increased the statistical power to detect any significant differences or associations.
However, one main limitation was the noise level range with a maximum level around 82 dBA for
the neuroimaging analysis which may not be high enough to cause and detect any brain structural
differences between low and relatively higher noise levels.

86
Chapter 5: Summary and Future Directions

This research work was mainly focused to further understand Alzheimer’s disease and its
progression. AD is a multifactorial disease with many complex etiologies that are not fully
understood. It is becoming a major health concern worldwide especially with the growing numbers
of the aging population. AD in its major LOAD form can be caused by many risk factors including
cardiovascular diseases, environmental pollution, lifestyle, genetics, and racial differences. Many
of these risk factors are controllable and modifiable at a personal and global governmental level.
Aging is the number one risk factor for AD, but normal aging does not always associate with
dementia. Many people can have normal aging without developing AD and its symptoms although
they may have many risk factors including genetics like APOE e4 allele. Living a healthy lifestyle
can be a good way to improve the cognitive reserve and maintain a healthy brain. Healthy lifestyle
means having a good healthy diet and exercise mentally and physically. Environmental pollution
is also considered a risk factor since it contributes to cardiovascular diseases and dementia
especially AD. However, environmental noise and air pollution are modifiable and controllable
risk factors. Everyone can avoid noisy or air polluted environments. Governmental regulations and
laws could limit the pollution levels and human exposure to high levels of pollution. The number
of deaths from Alzheimer’s dementia increased by 89% between 2000 and 2014, and the number
of Americans living with AD is projected to increase to 13.8 million by 2050. The number of AD
cases is also expected to increase worldwide to around 150 million by 2050. This increase in AD
numbers is not only due to genetics, but it is definitely associated with less healthy lifestyle and
environmental toxins which can be controlled. The familial AD is mainly caused by mutations in
genes that contribute to the accumulation of AB plaques and NFTs, but it is a rare form of AD.
87
Most AD cases are considered late-onset AD (LOAD) which can be controlled to slow the
progression of AD biological manifestation and symptoms.
Chapter two was focused on understanding the gene expressions and transcriptional
changes as the result of AD. Gene expression data was used from multiple brain regions that are
part of the cognitive system affected by AD. Meta-analysis studies on gene expression data have
reported some genes that were dysregulated in same direction in multiple AD brain regions that
can include both gray and white matter. This made more sense to integrate gene expression data
from multiple brain regions in addition to increasing the statistical power by taking all the samples
instead of focusing on one brain region or taking the average of the samples. Many dysregulated
genes were identified in this study that were never reported before. The dysregulated genes in this
study can be defined as the AD associated genes which are differentially expressed and reflect
statistically significant transcriptional changes in the brain of AD subjects compared to healthy
controls. These AD associated genes and their underlying pathways can help further understand
the pathogenesis of the disease, and they contribute to AD progression. It is crucial to mention that
previous research studies have shown that transcriptional changes for some genes may precede
any neuropathy, and transcriptional changes can follow up-down or down-up changes. This means
that some genes can be downregulated or upregulated before or after any AD symptoms or
neuropathy. The dysregulated genes in our analysis were mainly related to axon transport and
synaptic transmission which affect the neuronal connectivity between cognitive systems involved
in AD pathogenesis. These dysregulated genes can help in understanding the etiologies underlying
AD progression and open new ways to further explore AD treatment and early diagnosis.
Enrichment analyses revealed biological processes and pathways that are related to structural
constituents and organization of the axons and synapses in addition to lipoprotein metabolism and
88
neurotransmitter release cycle. Future research based on these results can possibly try to
understand how these dysregulated genes contribute to the progression of AD. Research can also
be done on how these genes can be used for early diagnosis of the disease especially that
transcriptional gene changes may precede any AD neuropathy or symptoms. Some research work
can target how to explore AD treatment methods based on these dysregulated genes.
Chapter three studied the ethno-racial disparities in Alzheimer’s disease. We showed that
disparities also exist in brain biomarkers and not just in the incidence and prevalence rates. It is
often estimated that there are higher rates of dementia among ethnic minorities. AA are about two
times higher than NHW to contract AD in their life span, and similar estimates often suggest that
Hispanics are 1.5 times higher than NHW to have AD. There are not many studies on ethno-racial
disparities using brain imaging data and using non-mixed AD participants. In this study, non-
mixed AD participants were mainly used from three ethnic groups to eliminate any factors and
mixed etiologies that may contribute to cognitive decline and brain atrophy besides AD. The
subjects were mainly AD subjects without cardiovascular, cerebrovascular, or psychological
disorders to understand and characterize the biological expression and manifestation of AD in
different ethnic groups. The brain measures were also adjusted for many main confounding factors
and variables that may contribute to some inter-subject variability in brain atrophy. Although it is
not possible to include or know all factors that affect the cognitive decline and brain atrophy, the
main factors were used. Most studies have not considered racial differences in the biological
expression of AD. Our study might be the first to investigate the ethno-racial AD disparities in
brain biomarkers using non-mixed AD participants. Significant differences were found in many
brain measures which suggested the existence of ethno-racial differences in AD expression and
progression. Brain atrophy and neurodegeneration were more similar between AA and HIS than
89
NHW AD participants. APOE e4 allele showed significant effects on NHW AD participants only.
This study revealed and confirmed that ethno-racial differences in brain AD biomarkers exist, and
future AD studies should take racial differences into consideration. Future research directions
based on these results may try to understand the main causes of these racial differences in brain
biomarkers. Genetics is considered the main first reason for these ethno-racial disparities, but it
may not be the only one. There might be some other causes behind these differences like access to
health care, environmental toxins, or cultural differences. Genetics is the number one reason after
matching these AD subjects and correcting for many confounding factors. GWAS studies may be
used to further explore this topic and possibly identify and detect some SNPs or genetic variants
that can be different between the three ethnic groups and contribute to disparities in AD onset and
progression.
Chapter four focused on studying the effects of environmental noise exposure on human
health and dementia. The effects of air pollution have been investigated in more depth than noise
pollution. Many studies on neuroimaging using magnetic resonance imaging scans have reported
an association of increased air pollution exposure with brain atrophy and smaller brain structures.
Animal studies have shown that ultrafine air pollutants can reach the brain through the olfactory
neuronal pathway and act as neurotoxins and cause neuroinflammation which may lead to
dementia and cognitive decline. Neuroimaging studies on noise exposure and dementia in humans
are very scarce and not well investigated. Studies on the effects of noise exposure on brain atrophy
and neuropathological brain changes have been limited to animal experiments. There is no more
doubt that long term exposure to high levels of noise has adverse health effects including
cardiovascular diseases and dementia especially AD. However, the extent of auditory and health
damage is not very well understood especially to the human brain. Some studies reported higher
90
incidence rates and cognitive decline in humans with higher exposure of noise, but none of these
studies considered any neuroimaging data and the effects of noise on human brain. Using the large
sample size of neuroimaging data from UK biobank, significant differences in structural brain
regions and even some cognitive and hearing scores were not found in our analysis between
relatively low and high noise levels independent of many confounding factors that may contribute
to cognitive decline. However, associations were found between noise and some cardiovascular
and dementia risk factors independent of air pollution, sleep disturbances, or socio-economic
status. White blood cells phenotypes showed significant associations with noise which are usually
linked to inflammation and endothelial dysfunction that can cause many health effects including
cardiovascular, cerebrovascular, and dementia type diseases. Additionally, noise was associated
with vitamin D, vitamin B6, and Apolipoprotein B (APOB) which were recently reported to be
associated with cardiovascular diseases and dementia especially AD. Noise exposure effects can
be explained by the arousal of the endocrine system and the release of stress hormones that are
linked to many detrimental health effects including Alzheimer’s disease. Noise exposure is a
modifiable risk factor that can be controlled through governmental regulations and even at a
personal level. Reducing exposure to high level of noise pollution is important to human health.
Future research directions based on these results can further investigate the relationship of noise
and air pollution with Alzheimer’s disease possibly by studying gene expression and
transcriptional changes as a result of environmental pollution exposure. Gene environment
interactions can also be studied to identify different effects of environmental pollution in subjects
with different genotypes or different ethno-racial background.

91
Papers and Abstracts

Published:
1. Arzouni N, Matloff W, Zhao L, Ning K, Toga AW (2020) Identification of Dysregulated
Genes for Late-Onset Alzheimer’s Disease Using Gene Expression Data in Brain. J
Alzheimers Dis Parkinsonism 10(6): 498.
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7717689/

In Preparation:

2. Nibal Arzouni, Lu Zhao, William Matloff, Yonggang Shi, Arthur Toga. Ethno-racial
disparities in brain biomarkers for late-onset Alzheimer’s disease.

3. Nibal Arzouni, Lu Zhao, William Matloff, Arthur Toga. The effects of environmental
noise exposure on human health and brain.

92
References

[1] Alzheimer’s Association (2021) Alzheimer’s disease facts and figures. 17, 1–108.
[2] Porteri C, Albanese E, Scerri C, Carrillo M, Snyder H, Martensson B (2017) The biomarker-
based diagnosis of Alzheimer’s disease. 1-ethical and societal issues. Neurobiol Aging 52,
132–140.
[3] Betram L, Tanzi R (2008) Thirty years of Alzheimer’s disease genetics: the implications of
systematic meta-analyses . Nat Rev Neurosci 9, 768–778.
[4] Lahiri DK, Maloney B, Basha MR, Ge YW, Zawia NH (2007) How and when
environmental agents and dietary factors affect the course of Alzheimer’s disease. Curr
Alzheimer Res 4, 219–228.
[5] Ya ping T, Elliot G (2003) Genetic Studies in AD. Dialogues Clin Neurosci 5, 17–26.
[6] Kang J, Lemaire H, UnterBeck A, Salbaum J, Masters C, Grzeschik K, et al. (1987) The
precursor of Alzheimer’s disease amyloid A4 protein resembles a cell-surface receptor.
Nature 325, 733–736.
[7] Selkoe D, Bell D, Podlisny M, Price D, Cork L (1987) Conservation of brain amyloid
proteins in aged mammals and humans with Alzheimer’s disease. Science 235, 873–877.
[8] Yankner B (2000) The pathogenesis of Alzheimer’s disease. Is amyloid β-protein the
beginning or the end? Ann N Y Acad Sci 924, 26–28.
[9] Talwar P, Sinha J, Grover S, Rawat C, Kushwaha S, Agarwal R, et al. (2016) Dissecting
complex and multifactorial nature of Alzheimer’s disease pathogenesis: a clinical, genomic,
and systems biology perspective. Mol Neurobiol 53, 4833–64.
[10] Yamagata K, Urakami K, Ikeda K, Ji Y, Adachi Y, Arai H, et al. (2001) High expression of
apolipoprotein E mRNA in the brains with sporadic Alzheimer’s disease. Dement Geriatr
Cogn Disord 12, 57–62.
[11] Haan M, Mayeda E (2010) Apolipoprotein E Genotype and Cardiovascular Diseases in the
Elderly. Curr Cardio Risk Rep 4, 361–368.
[12] Vincent G, Velkoff V (2010) The Next Four Decades, The Older Population in the United
States: 2010 to 2050. Current Population Reports, US Census Bureau P25-1138.
[13] (2019) United States Census Bureau Quick Facts. US Census Bureau web site Available at:
https://www.census.gov/quickfacts/fact/table/US/PST045219.
[14] (2021) The National Alzheimer’s Coordinating Center (NACC).
https://naccdata.org/requesting-data/data-summary/uds.
[15] Tang M, Cross P, Andrews H, Jacobs D, Small S, Bell K (2001) Incidence of AD in African-
Americans, Caribbean Hispanics, and Caucasians in northern Manhattan. Neurology 56, 49–
56.
[16] Mayeda E, Glymour M, Quesenberry C, Whitmer R (2016) Inequalities in dementia
incidence between six racial and ethnic groups over 14 years. Alzheimers Dement 12, 216–
224.
[17] Manly J, Tourdaji P, Tang M, Stern Y (2003) Literacy and memory decline among
ethnically diverse elders. J Clin Exp Neuropsychol 25, 680–690.
[18] Glymour M, Manly J (2008) Lifecourse social conditions and racial and ethnic patterns of
cognitive aging. Neuropsychology Review 18, 223–254.
93
[19] Gray H, Jimenez D, Cucciare M, Tong H, Gallagher-Thompson D (2009) Ethnic differences
in beliefs regarding Alzheimer disease among dementia family caregivers. Am J Geriatr
Psychiatry 17, 925–933.
[20] Gruber-Baldini A, Stuart B, Zuckerman I, Simoni-Wastila L, Miller R (2007) Treatment of
dementia in community-dwelling and institutionalized Medicare beneficiaries.". J AM
Geriatr Soc 55, 1508–1516.
[21] Vasefi M, Ghaboolian-Zare E, Abedelwahab H, Osu A (2020) Environmental toxins and
Alzheimer’s disease progression. Neurochemistry International 141,.
[22] Chin A, Negash S, Hamilton R (2011) Diversity and Disparity in Dementia: The Impact of
Ethnoracial Differences in Alzheimer’s Disease. Alzheimer Dis Assoc Disord 25, 187–195.
[23] Newman A, Fitzpatrick A, Lopez O, Jackson S, Lyketsos C, Jagust W, et al. (2005)
Dementia and Alzheimer’s disease incidence in relationship to cardiovascular disease in the
Cardiovascular Health Study cohort. J AM Geriatr Soc 53, 1101–7.
[24] Luchsinger J, Reitz C, Honig L, Tang M, Shea S, Mayeux R (2005) Aggregation of vascular
risk factors and risk of incident Alzheimer disease. Neurology 65, 545–551.
[25] Paul K, Haan M, Mayeda E, et al. (2019) Ambient Air Pollution, Noise, and Late-Life
Cognitive Decline and Dementia Risk. Annu Rev Public Health 203–220.
[26] Brook R, Franklin B, Cascio W, et al. (2014) Air Pollution and Cardiovascular Disease.
Circulation 2655–2671.
[27] Munzel T, Schmidt F, Steven S, et al. (2018) Environmental Noise and the Cardiovascular
System. J AM Coll Cardiol 688–97.
[28] Stampfer MJ (2006) Cardiovascular disease and Alzheimer’s disease: common links. J
Intern Med 260, 211–23.
[29] Fillit H, Nash D, Rundek T, et al. (2008) Cardiovascular risk factors and dementia. AM J
Geriatr Pharmacother 6, 100–18.
[30] Fitzpatrick A, Kuller L, Ives D, Lopez O, Jagust W, Breitner J, et al. (2004) Incidence and
prevalence of dementia in the Cardiovascular Health Study. J AM Geriatr Soc 52, 195–204.
[31] Block M, Garciduenas L (2009) Air pollution: mechanisms of neuroinflammation and CNS
disease. Trends Neurosci 32, 506–16.
[32] Elder A, Gelein R, Silva V, et al. (2006) Translocation of inhaled ultrafine manganese oxide
particles to the central nervous system. Environ Health perspect 114, 1172–8.
[33] Oberdorster G, Sharp Z, Atudorei V, et al. (2004) Translocation of inhaled ultrafine particles
to the brain. Inhal Toxicol 16, 437–45.
[34] Chen J-C, Wang X, Wellenius G, et al. (2015) Ambient air pollution and neurotoxicity on
brain structure: Evidence from women’s health initiative memory study. Ann Neurol 78,
466–76.
[35] Wilker E, Preis S, Beiser A, et al. (2015) Long-term exposure to fine particulate matter,
residential proximity to major roads and measures of brain structure. Stroke 46, 1161–66.
[36] Power M, Lamichhane A, Liao D, et al. (2018) The Association of Long-Term Exposure to
Particulate Matter Air Pollution with Brain MRI Findings: The ARIC Study. Environ Health
perspect 126,.
[37] Casanova R, Wang X, Reyes J, et al. (2016) A Voxel-Based Morphometry Study Reveals
Local Brain Structural Alterations Associated with Ambient Fine Particles in Older Women.
Front Hum Neurosci 495.
94
[38] Kempen E, Casas M, Pershagen G, et al. (2018) WHO Environmental Noise Guidelines for
the European Region: A Systematic Review on Environmental Noise and Cardiovascular
and Metabolic Effects: A Summary. Int J Environ Res Public Health 15, 379.
[39] Zijlema W, Cai Y, Mbatchou S, et al. (2016) Road traffic noise, blood pressure and heart
rate: Pooled analyses of harmonized data from 88,336 participants. Environmental Research
151, 804–813.
[40] Weuve J, D’Souza J, Beck T, et al. (2021) Long-term community noise exposure in relation
to dementia, cognition, and cognitive decline in older adults. Alzheimers Dement 17, 525–
533.
[41] Tzivian L, Dlugaj M, Winkler A, et al. (2016) Long-term air pollution and traffic noise
exposures and cognitive function: A cross-sectional analysis of the Heinz Nixdorf Recall
study. J Toxicol Environ Health A 79, 1057–1069.
[42] Cantuaria ML, waldorff FB, Wermuth L, et al. (2021) Residential exposure to transportation
noise in Denmark and incidence of dementia: national cohort study. BMJ 374,.
[43] Cui B, li K, Gai Z, et al. (2015) Chronic Noise Exposure Acts Cumulatively to Exacerbate
Alzheimer’s Disease-Like Amyloid-β Pathology and Neuroinflammation in the Rat
Hippocampus. Scientific Reports 12943.
[44] Pearson H, Peers C (2006) Physiological roles for amyloid β peptides. J Physiol 575, 5–10.
[45] Hardy J, Selkoe DJ (2002) The amyloid hypothesis of Alzheimer’s disease: progress and
problems on the road to therapeutics. Science 297, 353–356.
[46] Gu L, Guo Z (2013) Alzheimer’s Aβ42 and Aβ40 peptides form interlaced amyloid fibrils.
J Neurochem 126, 305–11.
[47] Spires-Jones T, Hyman B (2014) The intersection of amyloid beta and tau at synapses in
Alzheimer’s disease. Neuron 82, 756–771.
[48] Jackson J, Jambrina E, Li J, et al. (2019) Targeting the Synapse in Alzheimer’s Disease.
Front Neurosci.
[49] Yamamoto A, Shin R, Hasegawa K, Naiki H, Sato H, Yoshimasu F, et al. (2002) Iron (III)
induces aggregation of hyperphosphorylated tau and its reduction to iron (II) reverses the
aggregation: implications in the formation of neurofibrillary tangles of Alzheimer’s disease.
J Neurochem 82, 1137–1147.
[50] Braak H, Braak E (1991) Neuropathological staging of Alzheimer-related changes. Acta
Neuropathol 82, 239–259.
[51] Hardy J, Duff K, Hardy K, Perez-Tur J, Hutton M (1998) Genetic dissection of Alzheimer’s
disease and related dementias: amyloid and its relationship to tau. Nature Neurosci 1, 355–
358.
[52] Jack CJ, Knopman D, Jaqust W, Petersen R, Weiner M, Aisen P, et al (2013) Tracking
pathophysiological processes in Alzheimer’s disease: an updated hypothetical model of
dynamic biomarkers. Lancet Neurol 12, 207–216.
[53] Braak H, Del Tredici K (2011) The pathological process underlying Alzheimer’s disease in
individuals under thirty. Acta Neuropathol 121, 171–81.
[54] Zhang B, Gaiteri C, Bodea L, Wang Z, McElwee J, Podtelezhnikov A, et al (2013)
Integrated systems approach identifies genetic nodes and networks in late-onset
Alzheimer’s disease. Cell 153, 707–720.
[55] Patel H, Dobson RJB, Newhouse SJ (2019) A Meta-Analysis of Alzheimer’s Disease Brain
Transcriptomic Data. Journal of Alzheimer’s Disease 68, no. 4, 1635–1656.
95
[56] Annese A, Manzari C, Lionetti C, Picardi E, Horner D, Chiara M, et al. (2018) Whole
transcriptome profiling of Late-Onset Alzheimer’s Disease patients provides insights into
the molecular changes involved in the disease. Sci Rep 8, 4282.
[57] Miller J, Guillozet-Bongaarts A, Gibbons L, Postupna N, Renz A, Beller A, et al. (2017)
Neuropathological and transcriptomic characteristics of the aged brain. eLife 6, e31126.
[58] Smyth GK (2004) Linear models and empirical bayes methods for assessing differential
expression in microarray experiments. Statistical applications in genetics and molecular
biology 3, Article 3. 1-25.
[59] Smyth GK (2005) Limma: linear models for microarray data. In Bioinformatics and
Computational Biology Solutions using R and Bioconductor. Edited by Gentleman R, Carey
V, Dudoit S, Irizarry R, Huber W New York: Springer 397–420.
[60] Ritchie M, Phipson B, Wu D, Hu Y, Law C, Shi W, et al. (2015) Limma powers differential
expression analyses for RNA-sequencing and microarray studies. Nucleic acids research
43, e47.
[61] Lonnstedt I, Speed T (2002) Replicated microarray data. Statistica Sinica 12, 31–46.
[62] Law C, Chen Y, Shi W, Smyth G (2014) Voom: precision weights unlock linear model
analysis tools for RNA-seq read counts. Genome Biology 15, R29.
[63] Bates D, Mächler M, Bolker B, Walker S (2015) Fitting Linear Mixed-Effects Models Using
lme4. Journal of Statistical Software 67,.
[64] Ashburner M, Ball CA, Blake JA, Botstein D, Butler H, Cherry JM, et al. (2000) Gene
ontology: tool for the unification of biology. Nat Genet 25, 25–29.
[65] The Gene Ontology Consortium (2019) The Gene Ontology Resource: 20 years and still
GOing strong. Nucleic Acids Res 47, D330–D338.
[66] Huang D, Sherman B, Lempicki R (2009) Bioinformatics enrichment tools: paths toward
the comprehensive functional analysis of large gene lists. Nucleic Acids Res 7, 1–13.
[67] Huang D, Sherman B, Lempicki R (2009) Systematic and integrative analysis of large gene
lists using DAVID Bioinformatics Resources. Nature Protoc 4, 44–57.
[68] Subramanian A, Tamayo P, Mootha VK, Mukherjee S, Ebert BL, Gillete MA, et al. (2005)
Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide
expression profiles. Proc Natl Acad Sci USA 102, 15545–15550.
[69] Fabregat A, Jupe S, Matthews L, Sidiropoulos K, Gillespie M, Garapati P, et al. (2018) The
reactome Pathway Knowledgebase. Nucleic Acids Res 46, D649–D655.
[70] Cortes C, Vapnik V (1995) Support-Vector Networks. Machine Learning 20, 273–297.
[71] Hokama M, Oka S, Ninomiya T, Honda H, Sasaki K, Iwaki T, et al. (2014) Altered
Expression of Diabetes-Related Genes in Alzheimer’s Disease Brains: The Hisayama Study.
Cerebral Cortex 24, 2476–2488.
[72] Emilsson L, Saetre P, Jazin E (2006) Alzheimer’s disease: mRNA expression profiles of
multiple patients show alterations of genes involved with calcium signaling. Neurobiology
of Disease 21, 618–625.
[73] Furuya T, Silva P, Payao S, Bertolucci P, Rasmussen L, De Labio R, et al. (2012) Analysis
of SNAP25 mRNA expression and promoter DNA methylation in brain areas of
Alzheimer’s Disease patients. Neuroscience 220, 41–46.
[74] Braak H, Braak E (1996) Development of Alzheimer-related neurofibrillary changes in the
neocortex inversely recapitulates cortical myelogenesis. Acta Neuropathol 92, 197–201.
[75] Hyman B, Van Hoesen G, Damasio A, Barnes C (1984) Alzheimer’s disease: cell-specific
pathology isolates the hippocampal formation. Science 225, 1168–1170.
96
[76] Shannon P, Markiel A, Ozier O, Baliga N, Wang J, Ramage D, et al. (2003) Cytoscape: a
software environment for integrated models of biomolecular interaction networks. Genome
Research 13, 2498–2504.
[77] Galecki A, Burzykowski T (2013) Linear Mixed-Effect Models Using R, Springer.
[78] Wang L, Liu Z-P (2019) Detecting Diagnostic Biomarkers of Alzheimer’s Disease by
Integrating Gene Expression Data in Six Brain Regions. Front Genet 10, 157.
[79] Sherva R, Baldwin C, Inzelberg R, Vardarajan B, Cupples L, Lunetta K, et al. (2011)
Identification of novel candidate genes for Alzheimer’s disease by autozygosity mapping
using genome wide SNP data. Journal of Alzheimer’s Disease 23, 349–359.
[80] Petersen M, Kyu Ryu J, Akassoglou K (2018) Fibrinogen in neurological diseases:
mechanisms, imaging and therapeutics. Nat Rev Neurosci 19, 283–301.
[81] Alvarez V, Alvarez R, Lahoz C, Martinez C, Peña J, Guisasola L, et al. (1999) Association
between an alpha(2) macroglobulin DNA polymorphism and late-onset Alzheimer’s
disease. Biochem Biophys Res Commun 264, 48–50.
[82] Bruno E, Quattrocchi G, Nicoletti A, Le Pira F, Maci T, Mostile G, et al. (2010) Lack of
interaction between LRP1 and A2M polymorphisms for the risk of Alzheimer disease.
Neurosci Lett 482, 112–116.
[83] Mariani E, Seripa D, Ingegni T, Nocentini G, Ercolani S, Cherubini A, et al. (2006)
Interaction of CTSD and A2M polymorphisms in the risk for Alzheimer’s disease. J Neurol
Sci 247, 187–191.
[84] Huang R, Huang J, Cathcart H, Smith S, Poduslo S (2007) Genetic variants in brain-derived
neurotrophic factor associated with Alzheimer’s disease. J Med Genet 44, e66.
[85] Smith M, Rudnicka-Nawrot M, Richey P, Praprotnik D, Mulvihill P, Miller CA, et al. (1995)
Carbonyl-related posttranslational modification of neurofilament protein in the
neurofibrillary pathology of Alzheimer’s disease. j Neurochem 64, 2660–2666.
[86] Al-Chalabi A, Miller C (2003) Neurofilaments and neurological disease. BioEssays 25,
346–355.
[87] O’Day D, Eshak K, Myre M (2015) Calmodulin Binding Proteins and Alzheimer’s Disease.
J Alzheimer’s Disease 46, 553–569.
[88] Francis P (2005) The Interplay of Neurotranmitters in Alzheimer’s Disease. CNS Spectr 10,
6–9.
[89] Guerini FR, Agliardi C, Sironi M, Arosio B, Calabrese E, Zanzottera M, et al. (2014)
Possible association between SNAP-25 single nucleotide polymorphisms and alterations of
categorical fluency and functional MRI parameters in Alzheimer’s disease. J Alzheimer’s
Disease 42, 1015–1028.
[90] Dong Y, Zhang Y, Kang W, Wang G, Chen H, Higashimori A, et al. (2019) VSTM2A
suppresses colorectal cancer and antagonizes Wnt signaling receptor LRP6. Theranostics 9,
6517–6531.
[91] Blalock E, Geddes J, Chen KC, Porter N, Markesbery W, Landfield P (2004) Incipient
Alzheimer’s disease: Microarray correlation analyses reveal major transcriptional and tumor
suppressor responses. PNAS 101, 2173–2178.
[92] Fowler CJ, Cowburn RF, Garlind A, Winblad B, O’Neill C (1995) Disturbances in signal
transduction mechanisms in Alzheimer’s disease. Molecular and Cellular Biochemistry
149, 287–292.
97
[93] Zachariou V, Renthal W, Ring R, Rahman Z, Psifogeorgou K, Howland D, et al. (2010)
Brain Region Specific Actions of RGS4 Oppose Morphine Reward and Dependence but
Promote Analgesia. Biol Psychiatry 67, 761–769.
[94] Adie E, Adams R, Evans K, Porteous D, Pickard B (2005) Speeding disease gene discovery
by sequence based candidate prioritization. BMC Bioinformatics 6, 55.
[95] Huang X, Liu H, Li X, Guan L, Li J, Tellier L, et al. (2018) Revealing Alzheimer’s disease
genes spectrum in the whole-genome by machine learning. BMC Neurology 18, 5.
[96] Scheff S, Price D, Schmitt F, Dekosky S, Mufson E (2007) Synaptic alterations in CA1 in
mild Alzheimer disease and mild cognitive impairment. Neurology 68, 1501–1508.
[97] Hooijmans C, Kiliaan A (2008) Fatty acids, lipid metabolism and Alzheimer pathology.
European Journal of Pharmacology 585, 176–196.
[98] Strimbu K, Tavel JA (2010) What are Biomarkers? Curr Opin HIV AIDS 5, 463–466.
[99] Bossers K, Wirz K, Meerhoff G, Essing A, Dongen J, Houba P, et al. (2010) Concerted
changes in transcripts in the prefrontal cortex precede neuropathology in Alzheimer’s
disease. Brain 133, 3699–3723.
[100] Chawla N, Bowyer K, Hall L, Kegelmeyer W (2002) SMOTE: Synthetic Minority over-
sampling technique. Journal of Artificial Intelligence Research 16, 321–357.
[101] Harwood D, Ownby R (2000) Ethnicity and dementia. Current Psychiatry Reports 2, 40–
45.
[102] Gurland B, Wilder D, Lantigua R, Stern Y, Chen J, Killeffer E, et al. (1999) Rates of
dementia in three ethnoracial groups. International Journal of Geriatric Psychiatry 14, 481–
493.
[103] Mehta K, Yaffe K, Perez-Stable E, Stewart A, Barnes D, Kurland B, et al. (2008)
Race/ethnic differences in AD survival in US Alzheimer’s Disease Centers. Neurology 70,
1163–1170.
[104] Fillenbaum G, Heyman A, Huber M, Woodbury M, Leiss J, Schmader K (1998) The
prevalence and 3-year incidence of dementia in older Black and White community residents.
J Clin Epidemiol 51, 587–595.
[105] Morris J, Schindler S, McCue L, Moulder K, Benzinger T, Cruchaga C, et al. (2019)
Assessment of Racial Disparities in Biomarkers for Alzheimer Disease. JAMA Neurol 76,
264–273.
[106] McDonough I (2017) Beta-amyloid and Cortical Thickness Reveal Racial Disparities in
Preclinical Alzheimer’s Disease. NeuroImage: Clinical 16, 659–667.
[107] Garrett S, McDaniel D, Obideen M, Trammell A, Shaw L, Goldstein F, et al. (2019) Racial
Disparity in Cerebrospinal Fluid Amyloid and Tau Biomarkers and Associated Cutoffs for
Mild Cognitive Impairment. JAMA Netw Open 2,.
[108] Morris J (1993) The Clinical Dementia Rating (CDR): current version and scoring rules.
Neurology 43, 2412–2414.
[109] Aljabar P, Heckemann R, Hammers A, Hajnal J, Rueckert D (2009) Multi-atlas based
segmentation of brain images: atlas selection and its effect on accuracy. Neuroimage 46,
726–38.
[110] Fletcher E, Carmichael O, DeCarli C (2012) MRI non-uniformity correction through
interleaved bias estimation and B-spline deformation with a template. 2012 Annual
International Conference of the IEEE Engineering in Medicine and Biology Society 106–
109.
98
[111] Boccardi M, Bocchetta M, Apostolova L, Barnes J, Bartzokis G, Corbetta G, et al. (2015)
Delphi definition of the EADC-ADNI Harmonized Protocol for hippocampal segmentation
on magnetic resonance. Alzheimers Dement 11, 126–38.
[112] Desikan R, Segonne F, Fischl B, Quinn B, Dickerson B, Blacker D, et al. (2006) An
automated labeling system for subdividing the human cerebral cortex on MRI scans into
gyral based regions of interest. Neuroimage 31, 968–80.
[113] Alfaro-Almagro F, McCarthy P, Afyouni S, Andersson J, Bastiani M, Miller K, et al. (2021)
Confound modelling in UK Biobank brain imaging. NeuroImage 224,.
[114] Smith S, Nichols T (2018) Statistical Challenges in “Big Data” Human Neuroimaging.
Neuron 97, 263–268.
[115] Benjamini Y, Hochberg Y (1995) Controlling the False Discovery Rate: A Practical and
Powerful Approach to Multiple Testing. Journal of the Royal Statistical Society 57, 289–
300.
[116] Schoenhofen Sharp E, Gatz M (2011) The Relationship between Education and Dementia
An Updated Systematic Review. Alzheimer Dis Assoc Disord 25, 289–304.
[117] Beam C, Kaneshiro C, Jang J, Reynolds C, Pedersen N, Gatz M (2018) Differences Between
Women and Men in Incidence Rates of Dementia and Alzheimer’s Disease. J Alzheimer’s
Disease 64, 1077–1083.
[118] Scahill R, Frost C, Jenkins R, Whitwell J, Rossor M, Fox N (2003) A Longitudinal Study
of Brain Volume Changes in Normal Aging Using Serial Registered Magnetic Resonance
Imaging. Arch Neurol 60, 989–994.
[119] Fiford C, Ridgway G, Cash D, Modat M, Nicholas J, Manning E, et al. (2018) Patterns of
progressive atrophy vary with age in Alzheimer’s disease patients. Neurobiol Aging 63, 22–
32.
[120] Pini L, Pievani M, Bocchetta M, Altomare D, Bosco P, Cavedo E, et al. (2016) Brain atrophy
in Alzheimer’s Disease and aging. Ageing Res Rev 30, 25–48.
[121] Debette S, Markus H (2010) The clinical importance of white matter hyperintensities on
brain magnetic resonance imaging: systematic review and meta-analysis. BMJ 341,.
[122] Filon J, Inorcia A, Sue L, Arreola E, Wilson J, Davis K, et al. (2016) Gender Differences in
Alzheimer Disease: Brain Atrophy, Histopathology Burden, and Cognition. Journal of
Neuropath & Exp Neurology 75, 748–754.
[123] Laws K, Irvine K, Gale T (2018) Sex differences in Alzheimer’s disease. Curr Opin
Psychiatry 31, 133–139.
[124] Li R, Singh M (2014) Sex differences in cognitive impairment and Alzheimer’s disease.
Front Neuroendocrinol 35, 385–403.
[125] Sanchis-Segura C, Ibanez-Gual M, Adrian-Ventura J, Aguirre N, Gomez-Cruz A, Avila C,
et al. (2019) Sex differences in gray matter volume: how many and how large are they
really? Biology of Sex Differences 10,.
[126] Chen X, Sachdev P, Wen W, Anstey K (2007) Sex differences in regional gray matter in
healthy individuals aged 44–48 years: A voxel-based morphometric study. NeuroImage 36,
691–699.
[127] Perlaki G, Orsi G, Plozer E, Altbacker A, Darnai G, Nagy S, et al. (2014) Are there any
gender differences in the hippocampus volume afterhead-size correction? A volumetric and
voxel-basedmorphometric study. Neurosci Lett 570, 119–23.
[128] Yamazaki Y, Zhao N, Caulfield T, Liu C, Bu G (2019) Apolipoprotein E and Alzheimer
disease: pathobiology and targeting strategies. Nature Reviews Neurology 15, 501–518.
99
[129] Farrer L, Cupples L, Haines J, Hyman B, Kukull W, Mayeux R (1997) Effects of Age, Sex,
and Ethnicity on the Association Between Apolipoprotein E Genotype and Alzheimer
Disease. JAMA 278, 1349–1356.
[130] Saeed U, Mirza S, Macintosh B, Herrmann N, Keith J, Ramirez J, et al. (2018) APOE-ε4
associates with hippocampal volume, learning, and memory across the spectrum of
Alzheimer’s disease and dementia with Lewy bodies. Alzheimers Dement 14, 1137–1147.
[131] Konishi K, Joober R, Poirier J, MacDonald K, Chakravarty M, Patel R, et al. (2018) Healthy
versus Entorhinal Cortical Atrophy Identification in Asymptomatic APOE4 Carriers at Risk
for Alzheimer’s Disease. J Alzheimers Dis 61, 1493–1507.
[132] Dickerson B, Goncharova I, Sullivan M, Forchetti C, Wilson R, Bennett D, et al. (2001)
MRI-derived entorhinal and hippocampal atrophy in incipient and very mild Alzheimer’s
disease. Neurobiol Aging 22, 747–54.
[133] Nasreddine Z, Phillips N, Bedirian V, Charbonneau S, Whitehead V, Collin I, et al. (2005)
The Montreal Cognitive Assessment, MoCA: A Brief Screening Tool For Mild Cognitive
Impairment. J AM Geriatr Soc 53, 695–9.
[134] Hammer M, Swinburn T, Neitzel R (2014) Environmental Noise Pollution in the United
States: Developing an Effective Public Health Response. Environ Health perspect 122, 115–
119.
[135] Babisch W (2002) The Noise/Stress Concept, Risk Assessment and Research Needs. Noise
Health 4, 1–11.
[136] Basner M, McGuire S (2018) WHO Environmental Noise Guidelines for the European
Region: A Systematic Review on Environmental Noise and Effects on Sleep. Int J Environ
Res Public Health 15, 519.
[137] Munzel T, Daiber A, Steven S, et al. (2017) Effects of noise on vascular function, oxidative
stress, and inflammation: mechanistic insight from studies in mice. Eur Heart J 38, 2838–
2849.
[138] Kroller-Schon S, Daiber A, Steven S, et al. (2018) Crucial role for Nox2 and sleep
deprivation in aircraft noise-induced vascular and cerebral oxidative stress, inflammation,
and gene regulation. Eur Heart J 39, 3528–3539.
[139] Cui B, Zhu L, She X, et al. (2012) Chronic noise exposure causes persistence of tau
hyperphosphorylation and formation of NFT tau in the rat hippocampus and prefrontal
cortex. Exp Neurol 238, 122–9.
[140] Cui B, Wu M, She X, et al. (2012) Impulse noise exposure in rats causes cognitive deficits
and changes in hippocampal neurotransmitter signaling and tau phosphorylation. Brain Res
1427, 35–43.
[141] Cui B, Li K (2013) Chronic noise exposure and Alzheimer disease: Is there an etiological
association? Med Hypotheses 81, 623–6.
[142] Cui B, Wu M, She X (2009) Effects of chronic noise exposure on spatial learning and
memory of rats in relation to neurotransmitters and NMDAR2B alteration in the
hippocampus. J Occup Health 51, 152–8.
[143] Manikandan S, Padma M, Srikumar R, et al. (2006) Effects of chronic noise stress on spatial
memory of rats in relation to neuronal dendritic alteration and free radical-imbalance in
hippocampus and medial prefrontal cortex. Neurosci Lett 399, 17–22.
[144] Liu L, He T, Chang Y, et al. (2016) Noise induced hearing loss impairs spatial
learning/memory and hippocampal neurogenesis in mice. Scientific Reports 6, 20374.
100
[145] Bycroft C, Freeman C, Petkova D, et al. (2018) The UK Biobank resource with deep
phenotyping and genomic data. Nature 203–209.
[146] Littlejohns T, Holliday J, Gibson L, et al. (2020) The UK Biobank imaging enhancement of
100,000 participants: rationale, data collection, management and future directions. Nature
Communications 2624.
[147] Sudlow C, Gallacher J, Allen N, et al. (2015) UK Biobank: An Open Access Resource for
Identifying the Causes of a Wide Range of Complex Diseases of Middle and Old Age. PLoS
Med 12, e1001779.
[148] Kephalopoulos S, Paviotti M, Anfosso-Ledee F (2012) Common Noise Assessment
Methods in Europe (CNOSSOS‐EU). Publications Office of the European Union,
Luxembourg.
[149] Kephalopoulos S, Paviotti M, Anfosso-Ledee F, et al. (2014) Advances in the development
of common noise assessment methods in Europe: The CNOSSOS-EU framework for
strategic environmental noise mapping. Science of the Total Environment 400–410.
[150] Eeftens M, Beelen R, de Hoogh K, et al. (2012) Development of Land Use Regression
models for PM(2.5), PM(2.5) absorbance, PM(10) and PM(coarse) in 20 European study
areas; results of the ESCAPE project. Environ Sci Technol 46, 11195–205.
[151] Miller K, Alfaro-Almagro F, Bangerter N, et al. (2016) Multimodal population brain
imaging in the UK Biobank prospective epidemiological study. Nat Neurosci 19, 1523–
1536.
[152] Smith S (2002) Fast robust automated brain extraction. Hum Brain Mapp 17, 143–155.
[153] Jenkinson M, Bannister P, Brady M, et al. (2002) Improved optimization for the robust and
accurate linear registration and motion correction of brain images. Neuroimage 17, 825–41.
[154] Jenkinson M, Smith S A global optimisation method for robust affine registration of brain
images. Med Image Anal 5, 143–56.
[155] Andersson N, Smith S, Jenkinson M (2007) Non-linear optimisation., Oxford Centre for
Functional Magnetic Resonance Imaging of the Brain, Department of Clinical Neurology,
Oxford University, Oxford, UK.
[156] Andersson J, Jenkinson M, Smith S (2007) Non-linear registration aka spatial
normalisation., Oxford Centre for Functional Magnetic Resonance Imaging of the Brain,
Department of Clinical Neurology, Oxford University, Oxford, UK.
[157] Desikan R, Segonne F, Fischl B, et al. (2006) An automated labeling system for subdividing
the human cerebral cortex on MRI scans into gyral based regions of interest. NeuroImage
31, 968–80.
[158] Fischl B, Van der Kouwe A, Destrieux C, et al. (2004) Automatically parcellating the human
cerebral cortex. Cereb Cortex 14, 11–22.
[159] Klapwijk E, Van de Kamp F, Van der Meulen M, et al. (2019) Qoala-T: A supervised-
learning tool for quality control of FreeSurfer segmented MRI data. NeuroImage 116–129.
[160] Griffanti L, Zamboni G, Khan A, et al. (2016) BIANCA (Brain Intensity AbNormality
Classification Algorithm): A new tool for automated segmentation of white matter
hyperintensities. Neuroimage 191–205.
[161] Warrier C, Wong P, Penhune V, et al. (2009) Relating Structure to Function: Heschl’s Gyrus
and Acoustic Processing. J Neurosci 29, 61–69.
[162] Millard L, Davies N, Gaunt T, et al. (2018) Software Application Profile: PHESANT: a tool
for performing automated phenome scans in UK Biobank. Int J Epidemiol 47, 29–35.
101
[163] Basner M, Babisch W, Davis A, et al. (2014) Auditory and non-auditory effects of noise on
health. Lancet 383, 1325–32.
[164] Stansfeld S, Matheson M (2003) Noise pollution: non-auditory effects on health. British
Medical Bulletin 68, 243–257.
[165] Vos T, Flaxman A, Naghavi M, et al. (2012) Years lived with disability (YLDs) for 1160
sequelae of 289 diseases and injuries 1990–2010: a systematic analysis for the Global
Burden of Disease Study 2010. The Lancet 380, 2163–2196.
[166] Oishi N, Schacht J (2011) Emerging treatments for noise-induced hearing loss. Expert
Opinion on Emerging Drugs 16, 235–245.
[167] Song R, Pan K-Y, Xu H, et al. (2021) Association of cardiovascular risk burden with risk
of dementia and brain pathologies: A population-based cohort study. Alzheimers Dement
17, 1914–1922.
[168] Daulatzai M (2015) Evidence of neurodegeneration in obstructive sleep apnea: Relationship
between obstructive sleep apnea and cognitive dysfunction in the elderly. J Neurosci Res
93, 1778–94.
[169] Medic G, Wille M, Hemels M (2017) Short- and long-term health consequences of sleep
disruption. Nat Sci Sleep 9, 151–161.
[170] Osorio R, Gumb T, Pirraglia E, et al. (2015) Sleep-disordered breathing advances cognitive
decline in the elderly. Neurology 84, 1964–71.
[171] Sproston N, Ashworth J (2018) Role of C-Reactive Protein at Sites of Inflammation and
Infection. Front Immunol 9, 754.
[172] Kravitz B, Corrada M, Kawas C (2009) Elevated C-reactive protein levels are associated
with prevalent dementia in the oldest-old. Alzheimers Dement 5, 318–23.
[173] Hilal S, Ikram A, Verbeek M, et al. (2018) C-Reactive Protein, Plasma Amyloid-β Levels,
and Their Interaction With Magnetic Resonance Imaging Markers. Stroke 49,.
[174] Zhang X, Kang J, chen H, et al. (2018) PM2.5 Meets Blood: In vivo Damages and Immune
Defense. Aerosol and Air Quality Research 18, 456–470.
[175] Seaton A, Soutar A, Crawford V, et al. (1999) Particulate air pollution and the blood. Thorax
54, 1027–32.
[176] Walker A, Seibert S, Donato A, et al. (2010) Vascular Endothelial Function is Related to
White Blood Cell Count and Myeloperoxidase Among Healthy Middle-Aged and Older
Adults. Hypertension 55, 363–369.
[177] Eckrich J, Frenis K, Rodriguez-Blanco G, et al. (2021) Aircraft noise exposure drives the
activation of white blood cells and induces microvascular dysfunction in mice. Redox
Biology 46, 102063.
[178] Widmer R, Lerman A (2014) Endothelial dysfunction and cardiovascular disease. Glob
Cardiol Sci Pract 3, 291–308.
[179] Nezu T, Hosomi N, Aoki S, et al. (2015) Endothelial dysfunction is associated with the
severity of cerebral small vessel disease. Hypertension Research 38, 291–297.
[180] Stansfeld S (2015) Noise Effects on Health in the Context of Air Pollution Exposure. Int J
Environ Res Public Health 12735–12760.
[181] Guia-Galipienso F, Martinez-Ferran M, Vallecillo N, et al. (2021) Vitamin D and
cardiovascular health. Clin Nutr 40, 2946–2957.
[182] Littlejohns T, Henley W, Lang I, et al. (2014) Vitamin D and the risk of dementia and
Alzheimer disease. Neurology 83, 920–928.
102
[183] Friso S, Lotto V, Corrocher R, et al. (2012) Vitamin B6 and cardiovascular disease. Subcel
Biochem 56, 265–90.
[184] Morris M, Schneider J, Tangney C (2006) Thoughts on B-vitamins and dementia. J
Alzheimers Dis 9, 429–433.
[185] Behbodikhah J, Ahmed S, Elyasi A, et al. (2021) Apolipoprotein B and Cardiovascular
Disease: Biomarker and Potential Therapeutic Target. Metabolites 11, 690.
[186] Ishii M (2019) Apolipoprotein B as a New Link Between Cholesterol and Alzheimer
Disease. JAMA Neurol 76, 751–753.
[187] Picard C, Nilsson N, Labonte A, et al. (2022) Apolipoprotein B is a novel marker for early
tau pathology in Alzheimer’s disease. Alzheimers Dement 18, 875–887.
[188] Ising H, Braun C (2000) Acute and chronic endocrine effects of noise: Review of the
research conducted at the Institute for Water, Soil and Air Hygiene. Noise Health 2, 7–24.
[189] de Souza-Talarico J, Marin M-F, Sindi S, et al. (2011) Effects of stress hormones on the
brain and cognition Evidence from normal to pathological aging. Dement Neuropsychol 5,
8–16.
[190] Whitworth J, Williamson P, Mangos G, et al. (2005) Cardiovascular Consequences of
Cortisol Excess. Vasc Health Risk Manag 1, 291–299.
[191] Goyal S, Gupta V, Walia L (2010) Effect of noise stress on autonomic function tests. Noise
Health 12, 182–6.

Abstract (if available)

Abstract Alzheimer’s disease (AD) is a progressive neurodegenerative disease that affects many people and causes detrimental cognitive impairments and functional disabilities. Late onset AD (LOAD) is the most common form of AD which appears after the age of 65 years old. It is a multifactorial disease with mixed etiologies that are not fully understood. Aging is the highest risk factor for developing AD besides many other causes including genetics, lifestyle factors, and environmental pollution. Familial AD (FAD) is another rare form of AD that appears in early ages. Genetics is the main cause of FAD. AD is considered the sixth leading cause of death in the United States and the fifth leading cause of death in Americans who are older than 65 years. AD and dementia in general are becoming a major health concern worldwide which stems the importance of understanding AD especially with the continuing increase of the aging population. Normal aging is not typically associated with dementia and AD symptoms, but the risk of developing AD increases with aging. In this research work, we attempted to further understand AD and elucidate its progression, development, and biological implications. Gene expression analysis was performed using gene expression data from multiple brain regions to find dysregulated genes and understand their biological implications and pathways. Ethno-racial disparities in brain biomarkers were also studied for LOAD using a non-mixed AD subjects from three races: Non-Hispanic White, African American, and Hispanic AD subjects. We also studied the effects of environmental noise exposure on human health and brain using a large prospective cohort from the UK Biobank project.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Characterizing brain aging with neuroimaging, health, and genetic data

PDF

Vascular contributions to brain aging along the Alzheimer's disease continuum

PDF

Using neuroinformatics to identify genomic and proteomic markers of suboptimal aging and Alzheimer's disease

Pdf

Alzheimer's disease: risk factors, value, and alternative payment models

PDF

The role of inflammation in mediating effects of obesity on Alzheimer's disease

PDF

The effects of a fasting mimicking diet (FMD) on mouse models of Alzheimer's and Parkinson's disease

PDF

The role of blood pressure variability in cognitive decline, cerebrovascular disease and Alzheimer’s disease

PDF

From risk mitochondrial and metabolic phenotype towards a precision medicine approach for Alzheimer's disease

PDF

Blood-brain barrier pathophysiology in cognitive impairment and injury

PDF

Affective neuropsychiatric symptoms and neural connectivity in the early stages of Alzheimer’s disease

PDF

Alzheimer’s disease and related dementias research communication practices with minoritized racial and ethnic groups: a multiple case study

PDF

Neural crest-derived cranial pericytes and their dysfunction in disease

PDF

Neuroimaging markers of risk & resilience to brain aging and dementia

PDF

The role of the locus coeruleus in Alzheimer’s disease and cerebrovascular function: insights from neuroimaging, neuropsychology, and biofluid markers

PDF

The association of cerebrovascular disease risk factors with brain structure and its modification by genetic variation

PDF

Long-term blood pressure variability across the clinical and biomarker spectrum of Alzheimer’s disease

PDF

Bioenergetic fuel shift of reproductive aging: implication for late-onset Alzheimer's disease

PDF

TREM2 and C1q signaling regulates immunoproteostasis in Alzheimer's disease

PDF

Crosscultural differences in dementia diagnosis and care-seeking in Hispanic and non-Hispanic white outpatients

PDF

Longitudinal neurocognitive profiles of empirically-derived Alzheimer’s disease variants

Asset Metadata

Creator Arzouni, Nibal (author)

Core Title Alzheimer’s disease: dysregulated genes, ethno-racial disparities, and environmental pollution

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Computational Biology and Bioinformatics

Degree Conferral Date 2022-08

Publication Date 07/22/2022

Defense Date 06/15/2022

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

Tag aging,Air pollution,Alzheimer's disease,apolipoprotein E (APOE),biological pathways,brain,brain imaging,brain imaging biomarkers,cardiovascular diseases.,classification,cognition,dementia,ethno-racial disparities,gene expression data,linear mixed models,machine learning,MRI,noise exposure,OAI-PMH Harvest,sex differences,white blood cells

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Toga, Arthur (committee chair), Cabeen, Ryan (committee member), Nakano, Aiichiro (committee member), Shi, Yonggang (committee member), Sun, Fengzhu (committee member)

Creator Email arzouni@usc.edu,narzouni@umich.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-oUC111373915

Unique identifier UC111373915

Legacy Identifier etd-ArzouniNib-10913

Document Type Dissertation

Format application/pdf (imt)

Rights Arzouni, Nibal

Type texts

Source 20220722-usctheses-batch-961 (batch), 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 author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright. The original signature page accompanying the original submission of the work to the USC Libraries is retained by the USC Libraries and a copy of it may be obtained by authorized requesters contacting the repository e-mail address given.

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

Repository Email cisadmin@lib.usc.edu