Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
The association of cerebrovascular disease risk factors with brain structure and its modification by genetic variation
(USC Thesis Other)
The association of cerebrovascular disease risk factors with brain structure and its modification by genetic variation
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
THE ASSOCIATION OF CEREBROVASCULAR DISEASE RISK FACTORS
WITH BRAIN STRUCTURE AND ITS MODIFICATION BY GENETIC VARIATION
by
William J. Matloff
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
(NEUROSCIENCE)
May 2022
Copyright 2022 William J. Matloff
ii
DEDICATION
To my Mom and Dad, for their endless love, support, and wisdom.
iii
ACKNOWLEDGEMENTS
This work was supported by the following NIH Grants: U54 EB020406, P41 EB015922,
R01 MH094343, and the NIEHS T32 Environmental Genomics Training Grant (NIH Grant No.
T32ES013678).
I first thank my advisor, Dr. Arthur Toga. Through regular feedback and advice, Dr. Toga
helped me grow substantially as a researcher. He pushed me to develop a deep understanding of
the literature, to critically examine whether claims are supported by data, and to identify promising
areas of investigation. He gave me the opportunity to pursue a wide range of projects, through
which I gained many useful skills.
I am extremely grateful for the professors who have been part of either my Guidance of
Dissertation committees: Drs. Hosung Kim, Judy Pa, David Conti, and Yonggang Shi. Their expert
advice and mentorship have improved each project in this dissertation and helped me focus my
research direction. I also thank Dr. Jim Gauderman, who along with Dr. David Conti, served as a
faculty mentor within the Environmental Genomics T32 fellowship.
I am also very thankful to have had the opportunity to work with many great researchers at
the Laboratory of Neuro Imaging, including Lu Zhao, Kaida Ning, and Nibal Arzouni. In addition,
I am thankful for the leadership, peers, and administrators in the USC Neuroscience Graduate
Program. Furthermore, I am grateful to the current and former leadership of the USC-Caltech MD-
PhD Program, who gave me the opportunity to pursue a PhD in addition to my medical studies.
Finally, I am exceptionally thankful for my Mom, Dad, my sister Alexandra, my brother
Daniel, and my Grandma, who are always there for me and inspire me in countless ways.
iv
Chapters 2-5 are adapted from the following articles that are either published, in review, or
in preparation. I thank all coauthors for their contributions to these works.
Chapter 2:
Kim H*, Matloff WJ* {*Equal Contribution}, Zhao L, Arzouni N, Tanaka N, Heckbert SR,
Kwon Y, Toga, AW. Atrial fibrillation is associated with brain atrophy in regions associated
with Alzheimer disease in the UK Biobank. In review.
Chapter 3:
Matloff WJ*, Kim H* {*Equal Contribution}, Zhao L, Arzouni N, Haddad E, Toga AW. P-wave
duration is associated with MRI brain structure measurements: A cross-sectional study of 22,850
UK Biobank participants. In preparation.
Chapter 4:
Matloff WJ, Zhao L, Ning K, Conti DV, Toga AW. Interaction effect of alcohol consumption
and Alzheimer disease polygenic risk score on the brain cortical thickness of cognitively normal
subjects (2020). Alcohol, 85:1–12. https://doi.org/10.1016/j.alcohol.2019.11.002.
Chapter 5:
Matloff WJ, Zhao, L, Arzouni N, Conti DV, Gauderman J, Toga AW. Genome-wide interaction
of body mass index on hippocampal T2* value. In preparation
v
TABLE OF CONTENTS
Dedication ....................................................................................................................................... ii
Acknowledgements ........................................................................................................................ iii
List of Tables ................................................................................................................................ vii
List of Figures .............................................................................................................................. viii
Abstract .......................................................................................................................................... ix
Chapter 1: Introduction
1.1 Neuroimaging for preventive brain health .....................................................................1
1.2 Clinical and subclinical cerebrovascular disease ...........................................................1
1.3 Cerebrovascular disease risk factors and their association with brain structure ............2
1.4 Gene-environment interactions: a hypothesis-generating tool .......................................4
1.5 Research aims and overview of chapters .......................................................................5
Chapter 2: Atrial fibrillation is associated with brain atrophy in regions associated with
Alzheimer disease in the UK Biobank
2.1 Introduction ....................................................................................................................7
2.2 Methods..........................................................................................................................8
2.3 Results ..........................................................................................................................13
2.4 Discussion ....................................................................................................................20
Chapter 3: P-wave duration is associated with MRI brain structure measurements: A cross-
sectional study of 22,850 UK Biobank participants
3.1 Introduction ..................................................................................................................24
3.2 Methods........................................................................................................................25
3.3 Results ..........................................................................................................................30
3.4 Discussion ....................................................................................................................36
Chapter 4: Interaction effect of alcohol consumption and Alzheimer disease polygenic risk score
on the brain cortical thickness of cognitively normal subjects
4.1 Introduction ..................................................................................................................41
4.2 Methods........................................................................................................................44
4.3 Results ..........................................................................................................................50
4.4 Discussion ....................................................................................................................58
Chapter 5: Genome-wide interaction scan of body mass index on hippocampal T2* value
5.1 Introduction ..................................................................................................................65
5.2 Methods........................................................................................................................66
5.3 Results ..........................................................................................................................70
5.4 Discussion ....................................................................................................................75
vi
Chapter 6: Conclusions and Future Directions
6.1 Summary of findings ....................................................................................................79
6.2 Future directions ..........................................................................................................80
References ......................................................................................................................................82
vii
LIST OF TABLES
Table 2.1: Study population characteristics ...................................................................................14
Table 2.2: Association of AF/AFL with brain structure ................................................................16
Table 2.3: Association of AF/AFL with cognitive function test scores .........................................16
Table 3.1: Sample characteristics by P-wave duration group ........................................................31
Table 3.2: Association of IDP with P-wave duration group for FDR-significant IDPs ................34
Table 3.3: Association of covariates with P-wave duration group ................................................35
Table 3.4: Mediation analysis of covariates and brain structure by P-wave duration ...................36
Table 4.1: Study population characteristics ...................................................................................51
Table 4.2: Association of alcohol consumption with AD cortical thickness signature .................52
Table 4.3: Interaction of AD PRS and alcohol consumption on cortical thickness .......................54
Table 5.1: Sample characteristics by subset ...................................................................................70
Table 5.2: Statistics for significant SNPs ......................................................................................73
Table 5.3: Additional brain IDPs associated with the significant SNPs ........................................75
Table 5.4: Associations of the variants with blood values in the non-imaging subset ..................75
viii
LIST OF FIGURES
Figure 2.1: Mediation analysis for cognitive function tests ...........................................................17
Figure 2.2: Interaction of age and AF/AFL ...................................................................................18
Figure 2.3: Brain-wide AF/AFL associations ................................................................................19
Figure 3.1: Association of P-wave duration and brain structure ....................................................30
Figure 3.2: Nonlinear relationship between PWD and brain structure ..........................................32
Figure 3.3: Individual PWD association plots ...............................................................................33
Figure 4.1: Plots of association of alcohol consumption with AD Cortical Thickness Signature .53
Figure 4.2: Association of alcohol consumption and cortical thickness by AD genetic risk ........55
Figure 4.3: Interaction among weekly alcohol consumers ............................................................56
Figure 5.1: Association of BMI and both regional volume and subcortical median T2* value ....71
Figure 5.2: Manhattan plots ...........................................................................................................72
Figure 5.3: Interaction plots ...........................................................................................................74
ix
Abstract
Cerebrovascular disease risk factors, many of which are also risk factors for dementia,
contribute substantially to adverse brain changes even in the absence of clinically overt
cerebrovascular disease or dementia. Neuroimaging studies have found that these risk factors, such
as hypertension, obesity, and hyperlipidemia, are associated with brain atrophy, microinfarcts,
microbleeds, and white matter hyperintensities. These adverse changes increase cognitive
impairment, dementia, and stroke risk. Identifying novel and understanding existing
cerebrovascular risk factors is essential for preventing their negative impact. This dissertation
therefore has two main aims to advance this objective. The first is to investigate the association of
atrial fibrillation and abnormal P-wave duration, as measured on electrocardiogram, with brain
structure. Both of these heart-related factors have many unknowns in terms of their relationship
with brain structure. The second focus is to examine whether the association of well-known
cerebrovascular factors such as alcohol consumption and obesity is modified by genetic variation,
which may provide insight into how these factors lead to cerebrovascular disease. We used data
from the UK Biobank resource, a population-scale prospective cohort study in the United
Kingdom, to approach these aims. We found that atrial fibrillation and both an abnormally short
and long P-wave duration were associated with brain structure and composition. Second, we found
that the association of alcohol consumption and brain structure varied by Alzheimer disease
genetic risk and that there exist variants significantly associated with hippocampal T2* value, a
metric reflecting iron deposition. One such variant additionally had an association with serum
triglyceride levels that varied with body mass index. Overall, these results suggest that cardiac
pathology beyond coronary heart disease has a clear link with brain health and that there exist
genetic variants that modify the association of cerebrovascular risk factors with brain structure.
1
Chapter 1
Introduction and Background
1.1 Neuroimaging for preventive brain health
A major goal of neuroimaging research is to identify strategies to maintain brain health
throughout life [1]. With magnetic resonance imaging (MRI) allowing brain structure to be
quantified non-invasively without ionizing radiation, researchers have been able to evaluate how
the brain changes over time and how these changes are related to environmental and health factors
[2]. Factors that are modifiable and have a causal effect on brain structure represent intervention
targets for optimizing brain health.
Two main drivers of brain structural changes detected on MRI include cerebrovascular-
related changes and proteinopathy-related changes [3]. Cerebrovascular-related changes include
pathologies such as lacunar infarcts, white matter lesions, and microbleeds, whereas
proteinopathy-related changes include pathologies such as the deposition of amyloid, tau, or other
proteins [4]. While aging is strongly associated with dementia risk and brain structure changes,
there is evidence to suggest that the increased dementia risk with age may be mediated almost
entirely by the sum of both of these neuropathology types [5]. If this is indeed accurate, then
finding ways to prevent these neuropathologies will have a substantial impact on preventive brain
health.
1.2 Clinical and subclinical cerebrovascular disease
The cerebrovascular-related brain changes that can be detected on MRI are considered to
result from cerebrovascular disease. Cerebrovascular disease encompasses a wide range of brain
2
pathologies, from acute cerebrovascular events such as stroke to subclinical cerebrovascular
disease. While ischemic and hemorrhagic strokes typically have dramatic clinical presentations
and result in functionally evident brain damage, subclinical cerebrovascular disease (sCVD),
which is very common in the elderly, has much more subtle effects [6].
Brain changes associated with sCVD are reflected in a variety of MRI modalities. For
example, lacunar infarcts and cerebral microbleeds can be detected on susceptibility-weighted and
gradient echo MRI, and white matter hyperintensities and silent brain infarcts can be found on T2
MRI scans [6]. In addition, brain structural features such as cortical thickness, volume, and surface
area can be derived from T1-weighted MRI scans [7]. These types of structural changes may reflect
underlying sCVD. For instance, increased white matter hyperintensity volume has been found to
be associated with decreased frontal and temporal cortical thickness [8]. White matter tract
structural integrity can be quantified using diffusion tractography using metrics such as fractional
anisotropy (FA) [7]. Finally susceptibility weighted MRI can be used to quantify magnetic changes
characteristic of iron deposition [9]. There are number of MRI-derived cerebrovascular-related
brain features therefore that can be studied to help characterize how environmental and health
factors are linked to brain health.
1.3 Cerebrovascular disease risk factors and their association with brain structure
Neuroimaging studies have found that many of the risk factors for stroke are themselves
associated with these MRI-derived brain measurements found in sCVD, independent of stroke
[10]. Therefore, interventions that decrease stroke risk may also decrease the risk of adverse brain
changes and dementia [11–13]. Investigating the association of stroke risk factors on brain
3
structure in those without a history of stroke is therefore useful because it may reveal other
potential targets for intervention that might help to optimize brain health.
Established cerebrovascular disease risk factors include hypertension [14], obesity [15,16],
smoking [17], diabetes [18], dyslipidemia [19], congestive heart failure [20], excessive alcohol
consumption [21,22], and metabolic syndrome [23]. Consistent with the expectation that stroke
risk factors also are associated with brain changes prior to stroke, these factors have been found to
be associated with brain structure [10]. For instance, alcohol consumption has been found to be
associated with increased brain atrophy and deep cerebral microbleeds [24,25]. Obesity has been
found to be associated with decreased brain volumes, including hippocampal volume [26]
Many of these factors are also associated with Alzheimer disease. This could be due to the
factors independently increasing Alzheimer disease risk or a consequence of cerebrovascular-
related changes increasing Alzheimer dementia risk [27,28]. Results from genetics studies and
longitudinal studies suggest that cerebrovascular-related changes promote Alzheimer disease-
related changes [29,30]. Regardless, multiple brain pathology types are common among older
adults [4]. These stroke risk factors may affect the brain in the absence of stroke by leading to
blood flow reduction, alterations in brain vasculature, destruction of the blood brain barrier, or
cardiogenic emboli, in addition to other possibilities [27]. So, these cerebrovascular risk factors
can be expected to also have an association with AD-related brain regions.
One stroke risk factor that is need of further study in terms of its relation with brain
structure independent of stroke is atrial pathology of the heart, including atrial fibrillation and
abnormal P-wave duration [31,32]. Previous studies of the association of atrial fibrillation with
brain structure have been small and only focused on a few brain phenotype measurements [33,34].
4
Similarly, P-wave duration (PWD) has only been studied in relation to microbleeds, microinfarcts,
and white matter hyperintensities [35]. These factors are therefore a focus of this dissertation.
1.4 Gene-environment interactions: a hypothesis-generating tool
While there are many known cerebrovascular disease risk factors, exactly how these factors
contribute to adverse brain changes is not fully known. Genetics analyses can be a useful
hypothesis generating tool in this respect. Genome-wide association analyses have previously been
completed for MRI-derived sCVD markers, such as white matter hyperintensities [36]. An
interesting question though is whether the association cerebrovascular disease risk factors with
brain structure varies based on genetic variation. Or, equivalently, whether there are any risk
factors that only are associated with brain structure for a certain allele of a variant. Findings from
such gene-environment interaction analyses can therefore be hypothesis-generating by either
revealing novel variants associated with an specific MRI-derived brain metric that would not have
been found in a standard GWAS or based on the function of a variant giving an insight into how a
factor might contribute to cerebrovascular disease [37,38].
To date, gene-environment interaction analyses on either dementia risk or brain structure
measurements have been fairly sparse, especially for brain structure measurements. The apoE-4
allele, the strongest genetic risk factor for Alzheimer disease, has been found to modify the effect
of physical activity, alcohol consumption, pesticide exposure, and air pollution exposure on
dementia risk [39]. In addition, one study used brain morphometry and a set of AD-related SNPs
to find that ABCA7 (ATP-binding cassette, subfamily A, member 7) interacts with cardiovascular
disease risk factors on right superior parietal volume [40]. There has been at least one genome-
wide association study of brain structure accounting for an interaction term with a trait, a study
5
that investigated the genetics of white matter hyperintensity volume and additionally tested a
model with an interaction term with hypertension, though no additional variants were found using
a two degree-of-freedom test considering this interaction [41].
Despite the relative lack of investigation into gene-environment interactions on brain
structure measurements, searching for such interactions may be fruitful, and is the motivation for
the work in this dissertation on gene-environment interaction effects. The question of whether
there are any genetic variants that modify the association of cerebrovascular risk factors and brain
structure may generate some interesting hypotheses for why these factors are associated with
adverse brain associations.
1.5 Research aims and overview of chapters
A main goal of this research is to characterize the association with brain structure of two
atrial heart pathologies: atrial fibrillation and abnormal P-wave duration, which may reflect atrial
cardiomyopathy. The associations of these atrial pathologies with regional brain volumes,
thicknesses, white matter tract structural integrity measurements, subcortical susceptibility-
weighted MRI (swMRI) T2* values, and white matter hyperintensity volumes have not been
examined previously. The UK Biobank is used to study both of these pathologies, with the atrial
fibrillation variable based on self-reported history and ECG detection on the day of brain imaging.
Chapter 2 focuses on atrial fibrillation and Chapter 3 focuses on P-wave duration. Each of these
studies find significant associations with these atrial pathologies and identify a pattern of brain
measurements associated with these pathologies.
The second goal of this research is to evaluate whether genetic variation modifies the
association of known cerebrovascular risk factors, such as alcohol and obesity. The aim is to gain
6
further insight into these risk factors. Chapter 4 investigates the association of alcohol
consumption with cortical thickness in regions known to be associated with Alzheimer disease
(AD) and whether this association is modified by an AD polygenic risk score (PRS). Chapter 5
investigates the association of body mass index (BMI) and subcortical structures and subsequently
tests for gene-environment interaction effects across the whole genome on hippocampal T2* value,
the imaging-derived phenotype with the largest effect size with BMI. In the final chapter, Chapter
6, all work is summarized, and the significance is discussed along with possible future directions.
7
Chapter 2
Atrial fibrillation is associated with brain atrophy in regions associated with
Alzheimer disease in the UK Biobank
2.1 Introduction
Atrial fibrillation (AF) is the most common chronic arrhythmia, with an estimated
prevalence of at least 3-6 million people in the United States, increasing from approximately 0.1%
at ages <55 years to 8.8% at >80 years [42–45]. It is a major risk factor for cerebrovascular disease,
with AF being a presumed causal factor in about 1/3 of all strokes [46].
With the elderly population
percentage increasing, projections estimate that 6-12 million in the United States will be affected
by AF by 2050 and 17.9 million people in Europe by 2060 [31].
Beyond stroke risk, AF has been linked to cognitive impairment and dementia, even in the
absence of clinical stroke and accounting for vascular and genetic risk factors [47–49]. Possible
pathophysiological explanations for these risks include altered cerebral blood flow, microinfarcts,
and subclinical strokes [34,50,51]. Evidence additionally suggests that AF rhythm control by
catheter ablation and oral anticoagulation may decrease dementia risk [52–54].
Brain imaging can help elucidate underlying pathophysiological mechanisms linking AF
to neurocognitive impairment, which may be informative for optimal AF treatment strategy.
Indeed, AF has been associated with a number of structural and functional changes, including
cerebral microbleeds [55], decreased cerebral blood flow [56], silent cerebral infarctions [57,58],
and reduced volume in total brain, total gray matter, middle temporal lobe, hippocampus, and
entorhinal cortex [33,51,59–62].
8
These brain imaging studies, however, have been limited in sample size, brain structures
analyzed, and availability of cognitive test data. Our aim was to investigate using the UK Biobank
(UKB) resource the association of AF or atrial flutter (AF/AFL) with imaging-based
measurements of regional brain volume, cortical thickness, and white matter microstructural
integrity. In contrast to previous studies, we distinguish between AF/AFL history and ECG-
detected AF/AFL at the time of brain imaging. Furthermore, we aimed to investigate whether the
association of AF/AFL on brain structure varies with age, whether brain structure mediates the
association of AF/AFL and cognitive function, and whether self-reported warfarin oral
anticoagulation or antiarrhythmic use modifies the association of AF/AFL and hippocampal
volume.
2.2 Methods
2.2.1 Study population
We used data from the UKB (http://www.ukbiobank.ac.uk), a large prospective study
based in the United Kingdom [63]. From this resource, we studied a subset of participants who
attended an assessment visit during which brain MRI, cognitive function test, 12-lead ECG (10
seconds), and other health-related data were acquired. All participants provided written informed
consent, and ethics approval was given [64]. The initial assessment visit (no imaging or ECG data)
took place between 2006 and 2010 [65]. The imaging assessment visit took place between 2014
and 2019, a median of 9 years (min=4.2, max=13.5) after initial visit. Health information was
collected through a questionnaire, verbal interview, and linked inpatient hospital records [64].
9
2.2.2 Atrial fibrillation group definition
AF/AFL status was determined by either automatic electrocardiogram (ECG)
interpretation (GE Case Cardiosoft software) or history by self-reported or hospital-visit derived
data using a previously validated strategy [65,66]. For hospital-visit derived data, either a
diagnostic history or a procedural history, such as past cardioversion procedure, was used for
categorization into the AF/AFL history group. Table S1 contains variable definition details.
Participants were categorized into one of the following groups: 1) no history or ECG detection of
AF/AFL; 2) history of AF/AFL only; 3) AF/AFL on ECG only; and 4) AF/AFL history and on
ECG.
2.2.3 MRI acquisition and processing
MRI was performed for each participant on the same visit as the ECG, described in detail
previously [67]. MR images underwent quality control and processing by the UKB to extract brain
structure metrics, referred to as imaging-derived phenotypes (IDPs). These included subcortical,
ventricle, and cerebellar volumes, cortical thicknesses, and white matter (WM) tract fractional
anisotropy (FA) values [7]. The volume and cortical thickness measurements (UKB Categories
192 and 190) were derived through FreeSurfer processing of T1-weighted MRI scans using the
Desikan-Killiany parcellation [68]. Mean FA values for WM tracts (UKB Category 134) were
defined by the JHU ICBM-DTI-81 atlas [69] and derived from processing of diffusion tensor
imaging FA image with TBSS [70].
We focused on three main brain structure measurements: mean bilateral hippocampal
volume, fornix FA, and an AD Cortical Thickness Signature (AD-CTS), each of which is known
to be affected early in AD [71–73]. The fornix is the main outflow tract of the hippocampus, with
10
mean FA reflecting its structural integrity [72,74]. The AD Cortical Thickness Signature is the
average cortical thickness in regions known to be affected early in AD [71]. In addition, 139 IDPs
reflecting brain-wide cortical thicknesses, volumes, and WM tract FA values, were tested for their
association with AF/AFL.
2.2.4 Cognitive measurements
Participants completed touch-screen based cognitive tests at imaging assessment visit [64].
We investigated nine of these cognitive function tests: the reaction time, trail-making test B, pairs
matching test, numeric memory, fluid intelligence, matrix pattern, tower, symbol digit substitution,
and paired associate tests.
2.2.5 Covariates
Covariates included in regression models were assessment site, MRI head positioning and
motion (each for MRI-derived features only), sex, age, age
2
, education, apoE-4 genotype, tobacco
smoking, alcohol consumption, body mass index, intracranial volume (for brain volumes and FA
only), and medical or self-reported history of hypertension, hypercholesterolemia, diabetes,
coronary artery disease, heart failure, chronic kidney disease, and stroke. These covariates were
chosen based on their known association with both AF/AFL and brain structure [44]. Head motion
was estimated from resting state fMRI (UKB Data-Field 25741), measured in units of mm.
2.2.6 AF/AFL Medication Data
AF/AFL medication use, including anticoagulation and medications used for rhythm
control and rate control were also evaluated using self-reported data from either the initial visit or
imaging visit data. For anticoagulants, data was available only on warfarin, but not any of the novel
11
oral anticoagulants (NOACs). The variable for antiarrhythmic rhythm control medications
included sodium channel blockers and potassium channel blockers, and a variable for rate control
medications included beta blockers, calcium channel blockers, and digoxin.
2.2.7 Inclusion and Exclusion Criteria
We included participants having the three primary brain imaging metrics of interest, all the
covariate data described in 2.5, and with white British ancestry. Participants were excluded for
having history of dementia, Parkinson disease, or other central nervous system disease, and also if
the automated ECG analysis software reported any technical issue, such as poor data quality.
Participants were also removed to create a maximal subset of unrelated participants (greater than
3
rd
-degree) participants. Finally, participants with a mean hippocampal volume, AD cortical
thickness, or fornix FA
2
four standard deviations above or below the mean, likely due to MRI
artifacts, were also excluded.
2.2.8 Statistical analysis
The following analyses were performed using R v3.6.3 [75]. Demographic characteristics
were compared among the four AF/AFL groups using chi-squared tests for categorical variables
and ANOVA tests for continuous variables. A power analysis was conducted via simulation to
determine the statistical power to detect an association of meaningful size for the three main brain-
structural outcome variables of interest, decided arbitrarily to be four times the beta coefficient of
the linear age term (in years) on each outcome variable (i.e. a 4-year difference in age). 5,000
simulations were run for each combination of true associations by AF/AFL groups. The standard
12
deviation of the outcome variables in the simulation was defined to be the standard deviation of
the model residuals for a regression of these outcome variables on the covariates.
Linear regression models were used to estimate the association of AF/AFL with each of
the three brain-structure outcome variables, adjusting for the covariates described. The P-value for
the beta estimate of each AF/AFL group relative to the no-AF/AFL group (group 1) and the full
ANOVA p-value were evaluated. Cohen’s D effect sizes were computed from these models [76].
For the cognitive function variables, the comparison between AF/AFL on ECG (groups 3+4
combined) and no AF/AFL (group 1) was evaluated using linear regression models with covariates
described previously. A multiple mediation analysis was conducted using the lavaan (latent
variable analysis) R package to investigate whether there was evidence suggesting that the three
main brain structure measurements of interest were mediating a proportion of the association
between AF/AFL and cognitive function [77].
To investigate whether the association between AF/AFL and brain structure varies with
age, a regression model with an interaction term for age and AF/AFL was used. In this model, age
was modeled as a restricted cubic spline with 3 knots using the rms R package [78] and included
as an interaction with each AF/AFL group in separate models. The cubic spline fitting was used to
account for the nonlinear association of age with brain structure. We determined the significance
of the interaction term using an ANOVA test.
In addition to the three brain structure metrics of interest, a brain-wide association study
was completed for cortical thickness, subcortical volume, ventricle volume, cerebellar volume,
and white matter tract FA, evaluating the association of AF/AFL and each of the 139 IDPs. FDR-
adjusted P-values (for those <0.05) and signed Cohen’s D effect sizes were plotted on brain maps
13
using the ggseg and ggseg3d R packages [79], with the FA results plotted on the JHU ICBM-DTI-
81 WM Atlas [69].
Finally, we investigated whether the use of warfarin oral anticoagulation or antiarrhythmic
agents influences the association of AF/AFL and hippocampal volume. Specifically, we analyzed
the association of treatment with warfarin and the combined group of warfarin or an antiarrhythmic
agent for group 4, the group most likely to benefit from such medications and with the highest
proportion of participants reporting being on these medications. With a data subset including only
groups 1 and 4, we used a covariate-adjusted regression model containing terms for treatment,
group, and the group-by-treatment interaction. This model was used to test whether the association
of treatment and hippocampal volume were different between groups 1 and 4, and to estimate the
treatment association in these groups.
2.3 Results
2.3.1 Demographics
The final dataset satisfying the inclusion/exclusion criteria consisted of 24,332 participants
aged 46-81 (mean age 64.3 years, 52.7% female), 793 (3.3%) of whom had either ECG detected
AF/AFL or a history of AF/AFL. 447 (1.8%) had a history only (group 2), 153 (0.6%) had no
history of AF/AFL but AF/AFL on ECG (group 3), and 193 (0.8%) had both a history and AF/AFL
on ECG (group 4). The proportions of participants on warfarin were 0.3, 13.0, 3.9, and 37.8% for
groups 1-4, respectively. The proportions of antiarrhythmic use were 0.1, 18.8, 0.0, and 7.8%.
Demographic information is summarized in Table 2.1.
14
Table 2.1: Study population characteristics
Atrial Fibrillation Group:
None
(Group 1)
History Only
(Group 2)
ECG without
History
(Group 3)
ECG with
History
(Group 4)
p
1
Number of Participants 23539 447 153 193
Scan Site (%) 0.249
Cheadle 15300 (65.0) 293 (65.5) 83 (54.2) 126 (65.3)
Reading 2977 (12.6) 57 (12.8) 25 (16.3) 24 (12.4)
Newcastle 5262 (22.4) 97 (21.7) 45 (29.4) 43 (22.3)
rfMRI-derived Head Motion (mm) 0.12 (0.06) 0.13 (0.07) 0.14 (0.07) 0.15 (0.07) <0.001
Sex = Male (%) 10940 (46.5) 290 (64.9) 122 (79.7) 162 (83.9) <0.001
Age 64.11 (7.42) 67.83 (6.79) 69.34 (6.86) 71.58 (5.12) <0.001
Highest Education Level (%) 0.188
College or University 10739 (45.6) 198 (44.3) 69 (45.1) 76 (39.4)
Vocational or Professional
2
6729 (28.6) 122 (27.3) 47 (30.7) 62 (32.1)
Secondary or Pre-University
3
4595 (19.5) 88 (19.7) 25 (16.3) 36 (18.7)
None of the above 1476 (6.3) 39 (8.7) 12 (7.8) 19 (9.8)
Hypertension (%) 5680 (24.1) 200 (44.7) 57 (37.3) 109 (56.5) <0.001
Hypercholesterolemia (%) 3769 (16.0) 152 (34.0) 34 (22.2) 62 (32.1) <0.001
Diabetes (%) 1180 (5.0) 37 (8.3) 11 (7.2) 25 (13.0) <0.001
Coronary Artery Disease (%) 1208 (5.1) 101 (22.6) 13 (8.5) 52 (26.9) <0.001
Heart Failure (%) 99 (0.4) 29 (6.5) 3 (2.0) 27 (14.0) <0.001
Chronic Kidney Disease (%) 163 (0.7) 14 (3.1) 1 (0.7) 3 (1.6) <0.001
Stroke (%) 275 (1.2) 13 (2.9) 6 (3.9) 16 (8.3) <0.001
Tobacco Smoking (%) <0.001
Never 15030 (63.9) 255 (57.0) 76 (49.7) 101 (52.3)
Previous 7588 (32.2) 174 (38.9) 71 (46.4) 86 (44.6)
Current 921 (3.9) 18 (4.0) 6 (3.9) 6 (3.1)
Alcohol (%) 0.005
Infrequent 6208 (26.4) 123 (27.5) 37 (24.2) 40 (20.7)
Occasional 13184 (56.0) 226 (50.6) 83 (54.2) 103 (53.4)
Frequent 4147 (17.6) 98 (21.9) 33 (21.6) 50 (25.9)
BMI Category (%) <0.001
Normal Weight 9546 (40.6) 157 (35.1) 48 (31.4) 50 (25.9)
Underweight 169 (0.7) 1 (0.2) 2 (1.3) 1 (0.5)
Overweight 9652 (41.0) 194 (43.4) 67 (43.8) 94 (48.7)
Obese 4172 (17.7) 95 (21.3) 36 (23.5) 48 (24.9)
ApoE-4 (%) 0.194
0 16978 (72.1) 336 (75.2) 114 (74.5) 138 (71.5)
1 6031 (25.6) 102 (22.8) 32 (20.9) 53 (27.5)
2 530 (2.3) 9 (2.0) 7 (4.6) 2 (1.0)
Warfarin Anticoagulation (%) 74 (0.3) 58 (13.0) 6 (3.9) 73 (37.8) <0.001
Rate Control Medication (%) 1160 (4.9) 169 (37.8) 35 (22.9) 115 (59.6) <0.001
Rhythm Control Medication (%) 24 (0.1) 84 (18.8) 0 (0.0) 15 (7.8) <0.001
Warfarin or Rhythm Control Medication (%) 97 (0.4) 120 (26.8) 6 (3.9) 80 (41.5) <0.001
1
P-value for group differences. Chi-squared tests were used for the categorical variables, and ANOVA tests
were used for the continuous variables.
Values represent mean (standard deviation) for continuous variables and count (%) for categorical variables.
2
NVQ, HND, HNC or equivalent, or other professional qualifications such as nursing or teaching, without
having earned a College or University degree
3
O levels, GCSDs, A levels, AS levels, CSEs or equivalent (without college or vocational degree)
15
The power analysis found that for hippocampal volume, this study was estimated to have
nearly 100% power to detect meaningful differences if all AF/AFL groups (groups 2-4) are
different from the no-AF/AFL group (group 1), and 93% power to detect difference if only the
ECG groups (groups 3-4) are different from group 1. For fornix FA, these power estimates were
100% and 99.8%, and for AD-CTS, 99.8% and 96.7%.
2.3.2 Association of AF/AFL with main brain structure measurements of interest
The results are summarized in Table 2.2. Compared to participants with no AF/AFL (group
1), hippocampal volume was significantly smaller in the two groups with AF/AFL on ECG (groups
3-4: 89.4 and 73.3 mm
3
smaller; Cohen’s D: 0.22 and 0.18; p<0.005). The hippocampal volume
was not different between the groups with no AF/AFL (group 1) and AF/AFL history only (group
2). Similar group differences were found for both fornix FA and the AD-CTS (ANOVA p<0.05).
Results did not change when excluding stroke as a model covariate or with excluding participants
with a history of stroke.
2.3.3 Association of AF/AFL with cognitive measurements and mediation analysis
There were no cognitive function tests significantly associated with AF/AFL after
correction for multiple comparisons (Table 2.3). The mediation analysis, however, showed
evidence of a significant mediating (indirect) effect of hippocampal volume, fornix FA, and AD-
CTS on the association of AF/AFL on the reaction time, fluid intelligence, and symbol digit
substitution tests (p<0.005; Table 2.3; Figure 2.1). The indirect effect was driven by hippocampal
volume and fornix FA for each of these three cognitive function tests. The indirect effect was
16
estimated to be a small portion of the total effect: 25% for reaction time, 15% for fluid intelligence,
and 12% for symbol digit substitution.
Table 2.2: Association of AF/AFL with brain structure.
Brain Structure
Measurement Group
Signed
Cohen’s D Beta
*
95% CI P-value
ANOVA
P-value
Mean Hippocampal
Volume (mm
3
)
2. History Only -0.01 -5.90 (-34.6, 22.8) 0.69
3.29e-5 3. ECG Only -0.22 -89.4 (-138., -41.0) 2.9e-4
4. History + ECG -0.18 -73.3 (-117., -29.4) 1.1e-3
Fornix FA-squared
2. History Only 0.01 0.0010 (-0.0046, 0.0066) 0.72
0.017 3. ECG Only -0.12 -0.0084 (-0.0178, 0.0010) 0.080
4. History + ECG -0.16 -0.0116 (-0.0201, -0.0031) 7.8e-3
AD Cortical Thickness
Signature (mm)
2. History Only -0.08 -0.0089 (-0.0184, 0.0006) 0.066
0.012 3. ECG Only -0.15 -0.0168 (-0.0328, -0.0008) 0.040
4. History + ECG -0.13 -0.0148 (-0.0292, -0.0003) 0.046
*Beta coefficients represent difference in outcome variables relative to the no AF/AFL group (Group 1;
n=23,539) after adjustment for covariates.
Abbreviation: AF/AFL = Atrial fibrillation or flutter
Table 2.3: Association of AF/AFL with cognitive function test scores
Cognitive Function Test Group Number and Name Beta
*
95% CI P-value
†
Indirect
P-value
‡
Reaction Time (Proc. Speed) 3+4 ECG+ (n=322 of 23362) 0.054 (-0.051, 0.16) 0.31 3.2e-4
Numeric Memory (Working Mem.) 3+4 ECG+ (n=255 of 16696) -0.024 (-0.14, 0.097) 0.70 0.12
Fluid Intelligence
(Verbal and Numeric Reasoning)
3+4 ECG+ (n=316 of 23087) -0.055 (-0.16, 0.051) 0.31 2.8e-3
Trail-Making Test B (Executive
Function)
3+4 ECG+ (n=234 of 15874) 0.037 (-0.080, 0.15) 0.54 0.15
Matrix Pattern (Nonverbal Reas.) 3+4 ECG+ (n=250 of 16269) -0.12 (-0.24, -0.004) 0.043 0.086
Tower Test (Executive Function) 3+4 ECG+ (n=250 of 16136) -0.095 (-0.22, 0.028) 0.13 0.27
Symbol Digit Substitution
(Processing Speed)
3+4 ECG+ (n=248 of 16290) -0.078 (-0.19, 0.035) 0.17 3.0e-3
Paired Associate Learning
(Verbal Declarative Mem.)
3+4 ECG+ (n=253 of 16422) 0.023 (-0.096, 0.14) 0.71 0.018
Pairs Matching
(Visual Declarative Mem.)
3+4 ECG+ (n=326 of 23489) -0.062 (-0.17, 0.048) 0.27 0.051
*
Beta coefficients represent the difference in cognitive function test scores (in units of standard deviation)
relative to the no AF/AFL group after adjustment for covariates.
†
P-values here are for the total association of ECG+ AF/AFL on each cognitive function test not adjusted for
multiple comparisons
‡
P-values here represent SEM analysis-based test of indirect effect of ECG+ AF/AFL on cognitive function
through the three brain structure measurements
Abbreviation: AF/AFL = Atrial fibrillation or flutter, Proc. = Processing, Mem. = Memory, Reas. = Reasoning
17
2.3.4 Interaction with age
An interaction of age and
AF/AFL history only (group 2) on
mean hippocampal volume and
fornix FA was found and is shown
in Figure 2.2. In younger
participants with only a history of
AF/AFL, hippocampal volume and
fornix FA were estimated to be
greater than for no AF/AFL group
(group 1), whereas in older
participants, these metrics were
less than group 1 (the inflection
point between approximately 65-
70 years).
2.3.5 Brain-wide analysis
The full-brain exploratory
analysis found associations of AF/AFL with cortical thickness, regional volume, and white matter
tract FA in numerous regions (FDR p<0.05). P-value and signed Cohen’s D maps are shown in
Figure 2.3. For cortical thickness, six left hemisphere regions were significantly associated with
AF/AFL (FDR p<0.05), including precentral, supramarginal, superior parietal, isthmus cingulate,
posterior cingulate, and superior frontal cortices. The cortical thickness associations in most
Figure 2.1: Mediation analysis for cognitive function
tests. (A-C) show path diagrams for tests with a
significant indirect association with AF/AFL through
brain structure. Numbers represent effect estimates. *
and solid lines represent paths with a p-value <0.05.
18
regions displayed larger Cohen’s D values in the ECG-detected AF/AFL groups (groups 3-4) than
in the history only group (group 2). Among subcortical, cerebellar, and ventricle volumes, the
bilateral thalamus, ventral diencephalon, amygdala, and hippocampus, the cerebellar WM, and left
lateral ventricle and third ventricle volumes were associated with AF/AFL. The Cohen’s D value
of these associations relative to group 1 were also larger in groups 3-4 than in group 2. Among the
WM FA values, the right fornix, posterior thalamic radiation (PTR), posterior limb internal capsule
(PLIC), and left corticospinal tract (CST) were associated with AF/AFL. For the PLIC and CST
FA, FA values were increased in groups 3-4 relative to group 1. For fornix FA, however, FA values
in groups 2-4 were decreased relative to group 1, and the PTR FA was decreased in groups 2-3.
2.3.6 AF/AFL medication analysis
Among the three AF/AFL groups, there were differing proportions of AF/AFL-related
medication usage (Table 2.1). The ECG with History group (group 4) had the highest self-reported
medication usage, while the ECG only group (group 3) had the lowest. The association of taking
Figure 2.2: Interaction of age and AF/AFL. Interactions are shown for hippocampal volume
(A), fornix FA (B), and AD cortical thickness signature (C), separately for the three AF/AFL
groups. First row shows the AF/AFL history only group, second row shows the ECG-only
AF/AFL group, and the third row shows the AF/AFL on ECG and history group. Abbreviation:
AF/AFL = Atrial fibrillation or flutter.
19
warfarin or an antiarrhythmic and hippocampal volume was significantly different in groups 1 and
4 (F(1,23877) = 4.13, p = 0.042). In group 1, participants reporting such medication use did not
display a significant difference in hippocampal volume relative to those not reporting such
medication use (β = -21.0mm
3
, 95% CI: -81.5, 39.5; P=0.50), while in group 4, such participants
Figure 2.3: Brain-wide AF/AFL associations. Association of AF/AFL with cortical thickness
(A), subcortical volume (B), and mean white matter tract FA (C) across the whole brain. FDR-
adjusted p-values and beta estimates for the three AF/AFL groups relative to the “None” group
are shown. Beta estimates are in units of standard deviation of the outcome measurement.
Abbreviations: PreCG = Precentral Gyrus, SupMarg = Supramarginal Gyrus, SupPar = Superior
Parietal Gyrus, SupFront = Superior Frontal Gyrus, IsthCing = Isthumus of Cingulate Gyrus,
PostCing = Posterior Cingulate Cortex, LatVent = Lateral Ventricle, VentDC = Ventral
Diencephalon, Amyg = Amygdala, Hipp = Hippocampus, CerebWM = Cerebellum White Matter,
3
rd
Vent = 3
rd
Ventricle, PTR = Posterior Thalamic Radiation, PLIC = Posterior of Limb of
Internal Capsule, Fornix = Fornix Crestria Terminalis, CST = Corticospinal Tract
20
had a significantly increased hippocampal volume (β=88.3 mm
3
, 95% CI: 1.9, 174.7; P=0.045).
This estimate was similar considering only those reporting a history of taking warfarin, though did
not reach statistical significance (β=85.4 mm
3
,
95% CI: -2.4, 173.3; P=0.057).
2.4 Discussion
In this neuroimaging-based analysis, we found that AF/AFL detected on ECG at the time
of MRI is associated with decreased hippocampal volume, disrupted fornix microstructure, and
cortical thinning in regions associated with AD. The finding for hippocampal volume corroborates
previous studies [34,51]. The magnitude of the differences was clinically meaningful. In context
of the association of age with each metric, the beta estimates for group 3 relative to group 1
correspond approximately to a 4.9-year difference for hippocampal volume, 1.7-year for fornix
FA, and 3.4-year for AD-CTS. For group 4 relative to group 1, the estimates corresponded to a
4.0-year, 2.3-year, and 3.0-year difference, respectively. Notably, associations were not found for
the AF/AFL history only group. Since participants who exhibit AF/AFL on ECG at MRI scanning
likely represent those with persistent AF/AFL and/or a greater AF/AFL burden, AF/AFL severity
may be an important modifier of the AF/AFL-brain structure association. This matches with
findings showing an association of AF with decreased cognitive function in those with persistent
AF and not paroxysmal AF [80], and greater hippocampal atrophy in persistent AF [59]. Our
findings therefore enrich evidence that AF/AFL correlates with early structural alterations in brain
regions affected in AD.
While the associations were strongest for ECG-detected AF/AFL (groups 3-4), the
interaction between age and AF/AFL group suggests that the association is also present in the
AF/AFL history-only group (group 2) depending on age. In this group (group 2), hippocampal
21
volume and fornix FA were lower than group 1 at old ages only. So, despite no significant main
effect in this population, the association of AF/AFL with brain atrophy may be present in older
populations. No significant interaction of age with the ECG-detected AF/AFL groups (groups 3
and 4) was found, though these groups may be underpowered for detecting such an interaction
effect.
Previous studies have shown that oral anticoagulation and catheter ablation of AF are
associated with reduced dementia incidence [52–54]. However, no studies have investigated
warfarin anticoagulation and antiarrhythmic agents in relation to AD-affected brain structure. We
found that within the group with AF/AFL history and on ECG (group 4), participants reporting
taking either warfarin or an antiarrhythmic agent had a greater hippocampal volume. We could not
determine whether AF treatment modified the AF/AFL association in other groups (groups 2-3)
due to the small number with medication use (especially group 3: <4%). One large limitation to
this analysis is the lack of data on newer novel oral anticoagulants (NOACs). NOACs were
available in the UK starting in 2008, prior to the imaging visit (2014-2019). It is possible then that
some participants were taking a NOAC medication for up to 11 years. However, in the United
Kingdom NOACs did not account for over half of first-time oral anticoagulant prescriptions until
2015 [81]. In addition, incorporating medication data from the initial visit (2006-2010) may
capture participants who switched from warfarin to a NOAC by the imaging visit.
In contrast to several studies that have shown an association of AF with cognitive decline
[49,80,82], we did not find an association here, possibly due to measurement error of the cognitive
function tests used (touch-screen-based and abbreviated) and also the participant’s ages (<81
years). However, the mediation analysis showed that hippocampal volume and fornix FA
significantly mediate the association of ECG-detected AF/AFL with AD domain cognitive
22
functions, suggesting that brain structural changes associated with AF/AFL have an impact on
cognitive function.
The exploratory full-brain analysis revealed many additional brain regional structural
measurements associated with AF/AFL, suggesting that while AF/AFL is associated with AD-
related regions, it is not specific to these regions. Interestingly, no temporal lobe regions were
associated with AF/AFL. Since AD is associated with temporal lobe cortical thickness, this may
indicate that AF/AFL-dementia link is not specific to AD [71]. Furthermore, our finding of
associations with enlarged ventricular volumes matches a previous study finding an association of
AF with qualitatively-graded ventricular size [33]. While one study found no association between
AF and global white matter FA [83], we found that AF was associated with FA in the fornix and
three other regions. Finally, ECG-detected AF/AFL was more strongly associated with brain
structural abnormalities for most regions, supporting that uncontrolled AF/AFL imposes a greater
burden on brain structure.
Strengths of our study include: 1) distinguishing between an AF/AFL medical history and
AF/AFL detected via ECG on the day of MRI; 2) evaluating the association of AF/AFL with
whole-brain structural features and cognitive function tests; 3) investigating whether brain
structure mediates cognitive performance declines with AF/AFL; and 4) assessing effects of
AF/AFL medications on hippocampal volume. Since this is an observational study, however,
causality remains to be clarified. Other limitations include the self-reported nature of the medical
history and medication data, the unknown onset and frequency of AF/AFL, and the relatively
modest number of participants with AF/AFL. There may also be selection bias as a result of the
UKB sample selection process.
23
This study helps to shed light on the pathophysiological mechanism through which brain
structural alterations may mediate the adverse effect of AF on neurocognitive decline. The results
corroborate and expand AF/AFL-brain structure associations, elucidate the pattern of AF/AFL
associations across the brain, and suggest that AF/AFL burden influences the magnitude of brain
structure differences. They also suggest that the association may vary with age, mediate cognitive
function differences, and be modified by AF/AFL medication usage. Future longitudinal studies
are required to determine whether AF/AFL treatment prevents brain atrophy. If the association is
found to be causal, AF/AFL treatment may become an important element of preventive brain
health. Overall, the evidence supports the hypothesis that AF/AFL has an important role in brain
health and aging beyond stroke risk.
24
Chapter 3
P-wave duration is associated with MRI brain structure measurements
A cross-sectional study of 22,850 UK Biobank participants
3.1 Introduction
The P-wave on an electrocardiogram (ECG) reflects the depolarization wave of atrial
contraction. In addition to being essential for heart rhythm evaluation, its morphology is
informative of atrial structural health and can be quantified with P-wave indices such as P-wave
duration (PWD) [84,85]. A prolonged PWD, typically defined as >110 or >120 ms [86], can result
from pathologic structural changes characteristic of atrial cardiomyopathy that may slow the
depolarization wave or increase the distance it needs to travel [87]. A shortened PWD can also
reflect pathologic changes and possibly be a precursor to prolonged PWD [88].
Abnormal PWD is a risk factor for atrial fibrillation (AF), an arrhythmia linked to stroke,
dementia, and cognitive decline [42,48,82,89,90]. Both a PWD of <89 ms and >112 ms have been
associated with increased AF risk, with a dose-response relationship for increasing PWD on AF
risk [88,91–93]. The atrial structural changes producing these PWD abnormalities may also be
detrimental independent of AF [84,94–96]. For instance, prolonged and shortened PWD have been
associated with ischemic stroke risk [32,91]. Prolonged PWD has been associated with death from
cardiovascular causes and all-cause mortality [84,91,97,97,98], increased dementia risk
independent of stroke and AF [95], and mild cognitive impairment in elderly adults [99]. These
studies collectively suggest a PWD of 90-110 ms as representing optimal atrial health.
25
Brain MRI imaging has been used to investigate how atrial changes contributing to an
abnormal PWD might affect the brain even in the absence of AF. For instance, a prolonged PWD
(>120 ms) was associated with MRI-detected cortical and lacunar infarcts, suggesting a possible
pathophysiological pathway of emboli resulting from atrial cardiomyopathy leading to brain
infarcts which hasten neurocognitive decline [35]. P-wave terminal force in lead V1 has also been
associated with brain infarcts and white matter hyperintensities, further suggesting that atrial
structural abnormalities are associated with vascular-related brain changes [100].
The broader associations of PWD and MRI-derived brain-wide thickness, volume, T2*
value, and white matter tract structural integrity, however, are unknown. We therefore sought to
investigate whether PWD is associated with these brain structure measurements independent of
atrial fibrillation and cardiometabolic risk factors using the UK Biobank (UKB) resource. We also
tested whether both a short and a long PWD are associated with brain structure. In addition, we
evaluated the association of cardiometabolic risk factors with PWD and whether PWD is a
substantial mediator of the association of these factors on brain structure.
3.2 Methods
3.2.1 Study population
We used data from the UK Biobank (http://www.ukbiobank.ac.uk), a population-scale
prospective cohort study in the United Kingdom with a half million participants [63]. The UKB
study received ethics approval and participants provided informed consent [64]. The initial
assessment visit, consisting of collection of a comprehensive set of health-related variables, took
place from 2006 to 2010. A subset of these participants attended an additional visit starting in 2014
during which imaging and cardiac data were obtained [65].
26
3.2.2 P-wave duration
During the imaging visit, a 10-second 12-lead ECG was obtained using the GE Cardiosoft
v6.0 system [65]. The Cardiosoft software used the GE Marquette 12SL 12-lead resting
interpretation algorithm to automatically derive ECG metrics, including ventricular rate and PWD.
This software forms a median complex from the complexes in the ECG trace for calculating ECG
metrics [101]. Quality control on these automated metrics consisted of excluding participants with
missing ECG data, ECGs the Cardiosoft software labeled as problematic (poor data quality,
suspected arm lead reversal, less than four QRS complexes detected, or undetermined rhythm), a
ventricular rate of <40 or >110 beats per minute, and outlier PWD values, defined as being outside
of 3 standard deviations of the sample PWD mean [91].
3.2.3 Brain MRI imaging and processing
Several MRI scans were obtained at the imaging visit [67]. From these scans, imaging-
derived phenotypes (IDPs) were calculated by the UKB after quality control [7]. Regional cortical
thickness in addition to subcortical, cerebellar, and brain stem volumes were derived from T1-
weighted MRI scans using the FreeSurfer software using the Desikan-Killiany parcellation [68].
White matter tract mean fractional anisotropy (FA) was derived from TBSS processing of diffusion
tensor imaging scans using the JHU ICM-DTI-81 atlas [69,70]. Median T2* values in subcortical
structures were generated from susceptibility-weighted MRI scans [7,9,67]. White matter
hyperintensity volume was calculated from both T1 and T2 FLAIR scans using the BIANCA tool
[7,102]. 152 different IDPs among these IDP types were evaluated.
27
3.2.4 Additional Inclusion and Exclusion Criteria
Participants with ECG-detected atrial fibrillation or flutter (AF/AFL) or a self-reported or
hospital-record based history of AF/AFL were excluded [66]. Participants were also excluded if
not of white British ancestry, missing data for PWD or any of the covariates, or reporting central
nervous system-related disease, dementia, or Parkinson disease. Participants were also removed
using the genetic relatedness matrix and the ukbtools R package to ensure that there were no pairs
with closer than 3
rd
degree relatedness [103].
3.2.5 Statistical analysis
Statistical analysis was conducted using R v3.6.3. We represented PWD as quintiles and
as restricted cubic splines (rcs) with 5 knots placed at the 5
th
, 27.5
th
, 50
th
, 72.5
th
and 95
th
percentiles
[78,104]. White matter hyperintensity was log-transformed, FA values were square-transformed,
and T2* values were cube-transformed to satisfy the normality of residuals assumption of
regression models. Sample characteristics were computed and compared by P-wave quintile group.
Linear regression models (rms library) testing the association of PWD with the 152 IDPs
were completed using rcs (5 knots) for PWD, rcs (3 knots) for age, and adjusting for the additional
covariates: scan site, scanner head and table position, head motion estimated from resting fMRI,
sex, maximum education level, self-reported or hospital record derived hypertension,
hypercholesterolemia, diabetes, coronary artery disease (CAD), heart failure (HF), chronic kidney
disease (CKD), stroke, tobacco smoking, and alcohol consumption frequency, body mass index
(BMI), and apoE-4 genotype [78]. Intracranial volume, modeled using rcs (3 knots), was included
as a covariate for all but the cortical thickness IDPs. IDP outliers, defined as outside 4 standard
deviations of the mean, were removed prior to regression. P-values from each regression model
28
were computed with the Wald test and corrected for multiple comparisons using false-discovery
rate (FDR) adjustment. Adjusted P-values for IDPs with an FDR-adjusted P-value <0.05 were
plotted using the ggseg library [79].
To visualize the relationship between PWD and the significantly associated IDPs (FDR
P<0.05), the average Z-scores of IDPs of the same type, scaled to a unit standard deviation, were
used as outcome variables in the same linear regression model and the model-predicted outcome
variable values were then plotted against PWD. Similar plots for IDPs individually were also
created. The association of PWD and the significant IDPs, along with the Cohen’s D effect size,
were also evaluated using PWD as a categorical variable based on quintiles, with the 60-80% group
as reference [76].
Additional analyses on these same averaged Z-scores investigated whether the association
of brain structure with PWD varies by age and sex. This was done using the same regression
models but separately with an additional interaction term of age and PWD and an interaction term
of sex and PWD, with the P-value of these interaction terms evaluated for statistical significance.
Asymmetry analysis for cortical thickness IDPs and amygdala T2* value IDPs were completed by
deriving variables representing IDP asymmetry for the outcome variable (the difference of the
right and left IDP divided by the sum of the right and left IDP, multiplied by 100) and using the
same adjusted regression model.
Several sensitivity analyses were completed. First, the same analyses were completed on a
“Normal ECG” subset, which included only ECGs labeled as being normal or otherwise normal
with tachycardia or bradycardia. Multinomial logistic regression adjusting for age and sex was
used to test if having an abnormal ECG flag was associated with a greater odds of being in an
abnormal PWD group. Next, the analyses were completed on a subset of participants without
29
hypertension, hypercholesterolemia, diabetes, CAD, HF, CKD, stroke, current tobacco smoking,
frequent alcohol consumption, obesity, or underweight. This subset is referred to as a “Control”
subset and represents a healthier subset of participants. Finally, the analyses were completed using
the PWD variable after regressing out ventricular rate, body surface area, and left ventricle stroke
volume. This was done using the residuals from a regression model for PWD including these
predictors modeled with rcs with 3 knots and each with an interaction term with sex.
To investigate which covariates are independently associated with PWD, multinomial
logistic regression was used with the categorical PWD variable as the outcome variable and age
(in decades), hypertension, hypercholesterolemia, diabetes, CAD, HF, CKD, stroke, smoking,
alcohol consumption frequency, and BMI as predictors, with additional adjustment for education,
apoE-4 genotype, and sex. The likelihood ratio test was used to calculate a P-value for each
predictor.
For covariates associated with PWD, mediation analysis to evaluate whether the
association of these covariates with brain structure is mediated by PWD was completed using the
mediation R package [105]. PWD was represented in separate mediation models as a binary
variable for short PWD (1 for 0-20% PWD and 0 otherwise) and long PWD (1 for 80-100% PWD
and 0 otherwise). The average Z-score of the FDR-significant IDP regions of the same type were
used as the outcome variables. Linear regression was used for the outcome variable model and
logistic regression was used for the mediator model, each of which adjusted for the aforementioned
covariates. The average direct effect (ADE) and average causal mediation effect (ACME) effect
estimates and P-values were computed for each mediation model based on 1000 simulations.
30
3.3 Results
3.1 Demographics and P-wave duration quintiles
After application of the exclusion criteria, the final data subset consisted of 22,850
participants with ages ranging from 45 to 81 years (mean age of 64.1 years, 53.4% female) and
PWD ranging from 50 to 146 ms (mean: 98.3 ms, SD: 15.7 ms). The PWD ranges for the quintile
groups were: 50-86 ms (0-20%), 86-96 ms (20-40%), 96-104 ms (40-60%), 104-112 ms (60-80%),
and 112-146 ms (80-100%). Sample characteristics by quintile group, in addition to tests of group
differences, are shown in Table 3.1. Among the sensitivity analysis subsets, the “Normal ECG”
subset consisted of 10,682 participants and the “Control” subset consisted of 9629 participants,
each with similar demographics.
Figure 3.1: Association of P-wave duration and brain structure. Associations are
shown for (A) cortical thickness, (B) subcortical, cerebellar, and ventricle volumes, and
(C) median T2* value in subcortical regions. IDPs for which the FDR-adjusted P-value
was <0.05 are plotted. Abbreviations: PreCG, Precentral Gyrus; ParsOp, Pars
Opercularis; Inf. Temp, Inferior Temporal; Amyg, Amygdala.
31
3.2 Association of P-wave duration and brain-wide structural measurements
Out of the 152 IDPs evaluated, 11 were found to be significantly associated with PWD
(FDR-adjusted P < 0.05), as shown in Figure 3.1. Significant volume IDPs included the bilateral
Table 3.1: Sample characteristics by P-wave duration group
P-Wave Duration Percentile Group
Characteristics
0-20%
(50-86 ms)
20-40%
(86-96 ms)
40-60%
(96-104 ms)
60-80%
(104-112 ms)
80-100%
(112-146 ms) p
1
Number of Participants 4777 4677 5450 4197 3749
Scan Site (%)
<0.001
Cheadle 3093 (64.7) 3185 (68.1) 3707 (68.0) 2830 (67.4) 2527 (67.4)
Reading 535 (11.2) 535 (11.4) 725 (13.3) 579 (13.8) 549 (14.6)
Newcastle 1149 (24.1) 957 (20.5) 1018 (18.7) 788 (18.8) 673 (18.0)
Sex = Male (%) 1950 (40.8) 1888 (40.4) 2419 (44.4) 2097 (50.0) 2298 (61.3) <0.001
Age 63.92 (7.69) 63.37 (7.55) 63.85 (7.23) 64.45 (7.36) 65.30 (7.18) <0.001
Education (%)
<0.001
College/University 2059 (43.1) 2077 (44.4) 2482 (45.5) 1994 (47.5) 1802 (48.1)
Vocational/Professional
2
1387 (29.0) 1377 (29.4) 1543 (28.3) 1162 (27.7) 1063 (28.4)
Secondary/Pre-University
3
998 (20.9) 936 (20.0) 1062 (19.5) 794 (18.9) 694 (18.5)
None of Above 333 (7.0) 287 (6.1) 363 (6.7) 247 (5.9) 190 (5.1)
Hypertension (%) 1299 (27.2) 1195 (25.6) 1495 (27.4) 1306 (31.1) 1326 (35.4) <0.001
Hypercholesterolemia (%) 722 (15.1) 689 (14.7) 858 (15.7) 696 (16.6) 744 (19.8) <0.001
Diabetes (%) 257 (5.4) 227 (4.9) 275 (5.0) 193 (4.6) 204 (5.4) 0.351
Coronary Artery Disease (%) 262 (5.5) 210 (4.5) 231 (4.2) 201 (4.8) 260 (6.9) <0.001
Heart Failure (%) 21 (0.4) 16 (0.3) 16 (0.3) 21 (0.5) 28 (0.7) 0.018
Chronic Kidney Disease (%) 29 (0.6) 25 (0.5) 37 (0.7) 24 (0.6) 39 (1.0) 0.043
Stroke (%) 68 (1.4) 44 (0.9) 55 (1.0) 45 (1.1) 59 (1.6) 0.022
Tobacco Smoking (%)
0.032
Never 3042 (63.7) 3020 (64.6) 3517 (64.5) 2682 (63.9) 2313 (61.7)
Previous 1531 (32.0) 1458 (31.2) 1729 (31.7) 1369 (32.6) 1291 (34.4)
Current 204 (4.3) 199 (4.3) 204 (3.7) 146 (3.5) 145 (3.9)
Alcohol (%)
<0.001
Infrequent 1312 (27.5) 1307 (27.9) 1464 (26.9) 1085 (25.9) 861 (23.0)
Occasional 2665 (55.8) 2602 (55.6) 3045 (55.9) 2337 (55.7) 2115 (56.4)
Frequent 800 (16.7) 768 (16.4) 941 (17.3) 775 (18.5) 773 (20.6)
BMI Category (%)
<0.001
Normal Weight 1921 (40.2) 2018 (43.1) 2326 (42.7) 1624 (38.7) 1343 (35.8)
Underweight 37 (0.8) 37 (0.8) 44 (0.8) 30 (0.7) 12 (0.3)
Overweight 1918 (40.2) 1860 (39.8) 2215 (40.6) 1776 (42.3) 1634 (43.6)
Obese 901 (18.9) 762 (16.3) 865 (15.9) 767 (18.3) 760 (20.3)
ApoE-4 (%)
0.311
0 3418 (71.6) 3372 (72.1) 3947 (72.4) 3010 (71.7) 2722 (72.6)
1 1259 (26.4) 1185 (25.3) 1370 (25.1) 1091 (26.0) 960 (25.6)
2 100 (2.1) 120 (2.6) 133 (2.4) 96 (2.3) 67 (1.8)
1
P-values representing differences between groups, with chi-squared tests used for categorical
variables and ANOVA tests used for continuous variables. Continuous variables are represented as
mean (standard deviation) and categorical variables represented as count (%).
2
Represents NVQ, HND, or HNC qualifications (or their equivalents) or professional qualifications
(nursing/teaching/etc.), without college/university degree
3
Represents O levels, GCSDs, A levels, AS levels, or CSEs (or their equivalents, without
college/university/vocational degree)
32
thalamus, brain stem, bilateral
amygdala, left pallidum, and right
putamen. Significant mean thickness
IDPs were the left pars opercularis of
the inferior frontal gyrus, left
precentral, left inferior temporal
gyri. The one significant median T2*
value IDP was the left amygdala.
Five of these IDPs reached the
Bonferroni correction significance
threshold: right thalamus volume,
brain stem volume, median
amygdala T2* value, pars
opercularis thickness, and left
amygdala volume.
The relationships between
PWD and these 11 IDPs, with IDPs
of the same type averaged together,
are shown in Figure 3.2. Individual
plots for each significant IDP are
shown in Figure 3.3. Collectively,
both abnormally short and long
PWD had reduced IDP values
Figure 3.2: Nonlinear relationship between PWD
and brain structure. Plots of PWD and the average of
the IDPs significantly associated PWD for (A)
thickness, (B) volume, (C) and median T2* value in the
left amygdala. Gray around mean estimates represent
95% confidence limits.
33
relative to a normal PWD. The associations of these 11 IDPs and PWD using PWD as a categorical
variable are shown in Table 3.2. The asymmetry analyses for cortical thickness IDPs and median
T2* values in the bilateral amygdala found no significant asymmetry after FDR correction for
multiple comparisons. We also found no significant interactions with age or sex in the association
of PWD and the averaged Z-scores of significant IDPs of the same type.
Figure 3.3: Individual PWD association plots. The relation between PWD and the 11 IDPs
found the be significantly associated with PWD. Gray bars represent 95% confidence limits.
34
We found that an abnormal ECG label was associated with PWD quintile group (χ
2
(4) =
73.8, P=3.6e-15). Relative to the 60-80% group, the odds ratio (OR) of an abnormal ECG being
in the 0-20% group was 1.11 (95% CI: 1.02, 1.21; P=0.014); the 20-40% group, 0.86 (95% CI:
0.79, 0.94; P=4.3e-4); the 40-60% group, 0.94 (95% CI: 0.87, 1.02; P=0.14); and the 80-100%,
1.20 (95% CI: 1.10, 1.31; P=6.3e-5). The most frequent labeled ECG abnormalities were “possible
left atrial enlargement” (1155 participants), “marked sinus bradycardia (1053 participants), and
“left axis deviation” (335 participants). The brain-wide association analysis of PWD in the
“Normal ECG” subset found two significant IDPs: the brainstem (FDR P=0.03) and right
amygdala volume (FDR P=0.03). Short PWD had decreased volume and thickness, and long PWD
had a decreased T2* value in the left amygdala.
Table 3.2: Association of IDP with P-wave duration group for FDR-significant IDPs
P-Wave Duration Percentile Group:
0-20% 20-40% 40-60%
60-
80% 80-100%
IDP ROI Beta d
1
Beta d Beta d Beta Beta d
P-
Value
2
Thickness
(mm)
Left Pars
Opercularis
-0.0080*
(-0.014, -0.0021)
-0.054
-0.0089*
(-0.015, -0.0030)
-0.060
-0.0075*
(-0.013, -0.0017)
-0.050 Ref
0.0012
(-0.0051, 0.0075)
7.9e-3 8.2e-04
Left Precentral
-0.0093*
(-0.017, -0.0021)
-0.051
-0.0088
(-0.016, -0.0016)
-0.048
-0.0056
(-0.013, 0.0014)
-0.030 Ref
0.0043
(-0.0034, 0.012)
0.023 0.0010
Left Inferior
Temporal
-0.0073
(-0.013, -0.0013)
-0.050
-0.0046
(-0.011, 0.0013)
-0.032
-0.0047
(-0.010, 0.0011)
-0.032 Ref
0.0047
(-0.0016, 0.011)
0.032 0.0014
Volume
(mm
3
)
Right Thalamus
-30.72*
(-49.63, -11.81)
-0.046
-10.89
(-29.90, 8.11)
-0.016
-9.70
(-28.00, 8.59)
-0.014 Ref
14.71
(-5.37, 34.78)
0.022 1.8e-04
Brain Stem
-85.95
(-158.9, -12.97)
-0.035
-103.8*
(-177.1, -30.43)
-0.042
-10.27
(-80.89, 60.34)
-4.2e-3 Ref
64.95
(-12.52, 142.4)
0.026 4.6e-05
Left Amygdala
3.16
(-4.17, 10.49)
0.014
2.54
(-4.82, 9.91)
0.011
-4.83
(-11.92, 2.26)
-0.022 Ref
-8.70
(-16.48, -0.92)
-0.039 0.0073
Left Thalamus
-32.06*
(-52.76, -11.36)
-0.045
-13.53
(-34.33, 7.27)
-0.019
-14.21
(-34.24, 5.82)
-0.020 Ref
3.25
(-18.72, 25.22)
4.6e-3 0.0089
Right Amygdala
1.30
(-5.85, 8.46)
5.8e-3
3.01
(-4.18, 10.21)
0.013
-1.75
(-8.67, 5.18)
-7.7e-3 Ref
-10.09*
(-17.69, -2.49)
-0.045 0.0085
Left Pallidum
-12.48*
(-20.03, -4.92)
-0.055
-10.16*
(-17.76, -2.57)
-0.045
-2.36
(-9.67, 4.96)
-0.010 Ref
-2.88
(-10.90, 5.14)
-0.013 0.0029
Right Putamen
-27.20*
(-45.31, -9.10)
-0.051
-25.75*
(-43.95, -7.56)
-0.048
-8.31
(-25.83, 9.21)
-0.015 Ref
-19.87
(-39.09, -0.65)
-0.037 0.011
T2* Value Left Amygdala
1.32
(-1143.2, 1145.8)
4.7e-5
432.8
(-717.3, 1582.8)
0.015
258.6
(-849.7, 1366.8)
9.2e-3 Ref
-1344.5
(-2560.3, -128.6)
-0.048 0.038
1
Cohen’s D effect size
2
P-values are from ANOVA test evaluating if there is a difference among the P-wave duration groups
*Represents a P-value of <0.0125 (0.05/4) for the post-hoc comparison relative to the reference group (60-80%)
35
For the sensitivity analysis using the “control” dataset, the left amygdala volume (FDR
P=0.0096) and median T2* value (FDR P=0.022) were associated with PWD. The previous
relationship between thickness and PWD was not present, but short PWD did have decreased
volume and amygdala T2* value and high PWD had decreased amygdala T2* value. For the
sensitivity analysis using the adjusted PWD, 9 IDPs were significantly associated with PWD:
median T2* in the left amygdala, volumes of the bilateral amygdala, right thalamus, and left
pallidum, and thicknesses of the left pars opercularis, left inferior temporal, left superior parietal,
and left precentral gyrus.
3.3 Association of covariates with P-wave duration
Adjusted for the set of other covariates evaluated, age, hypertension, and BMI were
significantly associated with PWD after correction for multiple comparisons. Nominally
significant covariates included diabetes and CAD. The odds ratios of being in a specific PWD
percentile group for each variable are shown in Table 3.3. Age was associated with increased odds
Table 3.3: Association of covariates with P-wave duration group
Odds Ratio (95% CI) for each P-Wave Duration Percentile Group
Variable 0-20% 20-40% 40-60% 60-80% 80-100% P-Value
1
Age (per 10 years) 0.92* (0.87, 0.98) 0.84* (0.79, 0.89) 0.90* (0.85, 0.95) Ref 1.13* (1.06, 1.21) 7.0e-22
HTN 0.84* (0.76, 0.93) 0.83* (0.75, 0.92) 0.89 (0.81, 0.98) Ref 1.05 (0.95, 1.16) 1.5e-06
Hypercholesterolemia 0.94 (0.83, 1.07) 1.02 (0.90, 1.16) 1.05 (0.93, 1.18) Ref 1.05 (0.92, 1.19) 0.41
Diabetes 1.31* (1.07, 1.59) 1.26 (1.03, 1.55) 1.25 (1.03, 1.52) Ref 1.01 (0.82, 1.25) 0.011
CAD 1.36* (1.11, 1.66) 1.16 (0.94, 1.43) 0.99 (0.81, 1.22) Ref 1.20 (0.98, 1.47) 0.0072
HF 0.82 (0.44, 1.53) 0.74 (0.38, 1.44) 0.64 (0.33, 1.24) Ref 1.13 (0.63, 2.02) 0.40
CKD 1.07 (0.62, 1.85) 1.04 (0.59, 1.84) 1.28 (0.76, 2.16) Ref 1.60 (0.96, 2.69) 0.32
Stroke 1.44 (0.98, 2.12) 1.0 (0.65, 1.52) 1.02 (0.68, 1.52) Ref 1.25 (0.84, 1.85) 0.19
Smoking - Previous 1.01 (0.92, 1.11) 1.01 (0.92, 1.11) 1.00 (0.92, 1.10) Ref 1.01 (0.92, 1.12)
0.73
Smoking - Current 1.23 (0.99, 1.54) 1.22 (0.98, 1.53) 1.07 (0.86, 1.34) Ref 1.09 (0.86, 1.38)
Alcohol Consumption - Occasional 1.01 (0.92, 1.12) 0.97 (0.88, 1.07) 0.99 (0.90, 1.09) Ref 1.08 (0.97, 1.21)
0.33
Alcohol Consumption - Frequent 0.97 (0.85, 1.11) 0.92 (0.81, 1.06) 0.96 (0.85, 1.09) Ref 1.12 (0.97, 1.29)
BMI - Underweight 0.98 (0.60, 1.59) 0.93 (0.57, 1.51) 1.0 (0.62, 1.59) Ref 0.55 (0.28, 1.08)
1.6e-08 BMI - Overweight 0.96 (0.87, 1.06) 0.89 (0.81, 0.98) 0.89 (0.82, 0.98) Ref 1.02 (0.92, 1.13)
BMI - Obese 1.0 (0.88, 1.13) 0.80* (0.71, 0.91) 0.78* (0.69, 0.88) Ref 1.15 (1.01, 1.31)
1
P-values are from ANOVA test evaluating if there is a difference among the P-wave duration groups
*Represents a P-value of <0.0125 (0.05/4) for the post-hoc comparison relative to the reference group (60-80%)
Abbreviations: CAD, Coronary Artery Disease; HF, Heart Failure; CKD, Chronic Kidney Disease; BMI, Body Mass Index
36
of being in the highest PWD percentile group (80-100%), with an odds ratio of 1.13 (95% CI: 1.06,
1.21) per decade. Hypertension was also associated with greater odds of being in the long PWD
group. For BMI, there was decreased odds of obese participants being in 20-40% or 40-60% groups
relative to 60-80% group.
3.4 P-wave duration mediation analysis
We found no significant indirect effects for which PWD (short or long) mediated the
association of age, hypertension, BMI, or CAD on either of the averaged significant thickness,
volume, and T2* IDPs after correction for multiple comparisons. The estimated ADE and ACME
values are shown in Table 3.4. There were 4 nominally significant mediation effects, each
involving CAD and hypertension, with the proportion of mediation ranging from 1-3%.
3.4 Discussion
In this analysis, we used the UKB resource to investigate the association of PWD and brain
structure. We found that PWD was significantly associated with 11 IDPs, with an abnormally long
Table 3.4: Mediation analysis of covariates and brain structure by P-wave duration
Mediator: Low P-Wave Duration (0-20%) Mediator: High P-Wave Duration (80-100%)
Variable
1
IDP
2
ADE P ACME P ADE P ACME P
Age Thickness -0.035 <0.001 -2.7e-4 0.92 -0.035 <0.001 4.1e-4 0.36
Volume -0.057 <0.001 -1.3e-4 0.93 -0.057 <0.001 -1.7e-5 0.89
T2Star -7.2e-3 <0.001 1.1e-5 0.96 -7.1e-3 <0.001 -3.5e-5 0.40
BMI Thickness 0.016 <0.001 -2.4e-4 0.65 0.016 <0.001 9.1e-6 0.37
Volume 9.8e-3 <0.001 -2.2e-4 0.65 9.8e-3 <0.001 8.9e-6 0.84
T2Star -0.013 <0.001 -2.6e-6 0.97 -0.012 <0.001 -7.2e-5 0.40
CAD Thickness 0.055 0.10 -1.6e-3 0.020 0.052 0.15 9.0e-4 0.32
Volume -4.9e-3 0.82 -9.9e-4 0.066 -5.7e-3 0.81 -4.9e-5 0.84
T2Star -0.033 0.28 9.8e-5 0.87 -0.032 0.31 -5.6e-4 0.38
HTN Thickness -0.075 <0.001 7.5e-4 0.022 -0.076 <0.001 1.6e-3 0.0040
Volume 0.013 0.34 5.8e-4 0.068 0.014 0.35 -1.2e-4 0.75
T2Star -0.034 0.044 -5.2e-5 0.87 -0.033 0.026 -1.0e-3 0.018
1
Independent variable in mediation model
2
Outcome variable for the mediation model, representing the average Z-score of FDR-significant regions of the
specified IDP type
37
and short PWD estimated to have reduced IDP values for many of these IDPs. While a prolonged
PWD (>120 ms) has been associated with MRI-detected cortical and lacunar infarcts [35], our
finding of an association between abnormally short PWD and MRI-derived markers is novel. A
short PWD being associated with structural brain changes is consistent with its associations with
atrial fibrillation and death from cardiovascular causes [88,91].
The pattern of IDPs associated with PWD suggests that abnormal PWD may be associated
with vascular-type brain damage. For instance, prolonged PWD is associated with lacunar infarcts
[35], which most frequently occur in regions supplied by small perforating arteries [106]. These
regions include many of the volume IDPs associated with PWD: the thalamus, putamen, pallidum,
and brain stem [107–110]. Reduced volume in these regions may reflect the presence of lacunar
infarcts [111,112]. Amygdala volume was also found to be associated with PWD, though not
typically affected by lacunar infarcts [110]. It has, however, been found to be reduced in patients
with a history of stroke and transient ischemic attack [113] and is among the regions implicated in
vascular dementia [114,115]. The left amygdala was the only T2* IDP associated with PWD, with
a prolonged PWD having reduced median T2* value. Since a decreased T2* value can result from
local iron deposition, such as from a microbleed, the amygdala volume association may be related
to such pathology [9,116]. In contrast, white matter hyperintensity, a marker of cerebrovascular
disease, was not found to be associated with PWD, corroborating a separate study [35]. Overall,
though, the IDPs associated with PWD better matches a vascular brain injury pattern than an
Alzheimer disease atrophy pattern, which would likely present with lateral ventricle expansion and
hippocampal atrophy [117].
For the thickness IDPs associated with PWD, only a short PWD was estimated to have
decreased thickness. These IDPs may also be indicative of vascular-type brain atrophy. For
38
example, subcortical vascular dementia has been associated with inferior temporal gyrus atrophy
[118] and subcortical vascular mild cognitive impairment has been associated with inferior frontal
gyrus atrophy [119]. Furthermore, precentral gyrus atrophy rates have been found to be increased
in those with microinfarcts [120]. Finally, the association with the inferior temporal gyrus is
notable in that the inferior temporal gyrus is affected in mild cognitive impairment and
Alzheimer’s disease [121]. An Alzheimer disease-related IDP being associated with PWD is
consistent with PWD being associated with dementia risk [95].
Brain structure differences between the 60-80% group (104-112 ms) and the 40-60% group
(96-104 ms) were not substantial. The findings therefore corroborate results finding that in general
a PWD of <90 ms or >110 ms is associated with adverse health events [91]. The largest effect size
(Cohen’s D) for a PWD group relative to the 60-80% PWD group was 0.06. This is small, though
not unexpectedly small given that this sample has a maximum age of 81 and excludes participants
with dementia or AF. The small effect size also does not rule out the possibility of a substantial
number of microscopic brain changes, such as microinfarcts undetectable on MRI, in those with
abnormal PWD.
The association of PWD with brain stem and right amygdala volumes in the “Normal ECG”
subset helps to rule out the possibility that the association of PWD and brain structure is
confounded by comorbid heart pathology. The association of PWD with left amygdala volume and
T2* value in the “Control” subset suggests that PWD is associated with brain structure even in
those without cardiometabolic disease. Furthermore, since the relation between PWD and brain
structure did not change with the adjusted PWD variable, brain-heart allometry is unlikely to be
driving the PWD-brain structure associations.
39
Age, hypertension, and BMI were each found to be associated with PWD in this study.
Increasing age and hypertension have both been associated with increased PWD [85,122,123]. In
addition, not only is obesity associated with a longer PWD [124], but also studies have found that
weight loss leads to a decrease in PWD [125,126]. The association of CAD with PWD was only
nominally significant in this study, though other studies have found a significant association of
CAD with prolonged PWD [127,128].
Each of the variables associated with PWD are themselves associated with brain structure
[129–132]. In the causal mediation analysis, however, we did not find evidence that the association
of these variables and brain structure is mediated by an abnormal PWD. For nominally significant
indirect effects, the mediation by P-wave duration represented a very small percentage of the total
effect. So, while PWD is associated with brain structure and age, hypertension, and BMI, it is not
a substantial mediator of the association of these variables on brain structure.
Study strengths include the large sample size, the breadth of imaging-derived brain
structure measurements, and the availability of health-related variables. However, the study does
not inform on causality. While there are plausible pathways for which atrial cardiomyopathy,
reflected in PWD, leads to adverse brain changes, brain changes could also be impacting heart
function [133,134]. In addition, there could be a common cause of both heart and brain changes,
such as a tendency to form clots. Finally, there may be residual confounding. For example, those
with paroxysmal and undiagnosed AF/AFL may not be captured by the AF/AFL variable. The
association of PWD and brain structure could therefore be driven by participants with AF/AFL.
Overall, this study found that both an abnormally short and long PWD is associated with
brain structure, suggesting that atrial cardiomyopathy may have a role in brain health independent
of atrial fibrillation and other cardiometabolic risk factors. Future work includes longitudinal
40
studies of how brain structure changes in those with abnormal PWD and examining ECG indices
other than PWD. This study adds to the collective evidence that structural heart changes are
associated with adverse brain changes.
Given that PWD is a risk factor for atrial fibrillation, it is interesting that several IDP were
associated with both PWD and AF/AFL (Chapter 2). These matching IDPs included left precentral
gyrus thickness, thalamus, amygdala, and brain stem volumes. That these same IDPs were found
to be associated with both PWD and AF/AFL could indicate that the abnormal PWD group has an
overrepresentation of those with paroxysmal AF/AFL, a type of AF/AFL that would not have been
excluded in the P-wave duration study due to how AF/AFL was defined (history and ECG
detection at imaging visit). Though, there are a number of other possibilities for why these same
IDPs were found. Regardless, these four IDPs being found associated with both PWD and AF/AFL
suggest that they are particular liable to atrophy in the presence of cardiac pathology. Future work
will be needed to investigate what changes are mediating these thickness and volume differences.
Initial features to investigate within these regions include microbleed and microinfarcts.
41
Chapter 4
Interaction effect of alcohol consumption and Alzheimer disease polygenic risk
score on the brain cortical thickness of cognitively normal subjects
4.1 Introduction
Alcohol consumption is known to be associated with structural brain health [135]. Genetic
factors associated with Alzheimer’s disease (AD) are also known to be associated with brain
structure, even in cognitively healthy subjects [136]. However, the isolated presence of alcohol
consumption or genetic risk factors does not guarantee an increased level of brain atrophy or
dementia risk. It is possible then that the effect of alcohol consumption on the brain is modified
by the presence of genetic risk variants for AD.
Results from several studies have suggested the existence of an interaction effect of alcohol
consumption and AD genetic risk on cognition, brain structure, and dementia risk, though the
nature of this interaction is unclear. Some studies have found an increased risk of dementia and
poorer cognitive health with alcohol consumption among those at elevated risk for AD. For
instance, a study of midlife alcohol consumption and dementia risk by Anttila et al found an
interaction of alcohol consumption and apoE-4 on dementia risk, with the dementia risk of apoE-
4 carriers increased with increasing amounts of alcohol consumption [137]. Another study found
a similar relationship for cognitive abilities, in which alcohol consumption was associated with an
increased risk of cognitive decline among apoE-4 carriers, but a decreased risk in apoE-4
noncarriers [138].
Other studies, however, have suggested increased brain health with light-to-moderate
alcohol consumption in subjects with the AD genetic risk factor apoE-4. In a study by den Heijer
42
et al, for example, the evidence suggested an interaction effect of alcohol consumption and apoE
genotype on the hippocampal volume in cognitively healthy adults. In apoE-4 carriers, increased
alcohol consumption was found to be associated with increased hippocampal volume [139]. A
study of cognitive function in elderly men found that light alcohol consumption was associated
with increased cognitive function, and that this increase in cognitive function with light alcohol
consumption was even greater in apoE-4 carriers [140]. In a longitudinal study of cognition, a
more rapid decline in memory for carriers of apoE-4 relative to noncarriers was found only among
nondrinkers [141]. However, several studies have not found any such interaction effect [141–144].
Somewhat contradictory study results investigating alcohol consumption and AD genetic
risk alone also suggest the possibility of interactions with genetic variants and alcohol
consumption. For instance, while some studies show that alcohol in moderate quantities is
beneficial to the brain, others show that all levels of consumption are detrimental. One study
showed that alcohol consumption at all levels was associated with increased hippocampal atrophy
[135]. Another, however, showed that moderate consumption of alcohol had possible benefits to
the white matter of the brain [145]. Additionally, a study by Li et al found that cognitively healthy
subjects carrying the apoE-4 allele had a greater cortical thickness in AD-associated brain regions
than that of non-carriers, opposite to the expected trend of apoE-4 carriers having a decreased
cortical thickness [146].
Cell-based and mouse studies give further evidence of an interaction of alcohol
consumption and AD genetic risk factors. For instance, cell-based studies have found that apoE-4
synergistically increases the neurotoxicity of alcohol consumption, with apoE-4 leading to
increased oxidative stress within cells and apoptosis in the presence of high concentrations of
alcohol [147]. In addition, a recent mouse study found that low alcohol exposure has beneficial
43
effects on clearance of waste products in the brain [148]. Since AD genetic risk factors such as the
ApoE-4 allele are associated with impaired clearance of amyloid beta [149], low to moderate
alcohol consumption could potentially then be beneficial for those with the ApoE-4 allele if these
effects in mice also are present in humans.
Discrepant results in the literature may also be due to a combination of limited sample sizes
and unaccounted modifier effects. Studies of the effects of alcohol consumption on brain structures
have typically been assessed using limited sample sizes (~500 to 3000 participants) [24,139,150–
152]. In addition to the inconsistent results of the long-term effects of light-to-moderate alcohol
consumption on brain structure, existing results of AD-related genes’ influences on brain structure
in normal aging are often conflicting [135,153,154]. This is likely also due to a lack of sufficient
statistical power with which to detect subtle genetic effects, especially in non-pathological
populations.
We therefore sought to investigate whether the effect of alcohol consumption on regions
of the brain known to be affected in early AD varies based on AD genetic risk. Understanding
whether such an interaction exists is important in better understanding how alcohol consumption
leads to structural brain changes and increases in dementia risk. Our hypothesis was that there is
an interaction effect of a 33-SNP AD polygenic risk score and alcohol consumption on the cortical
thickness of AD-associated regions, known as the AD Cortical Thickness Signature. Data from
6213 cognitively healthy subjects from the UK Biobank were used to investigate this hypothesis.
The large sample size offered by the UK Biobank enables sufficient statistical power for detecting
such an interaction effect.
44
4.2 Methods
4.2.1 Study Population
To investigate these hypotheses, brain imaging, genetics, and health data for cognitively
healthy subjects were obtained from the UK Biobank. The UK Biobank is a large study enrolling
about 500,000 participants in the United Kingdom from 2006 to 2010 with baseline ages ranging
from 40-69 years (http://www.ukbiobank.ac.uk). Written informed consent was provided by all
UK Biobank subjects. Ethics approval for the UK Biobank was provided by the North West Multi-
Centre Research Ethics Committee [64]. The selection of only cognitively normal subjects with
appropriate imaging data followed the same procedure detailed in Zhao et al [155]. Briefly, only
those subjects with brain MRI data that passed quality control and image processing, without a
reported neurological or psychiatric disorder, with white British ancestry, and with a passable
genetic sample quality were included in the study. Beyond this selection process, only subjects
with complete cases for all of the studied variables were included in the final study data set.
Subjects with a value of “prefer not to answer” for the educational history categorical variable
were removed. In addition, subjects with a documented past medical history of heart attack or
stroke were removed from the study.
4.2.2 Alzheimer Disease Cortical Thickness Signature and Hippocampal Volume
Imaging data was processed using FreeSurfer v6.0 (https://surfer.nmr.mgh.harvard.edu/)
to extract region-of-interest based cortical thickness measurements and subcortical volumes [68].
Details are equivalent to those in Zhao et al [155]. We chose to investigate a set of regions
previously shown to be affected early in AD pathogenesis and that can be automatically measured
using FreeSurfer v6.0. Since the study population consists of only cognitively healthy subjects, we
45
expected that any changes in brain structure due to the factors studied were more likely to be in
these regions of brain than other regions of the brain. An AD Cortical Thickness Signature was
created based on that created by Sabuncu et al [156]. We selected a subset of these regions to
include in the AD Cortical Thickness Signature we studied. In particular, we omitted regions with
a tendency to have segmentation errors that when not manually corrected are more likely to lead
to altered study results [157]. So, the final outcome variable in this study consisted of the average
bilateral thickness of the superior temporal, middle temporal, inferior temporal, inferior parietal,
and posterior cingulate brain regions. Left and right hippocampal volumes were averaged to give
the average bilateral hippocampal volume. Due to the large sample size, scan-by-scan quality
control of the FreeSurfer output was unfeasible. Extreme outliers were examined, two of which
were removed due to severe errors.
4.2.3 AD Polygenic Risk Score
A weighted polygenic risk score (PRS) was computed for all subjects based on the 33 single
nucleotide polymorphisms (SNPs) that constitute the AD polygenic hazard score developed by
Desikan et al [158]. Two of the SNPs making up this PRS are the two SNPs that determine apoE
genotype, rs7412 and rs429358. Four SNPs in this polygenic hazard score were not available in
the genetics data of the UK Biobank study population. These SNPs were replaced by alternative
SNPs. The SNP in the abParts locus, rs7145100, was replaced by the SNP rs7146073, a SNP in
high linkage disequilibrium with rs7145100 (r
2
=1) based on the CEU population of the 1000
Genomes Project, Phase 3 [159]. The SNP in the CLU locus, rs9331888, was replaced by the SNP
rs11136000, another SNP in the CLU locus that has been found to be significantly associated with
progression from mild cognitive impairment to dementia [160]. The two SNPs of the polygenic
46
hazard score within the ABCA7 locus, rs3752246 and rs7408475, were replaced by two other
SNPs in the ABCA7 locus, SNPs rs3764650 and rs4147929, that have also been found to be
associated with AD [161].
Additionally, apoE-4 carrier status and a 31-SNP PRS not including the apoE-4 SNPs were
computed. ApoE genotype was determined from the genotyping of its two SNPs, rs7412 and
rs429358. It was assumed that subjects heterozygous at each of these SNPs had the apoE-2,4
genotype given the rarity of the apoE-1,3 genotype [162]. A 31-SNP polygenic risk score was
computed the same way as the 33-SNP PRS, except for not including the two apoE genotype SNPs.
The PRSice-2 software (version 2.1.0.beta, https://choishingwan.github.io/PRSice/) was
used for computing the polygenic risk scores [163]. Effect sizes used as weights in the computation
of these risk scores were obtained from The International Genomics of Alzheimer’s Project
(IGAP). The “International Genomics of Alzheimer's Project (IGAP) is a large two-stage study
based upon genome-wide association studies (GWAS) on individuals of European ancestry. In
stage 1, IGAP used genotyped and imputed data on 7,055,881 single nucleotide polymorphisms
(SNPs) to meta-analyse four previously-published GWAS datasets consisting of 17,008
Alzheimer's disease cases and 37,154 controls (The European Alzheimer's disease Initiative –
EADI the Alzheimer Disease Genetics Consortium – ADGC The Cohorts for Heart and Aging
Research in Genomic Epidemiology consortium – CHARGE The Genetic and Environmental Risk
in AD consortium – GERAD). In stage 2, 11,632 SNPs were genotyped and tested for association
in an independent set of 8,572 Alzheimer's disease cases and 11,312 controls. Finally, a meta-
analysis was performed combining results from stages 1 & 2” [164].
47
4.2.4 Health Data Variables
Alcohol consumption data was obtained in the same way as in Zhao et al and was treated
as a categorical variable using the following grouping: 0, <1, 1-6, 6-12, 12-24, 24-48, and >48
g/day of alcohol consumption [155]. Alcohol consumption frequency was used in some analyses
as a categorical variable with the values of “Never”, “Special occasions only”, “One to three times
a month”, “Once or twice a week”, “Three or four times a week”, and “Daily or almost daily”.
Body mass index (BMI) was also treated as a categorical variable. Subjects were defined as being
underweight (BMI of <18.5 kg/m
2
), normal weight (BMI of 18.5-24.9 kg/m
2
), overweight (BMI
of 25-29.9 kg/m
2
), or obese (BMI of >30 kg/m
2
). Underweight subjects were removed for all
analyses because of their small sample size (n=37) and the potential source of bias from their
inclusion, given that weight loss can be indicative of poor health and has been shown to precede
diagnosis of AD [165]. Variables representing hypertension and diabetes represent a self-reported
previous diagnosis of hypertension and diabetes, respectively. The extent of control of
hypertension and diabetes with medication is not accounted for in this study. Tobacco smoking is
represented by smoking intensity, calculated by dividing pack years by the number of years
smoked, yielding an average measure of packs per day.
4.2.5 Statistical Analysis
Mean and standard deviation for continuous variables and counts and percentages for
categorical variables were computed to characterize the study population. Tests for differences in
each variable among the alcohol consumption groups were conducted using a chi-squared test for
categorical variables and ANOVA for continuous variables. Linear models were used to test each
research question. Covariates included in every model that were selected a priori included sex,
48
age, education, tobacco smoking, alcohol consumption, hypertension, diabetes, BMI, AD genetic
risk, and the first 5 principal components (PC1-PC5) estimating population substructure. For
models of average hippocampal volume only, intracranial volume was also used as a covariate.
All continuous variables were mean-centered, except for intracranial volume (ICV), which was
median-centered. The reference level for daily alcohol consumption was set to “1-6 g/day” and the
reference level for alcohol consumption frequency was set to “Once or twice a week”. The
reference level for BMI was set to “Normal weight” and the reference level for the education
variable was set to “College or University degree”. When plotting the mean AD Cortical Thickness
Signature and hippocampal volume values for the various alcohol consumption and genetic risk
groups, fixed values for the covariates were used. In particular, the mean outcome variables were
evaluated at the reference levels for categorical variables, the median of ICV, the zero value for
the hypertension and diabetes variables, and the mean value for the remainder of the continuous
covariates.
Linear regression was first used to investigate the effect of alcohol consumption on both
AD Cortical Thickness Signature and mean hippocampal volume without consideration of an
interaction with genetic risk for AD. The effect of alcohol consumption was also examined among
only those who consume alcohol at least once per week, with adjustment for alcohol consumption
frequency. The associations of the outcome variables with alcohol consumption frequency were
also investigated. Linear regression was next used to test the hypothesis that there is an interaction
effect of alcohol consumption and the 33-SNP AD PRS on AD Cortical Thickness Signature. The
interaction effect was also investigated for average hippocampal volume. Each alcohol interaction
was additionally evaluated among only subjects who consume alcohol at least once per week, with
adjustment for alcohol frequency. A model with a triple interaction term for alcohol consumption
49
amount, alcohol consumption frequency, and AD 33-SNP PRS was used to investigate whether
the interaction varied among alcohol consumption frequency groups. Finally, a model with a triple
interaction term for grouped ages (45-60, 60-70, and 70-80 years old), alcohol consumption
amount, and AD 33-SNP PRS was used to test for differences in the interaction effect among
different age groups. P-values were obtained via either a likelihood ratio test (LRT) comparing the
model with the interaction term of interest to the reduced model without this interaction term or
an F-test from the ANCOVA model. The alpha level for determining statistical significance was
set to 0.05. Standard regression diagnostics were completed to ensure that assumptions for linear
regression were met and to determine whether any one data point would significantly alter the
result.
Since interactions between covariates and the modifiable risk factor of interest, in addition
to interactions between covariates and the genetic factor of interest, can confound the interaction
term of interest, the impact of the inclusion of such plausible interactions was investigated in a
separate modeling step, referred to as the fully-adjusted models [166]. For the model testing the
interaction of alcohol consumption and the 33-SNP AD PRS, these interaction terms include the
alcohol-by-sex, alcohol-by-age, alcohol-by-smoking, alcohol-by-hypertension, alcohol-by-
diabetes, alcohol-by-BMI, PRS-by-sex, PRS-by-age, PRS-by-smoking, PRS-by-hypertension,
PRS-by-diabetes, and PRS-by-BMI terms. The written-out model can be found in the
supplementary material.
For investigating the nature of the interaction effect for any significant interactions,
separate models were fit using categorical variables for the AD 33-SNP PRS. Specifically, the AD
33-SNP PRS was divided into three percentile groups: 0-20%, 20-80%, and 80-100%. The 80-
100% group represents those with the highest risk for AD. These ANCOVA models were used to
50
plot the estimated mean AD Cortical Thickness Signature values at each level of alcohol
consumption and each level of the genetic groupings. In addition, stratified analysis was used to
determine whether the AD genetic risk effect was significant within each alcohol consumption
group separately. Finally, for the alcohol consumption and 33-SNP PRS interaction, post-hoc
analysis was conducted to determine if this interaction was driven entirely by the two SNPs that
make up the apoE genotype, or if the other 31-SNPs together also have an interaction effect with
that factor. The 31-SNP PRS post-hoc analysis was conducted only among non-carriers of the
ApoE-4 allele. Given the relatively small effect sizes of the SNPs that make up the 31-SNP PRS
and the expectation that only those in the highest percentiles of this PRS would be at a very
significantly elevated risk for AD, the following percentile groups were examined: 0-5%, 5-25%,
25-75%, 75-95%, 95-100%. Groups with similar estimates were combined together when plotting
the results.
4.3 Results
4.3.1 Demographics
The dataset consisted of 6250 subjects that met the inclusion criteria. 37 underweight
subjects were removed, yielding a final dataset of 6213 subjects with a mean age of 62.4 years.
184 (3%) subjects reported consuming no alcohol per day, 321 (5.2%) less than 1 g/day, 1368
(22%) 1-6 g/day, 1243 (20%) 6-12 g/day, 1589 (25.5%) 12-24 g/day, 1106 (17.8%) 24-48 g/day,
and 402 (6.5%) >48 g/day. Descriptive statistics for the final study population by alcohol
consumption group are shown in Table 4.1. The maximum amount of alcohol consumption
reported by a subject was 161.2 g/day. Significant differences in age, sex, education, alcohol
consumption frequency, tobacco smoking, hypertension, diabetes, BMI, AD 33-SNP PRS, AD
51
Table 4.1: Study population characteristics
Alcohol Consumption
Group (g/day):
None <1 1-6 6-12 12-24 24-48 >48 P
1
Number of Subjects 184 321 1368 1243 1589 1106 402
Age 64.06 (7.51) 62.78 (7.75) 61.98 (7.48) 62.27 (7.39) 62.62 (7.49) 62.59 (7.35) 61.79 (6.98) 0.003
Age - Binned (%) 0.006
45-60 57 (31.0) 117 (36.4) 532 (38.9) 471 (37.9) 563 (35.4) 387 (35.0) 153 (38.1)
60-70 85 (46.2) 140 (43.6) 636 (46.5) 577 (46.4) 737 (46.4) 544 (49.2) 202 (50.2)
70-80 42 (22.8) 64 (19.9) 200 (14.6) 195 (15.7) 289 (18.2) 175 (15.8) 47 (11.7)
Sex = Male (%) 54 (29.3) 79 (24.6) 437 (31.9) 482 (38.8) 843 (53.1) 769 (69.5) 340 (84.6) <0.001
Education (%) 0.007
College/University degree 81 (44.0) 128 (39.9) 594 (43.4) 585 (47.1) 777 (48.9) 498 (45.0) 179 (44.5)
A levels/AS levels/equiv. 15 (8.2) 53 (16.5) 198 (14.5) 174 (14.0) 180 (11.3) 145 (13.1) 57 (14.2)
CSEs or equivalent 8 (4.3) 14 (4.4) 59 (4.3) 41 (3.3) 50 (3.1) 57 (5.2) 14 (3.5)
NVQ/HND/HNC/equiv. 13 (7.1) 16 (5.0) 83 (6.1) 58 (4.7) 104 (6.5) 79 (7.1) 41 (10.2)
O levels/GCSEs/equiv. 42 (22.8) 65 (20.2) 276 (20.2) 247 (19.9) 289 (18.2) 209 (18.9) 75 (18.7)
Other professional qual. 11 (6.0) 15 (4.7) 66 (4.8) 58 (4.7) 81 (5.1) 53 (4.8) 11 (2.7)
None of the above 14 (7.6) 30 (9.3) 92 (6.7) 80 (6.4) 108 (6.8) 65 (5.9) 25 (6.2)
Alcohol Cons. (g/day) 0.00 (0.00) 0.63 (0.20) 3.13 (1.46) 9.06 (1.78) 17.22 (3.42) 32.97 (6.41) 67.49 (19.36) <0.001
Alcohol Frequency (%) <0.001
Never 184 (100.0) 0 (0.0) 0 (0.0) 0 (0.0) 0 (0.0) 0 (0.0) 0 (0.0)
Special occasions only 0 (0.0) 218 (67.9) 195 (14.3) 5 (0.4) 0 (0.0) 0 (0.0) 0 (0.0)
One to three times a month 0 (0.0) 103 (32.1) 617 (45.1) 61 (4.9) 12 (0.8) 0 (0.0) 0 (0.0)
Once or twice a week 0 (0.0) 0 (0.0) 504 (36.8) 739 (59.5) 465 (29.3) 110 (9.9) 10 (2.5)
Three or four times a week 0 (0.0) 0 (0.0) 41 (3.0) 374 (30.1) 779 (49.0) 515 (46.6) 95 (23.6)
Daily or almost daily 0 (0.0) 0 (0.0) 11 (0.8) 64 (5.1) 333 (21.0) 481 (43.5) 297 (73.9)
Tobacco Smoking (pack/day) 0.11 (0.38) 0.16 (0.36) 0.16 (0.37) 0.17 (0.39) 0.24 (0.42) 0.36 (0.49) 0.51 (0.59) <0.001
Hypertension (%) 58 (31.5) 84 (26.2) 320 (23.4) 304 (24.5) 380 (23.9) 339 (30.7) 147 (36.6) <0.001
Diabetes (%) 18 (9.8) 22 (6.9) 61 (4.5) 55 (4.4) 59 (3.7) 46 (4.2) 18 (4.5) 0.005
BMI (kg/m
2
) 26.92 (5.15) 26.75 (5.02) 26.72 (4.75) 26.21 (4.21) 26.41 (4.04) 26.89 (3.84) 27.66 (3.81) <0.001
BMI - Binned (%) <0.001
Normal Weight 75 (40.8) 140 (43.6) 576 (42.1) 551 (44.3) 634 (39.9) 367 (33.2) 90 (22.4)
Overweight 68 (37.0) 115 (35.8) 507 (37.1) 473 (38.1) 710 (44.7) 535 (48.4) 220 (54.7)
Obese 41 (22.3) 66 (20.6) 285 (20.8) 219 (17.6) 245 (15.4) 204 (18.4) 92 (22.9)
ApoE-4 Carrier (%) 65 (35.3) 86 (26.8) 391 (28.6) 355 (28.6) 433 (27.2) 286 (25.9) 96 (23.9) 0.077
AD 31-SNP PRS -0.20 (0.47) -0.32 (0.44) -0.30 (0.46) -0.29 (0.45) -0.30 (0.45) -0.30 (0.47) -0.30 (0.49) 0.152
>95% AD 31-SNP PRS (%) 12 (6.5) 11 (3.4) 69 (5.0) 59 (4.7) 83 (5.2) 59 (5.3) 23 (5.7) 0.750
AD 33-SNP PRS 0.26 (0.93) 0.00 (0.81) 0.05 (0.84) 0.07 (0.86) 0.04 (0.86) 0.02 (0.84) -0.01 (0.85) 0.009
AD 33-SNP PRS - Bin (%) 0.228
0-20% 35 (19.0) 66 (20.6) 266 (19.4) 243 (19.5) 326 (20.5) 229 (20.7) 87 (21.6)
20-80% 96 (52.2) 196 (61.1) 835 (61.0) 743 (59.8) 937 (59.0) 664 (60.0) 249 (61.9)
80-100% 53 (28.8) 59 (18.4) 267 (19.5) 257 (20.7) 326 (20.5) 213 (19.3) 66 (16.4)
AD Cort. Thick. Sig. (mm) 2.63 (0.10) 2.64 (0.10) 2.66 (0.10) 2.65 (0.10) 2.65 (0.10) 2.63 (0.10) 2.61 (0.11) <0.001
Mean Hipp. Vol. (mm^3)
3983.52
(423.03)
3970.11
(391.30)
4019.20
(402.49)
4041.14
(409.23)
4072.59
(415.54)
4104.48
(425.38)
4097.27
(392.00)
<0.001
1
P-value for tests of group differences. Chi-squared test was used for categorical variables and ANOVA was used for continuous
variables.
For continuous variables, values represent: mean (standard deviation). For categorical variables, values represent: count
(percentage in category)
52
Cortical Thickness Signature, and mean hippocampal volume were found among the alcohol
consumption groups. The regression models all satisfied the assumptions for linear regression and
were not significantly influenced by any one subject.
4.3.2 Association of alcohol consumption and the AD Polygenic Risk Score with AD Cortical
Thickness Signature and Hippocampal Volume
We found a statistically significant difference in the AD Cortical Thickness Signature
among the binned alcohol consumption groups after adjustment for covariates (F(6,6187) = 12.0,
p = 2.0e-13). Comparison of the mean AD Cortical Thickness Signature of each group relative to
that of the 1-6 g/day group, after adjustment, can be found in Table 4.2. Model-predicted estimates
of the mean AD Cortical Thickness Signature for each alcohol consumption group is plotted in
Figure 4.1a. The 33-SNP PRS was not found to be significantly associated with AD Cortical
Thickness Signature (p = 0.24). The mean AD cortical thickness signature had an inverted-U-
shaped relationship with daily alcohol consumption, with the 1-6 g/day group having the highest
Table 4.2: Association of alcohol consumption with AD cortical thickness signature.
Variables Coefficient SE 95% CI t value Pr(>|t|) Sig
1
ANOVA P
2
Alcohol Consumption (g/day)
5
2.0e-13
None -0.020 7.2e-3 -0.034 to -5.6e-3 -2.74 0.0062 **
<1 -0.013 5.7e-3 -0.024 to -2.1e-3 -2.32 0.020 *
6-12 -4.9e-3 3.6e-3 -0.012 to 2.2e-3 -1.35 0.18
12-24 -8.9e-3 3.4e-3 -0.016 to -2.2e-3 -2.60 0.0095 **
24-48 -0.018 3.9e-3 -0.026 to -0.011 -4.77 1.9e-06 ***
>48 -0.042 5.5e-3 -0.053 to -0.031 -7.69 1.7e-14 ***
N 6213
R-Squared 0.177
1
*** p < 0.001; ** p < 0.01; * p < 0.05
2
Type III Sum of Squares P-value from ANCOVA model
3
Reference level set to College or University degree
4
Reference level set to Normal Weight
5
Reference level set to 1-6 g/da
53
estimated thickness. Relative to this 1-6 g/day group, the “None”, 24-48 g/day, and >48 g/day
groups each had a significantly lower thickness after correcting for multiple comparisons. We also
found a statistically significant difference in the average bilateral hippocampal volume among the
binned alcohol consumption groups after adjustment for covariates (F(6,6186) = 2.41, p = 0.025).
The 33-SNP PRS was also not found to be significantly associated with hippocampal volume (p =
0.72). Alcohol consumption frequency was not significantly associated with these metrics and its
addition as a covariate did not have a strong influence on the model output.
4.3.3 Interaction effect of alcohol consumption and genetic risk for AD
We found that the data supported our hypothesis that among cognitively healthy subjects,
there is an interaction effect of alcohol consumption and AD genetic risk on the AD Cortical
Thickness Signature. A model with an interaction term for alcohol consumption and the AD 33-
SNP PRS was found to fit the data significantly better than a model without this interaction term
(χ2(6) = 16.9, p = 0.0098). A summary of the model output can be found in Table 4.3. The same
Figure 4.1: Plots of association of alcohol consumption with AD Cortical Thickness
Signature. Model predicted AD Cortical Thickness Signature values among different alcohol
consumption groups in all subjects (A) and in only those who consume alcohol at least once per
week (B).
54
analysis was conducted using only apoE-4 carrier status, representing two SNPs of the full 33-
SNP AD PRS, in addition to using only the 31-SNP AD PRS among apoE-4 noncarriers. A model
with an interaction term for alcohol consumption and apoE-4 carrier status fit the data significantly
better than a model without this interaction term (χ2(6) = 13.5, p = 0.036). Among apoE-4
noncarriers, a model with an interaction term for alcohol consumption and the AD 31-SNP PRS
did not fit the data significantly better than a model without this interaction term (χ2(6) = 2.19, p
= 0.90). Using the binned versions of the AD 33-SNP PRS and the 31-SNP PRS, model-predicted
mean AD Cortical Thickness Signature values for each alcohol consumption group were plotted
as described in the methods section and are shown in Figure 4.2. Among subjects in the <1 g/day
Table 4.3: Interaction of AD PRS and alcohol consumption on cortical thickness.
Variables Coefficient SE 95% CI t value Pr(>|t|) Sig
1
ANOVA P
2
AD 33-SNP PRS -3.3e-4 3.0e-3 -6.1e-3 to 5.5e-3 -0.11 0.91 0.53
Alcohol Consumption (g/day)
5
1.4e-13
None -0.022 7.4e-3 -0.036 to -7.0e-3 -2.91 0.0036 **
<1 -0.014 5.7e-3 -0.025 to -2.7e-3 -2.44 0.015 *
6-12 -4.7e-3 3.6e-3 -0.012 to 2.3e-3 -1.31 0.19
12-24 g/day -8.9e-3 3.4e-3 -0.016 to -2.2e-3 -2.59 0.0096 **
24-48 g/day -0.019 3.9e-3 -0.026 to -0.011 -4.79 1.7e-6 ***
>48 g/day -0.042 5.5e-3 -0.053 to -0.031 -7.65 2.3e-14 ***
Alc. Cons. (g/day) * AD 33-SNP
PRS
0.010
None * AD 33-SNP PRS 0.010 7.9e-3 -5.3e-3 to 0.026 1.29 0.20
<1 g/day * AD 33-SNP PRS -0.014 7.0e-3 -0.027 to 2.9e-6 -1.96 0.050 .
6-12 g/day * AD 33-SNP PRS -2.7e-3 4.2e-3 -0.011 to 5.6e-3 -0.64 0.53
12-24 g/day * AD 33-SNP PRS 8.2e-3 4.0e-3 4.1e-4 to 0.016 2.06 0.039 *
24-48 g/day * AD 33-SNP PRS 8.1e-4 4.4e-3 -7.9e-3 to 9.5e-3 0.18 0.86
>48 g/day * AD 33-SNP PRS 7.4e-3 6.1e-3 -4.6e-3 to 0.019 1.21 0.23
N 6213
R-Squared 0.179
1
*** p < 0.001; ** p < 0.01; * p < 0.05
2
Type III Sum of Squares P-value from ANCOVA model
3
Reference level set to College or University degree
4
Reference level set to Normal Weight
5
Reference level set to 1-6 g/day
Abbreviations: AD = Alzheimer disease, PRS = Polygenic Risk Score, PC = Principle Component
55
group, for instance, those in
the 80-100% AD PRS group
had a lower estimated mean
AD Cortical Thickness
Signature than that of the 0-
20% and 20-80% AD PRS
group. Among subjects
consuming between 12 and
24 grams of alcohol per day,
however, those in the 80-
100% AD PRS group PRS
group had a higher mean AD
Cortical Thickness Signature
than that of the 0-20% and
20-80% AD PRS group. A
similar pattern of effect
estimates was found in the
post-hoc analyses of apoE
genotype alone and in the
analyses of the 31-SNP AD
PRS among ApoE-4 non-
carriers. So, moderate alcohol
consumption was estimated
Figure 4.2: Association of alcohol consumption and
cortical thickness by AD genetic risk. Model predicted AD
Cortical Thickness Signature values among different alcohol
consumption groups and among different AD 33-SNP PRS
percentiles (A), different ApoE-4 carrier statuses (B), and
different AD 31-SNP PRS percentiles (C).
56
to be associated with a greater
AD Cortical Thickness
Signature in those with a high
genetic risk for AD than in
those without a high genetic
risk for AD. Estimation of the
beta coefficient for the AD
33-SNP PRS term was also
conducted separately in
groups stratified by alcohol
consumption amount. After
adjustment for multiple
testing, only the 12-24 g/day
alcohol consumption group
had a significant effect for AD
33-SNP PRS on AD Cortical
Thickness Signature. Among
those who consume 12-24
g/day of alcohol and adjusting
for covariates, the estimated
mean AD Cortical Thickness
Signature increased by 6.7e-3
(95% CI: 2.1e-3, 0.011) mm
Figure 4.3: Interaction among weekly alcohol consumers.
Model predicted AD Cortical Thickness Signature values
among different alcohol consumption groups and among
different AD 33-SNP PRS percentiles (A), different ApoE-4
carrier statuses (B), and different AD 31-SNP PRS percentiles
(C) for only those subjects who consume alcohol at least once
per week.
57
per increase of 0.852 units of the AD 33-SNP PRS (1 standard deviation of the AD 33-SNP PRS
among all subjects) (p = 0.0043).
For average hippocampal volume, a model with an interaction term for alcohol
consumption and the AD 33-SNP PRS was not found to fit the data significantly better than a
model without this interaction term (χ2(6) = 6.62, p = 0.36). A model with an interaction term for
alcohol consumption and apoE-4 carrier status was also found to not fit the data significantly better
than a model without this interaction term (χ2(6) = 8.19, p = 0.22). Among ApoE-4 noncarriers, a
model with an interaction term for alcohol consumption and the AD 31-SNP PRS was not found
to fit the data significantly better than a model without this interaction term (χ2(6) = 2.34, p =
0.89). Considering only those who consume alcohol, though, among those in the 80-100% group
of the AD 33-SNP PRS, the 12-24 and 24-48 g/day groups had the highest estimated average
hippocampal volumes.
Adjustment for alcohol consumption frequency in the linear models yielded similar results
for both the AD cortical thickness signature and hippocampal volume, as shown in Figure 4.3. In
addition, using the model with a triple interaction of binned alcohol consumption, binned age, and
AD 33-SNP PRS, we found that the interaction of alcohol consumption and AD 33-SNP PRS on
AD Cortical Thickness Signature did not vary by age-group (F(12,6153) = 1.01, p = 0.44). The
interaction of alcohol consumption and AD 33-SNP PRS on average bilateral hippocampal volume
also did not vary by age-group (F(12,6152) = 0.997, p = 0.45).
In the fully adjusted model for AD Cortical Thickness Signature, the model including the
interaction term for binned alcohol consumption and AD 33-SNP PRS fit the data significantly
better than a model without this interaction term (χ2(6) = 19.5, p = 0.0034). Importantly, the overall
relationship of the interaction of alcohol consumption and AD 33-SNP PRS on AD Cortical
58
Thickness Signature did not change with adjustment for covariate-by-genotype and covariate-by-
factor interactions. In the fully adjusted model for average hippocampal volume, the model
including the interaction term for binned alcohol consumption and AD 33-SNP PRS did not fit the
data significantly better than a model without this interaction term (χ2(6) = 7.50, p = 0.277). The
overall trends were similar to that in the partially-adjusted model.
4.4 Discussion
In this study, we investigated the effect of alcohol consumption on brain structure and
tested whether this effect varies depending a person’s level of genetic risk for AD. For the
associations of both AD Cortical Thickness Signature and average hippocampal volume with
alcohol consumption, without consideration of any interaction effects, we found similar
relationships to those in other studies. Alcohol consumption was significantly associated with both
AD Cortical Thickness Signature and hippocampal volume. In the analyses of AD Cortical
Thickness Signature, an inverted-U-shaped relationship was found, similar to relationships found
previously with alcohol consumption [137]. Those in the 1-6 g/day alcohol consumption category
did not have a significantly different AD Cortical Thickness Signature than those in the <1, 6-12,
and 12-24 g/day groups, but had a significantly increased AD Cortical Thickness Signature relative
to the “None”, 24-48, and >48 g/day groups, suggesting that low-moderate alcohol consumption
is associated with better structural brain health in regions of the brain known to be affected early
in AD. In the analyses of hippocampal volume, this inverted-U-shaped relationship was much less
pronounced. Only those in the >48 g/day group had a significantly decreased volume relative to
the 1-6 g/day group. This inverted-U-shaped relationship, however, in which abstinence from
alcohol and high alcohol consumption groups have a lower cortical thickness than that of low-
59
moderate alcohol consumption groups, could potentially be due to reverse causation rather than a
protective effect of low-moderate alcohol consumption [167].
In addition to studying the main effect of alcohol consumption on brain structure, the main
effect of the AD 33-SNP PRS was investigated. It is notable that the AD 33-SNP PRS was not
significantly associated with either the AD Cortical Thickness Signature or average hippocampal
volume. This null finding has also been found by others using data from the UK Biobank. A study
of cognitively healthy subjects in the UK Biobank found that apoE-4 was associated with increased
white matter hyperintensity volumes, but not hippocampal volume, total grey matter volume, or
white matter integrity [168].
The data supported our hypothesis that there is an interaction effect of alcohol consumption
and genetic risk for AD on the AD cortical thickness signature among cognitively healthy adults.
Post-hoc analysis revealed that within the 12-24 g/day alcohol consumption group, those with a
higher genetic risk for AD had a greater AD Cortical Thickness Signature than those with a lower
genetic risk for AD. It is unclear whether this difference is due to a beneficial or detrimental effect
of moderate alcohol consumption in those with a high genetic risk for AD. In the groups consuming
<12 g/day of alcohol, the high-risk group (80-100% AD PRS) had a lower estimated thickness
than the medium-risk (20-80% AD PRS) group. In the groups consuming >12 g/day of alcohol,
though, the high-risk group had a higher estimated thickness than the medium-risk group. While
not statistically significant, there was a similar pattern in the average hippocampal volume
estimates. In the groups consuming <12 g/day of alcohol, the high-risk group had a lower estimated
hippocampal volume than the medium-risk group, but in the groups consuming >24 g/day, the
high-risk group had a higher estimated hippocampal volume than the medium-risk group. This is
60
a similar pattern to that found in a previous study, which found an increased hippocampal volume
among moderate alcohol consumers with apoE-4 [139].
The interaction effects we found were not statistically significant when considering only
those who consume alcohol at least once per week, but the overall pattern was not different from
the pattern found when considering the entire study population. Post-hoc analysis for apoE-4
carrier status and the AD 31-SNP PRS among apoE-4 noncarriers suggested that the interaction
effect was not driven entirely by the apoE gene, but rather AD genetic risk in general. In addition,
we found that this interaction effect did not vary significantly by age group, indicating that there
was no significant difference in the interaction of alcohol consumption and AD genetic risk among
45-60, 60-70, and 70-80 year old subjects. This suggests that this interaction effect is not strongly
age-dependent among those who are 45-80 years old. Furthermore, it is important to consider the
results of the fully-adjusted models that include each of the relevant covariate-by-PRS and
covariate-by-alcohol interactions to see if the interaction effect is changed by the inclusion of these
possible confounders. The nature of the interaction of alcohol consumption and the AD 33-SNP
PRS on brain structure did not change with their addition. Likely because of the increased variance
in brain structure explained by these additional interaction terms in the model, the p-value for these
interaction terms of interest were decreased slightly.
The direction of this interaction effect of alcohol and the 33-SNP AD PRS was different
from what we had expected. Heavy alcohol consumption is known to be associated with an
increased level of brain cortical atrophy [154]. In addition, while evidence has been mixed as to
whether cortical thickness is associated with genetic risk for AD in cognitively healthy subjects, a
longitudinal study of 497 cognitively normal subjects found that apoE-4 is associated with
increased amyloid deposition as early as middle age, in addition to a modest increase in medial
61
temporal lobe structural atrophy [169]. So, we had expected that alcohol consumers with a high
AD genetic risk would have a magnified level of atrophy. The data did not support this expectation,
and instead was consistent with the results of studies that have found reduced cognitive decline
and brain atrophy with low-moderate alcohol consumption among those with high genetic risk for
AD [139–141]. It is unclear why in the stratified analyses the AD 33-SNP PRS effect on AD
Cortical Thickness Signature was significant only within the 12-24 g/day groups. If alcohol is
having a protective effect among high AD genetic risk subjects, the data suggests that these
protective effects are only present with moderate consumption of alcohol, particularly 12-24 g/day.
At heavy levels of alcohol consumption, the known detrimental effects of alcohol likely
predominate any protective effects. Notably, the results do not suggest that having a high genetic
risk for AD exacerbates the negative effects of heavy alcohol consumption.
Even though the thickness was estimated to be greatest in the 12-24 g/day consumption
group among the high-risk subjects, there is a possibility that this could represent a detrimental
effect of alcohol consumption in this group instead of a protective effect. For example, some
studies have found an increased brain cortical thickness in cognitively healthy subjects in regions
with high amyloid beta deposition, a key pathological characteristic of AD, prior to the onset of
dementia. This may be due to an increase in thickness due to AD pathology prior to eventual
atrophy or a higher cognitive reserve among those with AD pathology but without dementia [170–
172]. So, the increase in thickness we found could also be due to a pathological change, whether
from amyloid beta deposition or another cause. Or, it is possible that this increase in cortical
thickness with regular alcohol consumption is a result of a compensatory increase in synaptic
sprouting, as has been seen in rats treated with alcohol [173].
62
Reverse causation and abstainer bias could also potentially explain the interaction. It is
possible that those who experience subtle memory difficulties indicative of early mild cognitive
impairment are more likely to reduce or stop their consumption of alcohol, leading to an abundance
of subjects with early dementia-related brain changes in the very low alcohol consumption groups
[174]. Alternatively, the subjects at high risk for AD that were included in the study could be those
whose brains are particularly resilient to the dementia for any number of reasons, while those with
less resilient brains could have been those not included in the study as a result of already having
converted to a state of cognitive impairment.
While this study has strengths including its large sample size, there are several limitations.
First, because the study is cross-sectional, it is impossible to determine whether alcohol
consumption leads to changes in cortical thickness or if changes in cortical thickness result in
changes in alcohol consumption habits. Second, because alcohol consumption data is self-reported,
it is subject to survey bias and social desirability bias. Furthermore, this alcohol consumption data
represented the reported alcohol usage at solely the imaging visit. The alcohol consumption data
therefore does not account for changes in alcohol consumption over a subject’s life. Some subjects
may report drinking small amounts per day despite having a former history of consuming large
quantities of alcohol. Other biases may be present that could make the results misleading. Selection
bias could result in only the healthiest of the high AD genetic risk subjects participating in the
study, for instance. Attrition bias could result in the loss of unhealthiest subjects who had begun
participation in the study, which could disproportionately affect subjects with high AD genetic
risk. It is also important to note that while FreeSurfer measures cortical thickness, other brain
structural changes unrelated to cortical thickness can cause changes to the measured thickness. For
example, measurements of the entorhinal cortex are often influenced by the proximity of the
63
meninges to the brain tissue in that region [175]. Nonetheless, the data showed that the association
of brain structure with alcohol consumption varies depending on one’s genetic risk for AD.
Replication of the results in this study is essential. No other currently available dataset that
we know of has sufficient statistical power to replicate this study. Similar studies using AD risk as
an outcome variable will also be useful. Amyloid PET imaging studies would also be very useful
in determining if alcohol consumption is associated with amyloid deposition in the brain, and
whether this deposition varies depending on one’s genetic risk for AD. Existing post-mortem
studies of the association of amyloid beta deposition and alcohol consumption so far have not
found an increase in AD pathology with excessive alcohol consumption, though these studies have
been limited in their sample size [176,177]. Amyloid beta blood biomarkers could also be used for
this purpose. Longitudinal studies may also be very useful for studying how the brain changes over
time with alcohol consumption and whether there is an interaction effect of alcohol consumption
and genetic risk for AD. Animal studies, such as with a mouse model with human apoE-4 and
APP, could be useful for better understanding this interaction of alcohol consumption and AD
genetic risk variants.
In conclusion, among cognitively healthy subjects, we found that alcohol consumption was
associated with both the AD Cortical Thickness Signature and mean hippocampal volume, similar
to what has been found in prior studies. An inverted-U-shaped relationship was found for AD
Cortical Thickness Signature. For mean hippocampal volume, this inverted-U-shaped relationship
was much less pronounced, with only the high alcohol consumption group having a significantly
lower hippocampal volume relative to the low alcohol consumption group. Moreover, we found a
significant interaction effect of alcohol consumption and a 33-SNP AD polygenic risk score.
Among those consuming 12-24 g/day of alcohol, those with the highest genetic risk for AD had
64
the greatest AD Cortical Thickness Signature. It is unclear whether the increased mean AD
Cortical Thickness Signature among moderate alcohol consumers with a high AD genetic risk
represents a protective effect of moderate alcohol consumption or if the increased cortical
thickness is due to other pathological mechanisms. Regardless, the data supports the hypothesis
that the association of brain structure and alcohol consumption varies based on one’s genetic risk
factors for AD.
65
Chapter 5
Genome-wide interaction scan of body mass index on hippocampal T2* value
5.1 Introduction
Body mass index (BMI) is associated with brain structure measurements derived from
magnetic resonance imaging (MRI) scans, with most studies finding obese participants to have
adverse brain changes such as hippocampal atrophy [26,131,178]. A key genetic risk variant for
obesity in the FTO gene has also been found to be associated with brain structure, further
supporting this relationship [179]. These associations are consistent with obesity being associated
with both stroke and dementia [16,180,181] in addition to subclinical markers of cerebrovascular
disease such as white matter hyperintensity volume [182]. The association of obesity with brain
structure, cerebrovascular disease, and dementia may be mediated by the negative health
consequences of obesity, including hypertension, dyslipidemia, and cardiovascular disease
[26,181–183].
Advances in the use of susceptibility-weighted MRI imaging has enabled MRI-based study
of tissue changes that may reflect iron deposition [9]. Indeed, a postmortem study confirmed a link
between iron in brain tissue and T2* value in susceptibility-weighted MRI (swMRI) [184]. This
provides a more specific measurement of brain changes compared to volume and thickness
measurements because it provides some clues about the composition of the brain tissue itself. One
of the most robust phenotypic associations with swMRI T2* in the brain is body mass index (BMI),
with the T2* value in the hippocampus having a particularly strong association with BMI [67,185].
The decreased T2* value with increasing BMI could represent hippocampal iron deposition
resulting from clinical or subclinical cerebrovascular disease, a possibility is supported by a study
66
finding that subcortical vascular mild cognitive impairment patients have swMRI measurements
consistent with iron deposition in the brain [186].
While it seems plausible that cerebrovascular changes downstream from obesity may
mediate hippocampal iron deposition in obese participants [16,187], understanding the genetic
variants associated with hippocampal T2* value and whether there exist variants that modify the
association BMI and hippocampal T2* may provide insight into this relationship. Here, we
conducted a genome-wide gene-by-environment interaction scan for BMI on hippocampal T2*
within the UK Biobank. For significant variants, we additionally tested whether these variants
were associated with other brain regions, and in a separate set of UK Biobank participants, whether
they were associated with a diverse set of blood biochemistry values. Our hypothesis was that there
exist genetic variants that modify the association of BMI on T2* value in the hippocampus.
5.2 Methods
5.2.1. Study population
We used data from the UK Biobank (UKB) resource, a large-scale, population-based
prospective cohort study based in the United Kingdom consisting of the collection of health-related
data for about 500,000 participants [63,64]. The initial assessment took place from 2006 to 2010
and the subsequent imaging assessment for which a subset of participants participated took place
starting in 2014 [65].
5.2.2. Imaging data
In the UKB, MRI scans of several different modalities were obtained for the participants
attending the imaging visit [67]. MRI acquisition details are described in the UK Biobank Brain
67
Imaging Documentation (http://biobank.ctsu.ox.ac.uk/crystal/refer.cgi?id=1977) and in the MRI
scan protocol document (http://biobank.ctsu.ox.ac.uk/crystal/refer.cgi?id=2367). From these
scans, the UK Biobank Imaging Working Group performed quality control and subsequently
calculated a variety of regional imaging-derived phenotypes (IDPs) [7]. In brief, cortical thickness
and subcortical, cerebellum, ventricle, brain stem, and corpus callosum volumes were computed
using FreeSurfer, white matter tract mean fractional anisotropy (FA) was derived from TBSS
processing of diffusion tensor imaging scans, median T2* values in subcortical structures were
computed from susceptibility-weighted MRI scans, and white matter hyperintensity volume was
computed with the BIANCA tool using T1 and T2 FLAIR scans [7].
5.2.3. Genetics data
For the genome-wide interaction scan (GWIS) completed in this study, we used version 3
of the UK Biobank imputed genetics data [63]. Variant quality control consisted of including only
variants with a minor allele frequency >0.001, an info score >0.8, a Hardy-Weinberg equilibrium
deviation test P-value >1e-7, and a missing proportion of <0.05. Sample quality control consisted
of including only participants in the white British ancestry subset, in the phasing input, without
excess relatives or relatives closer than 3
rd
degree relatedness, and without dementia, Parkinson
disease, or other central nervous system disease. In addition, participants were removed if missing
variables for BMI, median T2* value in the hippocampus, or head motion estimated from resting
fMRI. Genetics data was converted into the BinaryDosage format for the analyses [188].
68
5.2.4 Association of BMI with regional volume and hippocampal T2* value
The associations of BMI with the 26 regional brain volume metrics and 12 subcortical
median T2* values were evaluated using regression models adjusting for scanner head and table
position, head motion estimated from resting fMRI, intracranial volume, sex, age, and age
2
. BMI
was used as a categorical variable, with the underweight category having a BMI <18.5 kg/m
2
;
normal weight, 18.5-25 kg/m
2
; overweight, 25-30 kg/m
2
; and obese, >30 kg/m
2
. The outcome
variable was standardized, and the covariates were mean-centered. P-values for each IDP were
computed using the ANOVA test and adjusted for multiple comparisons using false-discovery rate
adjustment. Cohen’s D for underweight, overweight, and obese groups relative to the normal
weight group were also calculated [76]. P-values and Cohen’s D values were plotted on 2D brain
maps using the ggseg R library [79].
5.2.5 Genome-wide interaction scan
To test for the presence of variants that modify the association of BMI with hippocampal
T2* value, a genome-wide by BMI interaction scan was completed using GxEScanR [189]. The
average of the left and right hippocampal median T2* value was used as the outcome variable.
GxEScanR evaluated several regression models for each variant: (1) a model with covariates and
the variant (G) and environmental factor (BMI) as separate terms, (2) a model with covariates and
G, BMI, and G*BMI terms, and (3) a model with only the covariates and BMI. A P-value for the
G term was calculated based on a Wald test in the first model, a P-value for the GxBMI term was
calculated using a Wald test for the GxBMI term in the second model, and a P-value from a 2
degree-of-freedom (2df) test was calculated using the likelihood ratio test to compare models with
and without the G and GxBMI terms (Model 2 vs. Model 3). Covariates included in the models
69
included 10 principal components estimating population structure, scanner head and table position,
head motion estimated from fMRI, intracranial volume, sex, age, and age
2
. All variables were
standardized prior to completing the GWIS.
Significance testing of the variants was completed using two different approaches. In the
first approach, the 2df test was used to test for statistical significance, with variants with a 2df P-
value of <5e-8 considered statistically significant. In the second approach, a two-stage significance
testing approach accounting for linkage disequilibrium was used [37,190]. In the first stage, only
variants with a G p-value of <5e-4 were included in the second stage. In the second stage,
significance testing of these variants was based on Bonferroni adjustment accounting for linkage
disequilibrium of the variants. To determine the effective number of tests for which to correct for
multiple comparisons, a correlation matrix of these variants was created and the meff function of
the Poolr R package [191] was run using the Gao method (C=0.995) [192]. Variants in this second
stage were considered statistically significant if they had a GxBMI P-value of less than 0.05
divided by the estimated effective number of tests.
Manhattan plots for the G P-values and the 2df P-values were created using the qqman R
library. Using the variants found to be significant using either significance testing method, we then
tested whether there was an association with other IDPs as well, including volume, thickness, FA,
median T2* value, and white matter hyperintensity volume.
5.2.6 Testing for interactions on blood values in the non-imaging subset
A set of UK Biobank participants that did not participate in the imaging assessment was
used to test whether the variants identified in the GWIS were also associated with a variety of
blood values (UKB Category 17518), many of which reflect adverse health consequences of
70
obesity [193]. These blood values included albumin, alkaline phosphatase, alanine
aminotransferase, apolipoprotein A, apolipoprotein B, aspartate aminotransferase, direct bilirubin,
urea, calcium, cholesterol, creatinine, C-reactive protein, cystatin C, gamma glutamyl transferase,
glucose, glycated hemoglobin HbA1c, HDL cholesterol, IGF-1, LDL, lipoprotein A, phosphate,
rheumatoid factor, sex hormone binding globulin (SHBG), total bilirubin, total protein,
triglycerides, urate, and vitamin D. Values were log-transformed as appropriate to better satisfy
normality assumptions of regression models. For variants found to have a significant association
with hippocampal T2*, similar models adjusting for age, sex, BMI, and 10 principal components
estimating population structure were used to test if these variants have a significant GxBMI effect
on any of these blood values. Bonferroni adjustment was used to adjust the G and GxE P-values
calculated in this testing. For significant blood values, the estimated G effect in the normal and
obese categories were calculated.
5.3 Results
5.3.1. Demographics
Demographics of the
imaging and non-imaging subsets
are included in Table 5.1. After
application of the inclusion and exclusion criteria, the imaging subset consisted of 28,363
participants, and the distinct non-imaging subset consisted of 462,970 participants. The ages of the
imaging subset (mean: 64.2; SD: 7.5) were greater than those of the non-imaging subset (mean:
56.7; SD: 8.1)
Table 5.1: Sample Characteristics by Subset
Characteristics Imaging Subset Non-imaging Subset
n 28,363 452,970
Age (mean (SD)) 64.19 (7.47) 56.65 (8.13)
Sex = Male (%) 13439 (47.4) 206540 (45.6)
BMI (mean (SD)) 26.46 (4.35) 27.48 (4.82)
71
5.3.2 BMI-brain Association
Brain plots showing the associations of BMI with regional volumes and median T2* values
are shown in Figure 5.1. Most regions tested were significantly associated with BMI after false-
discovery rate adjustment. The IDPs with the largest Cohen’s D effect size estimates were the
hippocampus and amygdala T2* values. The signed Cohen’s D for the left hippocampus T2* value
in the obese group relative to the normal group was -0.46, and -0.48 for the right hippocampus
T2* value. For the left and right amygdala T2* values, these same Cohen’s D values were -0.28,
and -0.30, respectively. In contrast, the signed Cohen’s D for left and right hippocampal volume
Figure 5.1: Association of BMI and both regional volume and subcortical
median T2* value
72
in the obese group relative to the normal weight group was 0.08 and 0.06. For left and right
amygdala volume, these Cohen’s D values were 0.13 and 0.10.
5.3.3 GWIS results
For the GWIS of BMI on the bilateral median hippocampal T2* value, 6,749,795 variants
were considered after quality control. Manhattan plots for the G P-values and 2df P-values are
shown in Figure 5.2. One variant, rs9367218 (intronic variant of the SUPT3H gene), was genome-
wide significant for both the G test and the 2df test. Another variant, rs10413951 (intronic variant
Figure 5.2: Manhattan plots. Manhattan plots for the (A) association of SNPs
with hippocampal T2* without an interaction term with BMI and (B) for the 2df
test jointly considering the SNP and SNP*BMI terms
73
of the ZNF730 gene), was genome-wide significant for the 2df test and 2-stage filtering approach
accounting for linkage disequilibrium of variants. A third variant, rs460380 (an intronic variant of
the TSPEAR gene), was found to be significant using this same 2-stage filtering approach. For the
second stage of this approach, accounting for the correlation between the 5317 SNPs with a G P-
value of <5e-4 brought the effective number of tests to 1337, leading to a final significance
threshold of 3.7e-4 (0.05/1337) for the GxBMI test. Details on the variants with significant
associations with hippocampal T2* value are shown in Table 5.2. Plots showing the nature of the
interaction for each of these variants are shown in Figure 5.3. For each of these three SNPs, having
a copy of the minor allele was estimated to have increased median hippocampal T2* value in the
higher BMI groups, but not underweight or normal BMI groups.
Additional brain IDPs associated with the identified SNPs are shown in Table 5.3. The
SNP rs9367218 (SUPT3H) was associated with 5 additional IDPs, including bilateral corticospinal
tract FA, amygdala T2* value, and central corpus callosum volume. The association of this SNP
with these IDPs did not vary by BMI. In addition, the SNP rs10413951 (ZNF730) was also
associated with left amygdala T2* value, bilateral superior corona radiata FA, and the left posterior
Table 5.2: Statistics for significant SNPs
rsid Nearest Gene Chr:Pos
1
Alleles
(Ref/Alt)
Alt Allele
Freq
G β
2
G P GxE β
3
GxE P
2df
LRT P
4
rs9367218
SUPT3H
(intronic)
6:45088179 G/A 0.45 0.05 8.9e-10 0.02 0.03 6.2e-10
rs10413951
ZNF730
(intronic)
19:23316143 T/G 0.95 -0.08 4.5e-6 -0.07 1.9e-5 2.9e-9
rs460380
TSPEAR,
KRTAP10-7
(intronic)
21:46028864 C/A 0.64 -0.03 2.2e-4 -0.03 2.2e-5 1.3e-7
1
Coordinates in GRCh37 (hg19) assembly
2
Beta coefficient (in units of 1 SD hippocampal T2* value per additional copy of alternate allele) of the G term in model
adjusting for G and BMI with no interaction (G represents the number of copies of the alternate allele)
3
Beta coefficient for G*BMI term
4
P-value calculated by likelihood ratio test (LRT) comparing model with G and G*BMI terms to one without
74
limb of the internal capsule FA. The
association of rs10413951 with these
IDPs did vary by BMI (FDR-adjusted
P<0.05).
The association of these SNPs
with blood values in the non-imaging
subset is shown in Table 5.4. The SNP
rs9367218 (SUPT3H) was associated
with both cystatin C and urate, with the
minor allele being associated with
decreased cystatin C and increased
urate levels. The SNP rs10413951
(ZNF730) had a significant interaction
effect with BMI on triglyceride levels
(GxE P<5e-8). Among the obese group,
being a carrier of the minor allele of this
SNP was associated with decreased
triglyceride levels (P=1.6e-6), whereas
triglyceride levels did not significantly
differ in the normal weight group. The
SNP rs460380 (TSPEAR) was
associated with alanine
aminotransferase, though this association did not vary by BMI.
Figure 5.3: Interaction plots. (A-C) Model
predicted estimates of hippocampal T2* by BMI
group and minor allele carrier status for each variant.
75
5.4 Discussion
In this study,
we tested whether
there were any
genetic variants in
the UK Biobank
associated with
hippocampal T2*
value, accounting for
variant-by-BMI interaction effects. To better place the association of BMI and hippocampal T2*
value into context with nearby brain structures, we first evaluated the association between BMI
and brain volumes and subcortical T2* values. The majority of the IDPs evaluated were
significantly associated with BMI. While the basal ganglia, thalamus, cerebellum, corpus
callosum, and brainstem were estimated to have lower volumes with increasing BMI, the amygdala
and hippocampus were estimated to have larger volumes with increasing volumes. Median T2*
Table 5.3: Additional brain IDPs associated with the significant SNPs
rsid IDP GxE P
1
2df LRT P
1
rs9367218
(SUPT3H)
Right Corticospinal Tract FA 0.94 5.5e-5
Median T2* Value in Left Amygdala 0.94 8.3e-4
Left Corticospinal Tract FA 0.94 8.3e-4
Volume of Central Corpus Callosum 0.94 1.5e-3
Median T2* Value in Right Amygdala 0.94 1.5e-3
rs10413951
(ZNF730)
Median T2* Value in Left Amygdala 0.013 8.1e-3
Right Superior Corona Radiata FA 0.046 0.015
Left Superior Corona Radiata FA 0.031 0.037
Left Posterior Limb of Internal Capsule FA 0.031 0.071
1
P-values are FDR-adjusted separately for each variant (accounting for testing 152
IDPS). The GxE P-value is computed using the Wald test for the BMI*variant
interaction term. The 2df LTR P-value is computed using the likelihood ratio test
comparing a model with and without variant and BMI*variant terms.
Table 5.4: Associations of the variants with blood values in the non-imaging subset
rsid Blood Assay
Sample
Size
Variant
P-value
1
GxE P-
Value
2
Normal Weight Obese
β Estimate
3
P-Value β Estimate
3
P-Value
rs9367218
(SUPT3H)
Cystatin C* 430,847 5.8e-5 0.73 -0.007 0.06 -0.011 3.5e-3
Urate 430,364 1.0e-4 0.03 0.007 0.04 0.008 0.023
rs10413951
(ZNF730)
Triglycerides* 430,542 0.075 3.0e-8 0.005 0.53 -0.041 1.6e-6
rs460380
(TSPEAR)
Alanine
Aminotransferase*
430,704 0.85 2.5e-4 -0.005 0.20 0.011 8.4e-3
1
The variant P-value is computed using the Wald test for the variant term in a model adjusting for age, sex,
BMI, and the first 10 principal components estimating population structure.
2
The GxE P-value is computed using the Wald test for the BMI*variant interaction term. Either of the P-values
is considered significant if the P-value is <3e-4 (Bonferroni adjustment for 168 tests)
3
Estimates of the variant association with the specified blood assay in the normal weight and obese groups.
Estimates are in units of 1 standard deviation of the blood assay and for rs9367218 are per copy of A allele, for
rs10413951 per copy of T allele, and for rs460380 per copy of C allele.
*Values were log-transformed prior to evaluating regression model
76
values in the thalamus, basal ganglia, amygdala, and hippocampus, however, were uniformly
estimated to be decreased with increasing BMI. In addition, the largest Cohen’s D values were
found for amygdala and hippocampal T2*. It is unclear why an increased BMI was associated with
increased hippocampal volume, but a decreased T2* value. One previous study did find increased
amygdala and hippocampal volume among obese participants [194], though others have found
decreased hippocampal volume [131,178]. The decreased T2* value with increasing BMI was
consistent with prior studies [67,185].
The SNP rs9367218, an intronic variant of the SUPT3H gene, was found to be associated
with hippocampal T2* value independent of BMI. Variants of this SUPT3H gene have also been
found to be associated with corticospinal tract white matter microstructure [195], an association
that our analysis also found. So, our work suggests that in addition to these previously found brain
structure associations with SUPT3H genetic variation, variation in this gene is associated with
hippocampal and amygdala T2* value. There are many possibilities of how this variant may be
related to T2* value in these regions. First, genetic variants in the SUPT3H gene have been linked
to ADAMTS13 activity [196], a protein believed to have a role in maintaining cerebrovascular
integrity [197]. Second, the causal gene for this variant may actually be the nearby RUNX2 gene,
which has been shown to be regulated in part by the SUPT3H promoter [198]. RUNX2 is an
essential transcription factor for bone formation [199], and its genetic variation has been associated
with bone mineral density [200], skull shape [201], facial structure [202], and relative brain age as
assessed using brain MRI scans [203]. In addition to its involvement in osteogenesis, RUNX2 may
be involved in vascular disease as a result of calcification, osteogenic differentiation, and apoptosis
of vascular smooth muscle cells, in addition to vascular inflammation [204,205].
77
The SNP rs10413951, an intronic variant of ZNF730, was significantly associated with
both hippocampal T2* value and triglyceride levels using the 2df tests. Relative to the normal
weight group, obese participants with the minor allele of this variant (the T-allele) were estimated
to have both an increased T2* value (suggesting decreased hippocampal iron deposition) and
decreased triglyceride levels. The role of the variant is unclear, though zinc finger proteins are
known to have some involvement in antiviral response [206]. The finding of an association of this
variant with another adverse outcomes of obesity, triglyceride levels, in a distinct UK Biobank
subset serves as a form of replication for this variant. Given that studies have found that the risk
for an initial ischemic stroke among people with obesity is dependent on metabolic changes such
as altered triglyceride levels [181], triglyceride levels and hippocampal T2* value may be related
in some way. Moreover, this variant was found to have an interaction effect on other brain IDPs,
including white matter microstructure of the corona radiata and left posterior limb of the internal
capsule.
A third variant was identified using the two-stage filtering approach accounting for linkage
disequilibrium of variants passing the first filter. This variant, rs460380, is an intronic variant in
the TSPEAR and KRTAP10-7 gene locus. Deleterious TSPEAR gene variants have been found to
lead to congenital sensorineural deafness and epilepsy [207]. TSPEAR mutations have also been
found to be associated with tooth and hair follicle morphogenesis [208]. Furthermore, while not
genome-wide significant, TSPEAR variants have been found to be associated with increased odds
of small artery occlusion (OR=2.5; p=0.004) [209]. Another TSPEAR variant (rs11088961) was
found to be part of a significant SNP-by-SNP interaction on tau pathology, a pathology known to
occur in Alzheimer disease [210].
78
This genome-wide interaction scan of BMI on hippocampal T2* value, one of the brain
IDPs for which BMI has a substantial effect size estimate, identified three variants associated
hippocampal T2* value. Incorporating BMI information by including an interaction term with BMI
led to the identification of two additional variants than would be found in a standard genome-wide
association study. One novel finding of this study was that the SUPT3H intronic variant is
associated with both hippocampal and amygdala T2* value independent of BMI. An additional
novel finding is that the ZNF730 intronic variant, rs10413951, is associated with both hippocampal
T2* value and triglyceride levels in obese participants. Moreover, this study highlights the
usefulness of genome-wide interaction scans on MRI-derived brain structure measurements in
identifying genetic variants associated with brain structure that may not have been found with a
standard genome-wide association study. These findings, however, should be considered
hypothesis generating. Replication in a separate dataset other than the UK Biobank and additional
study will be needed further investigate these variants.
79
Chapter 6
Conclusions and Future Directions
6.1 Summary of findings
Chapters 2 and 3 present our studies of atrial fibrillation and P-wave duration on brain
structure. These studies collectively suggest that atrial pathology is associated with brain structure,
with a notable association of thalamus volume and brain stem volume found in both studies. The
association with these regions is interesting because they are regions known to be affected by
lacunar infarcts, and cardiogenic emboli emerging from the atria is a possible link between atrial
and brain pathologies [106]. The amygdala volume and precentral gyrus thickness were also found
to be associated with both atrial fibrillation and P-wave duration. The association with precentral
gyrus is interesting because two separate studies have found that precentral gyrus is associated
with cerebral small vessel disease [211], with one study finding that thickness in this region is
related to gait in patients with this disease [212]. If atrial pathology is actually causing these brain
changes, regular ECG monitoring for atrial fibrillation and P-wave abnormalities to enable early
treatment may become an important element of preventive brain health.
Chapters 3 and 4 both identified cerebrovascular risk factors whose association with brain
structure was modified by genetic variation. Chapter 3 described how the association between
alcohol consumption and the Alzheimer disease cortical thickness signature was modified by an
AD polygenic risk score (PRS). Interestingly, among those with moderate alcohol consumption, a
greater AD PRS was associated with greater cortical thickness, opposite from our hypothesis.
Chapter 4 described a genome-wide interaction scan of BMI on median hippocampal T2* value,
which identified 3 novel variants. One of these variants, an intronic variant in the ZNF730 gene,
80
was also found to have a significant interaction effect with BMI on triglyceride levels in a distinct
UK Biobank subset, corroborating that the variant modifies health outcomes of obesity. This
finding shows the value of genome-wide interaction scans on MRI-derived brain structure
measurements.
6.2 Future directions
Future work in investigating the association of atrial fibrillation and P-wave duration with
brain structure would benefit from a longitudinal study design with regular monitoring of ECG.
Such a longitudinal study may be important in testing whether there is a temporal order to the
association. In addition, a study with regular ECG monitoring would be better able to identify
participants with paroxysmal atrial fibrillation, record the total number of atrial fibrillation events,
and track P-wave duration over time. In addition, while we examined brain volumes, thicknesses,
FA values, and median T2* values, the results suggest that lacunar infarcts and microbleeds may
be involved. Future work will therefore involve developing methods for automated segmentation
of such lesions to test the hypothesis that atrial pathology is associated with lacunar infarcts in the
thalamus and brain stem. Other future directions include testing whether anticoagulation treatment
modifies the association of AF or PWD with brain structure in a larger dataset, investigating
whether microembolic signals detected on transcranial doppler ultrasound are more frequent
during atrial fibrillation episodes, and investigating other atrial properties such as P-wave axis.
Future work for the gene-environment interaction projects involve replicating the analyses
in separate datasets. In addition, the alcohol interaction project suggests the possibility of a
randomized experiment in which participants consume alcohol followed by blood testing for
amyloid and tau levels. Rodent studies have found that a low dose of ethanol increases the
81
clearance of waste molecules in the brain [213], so it would be interesting to test if this same effect
occurs in humans and whether it varies by genetic risk for Alzheimer disease. In addition, given
the promising findings of the BMI-Hippocampal T2* GWIS, a phenome-wide by variant
interaction scan may be illuminating. In addition, future work would involve completing GWIS
analyses for additional environmental or health factors. Functional studies of the variants found
may also be useful to try to understand why the variants found were associated with hippocampal
T2* value. The interaction of BMI and the ZNF730 variant on triglyceride levels if of particular
interest for replication in a separate dataset.
82
REFERENCES
[1] Raji CA, Eyre H, Wei SH, Bredesen DE, Moylan S, Law M, et al. Hot Topics in Research:
Preventive Neuroradiology in Brain Aging and Cognitive Decline. American Journal of
Neuroradiology 2015;36:1803–9. https://doi.org/10.3174/ajnr.A4409.
[2] Mazziotta JC, Toga AW, Frackowiak RSJ. Brain Mapping: The Disorders. Elsevier; 2000.
[3] Schneider JA. Vascular and Alzheimer’s pathology: A mixed affair. Alzheimer’s &
Dementia 2020;16:e038813. https://doi.org/10.1002/alz.038813.
[4] Schneider JA, Arvanitakis Z, Bang W, Bennett DA. Mixed brain pathologies account for
most dementia cases in community-dwelling older persons. Neurology 2007;69:2197–204.
https://doi.org/10.1212/01.wnl.0000271090.28148.24.
[5] Power MC, Mormino E, Soldan A, James BD, Yu L, Armstrong NM, et al. Combined
neuropathological pathways account for age-related risk of dementia. Annals of Neurology
2018;84:10. https://doi.org/10.1002/ana.25246.
[6] Havenon A de, Meyer C, McNally JS, Alexander M, Chung L. Subclinical Cerebrovascular
Disease: Epidemiology and Treatment. Current Atherosclerosis Reports 2019;21:39.
https://doi.org/10.1007/s11883-019-0799-1.
[7] Alfaro-Almagro F, Jenkinson M, Bangerter NK, Andersson JLR, Griffanti L, Douaud G, et
al. Image processing and Quality Control for the first 10,000 brain imaging datasets from
UK Biobank. NeuroImage 2018;166:400–24.
https://doi.org/10.1016/j.neuroimage.2017.10.034.
[8] Tuladhar AM, Reid AT, Shumskaya E, de Laat KF, van Norden AGW, van Dijk EJ, et al.
Relationship Between White Matter Hyperintensities, Cortical Thickness, and Cognition.
Stroke 2015;46:425–32. https://doi.org/10.1161/STROKEAHA.114.007146.
[9] Duyn J. MR susceptibility imaging. J Magn Reson 2013;229:198–207.
https://doi.org/10.1016/j.jmr.2012.11.013.
[10] Knopman DS, Roberts R. Vascular Risk Factors: Imaging and Neuropathologic Correlates.
Journal of Alzheimer’s Disease : JAD 2010;20:699. https://doi.org/10.3233/JAD-2010-
091555.
[11] DeCarli C. The link between blood pressure and Alzheimer’s disease. The Lancet
Neurology 2021;20:878–9. https://doi.org/10.1016/S1474-4422(21)00340-9.
[12] SPRINT MIND Investigators for the SPRINT Research Group, Williamson JD, Pajewski
NM, Auchus AP, Bryan RN, Chelune G, et al. Effect of Intensive vs Standard Blood
Pressure Control on Probable Dementia: A Randomized Clinical Trial. JAMA
2019;321:553–61. https://doi.org/10.1001/jama.2018.21442.
83
[13] Thacker EL, Gillett SR, Wadley VG, Unverzagt FW, Judd SE, McClure LA, et al. The
American Heart Association Life’s Simple 7 and incident cognitive impairment: The
REasons for Geographic And Racial Differences in Stroke (REGARDS) study. J Am Heart
Assoc 2014;3:e000635. https://doi.org/10.1161/JAHA.113.000635.
[14] Iadecola C, Davisson RL. Hypertension and Cerebrovascular Dysfunction. Cell Metabolism
2008;7:476. https://doi.org/10.1016/j.cmet.2008.03.010.
[15] Marini S, Merino J, Montgomery BE, Malik R, Sudlow CL, Dichgans M, et al. Mendelian
Randomization Study of Obesity and Cerebrovascular Disease. Annals of Neurology
2020;87:516–24. https://doi.org/10.1002/ana.25686.
[16] Letra L, Sena C. Cerebrovascular Disease: Consequences of Obesity-Induced Endothelial
Dysfunction. In: Letra L, Seiça R, editors. Obesity and Brain Function, Cham: Springer
International Publishing; 2017, p. 163–89. https://doi.org/10.1007/978-3-319-63260-5_7.
[17] Shah RS, Cole JW. Smoking and stroke: the more you smoke the more you stroke. Expert
Review of Cardiovascular Therapy 2010;8:917–32. https://doi.org/10.1586/erc.10.56.
[18] Lukovits TG, Mazzone T, Gorelick PB. Diabetes mellitus and Cerebrovascular Disease.
NED 1999;18:1–14. https://doi.org/10.1159/000026190.
[19] Yaghi S, Elkind MSV. Lipids and Cerebrovascular Disease: Research and Practice. Stroke;
a Journal of Cerebral Circulation 2015;46:3322.
https://doi.org/10.1161/STROKEAHA.115.011164.
[20] Haeusler KG, Laufs U, Endres M. Chronic Heart Failure and Ischemic Stroke. Stroke
2011;42:2977–82. https://doi.org/10.1161/STROKEAHA.111.628479.
[21] Hvidtfeldt UA, Frederiksen ME, Thygesen LC, Kamper-Jørgensen M, Becker U, Grønbæk
M. Incidence of Cardiovascular and Cerebrovascular Disease in Danish Men and Women
With a Prolonged Heavy Alcohol Intake. Alcoholism: Clinical and Experimental Research
2008;32:1920–4. https://doi.org/10.1111/j.1530-0277.2008.00776.x.
[22] Reynolds K, Lewis B, Nolen JDL, Kinney GL, Sathya B, He J. Alcohol Consumption and
Risk of Stroke: A Meta-analysis. JAMA 2003;289:579–88.
https://doi.org/10.1001/jama.289.5.579.
[23] Arenillas JF, Moro MA, Dávalos A. The Metabolic Syndrome and Stroke. Stroke
2007;38:2196–203. https://doi.org/10.1161/STROKEAHA.106.480004.
[24] Mukamal KJ, Longstreth WT, Mittleman MA, Crum RM, Siscovick DS. Alcohol
consumption and subclinical findings on magnetic resonance imaging of the brain in older
adults: the cardiovascular health study. Stroke 2001;32:1939–46.
84
[25] Lu D, Liu J, MacKinnon AD, Tozer DJ, Markus HS. Prevalence and Risk Factors of
Cerebral Microbleeds: Analysis From the UK Biobank. Neurology 2021;97:e1493–502.
https://doi.org/10.1212/WNL.0000000000012673.
[26] Raji CA, Ho AJ, Parikshak NN, Becker JT, Lopez OL, Kuller LH, et al. Brain structure and
obesity. Human Brain Mapping 2010;31:353–64. https://doi.org/10.1002/hbm.20870.
[27] Santos CY, Snyder PJ, Wu W-C, Zhang M, Echeverria A, Alber J. Pathophysiologic
relationship between Alzheimer’s disease, cerebrovascular disease, and cardiovascular risk:
A review and synthesis. Alzheimer’s & Dementia: Diagnosis, Assessment & Disease
Monitoring 2017;7:69–87. https://doi.org/10.1016/j.dadm.2017.01.005.
[28] Arvanitakis Z, Capuano AW, Leurgans SE, Bennett DA, Schneider JA. Relation of cerebral
vessel disease to Alzheimer’s disease dementia and cognitive function in elderly people: a
cross-sectional study. The Lancet Neurology 2016;15:934–43.
https://doi.org/10.1016/S1474-4422(16)30029-1.
[29] Broce IJ, Tan CH, Fan CC, Jansen I, Savage JE, Witoelar A, et al. Dissecting the genetic
relationship between cardiovascular risk factors and Alzheimer’s disease. Acta
Neuropathologica 2019;137:209. https://doi.org/10.1007/s00401-018-1928-6.
[30] Han JW, Maillard P, Harvey D, Fletcher E, Martinez O, Johnson DK, et al. Association of
vascular brain injury, neurodegeneration, amyloid, and cognitive trajectory. Neurology
2020;95:e2622–34. https://doi.org/10.1212/WNL.0000000000010531.
[31] Lippi G, Sanchis-Gomar F, Cervellin G. Global epidemiology of atrial fibrillation: An
increasing epidemic and public health challenge. International Journal of Stroke
2021;16:217–21. https://doi.org/10.1177/1747493019897870.
[32] He J, Tse G, Korantzopoulos P, Letsas KP, Ali-Hasan-Al-Saegh S, Kamel H, et al. P-Wave
Indices and Risk of Ischemic Stroke. Stroke 2017;48:2066–72.
https://doi.org/10.1161/STROKEAHA.117.017293.
[33] Berman JP, Norby FL, Mosley T, Soliman EZ, Gottesman RF, Lutsey PL, et al. Atrial
Fibrillation and Brain Magnetic Resonance Imaging Abnormalities. Stroke 2019;50:783–8.
https://doi.org/10.1161/STROKEAHA.118.024143.
[34] Moazzami K, Shao IY, Chen LY, Lutsey PL, Jack CR, Mosley T, et al. Atrial Fibrillation,
Brain Volumes, and Subclinical Cerebrovascular Disease (from the Atherosclerosis Risk in
Communities Neurocognitive Study [ARIC-NCS]). The American Journal of Cardiology
2020;125:222–8. https://doi.org/10.1016/j.amjcard.2019.10.010.
[35] Reyes JL, Norby FL, Wang W, Parikh R, Oldenburg NC, Lutsey PL, et al. Abstract 13193:
Association of Abnormal P-wave Indices With Brain MRI Infarcts: The Atherosclerosis
Risk in Communities Neurocognitive Study (ARIC-NCS). Circulation 2020;142:A13193–
A13193. https://doi.org/10.1161/circ.142.suppl_3.13193.
85
[36] Sachdev PS, Thalamuthu A, Mather KA, Ames D, Wright MJ, Wen W, et al. White Matter
Hyperintensities Are Under Strong Genetic Influence. Stroke 2016;47:1422–8.
https://doi.org/10.1161/STROKEAHA.116.012532.
[37] Gauderman WJ, Mukherjee B, Aschard H, Hsu L, Lewinger JP, Patel CJ, et al. Update on
the State of the Science for Analytical Methods for Gene-Environment Interactions. Am J
Epidemiol 2017;186:762–70. https://doi.org/10.1093/aje/kwx228.
[38] McAllister K, Mechanic LE, Amos C, Aschard H, Blair IA, Chatterjee N, et al. Current
Challenges and New Opportunities for Gene-Environment Interaction Studies of Complex
Diseases. Am J Epidemiol 2017;186:753–61. https://doi.org/10.1093/aje/kwx227.
[39] Dunn AR, O’Connell KMS, Kaczorowski CC. Gene-by-environment interactions in
Alzheimer’s disease and Parkinson’s disease. Neuroscience & Biobehavioral Reviews
2019;103:73–80. https://doi.org/10.1016/j.neubiorev.2019.06.018.
[40] Wang C, Sun J, Guillaume B, Ge T, Hibar DP, Greenwood CMT, et al. A Set-Based Mixed
Effect Model for Gene-Environment Interaction and Its Application to Neuroimaging
Phenotypes. Front Neurosci 2017;11. https://doi.org/10.3389/fnins.2017.00191.
[41] Sargurupremraj M, Suzuki H, Jian X, Sarnowski C, Evans TE, Bis JC, et al. Cerebral small
vessel disease genomics and its implications across the lifespan. Nat Commun
2020;11:6285. https://doi.org/10.1038/s41467-020-19111-2.
[42] Go AS, Hylek EM, Phillips KA, Chang Y, Henault LE, Selby JV, et al. Prevalence of
Diagnosed Atrial Fibrillation in AdultsNational Implications for Rhythm Management and
Stroke Prevention: the AnTicoagulation and Risk Factors In Atrial Fibrillation (ATRIA)
Study. JAMA 2001;285:2370–5. https://doi.org/10.1001/jama.285.18.2370.
[43] Hahne K, Mönnig G, Samol A. Atrial fibrillation and silent stroke: links, risks, and
challenges. Vasc Health Risk Manag 2016;12:65–74.
https://doi.org/10.2147/VHRM.S81807.
[44] Kornej J, Börschel CS, Benjamin EJ, Schnabel RB. Epidemiology of Atrial Fibrillation in
the 21st Century. Circulation Research 2020;127:4–20.
https://doi.org/10.1161/CIRCRESAHA.120.316340.
[45] Ryder KM, Benjamin EJ. Epidemiology and significance of atrial fibrillation. The
American Journal of Cardiology 1999;84:131–8. https://doi.org/10.1016/S0002-
9149(99)00713-4.
[46] Morillo CA, Banerjee A, Perel P, Wood D, Jouven X. Atrial fibrillation: the current
epidemic. J Geriatr Cardiol 2017;14:195–203. https://doi.org/10.11909/j.issn.1671-
5411.2017.03.011.
86
[47] de Bruijn RFAG, Heeringa J, Wolters FJ, Franco OH, Stricker BHC, Hofman A, et al.
Association Between Atrial Fibrillation and Dementia in the General Population. JAMA
Neurology 2015;72:1288–94. https://doi.org/10.1001/jamaneurol.2015.2161.
[48] Alonso A, Knopman DS, Gottesman RF, Soliman EZ, Shah AJ, O’Neal WT, et al.
Correlates of Dementia and Mild Cognitive Impairment in Patients With Atrial Fibrillation:
The Atherosclerosis Risk in Communities Neurocognitive Study (ARIC-NCS). Journal of
the American Heart Association 2017;6:e006014.
https://doi.org/10.1161/JAHA.117.006014.
[49] Nishtala A, Piers RJ, Himali JJ, Beiser AS, Davis-Plourde KL, Saczynski JS, et al. Atrial
fibrillation and cognitive decline in the Framingham Heart Study. Heart Rhythm
2018;15:166–72. https://doi.org/10.1016/j.hrthm.2017.09.036.
[50] Jacobs V, Woller SC, Stevens S, May HT, Bair TL, Anderson JL, et al. Time outside of
therapeutic range in atrial fibrillation patients is associated with long-term risk of dementia.
Heart Rhythm 2014;11:2206–13. https://doi.org/10.1016/j.hrthm.2014.08.013.
[51] Knecht S, Oelschläger C, Duning T, Lohmann H, Albers J, Stehling C, et al. Atrial
fibrillation in stroke-free patients is associated with memory impairment and hippocampal
atrophy. European Heart Journal 2008;29:2125–32.
https://doi.org/10.1093/eurheartj/ehn341.
[52] Friberg L, Andersson T, Rosenqvist M. Less dementia and stroke in low-risk patients with
atrial fibrillation taking oral anticoagulation. European Heart Journal 2019;40:2327–35.
https://doi.org/10.1093/eurheartj/ehz304.
[53] Friberg L, Rosenqvist M. Less dementia with oral anticoagulation in atrial fibrillation.
European Heart Journal 2018;39:453–60. https://doi.org/10.1093/eurheartj/ehx579.
[54] Kim D, Yang P-S, Sung J-H, Jang E, Yu HT, Kim T-H, et al. Less dementia after catheter
ablation for atrial fibrillation: a nationwide cohort study. European Heart Journal
2020;41:4483–93. https://doi.org/10.1093/eurheartj/ehaa726.
[55] Saito T, Kawamura Y, Tanabe Y, Asanome A, Takahashi K, Sawada J, et al. Cerebral
Microbleeds and Asymptomatic Cerebral Infarctions in Patients with Atrial Fibrillation.
Journal of Stroke and Cerebrovascular Diseases 2014;23:1616–22.
https://doi.org/10.1016/j.jstrokecerebrovasdis.2014.01.005.
[56] Gardarsdottir M, Sigurdsson S, Aspelund T, Rokita H, Launer LJ, Gudnason V, et al. Atrial
fibrillation is associated with decreased total cerebral blood flow and brain perfusion. EP
Europace 2018;20:1252–8. https://doi.org/10.1093/europace/eux220.
[57] Chen LY, Lopez FL, Gottesman RF, Huxley RR, Agarwal SK, Loehr L, et al. Atrial
Fibrillation and Cognitive Decline–The Role of Subclinical Cerebral Infarcts. Stroke
2014;45:2568–74. https://doi.org/10.1161/STROKEAHA.114.005243.
87
[58] Kalantarian S, Ay H, Gollub RL, Lee H, Retzepi K, Mansour M, et al. Association Between
Atrial Fibrillation and Silent Cerebral Infarctions. Ann Intern Med 2014;161:650–8.
https://doi.org/10.7326/M14-0538.
[59] Stefansdottir H, Arnar DO, Aspelund T, Sigurdsson S, Jonsdottir MK, Hjaltason H, et al.
Atrial Fibrillation is Associated With Reduced Brain Volume and Cognitive Function
Independent of Cerebral Infarcts. Stroke 2013;44:1020–5.
https://doi.org/10.1161/STROKEAHA.12.679381.
[60] Graff-Radford J, Madhavan M, Vemuri P, Rabinstein AA, Cha RH, Mielke MM, et al.
Atrial fibrillation, cognitive impairment, and neuroimaging. Alzheimer’s & Dementia
2016;12:391–8. https://doi.org/10.1016/j.jalz.2015.08.164.
[61] Qureshi AI, Saed A, Tasneem N, Adil MM. Neuroanatomical correlates of atrial
fibrillation: a longitudinal MRI study. J Vasc Interv Neurol 2014;7:18–23.
[62] Piers RJ, Nishtala A, Preis SR, DeCarli C, Wolf PA, Benjamin EJ, et al. Association
between atrial fibrillation and volumetric magnetic resonance imaging brain measures:
Framingham Offspring Study. Heart Rhythm 2016;13:2020–4.
https://doi.org/10.1016/j.hrthm.2016.07.004.
[63] Bycroft C, Freeman C, Petkova D, Band G, Elliott LT, Sharp K, et al. The UK Biobank
resource with deep phenotyping and genomic data. Nature 2018;562:203–9.
https://doi.org/10.1038/s41586-018-0579-z.
[64] Sudlow C, Gallacher J, Allen N, Beral V, Burton P, Danesh J, et al. UK Biobank: An Open
Access Resource for Identifying the Causes of a Wide Range of Complex Diseases of
Middle and Old Age. PLOS Medicine 2015;12:e1001779.
https://doi.org/10.1371/journal.pmed.1001779.
[65] Littlejohns TJ, Holliday J, Gibson LM, Garratt S, Oesingmann N, Alfaro-Almagro F, et al.
The UK Biobank imaging enhancement of 100,000 participants: rationale, data collection,
management and future directions. Nat Commun 2020;11:2624.
https://doi.org/10.1038/s41467-020-15948-9.
[66] Khurshid Shaan, Choi Seung Hoan, Weng Lu-Chen, Wang Elizabeth Y., Trinquart
Ludovic, Benjamin Emelia J., et al. Frequency of Cardiac Rhythm Abnormalities in a Half
Million Adults. Circulation: Arrhythmia and Electrophysiology 2018;11:e006273.
https://doi.org/10.1161/CIRCEP.118.006273.
[67] Miller KL, Alfaro-Almagro F, Bangerter NK, Thomas DL, Yacoub E, Xu J, et al.
Multimodal population brain imaging in the UK Biobank prospective epidemiological
study. Nat Neurosci 2016;19:1523–36. https://doi.org/10.1038/nn.4393.
88
[68] Fischl B, Salat DH, Busa E, Albert M, Dieterich M, Haselgrove C, et al. Whole brain
segmentation: automated labeling of neuroanatomical structures in the human brain.
Neuron 2002;33:341–55.
[69] Mori S, Oishi K, Jiang H, Jiang L, Li X, Akhter K, et al. Stereotaxic white matter atlas
based on diffusion tensor imaging in an ICBM template. NeuroImage 2008;40:570–82.
https://doi.org/10.1016/j.neuroimage.2007.12.035.
[70] Smith SM, Jenkinson M, Johansen-Berg H, Rueckert D, Nichols TE, Mackay CE, et al.
Tract-based spatial statistics: Voxelwise analysis of multi-subject diffusion data.
NeuroImage 2006;31:1487–505. https://doi.org/10.1016/j.neuroimage.2006.02.024.
[71] Dickerson BC, Bakkour A, Salat DH, Feczko E, Pacheco J, Greve DN, et al. The Cortical
Signature of Alzheimer’s Disease: Regionally Specific Cortical Thinning Relates to
Symptom Severity in Very Mild to Mild AD Dementia and is Detectable in Asymptomatic
Amyloid-Positive Individuals. Cereb Cortex 2009;19:497–510.
https://doi.org/10.1093/cercor/bhn113.
[72] Kantarci K. Fractional Anisotropy of the Fornix and Hippocampal Atrophy in Alzheimer’s
Disease. Front Aging Neurosci 2014;6. https://doi.org/10.3389/fnagi.2014.00316.
[73] Thompson PM, Hayashi KM, de Zubicaray GI, Janke AL, Rose SE, Semple J, et al.
Mapping hippocampal and ventricular change in Alzheimer disease. NeuroImage
2004;22:1754–66. https://doi.org/10.1016/j.neuroimage.2004.03.040.
[74] Thomas AG, Koumellis P, Dineen RA. The Fornix in Health and Disease: An Imaging
Review. RadioGraphics 2011;31:1107–21. https://doi.org/10.1148/rg.314105729.
[75] R Core Team. R: A Language and Environment for Statistical Computing. Vienna, Austria:
R Foundation for Statistical Computing; 2020.
[76] Cooper H, Hedges LV, Valentine JC. The Handbook of Research Synthesis and Meta-
Analysis. Russell Sage Foundation; 2009.
[77] Rosseel Y. lavaan: An R Package for Structural Equation Modeling. Journal of Statistical
Software 2012;48:1–36. https://doi.org/10.18637/jss.v048.i02.
[78] Harrell Jr FE. rms: Regression Modeling Strategies. 2021.
[79] Mowinckel AM, Vidal-Piñeiro D. Visualization of Brain Statistics With R Packages ggseg
and ggseg3d. Advances in Methods and Practices in Psychological Science 2020;3:466–83.
https://doi.org/10.1177/2515245920928009.
[80] Chen Lin Y., Agarwal Sunil K., Norby Faye L., Gottesman Rebecca F., Loehr Laura R.,
Soliman Elsayed Z., et al. Persistent but not Paroxysmal Atrial Fibrillation Is Independently
89
Associated With Lower Cognitive Function. Journal of the American College of Cardiology
2016;67:1379–80. https://doi.org/10.1016/j.jacc.2015.11.064.
[81] Loo SY, Dell’Aniello S, Huiart L, Renoux C. Trends in the prescription of novel oral
anticoagulants in UK primary care. British Journal of Clinical Pharmacology
2017;83:2096–106. https://doi.org/10.1111/bcp.13299.
[82] Chen LY, Norby FL, Gottesman RF, Mosley TH, Soliman EZ, Agarwal SK, et al.
Association of Atrial Fibrillation With Cognitive Decline and Dementia Over 20 Years:
The ARIC-NCS (Atherosclerosis Risk in Communities Neurocognitive Study). Journal of
the American Heart Association 2018;7:e007301.
https://doi.org/10.1161/JAHA.117.007301.
[83] Shao IY, Power MC, Mosley T, Jack C, Gottesman RF, Chen LY, et al. Association of
Atrial Fibrillation With White Matter Disease. Stroke 2019;50:989–91.
https://doi.org/10.1161/STROKEAHA.118.023386.
[84] Chen LY, Soliman EZ. P Wave Indices—Advancing Our Understanding of Atrial
Fibrillation-Related Cardiovascular Outcomes. Front Cardiovasc Med 2019;6:53.
https://doi.org/10.3389/fcvm.2019.00053.
[85] Magnani JW, Williamson MA, Ellinor PT, Monahan KM, Benjamin EJ. P Wave Indices.
Circulation: Arrhythmia and Electrophysiology 2009;2:72–9.
https://doi.org/10.1161/CIRCEP.108.806828.
[86] Bradley SM, Marriott HJL. Intra-Atrial Block. Circulation 1956;14:1073–8.
https://doi.org/10.1161/01.CIR.14.6.1073.
[87] Goette A, Kalman JM, Aguinaga L, Akar J, Cabrera JA, Chen SA, et al.
EHRA/HRS/APHRS/SOLAECE expert consensus on atrial cardiomyopathies: Definition,
characterization, and clinical implication. Heart Rhythm 2017;14:e3–40.
https://doi.org/10.1016/j.hrthm.2016.05.028.
[88] Auricchio A, Özkartal T, Salghetti F, Neumann L, Pezzuto S, Gharaviri A, et al. Short P-
Wave Duration is a Marker of Higher Rate of Atrial Fibrillation Recurrences after
Pulmonary Vein Isolation: New Insights into the Pathophysiological Mechanisms Through
Computer Simulations. Journal of the American Heart Association 2021;10:e018572.
https://doi.org/10.1161/JAHA.120.018572.
[89] Bunch TJ, Weiss JP, Crandall BG, May HT, Bair TL, Osborn JS, et al. Atrial fibrillation is
independently associated with senile, vascular, and Alzheimer’s dementia. Heart Rhythm
2010;7:433–7. https://doi.org/10.1016/j.hrthm.2009.12.004.
[90] Chen LY, Shen W-K. Epidemiology of atrial fibrillation: A current perspective. Heart
Rhythm 2007;4:S1–6. https://doi.org/10.1016/j.hrthm.2006.12.018.
90
[91] Nielsen JB, Kühl JT, Pietersen A, Graff C, Lind B, Struijk JJ, et al. P-wave duration and the
risk of atrial fibrillation: Results from the Copenhagen ECG Study. Heart Rhythm
2015;12:1887–95. https://doi.org/10.1016/j.hrthm.2015.04.026.
[92] Hari KJ, Nguyen TP, Soliman EZ. Relationship between P-wave duration and the risk of
atrial fibrillation. Expert Review of Cardiovascular Therapy 2018;16:837–43.
https://doi.org/10.1080/14779072.2018.1533814.
[93] Magnani JW, Zhu L, Lopez F, Pencina MJ, Agarwal SK, Soliman EZ, et al. P-wave indices
and atrial fibrillation: Cross-cohort assessments from the Framingham Heart Study (FHS)
and Atherosclerosis Risk in Communities (ARIC) study. American Heart Journal
2015;169:53-61.e1. https://doi.org/10.1016/j.ahj.2014.10.009.
[94] Barnes ME, Miyasaka Y, Seward JB, Gersh BJ, Rosales AG, Bailey KR, et al. Left Atrial
Volume in the Prediction of First Ischemic Stroke in an Elderly Cohort Without Atrial
Fibrillation. Mayo Clinic Proceedings 2004;79:1008–14. https://doi.org/10.4065/79.8.1008.
[95] Gutierrez A, Norby FL, Maheshwari A, Rooney MR, Gottesman RF, Mosley TH, et al.
Association of Abnormal P-Wave Indices With Dementia and Cognitive Decline Over
25 Years: ARIC-NCS (The Atherosclerosis Risk in Communities Neurocognitive Study).
Journal of the American Heart Association 2019;8:e014553.
https://doi.org/10.1161/JAHA.119.014553.
[96] Kamel H, Soliman EZ, Heckbert SR, Kronmal RA, Longstreth W t., Nazarian S, et al. P-
Wave Morphology and the Risk of Incident Ischemic Stroke in the Multi-Ethnic Study of
Atherosclerosis. Stroke 2014;45:2786–8.
https://doi.org/10.1161/STROKEAHA.114.006364.
[97] Magnani JW, Gorodeski EZ, Johnson VM, Sullivan LM, Hamburg NM, Benjamin EJ, et al.
P wave duration is associated with cardiovascular and all-cause mortality outcomes: the
National Health and Nutrition Examination Survey. Heart Rhythm 2011;8:93–100.
https://doi.org/10.1016/j.hrthm.2010.09.020.
[98] Maheshwari A, Norby FL, Soliman EZ, Alraies MC, Adabag S, O’neal WT, et al. Relation
of Prolonged P-wave Duration to Risk of Sudden Cardiac Death in the General Population
(From the Atherosclerosis Risk in Communities Study). Am J Cardiol 2017;119:1302–6.
https://doi.org/10.1016/j.amjcard.2017.01.012.
[99] Herrera C, Bruña V, Abizanda P, Díez-Villanueva P, Formiga F, Torres R, et al. Relation of
Interatrial Block to Cognitive Impairment in Patients ≥ 70 Years of Age (From the
CAMBIAD Case-control Study). The American Journal of Cardiology 2020;136:94–9.
https://doi.org/10.1016/j.amjcard.2020.09.008.
[100] Kamel H, Bartz TM, Longstreth W t., Okin PM, Thacker EL, Patton KK, et al.
Association Between Left Atrial Abnormality on ECG and Vascular Brain Injury on MRI
91
in the Cardiovascular Health Study. Stroke 2015;46:711–6.
https://doi.org/10.1161/STROKEAHA.114.007762.
[101] Healthcare G. Marquette 12SL ECG Analysis Program. Statement of Validation and
Accuracy 2007.
[102] Griffanti L, Zamboni G, Khan A, Li L, Bonifacio G, Sundaresan V, et al. BIANCA
(Brain Intensity AbNormality Classification Algorithm): A new tool for automated
segmentation of white matter hyperintensities. NeuroImage 2016;141:191–205.
https://doi.org/10.1016/j.neuroimage.2016.07.018.
[103] Hanscombe KB, Coleman JRI, Traylor M, Lewis CM. ukbtools: An R package to
manage and query UK Biobank data. PLOS ONE 2019;14:e0214311.
https://doi.org/10.1371/journal.pone.0214311.
[104] Harrell FE. Regression modeling strategies: with applications to linear models, logistic
and ordinal regression, and survival analysis. vol. 3. Springer; 2015.
[105] Tingley D, Yamamoto T, Hirose K, Keele L, Imai K. mediation: R Package for Causal
Mediation Analysis. J Stat Soft 2014;59. https://doi.org/10.18637/jss.v059.i05.
[106] Feekes JA, Hsu S-W, Chaloupka JC, Cassell MD. Tertiary microvascular territories
define lacunar infarcts in the basal ganglia. Ann Neurol 2005;58:18–30.
https://doi.org/10.1002/ana.20505.
[107] Fisher CM. LACUNES: SMALL, DEEP CEREBRAL INFARCTS. Neurology
1965;15:774–84. https://doi.org/10.1212/wnl.15.8.774.
[108] Giele JLP, Witkamp TD, Mali WPTM, van der Graaf Y. Silent Brain Infarcts in Patients
With Manifest Vascular Disease. Stroke 2004;35:742–6.
https://doi.org/10.1161/01.STR.0000117572.56058.2A.
[109] Gold G, Kövari E, Herrmann FR, Canuto A, Hof PR, Michel J-P, et al. Cognitive
Consequences of Thalamic, Basal Ganglia, and Deep White Matter Lacunes in Brain Aging
and Dementia. Stroke 2005;36:1184–8.
https://doi.org/10.1161/01.STR.0000166052.89772.b5.
[110] Smith EE, Schneider JA, Wardlaw JM, Greenberg SM. Cerebral microinfarcts: the
invisible lesions. The Lancet Neurology 2012;11:272–82. https://doi.org/10.1016/S1474-
4422(11)70307-6.
[111] Kloppenborg RP, Nederkoorn PJ, Grool AM, Vincken KL, Mali WPTM, Vermeulen M,
et al. Cerebral small-vessel disease and progression of brain atrophy: The SMART-MR
study. Neurology 2012;79:2029–36. https://doi.org/10.1212/WNL.0b013e3182749f02.
92
[112] Thong JYJ, Hilal S, Wang Y, Soon HW, Dong Y, Collinson SL, et al. Association of
silent lacunar infarct with brain atrophy and cognitive impairment. J Neurol Neurosurg
Psychiatry 2013;84:1219–25. https://doi.org/10.1136/jnnp-2013-305310.
[113] Sachdev PS, Chen X, Joscelyne A, Wen W, Brodaty H. Amygdala in Stroke/Transient
Ischemic Attack Patients and Its Relationship to Cognitive Impairment and
Psychopathology: The Sydney Stroke Study. The American Journal of Geriatric Psychiatry
2007;15:487–96. https://doi.org/10.1097/JGP.0b013e3180581fe6.
[114] Kalaria RN. Neuropathological diagnosis of vascular cognitive impairment and vascular
dementia with implications for Alzheimer’s disease. Acta Neuropathol 2016;131:659–85.
https://doi.org/10.1007/s00401-016-1571-z.
[115] Liu C, Li C, Gui L, Zhao L, Evans AC, Xie B, et al. The pattern of brain gray matter
impairments in patients with subcortical vascular dementia. Journal of the Neurological
Sciences 2014;341:110–8. https://doi.org/10.1016/j.jns.2014.04.017.
[116] Haacke EM, Mittal S, Wu Z, Neelavalli J, Cheng Y-CN. Susceptibility-Weighted
Imaging: Technical Aspects and Clinical Applications, Part 1. American Journal of
Neuroradiology 2009;30:19–30. https://doi.org/10.3174/ajnr.A1400.
[117] Apostolova LG, Green AE, Babakchanian S, Hwang KS, Chou Y-Y, Toga AW, et al.
Hippocampal Atrophy and Ventricular Enlargement in Normal Aging, Mild Cognitive
Impairment (MCI), and Alzheimer Disease. Alzheimer Disease & Associated Disorders
2012;26:17–27. https://doi.org/10.1097/WAD.0b013e3182163b62.
[118] Li M, Meng Y, Wang M, Yang S, Wu H, Zhao B, et al. Cerebral gray matter volume
reduction in subcortical vascular mild cognitive impairment patients and subcortical
vascular dementia patients, and its relation with cognitive deficits. Brain and Behavior
2017;7:e00745. https://doi.org/10.1002/brb3.745.
[119] Seo SW, Ahn J, Yoon U, Im K, Lee J-M, Tae Kim S, et al. Cortical thinning in vascular
mild cognitive impairment and vascular dementia of subcortical type. J Neuroimaging
2010;20:37–45. https://doi.org/10.1111/j.1552-6569.2008.00293.x.
[120] Raman MR, Preboske GM, Przybelski SA, Gunter JL, Senjem ML, Vemuri P, et al.
Antemortem MRI findings associated with microinfarcts at autopsy. Neurology
2014;82:1951–8. https://doi.org/10.1212/WNL.0000000000000471.
[121] Scheff SW, Price DA, Schmitt FA, Scheff MA, Mufson EJ. Synaptic Loss in the Inferior
Temporal Gyrus in Mild Cognitive Impairment and Alzheimer’s Disease. Journal of
Alzheimer’s Disease 2011;24:547–57. https://doi.org/10.3233/JAD-2011-101782.
[122] Dagli N, Karaca I, Yavuzkir M, Balin M, Arslan N. Are maximum P wave duration and P
wave dispersion a marker of target organ damage in the hypertensive population? Clin Res
Cardiol 2008;97:98–104. https://doi.org/10.1007/s00392-007-0587-8.
93
[123] Havmoller R, Carlson J, Holmqvist F, Herreros A, Meurling CJ, Olsson B, et al. Age-
related changes in P wave morphology in healthy subjects. BMC Cardiovascular Disorders
2007;7:22. https://doi.org/10.1186/1471-2261-7-22.
[124] Kosar F, Aksoy Y, Ari F, Keskin L, Sahin I. P-Wave Duration and Dispersion in Obese
Subjects. Annals of Noninvasive Electrocardiology 2008;13:3–7.
https://doi.org/10.1111/j.1542-474X.2007.00194.x.
[125] Duru M, Seyfeli E, Kuvandik G, Kaya H, Yalcin F. Effect of Weight Loss on P Wave
Dispersion in Obese Subjects. Obesity 2006;14:1378–82.
https://doi.org/10.1038/oby.2006.156.
[126] Russo V, Ammendola E, De Crescenzo I, Docimo L, Santangelo L, Calabrò R. Severe
Obesity and P-Wave Dispersion: The Effect of Surgically Induced Weight Loss. OBES
SURG 2008;18:90–6. https://doi.org/10.1007/s11695-007-9340-7.
[127] Buxton AE, Josephson ME. The Role of P Wave Duration as a Predictor of Postoperative
Atrial Arrhythmias. Chest 1981;80:68–73. https://doi.org/10.1378/chest.80.1.68.
[128] Yilmaz R, Demirbag R. P-wave dispersion in patients with stable coronary artery disease
and its relationship with severity of the disease. Journal of Electrocardiology 2005;38:279–
84. https://doi.org/10.1016/j.jelectrocard.2005.02.003.
[129] Beauchet O, Celle S, Roche F, Bartha R, Montero-Odasso M, Allali G, et al. Blood
pressure levels and brain volume reduction: a systematic review and meta-analysis. Journal
of Hypertension 2013;31:1502–16. https://doi.org/10.1097/HJH.0b013e32836184b5.
[130] Cox SR, Lyall DM, Ritchie SJ, Bastin ME, Harris MA, Buchanan CR, et al. Associations
between vascular risk factors and brain MRI indices in UK Biobank. European Heart
Journal 2019;40:2290–300. https://doi.org/10.1093/eurheartj/ehz100.
[131] Ho AJ, Raji CA, Becker JT, Lopez OL, Kuller LH, Hua X, et al. Obesity is linked with
lower brain volume in 700 AD and MCI patients. Neurobiology of Aging 2010;31:1326–
39. https://doi.org/10.1016/j.neurobiolaging.2010.04.006.
[132] Zhao L, Matloff W, Ning K, Kim H, Dinov ID, Toga AW. Age-Related Differences in
Brain Morphology and the Modifiers in Middle-Aged and Older Adults. Cereb Cortex
2019;29:4169–93. https://doi.org/10.1093/cercor/bhy300.
[133] Kang DO, Eo JS, Park EJ, Nam HS, Song JW, Park YH, et al. Stress-associated
neurobiological activity is linked with acute plaque instability via enhanced macrophage
activity: a prospective serial 18F-FDG-PET/CT imaging assessment. European Heart
Journal 2021;42:1883–95. https://doi.org/10.1093/eurheartj/ehaa1095.
94
[134] Chen Z, Venkat P, Seyfried D, Chopp M, Yan T, Chen J. Brain-heart interaction: cardiac
complications after stroke. Circ Res 2017;121:451–68.
https://doi.org/10.1161/CIRCRESAHA.117.311170.
[135] Topiwala A, Allan CL, Valkanova V, Zsoldos E, Filippini N, Sexton C, et al. Moderate
alcohol consumption as risk factor for adverse brain outcomes and cognitive decline:
longitudinal cohort study. BMJ 2017;357:j2353. https://doi.org/10.1136/bmj.j2353.
[136] Lupton MK, Strike L, Hansell NK, Wen W, Mather KA, Armstrong NJ, et al. The effect
of increased genetic risk for Alzheimer’s disease on hippocampal and amygdala volume.
Neurobiol Aging 2016;40:68–77. https://doi.org/10.1016/j.neurobiolaging.2015.12.023.
[137] Anttila T, Helkala E-L, Viitanen M, Kåreholt I, Fratiglioni L, Winblad B, et al. Alcohol
drinking in middle age and subsequent risk of mild cognitive impairment and dementia in
old age: a prospective population based study. BMJ 2004;329:539.
https://doi.org/10.1136/bmj.38181.418958.BE.
[138] Dufouil C, Tzourio C, Brayne C, Berr C, Amouyel P, Alpérovitch A. Influence of
apolipoprotein E genotype on the risk of cognitive deterioration in moderate drinkers and
smokers. Epidemiology 2000;11:280–4. https://doi.org/10.1097/00001648-200005000-
00009.
[139] den Heijer T, Vermeer SE, van Dijk EJ, Prins ND, Koudstaal PJ, van Duijn CM, et al.
Alcohol intake in relation to brain magnetic resonance imaging findings in older persons
without dementia. Am J Clin Nutr 2004;80:992–7. https://doi.org/10.1093/ajcn/80.4.992.
[140] Carmelli D, Swan GE, Reed T, Schellenberg GD, Christian JC. The effect of
apolipoprotein E epsilon4 in the relationships of smoking and drinking to cognitive
function. Neuroepidemiology 1999;18:125–33. https://doi.org/10.1159/000026204.
[141] Reas ET, Laughlin GA, Kritz-Silverstein D, Barrett-Connor E, McEvoy LK. Moderate,
Regular Alcohol Consumption is Associated with Higher Cognitive Function in Older
Community-Dwelling Adults. J Prev Alzheimers Dis 2016;3:105–13.
https://doi.org/10.14283/jpad.2016.89.
[142] Heffernan M, Mather KA, Xu J, Assareh AA, Kochan NA, Reppermund S, et al. Alcohol
Consumption and Incident Dementia: Evidence from the Sydney Memory and Ageing
Study. J Alzheimers Dis 2016;52:529–38. https://doi.org/10.3233/JAD-150537.
[143] Herring D, Paulson D. Moderate alcohol use and apolipoprotein E-4 (ApoE-4):
Independent effects on cognitive outcomes in later life. J Clin Exp Neuropsychol
2018;40:326–37. https://doi.org/10.1080/13803395.2017.1343803.
[144] Ruitenberg A, van Swieten JC, Witteman JC, Mehta KM, van Duijn CM, Hofman A, et
al. Alcohol consumption and risk of dementia: the Rotterdam Study. The Lancet
2002;359:281–6. https://doi.org/10.1016/S0140-6736(02)07493-7.
95
[145] McEvoy LK, Fennema-Notestine C, Elman JA, Eyler LT, Franz CE, Hagler DJ, et al.
Alcohol intake and brain white matter in middle aged men: Microscopic and macroscopic
differences. NeuroImage: Clinical 2018;18:390–8.
https://doi.org/10.1016/j.nicl.2018.02.006.
[146] Li C, Loewenstein DA, Duara R, Cabrerizo M, Barker W, Adjouadi M, et al. The
Relationship of Brain Amyloid Load and APOE Status to Regional Cortical Thinning and
Cognition in the ADNI Cohort. J Alzheimers Dis 2017;59:1269–82.
https://doi.org/10.3233/JAD-170286.
[147] Li J, Cheng J. Apolipoprotein E4 exacerbates ethanol-induced neurotoxicity through
augmentation of oxidative stress and apoptosis in N2a-APP cells. Neuroscience Letters
2018;665:1–6. https://doi.org/10.1016/j.neulet.2017.11.038.
[148] Lundgaard I, Wang W, Eberhardt A, Vinitsky HS, Reeves BC, Peng S, et al. Beneficial
effects of low alcohol exposure, but adverse effects of high alcohol intake on glymphatic
function. Scientific Reports 2018;8:2246. https://doi.org/10.1038/s41598-018-20424-y.
[149] Wildsmith KR, Holley M, Savage JC, Skerrett R, Landreth GE. Evidence for impaired
amyloid β clearance in Alzheimer’s disease. Alzheimers Res Ther 2013;5:33.
https://doi.org/10.1186/alzrt187.
[150] Ding Jingzhong, Eigenbrodt Marsha L., Mosley Thomas H., Hutchinson Richard G.,
Folsom Aaron R., Harris Tamara B., et al. Alcohol Intake and Cerebral Abnormalities on
Magnetic Resonance Imaging in a Community-Based Population of Middle-Aged Adults.
Stroke 2004;35:16–21. https://doi.org/10.1161/01.STR.0000105929.88691.8E.
[151] Gu Y, Scarmeas N, Short EE, Luchsinger JA, DeCarli C, Stern Y, et al. Alcohol intake
and brain structure in a multiethnic elderly cohort. Clin Nutr 2014;33:662–7.
https://doi.org/10.1016/j.clnu.2013.08.004.
[152] Kubota M, Nakazaki S, Hirai S, Saeki N, Yamaura A, Kusaka T. Alcohol consumption
and frontal lobe shrinkage: study of 1432 non-alcoholic subjects. Journal of Neurology,
Neurosurgery & Psychiatry 2001;71:104–6. https://doi.org/10.1136/jnnp.71.1.104.
[153] Fjell AM, Walhovd KB. Structural brain changes in aging: courses, causes and cognitive
consequences. Rev Neurosci 2010;21:187–221.
[154] Topiwala A, Ebmeier KP. Effects of drinking on late-life brain and cognition. Evidence-
Based Mental Health 2018;21:12–5. https://doi.org/10.1136/eb-2017-102820.
[155] Zhao L, Matloff W, Ning K, Kim H, Dinov ID, Toga AW. Age-Related Differences in
Brain Morphology and the Modifiers in Middle-Aged and Older Adults. Cerebral Cortex
2018.
96
[156] Sabuncu MR, Desikan RS, Sepulcre J, Yeo BTT, Liu H, Schmansky NJ, et al. The
Dynamics of Cortical and Hippocampal Atrophy in Alzheimer Disease. Arch Neurol
2011;68:1040–8. https://doi.org/10.1001/archneurol.2011.167.
[157] McCarthy CS, Ramprashad A, Thompson C, Botti J-A, Coman IL, Kates WR. A
comparison of FreeSurfer-generated data with and without manual intervention. Front
Neurosci 2015;9. https://doi.org/10.3389/fnins.2015.00379.
[158] Desikan RS, Fan CC, Wang Y, Schork AJ, Cabral HJ, Cupples LA, et al. Genetic
assessment of age-associated Alzheimer disease risk: Development and validation of a
polygenic hazard score. PLOS Medicine 2017;14:e1002258.
https://doi.org/10.1371/journal.pmed.1002258.
[159] Machiela MJ, Chanock SJ. LDlink: a web-based application for exploring population-
specific haplotype structure and linking correlated alleles of possible functional variants.
Bioinformatics 2015;31:3555–7. https://doi.org/10.1093/bioinformatics/btv402.
[160] Lacour A, Espinosa A, Louwersheimer E, Heilmann S, Hernández I, Wolfsgruber S, et al.
Genome-wide significant risk factors for Alzheimer’s disease: role in progression to
dementia due to Alzheimer’s disease among subjects with mild cognitive impairment.
Molecular Psychiatry 2017;22:153–60. https://doi.org/10.1038/mp.2016.18.
[161] Efthymiou AG, Goate AM. Late onset Alzheimer’s disease genetics implicates microglial
pathways in disease risk. Molecular Neurodegeneration 2017;12:43.
https://doi.org/10.1186/s13024-017-0184-x.
[162] Ordovas JM, Litwack-Klein L, Wilson PW, Schaefer MM, Schaefer EJ. Apolipoprotein E
isoform phenotyping methodology and population frequency with identification of apoE1
and apoE5 isoforms. J Lipid Res 1987;28:371–80.
[163] Euesden J, Lewis CM, O’Reilly PF. PRSice: Polygenic Risk Score software.
Bioinformatics 2015;31:1466–8. https://doi.org/10.1093/bioinformatics/btu848.
[164] Lambert JC, Ibrahim-Verbaas CA, Harold D, Naj AC, Sims R, Bellenguez C, et al. Meta-
analysis of 74,046 individuals identifies 11 new susceptibility loci for Alzheimer’s disease.
Nat Genet 2013;45:1452–8. https://doi.org/10.1038/ng.2802.
[165] Johnson DK, Wilkins CH, Morris JC. Accelerated Weight Loss May Precede Diagnosis
in Alzheimer Disease. Arch Neurol 2006;63:1312–7.
https://doi.org/10.1001/archneur.63.9.1312.
[166] Keller MC. Gene-by-environment interaction studies have not properly controlled for
potential confounders: The problem and the (simple) solution. Biol Psychiatry 2014;75.
https://doi.org/10.1016/j.biopsych.2013.09.006.
97
[167] Fekjær HO. Alcohol—a universal preventive agent? A critical analysis. Addiction
2013;108:2051–7. https://doi.org/10.1111/add.12104.
[168] Lyall DM, Cox SR, Lyall LM, Celis-Morales C, Cullen B, Mackay DF, et al. Association
between APOE e4 and white matter hyperintensity volume, but not total brain volume or
white matter integrity. Brain Imaging Behav 2019. https://doi.org/10.1007/s11682-019-
00069-9.
[169] Mishra S, Blazey TM, Holtzman DM, Cruchaga C, Su Y, Morris JC, et al. Longitudinal
brain imaging in preclinical Alzheimer disease: impact of APOE ε4 genotype. Brain
2018;141:1828–39. https://doi.org/10.1093/brain/awy103.
[170] Chételat G, Villemagne VL, Pike KE, Baron J-C, Bourgeat P, Jones G, et al. Larger
temporal volume in elderly with high versus low beta-amyloid deposition. Brain
2010;133:3349–58. https://doi.org/10.1093/brain/awq187.
[171] Fortea J, Sala-Llonch R, Bartrés-Faz D, Lladó A, Solé-Padullés C, Bosch B, et al.
Cognitively Preserved Subjects with Transitional Cerebrospinal Fluid ß-Amyloid 1-42
Values Have Thicker Cortex in Alzheimer’s Disease Vulnerable Areas. Biological
Psychiatry 2011;70:183–90. https://doi.org/10.1016/j.biopsych.2011.02.017.
[172] Johnson SC, Christian BT, Okonkwo OC, Oh JM, Harding S, Xu G, et al. Amyloid
burden and neural function in people at risk for Alzheimer’s Disease. Neurobiol Aging
2014;35:576–84. https://doi.org/10.1016/j.neurobiolaging.2013.09.028.
[173] Lundqvist C, Volk B, Knoth R, Alling C. Long-term effects of intermittent versus
continuous ethanol exposure on hippocampal synapses of the rat. Acta Neuropathol
1994;87:242–9. https://doi.org/10.1007/BF00296739.
[174] Hassing LB. Light Alcohol Consumption Does Not Protect Cognitive Function: A
Longitudinal Prospective Study. Frontiers in Aging Neuroscience 2018;10.
https://doi.org/10.3389/fnagi.2018.00081.
[175] Xie L, Wisse LEM, Das SR, Wang H, Wolk DA, Manjón JV, et al. Accounting for the
Confound of Meninges in Segmenting Entorhinal and Perirhinal Cortices in T1-Weighted
MRI. Med Image Comput Comput Assist Interv 2016;9901:564–71.
https://doi.org/10.1007/978-3-319-46723-8_65.
[176] Aho L, Karkola K, Juusela J, Alafuzoff I. Heavy alcohol consumption and
neuropathological lesions: a post-mortem human study. J Neurosci Res 2009;87:2786–92.
https://doi.org/10.1002/jnr.22091.
[177] Venkataraman A, Kalk N, Sewell G, Ritchie CW, Lingford-Hughes A. Alcohol and
Alzheimer’s Disease—Does Alcohol Dependence Contribute to Beta-Amyloid Deposition,
Neuroinflammation and Neurodegeneration in Alzheimer’s Disease? Alcohol Alcohol
2017;52:151–8. https://doi.org/10.1093/alcalc/agw092.
98
[178] Cherbuin N, Sargent-Cox K, Fraser M, Sachdev P, Anstey KJ. Being overweight is
associated with hippocampal atrophy: the PATH Through Life Study. International Journal
of Obesity 2015;39:1509–14. https://doi.org/10.1038/ijo.2015.106.
[179] Ho AJ, Stein JL, Hua X, Lee S, Hibar DP, Leow AD, et al. A commonly carried allele of
the obesity-related FTO gene is associated with reduced brain volume in the healthy
elderly. PNAS 2010;107:8404–9. https://doi.org/10.1073/pnas.0910878107.
[180] Kivipelto M, Ngandu T, Fratiglioni L, Viitanen M, Kåreholt I, Winblad B, et al. Obesity
and Vascular Risk Factors at Midlife and the Risk of Dementia and Alzheimer Disease.
Archives of Neurology 2005;62:1556–60. https://doi.org/10.1001/archneur.62.10.1556.
[181] Horn JW, Feng T, Mørkedal B, Strand LB, Horn J, Mukamal K, et al. Obesity and Risk
for First Ischemic Stroke Depends on Metabolic Syndrome: The HUNT Study. Stroke
2021;52:3555–61. https://doi.org/10.1161/STROKEAHA.120.033016.
[182] Lampe L, Zhang R, Beyer F, Huhn S, Masouleh SK, Preusser S, et al. Visceral obesity
relates to deep white matter hyperintensities via inflammation. Annals of Neurology
2019;85:194. https://doi.org/10.1002/ana.25396.
[183] Pi-Sunyer FX. The Obesity Epidemic: Pathophysiology and Consequences of Obesity.
Obesity Research 2002;10:97S-104S. https://doi.org/10.1038/oby.2002.202.
[184] Langkammer C, Krebs N, Goessler W, Scheurer E, Ebner F, Yen K, et al. Quantitative
MR Imaging of Brain Iron: A Postmortem Validation Study. Radiology 2010;257:455–62.
https://doi.org/10.1148/radiol.10100495.
[185] Pirpamer L, Hofer E, Gesierich B, De Guio F, Freudenberger P, Seiler S, et al.
Determinants of iron accumulation in the normal aging brain. Neurobiol Aging
2016;43:149–55. https://doi.org/10.1016/j.neurobiolaging.2016.04.002.
[186] Sun Y, Ge X, Han X, Cao W, Wang Y, Ding W, et al. Characterizing Brain Iron
Deposition in Patients with Subcortical Vascular Mild Cognitive Impairment Using
Quantitative Susceptibility Mapping: A Potential Biomarker. Frontiers in Aging
Neuroscience 2017;9.
[187] Guillemot-Legris O, Muccioli GG. Obesity-Induced Neuroinflammation: Beyond the
Hypothalamus. Trends in Neurosciences 2017;40:237–53.
https://doi.org/10.1016/j.tins.2017.02.005.
[188] BinaryDosage: Creates, Merges, and Reads Binary Dosage Files. USC Division of
Biostatistics; 2020.
[189] Morrison J, Kim A, Gauderman J. GxEScanR: An R package to detect GxE interaction in
a genomewide association study. USC Division of Biostatistics; 2018.
99
[190] Dai JY, Logsdon BA, Huang Y, Hsu L, Reiner AP, Prentice RL, et al. Simultaneously
testing for marginal genetic association and gene-environment interaction. Am J Epidemiol
2012;176:164–73. https://doi.org/10.1093/aje/kwr521.
[191] Cinar [aut O, cre, Viechtbauer W. poolr: Methods for Pooling P-Values from
(Dependent) Tests. 2022.
[192] Gao X, Starmer J, Martin ER. A multiple testing correction method for genetic
association studies using correlated single nucleotide polymorphisms. Genet Epidemiol
2008;32:361–9. https://doi.org/10.1002/gepi.20310.
[193] Bray GA. Medical consequences of obesity. J Clin Endocrinol Metab 2004;89:2583–9.
https://doi.org/10.1210/jc.2004-0535.
[194] for the PROSPER Study Group. Increased amygdalar and hippocampal volumes in
elderly obese individuals with or at risk of cardiovascular disease. The American Journal of
Clinical Nutrition 2011;93:1190–5. https://doi.org/10.3945/ajcn.110.006304.
[195] Zhao B, Zhang J, Ibrahim JG, Luo T, Santelli RC, Li Y, et al. Large-scale GWAS reveals
genetic architecture of brain white matter microstructure and genetic overlap with cognitive
and mental health traits (n = 17,706). Mol Psychiatry 2021;26:3943–55.
https://doi.org/10.1038/s41380-019-0569-z.
[196] de Vries PS, Boender J, Sonneveld MAH, Rivadeneira F, Ikram MA, Rottensteiner H, et
al. Genetic variants in the ADAMTS13 and SUPT3H genes are associated with
ADAMTS13 activity. Blood 2015;125:3949–55. https://doi.org/10.1182/blood-2015-02-
629865.
[197] Cao Y, Xu H, Zhu Y, Shi M-J, Wei L, Zhang J, et al. ADAMTS13 maintains
cerebrovascular integrity to ameliorate Alzheimer-like pathology. PLoS Biol
2019;17:e3000313. https://doi.org/10.1371/journal.pbio.3000313.
[198] Barutcu AR, Tai PWL, Wu H, Gordon JAR, Whitfield TW, Dobson JR, et al. The bone-
specific Runx2-P1 promoter displays conserved three-dimensional chromatin structure with
the syntenic Supt3h promoter. Nucleic Acids Research 2014;42:10360.
https://doi.org/10.1093/nar/gku712.
[199] Vimalraj S, Arumugam B, Miranda PJ, Selvamurugan N. Runx2: Structure, function, and
phosphorylation in osteoblast differentiation. International Journal of Biological
Macromolecules 2015;78:202–8. https://doi.org/10.1016/j.ijbiomac.2015.04.008.
[200] Medina-Gomez C, Kemp JP, Trajanoska K, Luan J, Chesi A, Ahluwalia TS, et al. Life-
Course Genome-wide Association Study Meta-analysis of Total Body BMD and
Assessment of Age-Specific Effects. Am J Hum Genet 2018;102:88–102.
https://doi.org/10.1016/j.ajhg.2017.12.005.
100
[201] Di Pietro L, Barba M, Palacios D, Tiberio F, Prampolini C, Baranzini M, et al. Shaping
modern human skull through epigenetic, transcriptional and post-transcriptional regulation
of the RUNX2 master bone gene. Sci Rep 2021;11:21316. https://doi.org/10.1038/s41598-
021-00511-3.
[202] Adhikari K, Fuentes-Guajardo M, Quinto-Sánchez M, Mendoza-Revilla J, Camilo
Chacón-Duque J, Acuña-Alonzo V, et al. A genome-wide association scan implicates
DCHS2, RUNX2, GLI3, PAX1 and EDAR in human facial variation. Nat Commun
2016;7:11616. https://doi.org/10.1038/ncomms11616.
[203] Ning K, Duffy BA, Franklin M, Matloff W, Zhao L, Arzouni N, et al. Improving brain
age estimates with deep learning leads to identification of novel genetic factors associated
with brain aging. Neurobiology of Aging 2021;105:199–204.
https://doi.org/10.1016/j.neurobiolaging.2021.03.014.
[204] Chen Y, Zhao X, Wu H. Transcriptional Programming in Arteriosclerotic Disease.
Arteriosclerosis, Thrombosis, and Vascular Biology 2021;41:20–34.
https://doi.org/10.1161/ATVBAHA.120.313791.
[205] Cobb AM, Yusoff S, Hayward R, Ahmad S, Sun M, Verhulst A, et al. Runx2 (Runt-
Related Transcription Factor 2) Links the DNA Damage Response to Osteogenic
Reprogramming and Apoptosis of Vascular Smooth Muscle Cells. Arteriosclerosis,
Thrombosis, and Vascular Biology 2021;41:1339–57.
https://doi.org/10.1161/ATVBAHA.120.315206.
[206] Wang G, Zheng C. Zinc finger proteins in the host-virus interplay: multifaceted functions
based on their nucleic acid-binding property. FEMS Microbiology Reviews
2021;45:fuaa059. https://doi.org/10.1093/femsre/fuaa059.
[207] Delmaghani S, Aghaie A, Michalski N, Bonnet C, Weil D, Petit C. Defect in the gene
encoding the EAR/EPTP domain-containing protein TSPEAR causes DFNB98 profound
deafness. Human Molecular Genetics 2012;21:3835–44.
https://doi.org/10.1093/hmg/dds212.
[208] Peled A, Sarig O, Samuelov L, Bertolini M, Ziv L, Weissglas-Volkov D, et al. Mutations
in TSPEAR, Encoding a Regulator of Notch Signaling, Affect Tooth and Hair Follicle
Morphogenesis. PLOS Genetics 2016;12:e1006369.
https://doi.org/10.1371/journal.pgen.1006369.
[209] Malik R, Chauhan G, Traylor M, Sargurupremraj M, Okada Y, Mishra A, et al.
Multiancestry genome-wide association study of 520,000 subjects identifies 32 loci
associated with stroke and stroke subtypes. Nat Genet 2018;50:524–37.
https://doi.org/10.1038/s41588-018-0058-3.
101
[210] Wang H, Yang J, Schneider JA, De Jager PL, Bennett DA, Zhang H-Y. Genome-wide
interaction analysis of pathological hallmarks in Alzheimer’s disease. Neurobiology of
Aging 2020;93:61–8. https://doi.org/10.1016/j.neurobiolaging.2020.04.025.
[211] Reid AT, van Norden AGW, de Laat KF, van Oudheusden LJB, Zwiers MP, Evans AC,
et al. Patterns of cortical degeneration in an elderly cohort with cerebral small vessel
disease. Human Brain Mapping 2010;31:1983–92. https://doi.org/10.1002/hbm.20994.
[212] de Laat KF, Reid AT, Grim DC, Evans AC, Kötter R, van Norden AGW, et al. Cortical
thickness is associated with gait disturbances in cerebral small vessel disease. NeuroImage
2012;59:1478–84. https://doi.org/10.1016/j.neuroimage.2011.08.005.
[213] Cheng Y, Liu X, Ma X, Garcia R, Belfield K, Haorah J. Alcohol promotes waste
clearance in the CNS via brain vascular reactivity. Free Radical Biology and Medicine
2019;143:115–26. https://doi.org/10.1016/j.freeradbiomed.2019.07.029.
Abstract (if available)
Abstract
Cerebrovascular disease risk factors, many of which are also risk factors for dementia, contribute substantially to adverse brain changes even in the absence of clinically overt cerebrovascular disease or dementia. Neuroimaging studies have found that these risk factors, such as hypertension, obesity, and hyperlipidemia, are associated with brain atrophy, microinfarcts, microbleeds, and white matter hyperintensities. These adverse changes increase cognitive impairment, dementia, and stroke risk. Identifying novel and understanding existing cerebrovascular risk factors is essential for preventing their negative impact. This dissertation therefore has two main aims to advance this objective. The first is to investigate the association of atrial fibrillation and abnormal P-wave duration, as measured on electrocardiogram, with brain structure. Both of these heart-related factors have many unknowns in terms of their relationship with brain structure. The second focus is to examine whether the association of well-known cerebrovascular factors such as alcohol consumption and obesity is modified by genetic variation, which may provide insight into how these factors lead to cerebrovascular disease. We used data from the UK Biobank resource, a population-scale prospective cohort study in the United Kingdom, to approach these aims. We found that atrial fibrillation and both an abnormally short and long P-wave duration were associated with brain structure and composition. Second, we found that the association of alcohol consumption and brain structure varied by Alzheimer disease genetic risk and that there exist variants significantly associated with hippocampal T2* value, a metric reflecting iron deposition. One such variant additionally had an association with serum triglyceride levels that varied with body mass index. Overall, these results suggest that cardiac pathology beyond coronary heart disease has a clear link with brain health and that there exist genetic variants that modify the association of cerebrovascular risk factors with brain structure.
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
Neuroimaging markers of risk & resilience to brain aging and dementia
PDF
Characterizing brain aging with neuroimaging, health, and genetic data
PDF
Genetic risk factors in multiple myeloma
PDF
Vascular contributions to brain aging along the Alzheimer's disease continuum
PDF
Alzheimer’s disease: dysregulated genes, ethno-racial disparities, and environmental pollution
PDF
Genetic and environmental risk factors for childhood cancer
PDF
Utility of polygenic risk score with biomarkers and lifestyle factors in the multiethnic cohort study
PDF
Blood-brain barrier pathophysiology in cognitive impairment and injury
PDF
The role of vascular dysfunction in cognitive impairment
PDF
Neuroimaging in complex polygenic disorders
PDF
Prostate cancer: genetic susceptibility and lifestyle risk factors
PDF
Independent and interactive effects of depression genetic risk and household socioeconomic status on emotional behavior and brain development
PDF
Shortcomings of the genetic risk score in the analysis of disease-related quantitative traits
PDF
Relationships between lifetime chronic stress exposure, vascular risk, cognition, and brain structure in HIV
PDF
Cardiovascular disease risk factors and cognitive function
PDF
Are life events differentially associated with dementia risk by gender? A twin study
PDF
Identifying genetic, environmental, and lifestyle determinants of ethnic variation in risk of pancreatic cancer
PDF
Associations between longitudinal loneliness, epigenetic age, and dementia risk
PDF
Genetic studies of cancer in populations of African ancestry and Latinos
PDF
The role of the locus coeruleus in Alzheimer’s disease and cerebrovascular function: insights from neuroimaging, neuropsychology, and biofluid markers
Asset Metadata
Creator
Matloff, William Joseph
(author)
Core Title
The association of cerebrovascular disease risk factors with brain structure and its modification by genetic variation
School
College of Letters, Arts and Sciences
Degree
Doctor of Philosophy
Degree Program
Neuroscience
Degree Conferral Date
2022-05
Publication Date
04/20/2022
Defense Date
03/08/2022
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
atrial fibrillation,brain structure,cerebrovascular disease risk factors,gene-environment interactions,neuroimaging,OAI-PMH Harvest,P-wave duration,UK Biobank
Format
application/pdf
(imt)
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Kim, Hosung (
committee chair
), Conti, David V. (
committee member
), Pa, Judy (
committee member
), Toga, Arthur W. (
committee member
)
Creator Email
matloff@usc.edu,william.j.matloff@gmail.com
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-oUC111037428
Unique identifier
UC111037428
Document Type
Dissertation
Format
application/pdf (imt)
Rights
Matloff, William Joseph
Type
texts
Source
20220421-usctheses-batch-931
(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
Tags
atrial fibrillation
brain structure
cerebrovascular disease risk factors
gene-environment interactions
neuroimaging
P-wave duration
UK Biobank